States Office Of Water
_ -.imental Protection (WH-550)
Agency
Pesticides And
Toxic Substances
EPA 570/9-91-020
January 1992
v>EPA
Another Look:
National Survey Of Pesticides
In Drinking Water Wells
Phase 2 Report
-------
ANOTHER LOOK:
National Pesticide Survey
PHASE II REPORT
January 1992
-------
-------
Table of Contents
Page
Executive Summary ix
Chapter One: introduction i
1.1 Review of Survey Implementation and Phase I Results 1
1.2 Phase II 3
Chapter Two: Data Sources and Statistical Approach 9
2.1 Data Sources 9
2.2 Statistical Approach 18
Chapter Three: Results 41
3.1 Stratification Variables 42
3.2 Temporal Allocation 45
3.3 Drastic as a Predictor of Drinking Water Well Contamination 50
3.4 Well Characteristics and Activities Conducted Near the Well 53
3.5 Pesticide Use Data 71
3.6 Fertilizer Sales Data 76
3.7 Well Water Characteristics: Temperature, pH, and Conductivity 80
3.8 Pesticide Chemical Characteristics 83
3.9 Relationship Between Nitrate Detections and Concentrations and
Pesticide Detections 90
3.10 Precipitation and Drought 91
Chapter Four: Evaluation of Results 101
4.1 Major Results 106
4.2 Multivariate Analyses 126
4.3 Potentially Limiting or Confounding Factors 133
4.4 Exposure and Risk Estimates 136
Chapter Five: Conclusions and Recommendations 155
5.1 Conclusions 155
5.2 Recommendations 162
Appendix A: Results with Significance Level (P-Value) Between 0.050
and 0.10 A-i
Bibliography
National Pesticide Survey: Phase II Report
-------
-------
List of Exhibits
Page
1-1 Estimated Number and Percent of Community Water System Wells Containing
NFS Analytes 4
1-2 Estimated Number and Percent of Rural Domestic Wells Containing NFS Analytes ... 4
1-3 Hypotheses Tested in the NFS Phase II Analysis 6
2-1 Type and Scope of Data Used in Phase II Analysis 9
2-2 Sources of Data Used in Phase II Analysis 10
2-3 Agricultural DRASTIC Subcomponent Definitions and Weights 12
2-4 DRASTIC Scores Used in Phase II Analyses 14
2-5 Effective Detection Sizes for Groups of Chemicals in the NFS Community Water
System and Rural Domestic Well Surveys 22
3-1 First Stage or County-Level Strata 42
3-2 Proportion of Sampled Wells with Nitrate and Pesticide Detections by First
Stage Strata 43
3-3 Proportion of Sampled Wells with Nitrate and Pesticide Detections by First
Stage Stratification Variables 44
3-4 Proportion of Nitrate and Pesticide Detections for the Rural Domestic Well
Survey Second Stage Strata 44
3-5 Association of Nitrate and Pesticide Detections with First and Second Stage
Stratification Variables 45
3-6 Binomial Confidence Intervals for Random Temporal Allocation 46
3-7 Monthly Distribution of Sampled Wells for the Community Water System Well
Survey 47
3-8 Monthly Distribution of Sampled Wells for the Rural Domestic Well Survey 47
3-9 Nitrate and Pesticide Detections by Month 48
3-10 Nitrate and Pesticide Detections by Temporal Variables in the Community Water
System Well Survey 49
3-11 Association of Pesticide and Nitrate Detections in Community Water System Wells
with County-Level DRASTIC Measures 51
3-12 Association of Pesticide and Nitrate Detections in Rural Domestic Wells
with County-Level DRASTIC Measures 51
National Pesticide Survey: Phase II Report
-------
List of Exhibits iv
3-13 Association of Pesticide and Nitrate Detections in Rural Domestic Wells
with Sub-County DRASTIC Measures 52
3-14 Models of Relationship of Sub-County DRASTIC Factors and Nitrate Concentrations
for Rural Domestic Wells 53
3-15 Factors from NFS Questionnaires Associated with the Probability of Detecting
Nitrate for Community Water System Wells 57
3-16 Models of the Relationship Between Nitrate Detections and Concentration for
Well Depth in Community Water System Wells 58
3-17 Factors from NPS Questionnaires Associated with the Probability of Detecting
Nitrate for Rural Domestic Wells 60
3-18 Models of the Relationship Between Nitrate Detections and Concentrations
and Well Depth and Year of Well Construction for Rural Domestic Wells 61
3-19 Factors from NPS Questionnaires Associated with the Probability of Detecting
at Least One Pesticide or Pesticide Degradate for Community Water System Wells ... 63
3-20 Logistic Regression Model for Well Depth for Community Water System Wells 63
3-21 NPS Pesticide Active Ingredients included in Analysis of Detections by
Reported Pesticide Use 64
3-22 Summary of Pesticide Detection by Reported Pesticide Use for Community Water
System Wells: Pesticide Level 67
3-23 Summary of Pesticide Detections by Reported Pesticide Use for Rural Domestic
Wells: Pesticide Level 68
3-24 Summary by Pesticide Detection and Reported Pesticide Use from the Local Area
Questionnaire: Well Level 69
3-25 Summary of the Number of Pesticide Active Ingredients Reported as Used from
the Local Area Questionnaires by Number of Rural Domestic and Community Water
System Wells Sampled 70
3-26 Summary by Pesticide Detection and Reported Pesticide Use from the Local Area
Questionnaire: Well by Chemical Level 70
3-27 Summary Statistics for Rate of DCPA Use In Agricultural Areas 73
3-28 Estimated Models of the Relationship Between DCPA Acid Metabolites Detections
and DCPA Use by Urban Application 74
3-29 Summary Statistics for Rate of DCPA Use by Urban Application 75
3-30 Estimated Model of the Relationship Between DCPA Acid Metabolites Detections
and DCPA Use on Golf Courses 76
3-31 Summary Statistics for Rate of DCPA Use on Golf Courses 76
3-32 Estimated Models for the Relationship of Nitrate Concentrations By Total Tons
of Nitrogen Sold per County Acre 78
National Pesticide Survey: Phase II Report
-------
List of Exhibits
3-33 Estimated Models for Selected Census of Agriculture Variables by Nitrate
Detections [[[ 79
3-34 Estimated Models for Selected Census of Agriculture Variables by Pesticide
Detections [[[ 79
3-35 Estimated Models for Selected Census of Agriculture Variables by Nitrate
Concentrations Above 0.15 mg/L ......................................... 80
3-36 Association of Detections with Well Water Temperature for Community Water
System Wells [[[ 81
3-37 Association of Detections with Well Water pH for Community Water System Wells ... 82
3-38 Association of Detections of Nitrate and Nitrate Concentrations with Well
Water Conductivity for Community Water System Wells ........................ 82
3-39 Association of Well Water pH with Detections of Nitrate for Rural Domestic
Wells 83
3-40 Means and Standard Errors of Soil Half-Life and K^ for Detected Analytes and
Non-Detected Analytes ................................................. 85
3-41 Soil Half-Life and K^, Effects for Detected Analytes and Non-Detected Analytes ..... 85
3-42 Soil Half-Life and K^ for Detected Pesticides ................. . ............. 86
3-43 Soil Half-Life and K for Pesticides that Were Not Detected .................... 87
3-44 Models of the Relationship Between Pesticides Detected and Half-Life or
in the NPS [[[ 89
3-45 Analysis of Association between Nitrate and Pesticide Detections ................. 90
3-46 Analysis of Association Between Nitrate Concentration and Pesticide Detections ..... 91
3-47 Estimated Models of the Probability of Nitrate Detections by Precipitation
Factors for Community Water System Wells ................................. 92
3-48 Estimated Models of the Probability of Nitrate Detections by Precipitation
Factors for Community Water System Wells ................................. 93
3-49 Estimated Models of Nitrate Concentration by Precipitation Factors for
Community Water System Wells .................. ........................ 94
3-50 Estimated Models of the Probability of Nitrate Detections by Precipitation
Factors for Rural Domestic Wells ........................................ 94
3-51 Estimated Models of Nitrate Concentration by Precipitation Factors for Rural
Domestic Wells [[[ 95
3-52 Frequency Distribution of Estimated Pesticide Detections by Drought and
-------
List of Exhibits vi
3-54 Frequency Distribution of Estimated Nitrate Detections by Drought and Moist
Spell Categories for Community Water System Wells 96
3-55 Frequency Distribution of Estimated Nitrate Detections by Collapsed Drought
and Moist Categories for Community Water System Wells 97
3-56 Means of Nitrate Concentrations For Collapsed Drought and Moist Categories
for Community Water System Wells 97
3-57 Frequency Distribution of Estimated Pesticide Detections by Drought and Moist
Spell Categories for Rural Domestic Wells 98
3-58 Frequency Distribution of Pesticide Detections by Collapsed Drought and Moist
Spell Categories for Rural Domestic Wells 98
3-59 Frequency Distribution of Estimated Nitrate Detections by Drought and Moist
Spell Categories for Rural Domestic Wells 99
3-60 Frequency Distribution of Estimated Nitrate Detections by Collapsed Drought
and Moist Categories for Rural Domestic Wells 99
3-61 Means of Nitrate Concentrations for Collapsed Drought and Moist Categories
for Community Water System Wells 100
4-1 Comparison of Selected Drinking Water Well Surveys 104
4-2 Summary of Ground-Water Sensitivity Variables Associated with Detections in
Drinking Water Wells 107
4-3 Summary of Pesticide and Fertilizer Use Variables Associated with Detections
in Drinking Water Wells 114
4-4 Summary of Transport Variables Associated with Detections in Drinking Water
Wells 118
4-5 Summary of Chemical Variables Associated with Detections in Drinking Water
Wells 121
4-6 Estimated NPS Logistic Regression Models for the GUS Index 123
4-7 Estimated NPS Logistic Regression Models for Half-Life and K^ 124
4-8 Physical Characteristics of Wells Associated with Detections in Drinking
Water Wells 125
4-9 Estimated Logistic Regression Model for the Probability of Detecting at Least One
Pesticide in Community Water System Wells 126
4-10 Alternative Estimated Logistic Regression Model for the Probability of Detecting at
Least One Pesticide in Community Water System Wells 127
4-11 Estimated Logistic Regression Model for the Probability of Detecting at Least
One Pesticide in Rural Domestic Wells 127
4-12 Alternative Estimated Logistic Regression Model for the Probability of
Detecting at Least One Pesticide in Rural Domestic Wells 128
National Pesticide Survey: Phase I! Report
-------
List of Exhibits vii
4-13 Estimated Logistic Regression Model for the Probability of Detecting Nitrate in
Community Water System Wells 128
4-14 Alternative Estimated Logistic Regression Model for the Probability of Detecting
Nitrate in Community Water System Wells 129
4-15 Estimated Logistic Regression Model for the Probability of Detecting Nitrate in
Rural Domestic Wells 129
4-16 Alternative Estimated Logistic Regression Model for the Probability of Detecting
Nitrate in Rural Domestic Wells 130
4-17 Estimated Linear Regression Model for the Logarithm of Nitrate Concentrations
in Community Water System Wells 130
4-18 Alternative Estimated Linear Regression Model for the Logarithm of Nitrate
Concentrations in Community Water System Wells 131
4-19 Estimated Linear Regression Model for the Logarithm of Nitrate Concentrations
in Rural Domestic Wells 132
4-20 Alternative Estimated Linear Regression Model for the Logarithm of Nitrate
Concentrations in Rural Domestic Wells 132
4-21 Alternative Estimated Linear Regression Model for the Logarithm of Nitrate
Concentrations in Rural Domestic Wells 133
4-22 Summary Statistics for NPS MRLs by Detected and Non-Detected NPS Pesticides ... 135
4-23 Estimated Number and Percent of Wells Containing Nitrate or DCPA Acid
Metabolites 136
4-24 Estimates of the Nitrate Concentration Distribution for Rural Domestic Wells 137
4-25 Estimated Distribution of Nitrate Concentration in Rural Domestic Drinking
Water Wells 139
4-26 Estimates of the DCPA Acid Metabolites Concentration Distribution for Rural
Domestic Wells 140
4-27 Estimated Distribution of DCPA Acid Metabolites Concentration in Rural Domestic
Drinking Water Wells 141
4-28 Estimates of the Nitrate Concentration Distribution for Community Water
System Wells 142
4-29 Estimated Distribution of Nitrate Concentration in Community Water System Wells . 143
4-30 Estimates of the DCPA Acid Metabolites Concentration Distribution for Community
Water System Wells 144
4-31 Estimated Distribution of DCPA Acid Metabolites Concentration in Community
Water System Wells 145
4-32 Plot of Predicted Versus Observed Nitrate Concentration in Rural Domestic Wells . . 147
National Pesticide Survey: Phase II Report
-------
List of Exhibits viii
4-33 Estimates of Population Exposed to Nitrate in Rural Domestic Wells by
Distribution Percentile 148
4-34 Estimates of Population Exposed to Nitrate by Concentration in Rural
Domestic Wells 148
4-35 Estimates of Population Exposed to Nitrate in Community Water System Wells
by Distribution Percentile 149
4-36 Estimates of Population Exposed to Nitrate by Concentration in Community
Water System Wells 149
4-37 Estimate of Population Exposed to DCPA Acid Metabolites in Rural Domestic
Wells at 99th Percentile 150
4-38 Estimates of Population Exposed to DCPA Acid Metabolites by Concentration
in Rural Domestic Wells 150
4-39 Estimates of Population Exposed to DCPA Acid Metabolites in Community Water
System Wells by Distribution Percentile 151
4-40 Estimates of Population Exposed to DCPA Acid Metabolites by Concentration
in Community Water System Wells 151
4-41 Estimates of Percentages of Wells and Number of Wells Containing At Least
One Pesticide Above Health-Based Levels 152
4-42 Estimates of Population Exposed to Pesticide Above Reporting and Health-
Based Levels 152
National Pesticide Survey: Phase II Report
-------
Executive Summary
The U.S. Environmental Protection Agency (EPA) has completed the Phase II Report for its National
Survey of Pesticides in Drinking Water Wells (NPS). Between 1988 and 1990, EPA sampled 1349 community
water system (CWS) wells and rural domestic drinking water wells for the presence of 101 pesticides, 25
pesticide degradates, and nitrate (127 analytes). The Phase II Report describes the outcome of an extensive
analysis of factors potentially affecting pesticide and nitrate occurrence in drinking water wells.
EPA designed the Survey with two principal objectives corresponding to its two phases. In Phase I,
completed in November 1990, EPA developed national estimates of the frequency and concentration of
pesticides and nitrate present in drinking water wells in the United States. In Phase II, EPA investigated the
NPS data alone and in combination with pertinent data from non-NPS sources. The Phase II analyses sought
to improve EPA's understanding of how the presence of pesticides and nitrate (measured as N) in drinking
water wells is associated with patterns of pesticide use, the sensitivity of areas surrounding drinking water wells
to ground-water contamination, transport of chemicals to well water, and other factors, including pesticide
chemistry and the physical condition of drinking water wells.
Discussion of Findings
The NPS, as a whole, represents the only national data currently available to EPA on the extent and
nature of the presence of pesticides and nitrate in drinking water wells. The NPS Phase II analysis investigated
possible correlations or associations between detections of pesticides or nitrate in drinking water wells and
a number of factors that might be related to their presence in those wells.
Statistical Methods
The Phase II analysis followed standard statistical procedures to evaluate the data gathered by the
Survey. EPA tested hypotheses about the relation between individual variables and pesticide and nitrate
detections and the degree of association between them. EPA used a criterion of significance that ensures that
there was less than a 5 percent probability that any single test result occurred by chance alone. The Phase II
analyses include tests of association which use chi square analysis, linear and logistic univariate regression, and
multiple regression to better understand the association of pesticide and nitrate detections in drinking water
wells with ground-water sensitivity, pesticide and nitrate use, transport mechanisms, pesticide chemistry, and
well condition, as each of them was defined for the NPS analysis.
The results obtained by the Phase II analysis represent associations between detections of pesticides
or nitrate in wells and other factors that may indicate the presence of chemicals in water from drinking water
wells. The findings reported in the Phase II Report have passed appropriate statistical tests for significance
and have been examined for consistency with current theories of well-water contamination as well as the results
of other selected similar surveys. Because the NPS was designed first and foremost to satisfy the objective of
generating national estimates of pesticide and nitrate contamination of drinking water wells, the statistical
analysis of Phase II was constrained in a number of ways. Most important, the set of hypotheses that could
be examined accurately was restricted by the number of observed detections. The number of detections were
limited by two factors: sample size and reporting limits. The number of samples taken was limited to the
number sufficient to produce the national estimates presented in the Phase I Report. Multi-residue chemical
analytic methods using stringent minimum reporting limits (MRLs) and confirmation of detections ensured
that false positive results were avoided when detections in drinking water wells were reported. The statistical
techniques used in Phase II require larger numbers of detections to perform effectively than the techniques
for estimating population proportions and their standard errors used in Phase I. The effective number of
detections for nitrate was approximately 220 in CWS wells and 232 in rural domestic wells; the effective
detection size for the "any pesticide" variable, the next largest chemical group included in the analysis, was
approximately 44 for CWS wells and 17 for rural domestic wells. In certain cases, the number of detections
of DCPA acid metabolites in wells also was sufficient to allow statistical analysis to be carried out. The other
detected analytes were detected in wells too few times to be included separately in the analyses.
National Pesticide Survey: Phase II Report
-------
Executive Summary
Data Sources
The data used in the Phase II analysis came from a broad range of sources. Generally either the
reported detections of pesticides or nitrate in wells or the concentrations of nitrate in wells were evaluated for
associations with factors that could serve as indicators of contamination in drinking water wells. Only
detections above the NFS MRLs reported by the primary laboratories were used in the analysis. In a few
cases, when concentrations could not be reliably measured above the MRL, the corresponding detections were
not included in the analysis to ensure consistency and accuracy in results. Detections reported only by NFS
referee laboratories are not included because those laboratories provided quality assurance review rather than
comprehensive analysis of all samples.
The other variables included in the analysis came both from data generated by the NFS and data
obtained from sources outside the Survey. Data sources include the following:
• Data collected from NFS questionnaires. Data on pesticide use, spills, and disposal;
agricultural activities; well age, depth, and construction; topography; and surface
water near the well were developed from information collected on NFS
questionnaires. (Copies of the NFS questionnaires and summaries of much of the
data collected from the questionnaires can be found in the NFS Phase I Report.)
• Data from NFS stratification. EPA developed county-level and sub-county level
DRASTIC measures of ground-water vulnerability, using a broad variety of geologic
and hydrogeologic data often obtained from U.S. Geologic Service sources. EPA
also developed measures of agricultural pesticide use intensity, using data on
pesticide sales and cropping, and a sub-county measure of "cropped and vulnerable"
used to identify sensitive sub-county areas. The pesticide sales data were from a
proprietary data base generated by Doanes Marketing Research, Inc. with data from
1984.
• Pesticide use data. EPA developed county-level estimates of pesticide use in
agricultural and non-agricultural applications from regional and state-level data
assembled for EPA by Resources for the Future.
• Fertilizer sales data. EPA developed county-level estimates of fertilizer sales from
materials collected under contract for EPA, including state-level data collected by the
Tennessee Valley Authority.
• Data from the U.S. Department of Agriculture's Census of Agriculture. EPA
obtained county-level data on agricultural activities in the United States from the
1987 Census of Agriculture from which measures of crop values, cropping activity,
and livestock operations were extracted.
• Rainfall and drought data. EPA obtained data from the National Climate Data
Service of the National Oceanic and Atmospheric Administration (NOAA) on
rainfall for five years prior to sampling at NFS well locations, and on drought
conditions, as estimated by NOAA using the Palmer Drought Index.
• Data on physical/chemical characteristics of pesticides. EPA obtained values for
physical/chemical properties of pesticides - half life and K^ - from data collected
by EPA's Office of Pesticide Programs.
• Data developed by other surveys of pesticides or nitrate in ground water or well
water. EPA reviewed the reports of selected recent surveys, including the National
Alachlor Well Water Survey and four state surveys, to identify similarities and
differences among the reported results.
National Pesticide Survey: Phase II Report
-------
Executive Summary xi
Associations Evaluated
EPA carried out over a thousand statistical tests evaluating hypotheses about the association of
pesticides and nitrate in drinking water wells with many different variables. These analyses included:
• Studies of the association between detections of NPS analytes in drinking water wells
and Agricultural DRASTIC scores and subscores for county-level and for sub-county
level evaluations of the sensitivity of ground water to contamination.
• Studies of the association between detections in wells and drinking water well
characteristics, reports of pesticide and fertilizer use on or near the property where
the well is located, fanning and animal husbandry near wells, the presence of surface
water, and other variables derived from questionnaire data. Two categories of data
that were not gathered and therefore were not included in the analysis are data on
the ground-water flow and recharge area for wells and data on the soil profile near
wells.
• Studies of the association between detections in wells and estimates of nitrogen
fertilizer and pesticide use and surrogate measures of agricultural activities and
livestock operations derived from data obtained by EPA from non-NPS sources such
as the Census of Agriculture.
• Studies of the association between detections in wells and the physical/chemical
characteristics of pesticides and ground water in the drinking water wells sampled.
The Survey did not collect data that would allow analysis of soil composition and
chemistry near the sampled drinking water wells.
• Studies of the association between detections in wells and measures of precipitation
and drought.
Individual results from these analyses were grouped according to their relationship to five factors
considered to affect well water contamination.
• Sensitivity is defined for this review of the analytic results as the intrinsic
susceptibility of an aquifer to contamination. Sensitivity addresses the hydrogeologic
characteristics of the aquifer and the overlying soil and geologic materials, and is
unrelated to agricultural practices, well construction, or pesticide characteristics.
• Pesticide and fertilizer use is defined for the NPS Phase II analysis as the scope and
amount of pesticides and nitrogen fertilizers applied near sampled drinking water
wells, and activities, such as cattle raising, that could contribute to the presence of
pesticides or nitrate in places where they could be transported into drinking water
wells.
• Transport is defined as factors that contribute, either directly or indirectly, to the
movements of pesticide and nitrate by water, including precipitation, irrigation,
surface water, drainage ditches near the well, other bodies of water such as rivers and
lakes, and nearby operating wells. This definition is narrower than customary
definitions of transport, since it does not include recharge or ground-water flow. It
applies only to the NPS Phase II analysis.
• Chemical characteristics are defined as characteristics of pesticides, such as organic
partition coefficients (K^) and half-life, and characteristics of well water, such as
water temperature, pH, and conductivity, that could affect the behavior of pesticides
or nitrate.
National Pesticide Survey: Phase II Report
-------
Executive Summary xii
• Physical characteristics of wells, are defined as age, depth, state of repair, and
protective devices, such as concrete pads, that could affect the presence of chemicals
in the well.
Evaluation Criteria
The results reported in Phase II have been evaluated in the context of several criteria, including the
accuracy, area of geographic coverage, and the time period in which the data were collected; the statistical
significance of the results, the size of the effect, the sample size, and patterns of results within the survey and
the comparison of results between NFS and other studies. The reported results should not be interpreted as
conclusions drawn from controlled experiments or as evidence of causation. The results are statistical
measures of association or correlation that were designed to test hypotheses developed from several scientific
and policy questions. The results should be used with caution, as indicators of topics that may be suitable for
further careful investigation and as information to supplement interpretation of previous studies. In several
cases, results that would have been expected, on the basis of other studies or theoretical understanding of the
processes involved, were not identified by the NFS Phase II analysis. Because of the many limits on the
analysis, this should not be taken as an indication that such factors are not important.
Key Results of Analysis of Single Variables
Key results for tests of association between detections in drinking water wells and individual factors
came primarily from analysis of nitrate detections in CWS wells. The relatively large number of associations
involving nitrate detections is partly due to the fact that a significantly larger number of nitrate detections than
pesticide detections were obtained. These key results include:
Sensitivity
• The Survey's county-level and sub-county level Agricultural DRASTIC assessments
did not prove to be a useful means of locating drinking water wells containing
pesticides or nitrate. Although DRASTIC did help to ensure that the Survey
achieved broad national coverage, total DRASTIC scores were not found to be
related to pesticide or nitrate detections. The failure of DRASTIC scores to be
positively associated with pesticide detections in drinking water wells is not a
measure of the validity of the DRASTIC method. EPA used a simplified and cost-
effective procedure to collect information for DRASTIC evaluations at the county
level. The NFS was not designed or implemented as a scientific test of DRASTIC.
The NFS was designed to use DRASTIC as a method to characterize counties in
relatively more or less vulnerable settings.
• Different DRASTIC factors often were associated with detections in CWS wells than
were associated with detections in rural domestic wells. County-level depth to water
table was associated with pesticide detections, and county-level hydraulic conductivity
was associated with nitrate detections in CWS wells. Accurate DRASTIC
measurements for specific well locations are not possible using county-level scores.
For example, intra-county variability is lost in county-level scoring. At the sub-
county level, different types of vadose zone media were associated with pesticide
detections in CWS wells.
• Stratification based on combined Agricultural DRASTIC scores and county cropping
patterns (the "cropped and vulnerable" strata) was not effective in improving the
precision of the Survey results. The Phase II analysis did not demonstrate a
relationship between stratification variables and pesticide or nitrate detections.
Oversampling of certain strata in future surveys should be undertaken only if the
criteria used are known to be closely associated with detections and can be measured
with sufficient accuracy to improve the survey estimates and precision.
National Pesticide Survey: Phase II Report
-------
Executive Summary xiii
Use of Pesticides and Nitrate
• The Survey's data on pesticide use, collected both from local landowners and from
experts on local cropping and agrichemical uses, in many cases did not indicate that
pesticides had been used within the previous 3 or 5 years near wells where they were
detected. These data, collected for the area within 300 or 500 feet of the well or
within one-half mile of the well, depending on the question and respondent, may not
have adequately reflected pesticide use within the recharge area for the well.
• Measures of agronomic activity based on state-level data, including the market value
of crops and livestock produced in the county where the well is located, were
associated with detections, particularly for nitrate.
• The degradates of the pesticide DCPA were detected in both CWS wells and rural
domestic wells. The amount of DCPA use on golf courses and in urban applications
was related to the possibility of detecting DCPA acid metabolite in CWS wells and
rural domestic wells, based on extrapolations of regional data on DCPA use.
Transport
• Variables relating to transport of chemicals to ground water provide suggestive
evidence concerning the processes affecting contamination. Nitrate detections were
more likely in CWS wells drawing water from an unconfined aquifer. The presence
of surface bodies of water near sampled wells generally was associated with a reduced
likelihood of detections.
• Survey results suggest that there is a lower probability of detecting pesticides or
nitrate in wells in counties that experienced high levels of rainfall. The result
suggests that high levels of rainfall may cause the pesticides to run off before
entering ground water or to dilute the concentration of pesticides and nitrate to the
point that they fall below the detection limits used by the Survey. In contrast, flood
irrigation was associated with a greater likelihood of detection.
• Evaluation of the associations between detections in drinking water wells and Palmer
Drought Index scores prepared by the National Oceanic and Atmospheric
Administration provided evidence that nitrate concentrations are lower in wells in
moist regions.
Chemical Characteristics
• The likelihood of detecting pesticides or nitrate is greater in wells with low water
temperature or low water pH.
• The likelihood of detecting nitrate is greater in wells delivering water with low
electrical conductivity. Higher conductivity of well water was found to correspond
to higher nitrate concentrations.
• The Survey examined the association of pesticide persistence, based on half-life in
soil, with detections. Persistent pesticides were more likely to be detected in
drinking water wells than pesticides with short half-lives.
• Estimates of concentrations for the chemicals most frequently detected in the NPS
suggest that there is a significant percentage of wells containing concentrations of
pesticides or nitrate below the MRLs that might have been detected without such
limits.
National Pesticide Survey: Phase II Report
-------
Executive Summary xiv
Well Construction and Condition
• Shallower wells were associated with nitrate detections in both CWS and rural
domestic wells and with pesticide detections in CWS wells. An association was also
found between detections of nitrate in rural domestic wells and older wells.
Results of Multivariate Analyses
The Phase II analysis carried out multivariate regression analyses to determine if combinations of
variables would be particularly strong predictors of pesticide and nitrate contamination of drinking water wells.
Because of the relatively small number of pesticide detections and substantial commonality among variables,
however, the predictive value of many of the factors considered in the analysis was difficult to evaluate. Very
few variables appeared in more than one model. The best models were identified on the basis of how well
they fit the data. The appearance of different variables in the models does not necessarily reflect physical or
theoretical interchangeability of the variables.
For pesticide detections, two variables -- fertilized pasture and rangeland, measured at the county level,
and well depth -- created the best model for detections of a pesticide in CWS wells. The presence of other
operating wells near the sampled wells can be used in place of the well depth variable without substantially
reducing model performance. An indirect measure of agricultural activity - market value of crops in
thousands of dollars -- by itself was the best model for pesticide detections in rural domestic wells. A variable
measuring the number of beef cattle per acre could also be included.
For nitrate detections in drinking water wells, a three-variable model, composed of fertilized pasture and
rangeland, average monthly precipitation, and well-water pH, created the strongest results for nitrate detections
in CWS wells. Farming on the property where the well is located also can be included in the model. A four-
variable model, composed of well age, maximum monthly precipitation in the five years prior to sampling, well
water pH, and the presence of an unlined drainage ditch within less than one-half mile, created the best model
for nitrate detections in rural domestic wells. A variable for fertilized pasture and rangeland could also be
included.
For nitrate concentrations in CWS wells, five variables - maximum monthly precipitation in the past
five years, well water electrical conductivity, total nitrogen sales by county for counties containing wells in
which nitrate was detected, well depth, and a categorical variable that reflects the Palmer Drought Index score
for the year prior to sampling - create the best model. Crop value can be used in place of total nitrogen sales
without substantially reducing model performance. For nitrate concentrations in rural domestic wells, four
variables are included in the best model - well depth, market value of crops, presence of an unlined body of
water within one-half mile, and the Agricultural DRASTIC subscore for topography measured at the sub-
county level. Total nitrogen sales can be used in place of crop value. A variable for the presence of a body
of water within 300 feet of the well could also be included.
Results of Population Exposure Analysis
The Phase II analysis prepared estimates of concentration distributions in drinking water wells for
DCPA acid metabolites and nitrate. Based on these concentration estimates, the frequency of occurrence of
these chemicals in CWS wells and rural domestic wells is somewhat higher than presented in the NPS Phase
I Report based on national estimates that were calculated from concentrations that exceeded minimum
reporting limits. Quantitative exposure estimates prepared for DCPA acid metabolites and nitrate using these
concentration estimates suggest that current potential health effects are low for these analytes detected most
frequently in the Survey. This result confirms the similar conclusion reached in the Phase I Report that Survey
results do not demonstrate any immediate widespread health concern.
Approximately 19 million people are estimated to be exposed to nitrate in rural drinking water wells,
with about 1.5 million exposed to levels of nitrate over the Maximum Contaminant Level of 10 mg/L (ppm),
and approximately 22,500 infants under 1 year old exposed to a concentration of greater than 10 mg/L.
National Pesticide Survey: Phase II Report
-------
Executive Summary xv
Persons with high levels of nitrate in their private wells should consult their pediatricians and may wish to
obtain water from alternate sources that have less than 10 mg/L of nitrate to help protect infants from the risk
of methemoglobinemia (blue-baby syndrome). Physicians are usually well informed a.bout the risks to infants
of high levels of nitrate in drinking water and are able to provide medical treatment. Approximately 85
million people are estimated to drink water from CWS wells that contain nitrate, with about 3 million exposed
to levels of nitrate over 10 mg/L (ppm), and approximately 43,500 infants under 1 year old exposed to a
concentration of greater than 10 mg/L. Public water supplies that violate the Maximum Contaminant Level
of 10 mg/L for nitrate are required to notify their customers about the violation, and the adverse health effects
caused by nitrate. (40 CFR 141.32) Local and state health authorities are the best source for information
concerning alternate sources of drinking water for infants. Systems that apply for variances or exemptions
while in violation of the standard may be required by the state to provide bottled water or point-of-use or
point-of-entry devices to avoid unreasonable risks to health. (40 CFR 141.62(f))
Results of Individual Risk Analysis
EPA estimates that about 10.4 percent of CWS wells and 4.2 percent of rural domestic wells contain
detectable levels of one or more pesticides. The Phase II study estimated the chance that a well that contains
one or more pesticides also exceeds a maximum contaminant level or health advisory. EPA estimates that no
more than 7.3 percent of the 10.4 percent of CWS wells that contain one or more pesticides could exceed an
MCL or health advisory. Similarly, no more than 28.3 percent of the 4.2 percent of rural domestic wells that
contain detectable levels of one or more pesticides are also expected to exceed a health based limit. In
summary, about 1 percent of all drinking water wells in the U.S. are estimated to exceed a health based limit.
EPA concluded that the overall chance of a given well exceeding a level of concern for a pesticide is low. If
a well contains a detectable amount of one or more pesticides, it has a slightly higher risk of also exceeding
a health based limit. EPA recommends that well owners that know or suspect that their well is affected by
pesticides have the water tested to ensure that any pesticides are present at levels below the MCLs or health
advisories.
Recommendations for Future Studies
The NPS was the first survey of the presence of pesticides and nitrate in public and private drinking
water wells throughout the United States. Its results provide several useful lessons for the design of future
studies. The Survey design functioned effectively to produce the data upon which the Phase I national
population estimates were based. To provide the best data for complex statistical analysis of results of future
surveys, however, four topics should be carefully evaluated.
First, features such as stratification should be used to control for possible factors that may confuse
or confound results, but the stratification variables must be known to have a substantial effect on what is being
measured. Oversampling of selected strata should be carried out only if the variables that define the strata
can be defined and measured with accuracy.
Second, pilot studies should test and evaluate statistical analysis approaches.
Third, the specifications for survey size and precision and the limits established in chemical analytic
procedures for reporting detections (such as minimum reporting limits) should be chosen to ensure that a
sufficient number of detections are likely to be obtained to support complex data analysis and statistical
modelling of the survey data. All laboratory results should be reported. Laboratory analyses must include
confirmation steps. The laboratory should then report the data with sufficient information to judge the
reliability of the reported concentration. For example, results reported below the minimum reporting limit
must be flagged to alert the data user to the inherent variability of concentrations reported at such low levels.
The frequency of detection of DCPA acid metabolites confirms the importance of considering pesticide
metabolites in survey design.
National Pesticide Survey: Phase II Report
-------
Executive Summary xvi
Fourth, site-specific data on ground-water sensitivity and pesticide use should be obtained. Data on
the recharge patterns for particular wells should be investigated. Locally precise pesticide use and distribution
data should include non-farm as well as farm pesticides, and data should be gathered on both farm and non-
farm uses of fertilizers. The data, which should be publicly available, will help to ensure that assessments of
the vulnerability of drinking water wells to contamination can be improved and made more reliable.
The Survey analysis also identified a number of additional topics for future study. They include studies
of seasonal and temporal effects on contamination, analysis of links between surface and ground-water
contamination, and collection and evaluation of site-specific data on soil characteristics and recharge and their
association with contamination patterns in wells.
In summary, the National Pesticide Survey provides useful national information for the formulation
and improvement of policies to protect drinking water wells from contamination by pesticides and nitrate.
The results suggest that many interacting factors affect the quality of drinking water wells.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
The U.S. Environmental Protection Agency's National Survey of Pesticides in Drinking Water Wells
(the National Pesticide Survey, the Survey, or NPS) was designed to determine the frequency and
concentration of the presence of pesticides and nitrate in drinking water wells in the United States. The
Survey's Phase I Report1 addressed this objective. The Phase I Report provides estimates of the number and
proportion of community water system (CWS) wells and rural domestic wells nationwide containing detectable
levels of 126 pesticides and pesticide degradates, plus nitrate. The Phase I Report also contains summary
estimates of drinking water well characteristics derived from questionnaire responses.
Survey results provide an opportunity to examine how detections of pesticides and nitrate in drinking
water wells are associated with a number of factors that could affect their presence in well water. In Phase
II of the Survey analysis, EPA carried out studies of associations between analyte detections in wells and
measures of the vulnerability of the wells' water to contamination; patterns of pesticide and fertilizer use near
wells; current and previous uses of property surrounding wells; hydrology; terrain and other natural features
near wells; depth, age, location, and construction of wells; and other factors. This Phase II Report has been
prepared for readers with a good understanding of statistics and fate and transport mechanisms in ground
water. It describes the data sources and analytic approaches used in Phase II; the key results of the analysis,
along with discussions of the statistical, scientific, or agronomic basis for their significance; and a discussion
of the Survey's implications for potential future analyses of similar issues. It also provides recommendations,
based upon the experience gained in the implementation of the Survey and analysis of its results, for future
work and for implementation of ground-water protection programs.
1.1 Review of Survey Implementation and Phase i Results
The National Pesticide Survey is a joint project of EPA's Office of Ground Water and Drinking Water
(OGWDW) (formerly the Office of Drinking Water) and Office of Pesticide Programs (OPP). Survey design
began in 1984. The design of the Survey was reviewed and endorsed by a special subpanel of EPA's
independent Federal Insecticide, Fungicide, and Rodenticide Act (FIFRA) Scientific Advisory Panel. EPA
conducted a pilot study in California, Mississippi, and Minnesota in 1987 to test Survey implementation and
analytical procedures. After extensive planning and preparation, full scale sampling for the National Pesticide
Survey began in April 1988. The final sample was collected in February 1990. Wells were sampled in every
State, and extensive data were collected about the wells' physical characteristics, in addition to surrounding
circumstances such as nearby pesticide use and agricultural activities.
EPA tested well water from two kinds of wells: wells at community water systems and rural domestic
wells. Community water systems are defined as systems of piped drinking water with at least 15 connections
or serving at least 25 permanent residents. To be eligible for the Survey, a system had to have at least one
operable well, at the time of sampling, that supplied drinking water. Rural domestic wells are defined as
drinking water wells supplying water for human consumption (drinking, cooking, or bathing) to an occupied
private household located in a rural area of the United States. Rural areas are defined as households outside
of incorporated or unincorporated places with a population of 2,500 or more and outside of urban areas.
A multistage stratified selection procedure was used to choose a nationally representative subset of
CWS wells and rural domestic wells for sampling. The results provide a national estimate and are not
representative of any State or local area. In choosing wells, EPA first characterized all counties in the U.S.
1 National Survey of Pesticides in Drinking Water Wells. Phase I Report (EPA 570/9-90-015, November 1990) (hereafter
NPS Phase I Report*). Copies may be obtained from the National Technical Information Service, Springfield, Virginia 22161
(1-800-336-4700) for $35 plus $3 handling. Request report number PB91-125765.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
according to proxies for pesticide use and relative ground-water vulnerability2 ("first-stage stratification").
Agricultural pesticide use was specified as high, moderate, low, or uncommon. EPA specified areas of greater
and lesser relative ground-water vulnerability by using a numerical classification system called DRASTIC,
which considers seven factors that may affect the vulnerability of ground water to contamination. The Survey
design called for a specified number of wells to be sampled from areas of greater and lesser pesticide use and
ground-water vulnerability.
To identify CWS wells, EPA randomly selected 7,083 community water systems from a list containing
information on all public water supply systems. EPA conducted telephone interviews with representatives of
the 7,083 selected systems and, based on the results of the screening process, identified systems eligible for
sampling. Water samples from 566 eligible CWS wells in 50 States were collected. The 540 samples that
passed quality assurance requirements were used in data analysis. When selecting rural domestic wells, EPA
randomly chose 90 counties located in 38 States as areas for sampling to represent the nation's wide range of
agricultural pesticide use and ground-water vulnerability. EPA assessed ground-water vulnerability within sub-
county areas and collected information from county agricultural extension agents on cropping intensity to
further subdivide the counties into areas that are more or less vulnerable to pesticide contamination in ground
water ("second-stage stratification"). Water samples from 783 eligible rural domestic wells were collected. The
752 samples that passed quality assurance requirements were used in data analysis.
Once the wells were selected, EPA scheduled sample collection so that wells were visited throughout
the 22 month sampling period from April 1988 to February 1990. This schedule provided well water samples
during all seasons and pesticide application cycles. This approach was used to minimize the effect that
seasonal variability may have on concentrations of pesticides or nitrate in drinking water wells.
EPA visited each well once, filling a sufficient number of sample bottles with well water to carry out
chemical analyses and QA/QC procedures. At each well sampled, questionnaires were used to collect data,
including:
• observations about the well and the surrounding area;
• information from the owner/operator about well construction and agricultural and
non-agricultural pesticide use on the property where the well was located; and
• information from county agricultural extension agents, who were knowledgeable
about local crops, pesticide use, and land use within one-half mile of the well.
Each water sample gathered in the National Pesticide Survey was tested for 127 analytes. Eight
chemical analytical methods (seven organic and one inorganic) were required. All positive detections of
organic analytes were confirmed by reanalyzing on a second (different) capillary gas chromatographic (GC)
column or a high performance liquid chromatographic (HPLC) column. These analyses provided both a
preliminary qualitative confirmation for all GC determinations and the final confirmation for HPLC analyses.
The analytes detected using a GC based method also were qualitatively confirmed using gas chromatography/
mass spectrometry (GC/MS). Ten percent of the well's samples were sent to referee laboratories to determine
2 When the stratification was carried out, the term "vulnerability" was used to describe the intrinsic susceptibility of an
aquifer to contamination, based on the hydrogeologic characteristics of the aquifer and the overlying soil and geologic
materials. EPA now uses the term "sensitivity" to refer to intrinsic susceptibility, and considers aquifer vulnerability to be a
more comprehensive term, encompassing the susceptibility of an aquifer to contamination from the combined effects of its
sensitivity, agricultural practices, and pesticide characteristics. In this report, the term vulnerability has been used when it
refers to activities undertaken at the time EPA used the term and to variables developed as a result. Thus, stratification is
discussed in terms of vulnerability because that term was used when stratification was carried out and results are reported
for DRASTIC scoring and assessments of vulnerability. The term sensitivity is used to report the evaluation of the Phase
II analyses because it represents the current concept that was analyzed. Irrespective of the term used, it must be noted that
the NFS is a survey of drinking water wells, not of ground water.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
the occurrence of false negatives. The measured concentration of the primary analysis of confirmed detections
was the concentration reported.
EPA established a minimum quantification limit (MQL) and a minimum reporting limit (MRL) for
each analyte. The MQL was based on the precision of the method and the sensitivity of the analytical
instrumentation. MQLs were established with concentrations for 112 of the analytes. Analyses of the
remaining 15 analytes were classified as unreliable because of the instability of the analytes in water or other
factors. The Agency chose to look for the presence of these analytes and only report them as "positive
detections," (i.e., if detected, their presence was reported to the well owner, but no concentration level was
reported). These "positive detections" were not included in the estimates presented in the Phase I report or
the other statistical analyses conducted for the Phase II report. Measured concentrations below the MQL,
down to a concentration of one-half the MQL, were not considered as reliable and also were reported as
"positive detections" but without a concentration level. The cutoff point of one-half the MQL was called the
minimum reporting limit (MRL). No detections or concentrations were reported below the MRLs.
The NPS sampled 1349 CWS wells and rural domestic wells for the presence of 101 pesticides, 25
pesticide degradates, and nitrate (measured as N) (127 analytes). In all, 17 analytes were detected in the
Survey. Thirteen were detected at levels above the NPS minimum reporting limits (MRLs) used by the
primary laboratories that performed the initial analysis of samples; three (alpha-chlordane, gamma-chlordane,
beta HCH) were detected by EPA laboratories at concentrations lower than the respective primary laboratory
MRLs. The concentration of one detected chemical (4-nitrophenol) could not be measured.
Exhibits 1-1 and 1-2 show the estimates generated in Phase I of the number and percent of wells of
the estimated 94,600 CWS wells and the 10.5 million rural domestic wells in the U.S. containing each of the
13 chemicals detected at concentrations above the Survey's MRLs.3 EPA estimated that about 10.4 percent
of the CWS wells nationwide (approximately 9,850 wells, although the number could be as high as 13,400 wells
or as low as 6,300 wells, based on the upper and lower 95% confidence bounds) and 4.2% of rural domestic
wells (approximately 446,000 wells, although the number could be as high as 647,000 wells or as low as 246,000
wells) contain detectable levels of one or more pesticides included in the Survey.
As the Phase I Report stated, the number of wells containing pesticides or nitrate could differ from
these results. Local or regional areas with particularly high or low levels of chemicals may not have been
randomly chosen for sampling and additional chemicals might have been present in the wells sampled, at levels
too low to be reported as detections by the Survey.
A detailed description may be found in the NPS Phase I Report of the Survey design and
implementation, national estimates of the number of wells containing NPS analytes, summaries of data derived
from the Survey questionnaires, and copies of the questionnaires. Knowledge of those topics is necessary for
complete understanding of the Phase II results.
1.2 Phase II
The Phase II analysis used data collected during the Survey and data from other sources to investigate
the relationship of pesticide and nitrate detections to five general factors. These factors are:
• Ground Water Sensitivity - the intrinsic susceptibility of an aquifer to
contamination, which addresses the hydrogeologic characteristics of the aquifer and
the overlying soil and geologic materials, but is not related to agricultural practices
or pesticide characteristics;
' NPS Phase I Report, p. 61. The MRL for each Survey analyte is provided in Appendix E of the Phase I Report.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
Exhibit 1-1
Estimated Number and Percent of Community Water System Wells
Containing NFS Analytes
Analyte
Nitrate
DCPA acid metabolites
Atrazine
Simazine
Prometon
Hexachlorobenzene*
Dibromochloropropane (DBCP)*
Dinoseb*
Estimated
Number
49,300
6,010
1,570
1,080
520
470
370
25
95% Confidence
Interval
(lower - upper)
(45,300 - 53,300)
(3,170-8,840)
(420 - 2,710)
(350 - 2,540)
(78-1,710)
(61 - 1,630)
(33-1,480)
(1 - 870)
Estimated
Percent
52.1
6.4
1.7
1.1
0.5
0.5
0.4
<0.1
95%
Confidence
Interval
(lower • upper)
(48.0 - 56.3)
(3.4 - 9.3)
(0.5 - 2.9)
(0.5 - 2.7)
(0.1 - 1.8)
(0.1 - 1.7)
(0.1 - 1.6)
(<0.1 -0.9)
Registration canceled by EPA.
Exhibit 1-2
Estimated Number and Percent of Rural Domestic Wells
Containing NPS Analytes
Analyte
Nitrate
DCPA acid metabolites
Atrazine
Dibromochloropropane (DBCP)*
Prometon
Simazine
Ethylene dibromide*
Lindane
Ethylene thiourea
Bentazon
Alachlor
Estimated
Number
5,990,000
264,000
70,800
38,400
25,600
25,100
19,200
13,100
8,470
7,160
3,140
95% Confidence
Interval
(lower - upper)
(5,280,000 - 6,700,000)
(129,000 - 477,000)
(13,300 - 214,000)
(2,740 - 164,000)
(640 - 142,000)
(590 - 141,000)
(160-131,000)
(14-120,000)
(1 -111,000)
(1 - 109,000)
(1 - 101,000)
Estimated
Percent
57.0
2.5
0.7
0.4
0.2
0.2
0.2
0.1
0.1
0.1
<0.1
95%
Confidence
Interval
(lower - upper)
(50.3 - 63.8)
(1.2-4.5)
(0.1 -2.0)
(<0.1 - 1.6)
(<0.1 - 1.4)
(<0.1 - 1.3)
(<0.1 - 1.2)
(<0.1 - 1.1)
(<0.1 - 1.1)
(<0.1 - 1.0)
(<0.1 - 1.0)
* Registration canceled by EPA.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
• Agricultural Activity and Pesticide and Fertilizer Use -- the scope of agricultural and
non-agricultural use of pesticides and nitrogen fertilizers and the scope of
agricultural practices, that could contribute to the presence of pesticides or nitrate
in well water;
• Transport -- factors that contribute, either directly or indirectly, to the movement of
pesticides and nitrate, such as precipitation;
• Chemical Characteristics - factors that are characteristic of pesticides (K^, half life),
and factors that are characteristic of the soil or ground water in the sampled wells
(temperature, pH, conductivity); and
• Physical Characteristics of Wells -- factors that are characteristic of wells (depth, age,
construction).
EPA specified an initial set of hypotheses to focus the Phase II analysis on important scientific or
policy questions. These hypotheses are shown in Exhibit 1-3. EPA tested these hypotheses and conducted
exploratory analyses using a variety of data analysis methods that are described in Chapter 2. Chapter 3
describes the significant findings for specific variables.
Nitrate, DCPA acid metabolites, and atrazine are the three most frequently detected analytes in both
community water system wells and rural domestic wells. Both DCPA (Dacthal), the parent compound of
DCPA acid metabolites, and atrazine were registered for use in the late 1950s. The uses of the two chemicals,
however, are substantially different. DCPA is used to control annual grasses and broadleaf weeds in turf and
on ornamentals and a number of fruits and vegetables, as well as cotton, soybeans, and field beans. It has both
agricultural and non-agricultural uses. Atrazine, in contrast, is most frequently used for agricultural
applications, particularly on corn, sorghum, sugarcane, and for weed control on non-cropped and fallow land.
In general, pesticides, including atrazine and DCPA acid metabolites, were detected more frequently in CWS
wells than in rural domestic wells, which is contrary to the expectations upon which the Survey was designed.
The concentrations of pesticides were generally lower in CWS wells than in rural domestic wells. The high
concentrations found in rural domestic wells were substantially higher than the high concentrations found in
CWS wells, except for DCPA acid metabolites. A number of questions that arise from these results were
examined in Phase II, including whether the Survey data provide an explanation for the difference between
CWS and rural domestic well findings and whether explanations for the relatively large numbers of DCPA acid
metabolite and atrazine detections can be developed and tested. EPA also examined the degree of similarity
between the NPS results and results obtained by other studies of pesticides and nitrate in drinking water wells.
The analyses performed in Phase II are limited by the following:
• On a national scale, contamination of drinking water wells by any given pesticide is,
fortunately, relatively infrequent. The reliability with which any relationship between
detection and another factor can be estimated depends on the frequency of
detections. As discussed in detail in Chapter 2, the only individual NPS analytes with
sufficient detections for reliable statistical analysis are nitrate in CWS wells and rural
domestic wells and DCPA acid metabolites for CWS wells. Other analyses are based
on a subset of all the analytes consisting of all pesticides detected by the primary
analytic laboratories.
• The Phase II analyses identify factors that are potentially useful topics for additional
detailed research and provide information to be considered together with the results
of previous studies. The Phase II analyses are not designed to provide results that
are suitable by themselves to serve as the basis for policy development.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
Exhibit 1-3
Hypotheses Tested in the NFS Phase II Analyses
Ground-Water Sensitivity
• Analytes are detected more frequently in counties with high DRASTIC scores.
• Analytes are detected more frequently in areas where aquifers are more vulnerable to
contamination.
• Analytes are detected more frequently in vulnerable areas if the area is also cropped.
Pesticide and Nitrogen Fertilizer Use
• Analytes are detected more frequently in areas with high pesticide use.
• There is a significant difference in analyte detection by month or season.
• Analytes are detected more frequently in areas with pesticide use within one-half mile of
the well.
• Analytes are detected more frequently in areas of pesticide use if the area is also an area
of high vulnerability.
• Analytes are detected more frequently in areas where non-farm pesticides are used.
• Detection of nitrate is a good indicator of pesticide contamination.
• Crops or crop types are good indicators of pesticide or nitrate contamination.
• Point sources of contamination (septic tanks, pesticide spills, pesticide and fertilizer
dealerships, and disposal sites) are good indicators of contamination.
• Analytes are detected more frequently in areas of livestock production (high acreage in
pasture, feed lots).
Transport
• Analytes are detected more frequently in wells tapping unconfined aquifers.
• Analytes are detected more frequently in areas where irrigation is used.
• Analytes are detected more frequently in areas with greater precipitation.
Chemical Characteristics
• Analytes are detected more frequently if they are mobile (low Koc values) and persistent
(long biodegradation and hydrolysis half-lives).
Analytes are detected more frequently in wells with low water temperature.
» Analytes are detected more frequently in wells with water with low pH.
• Analytes are detected more frequently in wells with water with high electrical conductivity.
Physical Characteristics
• Analytes are detected more frequently in shallow and/or not optimally constructed and/or
old wells.
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction
• The Survey's statistical design was chosen to provide national estimates of wells
containing pesticides or nitrate. The choice of wells to be sampled and the number
of water samples obtained were not designed specifically to test the hypotheses
investigated in the Phase II analyses. The Phase II analysis was unable to .test or
control for many confounding factors that are known to influence pesticide
occurrence in drinking water wells. What is understood about the mechanisms
involved in well water contamination suggests that the process is quite complex,
involving many factors, including some factors that were not measured by the NFS.
In general, the NFS collected data on many of the most important factors generally
considered to affect ground-water and well water contamination. Very little data is
available on other factors, such as soil types and characteristics in areas of pesticide
application near tested wells, recharge areas of tested wells, and well location up-
gradient or down-gradient of pesticide or fertilizer use areas.
• The NFS is a one-time "snapshot" of the nation's community water system wells and
rural domestic drinking water wells. The presence and concentrations of pesticides
and nitrate in drinking water wells are thought to vary temporally and spatially.
Temporal variation involves a complex interaction among rainfall patterns, crop life
cycles, pesticide and fertilizer application patterns, and local soil characteristics. This
pattern may vary regionally. As a one-time sample, however, the NFS cannot reliably
estimate any temporal trends. In addition, because the NFS is a "snapshot," the
number of wells containing pesticides and nitrate may differ over time from the
number estimated from NFS results. As a national sample, too few wells were
sampled in each region of the country to perform regional studies.
• The Phase II analysis does not address causation. A statistical association between
two variables by itself does not imply that one causes the other. Unambiguous
attribution of causality generally requires not just a statistical association, but also
a plausible causal mechanism. An observational study, such as the NFS, cannot
alone provide proof of causality.
• The Survey followed stringent quality assurance/quality control procedures in the
collection of questionnaire data as well as in chemical analyses. In a few cases, the
quality of the data for factors that were measured by the NFS questionnaires may
have been reduced as a result of measurement error, respondent error, or for other
reasons. The QA/QC results for the chemical analyses presented in Appendix D to
the Phase I report indicate that those data meet all QA/QC standards. When
pertinent, this Report discusses the quality of the data used in the analyses.
The balance of this Phase II Report consists of four chapters:
• Chapter Two describes the data sources and statistical approaches used to investigate
the extent to which analyte detections are associated with factors such as ground-
water vulnerability, pesticide or fertilizer use, and a number of other characteristics;
• Chapter Three presents the results of statistical analyses organized by data sources.
Appendix A reports additional results that did not fully satisfy statistical screening
criteria but that may be useful to help identify trends or to stimulate additional
research;
• Chapter Four summarizes the results of the analyses reported in Chapter 3 in the
context of the four key factors thought to affect contamination, and compares the
findings to those of previous studies as well as other selected surveys. Chapter Four
also presents the results for multivariate regression analyses, estimates of
concentration distributions for the most frequently detected analytes, and exposure
estimates for those analytes; and
National Pesticide Survey: Phase II Report
-------
Chapter One: Introduction 8
Chapter Five presents the major conclusions of Phase II and makes several
recommendations for future environmental monitoring work.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach
This chapter describes the data sources used in the Phase II analysis and the statistical techniques used
to perform the analyses.
2.1 Data Sources
The Phase II analysis used several types and sources of data, including pesticide and nitrate detections;
measures of ground-water vulnerability and pesticide use developed during the stratification process that
preceded selection of the wells to be sampled; information from questionnaires administered at the time of
well sampling; and data on precipitation and drought, pesticide use, and potential sources of nitrate such as
fertilizer use and animal husbandry. As Exhibit 2-1 indicates, the scope of the data extended from broad
regional or state-wide aggregation to a narrow focus on individual wells or areas in close proximity to wells.
Exhibit 2-2 provides the sources for each category of data.
Exhibit 2-1
Type and Scope of Data Used in Phase II Analysis
Analyte Detections
Precipitation
Ground-Water Vulnerability
Pesticide Use
Fertilizer Use
Farming Activities
Non-Farming Activities
Well Construction/Conditions
Well Water Characteristics
Pesticide Characteristics
EPA
Region
•
•
State
•
•
County
•
•
•
•
•
Sub-
County
•
•
•
•
•
Well or
Well
Area
•
•
•
•
•
•
•
•
•
Chemical
•
Well or well area data were all generated by the NFS. Analyte detection data came from the chemical
analysis of well water samples; well water characteristics data (temperature, pH, electrical conductivity) were
collected during sampling; all other well-level data were collected from Survey questionnaires. Sub-county
level data, corresponding to areas larger than the immediate vicinity of the well but smaller than a county,
includes data from weather stations, data used in defining "cropped and vulnerable" areas, and questionnaire
data. County-level data includes data used for first-stage stratification, and data from the Census of
Agriculture. In addition, as described in Sections 2.1.4 and 2.1.5 of this chapter, as well as in Chapter 3,
county-level measures were developed from regional and state-level data. Regional and state-level data
included estimates of pesticide use and nitrogen fertilizer sales.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 10
«*
CM
.t±
n
CO
s
CO
CD
CO
CO
CO
CO
CO
0)
B
o
w
3*
Ml
0. Q
DC 1 g
$
III
c
ri § 3
« 8 -e
2
S
z
S *
2 E J?
T* C
0<
« °*
II
«P=TjS§
75
||
O
9
Analyte Detections
•
1 Precipitation and
Drought
•
•
1 Ground-Water
Vulnerability
.
•
I! Pesticide Use
•
•
II Fertilizer Use
•
•
9
1
o>
u.
1 Non-Farming
Activities
•
[Well Construction/
Conditions
9
*
[Well Water
Characteristics
•
1 Pesticide
Characteristics
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 11
2.1.1 Laboratory Analytical Data
An analytical results data file was created for the Phase II analysis for individual analytes and groups
of analytes. Analytes with less than five detections were included in analyses of selected groups of chemicals
but individually could not be used for statistical analysis.1
Background on the analytical results data file is provided in the NFS Phase I Report. The procedures
followed by the NFS for selecting the chemical analytic procedures, establishing minimum quantification limits
and minimum reporting limits, carrying out confirmation of positive detections, and performing quality control
are described in Chapters 4 and 5 of the Phase I Report. Appendix E of that report lists each Survey analyte,
along with the minimum quantification limit for that analyte. Appendix E also provides data on the numbers
of analyses deleted due to failure to meet quality control guidelines; results for laboratory control standards
fortified at 10 times the MQLs; a comparison of the results for laboratory control standards and for spiked
field samples; and a false negative rate for each analyte.
2.1.2 Ground-Water Vulnerability Characteristics
To analyze the influence of ground-water vulnerability characteristics, NPS collected data using the
Agricultural DRASTIC scoring system and questionnaires that obtained information on land features and
aquifer type.
DRASTIC scoring. During the implementation of the statistical design of the Survey, EPA obtained
data on hydrogeological conditions throughout the nation at the county level and for sub-county areas of 90
counties. Although these data were obtained initially to cany out stratification, they were also used in Phase
II of the Survey to test the success of stratification in identifying vulnerable areas as well as to test other
hypotheses about ground-water vulnerability and contamination.
Hydrogeological data from sources such as State geologic survey reports and United States
Department of Agriculture Soil Conservation Service soil surveys were used to derive county-level scores for
each of the 3,137 counties or county equivalents in the U.S. Prior to sampling, each county was assigned to
the category of relatively high, moderate, or low ground-water vulnerability (first-stage stratification) on the
basis of its Agricultural DRASTIC classification. This was done for each county by characterizing the seven
hydrogeologic parameters of DRASTIC: depth to water, recharge (net), aquifer media, soil media, topography
(slope), impact of vadose (unsaturated) zone, and conductivity (hydraulic) of the aquifer. These county-level
DRASTIC first-stage data do not account for site-specific detail such as topography, well depths, and soils.
A dataset was created that identified the seven hydrogeologic subscores and the total weighted score for each
county.
In order to obtain ground-water vulnerability information at a more refined spatial scale for the
sampling of rural domestic wells, the 90 counties selected for rural domestic well sampling were divided into
sub-county regions with common hydrogeologic settings. DRASTIC scores and subscores were then derived
for these sub-county regions. Because the sub-county DRASTIC scores and subscores are more site-specific
than those at the county level, they may be more accurate in capturing the degree of ground-water vulnerability
associated with the sampled wells. Data used for sub-county scores included the sources used for the county-
level scores, well drilling logs and information on soil and aquifer media obtained from local experts and State,
county or regional sources.
1 See Section 2.2 of this chapter.
2 The DRASTIC system was developed by the National Water Well Association (NWWA) for the EPA and its scoring
is described in detail in Allen, L., T. Bennett, J. Lehr, R. Petty, and G. Hackett, DRASTIC: A Standardized System for
Evaluating Ground Water Pollution Potential Using Hydrogeologic Settings. April 1987, EPA-600/2-87-035 (hereafter,
DRASTIC: A Standardized System").
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 12
The DRASTIC scoring system is based upon a series of ratings and weights. Each DRASTIC factor
is assigned a weight between one and five based upon its importance relative to the other factors. Definitions
of the seven DRASTIC components and their relative weights for the NFS are provided in Exhibit 2-3.3 The
weights are those generally used for Agricultural (pesticide) DRASTIC. Agricultural DRASTIC is a
modification of DRASTIC that addresses the potential degradation of pesticides within soil; consequently it
weights soil media and topography factors more heavily than "normal" DRASTIC.
Exhibit 2-3
Agricultural DRASTIC Subcomponent Definitions and Weights
DRASTIC Factor and Acronym
Depth to water
Net Recharge
Aquifer media
Soil media
Topography
Impact of vadose zone
Hydraulic Conductivity
D
R
A
S
T
I
C
Weight
5
4
3
5
3
4
2
Definition
Depth to static water levels in unconfined aquifers;
to base of aquitard in confined aquifers. Effects
of artificial recharge removed. As depth
increases, the score decreases.
Natural recharge to water table or to confined
aquifer. Effects of artificial recharge removed. As
net recharge increases the score increases.
Lithology and structure of aquifer; emphasis upon
attenuation and hydraulic properties. More
porous media have higher socres than less
porous media.
Texture of the most significant soil layer; emphasis
upon attenuation and infiltration. As permeability
of soil type increases, the score increases.
Degree of slope determined from large scale
topographic maps or published soil surveys. As
the steepness of the topography (percent slope)
increases, the score decreases.
Lithology of unsaturated zone for unconfined
aquifer or material above confined aquifer;
emphasis on attenuation and hydraulic properties.
Less attenuating and more porous media have
higher scores.
Ease of ground-water flow as inferred from well
data or from lithology. As the conductivity
increases, the score increases.
The DRASTIC factors consist of continuously distributed values such as depth to water table, and sets
of classes such as aquifer media. Both types of data are categorized into ranges or groups. The categories are
evaluated for their relative impact on ground-water vulnerability and assigned a rating of one to ten.
3 Ranges and weights for Agricultural (Pesticide) DRASTIC scores are provided on pages 20 to 33 of DRASTIC: A
Standardized System. Detailed descriptions of each DRASTIC factor are provided on pages 44 to 62, and those pages should
be consulted for full definitions of the factors.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 13
• Depth to water determines the depth of material through which a contaminant must
travel before reaching the aquifer. Greater depths are typically associated with lower
potential for ground-water contamination.
• Net Recharge of an aquifer is measured as the difference between the amount of
water available to enter the aquifer (e.g., local mean annual precipitation, and local
irrigation application) and the amount of water that does not reach the aquifer (e.g.,
potential evapotranspiration, and estimated runoff).
• For Aquifer media an increase in grain size and fractures or openings within an
aquifer is associated with an increase in permeability and a decrease in attenuation
capacity (sorption, reactivity, and dispersion) for contaminants.
• The Soil media is defined as the uppermost portion of the vadose zone characterized
by significant biological activity (usually less than three feet in depth). Soil type is
a major controlling factor of the infiltration process and contaminant attenuation
through filtration, biodegradation, sorption, and volatilization. In general, the effect
of soil on pollution potential depends on the clay content, the shrinking or swelling
potential of the clay, the grain size of the soil, and the presence of organic material.
• The chance that a chemical remains on the land surface for subsequent infiltration
is partly controlled by the Topography, measured as the percent of local slope. In
general, steeper slopes signify a high run-off capacity and a lower chance of
infiltration and ground-water pollution potential.
• The vadose zone is defined as the zone above the water table and below the soil
horizon that is unsaturated or discontinuously saturated. When evaluating a confined
aquifer the Impact of the vadose zone is expanded to include saturated zones that
overlie the aquifer. The type of vadose zone media controls the attenuation of
contaminants below the soil horizon and above the water table. The vadose zone
media also controls the flow path and path length of contaminants. An increase in
grain size or fractures and openings in the vadose zone is associated with an increase
in permeability and a decrease in attenuation capacity through biodegradation,
neutralization, mechanical filtration, chemical reaction, volatilization, and dispersion.
• Hydraulic Conductivity of an aquifer is associated with interconnections of void
spaces within the aquifer, which facilitate rapid ground water and contaminant
movement. The greater the hydraulic conductivity the greater the pollution
potential. The DRASTIC scoring system reflects the close relationship between
hydraulic conductivity and aquifer media.4
The DRASTIC weights and ratings are combined in an additive model to produce the DRASTIC index
(or total DRASTIC score). In the model, the weight of each DRASTIC factor is multiplied by the selected
rating for that factor, and the products summed to form the total DRASTIC score.
The second component of the DRASTIC method, hydrogeologic settings, are defined within the
DRASTIC system as having common hydrogeologic characteristics and common vulnerability to ground-water
contamination. Settings represent areas larger than 100 acres in size. In the United States, the DRASTIC
system recognizes 111 hydrogeologic settings. The seven DRASTIC factors are evaluated for each
hydrogeologic setting identified in a local area, such as within a county, to produce a more precise ground-
water vulnerability index than a broad area DRASTIC score can provide.
4 DRASTIC: A Standardized System, p. 68.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 14
County and sub-county areas were evaluated using available data and expert opinion and DRASTIC
scores were recorded on coding sheets. The percent of each county that corresponds to each individual
DRASTIC subcomponent rating was entered on the coding forms, along with the hydrogeologic settings and
a variability index. Two initial methods of forming subcomponent scores were used. The first used the most
likely rating for each component. The second method used a weighted average of the different responses for
the component. These scores were incorporated into the DRASTIC index model to produce both a "most
likely" total score and a Veighted average" total score. A variability index (related to aquifer media and
vadose zone) was also included on the DRASTIC coding sheets. This was used to measure hydrogeologists'
opinions on the variability that may be associated with the previously determined "weighted average" total
DRASTIC score. The variability index was used to adjust the Veighted average" total DRASTIC score to
produce an adjusted weighted average. Of the three total DRASTIC scores available, this DRASTIC score,
termed the VARSCORE, was thought to be the most likely to reflect ground-water vulnerability.5
To prepare the DRASTIC scores for the Phase II analysis, data from the coding forms were entered
into a computerized database. The coding forms were reviewed for inconsistencies and adjustments were
made, where necessary, to remove the inconsistencies. This operation was performed separately for county
level DRASTIC, used at the first stage of both the CWS and rural domestic well surveys, and sub-county
DRASTIC, used at the second stage of the rural domestic well survey. Exhibit 2-4 summarizes the first and
second stage DRASTIC variables included in the Phase II analysis.
Exhibit 2-4
DRASTIC Scores Used in Phase II Analyses
DRASTIC Related Score
Total depth to water
Net recharge
Aquifer Media
Soil Media
Topography
Impact of Vadose Zone
Hydraulic Conductivity
"Weighted average" total
Adjusted "weighted average1
total
DRASTIC
Variable Name
DEPTH
RECHARGE
AQUIFER
SOIL
TOPOGRAPHY
IMPACT
CONDUCTIVITY
WGTSCORE
VARSCORE
Data Definition
Weighted average depth index.
Weighted average recharge index.
Weighted average aquifer index. Not
adjusted for index variability.
Weighted average soil index.
Weighted average topography index.
County weighted average. Not adjusted for
index variability.
Weighted average conductivity index.
DRASTIC index using the weighted average
DRASTIC ratings.
WGTSCORE adjusted for the variability index.
2.1.3 Land Features
Sampling teams completed on-site well observation forms for CWS wells and rural domestic wells,
recording information about the topography around the well, the soil texture, land uses within 300 feet of the
well, and the presence within 300 feet of the well of such features as drainage ditches, bodies of water, septic
DRASTIC: A Standardized System, pp. 75-83.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 15
systems, farmland, buildings, livestock waste storage pits, pesticide storage areas, and other wells. (Appendix
D of the NFS Phase I Report provides the full text of questionnaires.)
2.1.4 Data on Pesticides and Fertilizer Use and Agronomic Activities
Questionnaires were administered to community water system owner/operators, rural domestic well
owners/residents/farmers, and county agricultural extension agents. Data were collected on farm pesticide uses,
spills, and disposal practices; crops grown; animal husbandry practices; fertilizer applications; and irrigation
practices on the property and within one-half mile of the well. Farming-related questionnaire items were
designed to gather specific information relating to potential sources of pesticides and nitrate in well water.
The types of fanning data provided by the respondents are:
• Community water system owners/operators provided information on crops grown and
pesticides used, including pesticide storage and disposal areas on the property.
• Rural domestic well owners or the tenants residing on the property provided
information on crops grown and pesticides used; pesticide storage and disposal areas;
animal husbandry practices; nitrogen fertilizer practices; and sources of irrigation
water on the property.
• County extension agents provided information pertaining to the area within one-half
mile of the well on crops grown and pesticides used; accidental spills of pesticides;
irrigation practices; and farming management practices.
In addition to the NFS field questionnaires, three additional non-NFS databases were used to provide
data on pesticide use, nitrogen fertilizer use, crops grown, and farm animals raised.
During first-stage stratification, EPA developed pesticide use estimates for each county based on
information about crop acreages from the 1982 Census of Agriculture and on private marketing data on
pesticides provided by Doanes Marketing Research, Inc.6
The second database was compiled for EPA by Resources for the Future, Inc. of Washington, D.C.
(RFF) on the estimated use of 96 pesticides by crop for each of the 3,137 counties and county equivalents in
the U.S. This database includes county-level estimates derived from state-level data on: (1) use estimates for
42 herbicide active ingredients primarily used on farm crops; (2) use estimates for 16 insecticide active
ingredients primarily used on farm crops; (3) use estimates for 10 selected active ingredients used by urban
applicators; and (4) use estimates for 10 selected active ingredients used on golf courses.7
The third database consists of state-level nitrogen fertilizer sales data prepared at the Division of
Resources Management at West Virginia University in conjunction with the National Fertilizer and
Environmental Research Center (NFERC) at the Tennessee Valley Authority in Alabama and county-level
1987 Census of Agriculture data on field animals raised and acreage of crops grown.8
The nitrogen fertilizer sales database consists of nitrogen fertilizer sales totals for each state. States
require fertilizer dealers to report sales to state regulatory agencies. This information is submitted in the form
* Derivation of County-Level Pesticide Usage Estimates for Design of the Groundwater Pesticide Survey, EPA Office
of Pesticide Programs, September 1985, and NFS Phase I Report, p. B-4.
7 The data are described in Gianessi, L.P., and Puffer, C.A, November 1990, The Use of Herbicides in the United States.
Resources for the Future, Inc., Washington, D.C., and Gianessi, L.P., and Puffer, C.A, January 1991, Estimation of County
Pesticide Use on Golf Courses and by Urban Applicators. Resources for the Future, Inc., Washington, D.C.
8 County-Level Fertilizer Sales Data. United States Environmental Protection Agency, Policy Planning and Evaluation
(PM-221), September, 1990.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 16
of tonnage reports as inspection or tonnage fees are paid. The nitrogen sales data is also broken down into
subgroups of nitrogen fertilizer. These subgroups are ammonium nitrate, anhydrous ammonia, miscellaneous
forms, nitrogen in solution, and urea. The subgroups are mutually exclusive and exhaustive, meaning that the
total nitrogen sales measured in tons is the sum of these five component parts. Annual data are currently
available for the past six years (July 1, 1984 through June 30, 1990). A subset of the annual data
corresponding to nitrogen sales in the fall months is also available for each nitrogen group and year. The data
can be used to generate similar data for the spring season (for the nitrogen fertilizer sales data the year is
considered in two parts corresponding to fall and spring).
Two sets of nitrogen sales data are available for each year and nitrogen group. The first corresponds
to the raw data submitted by the states to NFERC at the Tennessee Valley Authority. The second consists
of estimated data using a model developed at West Virginia University. The 1987 Census of Agriculture data
on fertilizer expenditures by farmers is used to develop the county estimates. The estimates are generated by
allocating NFERC state totals, which are available for all states, among counties in proportion to reported
expenditures on all fertilizers, as reported in the 1987 Census of Agriculture.
The second part of the NFERC/EPA database consists of county-level information gathered from the
1987 Census of Agriculture. The data are organized into three categories: general crop acreage and crop and
livestock value; specific crop acreage and crop production; and animal counts.
2.1,5 Data on Non-Farming Activities Involving Pesticides
Questionnaire data were collected on non-farm pesticide uses, spills, storage, and disposal practices;
household, lawn, and garden chemicals used; and land uses on the property and within one-half mile of the
well. Additional questions were also administered on the age and depth of the well and well construction.
The same questionnaires used in collecting the farming data were used in collecting the non-farm data.
Questionnaires were administered to CWS owner/operators, rural domestic well owners/residents, and county
agricultural extension agents. Questionnaire items were designed to gather information about potential
sources of pesticide use, land features, and well construction practices that could affect the detection of the
NFS analytes. Data provided by the respondents include:
• Community water system owners/operators provided information about the
construction of the well, the well casing, including screens and grouting; the type of
aquifer (confined or unconfined) from which the well draws water; other
operating/non-operating wells within 500 feet of the sampled well; non-farm
pesticides used or disposed of on the property; and septic systems located on the
property.
• Rural domestic well owners or household tenants provided information on the
construction of the well, the well casing, including screens and grouting; other
operating/non-operating wells within 500 feet of the sampled well; non-farm
pesticides used or disposed of on the property; septic systems located on the
property, types of animals raised, and types of animal husbandry practices (feedlots,
dairy, grazing) on the property.
• County extension agents provided information pertaining to the area within one-half
mile of the well on bodies of water, hazardous chemical spills, non-agricultural
pesticide spray applications, and facilities such as military bases, golf courses,
pesticide retail outlets, and septic systems.
In addition, the RFF pesticides database also includes estimates for urban and golf course use
constructed from data from 1982 national surveys conducted by EPA's Office of Pesticide Programs. The
urban applicators survey included three groups of applicators; tree, lawn, and structural. Further information
used to construct the golf course database came from the 1980 Census of Housing and the National Golf
Foundation.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 17
2.1.6 Data on Well Water Characteristics
Well water pH, temperature, and electrical conductivity (measured as ppm total dissolved solids) were
measured during the sampling process to ensure that fresh ground water rather than water that had been
standing in the well or piping was being collected. Sampling teams recorded the initial pH, temperature, and
conductivity of well water, and remeasured each at five minute intervals up to a maximum elapsed time of 30
minutes. A data set consisting of the initial, stabilized,9 and final (after the last sample bottle was filled)
readings of well water temperature, pH, and conductivity was derived from the well records. The stabilized
and final readings of temperature, pH, and conductivity were considered the most representative of ground
water in the aquifer. The stabilized readings were used in the Phase II analysis.
2.1.7 Data on Chemical Properties of Pesticides
The fate of pesticides in soil is partially dependent on two characteristics: half-life of the pesticide in
soil and soil adsorption coefficient (organic carbon partition coefficient, or K^,). The half-life is used to
measure the time for a chemical to dissipate and/or degrade to half its initial concentration. The Koc is related
to the relative mobility of the compound in soils. A database was compiled of 64 of the 127 analytes included
in the NFS with their known soil half-life and
2.1.8 Precipitation and Drought Data
Seasonal and yearly precipitation levels may influence the leaching potential of pesticides and
fertilizers applied to soil. To analyze the influence of precipitation and drought conditions, NFS collected
precipitation records and data on the Palmer Drought Index scores for counties in which wells were sampled.
Precipitation. Precipitation information was retrieved from a database maintained by the National
Climatic Data Center (NCDC) of the National Oceanic and Atmospheric Administration (NOAA).
Precipitation data were extracted from a data set called Climatedata distributed by Earthlnfo, Inc. of Boulder,
Colorado. Climatedata is identical to the TD-3200 Summary of the Day Cooperative Observer Network
database maintained by NCDC.
NPS collected precipitation data from weather stations across the U.S., using counties as the basic
unit. Data were obtained for 405 counties for community water system wells (399 counties where sampling
occurred and 6 counties contiguous to counties where sampling occurred when the latter had no eligible
weather station within its borders) and 90 counties for rural domestic water wells. Using station descriptions
and maps from the Climatological Data Annual Summary published by NCDC, weather stations were selected
based on the following two criteria:
• precipitation information is available for the station from 1980 to 1989; and
• the station is located in a county included in NPS or contiguous to such a county if
that county has no eligible weather station.
9 Well water was defined as stabilized when: (1) two water temperatures, taken five minutes apart, were within one degree
Celsius of each other; (2) two pH readings, taken five minutes apart, were within 0.2 units of each other; and (3) two
conductivity readings (reported as TDS in ppm), taken five minutes apart, were within 1 ppm of each other.
10 Values reviewed and approved by the EPA Office of Pesticide Programs were used when available for K^ and half-life.
Sources included values prepared in 1989 for the U.S. EPA Environmental Criteria and Assessment Office as revisions to
the Superfund Public Health Evaluation Manual. 1986, and tables included in R. Don Wauchope, "Selected Values for Six
Parameters from the SCS/ARS/CES Pesticide Properties Database: A Brief Description," February 20, 1991.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 18
Once weather stations were selected, total monthly precipitation for each station was derived from available
daily precipitation data using the standard NCDC method.11 For about 50 percent of counties sampled by
NFS, multiple weather stations were available for the precipitation analysis. In these cases, the derived total
monthly precipitation from these stations were averaged to provide a single value for the county. A new set
of variables depicting short-term and long-term precipitation characteristics, based on the sampling date
(month and year), was then extracted from the monthly precipitation data set, This set of precipitation
variables was used in NFS to assess the relationships between long and short term precipitation characteristics
and the occurrences of pesticides and nitrate in NFS well water samples. The derived total monthly
precipitation is highly generalized and may not be representative of precipitation patterns at a highly localized
level such as the wellhead area.
Palmer Drought Index. Information on drought was extracted from "Drought Severity Index by
Division" maps generated by the Climate Analysis Center of NOAA The Drought Severity Index or Palmer
Drought Index (PDI) was developed by Palmer in 1965 to evaluate drought. PDI is used jointly by NOAA
and United States Department of Agriculture to depict prolonged abnormal dryness or wetness on a regional
scale. PDI is calculated for each climatic division designated by NOAA and takes into account long-term
values of evapotranspiration, recharge, runoff, and net loss of soil moisture for the specific climatic division.
Because of the autoregressive nature of PDI (i.e., inclusion of past values in its formulation), PDI does not
respond quickly to short-term moisture input and thus is generally not representative of short-term drought
or moist spell conditions. In the NFS, data on drought and moist spells were extracted from the "Drought
Severity Index by Division" maps associated with the first week of each month.
On the basis of sampled well locations and climatic division maps from the Climatological Data
Annual Summary published by the National Climatic Data Center (NCDC) of NOAA, corresponding climatic
divisions were obtained for all sampled wells. A data set consisting of drought severity categories for the
Divisions in which wells were located was then extracted from the "Drought Severity Index by Division" maps
for data analysis.
2.1.9 Data Availability
Data collected by EPA during the NFS, including chemical analysis results, QA/QC records,
stratification, and questionnaire data, are available from the Office of Pesticides Programs docket. The data
are in SAS files on tape for IBM mainframe use, together with a data dictionary and other documentation.
For access to the data, write to: U.S. EPA, Office of Pesticides Programs Docket, H7506c, 1921 Jefferson
Davis Highway, Arlington, Virginia 22202.
2.2 Statistical Approach
This section describes the procedures that were followed to analyze NFS data for this Phase II Report.
Section 2.2.1 provides an introduction to the section and outlines the procedures, first listing them and then
discussing the statistical software packages used to carry them out. Section 2.2.2 discusses the effective sample
size and effective detection size available for the analyses and constraints that these sizes imposed on the
analyses. Section 2.2.3 describes the data reduction and receding that was carried out. Finally, Section 2.2.4
discusses each of the statistical procedures in detail.
2.2.1 Introduction and Outline of Procedures
A stratified, multistage procedure with clustering and disproportionate sampling was used to select
wells for the National Pesticide Survey (the design is fully described in the Phase I Report). Consequently,
procedures that account for the complex sampling design, particularly weights corresponding to the inverse
probability of well selection, were used to analyze the Survey data.
11 NCDC computes monthly statistics if a particular month contains no more than 10 days of missing values. If this
criterion is met, monthly precipitation is calculated as the sum of all daily precipitation in that month.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 19
The Phase II analyses sought to identify associations or relationships between detections of NFS
analytes in well water and ground-water vulnerability, pesticide use, farming practices, and other factors. The
following procedures were used to analyze the NFS data:
• Cross tabulation and chi-square (X2) tests for independence;
• Exploratory analysis (graphing, means, medians and quantiles);
• Univariate logistic regression;
• Univariate linear regression;
• Analysis of Variance and t-tests;
• Multivariate regression analyses; and
• Models to estimate concentration distributions.
The Phase II analysis used Taylor linearization to provide variance estimates for parameters of the
statistical models.12 These procedures properly account for complex survey designs and incorporate
procedures for performing chi square tests for independence and estimating coefficients in linear and logistic
regression models.13
Additional statistical software that was used to analyze the NFS data included a specially written
maximum likelihood estimation procedure that was used to estimate parameters of a mixture model consisting
of a lognormal distribution with a point mass at zero. This software was used for developing concentration
distributions for nitrate and DCPA acid metabolites.
The dependent variable for the various statistical analyses reported in Chapter 3 is usually a function
of chemical detections or concentrations of nitrate or pesticides in well water samples. These analyses pertain
to detection of NFS chemicals above the NFS minimum reporting limits, i.e., nondetection does not imply
noncontamination with certainty, it implies that the chemical is not found in the sampled water at levels above
12 Shah, B.V., LaVange, L.M., Barnwell, E.G., Killinger, I.E., and Wheeless, S.C., SUDAAN: Professional Software for
SUrvey DAta ANalysis, Research Triangle Institute, Research Triangle Park, NC, March, 1989 (hereafter SUDAAN
documentation). The primary package used for analysis of NFS data was version 5.50 of SUDAAN on an IBM PC. The
SUDAAN procedures are both time consuming to set up and to run, requiring the maximum memory available on a standard
IBM PC (640k bytes memory available) to run the simplest programs. Consequently, exploratory analysis was generally
performed using version 6.04 of the SAS statistical software package on an IBM PC. SAS/STAT User's Guide, Release 6.03
Edition, SAS Institute, Inc., SAS Circle, Gary, NC, 1988 (SAS Technical Report P-200, SAS/STAT Software: CALIS and
LOGISTIC Procedures, Release 6.04, SAS Institute, Inc., SAS Circle, Gary, NC, 1990). Other analyses were performed in
SAS when they did not require that the sampling weights associated with each well be included. For instance, some analyses
were performed at the analyte level instead of the well level.
13 For the analyses performed in Phase II, the Taylor linearization method has been shown to perform at least as well
as other methods based on resampling approaches (such as jackknife, balanced repeated replicates, and Fay's method (Dippo,
C.S., Fay, R.E., and Morganstein, D.H., Computing Variances from Complex Samples with Replicate Weights. Proceedings
of the American Statistical Association, Section on Survey Research Methods, Washington, D.C., August, 1989) when
estimation procedures are fairly simple, and the range of items for which variance estimates are required is fairly limited (See
Judkins, D.R., Fay's Method for Variance Estimation. Journal of Official Statistics, Vol. 6, No. 3, pp. 223-239, Sweden, 1990).
The Taylor linearization method requires considerable computer resources for most analyses, which limits its capacity for
complex analysis. Variance estimation of loglinear models currently has not been developed for the SUDAAN package, which
limits the scope of categorical data analysis to chi square tests for independence and logistic regression models. Taylor
linearization also cannot be used to provide reliable variance estimates for nonlinear parameters, such as medians.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 20
the NFS minimum reporting limits.14 In this report, detection is generally used to mean concentration above
the NFS minimum reporting limit, and occurrence is used to mean concentration above zero.
2.2.2 Samples Used for Analysis
Nitrate was the only chemical detected with sufficient frequency in the NFS that the statistical analysis
could be performed without great concern that the statistical assumptions required to perform the analyses
might not be satisfied. Analysis of NFS data was carried out using the following groups of chemicals:
• Nitrate;
• Any pesticide;
• DCPA acid metabolites;
• Atrazine;
• Triazines; and
• Herbicides.
Analytical results for the groups that contain more than one chemical pertain to all NFS analytes that
fall into those groups. For example, the triazine group consists of ametryn, atrazine, deethylated atrazine,
cyanazine, hexazinone, metribuzin, prometon, prometryn, propazine, simazine, and terbutryn. However, the
analysis focused on those chemicals within a group that were detected. For the triazine group, these are
atrazine, prometon, and simazine only (in both the CWS and rural domestic well surveys). For the herbicide
group, chemicals that were detected are:
• CWS wells: atrazine, dinoseb, DCPA acid metabolites, prometon, and simazine.
• Rural domestic wells: alachlor, atrazine, bentazon, DCPA acid metabolites, prometon, and
simazine.
Although analysis using the "any pesticide" group centers on those pesticides or pesticide degradates
that were detected, the analysis pertains to all NFS pesticides and pesticide degradates and not just to the
detected analytes. That is, the hypotheses being investigated for the groups of analytes pertain, respectively,
to all NFS triazines, all NFS herbicides, and all NFS pesticides and pesticide degradates, and not only to the
detected analytes in each of these groups.
Chi-square tests for independence and logistic regression modeling were performed using detections
of pesticides or nitrate as the dependent variable. (All analyte groups listed above were analyzed but the only
two groups that provided acceptable results were any pesticide and nitrate.) Although the chi-square
procedures have no dependent variable per se, the independence hypothesis concerned detections of a pesticide
or nitrate and a categorical (usually binary) response variable that could possibly explain detections. The
descriptive significance level resulting from a 2X2 chi-square analysis is identical to the value obtained using
univariate logistic regression with categorical variables. Detections thus may be thought of as the dependent
variable in the chi-square tests of independence. However, no regression coefficient indicating the amount
of influence of the independent variable on the dependent variable is produced by the X2 test.
Linear regression models were used to analyze nitrate concentrations. A logarithmic transformation
of the nitrate concentrations was necessary for all linear regression models where nitrate concentrations are
the dependent variable. For all regression models presented in Chapter 3 the results are valid only for within
the range of the data. When nitrate concentrations are the dependent variable the range of the data is from
0.15 mg/L (i.e., the minimum reporting limit for nitrate in the NFS) to the highest detected concentration.
14 Tables of the NFS minimum reporting limits can be found in the NFS Phase I Report,
Appendix E.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 21
The corresponding regression models may not be appropriate outside this range. The linear regression models
involving nitrate concentrations are also conditional on detection above 0.15 mg/L. That is, the wells that did
not contain nitrate at or above detectable levels are not included in these analyses. Analysis of nitrate
occurrence in CWS and rural domestic wells often began by identifying a relationship with nitrate detections
through logistic regression. If a relationship was found, a linear regression model then was used to determine
if a similar relationship could be found with nitrate concentrations above 0.15 mg/L.
Whenever data were logarithmically transformed, normal probability plots were prepared for the
transformed data, and in each case normality was verified.
2.2.2.1 Effective Sample Size
The primary goal of the NFS, carried out in Phase I, was to assess the proportion of CWS and rural
domestic wells nationally that contain pesticides or pesticide degradates. In Phase II, chi-square tests for
independence, logistic regression, and linear regression were used to analyze relationships between pesticide
or nitrate detections and factors that may affect detection, such as well characteristics, pesticide use, ground-
water vulnerability, location, characteristics within the vicinity of the well, and precipitation. These statistical
methods are more complex than methods for estimating simple proportions and their standard errors such as
those in the Phase I analysis. The NPS Data Quality Objectives (DQOs), which established the precision
requirements for sampling water supply wells led to the sample allocation for the Survey. The DQOs were
adequately satisfied for Phase I of the Survey, principally because the number of pesticide detections realized
was far greater than had been anticipated. During the design of the study, EPA decided to accept the
sample sizes set by the DQOs for Phase I. Because the sample sizes were set to meet sample sizes for Phase
I, no explicit DQO's for Phase II were established. Given the time and cost constraints, the 1985 Scientific
Advisory Panel concurred, noting that it would be difficult to increase the sample sizes by a significant enough
number to affect the confidence of the interpretations. EPA therefore maintained the sample size chosen for
Phase I.16 In general, the Phase II analyses require larger sample sizes than Phase I analyses.
Appendix B of the Phase I Report discusses the effective sample size for the CWS and rural domestic
well surveys. The effective sample size is the sample size which would be required to achieve the same level
of precision in the analyses if a simple random sample had been taken (given that the proportions of pesticide
detections remain constant). The ratio of the actual sample size to the effective sample size is termed the
design effect. In the NPS, the design effect measures the extent to which stratification, clustering, and
disproportionate sampling cause the information contained in the NPS sample to differ from the information
provided in a simple random sample. The design effects for the CWS and rural domestic well surveys were
approximately 1.27 and 1.84, leading to effective sample sizes of 426 and 410 respectively. That is, the CWS
and rural domestic well samples contain the equivalent information of simple random samples with 426 and
410 sampled wells. These effective sample sizes achieved by the NPS were not always capable of achieving the
necessary number of observations for the Phase II analyses. A guideline for adequacy of the chi-square test
for independence procedures is that the equivalent information of at least five observations must be expected
in each cell of the cross tabulations to satisfy the test's assumptions. This guideline is used to meet
requirements of the chi-square approximation to the hypergeometric distribution (or normal approximation
to the binomial distribution) that underlies the testing procedure.
2.2.2.2 Effective Detection Size
Effective detection size is related to effective sample size. That is, the effective sample consists of
pesticide detections and non-detections. The proportion of rural domestic wells with pesticide detections is
15 Black, P.K, Johnson, L., and Lester, H., National Pesticide Survey Data Quality Objectives: Evaluation and Results,
Proceedings of the Fourth Annual Ecological Quality Assurance Workshop, Cincinnati, Ohio, February 1991.
16 NPS Phase I Report; Appendix A, p. A-ll.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 22
4.2% (see the Phase I Report). This corresponds to an effective detection size of 17 (to the nearest integer).
Effective detection sizes for the full design for the chemical groups analyzed are provided in Exhibit 2-5.
Exhibit 2-5
Effective Detection Sizes for Groups of Chemicals in the NFS
Community Water System and Rural Domestic Well Surveys
Chemical Group
CWS:
Nitrate
Any Pesticide
Herbicides
DCPA acid metabolites
Triazines
Atrazine
Rural Domestic:
Nitrate
Any Pesticide
Herbicides
DCPA acid metabolites
Triazines
Atrazine
Estimated National
Proportion of Wells
With Detections
51.9%
10.4%
9.5%
6.4%
3.2%
1.7%
56.9%
4.2%
3.6%
2.5%
1.1%
0.7%
Effective
Detection
Size
220.6
44.3
40.5
27.0
13.4
7.0
232.8
17.4
14.9
10.3
4.4
2.7
As Exhibit 2-5 demonstrates, a sufficient number of effective detections to satisfy the assumptions or
guidelines required for the statistical analyses was achieved for only a few groups of analytes. These groups,
for the CWS well survey, are nitrate, any pesticide, herbicides, and sometimes DCPA acid metabolites; for the
rural domestic well survey they are nitrate, and perhaps any pesticide. Detected chemicals in the any pesticide
group consist mainly of herbicides, and detected chemicals in the herbicides group consist mainly of DCPA
acid metabolites and atrazine. Statistical analyses for any pesticide are likely to produce more powerful results
than those for herbicides, and the direction of any effect is likely to be the same. Consequently, most results
for pesticides are for any pesticide rather than for a smaller group of pesticides. There are sufficient nitrate
detections for reasonable analysis of nitrate detections and nitrate concentrations above 0.15 mg/L.
To the extent that data are missing in a particular chemical group, the effective detection sizes
available for analysis are smaller than those given. This occurred, however, in very few cases. Assuming the
relative proportions in a cross tabulation remain constant, then as the sample size for a chi-square test of
independence decreases, the potential for realizing an expected value of at least five observations in each cell
also decreases.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 23
2.2.2.3 Statistical Quality Control and Interpretation
For analyses reported in the succeeding chapters the sample sizes have been judged to be sufficient
to satisfy statistical assumptions and to adequately control the chance of not rejecting hypotheses when, in fact,
they should be rejected. Analysts not familiar with using survey data from complex sample designs should note
that in order to satisfy the statistical criteria for specific procedures they must incorporate the survey design,
including selection probabilities, to correctly estimate standard errors and descriptive significance levels (p-
values). As discussed in Sections 2.2.2 and 2.2.3 effective sample and detection sizes were calculated based
on the Survey design for all analyte groups and questionnaire items used in the analyses. The effective sample
sizes and effective detection sizes were used in NFS as an efficient screening criteria to judge analyses with
respect to statistical assumptions and Type II error concerns. Analyses reported in the succeeding chapters
are judged to adequately satisfy the statistical assumptions for X2, logistic regression, and linear regression
procedures. In addition, each result has been judged to adequately control for Type II error (i.e., the
probability of not rejecting the hypothesis when it should, in fact, be rejected) or, equivalently, the power of
the test result (i.e., the probability of rejecting the hypothesis when it should be rejected).
The power of each test is related to the amount of information contained in the data. For example,
analyses of NFS data are measured by the effective sample size and the effective detection size (or effective
number of observations in categories under investigation). Two distinct analyses may yield equal p-values
indicating significance of a hypothesis, but one result may be stronger than the other because of the amount
of information contained in the data. The relative importance of each analysis may be measured by the power
of the test, but in practice the power is often very difficult to calculate. Consequently, guidelines for adequacy
of the procedures are often used to control for the effect of power. For example, for X2 tests for
independence of attributes an expected value of at least five observations per category is often required.
Exhibit 2-5 shows the effective detection sizes for specific analyte groups. It is important to
understand that effective sample sizes will vary with each specific test that is conducted. This occurs in the
NFS for two closely related reasons. First, the amount of missing data for each analysis varies. Missing data
is most often a result of "don't know" responses to questionnaire items and incomplete national data from
external databases. Second, effective sample sizes are smaller for response to questionnaire items that are
asked only when a prior question has a specific response (i.e., due to a skip pattern). Although the effective
sample and detection sizes may vary for each analysis, results are presented in Chapters 3 and 4 only when
Type II error is judged to be adequately controlled and statistical assumptions have been adequately satisfied.
The following examples help to illustrate the points raised in this section:
CWS Local Area Questionnaire items 12e (pertaining to the presence of a lined drainage
ditch within one half mile of the sampled well) and 14b (pertaining to use of flood irrigation
within one half mile of the sampled well) were analyzed for associations with nitrate
detections. Both X2 analyses yielded p-values of 0.019. However, the result for item 12e is
stronger than the result for item 14b because the effective sample sizes for these two analyses
differ dramatically. The effective sample size for the drainage ditch analysis is approximately
369 and for the flood irrigation analysis is approximately 99. Answers to the flood irrigation
question occur as the result of a skip pattern and are obtained only if the respondents use
irrigation. Although both analyses satisfy the statistical assumptions required for the testing
procedure, the power of the test for the drainage ditch item is greater than the power of the
test for the flood irrigation item.
The rural domestic well Local Area Questionnaire item llj pertaining to the presence of golf
courses within one half mile of the sampled well was analyzed for an association with "any
pesticide." A X2 analysis of the test of independence of the two attributes (pesticide
detection and presence of golf course) revealed a p-value of 0.002, however, the expected
number of effective observations in the category corresponding to detection and presence of
golf course was less than 1. Consequently, the statistical assumptions for the test are not
satisfied and the test has low power.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 24
2.2.3 Data Reduction and Receding
At the beginning of the Phase II analysis, data reduction and receding were performed to set up
datasets that could be more easily manipulated for different series of analyses and that were more appropriate
for some hypotheses of interest. Responses of "Don't know" to questionnaire items were excluded from the
analyses. (EPA also analyzed the "don't know" responses as a separate category. See footnote 17 in this
section.) Responses of "Other" to questionnaire items, where respondents specified a text response, were
appropriately categorized. Some questionnaire items were combined because those questions were highly
related, and some items were receded to test an associated hypothesis of interest. Finally, external databases
were used to create variables that were relevant to some of the hypotheses tested.
Imputation procedures were performed for missing questionnaire and analyte data in Phase I to
produce national population estimates. These imputed values have been included in the Phase II analysis for
consistency with the Phase I results. In addition, many of the questionnaire items contain "Don't know" as
possible responses. "Don't know" responses were excluded from Phase II analyses. Imputation for these
responses was not performed. Imputing values for the "Don't know" responses would require an assumption
that the "Don't know" responses were random with respect to the population distribution. While this
assumption is hard to justify for the "Don't know" responses, which are legitimate responses, they generally
account for only a small percentage of responses and are therefore unlikely to affect the findings. Imputation
for "Don't know" responses would also increase the sample size available for many of the statistical procedures,
resulting in an artificial increase in statistical power. The potential for introducing bias due to any non-
random effects could result in substantiation of hypotheses purely on the basis of "Don't know" responses.17
For the reported findings with the highest "Don't Know" rates, a sensitivity analysis indicated the results were
not affected.
Data from the external databases, such as the RFF pesticides database, the NFERC nitrogen database,
precipitation data, and drought data, were county-specific rather than well specific. Analysis of these data was
performed by first associating wells with their counties of origin, and then matching the counties with-county
level information from the external data. In general this leads to averaging of these data. For example, the
nitrogen database contains information on crops grown and animals raised in each county. The fact that,
according to that database, crops are grown in a county in which an NPS sampled well is located does not
mean that the particular well is located close to cropland. Cropping values are averaged across the county to
be able to associate cropping data with wells, but the averaging that this causes could affect conclusions
resulting from their analysis. In addition, the RFF pesticide and NFERC nitrogen databases contain data at
the county level that are prorated from the state or EPA Regional level. This proration also leads to
averaging that can affect the quality of the conclusions reached from the analysis.
2.2.4 Description of Procedures
NPS data consisted of both categorical and continuous responses. Section 2.2.4.1 describes the
procedures used for analyses involving categorical data; Sections 2.2.4.2 through 2.2.4.6 describe the procedures
that also include continuous data.
17 An alternative approach to the "Don't know" responses is to include them as a separate category in the statistical
analysis. Generally, there are too few of these responses to particular questionnaire items to warrant inclusion in this way.
One problem with including "Don't know" responses as a separate category for analysis is that the small number of pesticide
detections already available for analysis are spread even more thinly, resulting in violation of the statistical assumptions. When
there were sufficient "Don't know" responses so that conclusions could be affected, further analysis was performed to verify
that the initial conclusions were appropriate.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 25
2.2.4.1 Analysis of Categorical Data: Cross Tabulation and
Chi-Square Tests for Independence
Much of the NFS data consist of categorical responses, including detections of pesticides and nitrate,
many questionnaire items, stratification variables, seasonal categories, and Palmer Drought Indices. Potential
associations between detections of pesticides or nitrate and other categorical data were examined by
performing chi-square tests for independence for two dimensional contingency tables.
Although univariate logistic regression models for categorical independent and dependent variables
produce the same numerical results as chi-square tests for independence, the hypotheses may be interpreted
differently. Regression models are often used to measure predictive or causal relationships. Chi-square results
can be regarded as providing information on the independence or association of variables. Loglinear modeling,
a generalization of both chi-square analysis and logistic regression modeling for categorical data, is not
generally available in SUDAAN. However, the chi-square test statistics used in the NPS analysis are based
on log odds ratios and are equivalent to a test for no interaction in the loglinear model fitted to the logarithm
of the estimated cell proportions18.
Prior to performing the chi-square tests, cross tabulations of the data were examined to verify that the
tests were appropriate. Most of the data were binary, comprised of positive or negative responses to the ques-
tionnaire items. For these data the main concern was adequate satisfaction of the statistical assumptions for
the testing procedure. For 2x2 tables the cross-tabulations were reviewed to indicate the direction of potential
associations, i.e., whether chemical detections are positively or negatively associated with the second variable.
For larger tables the cross-tabulations were reviewed to determine differences between categories.
The small number of effective detections for the pesticide groups requires a fairly even distribution
across the second variable for reasonable chi-square analysis. Often, however, the responses to the
questionnaire items also were unbalanced. For example, the effective number of pesticide detections in rural
domestic wells is 17 for the full survey (corresponding to 4.2 percent of rural domestic wells). To achieve an
expected number of observations of at least five in each cell of the contingency table requires an average of
at least 220 questionnaire responses in each response category (corresponding to approximately 121 effective
responses) of which 4.2 percent are detections. Furthermore, if the number of responses is not substantially
more than 220 in the category corresponding to the smallest number of detections, a significant difference
cannot be found. In summary, both a reasonable overall balance is required in the questionnaire responses
to satisfy the statistical assumptions and a reasonable imbalance in the questionnaire responses is required to
provide a significant result. The specifics of the requirement are different for each survey and pesticide group,
but the requirements are satisfied for very few of the questionnaire items examined. Some of the questionnaire
responses are so unbalanced that analysis of nitrate detections is not possible.
The guideline of five expected observations per cell is a conservative criterion, particularly in large
contingency tables. With respect to pesticide detections, however, the expected number in a particular cell
was often less than one for questionnaire items. Several approaches were taken to alleviate the problem.
First, if the item in question provided binary responses (e.g., positive and negative, "yes" and "no"), similar
questionnaire items were found and these items were combined using a logical "or" operation on the positive
responses, and a logical "and" operation on the negative responses, to produce a new variable.
Second, when the allowable responses of the questionnaire variable were not binary, but allowed
categorical responses with more than two possibilities, some of the groups were collapsed to form new groups.
Collapsing of the cells of the contingency tables was based on commonalities in the responses and the need
to satisfy the statistical assumptions as fully as possible. The procedure of collapsing cells has several
disadvantages: some information may be lost, the unbiased nature of the sample may be affected because this
procedure is performed post hoc, and the manner in which the categories are pooled can affect the conclusions
18 SUDAAN Professional Software for SUrvey DAta ANalysis, documentation, Research Triangle Institute.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 26
that may be drawn. However, collapsing cells is often the best alternative for addressing the problem of
insufficient data.
Third, for large contingency tables, all 2x2 subtables can be analyzed. Generally, in this approach an
adjustment to the chi-square statistic may be carried out to reflect the number of tests being performed on
the same data. Brunden (1972)19 provides an adjustment for chi-square analysis for 2xc contingency tables:
A similar adjustment is made for larger contingency tables from which several chi-square analyses have
been performed. For example, such an adjustment was necessary for analysis of temporal variables, which were
regrouped into 3x2 and 4x2 contingency tables.20
Finally, in cases when the assumptions for chi-square analysis of 2x2 tables are not satisfied, EPA
assessed the degree of assumption violation together with the supplied p-value to determine the
appropriateness of the conclusion.
Significance levels, or p-values, represent the degree to which the hypotheses are judged to be
supported by the data. The lower the p-value the more evidence there is in favor of an association between
the variables' attributes, e.g., detections and questionnaire items. Statistical tests are usually performed with
respect to a specified significance level (a-level) to determine whether a null hypothesis (in this case
independence of attributes) should be rejected or not. For chi-square tests of independence rejection of the
null hypothesis (of independence) implies that there is evidence of an association. Significance levels of 0.05
and 0.01 are most often used to determine statistical significance.
Rejection of a hypothesis at the 0.05 significance level also corresponds to a one in twenty chance of
rejecting the null hypothesis of independence when it should not be rejected. That is, a one in twenty chance
of producing a false positive result is accepted by specifying a significance level of 0.05. For every 1,000 tests
performed, 50 can be expected, by random chance alone, to provide significant results at the 0.05 level of
significance. The chi-square test for 2x2 tables is a two-tailed test (in the sense that a significant effect can
be found in one of two directions corresponding to positive and negative coefficients in the equivalent logistic
regression model). The NPS scientific hypotheses were often intended to be unidirectional only (for instance,
pesticide use was anticipated to be associated with more pesticide detections). The implication is that for
every 1,000 tests performed, 25 can be expected to yield a statistically significant result in the direction of
accepted scientific hypotheses and 25 can be expected to be in the opposite direction by random chance alone
at the 0.05 significance level. Although the number of statistical tests performed in the NPS has not been
counted, it is likely that it greatly exceeded 1,000. The cutoff of p <. 0.05 was used to determine which results
were deemed to be statistically significant. A sizeable proportion of these results had p-values much less than
19 Brunden, M.N., The Analysis of Non-independent 2x2 Tables Using Rank Sums. (Biometrics, Vol. 28,1972, pp. 603-
607, reprinted in Everitt, B.S., The Analysis of Contingency Tables: Monographs on Applied Probability and Statistics.
Chapman and Hall, London, 1977.
20 EPA also considered dealing with small expected cell frequencies through use of Fisher's exact test. This procedure
uses the exact hypergeometric distribution to determine the probability of obtaining a result at least as extreme as that
provided by the data. Fisher's exact test requires counts for input, e.g., numbers of detections and non-detections, or numbers
of positive or negative responses. This appears to correspond exactly to NPS data. However, there are several obstacles to
using Fisher's exact test in the NPS. In particular, there is no recognized direct method for employing the survey weights
or design in the Fisher test calculations. For this reason Fisher's exact test is not available in SUDAAN. Fisher's test can
be performed on the unweighted raw data, or a method for using the weights and the design must be created and
implemented. The usual effect of Fisher's exact test on the descriptive significance level (p-value) is to provide more
conservative results (higher p-values).
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 27
0.05. This directly reduces the overall expected rate of associations deemed to be significant due to chance
to below 50 per 1,000 tests. Comparatively few results are statistically significant in the sense used here.
Furthermore, nearly as many are in the opposite direction to the scientific hypothesis as are in the expected
direction. Consequently, apparently significant results should be regarded with some caution. Results that
are substantiated by analysis of similar variables provide the most likely indicators of pesticide or nitrate detec-
tions. To the extent that the hypotheses tested reflect a summary of data available before completion of NFS,
the results that support the hypotheses are stronger than those that contradict the tested hypotheses, even
when the p-values are equal.
Another concern for significance testing is that the p-value is inversely related to the amount of data.
For example, under simple random sampling for a test of the difference between two means, and assuming
normality is invoked, the p-value is inversely related to the square root of the sample size. There are at least
two ways to interpret this: As the sample size increases to infinity the p-value decreases to zero irrespective
of the actual difference between means; or as the sample size increases the power to detect differences also
increases. When comparing two tests with unequal but relatively small sample sizes, the "power" argument
is probably more persuasive, since the p-value is probably not being forced to zero by a large sample size.
However, when both tests have relatively large sample sizes, the p-value could become small, implying a
significant difference, purely because of this inverse relationship. In general it is possible to select a sample
allocation that will guarantee a significant result given a difference in sample means, and the larger the sample
size the more likely that significant results are to occur. Although the NFS does not contain analyses of such
simple differences between sample means, the arguments are analogous. Essentially, the amount of
information (data) contained in the sample is inversely related to the p-value for analyses that use that data.
That is, as the amount of data increases the chance of obtaining a statistically significant result increases and
the potential for false negative results also increases. Because of the relatively large number of nitrate
detections, it is not surprising that most of the analyses that produce low p-values are for nitrate detections
rather than pesticide detections. Thus, two trends of equal statistical significance (which always would have
equal p-values) could have different strengths of association, depending on the amount of information (the
number of data points and the number of detections) included in the estimate.
A significance level, or p-value, cut-off of 0.05 was used to determine inclusion of an analysis in
Chapters 3 and 4, with results from 0.05 to 0.1 presented in Appendix A of this report. Such a broad range
of results was presented to allow reviewers to determine for themselves which results are important. A higher
cut-off would have allowed inclusion of many results of questionable significance.
Much of the discussion regarding p-values in this section also applies to the univariate logistic
regression, univariate linear regression, multivariate regression, and analysis of variance procedures discussed
in the following sections. These sections are used primarily to describe the further capabilities of the
corresponding analyses.
2.2.4.2 Exploratory Data Analysis for Continuous NFS Variables
Continuous variables available for analysis in NFS include nitrate concentrations, minimum reporting
limits, well depth and age of well from the questionnaires, DRASTIC factors, well conductivity, well water pH
and temperature, estimates of pesticide and nitrogen sales, K^, half-life, and precipitation. For most analyses
involving continuous variables, the dependent variable is a function of pesticide or nitrate detections for
logistic regression models, and nitrate concentrations for linear regression models. The remaining data are
used as independent variables. For both types of regression, analysis of the residuals is often used to provide
indications for transformations of variables, identification of outliers, and model validation. Exploratory
analyses were used prior to regression modeling to provide initial insights into problems that could arise.
Exploratory analyses also were used to provide summary statistics and relevant plots of the data. Means,
standard errors, order statistics, stem and leaf plots, box plots, and normal probability plots were examined
for most continuous variables prior to inclusion of these variables in regression analysis.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 28
2.2.4.3 Univariate Logistic Regression Modeling
Relationships between pesticide or nitrate detections and continuous NFS variables were examined
using univariate logistic regression models. Logistic regression models are models for the probability of
occurrence of a specified event, in this case pesticide or nitrate detections. The univariate logistic regression
model for the probability of detection in the NFS can be expressed as:
tn
1 -Pr (detection))
The left side of the equation is the logistic or logit transformation of the probability of detections,
and the variable X represents the independent factor being investigated. The predictive capability of the
independent factor with respect to the probability of detection is determined through the descriptive sig-
nificance level (p-value) of the ft coefficient. The lower the p-value the more evidence there is to suggest that
the independent variable is a good predictor of the probability of detection. The intercept term is not analyzed
for statistical significance.
The logistic regression analyses were performed in SUDAAN in order to account for the survey design
and the survey weights when estimating the standard errors of the 6 coefficients. Residual analysis was used
to identify outliers. When obvious outliers were identified, the logistic procedure was repeated on the reduced
data set. If removal of the outliers did not result in substantially different results the model for the full data
is presented. Transformations of the continuous independent variable used in the univariate models sometimes
were performed for consistency with use of that variable in the linear regression procedures. Typically,
logarithmic transformations were used. SUDAAN does not currently provide useful goodness-of-fit statistics
for logistic regression models. Consequently, it is difficult to determine if transforming an independent
variable provides a better model. As transformations of the independent variable do not alter the variability
or shape of the error distribution it is unlikely that the logarithmic transformations of independent variables
result in substantially improved models.
A further rationale for transforming the independent variable can be found by considering t-tests for
detections. Logistic regression models have been developed to analyze the NFS data because the assumed
model is that detections are caused by other factors. To the extent that detections and other factors are
related without implied causality from factors to detections, t-tests provide a potential alternative analysis.
Once the continuous variables are cast as the dependent variable the need for transformations to control
variance is apparent.
Results of univariate logistic regression analyses that are presented in Chapter 3 are provided in terms
of estimates of the intercept term, the ft coefficient, the standard error of the ft coefficient, and the p-value
associated with the ft coefficient, whenever the units of the independent variable can be reasonably interpreted.
Many of the univariate logistic regression analyses use categorical data for the independent variable (i.e., chi-
square analysis), or have transformed continuous independent variables. In both cases, if the magnitude of
the B coefficients cannot be interpreted naturally it has not been provided. For these logistic regression
models the direction of the implied effect is reported instead of the ft coefficient.
2.2.4.4 Univariate Linear Regression Modeling
Relationships between nitrate concentrations and continuous variables were examined using univariate
linear regression models. The univariate linear regression model for nitrate concentration can be expressed
as:
In (Nitrate Concentrations) = a + px
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 29
The function of nitrate concentration most often employed involves a logarithmic transformation of
the data. The variable X represents the independent factor being investigated. The predictive capability of
the independent variable with respect to the nitrate concentrations is determined through the descriptive
significance level (p-value) of the B coefficient. The lower the p-value the more evidence there is to suggest
that the independent variable is a good predictor of nitrate concentrations. The intercept term is not analyzed
for statistical significance.
The linear regression model for nitrate concentrations may be regarded as a conditional analysis,
because only wells that recorded detections are included in the analysis. That is, the analysis is conditional
on detection of nitrate above the minimum reporting limit of 0.15 mg/L.
The linear regression analyses were performed in SUDAAN to account for the survey design and the
survey weights for estimating the standard errors of the 6 coefficients. Residual analysis was used to identify
outliers, and to verify the need for transforming the nitrate concentration data. When obvious outliers were
identified the linear regression procedure was repeated on the reduced data set. If removal of the outliers did
not result in substantially different results, the model for the full data is presented. Transformations of the
continuous independent variable used in the univariate models were based primarily on exploratory analysis.
An advantage of performing a logarithmic transformation on the independent variable as well as the dependent
variable is that it allows the intercept term to be interpreted as the percent change in nitrate concentrations
per one percent change in the independent variable. However, transformations of the independent variable
do not affect the variability or shape of the error distribution, and are unlikely to substantially alter qualitative
results.
The only goodness-of-fit statistic provided in SUDAAN for linear regression models is an R2 statistic.
This statistic is often used for comparison of nested models rather than to assess adequacy of univariate
models. However, R2 for a univariate linear regression model corresponds to the square of the correlation
between the dependent and independent variables. Consequently, it can be used to help determine the
adequacy of the models, and may provide an indication of the effect of transforming the variables.
Results of univariate linear regression analyses that are presented in Chapter 3 provide estimates of
the intercept term, the 6 coefficient, the standard error of the B coefficient, and the p-value associated with
the B coefficient, whenever the units of the independent variable can be reasonably interpreted. Some of the
univariate linear regression analyses use transformed variables. In these cases, if the magnitude of the 8 coef-
ficients cannot be interpreted naturally, it is not provided, but the direction of the implied effect is reported
instead.
2.2.4.5 Analysis of Variance and T-Tests
A limited number of analysis of variance and t-tests were performed in the analysis of the NFS data.
Analysis of variance (ANOVA) was performed in SUDAAN to examine the relationship between nitrate
concentrations and Palmer Drought Index data; unweighted t-tests were performed in SAS to examine the
relationship between K^ or half-life and chemical detections in the NFS.
The formal regression model underlying ANOVA or t-tests procedures at the well level specifies
nitrate concentrations as the dependent variable and indicators of Palmer Drought Index as the independent
variables. At the chemical level, K^ and half-life were, separately, specified as the dependent variable.
Results of the ANOVA and t-test analyses are presented in terms of an overall F-test for the
hypothesis of no difference between the means of the groups. Means and standard errors are also provided.
2.2.4.6 Multivariate Analyses
Multivariate logistic regression modeling was performed for detections of pesticides and nitrate, and
multivariate linear regression modeling was performed for nitrate concentrations. The procedures do not differ
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 30
substantially from those presented in Sections 2.2.4.3 and 2.2.4.4. The purpose of the multivariate analyses
was to examine how dependent variables were affected by a combination of independent variables.
Once the univariate analyses described in the previous sections were completed, a list of potential
variables was established for use in multivariate analyses. The best regression models were determined
primarily by examining correlation tables and through manual forward and backward elimination of variables.
The SUDAAN procedure does not provide a procedure for stepwise regression, so the p-values and R2
statistics were examined along with the correlation tables to determine which independent variables constitute
the best models.
Two categories of models were used. In both, the terms 8 and X denote vectors that represent the
multiple independent variables. The first category contains logistic regression models with the "log-odds" of
detection as the dependent variable:
In ( ) = « + p*
\-Pr(Detection)
The second category of models, for nitrate concentrations, are linear regression models with the
natural logarithm of nitrate concentrations as the dependent variable:
tn (Nitrate Concentration) = a +• p*
These models apply only to the NFS and may be invalid outside the range of the data. This concern
is particularly important for the nitrate concentration analyses, where the data are limited to concentration
values above the MRL of 0.15 mg/L. Other caveats regarding measurement error in the data should also be
considered.
Independent variables may be categorical or continuous. For categorical (binary) variables the
corresponding columns of the design matrix consist of indicators (zero or one). The model is fit to the data
using the SUDAAN package. A maximum likelihood approach is used for estimation in this procedure.
There are no statistics readily available from the SUDAAN output for logistic regression that can be used for
diagnostic checks of the estimated model. Because automatic stepwise procedures are not available in
SUDAAN the selected model was chosen based on a manual stepwise approach that consisted of backward
elimination based on t-statistics or p-values of the /3-coefficient for variables that were considered.21 This
backward elimination step was followed by repeated forward and backward elimination using the variables that
were eliminated in the initial manual backward elimination phase. The main criterion used for inclusion or
exclusion from the model was the p-value of the ^-coefficients themselves. Correlations among the model
variables and the p-value of the overall model also were considered for model selection.
2.2.4.7 Concentration Distributions
This section describes the general procedures used for estimating concentration distributions for
nitrate and DCPA acid metabolites in both rural domestic and CWS wells. Results of the concentration
distribution analysis are presented in Section 4.4.2 of this report.
Pollutants generally occur predominately at low concentrations with relatively few occurrences at high
concentrations. The lognormal distribution is often used to model such data. The NFS concentration
distributions are assumed to follow a mixture model that further accounts for the possibility that some wells
contain no contaminants. The model includes a component estimating the probability of non-occurrence, and
a component measuring concentrations given occurrence. The former component is represented by a binomial
21 Using p-values is equivalent to using t-statistics for this process because the variables used in the multivariate analysis
have 1 degree of freedom (except for the Palmer Drought Index which has 2 degrees of freedom).
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 31
probability (i.e., contaminants either occur or do not occur in wells), whereas a lognormal model is used to
characterize the latter component. Consequently, the mixture model consists of three parameters:
TT The probability of non-occurrence;
H The mean of the Cn(concentrations) given occurrence; and
a The standard deviation of the to(concentrations) given occurrence.
Given occurrence, the natural logarithm of concentrations, denoted ^(concentrations), follows a
normal distribution with mean n, and standard deviation a, if the concentrations are lognormal(/i,<7). The
model also can be presented in terms of its cumulative distribution function, F( •), as follows:
F(x\n,n,o) = it + (1-*) G(x\\i,o)
The function G( •) is the cumulative distribution of the lognormal component. Maximum likelihood
estimation procedures were used to estimate the parameters of the model for nitrate and DCPA acid
metabolites. Although commercial statistical software packages include maximum likelihood estimation
procedures, none of these packages include a procedure that can perform maximum likelihood estimation for
the types of mixture models used here and where minimum reporting limits are used. No methods or software
currently exist for estimating concentration distributions using binomial/lognormal mixture models and
censored data from complex sample surveys. A maximum likelihood estimation program was developed
specifically to estimate distributions using censored data. This program does not account for complex survey
designs or survey weights.22
A modified bootstrapping technique was used to produce weighted concentration distributions based
on the binomial/lognormal model.23 Monte Carlo simulation of the empirical population distribution was
used to generate 40 samples of size equal to the effective sample size. Survey weights are accounted for by
re-sampling from the empirical distribution and the effective sample size is used to appropriately control the
variance. Estimates of relevant parameters of the concentration distributions, including n, p, and a, were cal-
culated for each of the 40 random re-samples using the maximum likelihood procedure. The median of the
40 values of ?r, p, and a were produced as estimates of the concentration distribution parameters. Other
parameters such as quantiles and means were similarly estimated using the median of the 40 values resulting
from the maximum likelihood estimation of the re-samples.
Confidence intervals were also calculated from the results of the maximum likelihood estimation of
the 40 re-samples. The 2.5 and 97.5 quantiles of the 40 re-sample estimates were used to estimate the 95%
confidence intervals. An alternative approach to using the median and quantiles as estimates of parameters
and confidence intervals is to use the mean and the standard error under an assumption that the 40 re-sample
estimates follow a normal distribution. This assumption does not hold for many of the parameters of interest,
therefore the approach using medians and quantiles was used. When the normality assumption is reasonably
satisfied the results should not be very different.
Although the basic procedure used to estimate concentration distributions is similar to a traditional
bootstrap procedure, it differs in some important aspects. Re-sampling in the traditional bootstrap is
performed with replacement and with equal probabilities on each data point, whereas, for this exercise, re-
sampling is performed with probabilities proportional to the inverse of the survey weights, though still with
replacement. Also, traditional bootstrap methods create re-samples of size equal to the sample size, whereas
22 An alternative to such techniques is reporting of all data, including data below the MRL and, when there are data
confirmed by mass spectroscopy, data approaching the limit of detection.
23 For a description of bootstrapping techniques see, for example, Efron, B., The Jackknife, the Bootstrap and Other
Resampling Plans. SIAM, CBMS-National Science Foundation Monograph, 38, 1982.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 32
re-samples are of size equal to the effective sample size here. The effective sample size is used to try to
provide appropriate control over the variance estimates. These modifications to the traditional bootstrap seem
reasonable to the extent that this approach accurately reflects re-sampling from the nonparametric maximum
likelihood estimate of the empirical population distribution and the variance components are appropriately
controlled, but more work is required to verify the statistical properties of this method.
Bootstrap procedures often involve generation of more than 40 re-samples. This choice for the
number of re-samples was based primarily on computational and time resource considerations, and also
because of the ease afforded for calculating confidence intervals through the 2.5 and 97.5 quantiles of the 40
re-sample estimates (corresponding to the average of the lowest two and the highest two estimates,
respectively).
The basic approach consists of five steps:
• Assume a binomial/lognormal mixture model for the concentration distributions;
• Generate appropriate probabilities as the inverse of Survey weights for the re-
sampling plan;
• Generate 40 re-samples of size equal to the effective sample size;
• Use the maximum likelihood estimation procedure to estimate the parameters of the
mixture model for each of the 40 re-samples; and
• Calculate estimates and 95% confidence intervals for relevant parameters using the
40 re-sample estimates.
Due to the comparatively large number of nitrate detections in the NFS, the procedure used for
estimating nitrate concentration distributions is fairly robust. Minor modifications to the procedure, such as
using means as opposed to medians from the 40 re-sampled estimates, do not substantially alter the numerical
results.
The case for DCPA acid metabolites is not as straightforward as that for nitrate. There are sufficient
numbers of DCPA acid metabolites detections in the CWS well survey so that the current methods can be
used. However, there are not sufficient DCPA acid metabolites detections to enable the maximum likelihood
procedure to converge for all the re-samples. The maximum likelihood procedure performs poorly or does
not converge at all if there are a relatively small number of positive data points (detections), or the data points
are grouped closely together or are all in the tail of the distribution. Discussion of the re-samples that do not
converge and their effect on the analysis can be found in Chapter 4.
2.2.4.8 Method for Generating National Population Exposure and
Risk Estimates
This section provides a discussion of methods used to develop population exposure estimates from
information about the populations that drink CWS or rural domestic well water.
Survey weights are employed in the modified-bootstrap procedure to estimate CWS and rural domestic
well concentration distributions. Two possible methods can be used to modify that basic approach to arrive
at population estimates. The first, and most simple, is to assume a constant average number of people served
by each well nationally, and to make the appropriate transformation of the estimated well concentration
distribution to generate a "people" concentration distribution. This approach can provide biased estimates of
population exposure to the extent that the number of people associated with a well is related to pesticide or
nitrate detections or concentrations.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 33
The second approach is to use available population data to create "people weights" from the Survey
well weights. This method explicitly accounts for the number of people associated with each well, and provides
more precise population exposure estimates when the number of people associated with a well is a good
predictor of analyte detections or concentrations. The population of the household supplied by the well was
addressed specifically in the Main Questionnaire for the rural domestic well survey. A similar question was
not asked in the CWS well survey. The most appropriate surrogate in the CWS well survey is a question
concerning the size of the system in terms of its total number of wells. External data from a report of a 1986
national survey of community well systems conducted by Research Triangle Institute24 provides sufficient
information to estimate an average value for the CWS population served given the number of wells in the
system. Once "people weights" are developed, the modified bootstrap and maximum likelihood estimation
procedures can be employed, as they were used for well concentration distributions.
Although population data are available from the rural domestic well Main Questionnaire, they are
not in a form that is entirely conducive for use in assessment of population exposure. Two relevant questions
pertain to the number of people in the household served by the well (wells were identified through telephone
interviews directed at households in the rural domestic well survey), and whether or not the well water is used
for drinking. Other issues concerning the extent of use, the number of people who drink the well water, the
number of households served by the well, and the number of wells that serve the household, were not
addressed. If the well is used to provide drinking water then all members of the household are assumed to
drink the water, and exactly one well is assumed per household. The extent to which these assumptions are
violated is not likely to significantly alter conclusions resulting from the rural domestic well exposure analysis.
Assumptions related to developing "people weights" for the CWS well survey are not as
straightforward. The information available to perform this task consists of the following three items:
• The number of operating wells in the CWS represented by the sampled well (from
question 9 in the Team Leader Introduction Questionnaire);
• The CWS survey well weights; and
• Information on the median number of wells associated with systems that serve certain
population categories from EPA's 1986 Survey of Community Water Systems.
The first two items are used initially to create "system weights" by dividing the well weight by the
number of operating wells per system to avoid double counting. This adjustment ensures that population
estimates are not biased upward.25
EPA's 1986 Survey provides an estimated range of the number of wells per system for 12 population
categories ranging from 25-100 people to over 1,000,000 people.26 The average number of wells in systems
serving those categories ranged from 1.5 wells per system for the first category (serving 25-100 people), to 204
wells per system for the largest category (serving over 1,000,000 people). Two regression lines were fit to these
data, one for the low value in each population category, and one for the high end of each population category.
Logarithmic transformations were necessary to provide a good fit. For each CWS well sampled in the NPS,
the number of wells in the system, according to the Team Leader Introduction Questionnaire item, was used
to predict the population by using each of the two estimated regression models. These regression models
provide two extreme estimates for the population served by the system. A geometric mean was calculated from
24 Final Descriptive Summary. 1986 Survey of Community Water Systems. Report number RTI/7805/02-02F, Research
Triangle Institute, October 1987.
25 The system weights lead to an estimate of 35,800 community water systems nationally that provide ground water. This
is similar to the estimate of 38,300 provided in the Phase I Report.
26 The full list of population categories consists of: 25-100; 101-500; 501-1,000; 1,000-3,300; 3,301-10,000; 10,001-25,000;
25,001-50,000; 50,001-75,000; 75,001-100,000; 100,001-500,000; 500,001-1,000,000; and over 1,000,000.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 34
the two extreme values. This value served as the number of people served by the system of the sampled well,
and was multiplied by the "system weights" to arrive at population weights. The "people weights" reflect the
estimated number of people served by a CWS well and assume that each well serves only those people without
any mixing of water from other wells in the system or from other surface water sources in the system. For
rural domestic wells, each well serves a given number of people, estimated for each well from the Household
Questionnaire. For CWS wells, each well independently serves a given number of people, estimated from the
1986 RTI Survey and the Team Leader Introduction Questionnaire. The only required information on other
wells was the total number of wells in the system which was needed to estimate the total number of individuals
served by the system, using RTI's data. Estimating the total number of individuals in the system makes it
possible to estimate the mini-population of each well. This was regarded as the most reasonable approach
to assessing CWS exposure, given that information on the quality of water in other wells in the system or other
water sources in the system were unavailable.
For the purposes of the risk analysis the infant population is defined as children under one year of
age. Infant data are available in the form of estimates of the average number of children per household by
household population size. This data comes from the the Bureau of the Census "Current Population Reports"
of the 1990 Census of Household and Family Characteristics. The data is reported in categories of number
of people living in a household. For example households in which two people live comprise 55,212,000 people,
206,000 of which are infants. The overall population living in households is 235,499,000, of which 3,443,000
are infants. This corresponds to a population with a proportion of 1.46% infants. This proportion could be
used to adjust the concentration distributions developed from the "people weights" to arrive at estimates for
the infant population.
To generate infant population exposure estimates relevant quantile estimates from the estimated
nitrate concentration distributions based on "people weights" were multiplied by the infant proportion factor
of 1.46%
Method for Calculating Drinking Water Exposures
Adult exposures to nitrate and DCPA in drinking water obtained from NFS rural domestic wells were
calculated using the following equation:
ADE = (C x IR x ED)/(BW x AT)
where
ADE = lifetime average daily exposure (mg/kg/day)
C = concentration of the analyte (mg/L)
IR = intake rate
ED = exposure duration
BW = body weight
AT = averaging time
For nitrate and DCPA, which have not been associated with the potential development of cancer
Averaging Time (AT) in the above equation equals the exposure duration. This equation yields Average Daily
Exposures (ADE). Adult exposures were calculated for DCPA; however, infant exposures were calculated for
nitrate, since infants are especially sensitive to the effects of nitrate exposure.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 35
The following assumptions were made in the development of drinking water exposure estimates:
• Adults drink 2 liters of water per day and infants drink 0.64 liters of water per day;
• Values for the exposure duration parameter in the LADE/ADE equation were
assumed to be 70 years for adults and 1 year for infants;
• Values for the body weight parameter in the LADE/ADE equation were assumed to
be 70 kg for adults and 4 kg for infants; and
• Values for the lifetime (averaging time) parameter in the LADE/ADE equation were
assumed to be 70 years for adults and 1 year for infants.
The values used for the above terms (intake rate, exposure duration, body weight, and lifetime) are
EPA's standard assumptions used when determining health-based limits for drinking water.
Drinking water from rural domestic wells and community water systems is assumed to undergo no
treatment prior to consumption. That is, treatment efficacy is assumed to be zero; the concentration of an
analyte in drinking water at the tap is assumed to be the same as that detected at the wellhead; and, no loss
of analyte due to volatilization or chemical degradation (e.g., hydrolysis) from the wellhead to the tap is
assumed. These are reasonable assumptions given the information collected in the questionnaires regarding
water treatment and the chemical properties of analytes used in the analysis.
Method for Calculating Noncancer Risks for Nitrate and DCPA Acid
Metabolites
Noncancer risks were calculated for the analytes DCPA acid metabolites and nitrate using the
following equation:
HI = ADE/RfD
where
HI = the Hazard Index
ADE = average daily exposure (mg/kg/day), as defined above
RfD = the oral reference dose for chronic exposures (mg/kg/day)
The RfD indicates a level below which adverse, noncancer effects are not expected to occur over a
lifetime exposure to the chemical. The Hazard Index (HI) is used to measure noncancer risks; when HI >
1, a risk of developing adverse, noncancer effects may exist.
The RfD values for nitrate (1.6 x 10° mg/kg/day) and DCPA acid metabolites (5.0 x 10"1 mg/kg/day)
were obtained from: (1) "Health Effects Assessment Summary Tables" (HEAST) (USEPA 1990); (2) "Health
Advisories" (HA), (USEPA 1987 and 1988); (3) Integrated Risk Information System (IRIS 1991); and (4)
EPA's Office of Drinking Water "Reference Dose Cover Sheet for Nitrate," 1990. Note that the RfD value
used for DCPA acid metabolites is actually the RfD for the parent compound DCPA No RfD is available
for the acid metabolites; therefore, it was assumed that the RfD for these compounds would be similar to that
for the parent compound. A high confidence rating is given for this RfD since there are several studies that
support the data used in its derivation (IRIS 1991).
The chronic oral RfD for DCPA of 0.5 mg/kg/day is based upon kidney and adrenal gland effects in
long-term laboratory studies with rats and dogs. A medium confidence rating is given for this RfD because
the rat bioassays are flawed to some extent (IRIS 1988).
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 36
The chronic oral RfD for nitrate of 1.6 mg/kg/day is based upon development of methemoglobinemia
in infants (0-3 months old).
Method for Calculating Any Pesticide Exposure
Estimates of numbers of wells and percentages of wells containing at least one pesticide above health-
based levels and of populations exposed to at least one pesticide above health-based levels were not based on
concentration models, but on estimates of the number of sampled wells containing pesticides and estimates
of the populations served. The estimates and confidence intervals were generated by application of the sample
design through the sample weights to appropriate sample statistics. The estimates are the weighted sum of
the sample statistics, calculated using the same procedures as were followed in preparing the similar estimates
in the Phase I Report.27
Limitations of the Basic Approach for Generating Concentration Distributions
Development of concentration distributions is limited by two factors: the NFS data and the limitations
of the statistical method.
First, the NFS was not designed for the purpose of generating concentration distributions for use in
exposure and risk assessment. Although concentration distributions can be developed for wells, it is not clear
how to incorporate information on populations that drink the well water. Neither the CWS or rural domestic
well surveys were designed to obtain direct population information to perform this task, though the rural
domestic well survey does provide some useful information (i.e., the number of people in the household served
by the sampled well and whether or not the sampled well provides water used for drinking). Assumptions and
approximations, many of them standard EPA procedure, had to be made to overcome the inadequacy of the
NFS data for assessing population exposure and risk estimates.
In addition, NPS well water samples were analyzed by contract laboratories that maintained a
minimum reporting limit for each analyte (see the NPS Phase I Report). All positive detections above the
minimum reporting limit for an analyte also were analyzed on a second GC column. The primary
quantification for a detected analyte was retained if the secondary results were within 25% of the primary
result, otherwise the detection was reported without quantification. In one case, quantification was
considerably different in the primary and secondary analyses, which could affect the estimated concentration
distributions.
Apart from the primary analysis detections, approximately ten percent of the NPS samples were
reanalyzed by referee laboratories to determine the extent to which primary analysis resulted in non-detection,
and secondary analysis resulted in detection. Of these referee analyses, two cases recorded detections and
these were near the MRLs for these analytes. Exhibit 6 of Appendix E of the NPS Phase I Report, reports
the false negative rate for atrazine as 1 out of 108, and the rate for DCPA acid metabolites as 1 out of 133.
Second, although the modified bootstrap procedure seems reasonable, given appropriate weights, the
procedure has not been tested extensively beyond its application here. Furthermore, the maximum likelihood
procedure can produce unexpected results when the concentration data consists of too few positive
observations or the data are grouped too closely together. This is particularly true if the data are in the tail
of the distribution. There are too few DCPA acid metabolites detections for the maximum likelihood
estimation procedure to converge28 to correct estimates of the concentration models for rural domestic wells.
This problem is exacerbated by using the effective sample size, and hence effective detection size, in the re-
27 NPS Phase I Report, Appendix B, pp. B-52 to B-62.
28 The maximum likelihood estimation procedure is an iterative procedure that converges, under appropriate conditions,
to the maximum likelihood estimates. The amount of positive data and the grouping and position of those data relative to
the mode of the lognormal component of the concentration distribution determine if appropriate conditions are present.
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 37
sampling plan to appropriately control variance estimation. Therefore, a traditional bootstrap procedure was
used to estimate the DPCA acid metabolites distribution for rural domestic wells.
The necessity of using the Survey weights to develop concentration profiles is not clear. The Survey
weights were developed based on stratification variables that were anticipated to affect pesticide occurrence
in well water. The corresponding statistical assumption is that wells within strata are exchangeable with
respect to pesticide detection, but wells between strata are not. The stratification variables, however, were not
intended to delineate nitrate detections or concentration patterns, though they were expected to be associated
with changes in pesticide detections and concentrations. The stratification variables did not, however, predict
pesticide detections (see Chapters 3 and 4), and they predicted nitrate detections in unexpected ways. This
suggests that the stratification variables cannot be used to distinguish wells that have greater pesticide or
nitrate concentrations, and that weights based on the stratification variables may confound, rather than assist,
development of concentration distributions.
For consistency with other NFS analyses, the Survey weights were incorporated into the development
of concentration distributions. The design, however, was not incorporated. That is, the modified bootstrap
procedure re-samples in proportion to the inverse of the weights for the whole population to generate a re-
sample of size equal to the effective sample size. An alternative approach, which incorporates the design, is
to re-sample in proportion to the inverse of the weights within primary sampling units. Re-samples can then
be generated for each primary sampling unit of size equal to the effective sample size of that unit. The
primary sampling unit re-samples can then be combined to form a single re-sampled data set. For the rural
domestic well data, this approach can be expected to yield qualitatively similar results to the more simple
approach actually used, since design effects are similar across primary sampling units. For the CWS well
survey, this approach cannot be used since primary sampling units often consist of a single community water
system that contains one well. Other similarly motivated re-sampling plans could be used, (for example based
on the strata), but considering that the statistical method is in its infancy and that results for these types of
modified bootstrap re-sampling methods are likely to be similar, the more simple approach was adopted.
Distributions other than the lognormal distribution also are characterized by long right tails (for
example, the Gamma and Weibull distributions) and may provide better fits for the data. Use of different
basic models could substantially alter the results, including those for population exposure and risk. Finally,
the model used for this analysis is a binomial/lognormal mixture model, but this model does not account for
other predictive factors (for example, the results presented in Chapter 3 indicate that there are many factors
that influence nitrate detections and nitrate concentrations). A better model may incorporate predictive
factors in the modeling process.
As previously discussed, the models incorporate a parameter representing the probability of non-
occurrence. For these models there is an implicit assumption that some wells do not contain contaminants.
For DCPA acid metabolites this is clearly a reasonable assumption, since it does not occur naturally. The
assumption applied to nitrate, which does occur naturally, implies that nitrate does not occur in all wells.
Available data from the NFS and other monitoring surveys indicate that this is a reasonable assumption for
nitrate.
Estimation of concentration distributions for CWS wells requires assumptions about how the
occurrence of NFS analytes is correlated among multiple wells within systems. A CWS well belongs to a
system which may contain any number of wells (the largest system sampled in NFS contained 228 operating
wells at the time of sampling).
Estimating concentration distributions for CWS wells requires making an assumption about the intra-
system correlation of pesticide and nitrate occurrence in CWS wells. Assumptions of zero or complete
correlation can be made to provide bounds for the estimates. CWS wells within systems are further defined
by the clusters or groups within systems. Typically there are 4-6 wells in a cluster for large systems. Wells
within a cluster are in close proximity (< 1/2 mile) and often draw from the same recharge area. The
correlation between these wells can be expected to be greater than the correlation between wells from different
clusters. Clusters of wells also vary in their proximity and correlations between clusters probably depend on
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 38
the distance between them. NFS does not contain enough information to adequately model this correlation
structure. For the purpose of developing concentration profiles a simple assumption of zero correlation (i.e.,
independence) between CWS wells was made.
Limitations of Exposure and Risk Methodology
The exposure and risk methodology provides an estimate of the number of people potentially served
by a sampled well. A number of problems were encountered in trying to use the results of the RTI 1986
Survey of community water systems:
• The empirical relationship between categories of number of people served and the
mean number of wells in community water systems is a non-increasing function in
the RTI 1986 Survey;
• There are very few observations in the RTI 1986 Survey for large systems (only 2 in
the final category of over 1,000,000, and only 5 in the 500,001-1,000,000 category).
That is, the RTI data, used in this way, are subject to large sampling errors that are
not accounted for in the concentration distribution estimation procedure. These
categories ultimately drive the exposure numbers yet they are estimated with the least
precision;
• A population of 4,000,000 was used as the upper bound estimate of the number of
people served by the largest systems because this was the largest population served
in the NFS;
• The choice of 4,000,000 as an upper bound for the largest population category may influence
population exposure results by several hundred thousand people for each contaminated well
sampled from systems serving at least 1,000,000 people. For example, use of 2,000,000 instead
of 4,000,000 results in a decrease in the population estimate of approximately 600,000 people
for each contaminated well in systems serving over 1,000,000 people; and
• Some community water systems use both ground water and surface water. Estimating
the number of people served by wells in these systems is more difficult. The analysis
assumed that the population that drinks water from CWS wells does not also drink
from surficial sources.
The extent to which community water systems that use ground water use other sources of water makes
estimating exposure more difficult. The RTI 1986 Survey indicates that fewer than 50% of CWSs use 100%
ground water. Large systems get most of their water from other sources such as surface water. If other
sources of water are mixed with the ground water then the exposure due to ground water only is lessened.
Either fewer people than the CWS population of drinkers are exposed (because some drink ground water and
some drink surface water) or the concentration to which they are exposed due to ground water is lower than
the concentration in one well (unless other wells in the system have higher concentration levels). The RTI
1986 Survey Report does not contain sufficient information to assess adequately the percent of a CWS that
uses ground water. For the development of concentration profiles, a simple assumption of no mixing of
ground water with surface water was made.
The RTI Survey Report indicates that, on average, there are 2.8 wells per CWS that provide
predominantly ground water, whereas NFS estimates suggest that there are approximately 94,600 wells in
35,800 community water systems nationally, i.e., approximately 2.6 wells per system. This difference may be
due to the exclusion of "primarily surface water" systems from the RTI 1986 Survey data used for this analysis.
Issues related to intra-system correlation structures and mixing of both well water and ground and
surface water were treated by using standard EPA assumptions in the concentration distribution analysis.
Population served by CWSs, estimated from the regressions using RTFs 1986 Survey results, were pro-rated
National Pesticide Survey: Phase II Report
-------
Chapter Two: Data Sources and Statistical Approach 39
to the system's wells, and the wells were assumed to be independent with respect to nitrate and DCPA acid
metabolites occurrence.The wells are assumed to serve a separate mini-population. That is, each well is
assumed to serve only a particular population, without mixing of water form other wells in the system or
mixing of water from surface sources in the system. The effect of these assumptions on the results presented
in Chapter 4 is not clear, though it seems likely that concentration levels to which people are exposed due to
contaminated ground water are likely to be lower than concentration levels in individual wells that contain
comparatively high levels of NFS analytes due to the potential for mixing of water supplies (assuming that the
surface water is untainted).
For the CWS well survey, creation of infant weights that adjust for the small differences in proportions
of infants in such large populations is not likely to result in more precise estimates. For the rural domestic
well survey, however, "infant weights" could be created by adjusting the "people weights" for the proportion
of infants in each category of household size. For example, zero infants would be assigned to households with
only one person, and according to the 1990 Census data, 0.37% is the proportion of infants in households with
two people, 2.54% is the proportion of infants in households with three people, after which the proportion
of infants decreases to 1.11% for households with at least seven people. To the extent that occurrence of
contaminants in rural domestic well water is associated with the number of infants in rural domestic well
households an approach using "infant weights" may produce more accurate estimates. However, for consistency
with the approach used for CWS wells, a transformation assuming an average proportion of 1.46% infants in
the population was performed.
A further factor influencing the decision to use a constant proportion to represent infant populations
was the frame used to construct the Census data. Particularly, only data for families living in households and
single persons living alone are available to estimate the proportion of infants in the population. That data
comprises approximately 235,499,000 people, which does not represent the entire population of the United
States. A further factor that may introduce biases with respect to the exposure analysis is that the populations
that drink rural domestic or CWS well water are substantially less than the population used to estimate the
infant proportion. If the characteristics of either NFS Survey population are different from those for the
nation as a whole, the infant proportion estimate may not be appropriate.
National Pesticide Survey: Phase II Report
-------
-------
Chapter Three: Results
This chapter presents the results of analyses conducted to test hypotheses about the association of
pesticide and nitrate detections1 with a broad range of variables. Readers are expected to possess a good
understanding of statistical concepts. Results are presented for major databases generated by the NFS,
corresponding to data collected and used in DRASTIC coding, obtained from questionnaires, and obtained
during the sampling and chemical analysis processes. This chapter also contains results for data obtained by
EPA from sources outside the NFS that provide supplementary information about topics useful to the analysis,
such as long-term precipitation patterns and estimates of pesticide use. Results are presented in this chapter
for each database in order to provide a clear record of the analysis and results for subsequent researchers.
Results are presented in Chapter 4 organized by four major factors expected to affect contamination: ground-
water sensitivity; agricultural activities and pesticide and fertilizer use; transport; and physical/chemical
characteristics. Readers whose primary interest is in the evaluation rather than derivation of results may wish
to proceed directly to Chapter 4.
This chapter first presents the results of two analyses addressing the design and implementation of
the Survey:
• Analysis of the association of stratification variables with detections (Section 3.1);
and
• Analysis of temporal factors and their possible effect on Survey results (Section 3.2).
The chapter then presents the results of analyses of associations between:
• Detections and DRASTIC scores and subscores (Section 3.3);
• Detections and well characteristics or activities near the well (Section 3.4);
• Detections and pesticide applications in agricultural areas, urban areas, and golf
courses (Section 3.5);
• Detections and measures of crop production and animal husbandry (Section 3.6);
• Detections and physical/chemical characteristics of well water (temperature, pH,
electrical conductivity) at the time of sample collection (Section 3.7);
• Detections and chemical characteristics of pesticides (Section 3.8);
• Nitrate detections and pesticide detections (Section 3.9); and
• Detections and precipitation (Section 3.10).
Results are reported if they met a screening criterion of a significance level of 0.05 or below.
Descriptive significance levels (p values) are also presented for results reported in this chapter. Appendix A
reports additional results that did not satisfy this screening criterion but did have a significance level of greater
than 0.05 and less than 0.10.
1 The term pesticide detections is used to represent detection of at least one pesticide or pesticide degradate for the
remainder of this section.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 42
3.1 Stratification Variables
The NFS used a stratified design by which 1349 wells were selected for sampling using probability-
based sample selection techniques. First-stage stratification variables were a proxy for ground-water vul-
nerability measured using the Agricultural DRASTIC scoring method, and estimates of pesticide use were
determined from proprietary data held by the EPA Office of Pesticide Programs and other sources. Exhibit
3-1 shows the 12 stratification categories, which were used to classify all counties in the U.S.
Exhibit 3-1
First Stage or County-Level Strata
Ground-Water
Vulnerability
High
Moderate
Low
Pesticide Use
High
1
2
3
Moderate
4
5
6
Low
7
8
9
Uncommon
10
11
12
The NPS was conducted as two parallel surveys, one for community water system (CWS) wells and
one for rural domestic wells. For the rural domestic well survey only, ninety counties were selected at the first
stage and wells were further stratified at the second stage using an index that combined information from sub-
county DRASTIC scores and information on cropping intensities. Areas within each county were defined by
this index as "cropped and vulnerable".
A stratified design serves two purposes: to increase sampling efficiency over simple random sampling
by grouping sampling units that are expected to have similar characteristics, and to increase the number of
sampled units with desired characteristics. EPA chose a stratified design for the NPS believing that differences
in hydrogeologic conditions and pesticide use would affect the percentage of wells with detectable levels of
pesticides or nitrate. Oversampling certain strata would increase the number of sampled wells containing
pesticides or nitrate. The analysis reported in this section tested that hypothesis. The three stratification
variables, Agricultural DRASTIC score (hereafter DRASTIC), pesticide use, and the "cropped and vulnerable"
indicator, were included in separate analyses to determine their level of association with nitrate or pesticide
detections.
Exhibit 3-2 presents the proportions of nitrate and pesticide detections in each of the first stage strata
for CWS wells and rural domestic wells, while Exhibit 3-3 shows the proportions of nitrate and pesticide
detections associated with each of the first stage stratification variables.
Exhibits 3-2 and 3-3 show some surprising results, considering the design of the Survey. In particular,
compared to the moderate, low, and uncommon categories, there are smaller proportions of nitrate detections
in the high pesticide use categories for the rural domestic well survey. With the exception of stratum one,
there also are smaller proportions of nitrate detections in the high ground-water vulnerability categories in
both surveys. There is also a smaller proportion of pesticide detections in the high vulnerability category for
the rural domestic well survey. These results suggest that stratification did not function as hypothesized, at
least at the first stage. In contrast, Exhibit 3-4 suggests that the rural domestic well second stage stratification
variable may have performed as expected for pesticides, although not for nitrate.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 43
Exhibit 3-2
Proportion of Sampled Wells with Nitrate
and Pesticide Detections by First Stage Strata
Pesticide
Use Index
High
High
High
Moderate
Moderate
Moderate
Low
Low
Low
Uncommon
Uncommon
Uncommon
Total
Ground-Water
Vulnerability
Index
High
Moderate
Low
High
Moderate
Low
High
Moderate
Low
High
Moderate
Low
First
Stage
Stratum
1
2
3
4
5
6
7
8
9
10
11
12
Proportion of Sampled Wells with Detections
Nitrate
CWS
Wells
0.346
0.482
0.200
0.284
0.434
0.576
0.408
0.498
0.643
0.342
0.584
0.700
0.519
Rural
Domestic
Wells
0.397
0.179
0.374
0.356
0.602
0.748
0.536
0.634
0.740
0.498
0.681
0.579
0.569
Pesticides
CWS
Wells
0.023
0.089
0.200
0.046
0.138
0.105
0.062
0.138
0.096
0.171
0.048
0.135
0.102
Rural
Domestic
Wells
0.007
0.037
0.076
0.014
0.067
0.039
0.033
0.052
0.083
0.036
0.040
0.000
0.042
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 44
Exhibit 3-3
Proportion of Sampled Wells with Nitrate and Pesticide
Detections by First Stage Stratification Variables
Stratification
Variables
Pesticide Use
High
Moderate
Low
Uncommon
Vulnerability
High
Moderate
Low
Total
Nitrate
CWS
Wells
0.384
0.448
0.533
0.574
0.356
0.523
0.619
0.519
Rural
Domestic
Wells
0.270
0.561
0.634
0.604
0.468
0.589
0.658
0.569
Pesticides
CWS
Wells
0.100
0.105
0.103
0.100
0.093
0.092
0.121
0.102
Rural
Domestic
Wells
0.033
0.046
0.054
0.029
0.027
0.050
0.044
0.042
For the rural domestic well survey, Exhibit 3-4 shows the proportion of nitrate and pesticide detections
for the "cropped and vulnerable" indicator (i.e., sub-county areas that have high ground-water vulnerability and
greater than 25 percent of the land area used for agricultural production or moderate ground-water
vulnerability and greater than 50 percent of the land area used for agricultural production).
Exhibit 3-4
Proportion of Nitrate and Pesticide Detections for
the Rural Domestic Well Survey Second Stage Strata
•Cropped and Vulnerable"
Indicator
•Cropped and vulnerable'
Not "cropped and vulnerable"
Rural Domestic Wells
Proportion
of Nitrate
Detections
0.525
0.593
Proportion
of Pesticide
Detections
0.054
0.036
Exhibit 3-5 presents the results for tests of association between detections and stratification variables
(p-values in this exhibit are not restricted to <. 0.05). The tests confirm that there are few significant
relationships between nitrate or pesticide detections and the stratification variables. Only the differences in
proportions of nitrate detections across ground-water vulnerability categories in the CWS well survey and
across pesticide use categories in the rural domestic well survey are significant at the <. 0.05 level but these
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 45
Exhibit 3-5
Association of Nitrate and Pesticide Detections with
First and Second Stage Stratification Variables
Stratification Variable
Pesticide Use
Ground-Water Vulnerability
"Cropped and Vulnerable"
CWS Wells
Nitrate
Detections
p-value
0.087
< 0.0005
NA
Pesticide
Detections
p-value
0.994
0.734
NA
Rural Domestic Wells
Nitrate
Detections
p-value
< 0.0005
0.235
0.304
Pesticide
Detections
p-value
0.705
0.524
0.376
NA = Cropped and vulnerable stratification not performed for CWS wells.
are not in the expected direction. They correspond to a greater proportion of nitrate detections in low use
and low vulnerability strata). The differences in proportions for pesticide detections for ground-water
vulnerability and for the second stage strata in the rural domestic well survey are not statistically significant.
These results show that the NFS stratification did not effectively reduce the variance of sample
estimates. Section 3.3 presents a more detailed investigation of the association of DRASTIC scores to
pesticide and nitrate detections. Sections 3.4.3, 3.5, and 3.6 describe analyses of pesticide and fertilizer use
measures.
3.2 Temporal Allocation
The EPA Scientific Advisory Panel (SAP), in its review of the NFS pilot study and final survey design
conducted in September 1987, recommended that EPA design the study to minimize the effect of temporal
(seasonal) variability on the Survey results. Two methods are available for controlling the effect of
confounding factors such as temporal variability: stratification, and randomization. Both of these were used
to control the effect of temporal variability in the NFS. Within each first-stage stratum, the water samples
were spread throughout the term of the Survey. Drinking water wells to be sampled were randomly allocated
to available two-week time periods covering the duration of the Survey. The analysis reported in this section
evaluated whether the Survey successfully minimized confounding due to disproportionately sampling during
a single month or season.
3.2.1 Temporal Randomization
The initial NFS temporal design was defined by 28 available two-week time periods. The initial CWS
well sample design specified two-week time slots from June 1988 to August 1989. The initial rural domestic
well sample design specified time periods from August 1988 to December 1989. Within each first-stage
stratum, wells were randomly allocated to these time slots. This assured that laboratories could analyze a
maximum of 40 samples within such time slot, but would not be overloaded by an excess of samples during
any period. This temporal design does not correspond to full temporal randomization but was used to
minimize the effects of temporal variability given the time and logistical constraints on the Survey.
During implementation, the Survey deviated from the initial temporal allocation for a variety of
reasons (see the NFS Phase I report, Appendix B), but the underlying approach to temporal allocation was
retained. How well the Survey minimized the effect of temporal variability was measured by comparing actual
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 46
monthly allocation of wells to expectations under hypothetical random temporal allocation. Confidence
intervals were calculated for the number of wells that would be expected to be sampled under random
allocation, and these confidence intervals were compared to the actual NFS temporal allocation. The actual
temporal variability of the NFS is less than indicated by these results, due to the control by first-stage strata.
Sample sizes are too small to allow examination of temporal variability by stratum.
Under randomization (without regard to the planned Survey data collection period, laboratory capacity
limits, or holiday seasons) the proportion of wells allocated to a particular month is expected to be one twelfth
of the sample, but with a standard error related to the sample size as well as the proportion. The proportion
may be regarded as a Binomial proportion2. The total normalized weighted frequency is the appropriate
sample size for calculating the binomial confidence intervals required for comparison to the NFS sample. The
confidence intervals were calculated using the normal distribution approximation to the binomial distribution.
The 95% and 99% confidence intervals are presented in Exhibit 3-6.
Exhibit 3-6
Binomial Confidence Intervals for Random Temporal Allocation
Domain
cws
cws
Rural
Rural
Effective
Sample
Size
426.10
426.10
409.72
409.72
Proportion of
Wells Allocated to
Particular Month
0.083
0.083
0.083
0.083
C.I.
Level
95%
99%
95%
99%
Lower
Bound
24.33
22.22
23.18
21.11
Upper
Bound
46.69
48.80
45.11
47.18
Wells were sampled during twenty-three months from April 1988 to February 1990. The monthly
distribution of sampled wells, collapsed over the years spanned by the Survey, is provided in Exhibits 3-7 (CWS
well survey) and 3-8 (rural domestic well survey). The normalized weighted frequencies are also provided in
these exhibits. These frequencies employ the design effect to calculate the sample size required in a simple
random sample that would provide the same Survey precision.
Exhibits 3-7 and 3-8 show that the CWS well survey sampled a comparatively high proportion of
drinking water wells in the months of September, October, and November, and the rural domestic well survey
sampled a comparatively high number of wells in January, February, October, and November.
2 The distribution of samples across months under randomization would be expected to follow a multinominal distribution
with £ = 1/12 (i = 1 through 12). The single parameter marginals of the multinominal distribution are binomial with the
same parameter value.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 47
Exhibit 3-7
Monthly Distribution of Sampled Wells for the Community Water System Well Survey
Month
January
February
March
April
May
June
July
August
September
October
November
December
Total
NPS Implementation
Wells
34
39
31
40
45
27
21
33
78
88
73
31
540
Percent
6.3
7.2
5.7
7.4
8.3
5.0
3.9
6.1
14.4
16.3
13.5
5.7
100.0
Comparable Random Sample
Weighted Frequency
26.60
33.94
25.29
32.14
34.96
19.67
17.66
27.45
61.49
69.70
54.57
22.62
426.10
Percent
6.2
8.0
5.9
7.5
8.2
4.6
4.1
6.4
14.4
16.4
12.8
5.3
100.0
Exhibit 3-8
Monthly Distribution of Sampled Wells for the Rural Domestic Well Survey
Month
January
February
March
April
May
June
July
August
September
October
November
December
Total
NPS Implementation
Wells
114
86
29
43
33
53
41
34
60
102
115
42
752
Percent
15.2
11.4
3.9
5.7
4.4
7.0
5.5
4.5
8.0
13.6
15.3
5.6
100.0
Comparable Random Sample
Weighted Frequency
56.97
43.28
19.77
28.19
21.36
26.82
23.64
19.25
33.44
52.44
64.49
20.10
409.72
Percent
13.9
10.6
4.8
6.9
5.2
6.5
5.8
4.7
8.2
12.8
15.7
4.9
100.0
If the NPS temporal allocation had been random most of the months would be likely to have sample
sizes between the calculated bounds shown in Exhibit 3-6.3 Comparisons of the 95% confidence intervals
presented in Exhibit 3-6 and the weighted sample allocation presented in Exhibits 3-7 and 3-8 show that at
the 95% confidence interval for both the CWS and the rural domestic well survey the sample sizes in half of
3 The confidence intervals presented in Exhibit 3-6 were calculated using the effective sample sizes. The confidence
intervals apply to any particular month rather than all months together. Without these assumptions the confidence intervals
would be narrower than those presented.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 48
the months did not fall within these bounds. As the next section shows, however, there is insufficient evidence
to suggest a temporal effect on pesticide or nitrate contamination in the NFS.
3.2.2 Temporal Analysis
To determine if there is a relationship between the month or season of well water sampling and
pesticide or nitrate detections, EPA evaluated detections by month and for each of seven seasonal variables.
The "all pesticides" group was used for this analysis. Exhibit 3-9 shows the proportion of nitrate and pesticide
detections by month.
Exhibit 3-9
Nitrate and Pesticide Detections by Month
Month
January
February
March
April
May
June
July
August
September
October
November
December
CWS Wells
Proportion
of Nitrate
Detections
0.53
0.61
0.53
0.62
0.39
0.38
0.45
0.62
0.57
0.48
0.50
0.50
Proportion
of Pesticide
Detections
0.101
0.063
0.124
0.109
0.168
0.029
0.183
0.101
0.087
0.064
0.125
0.165
Rural Domestic Wells
Proportion
of Nitrate
Detections
0.51
0.50
0.52
0.45
0.43
0.58
0.63
0.69
0.59
0.63
0.63
0.66
Proportion
of Pesticide
Detections
0.001
0.077
0.045
0.039
0.032
0.066
0.019
0.100
0.013
0.057
0.048
0.030
Exhibit 3-10 presents the significance levels for tests of associations between detections and the
monthly and seasonal variables, performed to determine whether pesticide or nitrate detections are associated
disproportionately with particular months or seasons.4 There is insufficient evidence based on these analyses
to suggest a temporal effect on pesticide or nitrate detections in the NFS.
4 The analysis of pesticide detections by temporal variables was performed for the "all pesticides" group rather than on
any groups of pesticides with smaller numbers of detections. Even so, the assumptions for the corresponding statistical
analysis were not met because of the comparatively large number of groups (24) and the small number of pesticide detections.
Other analyses of pesticide detections by seasonal variables also did not fully maintain the underlying assumptions.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 49
Exhibit 3-10
Nitrate and Pesticide Detections by
Temporal Variables in the Community Water System Well Survey
Temporal Group1
Month
Seasonl
Season2
Seasons
Season4
Seasons
Seasons
Season?
cws
Nitrate
p-value
0.630
0.467
0.107
0.910
0.232
0.722
0.742
0.499
Pesticide
p-value
0.8052
0.978
0.522
0.763
0.782
0.871
0.734
0.662
Rural Domestic Wells
Nitrate
p-value
0.916
0.361
0.564
0.297
0.311
0.215
0.370
0.410
Pesticide
p-value
0.0802
0.947
0.573
0.856
0.813
0.717
0.930
0.807
1 Seasonl = {(JAN, FEB, MAR), (APR, MAY, JUN), (JUL, AUG, SEP), (OCT, NOV, DEC)}
Season2 = {(FEB, MAR, APR), (MAY, JUN, JUL), (AUG, SEP, OCT), (NOV, DEC, JAN)}
Seasons = {(MAR, APR, MAY), (JUN, JUL, AUG), (SEP, OCT, NOV), (DEC, JAN, FEB)}
SeasonA = {(JAN, FEB, MAR, APR), (MAY, JUN, JUL, AUG), (SEP, OCT, NOV, DEC)}
Seasons = {(FEB, MAR, APR, MAY), (JUN, JUL, AUG, SEP), (OCT, NOV, DEC, JAN)}
Seasons = {(MAR, APR, MAY, JUN), (JUL, AUG, SEP, OCT), (NOV, DEC, JAN, FEB)}
Season? = {(APR, MAY, JUN, JUL), (AUG, SEP, OCT, NOV), (DEC, JAN, FEB, MAR)}.
2 May not be a valid test as the statistical assumptions underlying the test have not been
met due to insufficient numbers of detections in some months.
The lowest significance level presented in Exhibit 3-10 is 0.080 for the analysis of pesticide detections by
month in the rural domestic well survey.5
5 The procedures used are not adjusted for the number of hypotheses considered for the temporal analysis. It may be
possible to find single comparisons that have low descriptive significance levels, but when many comparisons are performed
on the same data set an adjustment to account for the number of comparisons is appropriate. For instance, an analysis of
the difference in proportions of nitrate detections for the first six months versus the last six months in the rural domestic well
survey yields a descriptive significance level of 0.082 based on that analysis alone. However, when numerous post hoc
hypotheses are statistically analyzed using the same data the descriptive significance level over the set of comparisons must
be considered, rather than the descriptive significance level of the individual comparison alone. An adjustment that accounts
for the number of hypotheses tested on the same data has the effect of decreasing the significance obtained from a single
comparison alone (i.e., increasing the descriptive significance level). While single comparisons can be found in the temporal
data that yield low descriptive significance levels, the apparent significance may be attributed as much to maximizing the
possibility of finding a single significant comparison as to evidence of an effect that warrants further study. Due to the
comparatively high descriptive significance levels found in the temporal analysis, values are presented without adjustment.
These results do not provide sufficient evidence to suggest a temporal effect.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 50
3.3 Drastic as a Predictor of Drinking Water Well Contamination
The analyses reported in this section tested the hypothesis that a higher occurrence of analyte
detections in drinking water wells is found in counties and sub-county areas with higher Agricultural
DRASTIC scores. Higher positive total scores and subscores indicate greater pollution potential of aquifers.
County-level DRASTIC scores for CWS wells and DRASTIC scores at both the county and sub-county
levels for rural domestic wells were examined for potential relationships with pesticide and nitrate detections.
(Although DRASTIC is designed to score areas larger than 100 acres, an area as large as a county can be
scored, with the results dependent in part on the time and effort expended and in part on the complexity of
the area and the availability of data.) Both overall DRASTIC scores and scores for each of the seven
individual DRASTIC subcomponents (DRASTIC subscores) were included. The following sections describe
the analysis and results for the CWS well survey and the rural domestic well survey.
3.3.1 Evaluation of County-Level DRASTIC
County-level Agricultural DRASTIC scores and subscores were developed as part of the first stage
stratification process for all 3,137 counties and county equivalents in the U.S. For the rural domestic well
survey 90 counties were selected; 399 counties were selected for the CWS survey. Although the DRASTIC
scores presented in this section differ from those used to develop the first stage strata, the qualitative results
may be expected to be similar.6 As Section 3.1 discusses, DRASTIC measured at the county level did not
perform as anticipated. DRASTIC did not successfully identify areas of greater vulnerability to pesticide
contamination. DRASTIC did identify areas of greater likelihood of nitrate contamination, but in the opposite
direction to that expected.
The county-level DRASTIC scores and subscores were used to identify relationships between
DRASTIC and pesticide detections for the "all pesticides" group, nitrate detections, and nitrate concentrations
in CWS wells. Exhibit 3-11 provides a summary of the results of analyses between DRASTIC factors and
pesticide and nitrate detections in CWS wells for the counties included in the CWS study.
The results presented in Exhibit 3-11 generally show that DRASTIC factors are poor predictors of
pesticide and nitrate detections in CWS wells. Notable exceptions include the relationship between depth to
water and pesticide detections, and between hydraulic conductivity and nitrate detections. Other effects with
low significance levels (p-values) include the relationship of net recharge and aquifer media with nitrate
detections. The correlation of these DRASTIC variables with nitrate detections is, however, in the opposite
direction to that expected.
Analyses were performed to identify relationships between DRASTIC factors and nitrate
concentrations given detection above the NFS minimum reporting limit for nitrate of 0.15 mg/L. These
analyses used the logarithmic transformation of the nitrate concentrations. This series of analyses did not
provide sufficient evidence of a relationship in the hypothesized direction for any of the DRASTIC variables.
These results confirm the findings, presented in Section 3.1, that Agricultural DRASTIC at the county
level as applied by NFS is a poor predictor of pesticide detections in CWS wells. (It should be noted that
Agricultural DRASTIC was designed to predict pesticide occurrence in ground water, not in individual wells.)
These results also confirm the results for nitrate detections in CWS wells, presented in Section 3.1, that nitrate
detections are inversely related to Agricultural DRASTIC factors.
6 The procedure followed in the DRASTIC scoring for the first-stage stratification is described in the NFS Phase I
Report, pp. B6-B7. Stratification was based on weighted scores, although VARSCORES were prepared. Because the
VARSCORE best accounts for intracounty variability, it was used for the analysis reported in this section.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 51
Exhibit 3-11
Association of Pesticide and Nitrate Detections
in Community Water System Wells with County-Level DRASTIC Measures*
DRASTIC Factors
Depth to water
Net Recharge
Aquifer media
Hydraulic Conductivity
Total DRASTIC score
Adjusted WGTSCORE
Variable Name
DEPTH
RECHARGE
AQUIFER
CONDUCTIVITY
WGTSCORE
VARSCORE
Pesticide Detection
Direction
of Effect
+
p-value
0.011
Nitrate Detection
Direction
of Effect
-
-
+
-
-
p-value
0.003
0.004
0.001
< 0.0005
< 0.0005
Blanks indicate no significant associations.
The procedures for scoring DRASTIC at the first stage were identical for both CWS and rural
domestic well surveys. For the rural domestic well survey, ninety counties were selected for sampling.
Sampled wells were selected from those counties. The first stage stratification variable developed from the
DRASTIC scoring system did not perform as anticipated in the rural domestic well survey. That is,
Agricultural DRASTIC as measured through the first stage stratification variable did not successfully identify
areas of greater incidence of pesticide contamination. The topography variable was associated with nitrate
detections. Exhibit 3-12 provides a summary of the results of analyses between DRASTIC factors and pesticide
and nitrate detections in rural domestic wells.
Exhibit 3-12
Association of Pesticide and Nitrate Detections in
Rural Domestic Wells with County-Level DRASTIC Measures*
DRASTIC Factors
Aquifer media
Topography
Variable
Name
AQUIFER
TOPOGRAPHY
Pesticide
Detection
Direction
of Effect
p-value
Nitrate
Detection
Direction
of Effect
-
+
p-value
0.034
0.044
* Blanks indicate no significant associations.
The results presented in Exhibit 3-12 generally show that DRASTIC factors are poor predictors of
pesticide and nitrate detections in rural domestic wells. The relationship between aquifer media and nitrate
detections is contrary to the intended hypotheses.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 52
Further analyses were performed to identify relationships between DRASTIC factors and nitrate
concentrations. This series of analyses did not provide sufficient evidence of a relationship with pesticide
concentrations in the hypothesized direction for any of the DRASTIC variables. There was, however, evidence
of a relationship between topography and nitrate concentrations (significance level < 0.00005).
These results confirm the findings, presented in Section 3.1, that Agricultural DRASTIC at the county-
level is a poor predictor of pesticide detections in rural domestic wells, even though Agricultural DRASTIC
was designed to predict pesticide occurrence.
3.3.2 Evaluation of Sub-County-Level DRASTIC
Sub-county DRASTIC scores and subscores were developed as part of the second stage stratification
process for the 90 counties in which sampling of rural domestic wells occurred.7 Analyses of sub-county-level
DRASTIC paralleled the procedures used for first stage DRASTIC analysis. Exhibit 3-13 summarizes the
results of these analyses for pesticide detections and Exhibit 3-14 presents results for association of DRASTIC
scores with nitrate detections.
Exhibit 3-13
Association of Pesticide and Nitrate Detections in
Rural Domestic Wells with Sub-County DRASTIC Measures*
DRASTIC Factors
Aquifer media
Impact of vadose zone
Hydraulic Conductivity
Variable
Name
AQUIFER
IMPACT
CONDUCTIVITY
Pesticide Detection
Direction
of Effect
+
p-value
0.020
Nitrate Detection
Direction
of Effect
-
-
p-value
0.036
0.006
Blanks indicate no significant associations.
The results presented in Exhibit 3-13 generally show that DRASTIC factors at the sub-county-level
are also poor predictors of pesticide and nitrate detections in rural domestic wells. The most noticeable
exception concerns the relationship between the impact of vadose zone media and pesticide detections. Other
effects that are apparent are the relationship of aquifer media, and hydraulic conductivity with nitrate
detections. The correlation of these DRASTIC variables with nitrate detections is in the opposite direction
to that expected.
Further analyses were performed to identify relationships between sub-county DRASTIC factors and
nitrate concentrations in wells. This series of analyses used the logarithmic transformation of nitrate
concentration. Results are reported in Exhibit 3-14.
7 The second-stage stratification for rural domestic wells is described in the NFS Phase I Report, pp. 18-21. In the 90
counties, "cropped and vulnerable" and not "cropped and vulnerable" sub-county areas were identified using DRASTIC scores
and estimates of agricultural activity. Sub-county DRASTIC scores were not prepared for the CWS survey.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 53
Exhibit 3-14
Models of Relationship of Sub-County DRASTIC Factors
and Nitrate Concentrations for Rural Domestic Wells*
DRASTIC Factors
Aquifer media
Topography
Total DRASTIC score
Variable Name
AQUIFER
TOPOGRAPHY
WGTSCORE
Estimated
Intercept
-0.542
-0.714
-0.760
Estimated
DRASTIC
Factor
Coefficient
0.047
0.046
0.007
p-value
0.006
< 0.0005
0.014
Nitrate concentrations have been logarithmically transformed.
Although there was insufficient evidence of associations in the hypothesized direction between
DRASTIC factors at the sub-county-level and nitrate detections in drinking water wells, there is some evidence
in support of associations with nitrate concentrations in the manner expected by the DRASTIC system. In
particular, there is strong evidence that nitrate concentrations are related to aquifer media, topography, and
total DRASTIC scores. Possible explanations of the difference in performance of the two approaches to
modeling nitrate associations include: (1) nitrate occurs in groundwater at very low concentrations, the
DRASTIC scoring system may not be effective in predicting the presence of nitrate in drinking water wells
at low concentrations (including zero); and (2) logistic regression models are not appropriate for the full range
of nitrate concentrations (i.e., greater than zero concentration). Since the sub-county DRASTIC results are
not supported by the county-level DRASTIC results for rural domestic wells or CWS wells, these possible
explanations pertain to sub-county DRASTIC only. Because nitrate detections and concentrations are
recorded only at levels above the minimum reporting limit of 0.15 mg/L, resolution of the possible
explanations using NFS data is not possible.
The sub-county DRASTIC results parallel the findings, presented in Section 3.1 for second stage
stratification, that no relationship could be demonstrated between Agricultural DRASTIC scores at the sub-
county level and pesticide detections in rural domestic wells. These results similarly confirm the results for
nitrate detections in rural domestic wells at the second stage that nitrate detections are not related to
Agricultural DRASTIC factors. However, nitrate concentrations were found to be related to three DRASTIC
factors.
3.4 Well Characteristics and Activities Conducted Near the Well
This section presents the results of the statistical analyses of the relationships between detections of
nitrate or at least one pesticide or pesticide degradate in CWS and rural domestic wells and factors such as
well location and construction, local land uses, and local land features. The data used for these analyses
consist of pesticide and nitrate detections above the minimum reporting limits of the NFS and responses to
detailed questionnaires obtained from homeowners or residents, community water system owners or operators,
local area experts, and well samplers. The questionnaires were reviewed by hydrogeologists and soil scientists
to identify site-specific cropping practices and pesticide use patterns that could generate useful hypotheses for
statistical testing. The hypotheses tested include the following:
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 54
• Analytes are detected more frequently in areas of high pesticide use;
• Analytes are detected more frequently in shallow and/or not optimally constructed
and/or old wells;
• Analytes are detected more frequently in areas where non-farm pesticides are used;
• Crops or crop types are good indicators of pesticide or nitrate contamination;
• Point sources of contamination (septic tanks, pesticide spills, pesticide and fertilizer
dealerships, and disposal sites) are good indicators of contamination; and
• Analytes are detected more frequently in areas where irrigation is used.
Section 3.4.1 provides a brief discussion of the data sources and statistical methods used, Section 3.4.2
provides details of the results of the ensuing statistical analyses, and Section 3.5.3 describes a review of
reported pesticide use activities and the relationship between reported pesticide use and pesticide detections.
3.4.1 Data Sources and Statistical Methods
Data regarding well location and construction, activities occurring near the well, and local land
features were obtained primarily from questionnaires administered prior to, or at the time of, well sampling.
Three detailed questionnaires were completed during well sampling at both CWS and rural domestic wells.
These sets of questionnaires contained similar items for the two surveys8:
• Team Leader Introduction and Well Observation Record. The Team Leader
Introduction (TLI) contains questions about well water treatment, septic systems, and
diagrams of local land features within 300 feet of the well (such as drainage ditches,
farmland, wooded areas, bodies of water, paved areas, buildings, or other wells). The
Well Observation Record (WOR) contains questions about the well and the
surrounding area, including topography, soils, and well protection. The TLI and
WOR were administered together as one questionnaire. Responses for this
questionnaire were provided primarily by well operators and well owners.
• Main Questionnaire. These questionnaires contain detailed questions about well
construction, non-farm and farm pesticide use, and nitrogen fertilizer use on the
property where the well is located. This questionnaire was administered primarily
to well operators and well owners.
• Local Area Questionnaire. This questionnaire was predominantly administered to
county agricultural extension agents following well sampling to collect information
about local conditions within one-half mile of the well. Information was collected
on crops grown, pesticide use, local land features such as water bodies, landfills, and
chemical manufacturing facilities, and irrigation practices.
In addition to these questionnaires, a fourth questionnaire was administered prior to the sampling of
rural domestic wells. This instrument, the Second Stage County Agent Questionnaire, was administered to
county agricultural extension agents as part of the second stage stratification process. Information was
collected on crops grown, pesticides used, and local soil characteristics within the United States Geological
Survey (USGS) 7.5 minute quadrangle area containing the sampled wells.
See NFS Phase I Report Appendix D for copies of the questionnaires.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 55
Most questions permitted only yes or no as a response. For example, a typical question was "Is the
property on which the well is located used for farming?". Other questions allowed a broader set of categories
as possible responses. For example the question "What is the predominant soil condition within 300 feet of
the well?" allowed seven different responses ranging from clay soil to sandy/silty soil. A few questions required
numerical answers describing, for instance, the depth of the well or the year of well construction. Analysis
of the relationship between items corresponding to the first two cases (categorical responses) and pesticide
or nitrate detections consisted primarily of chi-square procedures. Analysis of the relationship between well
depth or year of construction and pesticide or nitrate detections consisted primarily of univariate logistic
regression analysis with the continuous variables as the explanatory (independent) variable. Exploratory
analyses were performed in all cases to verify that statistical assumptions underlying these tests were adequately
satisfied. The procedures used are described in more detail in Chapter 2 of this report.
3.4.2 Results of Analysis of Questionnaire Items
This section presents the results of analyses of associations of pesticide and nitrate detections with
well characteristics and activities reported near sampled wells. Results are presented in the following order:
• Analyses of the relationship of questionnaire variables and nitrate detections for
CWS and rural domestic wells;
• Analyses of the relationship of questionnaire variables and pesticide detections for
CWS and rural domestic wells; and
• Summary of results including comparisons of results for pesticide detections and
nitrate detections.
The first two subsections include discussions of the analyses involving questionnaire items in the
following general categories of well characteristics and activities conducted near the well. Questionnaire items
are described and the source questionnaire and item number are provided in the table of results. The
questionnaires addressed the following topics:
Farming practices, including agricultural pesticide use and nitrogen fertilizer use. Questions
concerning farming practices were posed in the Main and Local Area Questionnaires.
Questions included Main Questionnaire items about farming, pesticides used for farming,
farm fertilizer use, and pesticide storage on the property where the well is located, and Local
Area Questionnaire items about crops farmed and land used for pasture within one-half mile
of the well.
Non-farm pesticide use. Questions concerning non-farm pesticide use around the home or
in the vicinity of the sampled wells came from both the Main and Local Area Questionnaires.
Well construction characteristics. Well characteristic information was collected using the
Well Observation Records and the Main Questionnaires. These questions prompted
responses about well protection, well construction (casing, grouting, surface closure, drilling,
and age), well depth, and recent problems with well operations.
Use of septic systems. Information about septic systems used near the well was collected
from the Team Leader Introductions and Local Area Questionnaires, and from the rural
domestic well Main Questionnaire. These questions pertained explicitly to septic tanks, septic
fields, cesspools, and water disposal ponds.
Local land area conditions. Information about land features near the well was drawn from
the Well Observation Records, and included questions about local topography and
predominant soil condition.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 56
Type of aquifer (CWS well survey only). The CWS Main Questionnaire asked whether the
CWS wells draw water from confined or surficial aquifers.
Water bodies near the well. The Local Area Questionnaire contained a series of questions
pertaining to water bodies within one-half mile of the sampled wells. Water bodies included
streams, rivers, irrigation canals, reservoirs, lakes, bays, springs, ponds, and drainage ditches.
Some questions distinguished between lined and unlined water bodies. Questions about
drainage ditches and bodies of water in general were also contained in the Well Observation
Record.
Irrigation. Questions about irrigation practices were included in the Local Area
Questionnaires. These questions covered types of irrigation and sources of water for
irrigation, but did not ask about amounts of water used for irrigation.
Facilities near the well. The Local Area Questionnaire contained a series of questions
pertaining to installations within one-half mile of the sampled wells. Installations considered
included chemical plants, airports, military bases, waste sites, landfills, waste treatment
facilities, golf courses, and pesticide retail outlets.
Water treatments applied to well water. A series of questions about treatment of the well
water prior to drinking was administered in the Team Leader Introductions for both surveys,
and included questions about use of water softeners, filters, chlorine treatment, and carbon
or charcoal treatment. (The water samples in the NFS were always taken prior to treatment.)
The responses obtained on NFS questionnaires and used in the analysis of questionnaire items were
not validated from other independent sources. Questionnaire responses were reviewed for internal consistency.
Results of field interviewer debriefings that provide information about the quality of the questionnaire results
are provided in Chapter 4.
3.4.2.1 Analysis of Nitrate Detections and Questionnaire Items
Nitrate is the analyte most frequently detected in both the CWS and rural domestic well surveys.
Survey results indicate that over half of the CWS and rural domestic wells in the United States contain nitrate
(measured as N) at or above the Survey's minimum reporting level (MRL) of 0.15 mg/L. This section
discusses the analysis of associations between nitrate detections and factors such as well construction, well
location, and activities conducted in the vicinity of the well. Results are presented first for the CWS well
survey, and then for the rural domestic well survey. This section concludes with a summary of the results
including a comparison of CWS and rural domestic well survey results.
CWS Well Survey Results
Exhibit 3-15 provides a summary of factors that are related to nitrate detections in CWS wells.
Farming (CWS Main Questionnaire (item C.I)) is related to nitrate detections. This conclusion is not
substantiated by analysis of the other farming-related questions.
Analysis of responses to questions about well characteristics (such as well casing, grouting, etc.) did
not, in general, show evidence of a relationship with nitrate detections.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 57
Exhibit 3-15
Factors from NFS Questionnaires Associated with the Probability of
Detecting Nitrate for Community Water System Wells
Factor (with Source Questionnaire and Item)
Property where well is located farmed since Jan. 1,
1984.*
(Main C.1)
Flood irrigation methods used within one-half mile of the
well.
(LAQ 14b)
Well draws water from unconfined aquifer(s).
(Main A.25)
Drainage ditch within 300 feet of the well.
(WOR 6a)
Lined drainage ditch within one-half mile of the well.
(LAQ 12e)
Unlined water bodies within one-half mile of the well.
(LAQ 12a, b, d, g, h, i, j, k, I)
Water Disposal Pond within one-half mile of the well.
(LAQ 11e)
Water treatment applied to the water from the well.
(TLI 13a, b, c, d, e, f)
Direction
of
Effect
Significance
Level
(p-value)
0.022
0.019
0.006
0.030
0.019
0.010
0.003
< 0.0005
TLI = Team Leader Introduction
WOR = Well Observation Record
Main = Main Questionnaire
LAQ = Local Area Questionnaire
* For wells sampled after March 5, 1989, this question was asked about farming since
January 1, 1985.
Analysis of well depth indicated a relationship with nitrate detections and nitrate concentrations. The
total distance from the ground surface to the bottom of the well was collected from question A.5 in the CWS
Main questionnaire. Results for the relationship between well depth and nitrate detections are presented in
Exhibit 3-16. There is a clear relationship between nitrate detections and well depth in the CWS well survey,
i.e., there is a greater chance of nitrate contamination in shallow wells. Furthermore, there is sufficient
evidence to suggest that higher concentrations of nitrate are found in shallow wells. The average well depth
for wells containing nitrate at concentrations above the minimum reporting limit is 310 feet for CWS wells,
compared to an average depth of 434 feet for wells in which nitrate was not detected.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 58
Exhibit 3-16
Models of the Relationship Between Nitrate Detections* and
Concentration and Well Depth in Community Water System Wells
Dependent
Variable
Nitrate
Detection
dn(Nitrate
Concentration)
Explanatory
Variable
flnfWell
Depth)
{nfWell
Depth)
Estimated
Intercept
348
1.388
Standard
Error
0.621
0.467
Estimated
Well Depth
Coefficient
(in Feet)
-0.414
-0.213
Standard
Error
0.132
0.087
4n(Feet)
Coefficient
Significance
Level
(p-value)
< 0.0005
0.015
Nitrate concentrations have been logarithmically transformed.
Information on the year in which sampled wells were originally constructed or drilled was collected
from question A.1 in the CWS Main Questionnaire. Logistic regression analyses were performed to determine
the extent of the relationship between year of well construction and both nitrate detections and nitrate
concentrations. There is insufficient evidence based on this analysis to suggest that the year of construction
is associated with nitrate detections. The average number of years since construction for wells containing
nitrate at concentrations above the minimum reporting limit is 24.5 years in CWS wells, compared to an
average age of 23.3 years for wells in which nitrate was not detected.
Statistical analysis involving data collected about topography, soil type, and other land features within
300 feet of the well did not show evidence of an association with nitrate detections.
Analysis of the CWS aquifer question revealed that the proportion of nitrate detections in the
unconfined aquifer group (62.6%) is significantly greater than the proportion in the confined aquifer group
(44.7%). These results indicate that unconfined aquifers are more at risk of nitrate contamination than
confined aquifers.
Analysis of questions related to water bodies indicate that there is a lower chance of detecting nitrate
in CWS wells near unlined water bodies, drainage ditches or water disposal ponds within one-half mile of the
well. Similar results are obtained from analysis of individual factors as for the overall unlined water body
factor presented in Exhibit 3-15. For instance, the presence of streams or rivers, unlined drainage ditches,
natural lakes, man-made lakes, ponds, and bays or estuaries, within half mile of the well, are individually, as
well as collectively, inversely associated with nitrate detections.
Of the irrigation methods included for analysis in the NFS, only flood irrigation shows an association
with nitrate detections. Other types of irrigation and sources of water for irrigation within one-half mile of
the well generally were not found to be associated with nitrate detections.
There is a lower chance of detecting nitrate in CWS wells with water treatment. The dominant factor
for this relationship is use of water softeners, although there is evidence of an association with nitrate
detections for all individual water treatments as well as for water treatments collectively.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 59
Rural Domestic Well Survey Results
Exhibit 3-17 provides a summary of factors that analysis suggests are related to nitrate detections in
rural domestic drinking water wells. Nitrate is different from pesticides with respect to its origins in ground
and surface water. While pesticides always occur ultimately as a result of preceding human activity, and do
not occur naturally, nitrate may be either anthropogenic in origin or naturally occurring. Natural forms of
nitrate can be produced by certain algae, species of bacteria, and other organisms that have the capacity of
photosynthesis. Nitrate also can occur in rainwater and may be of geologic origin. Detection of nitrate in well
water, therefore, may not in all cases indicate contamination by anthropogenic sources. According to the U.S.
Geological Survey, concentrations of nitrate can be classified as follows:
"• Less than 0.2 mg/L - Assumed to represent natural background concentrations.
• 0.21 to 3.0 mg/L - Transitional; concentrations that may or may not represent
human influence.
• 3.1 to 10 mg/L - May indicate elevated concentrations resulting from human
activities.
• More than 10 mg/L - Exceeds maximum concentration in National Interim Primary
Drinking Water Regulations."9
Results are presented in this section for associations with both nitrate detections and concentrations.
Many questions were administered in the rural domestic well survey concerning farming and farming
practices, such as pesticide use and storage and nitrogen fertilizer use. Of these, only use of the property
where the well is located for pasture shows a relationship with nitrate detections. Use of non-farm pesticides
on home lawns (Main B.2) and, more generally, use within one-half mile of the well (LAQ 5), are inversely
related to nitrate detections.
Factors associated with well construction that were measured categorically did not, in general, show
a relationship to nitrate detections. Use of a covered pit to protect the well surface was found to be associated
with an increased chance of nitrate detections, that is, the result contradicted the hypothesis that there is a
smaller chance of detecting nitrate in wells that have a covered pit to protect the well surface. The analysis
provided evidence that the presence of other operating wells within 500 feet of the sampled well was directly
related to nitrate detections. For the rural domestic well survey the quality of the water was also measured
through responses to item A.10 of the Main Questionnaire. Analysis of this data indicates there may be a
greater chance of detecting nitrate in wells in which the water is hard. There also may be a greater chance
of detecting nitrate in wells that do not have water with color or odor.
Analysis of well depth and year of well construction showed evidence of a relationship with nitrate
detections. Well depth information on the total distance from the ground surface to the bottom of the well
was collected from question D.6 in the Main questionnaire. Analysis was performed to determine the extent
of the relationship between well depth and nitrate detections and nitrate concentrations above 0.15 mg/L.
Results of these analyses are presented in Exhibit 3-18. There is a clear relationship between nitrate
detections and concentrations and well depth in the rural domestic wells, i.e., there is a greater chance of
nitrate contamination in shallow wells and greater nitrate concentrations can be expected in shallow wells.
9 Madison, R J. and J.O. Brunett, Overview of the Occurrences of Nitrates in Groundwater of the United States, National
Water Summary 1984, U.S. Geological Survey Water Supply Paper #2275, 1985, p. 95; see also Power, J.F. and J.S.
Schepers, Nitrate Contamination of Groundwater in North America. 26 (1989), Agriculture, Ecosystems and Environment,
pp. 165-187. National Interim Primary Drinking Water Regulations became permanent Maximum Contaminant Levels
following preparation of the 1984 U.S.G.S. paper.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 60
Exhibit 3-17
Factors from NFS Questionnaires Associated with the Probability of
Detecting Nitrate for Rural Domestic Wells
Factor (with Source Questionnaire and Item)
Direction
of Effect
Significance
Level
(p-value)
Other operating wells within 500 feet.
(Main B.12)
Covered pit protects the well at the surface.
(WOR 3c)
Land used for pasture within one-half mile of the well since Jan 1,
1986.*
(LAQ3)
Pesticide retail outlet within one-half mile of the well.
(LAQ 11n)
Unlined body of water within 300 feet of the well.
(WOR 6b)
Grain elevator within one-half mile of the well.
(LAQ 11o)
Ground water used for irrigation.
(LAQ 15a)
Well water is hard (high iron or mineral content).
(Main A.10c)
Unlined body of water within one-half mile of the well.
(LAQ 12a, b, d, g, h, i, j, k, I)
Unlined drainage ditch within one-half mile of the well.
(LAQ 12d)
Water treatment applied to the water from the well.
(TLI 2a, b, c, d, e, f)
0.043
0.015
0.013
0.013
0.037
0.036
0.033
0.033
0.015
0.0005
0.0005
TLI = Team Leader Introduction
WOR = Well Observation Record
Main = Main Questionnaire
LAQ = Local Area Questionnaire
* For wells sampled after March 15,1989, this question asked about pastures since January 1,
1987.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 61
Exhibit 3-18
Models of the Relationship Between Nitrate Detections*
and Concentrations and Well Depth and Year of
Well Construction for Rural Domestic Wells
Dependent
Variable
Nitrate
Detection
0n(Nitrate
Concentration)
Nitrate
Detection
Explanatory
Variable
{n(Well
Depth)
0n(Well
Depth)
4n(Age)
Estimated
Intercept
2.646
2.617
-0.577
Standard
Error
0.616
0.489
0.440
Estimated
Explanatory
Variable
Coefficient
(£n Scale)
-0.499
-0.513
0.325
Standard
Error
0.111
0.098
0.130
Significance
Level
(p-value)
< 0.0005
< 0.0005
0.013
Nitrate concentrations have been logarithmically transformed.
The average well depth for wells containing nitrate at concentrations above the minimum reporting limit is
144 feet for rural domestic wells, compared to an average depth of 213 feet for wells in which nitrate was not
detected.
Information on the year in which sampled wells were constructed was collected from question D.3 in
the rural domestic well Main questionnaire. As Exhibit 3-18 shows, there is some evidence of a relationship
between nitrate detections and year of construction in the rural domestic well survey, but there is no evidence
of a similar relationship for CWS wells.
Analysis of questions related to water bodies indicate that the presence of unlined water bodies or
drainage ditches within one-half mile of rural domestic wells is negatively associated with nitrate detections.
The result for unlined water bodies is largely influenced by the data for unlined drainage ditches. Nitrate
detections in rural domestic wells are less likely if there are bodies of water present within 300 feet of the
wells.
Irrigation and irrigation methods within one-half mile of the well were not found to be associated with
nitrate detections. Analysis of the sources of water used for irrigation shows some evidence of a lower chance
of detecting nitrate if groundwater near the well is used as an irrigation source.
The Local Area Questionnaire included items for many types of installations in the vicinity of the well.
Of these, the data provide some evidence that there is a lower chance of detecting nitrate if a grain elevator
is within one-half mile of the well. The data also provide some evidence that the presence of pesticide retail
outlets within one-half mile of the well is directly related to nitrate detections.
Water treatment is found to be inversely correlated with nitrate detections in rural domestic wells.
The dominant factor for this relationship is use of water softeners.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 62
Summary of Nitrate Results
In general, there were an adequate number of nitrate detections to satisfy the statistical assumptions
required for the procedures used. Exhibits 3-15 through 3-18 portray the results of analyses for which the
descriptive significance level was less than 0.05. Of the many statistical analyses that were performed, these
are the only ones for which preliminary conclusions can readily be made.
A few themes recur throughout the analysis:
• There is recurring evidence that nitrate is less likely to be detected if there are
unlined bodies of water in the vicinity of the wells;
• Both the CWS and rural domestic well surveys indicate that there is a greater chance
of nitrate detection in shallow wells and greater concentrations of nitrate can be
expected in shallow wells; and
• Water treatment (after the point at which the wells were sampled) may be negatively
associated with nitrate detections.
Another statistically significant factor concerns the generic type of aquifer from which CWS wells draw
water:
• The CWS well survey indicates that there is a greater chance of nitrate detection
from wells that draw water from unconfined aquifers.
3.4.2.2 Analysis of Pesticide Detections and Questionnaire Items
The Survey detected pesticides and pesticide degradates much less frequently than nitrate. Two
analytes, DCPA acid metabolites (degradates of DCPA) and the pesticide atrazine, were the most frequently
detected pesticide analytes in both CWS and rural domestic wells. Other analytes that were detected above
minimum reporting limits in CWS wells include dibromochloropropane, dinoseb, hexachlorobenzene,
prometon, and simazine. The rural domestic well survey also recorded detections of dibromochloropropane,
prometon, simazine, EDB, lindane, ETU, bentazon, and alachlor. These results are reported in detail the NFS
Phase I Report and summarized in Exhibits 1-1 and 1-2 in Chapter 1. This section presents the results of the
analysis of relationships between detections of pesticides and pesticide degradates and well characteristics, local
land features, and land uses on the property where the well is located or within one-half mile of the well.
Extensive analysis of possible relationships among these factors was not possible because there were fewer
detections of pesticides than nitrates.
The description of results follows the same pattern as those for the nitrate analysis presented in
Section 3.5.2.1. Results are presented primarily for the detection of at least one pesticide or pesticide
degradate, denoted by "any pesticide". There are two basic reasons for this approach. First, the NPS was
designed to collect data on the presence of at least one pesticide or pesticide degradate rather than on any
individual chemical. Second, generally there are too few detections of any single pesticide analyte to perform
reasonable statistical analyses.
CWS Well Survey Results
Exhibit 3-19 provides a summary of factors that the analysis suggests are related to pesticide detections
in CWS wells.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 63
Exhibit 3-19
Factors from NFS Questionnaires Associated with the Probability of Detecting
at Least One Pesticide or Pesticide Degradate
for Community Water System Wells
Factor (with Source Questionnaire and Item)
Other operating wells within 500 feet of the well.
(Main A.35)
Use of well house or shed as protection at the well surface.
(WOR 3a)
Unlined body of water within 300 feet of the well.
(WOR 6b)
Direction
of Effect
*
Significance
Level
(p-value)
0.006
0.001
0.014
WOR = Well Observation Record
Main = Main Questionnaire
LAQ = Local Area Questionnaire
Questions related to farming practices and pesticide use and non-farm pesticide use did not provide
evidence of a relationship with pesticide detections. Well construction characteristics did not show a
relationship with pesticide detections in general; however, there is some evidence that use of a well house or
shed and the presence of other operating wells within 300 feet of the sampled wells are associated with an
increased likelihood of pesticide detections. Well depth apparently also is related to pesticide detections.
Results for the relationship between pesticide detections and well depth are presented in Exhibit 3-20.
Exhibit 3-20
Logistic Regression Model for Well Depth for Community Water System Wells
Explanatory
Variable
Cn(Well Depth)
Estimated
Intercept
0.603
Standard
Error
0.938
Estimated
Explanatory
Variable
Coefficient
((n Scale)
-0.518
Standard
Error
0.176
Significance
Level
(p-value)
0.003
The average well depth in wells containing pesticides at concentrations above the minimum reporting
limits is 230 feet in CWS wells, compared to an average depth of 385 feet for wells not containing at least one
pesticide.
Analyses involving data collected about topography, soil type, and other land features within 300 feet
of the well did not show evidence of a direct association with pesticide detections. Analyses of questions
related to water bodies indicate that there is a lower likelihood of detecting pesticides in CWS wells if unlined
bodies of water are within 300 feet of the well.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 64
Irrigation, irrigation methods, and sources of irrigation water within one-half mile of the well were
not found to be associated with pesticide detections, and there is no evidence of a relationship between water
treatment and pesticide detections. There was also no evidence of a relationship between the presence of
installations within one-half mile of CWS wells and pesticide detections.
Rural Domestic Well Survey Results
The rural domestic well survey recorded too few detections of pesticides and pesticide degradates to
satisfy the assumptions underlying the procedures used for analyses. This problem was compounded by the
small number of responses recorded for some possible answers to questionnaire items (i.e., the responses were
very unbalanced). The problems caused by too few detections or small numbers of responses in some
categories of answers can result in spurious or unreliable results. Chapter 2 provides a detailed explanation
of the issues involved in determining appropriateness of the modeling procedure. Consequently, no result
from the pesticide analysis in the rural domestic well survey is statistically significant when the statistical
assumptions have been adequately satisfied and the conclusion agrees with the hypothesis of interest.
Summary of Pesticide Results
The three strongest associations of pesticide detections and questionnaire variables resulted from
analysis of well depth for CWS wells, the use of a well house or shed to protect the well at the surface, and
the presence of other operating wells in the vicinity of CWS wells, although the results for well house or shed
protection at the surface do not support the intended hypothesis. Exhibits 3-18 and 3-19 portray results of
analyses for which the statistical assumptions were adequately satisfied and for which descriptive significance
level was less than 0.05. Of the many statistical analyses that were performed, these are the only ones for
which preliminary conclusions can readily be made. There were, in many cases, too few pesticide detections
to satisfy the statistical assumptions required for the procedures used. Although statistically significant results
are not supported by similar results for other questionnaire items, this is partly due to the few detections or
imbalance in the number of responses in different categories of the questions.
Responses to questions administered in the Second Stage County Agent Questionnaire were also
analyzed for possible associations with pesticide detections in rural domestic wells. This questionnaire was
administered to county agricultural extension agents prior to selecting rural domestic drinking water wells as
part of the second stage stratification process. Though the information on local agricultural practices was
considered to be more reliable than similar information gathered from questionnaires administered to
homeowners, variables from this questionnaire also failed to provide evidence of associations with pesticide
detections. Responses from this questionnaire were also analyzed for possible associations with nitrate
detections, but there was no evidence to suggest relationships between questionnaire variables and detections.
3.4.3 Review of Pesticide Use Patterns
Information about particular pesticides used near sampled wells was gathered for both CWS and rural
domestic wells10. These data were analyzed to determine if there was a relationship between reported
pesticide use and pesticide detections. Based on the results of the analysis, there is insufficient evidence to
suggest that pesticide use as reported on NFS questionnaires is associated with pesticide detections. Possible
explanations for the lack of association include measurement error of the respondents, failure to fully capture
data representing pesticide use in the recharge area of the well, and application of pesticides prior to the
period for which information was obtained.
10 The CWS and rural domestic well questionnaires imposed different restrictions on the size of the local area and the
number of years of use to be reported upon. For example, the Local Area Questionnaires required responses pertaining to
an area within one-half mile of the sampled well and within three years prior to sampling, while the Main Questionnaire
required responses for farm pesticide use on the property within the past five years.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 65
Questionnaire responses about specific pesticide use generally provided brand names for pesticides.
The brand names were matched with active ingredients from the NFS pesticides. Pesticide degradates, such
as DCPA acid metabolites, were matched to their NFS pesticide active ingredients. Some chemicals could not
be distinguished through the brand names supplied as responses to the questionnaire items. For example,
chlordane could not be distinguished from the alpha and gamma chlordane isomers. NFS analytes designated
as "qualitative only" were not included. One detected chemical, ETU, is a pesticide degradate that does not
have a parent compound that was included in the NFS. Consequently, ETU is not included in the analyses
reported in this section. Because ETU was detected very infrequently, it is unlikely that its exclusion has a
significant impact on the analysis presented in this section. Full details of the 82 pesticide active ingredients
(hereafter pesticides) included for analysis are provided in Exhibit 3-21.
Questions about both farm and non-farm pesticide use, storage, disposal, and spillage were analyzed
from main and local area questionnaires from the CWS and rural domestic well surveys. Reported pesticide
use information from the second stage county agent questionnaire also was analyzed.
Several approaches were taken to determine the extent of the relationship between pesticide use
reported on NFS questionnaires and detections of pesticides. Exhibits 3-22 through 3-26 summarize reported
pesticide use by pesticide detections, dividing them into four categories: (1) detected and reported as used;
(2) detected and not reported as used; (3) not detected and reported as used; and (4) not detected and not
reported as used.
Although these categories are mutually exclusive for individual well by pesticide observations, they
are not mutually exclusive at the pesticide or well levels individually. For example, Exhibit 3-23 indicates that
four pesticides were detected in and reported as used near rural domestic wells. This does not mean that every
time these pesticides were detected they were reported as used. Instead, each of the four pesticides was
reported as used in the vicinity of at least one well in which it was detected. The same four pesticides,
however, could also appear in other categories. For example, atrazine falls into all four groups: atrazine was
reported as used in some wells in which it was detected; it was not reported as used in other wells in which
it was detected; it was reported as used in many wells in which it was not detected; and it was not reported
as used in many wells in which it was not detected. At the pesticide level, observations (i.e., pesticides) are
assigned to the first category into which they fell. For example, atrazine is included in the first group because
it was reported as used in at least one well in which it was detected. Similarly, at the well level, wells are
assigned to the first group into which they fall regardless of the possibility of use of multiple pesticides at a
well site.
Analyses of reported pesticide use and detections of NFS pesticide active ingredients were performed
at three levels:
1. Pesticide - The total number of observations is 82. Each observation corresponds to
an NFS active ingredient and indicates to which of the above four groups a pesticide
belongs;11
2. Well - The estimated percent of wells at which use of at least one NFS pesticide
active ingredient at the well site was reported; and
3. Well by Chemical - Percents are provided that combine the previous two factors and
correspond to estimated proportions of NFS wells nationally aggregated for each
chemical.
11 NFS weights are not included in this analysis as weights correspond to wells rather than pesticides.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 66
Exhibit 3-21
NFS Pesticide Active Ingredients included
in Analysis of Detections by Reported Pesticide Use
Alachlor
Aldicarb
Aldrin
Ametryn
Atraton
Atrazine
Barban
Baygon8
Bentazon
Bromacil
Butachlor
Butylate
Carbaryl
Carbofuran
Carboxin
Chlordaneb
Chlorneb
Chlorothalonil
Chlorpropham
Cyanazine
Cycloate
DBCP
DCPAC
Dicamba
Dichlorprop
Dichlorvos
Dieldrin
Dinoseb
Diphenamid
Diuron
EDB
Endosulfand
Endrin
EPIC
Ethoprop
Etridiazole
Fenamiphos
Fenamirol
Fluometuron
Fluridone
Heptachlor
Hexachlorobenzene
Hexazinone
Lindane6
U'nuron
Methiocarb
Methomyl
Methoxychlor
Metolachlor
Metribuzin
Mevinphos
MGK-264
Molinate
Napropamide
Neburon
Norflurazon
Oxamyl
POP
Pebulate
Permethrinf
Piclcram
Prometon
Prometryn
Propachlor
Propanil
Propazine
Propham
Simazine
Simetryn
Stirofos9
SWEP
Terbuthiuronh
Terbutryn
Triademefon
Tricyclazole
Trifluralin
Vernolate
2,4-D
2,4-DB
2,4,5-T
2,4,5-TP
4,4-DDT
a Pesticide use information for Baygon was also collected under the name Propoxur.
b Chlordane data consists of chemical analysis of alpha and gamma chlordane
isomers, and use of chlordane, alpha chlordane or gamma chlordane.
c DCPA data includes detections of its degradate DCPA acid metabolites. Use data for
DCPA was also collected under the names Chlorthal-Dimethyl and Dacthal.
d Endosulfan data consists of chemical analysis of endosulfan I and II, and use of
endosulfan, endosulfan I, or endosulfan II.
e Lindane data consists of chemical analysis of alpha, beta and gamma HCH, and use
of Lindane, alpha HCH, beta HCH, or gamma HCH.
f Permethrin use data was collected under the names permethrin, cis-permethrin and
trans-permethrin. Chemical analysis was performed on cis-permethrin and trans-
permethrin.
g Stirofos use information was also collected under the name tetrachlovinphos.
h Terbuthiuron data was collected under the names Terbuthiuron, Tebuthiuron, and
Terbacil. Chemical analysis was performed on Tebuthiuron and Terbacil.
The pesticide degradates ETU and methyl paraoxon were not included in the analysis.
Parent compounds for these two chemicals degrade very rapidly and were not included
in the list of NPS analytes.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 67
Exhibit 3-22 shows that none of the seven pesticide active ingredients detected in CWS well samples
are reported as used in the vicinity of the sampled CWS wells. Over half of the pesticide active ingredients
included in this analysis were reported as used in the vicinity of CWS wells (Local Area Questionnaire q2).
However, none of the 45 active ingredients reported as used within one half mile of CWS wells sites in the
three years prior to sampling were detected in any sampled CWS well. In contrast, seven of the 37 NFS active
ingredients not reported as used within one half mile of the well in the last three years were detected in at
least one CWS well. These results are contrary to the expectation that the probability of detection in wells
should be greater for pesticides that are used near those wells. Results for storage, disposal, and spillage of
pesticides were inconclusive, because very few pesticides were reported as stored, disposed, or spilled in the
vicinity of CWS well sites. No pesticides that were reported as stored, disposed, or spilled near CWS well sites
were detected at those well sites.
Exhibit 3-22
Summary of Pesticide Detection by Reported Pesticide Use
for Community Water System Wells: Pesticide Level
Questionnaire Item
Farm pesticide use in the
past three years within one
half-mile (LAQ q2).
Non-farm pesticide use on
property in past three years
(Main qB2)
Farm pesticide use on
property in the past five
years (Main qC3)
Number of Pesticides
Detected
Reported as
Used
0
0
0
Not Reported
as Used
7
7
7
Number of Pesticides Not
Detected
Reported as
Used
45
8
6
Not Reported
as Used
30
67
69
Results for rural domestic wells are similar to those for CWS wells. There is little evidence to suggest
that reported pesticide use from NFS questionnaires is related to pesticide detections. Exhibit 3-23 shows that
four of the nine pesticide active ingredients detected in rural domestic well samples were reported as used
within one-half mile of at least one rural domestic well in the three years prior to sampling, and that
approximately half of the NFS active ingredients included in this analysis were reported as used within one
half mile of rural domestic wells in the three years prior to sampling. Results using responses from the second
stage county agent questionnaire were very similar to those presented from the Local Area Questionnaire
presented in Exhibit 3-23. Results for storage, disposal, and spillage of pesticides were inconclusive because
very few pesticides were reported as stored, disposed, or spilled in the vicinity of rural domestic well well sites.
No pesticides that were reported as stored, disposed, or spilled near rural domestic well sites were detected
at those well sites.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 68
Exhibit 3-23
Summary of Pesticide Detections by Reported Pesticide Use
for Rural Domestic Wells: Pesticide Level
Questionnaire Item
Farm pesticide use in the
past three years within one
half-mile (LAQ q2)
Pesticide use on the lawn
in the past three years
(Main qB2)
Pesticide use on the
garden in the past three
years (Main qB3)
Farm pesticide use on
crops on the property in
the past five years (Main
qCT)
Number of Pesticides
Detected
Reported as
Used
4
0
0
3
Not Reported
as Used
5
9
9
6
Number of Pesticides
Not Detected
Reported as
Used
37
5
8
27
Not Reported
as Used
36
68
65
46
The four pesticides that were detected and reported as used were alachlor, atrazine, bentazon, and
DCPA12. The minimum reporting limits (MRLs) for these chemicals are 0.50, 0.12, 0.25, and 0.1013 /zg/L
respectively. These MRLs are lower than the average MRLs for detected and non-detected pesticides which
are 1.23 and 2.09 pg/L respectively (see Section 4.3.2.1)
Results presented at the pesticide level do not fully reflect the extent of the unexpected relationship
between reported pesticide use and pesticide detections because a pesticide is categorized by the first category
into which it falls. Although four pesticides detected in wells were reported as used near rural domestic wells,
they may have been detected more than once, but may have been reported as used at comparatively few of the
well sites at which they were detected. Analysis leading to the results presented in Exhibit 3-24 partially
addresses this problem. Results are presented at the well, rather than the pesticide, level. The most
noticeable difference between the results presented at the well level and the results presented at the pesticide
level are those for rural domestic wells. Whereas four out of nine detected pesticides were reported as used
(i.e., 44.4%), only 0.66% of the 4.14% estimated proportion of wells with detections (i.e., 15.9%) had reported
use of the detected pesticides. The difference in these proportions (44.4% versus 15.9%) is a consequence of
moving from a pesticide level analysis where pesticides were categorized in the first possible category to a well
level analysis. The overall conclusion remains the same, i.e., that the reported pesticide use from NFS
questionnaires and pesticide detections is not as expected.
12 DCPA acid metabolites, the pesticide degradate of DCPA, was the chemical actually detected in rural domestic wells.
13 This is the MRL for DCPA acid metabolites, the degradate of DCPA that was detected in the NFS.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 69
Exhibit 3-24
Summary by Pesticide Detection and Reported Pesticide Use
from the Local Area Questionnaire: Well Level
Survey
cws
Rural Domestic
National Estimate of
Proportion of Wells with
Detections
Reported as
Used
0%
0.66%
Not Reported
as Used
10.01%
3.48%
National Estimate of
Proportion of Wells with
Non-Detections
Reported as
Used
24.42%
54.44%
Not Reported
as Used
65.58%
41.42%
Even at the well level the four categories of detections by use are not mutually exclusive, i.e., often
more than one pesticide is used at a well site, but a well by itself is used only as an indicator for all pesticides
used at the well site. Although 0.66% of wells recorded detections of at least one pesticide that was reported
as used, other pesticides, which were not detected, may also have been reported as used at that well site.
Exhibit 3-25 shows the distribution of the number of pesticides used at CWS and rural domestic well sites
according to Local Area Questionnaire responses14. The average number of pesticides reported as used at
rural domestic well sites is 5.2, and at CWS well sites is 1.4, and the greatest number of pesticide active
ingredients used at a CWS or rural domestic well site is 19.
A further analysis of reported pesticide use by pesticide detections was performed at the well by
pesticide level. That is, each reported use of a pesticide near a well may be regarded as a "well-chemical" and
represents the use and detection status for an individual chemical at a single well. Exhibit 3-26 shows the
estimated proportion of detections by use across well-chemicals for both the CWS and rural domestic well
surveys. The exhibit contains the same four groups as the above exhibits, but now each pesticide active
ingredient is included in each of the four groups in proportion to the number of times it was reported in each
use and detection category. The groups in Exhibit 3-26 are mutually exclusive across wells and pesticide active
ingredients.
The results presented in Exhibit 3-26 confirm the earlier finding that reported pesticide use from NFS
questionnaires is not related to pesticide detections. Of the 0.05% of well-chemicals that are estimated to
occur above NFS minimum reporting limits in rural domestic drinking water wells, only 20% are estimated
to be classified under "reported as used", whereas 80% are estimated to be classified under "not reported as
used". This is contrary to initial expectations that detections would generally occur when pesticides are
reported as used. The situation is more clear for CWS wells. None of the well-chemicals are both detected
and reported as used.
Exhibit 3-26 also indicates the rarity of detecting an individual pesticide, and consequently helps
explain why the NFS was not designed to measure and statistically analyze the presence of individual analytes.
That is, the proportion of well-chemicals that are detected is 0.13% for CWS wells and 0.05% for rural
domestic wells nationally.
14 The Local Area Questionnaire is used because there are substantially more reports of pesticide use from this
questionnaire than from the Main Questionnaires.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 70
Exhibit 3-25
Summary of the Number of Pesticide Active Ingredients
Reported as Used from the Local Area Questionnaires
by Number of Rural Domestic and Community Water System Wells Sampled
Number of Pesticide
Active Ingredients
Reported as Used
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Number of Rural
Domestic Wells
Sampled
279
29
27
30
26
29
30
42
37
47
32
46
22
14
9
8
27
6
10
2
Number of
CWS Wells
Sampled
391
24
17
19
16
15
13
12
8
10
5
2
1
2
0
0
0
2
2
1
Exhibit 3-26
Summary by Pesticide Detection and Reported Pesticide Use
from the Local Area Questionnaire: Well by Chemical Level
Survey
CWS
Rural Domestic
National Estimate of
Proportion Detected
Reported
as Used
0%
0.01%
Not Reported
as Used
0.13%
0.04%
National Estimate of
Proportion Not Detected
Reported
as Used
1.49%
4.38%
Not Reported
as Used
98.38%
95.57%
Note that the proportion of detected well-chemicals in rural domestic wells that are reported as used
is approximately 20% (0.01% + 0.05%), whereas the proportion of non-detected well-chemicals that are
reported as used is approximately 4.4%. This implies that the probability of detecting a pesticide is greater
if the pesticide is reported as used than if it is not reported as used. Due to the small number of detections
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 71
a formal statistical test comparing these two proportions does not reveal a statistical difference. The
corresponding proportions for CWS wells (i.e., 0% and 1.5%) do not allow concurrence with the conclusion
for rural domestic wells.
The data used for the proportions presented in Exhibits 3-24 and 3-26 come from the Local Area
Questionnaire items relating to farm pesticide use. The results indicate that, in general, pesticides were not
reported as used and were not detected. There is insufficient data to be able to determine if there is a
statistically significant relationship between reported pesticide use and pesticide detections, mainly because
there are too few detections coupled with too few reports of pesticide use. However, the data suggest that
reported pesticide use is not associated with pesticide detections, which is contrary to initital expectations.
In particular, it is surprising to find that pesticides that were detected were most often not reported as used.
Detections of NFS pesticides in sampled wells for which those pesticides are not reported as used could be
explained by a number of factors, including the comparatively short period of time (up to five years) for which
pesticide use information was obtained, the location of the recharge area with respect to the sampled wells,
and measurement error in the responses.
3.5 Pesticide Use Data
This section reports results for analyses of associations between pesticide and nitrate detections and
estimates of herbicide use for agricultural, urban and golf course purposes compiled by Resources for the
Future, Inc. (RFF). The data have been gathered from a variety of sources including mail surveys conducted
by RFF, national censuses, and national and state surveys.
The database on agricultural use of pesticides contains information on 96 active ingredients and 84
crops. This information was collected through a mail survey of Extension Service weed scientists, and was
supplemented with other sources of state-specific herbicide use such as a 1984 survey of weed scientists
conducted by the Agricultural Research Service, individual state surveys and reports, and published surveys
for specific crops.15 When information for states was not available from any of these sources, imputation
was performed by assuming that the missing states's information is identical to that of a neighboring state.
The databases on urban and golf course use of pesticides were constructed using data from 1982
national surveys conducted by EPA's Office of Pesticide Programs. The urban applicators survey included
three groups of applicators; tree, lawn, and structural. Further information used to construct the golf course
database came from the 1980 Census of Housing and the National Golf Foundation.16
3.5.1 Pesticide Application for Agricultural Purposes
The agricultural pesticide use database was compiled at the state level for the 48 contiguous states.
The data consist of the number of harvested acres in the state, the number of treated acres, and the quantity,
measured in pounds of active ingredient, used within the state. Other factors are calculated from this
information, such as the percent of harvested acres that are treated and the rate of application per treated
acre. Harvested crop acreage for 84 crops was estimated from the 1987 Census of Agriculture.
RFF constructed a database at the county level from the state-level information by proportionately
assigning the harvested acres of specific crops within a state according to the counties' relative sizes. Counties
in which the crops are not grown were omitted from this process and were assigned a zero value. The
procedure assumes the same rate of application of pesticides to individual crops throughout the state and
15 Gianessi, L.P., and Puffer, C.A., The Use of Herbicides in the United States. Resources for the Future, Inc.,
Washington, D.C., paper presented at the National Research Conference, "Pesticides in the Next Decade: The Challenges
Ahead," sponsored by Virginia Water Resources Research Center, Richmond, Virginia, November 9, 1990.
16 Gianessi, L.P., and Puffer, C.A., Estimation of County Pesticide Use on Golf Courses and by Urban Applicators.
Resources for the Future, Inc., Washington, B.C. report submitted to U.S. EPA, January 1991.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 72
assumes that all counties within a state have the same percent of individual crops grown (except for counties
in which the crops are not grown). For example, atrazine is assumed to be applied to sorghum at a rate of
two pounds per acre in all counties in Alabama that grow sorghum, and if sorghum is grown in 10 percent of
the state then each county in that state is assumed to have 10 percent sorghum (adjusted for counties with no
sorghum cropping).
The analyses reported in this section use county-level information that is associated with the sampled
wells by well location. Consequently, there are at least two levels of averaging of data that could affect the
validity of the results. First, a well is associated with a county. Even though counties may have extensive
pesticides use, the pesticides may not be applied near the sampled wells or the wells' recharge areas. Second,
county-level information is constructed through state-level averaging in proportion to the counties' relative
sizes. Large counties with minimal pesticide use or small counties with considerable pesticide use are not
accounted for through the averaging process. Finally, measurement error associated with collection of the
RFF data is not taken into account in the analyses presented in this section. Because NFS questionnaires were
designed specifically to obtain information about pesticide use in the vicinity of the well, they should, given
reliable responses, provide more locally precise measures of pesticide use at the well sites.
The analyses presented in this section use county-level data to assess relationships between pesticide
use, measured through the RFF database, and pesticide detections. The measure of pesticide use required for
analysis is the total pounds of pesticide applied per county acre. County size data are not available from the
RFF database. To perform the analyses county size data were obtained from the County and City Data Book
using 1986 data. The data were provided in units of square miles, and the appropriate multiplication was
performed to transform land area in square miles to acres. Total pesticide use within a county was calculated
by summing, for all crops, the rate of application measured in pounds per county acre.17
The only pesticide product that was detected sufficiently often for reasonable statistical analysis was
DCPA acid metabolites. RFF data were available for DCPA, the parent compound. An analysis also was
performed for aldicarb, which was not detected but is included in the RFF database, to compare rates of
pesticide use in counties included in the NFS and counties not included in the NFS to help determine if the
NFS sampled wells in comparatively high use areas.
The distributions of the rate of application variables for the different pesticides used in the analyses
had many low application rates with a few cases of very high rates. A logarithmic transformation was used
to obtain more symmetric distributions. This transformation was used for all regression models reported in
this section.
Analysis of the relationship between agricultural DCPA use and detections of DCPA acid metabolites
did not show sufficient evidence to suggest that the rate of DCPA use in agriculture (DCPA also has major
non-agricultural uses) predicts DCPA acid metabolites detections. A similar analysis for the rural domestic
well survey yielded qualitatively similar results, although this analysis may have been affected by the small
number of DCPA acid metabolite detections in rural domestic wells. Selected means and standard errors are
presented in Exhibit 3-27.
17 Pounds per county acre was determined by dividing the pounds of active ingredients variable from the RFF database
by the county size. Given that county information in the RFF database is already prorated by county size and a rate of
application variable is supplied, the need for a county size component is not immediately apparent. However, the rate of
application variable supplied by RFF is assumed to be constant for a given state and crop. An individual pesticide is generally
applied at different rates for different crops, hence the need for county size to provide a more appropriate measure of rate
of total pesticide use throughout the county.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 73
Exhibit 3-27
Summary Statistics for Rate of DCPA Use In Agricultural Areas
Survey
cws
Rural
Domestic
Status
Detected
Non-Detected
Detected
Non-Detected
Mean Rate of
Use
(Ibs/acre)
0.011
0.006
0.002
0.002
Standard
Error
0.005
0.001
0.001
< 0.0005
Mean
Logarithm of
Rate of Use
-6.72
-7.81
-7.58
-8.42
Standard
Error
0.62
0.22
0.53
0.34
Further analyses were considered to determine whether or not selected NFS chemicals are used more
heavily in counties included in the NFS study than in other counties. This analysis could shed light on possible
explanations for the pattern of which pesticides were detected by the survey. For example, NFS did not detect
aldicarb, yet data from other sources such as EPA's Pesticide in Groundwater Database indicate that in certain
areas, contamination of drinking water wells by aldicarb is of concern. The RFF database provides information
that could be used to indicate if the NFS sampled in comparatively high pesticide use counties.
Before completion of this analysis, the data in the RFF database was compared to information from
the manufacturer to ensure that the distribution of pesticide sales is accurately represented by the RFF
database. For one chemical with relatively lower sales, this comparison uncovered significant differences
between the RFF database and the manufacturer's data. For one chemical with relatively higher sales, the
publicly available data adequately represented the confidential data held by the manufacturer.
In particular, the Phase II analysis found significant differences between the manufacturer's
confidential database and the publicly accessible data on aldicarb in the RFF database. The targeting of areas
of aldicarb use was evaluated using confidential sales distribution data supplied by the registrant of aldicarb.
These data are counties' totals of aldicarb, (in pounds of active ingredient) sold to distributors in 1988. The
data indicate that 6 of the 90 counties included in the rural domestic well survey and 80 of he 299 counties
included in the CWS well survey have aldicarb distribution. In contrast, data from RFF indicate that 66 of
the 90 counties in the rural domestic well survey and 267 of the 399 counties in the CWS survey have aldicarb
use. Similar differences between RFF data and manufacturer's data were also found for alachlor in a review
of the NAWWS data. These differences may be due to several factors: Distribution points within counties
may provide aldicarb to adjacent counties, and the RFF database is constructed, in part, by assuming that
counties have aldicarb use if crops for which aldicarb is recommenced are grown.
The pesticide sales distribution data for the relatively high volume chemical atrazine in the RFF
database appears to match the manufacturer's data fairly well.
Because of the uneven reliability of information in the public domain on the distribution of pesticides
within states, the NFS analysis could not determine whether the counties sampled by NFS were nationally
representative of the use of individual pesticides.
3.5.2 Pesticide Applications in Urban Areas
EPA's Office of Pesticide Programs commissioned a national survey of urban commercial tree, lawn,
and structural pest control applicators in 1982. The survey reports the aggregate use within each EPA Region
of ten selected active ingredients by urban pesticide applicators. A further component of the RFF urban
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 74
applicators database consists of estimates of the number of single family housing units taken from the 1980
Census of Housing.18
To construct a county-level database, the aggregate use estimate for each active ingredient for each
EPA Region was divided by the single family housing unit estimate for the region to provide a rate of
application measured in units of pounds of active ingredient per housing unit. These estimates were then
prorated to counties in proportion to the number of housing units in the county. This procedure uses
averaging on a greater scale than is used to construct the RFF county agricultural database (since there are
10 EPA Regions nationally compared to 48 contiguous states).
Analysis at the well level involves associating a well with a county by well location. The variable
required for analysis measures the total pesticide use by urban applicators within a county, adjusted by the
county size.
DCPA, the parent compound of the DCPA acid metabolites detected in the Survey, was the only
pesticide contained in the RFF urban applicators database that recorded sufficient detections in the NPS for
reasonable analysis. A logarithmic transformation of the DCPA use data was used in the analyses because the
distribution of the use data was heavily skewed. Exhibit 3-28 provides details of the estimated models of the
relationship between the logarithm of DCPA use and the probability of detecting DCPA acid metabolites.
Exhibit 3-28
Estimated Models of the Relationship Between DCPA Acid Metabolites Detections
and DCPA Use by Urban Application
Survey
cws
Independent
Variable
Cn(Rate of DCPA
Use) (pounds per
county acre)
Intercept
Estimate
2.213
Standard
Error
1.086
Beta
Coefficient
Estimate
0.729
Standard
Error
0.180
Significance
Level
(p-value)
< 0.0005
Exhibit 3-29 provides summary statistics of DCPA use in both CWS and rural domestic wells. These exhibits
present strong evidence of a positive relationship between the rate of DCPA use for urban application and
the probability of detecting DCPA acid metabolites in both CWS and rural domestic wells.
18 Gianessi, L.P., and Puffer, C.A., The Use of Herbicides in the United States, Resources for the Future, Inc.,
Washington, D.C.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 75
Exhibit 3-29
Summary Statistics for Rate of DCPA Use by Urban Application
Survey
cws
Rural
Domestic
Weils In Which
DCPA was
Detected or Not
Detected
Mean Rate
of Use
(Ibs/acre)
Detected 0.010
Non-Detected || 0.002
Detected
Non-Detected
0.005
0.002
Standard
Error
0.002
< 0.0005
0.001
< 0.0005
Mean
Logarithm of
Rate of Use
-5.67
-8.02
-5.86
-7.49
Standard
Error
0.39
0.12
0.33
0.17
3.5.3 Pesticide Applications on Golf Courses
Data pertaining to use of selected pesticide active ingredients on golf courses was obtained by RFF
from a national survey commissioned by EPA's Office of Pesticide Programs in 1982. This survey reports the
aggregate use of pesticides on golf courses by EPA Region. Information about golf courses in each state was
compiled from the Statistical Profile of Golf in the United States published by the National Golf Foundation.
An assumption of 7.9 acres per golf hole was used to estimate the number of golf courses acres within a
state.19
To construct a county-level database the aggregated use estimate was first divided by the golf course
acreage estimate for each EPA Region. County-level estimates were then created by prorating the EPA
Region totals to counties in proportion to county golf course acreage. The variable required for analysis
measures the total pesticide use on golf courses within a county, adjusted by the county size. Analysis was
carried out by associating NPS wells with counties by well location.
DCPA, the parent compound of the DCPA acid metabolites detected in the Survey, was the only
pesticide that is contained in the RFF golf course database that recorded sufficient detections in the NPS for
reasonable analysis. A logarithmic transformation of the DCPA use data was used in the analyses because the
distribution of the use data was strongly skewed. Exhibit 3-30 provides details of the estimated models of the
relationship between the logarithm of the rate of DCPA use and the probability of detecting DCPA acid
metabolites. Exhibit 3-31 provides summary statistics of DCPA use in both CWS and rural domestic wells.
These exhibits present strong evidence of a positive relationship between the rate of DCPA use on golf courses
and the probability of detecting DCPA acid metabolites in both CWS and rural domestic wells.
19 Gianessi, L.P., and Puffer, C.A., Estimation of County Pesticide Use on Golf Courses and by Urban Applicators.
Resources for the Future, Inc., Washington, B.C.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 76
Exhibit 3-30
Estimated Model of the Relationship Between DCPA Acid Metabolites Detections
and DCPA Use on Golf Courses
Survey
cws
Independent
Variable
0n(Rate of DCPA
Use) (pounds per
county acre)
Intercept
Estimate
0.817
Standard
Error
0.685
Beta
Coefficient
Estimate
0.590
Significant
Error
0.126
Significance
Level
(p-value)
< 0.0005
Exhibit 3-31
Summary Statistics for Rate of DCPA Use on Golf Courses
Survey
CWS
Rural
Domestic
Wells in Which
DCPA was
Detected or
Not Detected
Detected
Non-Detected
Detected
Non-Detected
Mean Rate
of Use
(Ibs/acre)
0.018
0.005
0.014
0.004
Standard
Error
0.005
0.001
0.005
0.001
Mean
Logarithm of
Rate of Use
-5.04
-6.85
-5.04
-6.54
Standard
Error
0.31
0.10
0.43
0.16
3.6 Fertilizer Sales Data
This section describes studies of the relationship between nitrate and pesticide detections and county-
level data on nitrogen fertilizer sales, agricultural crop production and animal stock levels obtained by EPA
(the NFERC/EPA Fertilizer Sales Data).20 The data cover the six "fertilizer years" between July 1, 1984 and
June 30, 1990. Each "fertilizer year" corresponds to the period over which nutrients are most commonly
applied in the production of a crop. These data were prepared at the Division of Resources Management at
West Virginia University in conjunction with the National Fertilizer and Environmental Research Center
(NFERC) at the Tennessee Valley Authority in Alabama. The Fertilizer Sales Data are maintained as two
separate databases. The first concerns nitrogen fertilizer sales; the second concerns agricultural crop
production and livestock production. The data were used to test the hypotheses that higher occurrences of
nitrate and pesticide detections occur in areas with high nitrogen sales, high levels of crop production, and/or
high levels of livestock production.
3.6.1 Nitrogen Fertilizer Sales
The analyses reported in this section use estimated county-level nitrogen fertilizer sales information
that is associated with the sampled wells by well location. The county-level information was constructed
20 County-Level Fertilizer Sales Data, United States Environmental Protection Agency, Policy Planning and Evaluation
(PM-221), September 1990.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 77
through state-level averaging in proportion to the counties' relative expenditures on fertilizers. That is,
expenditures are used as a surrogate for sales. The measure of fertilizer sales required for analysis is the total
tons of nitrogen fertilizer sales per county acre. County sizes are not available from the NFERC database.
To perform the analyses, data on the number of acres in each county were obtained from the County and City
Data Book using 1986 data.21 These data were provided in units of square miles, and the appropriate
multiplication was performed to transform land area in square miles to acres.
The NFERC data reflect total sales of fertilizer without regard to where or when the fertilizer is used.
Sales figures can only be a proxy for use. Also, fertilizer may be bought in one county and used in another;
similarly, it may be purchased in one season and used in a subsequent season. Although manure is a
significant contributor to fertilizer use, it is not measured with the NFERC data. Finally, even though
counties may have large nitrogen fertilizer sales, the fertilizer may not be applied near the sampled wells or
their recharge areas.
The agricultural crop production and animal stock levels database (see next section) uses 1987 Census
of Agriculture data. Consequently, 1987 nitrogen fertilizer sales data were used for consistency of analysis.
This has the added advantage of using data that precedes the NFS survey, so that analysis of the NFS and
NFERC data is not confounded by the time of well sampling. The annual NFERC data are, however, highly
correlated across years. Correlation coefficients for total nitrogen fertilizer sales exceed 0.99 for all pairs of
annual variables. Correlations across years for the six groups of nitrogen fertilizer were also high.
Consequently, use of 1987 annual nitrogen fertilizer sales data is likely to produce qualitatively similar results
to data from other years. Similarly, correlations between annual and fall sales data were very high (usually
exceeding 0.9).
Seven NFERC variables (logarithmically transformed) were ultimately evaluated for correlation with
NFS pesticide or nitrate detections, and nitrate concentrations: 1987 nitrogen fertilizer sales data for all six
nitrogen fertilizer groups (ammonium nitrate, anhydrous ammonia, miscellaneous forms, nitrogen in solution,
urea, and total nitrogen), and commercial fertilizer use, measured in acres on which used, from the 1987
Census of Agriculture. These variables were adjusted for county size.
None of the analyses by nitrate detections provided evidence of a relationship between NFERC
variables and the probability of nitrate detections. Nitrate concentrations (logarithmically transformed),
however, were found to be positively related to all the NFERC variables. Exhibit 3-32 presents results for
total nitrogen fertilizer sales for both the CWS and rural domestic well surveys. Tons of nitrogen sold per
county acre was found to be associated with nitrate concentrations for both CWS and rural domestic wells.
The remaining NFERC variables are highly correlated with total nitrogen fertilizer sales, and provide
qualitatively similar results.
One NFERC variable provided evidence of a relationship with pesticide detections in each of the CWS
and rural domestic well surveys. For the CWS well survey this variable was ammonium nitrate sales per county
acre (p-value = 0.010), but the effect is in the opposite direction to that expected. For the rural domestic well
survey, the significant predictor variable corresponded to miscellaneous nitrogen fertilizer sales per acre (p-
value = 0.012). The overall case for a relationship between pesticide detections and nitrogen fertilizer sales
is weak, since only one of fourteen analyses yielded a significant result in the direction of the hypothesis, and
the NFERC variables are highly correlated.
21 U.S. Department of Commerce, Bureau of the Census, County and City Data Book, 1988. All areas are defined as
of October 18, 1986.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 78
Exhibit 3-32
Estimated Models for the Relationship of Nitrate Concentrations*
By Total Tons of Nitrogen Sold per County Acre
Survey
cws
Rural Domestic
Estimated
Intercept
-0.832
-1.526
Standard
Error
0.214
0.572
Estimated
Independent
Variable
fti (Tons/Acre)
0.144
0.261
Standard
Error
0.029
0.084
Tons/Acre
Coefficient
Significance
Level
(p-value)
< 0.0005
0.002
Nitrate concentrations are logarithmically transformed.
In summary, there is strong evidence of a relationship between nitrate concentrations and nitrogen
fertilizer sales data. This relationship is not supported by analyses of nitrate detections. Although nitrate
detections offer a comparison of nitrate contamination above and below the NFS minimum reporting limit
of 0.15 mg/L, rather than a continuous measure above 0.15 mg/L, the results for these two sets of analyses
should be expected to be qualitatively similar. The overall evidence to support a relationship between
pesticide use and nitrogen fertilizer sales per county acre is too weak to be used to provide predictive models.
3.6.2 Agricultural Crop Production and Livestock Production
The second part of the NFERC/EPA database consists of county-level information gathered from the
1987 Census of Agriculture. The data are organized into three categories: general crop acreage and crop and
livestock value; specific crop acreage and crop production; and animal counts.22
Though more than 50 variables are available from the Census of Agriculture data, many of these
variables are highly correlated. The variables fall into five main categories:
• Crop farming;
• Fertilizer use on cropland;
• Livestock production;
• Fertilizer use on range or pasture land; and
• Miscellaneous variables (such as vegetable and orchard farming).
Five variables (logarithmically transformed) were chosen for analysis:
CROPVAL
CROPFERT
Market value of crops, including greenhouses and market gardens
in $1,000;
Acres of cropland fertilized, excluding range and pastureland;
22 The crops reported in the database include corn for seed and silage, soybeans, wheat, barley, oats, sorghum, cotton,
beans, rice, sunflower, vegetables, sugar beets, potatoes, peanuts, sugar cane, hay, and tobacco. Combined orchards and
vegetable acreage variables are also available. Animal inventories are included for livestock consisting of chicken, beef and
milk cows, sheep, horses, and hogs.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 79
PASTFERT
LVSTVAL
BEEFCOW
Acres of range and pastureland fertilized;
Market value of livestock in $1,000; and
Number of beefcows.
Several variables were highly correlated and yielded qualitatively similar predictive capabilities.
CROPVAL, for example, was highly correlated with other total measures of crop farming such as harvested
cropland and market value of agricultural products, and with commonly grown crops such as corn, wheat,
soybeans, and hay. Three variables related to fertilizer use on crop farming land were highly correlated.
Variables related to animal farming, including the total live stock value and variables measuring the number
of milk cows, horses, hogs, chickens, and sheep were all positively correlated. The BEEFCOW variables,
however, was found to be negatively correlated with the other animal farming variables. The livestock value
variable (LVSTKVAL) was also found to be correlated with crop farming variables.
Exhibits 3-33 and 3-34 present results of analyses of the relationship between the five Census of
Agriculture variables and detections in wells. Exhibit 3-35 presents results of analysis of the relationship
between Census of Agriculture variables and nitrate concentrations.
Exhibit 3-33
Estimated Models for Selected Census of Agriculture Variables
by Nitrate Detections
Survey
cws
cws
cws
Independent
Variable
BEEFCOW
LVSTKVAL
PASTFERT
Estimated
Intercept
0.51
-0.07
0.32
Standard
Error
0.15
0.12
0.13
Estimated
Independent
Variable
Coefficient
-20.51
2.14
-26.31
Standard
Error
6.77
0.92
7.82
Descriptive
Significance
Level
(p-value)
0.003
0.020
0.001
Exhibit 3-34
Estimated Models for Selected Census of Agriculture Variables by
Pesticide Detections
Survey
CWS
CWS
Rural
Domestic
Rural
Domestic
Independent
Variable
PASTFERT
BEEFCOW
CROPVAL
BEEFCOW
Estimated
Intercept
-1.74
-1.59
-1.14
-5.63
Standard
Error
0.21
0.26
0.60
1.29
Estimated
Independent
Variable
Coefficient
-74.12
-41.98
0.58
-0.50
Standard
Error
27.50
18.11
0.19
0.25
Descriptive
Significance
Level
(p-value)
0.007
0.021
0.002
0.043
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 80
Exhibit 3-35
Estimated Models for Selected Census of Agriculture Variables by
Nitrate Concentrations Above 0.15 mg/L*
Survey
cws
cws
cws
Rural
Domestic
Rural
Domestic
Rural
Domestic
Independent
Variable
CROPVAL
CROPFERT
LVSTKVAL
CROPVAL
CROPFERT
LVSTKVAL
Estimated
Intercept
0.06
-0.06
0.10
1.40
1.06
1.15
Standard
Error
Estimated
Independent
Variable
Coefficient
0.09 I 3.14
0.10
0.09
0.29
0.31
0.40
2.67
2.03
0.27
0.26
0.23
Standard
Error
0.74
0.56
0.50
0.06
0.09
0.10
Descriptive
Significance
Level
(p-value)
< 0.0005
< 0.0005
< 0.0005
< 0.0005
0.005
0.022
Nitrate concentrations are logarithmically transformed.
In general, crop fanning variables are positively associated with nitrate concentrations, though there
is insufficient evidence to support a similar hypothesis of a relationship with nitrate detections. There is some
evidence of a relationship between crop farming variables and pesticide detections in rural domestic wells, but
that evidence is not supported in CWS wells or by results for other crop farming variables. Fertilizer use on
crop land is associated with nitrate concentrations for CWS wells and rural domestic wells.
Analysis of other variables that measure vegetable and orchard farming indicate positive correlations
with nitrate concentrations. These results are in line with results for crop fanning variables.
Animal inventory variables fall into two categories. The market value of livestock variable provides
qualitatively similar results to the crop farming variables, whereas the beef cow variable yields opposite effects.
Analysis of the fertilizer use on range and pasture land variable yields similar results to those for the beef cow
data. That is, fertilizer use on range and pasture land is negatively associated with pesticide and nitrate
detections.
3.7 Well Water Characteristics: Temperature, pH, and Conductivity
This section presents results of the analyses of the association of detections with well water
temperature, pH, or electrical conductivity (measured as the concentration of total dissolved solids) as
potential indicators of the presence of pesticides or nitrate in well water. Two hypotheses were tested: (1)
that pesticide persistence, and thus the likelihood of detection, increases at lower temperatures and low pH
conditions; and (2) that nitrate persistence, and thus the likelihood of detection, increases at lower
temperatures and in both low and high pH conditions.
Results are presented first for CWS wells and second for rural domestic wells. Logistic regression
models were developed to examine the relationship of water temperature, pH, and conductivity of pesticide
and nitrate detections. The statistical analyses for electrical conductivity involved a logarithmic transformation
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 81
of the conductivity data. Although the estimated coefficients of the regression models are included, their
interpretation must account for the transformation of the conductivity data.
A data set consisting of the initial, stabilized, and final readings of well water temperature, pH, and
conductivity was derived from the records of well purging conducted at the time of sampling. These were used
as indicators of the availability of fresh ground water for sampling. The instruments used for these
measurements were calibrated frequently to minimize calibration errors. However, the well purging data set
was not subject to the same level of scrutiny, on the basis of established quality assurance procedures, as most
of the NFS data. The initial, stabilized, and final readings were compared and aggregated to derive the data
used for analysis. The stabilized reading was treated as the primary reading for analysis.23
3.7.1 Well Water Characteristics for CWS Wells
The results for CWS wells presented in Exhibits 3-36 to 3-38 indicate that both pesticide and nitrate
detections in CWS wells are negatively associated with well water temperature and pH levels. The pesticide
results were presented for "any pesticide" rather than any smaller group of pesticides because there were too
few detections in the smaller groups to satisfy the statistical assumptions of the analysis, though the results
were qualitatively similar to those for any pesticide group. Consistent with the prior hypotheses, these findings
imply that:
Exhibit 3-36
Association of Detections with Well Water Temperature
for Community Water System Wells
Analyte
Detection
Group
Any Pesticide
Nitrate
Estimated
Intercept
-0.466
0.824
Standard
Error
0.458
0.312
Estimated
Temperature
Coefficient
-0.113
-0.047
Standard
Error
0.030
0.018
Temperature
Coefficient
Significance
Level
(p-value)
< 0.0005
0.010
Based on logistic regression model of detections.
23 Measured well water temperatures ranged from a low of 1 degree Celsius to a high of 37 degrees celsius, with a mean
temperature of approximately 17 degrees. Well water pH ranged from 4.7 to 9.6 with a mean of approximately 7.4.
Conductivity values ranged from 0 to 2500 ppm Total Dissolved Solids (IDS) with a mean of approximately 230 ppm IDS.
Data were treated as missing when the three readings were too disparate or when the readings suggested nonsensical results
(e.g., one reading of a pH of 0.8 was reported). The statistical results reflect those data ranges, and may not be appropriate
for data outside those ranges.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 82
Exhibit 3-37
Association of Detections with Well Water pH
for Community Water System Wells
Analyte
Detection
Group
Any Pesticide
Nitrate
Estimated
Intercept
0.504
2.994
Standard
Error
0.888
1.216
Estimated
PH
Coefficient
-0.359
-0.392
Standard
Error
0.120
0.161
pH
Coefficient
Significance
Level
(p-value)
0.003
0.015
Based on logistic regression model of detections.
• the probability of detecting at least one pesticide or nitrate is greater in wells with
low water temperature; and
• the probability of detecting at least one pesticide or nitrate is greater in wells with
low water pH (i.e., high acidity).
Conductivity was not found to be associated with pesticide detections, but was found to be associated
with nitrate detections. That is, the probability of detecting nitrate is greater in wells with low electrical
conductivity. In contrast, however, analysis of nitrate concentration analysis of nitrate concentrations showed
that higher electrical conductivity corresponds to greater concentrations. These results are presented in
Exhibit 3-38.
Exhibit 3-38
Association of Detections of Nitrate and Nitrate Concentrations
with Well Water Conductivity* for Community Water System Wells
Dependent
Variable
Nitrate
fn (Nitrate
concentrations)
Estimated
Intercept
2.083
-1.103
Standard
Error
0.635
0.390
Estimated
Cn(Conductfvtty)
Coefficient
-0.401
0.274
Standard
Error
0.122
0.081
Cn(Conductivity)
Coefficient
Significance
Level
(p-value)
0.001
0.001
Well water electrical conductivity and nitrate concentrations are logarithmically transformed.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 83
3.7.2 Well Water Characteristics for Rural Domestic Wells
The results for rural domestic wells presented in Exhibit 3-39 indicate that nitrate detections in rural
domestic wells are negatively associated with pH levels. Results for temperature and conductivity for rural
domestic wells were qualitatively similar to results for CWS wells (see Appendix A) but did not satisfy the 0.05
significance criterion. The significant findings imply that the probability of detecting nitrate is greater in wells
with low water pH (i.e., high acidity).
Exhibit 3-39
Association of Well Water pH* with Detections of
Nitrate for Rural Domestic Wells
Analyte
Detection
Group
Nitrate
Estimated
Intercept
7.964
Standard
Error
1.551
Estimated
pH
Coefficient
-1.043
Standard
Error
0.204
PH
Coefficient
Significance
Level
(p-value)
< 0.0005
Well water pH has been logistically transformed.
In summary, detection of pesticides and nitrate in drinking water wells are found to be related to some
characteristics of well water. These relationships are generally stronger for the CWS well survey than for the
rural domestic well survey, though the results are generally in the same direction. Results from both surveys
indicate negative correlations of well water temperature, pH, and conductivity with the probability of pesticide
and nitrate detections in wells. However, the results suggest that there is a positive correlation between
conductivity and nitrate concentration.
3.8 Pesticide Chemical Characteristics
This section reports the results of an examination of the relationship between pesticide detections and
soil half-life or soil adsorption partition coefficients. Statistical analyses were conducted to determine if
pesticide detections could be related to the soil half-life or soil adsorption coefficient (organic carbon
coefficient or
Degradation half-life is defined as the time required for half the concentration of a chemical to
degrade. The principal mechanism of degradation of pesticides in soil is microbial degradation. The process
of photolysis (degradation in sunlight) and hydrolysis (degradation in water) also can be important for certain
pesticides in certain conditions. The rate of degradation for a pesticide is not only an inherent characteristic,
it is also very dependent on site-specific conditions. Higher soil temperatures and moisture are more
conducive to microbial degradation; exposure to sunlight is critical to photolysis. The NFS Phase II analysis
could not account for such site-specific conditions.
The organic carbon partition coefficient K^ characterizes the mobility of pesticides in soils. It is
defined as the ratio of the concentration of pesticide sorbed to organic carbon in soil divided by the
concentration in solution at equilibrium. It is a property known for most pesticides and is generated in
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 84
laboratory conditions.24 The mobility of a pesticide is not only an inherent property, it is also a function
of site-specific conditions. The higher the K^ the more strongly adsorbed the pesticide is to soil, organic
matter, and sediment. Pesticides were originally selected to be included in the Survey based in part on
sorption and degradation properties.25
Both soil half-life and K^ information is available for 64 of the 126 pesticides included in the NFS.
Soil half-life information alone is available for one more pesticide, as is K^ information alone.
Three statistical analyses were performed to study the relationship between soil half-life, K^, and
detected chemicals.
• Exploratory analyses, including means, standard errors, stem and leaf plots, and box
plots;
• Analysis of variance or t-tests to determine if detected pesticides differed from
nondetected pesticides in terms of soil half-life or K^; and
Logistic regression to determine if soil half-life or K^ predicted detected chemicals.
The data for these analyses pertain only to those pesticides for which soil half-life or K^ data are
readily available. For each study, these pesticides were divided into two groups corresponding to pesticides
that were detected or not detected in the Survey. The statistical analyses were performed on the 65 soil half-
life and the 65 K^ observations, not NFS samples of wells. The variable of interest is "detection of pesticide
in the NPS" rather than "detection of pesticide in a well." Accordingly, the analysis does not rely on the NFS
design and was performed unweighted.
Because the variable of interest is pesticides detected in the NPS, data were collapsed across both the
CWS and rural domestic well survey. The mean soil half-life and mean K^ and the standard errors of the two
groups of pesticides (detected and not detected) are provided in Exhibit 3-40. This exhibit also includes the
means and their standard errors for the logarithmically transformed data. Exploratory analysis of both the soil
half-life and K^ data indicated that the distributions are heavily skewed to the right. Accordingly, results of
t-test and logistic regression analyses results are provided only for the transformed data.
Unlike most analyses presented in other sections of this report, analyses involving half-life and
are necessarily chemical-specific rather than well-specific. The NPS was designed to investigate the presence
of pesticides and nitrate in wells, and the survey was designed to control for potential confounding factors.
Analyses of half-life and K^ were not the focus of the NPS and consequently are subject to possible
uncontrolled confounding factors. Two factors in particular - individual pesticide use and detection limits -
- may affect the conclusions concerning whether soil half-life or K^ differ for the two analyte groups. If an
active ingredient was not used in the location of any wells in the NPS, then that analyte would not have been
detected, and if the analyte is present at concentrations that are less than the minimum reporting limit (MRL)
it would not have been detected. Atrazine and DCPA acid metabolites were the most commonly detected
pesticide and pesticide degradate in the NPS, but both are known to be widely used, and their MRLs were
relatively low. Other active ingredients, such as Bromacil, are not widely used, or their MRLs are relatively
high, so regardless of their soil half-life or K^ they still are unlikely to have been detected. Statistically
significant results presented in this Section should be regarded as preliminary when considering whether they
are applicable beyond the NPS study.
24 The KO,. is calculated by measuring the ratio of sorbed to solution pesticides concentrations after equilibrium of a
pesticide in a water/soil slurry, and then dividing by the weight fraction of organic carbon present in the soil.
25 See Chapter 4.0 of the NPS Phase I Report for more information on analyte selection.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 85
Exhibit 3-40
Means and Standard Errors of Soil Half-Life and
K for Detected Analytes and Non-Detected Analytes
Detected Analytes8
Nondetected Analytesb
Mean
Standard Error
Mean
Standard Error
Soil
half-life
220
86
54
10
^(half-
life)
4.4
0.5
3.4
0.2
KOC
1,340
1,160
6,730
3,500
Cn(Koc)
4.5
0.7
5.7
0.3
For two analytes in this group the soil half-life was provided as a range. The lowest value of the ranges was
used to provide maximum conservatism (least significance) in the testing and regression procedures. For two
analytes in this group the K^. was provided as a range. The lowest value of the range was used to provide the
most significant results for K.x.
For one analyte in this group the soil half-life was provided as a range. The highest value of the range was
used to provide maximum conservatism in the testing and regression procedures. Also for two analytes in this
group the K^. was provided as a range. The highest value of the ranges was used to provide the most
significant results for K^..
Pesticides for which quantification of the concentration detected was not possible are not included
in the analyses presented in this section. The Phase I Report provides a complete listing of pesticides included
in the NPS, including whether or not concentrations of the pesticides were quantifiable.
For each study, the pesticides were divided into two groups corresponding to pesticides that were
detected and pesticides that were not detected in the Survey. For both soil half-life and Koc there are 12
chemicals in the former group and 53 in the latter group.
The t-test provides a measure of the differences between the two groups of analytes in terms of soil
half-life and K^,. A summary of the results is provided in Exhibit 3-41. Soil half-life tends to be higher for
detected analytes than for nondetected analytes (as indicated by the descriptive significance level or p-value
of 0.02), whereas Koc is not significantly different for the two groups of chemicals (p-value = 0.22).
Exhibit 3-41
Soil Half-Life and Knr Effects for
UG
Detected Analytes and Non-Detected Analytes
t-statistic
p-value
fti (half-life)
2.51
0.01
NKOC)
1.68
0.10
The list of chemicals used in these analyses and their corresponding soil half-life and K^ values are
presented in Exhibit 3-42 for detected NPS chemicals and Exhibit 3-43 for non-detected chemicals. Other
sources may provide different values but values used here are expected to be representative for the chemicals
listed.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 86
Exhibit 3-42
Soil Half-Life and Koc for Detected Pesticides
Detected Pesticide
Alachlor
Atrazine
Bentazon
DCPA and acid metabolites8
Dibromochloropropane (DCBP)
Dinoseb
Ethylene Dibromide (EDB)
Ethylene thiourea (ETU)
Hexachlorobenzene
Lindaneb
Prometon
Simazine
Soil Half-Life
(Days)
15
60
20
365
180
20
28-180
7
990-2080
400
500
80
KOC
170
70
21-34
4-90
129
124
1
50
14100
1081
200
103
a The minimum soil half-life for DCPA acid metabolites is 365 days.
b The NFS detected both gamma-HCH and beta-HCH. However, soil half-life and
K^. information is available for Lindane only, without distinguishing between
gamma-HCH and beta-HCH.
c Nitrate is not included in this table or in the reported analysis because it is not an
organic compound. Nitrate has a K^ of zero and no half-life value.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 87
Exhibit 3-43
Soil Half-Life and Koc for Pesticides
that Were Not Detected
Detected Pesticide
Aldicarb
Aldicarb Sulfone
Ametryn
Butylate
Carbaryl
Carbofuran
Carboxin
Chlordane8
Chlorneb
Chlorothalonil
Chlorpropham
Cyanazine
Cycloate
2,4-D
2,4-DB
DCPA
Dicamba
cis-1 ,3-dichloropropene
Diphenamid
Diuron
Endosulfan
EPIC
Ethoprop
Etridiazole
Fenamiphos
Fenarimol
Fluometuron
Fluridone
Hexazinone
Linuron
Soil Half-Life
(Days)
30
20
60
13
10
50
7
N/A
130
30
30
14
30
10
7-10
100
14
10
30
90
50
6
25
20
50
360
85
21
90
60
KOC
30
10
300
400
300
22
260
140000
1650
1380
400
190
430
20-100
20-500
NA
2
32
210
480
12400
200
70
1000
100
600
100
1000
54
400
a Although chlordane is comprised of alpha-chlordane and gamma-chlordane,
KO,. information is available for chlordane only.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 88
Exhibit 3-43 (continued)
Soil Half-Life and K._ for Pesticides
UL*
that Were Not Detected
Detected Pesticide
Methiocarb
Methomyl
Methoxychlor
Metolachlor
Metribuzin
Mevinphos
Molinate
Napropamide
Norflurazon
Oxamyl
Pebulate
Permethrin
Prometryn
Propachlor
Propanil
Propazine
Propham
2,4,5-T
Tebuthiuron
Terbacil
Terbutryn
Triademefon
Trifluralin
Vernolate
Soil Half-Life
(Days)
30
30
120
90
40
3
21
70
90
4
14
30
60
6.3
1
135
10
24
360
120
42
21
60
12
Koc
300
72
80000
200
60
44
190
400
600
25
430
100000
400
80
149
154
200
80
80
55
2000
300
8000
260
Source: Wauchope, R. Don, Selected Values for Six
Parameters from the SCS/ARS/CES Pesticide
Properties Database: A Brief Description, February
20, 1991 and selected values prepared for the U.S.
EPA Environmental Criteria and Assessment Office
as revisions to the Superfund Public Health
Evaluation Manual. 1986.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 89
The logistic regression analyses consider the extent to which soil half-life or K^ can be regarded as
predictors of the probability of detection of an analyte in the NPS, and should be expected to give qualitatively
the same results as the t-tests presented above. Exhibit 3-44 provides details of the estimated coefficients for
the logistic regression models for to(half-life) and
Exhibit 3-44
Models of the Relationship Between Pesticides Detected and
Half-life or Knr in the NPS
OC
Explanatory
Variable
MHalf-life)
NKOC)
Estimated
Intercept
-3.96
-0.01
Standard
Error
1.19
0.95
Estimated
Explanatory
Variable
Coefficient -
In Scale
4-0.64
-0.29
Standard
Error
0.28
0.18
Significance
Level
(p-value)
0.021
0.100
The results of the logistic regression indicate that there is a relationship between soil half-life and the
probability of a pesticide's detection in at least one well in the NPS, but there is no evidence of a similar
relationship for
An additive model involving both soil half-life and K^ as independent variables showed strong
evidence for a positive relationship between soil half-life and the probability of detecting a pesticide
(p-value = 0.008), and weak evidence of a relationship for K^ (p-value = 0.034).26 This model provides
a marginally better fit for the data than the model involving half-life only. A further model including an
interaction term for half-life and K^ does not provide a better fit. Some of the half-life and K^ data were
provided as a range because the sources used to compile the data provided differing values. The results
presented in this section are based on analyses that use extreme values from these ranges. This corresponds
to the least possible significance for half-life and the most possible significance for K^. If the opposite
assumption is made, the results for K^. in the additive model yield a descriptive significance levels greater than
0.1, providing further evidence that the best model based on NPS data contains a half-life term only.
The soil half-lifes for detected pesticides were higher those for the nondetected pesticides. The soil
half-life values used for the detected pesticides ranged from 7 to 990 days. The soil half-life values for the
nondetected pesticides ranged from 3 to 360 days. As Exhibit 3-44 and the subsequent statistical analyses
demonstrate, the means of the soil half life values for detected and not detected pesticides differ significantly.
This finding can be interpreted as providing evidence that pesticides that persist longer in the environment
are more likely to be detected in drinking water wells. However, these analyses are subject to the influence
of the small number of detections and other uncontrolled confounding effects and are applicable to the range
of the data only (that is the range of half-life values and, more importantly, the chemicals used in this
analysis). Extrapolation to other groups of chemicals should be treated with caution. Similar evidence for
a relationship between pesticide detections and K^ was not found.
26 Chlordane has the highest K^ value used in the analyses presented above. Chlordane was not detected at
concentrations above the NPS minimum reporting levels for a-chlordane and y-chlordane of 0.060 yg/L. Both isomers of
chlordane were detected at much lower levels by the EPA laboratories. If Chlordane is treated as a detected chemical instead
of a non-detected chemical in the analyses reported in this section, the p-value for K^ in the univariate logistic regression
model changes from 0.100 to 0.480. A similar comparison can not be made for the additive model since a half-life for
chlordane was not specified. An increase in p-value, or decrease in significance, could be expected.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 90
3.9 Relationship Between Nitrate Detections and Concentrations and Pesticide
Detections
This section provides the results of an analysis of the extent of the association between nitrate and
pesticide detections in drinking water wells. The underlying hypothesis for these analyses is that pesticide
detections in drinking water wells follows the same pattern as nitrate detections. Results of the analysis of
the relationship between pesticide detections and nitrate detections are presented first, followed by results of
the analysis of pesticide detections with nitrate concentration given nitrate detection above the NFS minimum
reporting limit of 0.15 mg/L. Analysis of the relationship between nitrate and pesticide concentrations is not
feasible as there are insufficient detections for any single pesticide and the pesticides generally have different
NFS minimum reporting limits.
The results of the analysis of nitrate and pesticide detections presented in Exhibit 3-45 provide weak
evidence that there is a greater chance of at least one pesticide detection when nitrate is detected. Results
are presented in this section rather than in Appendix A even though the p-values exceed 0.05 to enable
discussion of the topic to be included. The effects are not strong enough to fully substantiate a hypothesis
of association between nitrate and pesticide detections but are mildly suggestive. The weakness of the result
may be due to the small number of pesticide detections that are distributed among four categories for this
analysis. Analysis of smaller groups of pesticides leads to the same type of results when there are sufficient
data to satisfy the underlying statistical assumptions required.
Exhibit 3-45
Analysis of Association between Nitrate and Pesticide Detections
Survey
CWS Wells
Rural Domestic Wells
Proportion of Pesticide Detections
When Nitrate
Detected
12.8 percent
5.6 percent
Not Detected
7.4 percent
2.4 percent
Significance
Level
(p-value)
0.093
0.061
Results of the analysis relating nitrate concentrations and pesticide detections are presented in Exhibit
3-46. These results do not substantiate the results presented above, i.e., there is insufficient evidence to
suggest that pesticide detections are related to nitrate concentrations. These results are based on a logistic
regression analysis with the logarithm of nitrate concentration as the independent variable, and the logistic
function of the probability of pesticide detection as the dependent variable. The models are not useful for
predictive purposes due to the comparatively large p-values for the estimated independent variable coefficients.
Analysis of smaller groups of pesticides leads to the same type of results when there is sufficient data to satisfy
the underlying statistical assumptions required.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 91
Exhibit 3-46
Analysis of Association Between Nitrate Concentration
and Pesticide Detections
Survey
CWS Wells
Rural Domestic Wells
Estimated
Intercept
-2.26
-3.20
Standard
Error
0.24
0.31
Estimated
Coefficient
for
0n (Nit rate)
0.24
0.14
Standard
Error
0.20
0.13
Significance
Level
(p-value)
0.225
0.292
In summary, there is insufficient evidence to suggest a relationship between nitrate concentration and
pesticide detections, and weak support for an association between nitrate and pesticide detections. These
results alone do not provide sufficient evidence to support a hypothesis of a correlation between nitrate and
pesticide contamination hi drinking water wells nationally.
3.10 Precipitation and Drought
This section presents the results obtained from analyses of the relationship between pesticide and
nitrate detections and two measures of precipitation: (1) data on precipitation patterns and (2) an index of
drought conditions.
In order to assess the impact of precipitation patterns on the occurrences of pesticides and nitrate in
well water, NPS obtained precipitation data from weather stations across the U.S. for 405 counties with CWS
wells (399 counties in which wells were sampled and 6 counties contiguous to counties in which wells were
sampled but for which precipitation data were unavailable) and 90 counties in which rural domestic water wells
were sampled. A new set of variables depicting short-term and long-term precipitation characteristics, based
on the sampling date (i.e., month and year), were then extracted from the monthly precipitation data set.
These variables are:
• Quantity of precipitation during the month of well sampling (PCUR);
• Quantity of precipitation during the month prior to sampling (PPREV);
• Average monthly precipitation for the year prior to sampling (PAVG1);
• Maximum monthly precipitation for the year prior to sampling (PMAX1);
• Average monthly precipitation for the two years prior to sampling (PAVG2);
• Maximum monthly precipitation for the two years prior to sampling (PMAX2);
• Average monthly precipitation for the five years prior to sampling (PAVG5); and
• Maximum monthly precipitation for the five years prior to sampling (PMAX5).
In order to determine if the 1988-1989 drought in the central and western United States had an impact
on the Survey detection estimates, NPS obtained information on drought severity from the Climate Analysis
Center of the National Oceanic and Atmospheric Administration (NOAA/CAC). These data were based on
the Drought Severity Index or Palmer Drought Index (PDI). The PDI was developed by Palmer in 1965 to
evaluate meteorological drought. PDI is currently being used jointly by NOAA and the United States
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 92
Department of Agriculture (USDA) to depict prolonged abnormal dryness or wetness on a regional scale
across the United States27.
A data set consisting of well-specific drought severity and moist spell categories was then extracted
from the "Drought Severity Index by Division" maps generated by NOAA/CAC. In these maps, categories of
drought and moist spell are defined as:
• Extreme drought: PDI less than or equal to -4.0;
• Severe drought: PDI from -3.0 to -3.9;
• Moderate drought: PDI from -2.0 to -2.9;
PDI from -1.9 to 1.9;
PDI from 2.0 to 2.9;
PDI from 3.0 to 3.9; and
PDI greater or equal to 4.0.
Near normal:
Unusually moist:
Very moist:
Extremely moist:
3.10.1 Results of Analysis of Precipitation Data
Community Water System Wells. This section describes results of the analyses performed to examine
the relationship between precipitation and pesticide and nitrate detections and nitrate concentrations in CWS
well water. Analyses for wells with pesticide and nitrate detections were examined for evidence of associations
between precipitation variables and detections. The results for pesticides and nitrate detections are
summarized in Exhibits 3-47 and 3-48 respectively. Herbicide detections and DCPA acid metabolites
detections yield qualitatively similar results to those presented for pesticide detections. Smaller subgroups of
pesticides do not contain sufficient detections to satisfy the statistical assumptions.
Exhibit 3-47
Estimated Models of the Probability of Pesticide Detections by Precipitation
Factors for Community Water System Wells
Precipitation
Variable
Maximum monthly
2 years prior
Estimated
Intercept
-1.495
Standard
Error
Estimated
Precipitation
Variable
Coefficient
0.377 HI -0.093
I
Standard
Error
0.043
Significance
Level
(p-value)
0.032
Based on results of the analyses presented in Exhibit 3-47, the probability of detecting pesticides in
CWS wells is not generally related to precipitation factors. There is some evidence of exceptions for the
occurrence of intense precipitation events over an extended period as indicated by the results for precipitation
variable PMAX2. The precipitation variables are, however, highly correlated.
27 Palmer, Wayne C., Meteorological Drought, U.S. Department of Commerce, Weather Bureau, Research Paper No.
45, Washington, D.C., February, 1965. The Palmer Drought Index (PDI) is calculated for each climatic division designated
by NOAA. The number of climatic divisions per State ranges from 2 to 10 with a mode of 8.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 93
Exhibit 3-48
Estimated Models of the Probability of Nitrate Detections by Precipitation
Factors for Community Water System Wells
Precipitation
Variable
During month of
sampling
During prior month
Monthly average
for 1 year prior
Monthly average
for 2 years prior
Monthly average
for 5 years prior
Monthly maximum
for 1 year prior
Monthly maximum
for 2 years prior
Monthly maximum
for 5 years prior
Estimated
Intercept
0.526
0.431
1.463
1.763
2.210
1.141
1.446
1.841
Standard
Error
0.163
0.171
0.286
0.311
0.361
0.258
0.291
0.342
Estimated
Precipitation
Variable
Coefficient
-0.160
-0.136
-0.495
-0.589
-0.700
-0.156
-0.166
-0.171
Standard
Error
0.040
0.047
0.094
0.101
0.111
0.034
0.034
0.032
Significance
Level
(p-value)
< 0.0005
0.004
< 0.0005
< 0.0005
< 0.0005
< 0.0005
< 0.0005
< 0.0005
Analysis of the relationship between precipitation factors and nitrate concentrations in CWS wells
yields results that substantiate those for nitrate detections. Results of the concentration-based analyses are
presented in Exhibit 3-49. With the exception of the previous month precipitation factor, all precipitation
factors exhibit a negative correlation with nitrate concentration. The results presented are for models for the
logarithmic transformation of nitrate concentration.
Rural Domestic Wells. This section describes results of the analyses performed to examine the
relationship between precipitation and pesticide and nitrate detections and nitrate concentrations in rural
domestic well water. The results for nitrate detections are summarized in Exhibit 3-50. Smaller subgroups
of pesticides do not contain sufficient detections to satisfy the statistical assumptions necessary for correlation
or regression analysis.
Based on results of these analyses, the probability of detecting pesticides or nitrate is not generally
related to precipitation factors. There is some evidence of an exception for the relationship between nitrate
detections and intense precipitation events during the previous five years, as indicated by the results for the
precipitation variable PMAX5 (see Exhibit 3-50). The precipitation variables are, however, highly correlated.
Analysis of the relationship between precipitation factors and nitrate concentrations in rural domestic
wells is presented in Exhibit 3-51.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 94
Exhibit 3-49
Estimated Models of Nitrate Concentration* by Precipitation
Factors for Community Water System Wells
Precipitation Variable
During month of
sampling
Monthly average for prior
year
Monthly average for prior
2 years
Monthly average for prior
5 years
Monthly maximum for
prior year
Monthly maximum for
prior 2 years
Monthly maximum for
prior 5 years
Estimated
Intercept
0.450
0.672
0.729
0.749
0.667
0.825
0.787
Standard
Error
0.116
0.175
0.88
0.200
0.172
0.189
0.223
Estimated
Precipitation
Variable
Coefficient
-0.083
-0.161
-0.198
-0.194
-0.072
-0.082
-0.060
Standard
Error
0.034
0.062
0.066
0.068
0.025
0.024
0.022
Significance
Level
(p-value)
0.015
0.010
0.003
0.004
0.005
0.001
0.008
Nitrate concentrations are logarithmically transformed.
Exhibit 3-50
Estimated Models of the Probability of Nitrate Detections by Precipitation
Factors for Rural Domestic Wells
Precipitation
Variable
Monthly maximum
for 5 years prior
Estimated
Intercept
1.873
Standard
Error
Estimated
Precipitation
Variable
Coefficient
0.584 If -0.156
Standard
Error
0.055
Significance
Level
(p-value)
0.005
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 95
Exhibit 3-51
Estimated Models of Nitrate Concentration* by Precipitation
Factors for Rural Domestic Wells
Precipitation
Variable
Monthly average for
prior year
Estimated
Intercept
0.831
Standard
Error
0.305
Estimated
Precipitation
Variable
Coefficient
-0.140
Standard
Error
0.067
Significance
Level
(p-value)
0.040
Nitrate concentrations are logarithmically transformed.
3.10.2 Palmer Drought Index
Community Water System Wells. Results of statistical analyses involving the Palmer Drought Index
are presented first for pesticide detections, next for nitrate detections, and finally for nitrate concentrations.
A frequency distribution for wells with pesticide detections was examined for evidence of an association
between drought conditions and the detection of any pesticide in CWS well water. The distribution of wells
with pesticide detections across various drought and moist spell categories is presented in Exhibit 3-52.
Exhibit 3-52
Frequency Distribution of Estimated Pesticide
Detections by Drought and Moist Spell Categories
for Community Water System Wells
Drought and Moist
Categories
Extreme Drought
Severe Drought
Moderate Drought
Near Normal Condition
Unusually Moist
Very Moist
Extremely Moist
Estimated
Number of
Wells with
Pesticide
Detections
1,140
180
1,300
6,000
190
570
0
Estimated
Total
Number of
Wells
8,080
5,980
13,200
55,100
6,410
4,590
1,200
Estimated
Percent of
Wells with
Pesticide
Detections
14.1
3.0
9.8
10.9
3.0
12.4
0
No observable trend relates the proportion of wells with pesticide detections to drought or moist
conditions measured by the PDI. The apparent fluctuations are a function of the relatively small number of
wells estimated to contain detectable levels of pesticides for some of the PDI categories. There are too few
pesticide detections to be spread across the seven PDI categories in a statistical analysis. To determine if there
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 96
is sufficient evidence of a drought effect on pesticide detections the seven categories were collapsed into three
categories signifying drought, normal, and moist conditions. As shown in Exhibit 3-49, when drought and
moist conditions are collapsed the percents of pesticide detections by drought categories become very similar.
A statistical analysis of the differences between the three mean proportions presented in Exhibit 3-53 indicates
there is insufficient evidence to suggest drought conditions are related to pesticide detections.
Exhibit 3-53
Frequency Distribution of Estimated Pesticide
Detections by Collapsed Drought and Moist Spell Categories
for Community Water System Wells
Drought and Moist Categories
Drought (PDI less than or equal -2)
Near Normal (PDI from -1 .9 to 1 .9)
Moist (PDI greater than or equal 2)
Estimated
Number of
Wells with
Pesticide
Detections
2,610
6,000
770
Estimated
Total
Number
of Wells
27,300
55,100
12,200
Estimated
Percent of
Wells with
Pesticide
Detections
9.6
11.4
6.3
A frequency distribution for wells with nitrate detections was also examined for evidence of an
association between drought conditions and the detection of nitrate in CWS well water. The distribution of
wells with nitrate detections is presented in Exhibit 3-54.
Exhibit 3-54
Frequency Distribution of Estimated Nitrate
Detections by Drought and Moist Spell Categories
for Community Water System Wells
Drought and Moist
Categories
Extreme Drought
Severe Drought
Moderate Drought
Near Normal Condition
Unusually Moist
Very Moist
Extremely Moist
Estimated Number
of Wells Nationally
with Nitrate
Detections
5,160
3,680
6,440
28,700
1,910
2,430
710
Estimated Total
Number of Wells
Nationally
8,080
5,980
13,200
55,100
6,410
4,590
1,200
Estimated Percent
of Wells Nationally
with Nitrate
Detections
64
62
49
52
30
53
59
The percent of wells with nitrate detections appears to be greater in drought areas than in moist areas.
To determine if there is sufficient evidence of a drought effect on nitrate detections the seven categories were
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 97
collapsed into three categories signifying drought, normal, and moist conditions. As shown in Exhibit 3-55,
when drought and moist conditions are collapsed the difference in percents of nitrate detections by drought
categories becomes more apparent. However, a statistical analysis of the differences between the three mean
proportions presented in Exhibit 3-55 does not yield a statistically significant result, i.e., there is no evidence
based on this analysis to suggest drought conditions are related to nitrate detections in CWS wells.
Exhibit 3-55
Frequency Distribution of Estimated Nitrate
Detections by Collapsed Drought and Moist Categories
for Community Water System Wells
Drought and Moist Categories
Drought (PDI less than or equal to -2)
Near Normal (PDI from -1 .9 to 1 .9)
Moist (PDI greater than or equal to 2)
Estimated
Number of
Wells
Nationally
with Nitrate
Detections
15,300
28,700
5,050
Estimated
Total
Number of
Wells
Nationally
27,300
55,100
12,200
Estimated
Percent of
Wells
Nationally
with Nitrate
Detections
56
52
41
A further analysis was performed to determine the relationship between nitrate concentrations and
the three collapsed categories of PDI (drought, normal, and moist). The average nitrate concentrations for
these drought categories are presented in Exhibit 3-56. Statistical analyses were performed on the logarithm
of nitrate concentrations that correctly account for the standard errors of the mean values. Analysis of the
differences in means of the logarithm of nitrate concentrations yields a descriptive significance level of 0.003,
providing strong evidence that nitrate concentrations are lower in CWS wells in moist regions.
Exhibit 3-56
Means of Nitrate Concentrations
For Collapsed Drought and Moist Categories
for Community Water System Wells
Drought and Moist Categories
Drought
(PDI less than or equal to -2)
Near Normal (PDI from -1 .9 to 1 .9)
Moist
(PDI greater than or equal to 2)
Mean Nitrate
Concentration
(mg/L)
2.51
2.49
1.13
Mean
fn (Nitrate
Concentration)
0.34
0.33
-0.44
Rural Domestic Wells. A frequency distribution for wells with pesticide detections was examined for
evidence of an association between drought conditions and the detection of any pesticide in rural domestic
well water. The distribution of wells with pesticide detections across various drought and moist spell
categories is presented in Exhibit 3-57.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 98
Exhibit 3-57
Frequency Distribution of Estimated Pesticide
Detections by Drought and Moist Spell Categories
for Rural Domestic Wells
Drought and Moist
Spell Categories
Extreme Drought
Severe Drought
Moderate Drought
Near Normal
Unusually Moist
Very Moist
Extremely Moist
Estimated
Number of
Wells with
Pesticide
Detections
18,700
38,400
23,800
234,000
81,400
29,100
0
Estimated
Total
Number of
Wells
415,000
956,000
719,000
4,860,000
1,500,000
1,150,000
907,000
Estimated
Percent of
Wells with
Pesticide
Detections
4.5
4.0
3.3
4.8
5.4
2.5
0
The occurrences of pesticides in rural domestic well water do not appear to be related to the presence
of drought or moist spells. The estimated percents of wells with pesticide detections for various drought or
moist spell categories (Exhibit 3-57) are not associated with any observable trend. There is no increase or
decrease in the percents of detected pesticides with changes in moisture conditions. In addition, when the
three drought categories and the three moist categories are collapsed into two categories, the estimated
percents of pesticide detections by drought categories are very similar (Exhibit 3-58). It appears unlikely that
the presence of drought had any influence on the occurrence of pesticides in rural domestic well water.
Exhibit 3-58
Frequency Distribution of Pesticide Detections
by Collapsed Drought and Moist Spell Categories
for Rural Domestic Wells
Drought and Moist Spell Categories
Drought (PDI less than or equal -2)
Near Normal (PDI from -1 .9 to 1 .9)
Moist (PDI greater than or equal 2)
Estimated
Number of
Wells with
Pesticide
Detections
80,900
234,000
110,000
Estimated
Total
Number of
Wells
2,090,000
4,860,000
3,560,000
Estimated
Percent of
Wells with
Pesticide
Detections
3.9
4.8
3.1
A frequency distribution for wells with nitrate detections was also examined for evidence of an
association between drought conditions and the detection of nitrate in rural domestic well water. The
distribution of wells with nitrate detections is presented in Exhibit 3-59.
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 99
Exhibit 3-59
Frequency Distribution of Estimated Nitrate
Detections by Drought and Moist Spell Categories
for Rural Domestic Wells
Drought and Moist
Spell Categories
Extreme Drought
Severe Drought
Moderate Drought
Near Normal
Unusually Moist
Very Moist
Extremely Moist
Estimated Number
of Wells Nationally
with Nitrate
Detections
230,000
589,000
516,000
2,510,000
772,000
741,000
620,000
Estimated Total
Number of Wells
Nationally
415,000
956,000
719,000
4,860,000
1,500,000
1,150,000
907,000
Estimated Percent
of Wells Nationally
with Nitrate
Detections
56
62
72
52
51
65
68
The percent of wells with nitrate detections does not appear to differ significantly across PDI
categories. To verify that there is insufficient evidence of a drought effect on nitrate detections the seven
categories were collapsed into three categories signifying drought, normal, and moist conditions. A statistical
analysis of the differences between the three mean proportions presented in Exhibit 3-60 shows that there is
insufficient evidence (p-value > 0.1) that drought conditions are related to nitrate detections in rural domestic
wells.
Exhibit 3-60
Frequency Distribution of Estimated Nitrate
Detections by Collapsed Drought and Moist Categories
for Rural Domestic Wells
Drought and Moist Categories
Drought (PDI less than or equal to -2)
Normal (PDI from -1 .9 to 1 .9)
Moist (PDI greater than or equal to 2)
Estimated
Number of
Wells
Nationally
with Nitrate
Detections
1,340,000
2,510,000
2,130,000
Estimated
Total
Number
of Wells
Nationally
2,090,000
4,860,000
3,560,000
Estimated
Percent of
Wells
Nationally
with Nitrate
Detections
57
51
60
National Pesticide Survey: Phase II Report
-------
Chapter Three: Results 100
A further analysis was performed to determine the relationship between nitrate concentrations and
the three collapsed categories of PDI (drought, normal, and moist). The average nitrate concentrations for
these drought categories are presented in Exhibit 3-61. Statistical analyses were performed on the logarithm
of nitrate concentrations that correctly account for the standard errors of the mean values. Analysis of the
differences in means of the logarithm of nitrate concentrations shows no evidence that nitrate concentrations
are lower in rural domestic wells in moist regions p-value > 0.5).
Exhibit 3-61
Means of Nitrate Concentrations
for Collapsed Drought and Moist Categories
for Community Water System Wells
Drought and Moist Categories
Drought
(PDI less than or equal to -2)
Near Normal (PDI from -1 .9 to 1 .9)
Moist
(PDI greater than or equal to 2)
Mean Nitrate
Concentration
(mg/L)
3.72
3.17
2.54
Mean
ln(Nftrate
Concentration)
0.35
0.36
0.29
Pesticide and nitrate detections in rural domestic wells are not found to be related to drought
conditions measured through the Palmer Drought Index. Though the lack of a relationship with pesticide
detections can possibly be attributed in part to the small number of detections, the same cannot be said for
nitrate detections. The results for rural domestic wells do not corroborate the results for CWS wells,
suggesting that the effect of drought conditions nationally is different for CWS wells than for rural domestic
wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results
In designing the National Pesticide Survey, EPA sought to develop information on five major factors
that affect the presence of pesticides and nitrate in ground water -- ground-water sensitivity, agricultural
activity and pesticide and fertilizer use, contaminant transport mechanisms, chemical properties of pesticides,
and physical characteristics of wells. EPA collected data on these key factors using a number of different
variables and measurement levels. For both CWS wells and rural domestic wells, the data can be viewed as
describing three successive zones surrounding the sampled wells: the area near the well on the well owner's
property (ranging from 300 to 500 feet, depending on the information sought); the area within one-half mile
of the well; and the area within the county where the well is located. Data were also collected for rural
domestic wells for a fourth zone, sub-county "cropped and vulnerable" areas surrounding the well.
Chapter 3 describes the results of more than one thousand data analysis tests carried out to identify
variables that are associated with nitrate and pesticide detections in the NPS. EPA's objective in conducting
these analyses was to improve its understanding of the relationship of ground-water sensitivity, fertilizer and
pesticide use, transport, chemical properties of pesticides, and physical characteristics of wells to pesticide and
nitrate detections. EPA sought to identify strong evidence to develop indicators of vulnerable wells, in order
to support several objectives. The Phase II results may be used both to guide program and policy, as support
for the implementation of the EPA Pesticide in Ground Water Strategy, data for assessment of pesticide
registration and review, information to support improvements in the control of nitrate pollution sources, and
to identify additional research needs.
This chapter evaluates the results of the Phase II analysis presented in Chapter 3 from several different
standpoints. First, EPA regrouped the significant results in terms of the five key factors and sought to identity
trends in each group, as well as conflicting or inconsistent results. Possible trends were then evaluated with
respect to the quality of the data upon which they were based, and compared to results obtained by other
surveys and studies of similar topics. Next, EPA conducted multivariate analyses using combinations of results
to test models composed of the major findings as potential indicators of contamination. Third, EPA assessed
the potential effects of factors that could limit or confound the identification of significant variables.
The purpose of this chapter is, in part, to present an explicit discussion of the quality of the individual
results. Chapter 3 stated the hypotheses tested and identified whether the results are considered to support
or contradict the hypothesis. This chapter evaluates those supporting or contradictory results in the following
ways:
• By assessing the amount and quality of data upon which the result is based;
• By providing additional details concerning why the result was evaluated as supporting or not
supporting particular hypotheses; and
• By comparing the results to the results of other selected recent surveys and studies of ground-
water contamination by pesticides and nitrate.
Much of the background necessary for assessing the amount and quality of data supporting particular
results is provided in the NPS Phase I Report. In particular, the Phase I Report provides important
background on the planning and background of the Survey, including choice of analytes, development of multi-
residue analytic methods, choice of stratification variables, and well sampling and data collection techniques.
Copies of the questionnaires used in the Survey and national estimates, including confidence limits, for many
of the questionnaire items are provided in Appendix D of that report.
An overall comparison of the key results of the NPS and of selected surveys of the presence of
pesticides and nitrate in ground water or well water are presented in Exhibit 4-1. The survey most closely
resembling the NPS is the National Alachlor Well Water Survey (NAWWS), which was a statistically-based
survey based on a design similar to the NPS that sought to estimate the proportion of private, rural domestic
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 102
wells with detectable concentrations of alachlor, four other herbicides, and nitrate, located in counties
nationwide in which alachlor is sold.1
Four other surveys also are included, out of the large number of state and local ground water studies
that have been conducted, because of their scope and the period of time in which they were carried out. They
are the Iowa State-Wide Rural Well-Water Survey; the Wisconsin Grade A Dairy Farm Well Water Survey;
the Rhode Island Private Well Water Survey; and the Minnesota Surveys of Selected Minnesota Wells. These
surveys were selected for comparison to the National Pesticide Survey because they were carried out at
approximately the same time as the NFS, each involved sampling of private rural wells (although several
surveys sampled other categories of wells in addition), and each analyzed water samples for selected pesticides
and nitrate. These other surveys were generally limited in scope to a single state or portion of a state.
They are not probability-based, but chose special geographic areas, such as karst regions, or areas associated
with selected activities, such as dairy farming, for sampling. Several carried out extensive sampling, and some
used repeat sampling of the same well or location. Like the NPS, most analyzed samples using multiresidue
chemical analytic methods, sometimes accompanied by single tests for specific analytes. The NPS analyzed
for the largest number of analytes.
As Exhibit 4-1 indicates, all of these surveys detected nitrate in about half of the wells sampled. The
proportions of wells containing at least one pesticide varied more widely among surveys, from a low of 4.2
percent for the NPS rural domestic well survey to a high of 51 percent for the Minnesota Department of
Agriculture study. Differences in detection rates could be due to several factors, including random versus
targetted sampling, different reporting limits, and absence of mass spectrometry confirmation.
Comparison of the results for the NPS community water system well survey with the results for the
NPS rural domestic well survey provide some inconclusive evidence that CWS wells may be more vulnerable
to contamination than rural domestic wells. In general, pesticides, including atrazine and DCPA acid
metabolites, were detected more frequently in CWS wells than in rural domestic wells. However, the
concentrations of pesticides were generally lower in CWS wells than in rural domestic wells. The highest
concentrations found in rural domestic wells were substantially higher than the highest concentrations of the
same pesticide found in CWS wells. The highest concentrations of DCPA acid metabolites, in contrast, were
found in CWS wells.
The Phase II analyses were not able to identify factors that were associated with these differences
between CWS wells and rural domestic wells. Three possible factors were considered, but the Phase II analysis
could not conclusively identify significant results relating to them. First, the possibility of greater drawdown
by large CWS wells could be associated with the presence of larger numbers of analytes at lower
concentrations. Insufficient data on system sizes and well pumping rates were available to fully evaluate this
factor. Large community well systems, defined as systems with more than two wells, were found to have a
proportion of detections of any pesticide of 15.0 percent, while small systems, defined as systems with one or
two wells, had a proportion of detections of 5.0 percent, with a p-value for the difference in proportions of
0.0002. For nitrate, the difference was not significant. Second, the recharge areas of CWS wells and rural
domestic wells, a factor that the NPS could not examine because data on recharge areas were not collected,
could differ significantly. Third, the differences in the results might be related to differences between
1 Holden, L.R. and J.A. Graham, February 1990. The National Alachlor Well Water Survey: Project Summary.
Monsanto Agricultural Company, St. Louis, Missouri, Volume 1 of 7. (Hereafter NAWWS Project Summary)
2 Iowa Department of Natural Resources, [1989], Iowa State-Wide Rural Well-Water Survey: Summary of State-wide
Results and Study Design: Perspectives on the SWRL Results: Summary of Results: Pesticide Detections: Summary of
Results: Atrazine Detections; Summary of Results: Nitrate and Bacteria. LeMasters, G. and D J. Doyle, April 1989, Grade
A Dairy Farm Well Water Quality Survey. Wisconsin Department of Agriculture, Trade, and Consumer Protection and
Wisconsin Agricultural Statistics Service. Groundwater Section, Rhode Island Department of Environmental Management,
May 1990, Rhode Island Private Well Survey Final Report. Providence, Rhode Island. Klaseus, T.G., G.C. Buzicky, and E.G.
Schneider, February 1988, Pesticides and Groundwater: Surveys of Selected Minnesota Wells. Minnesota Department of
Health and Minnesota Department of Agriculture.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 103
agricultural and non-agricultural uses of pesticides, and particularly to non-agricultural uses of DCPA.
However, conclusive information on the balance between agricultural and non-agricultural uses of DCPA was
not available. As Exhibit 4-1 indicates, DCPA and its acid metabolites were detected in only one of the
surveys described (the Iowa State-Wide Rural Well-Water Survey) in addition to the NPS.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 104
Exhibit 4-1: Comparison of Selected
Variable
Type of wells sampled
Study area
Sampling period
Number of wells sampled
Wells containing nitrate (%)
Nitrate minimum reporting limit
Wells containing nitrate above 10 mg/L (%)
Wells containing at least one pesticide (%)
Wells containing pesticides above drinking
water standards (%)
Wells containing atrazine (%)
Atrazine minimum reporting limit
Wells containing DCPA or DCPA acid
metabolites (%)
Number of different pesticides
detected/number pesticide analytes
Pesticides detected
Analytical method:
single column chromatography
second column chromatography
confirmation
mass spectrometry
National Pesticide Survey
Community Water
System Wells
Public
Throughout nation
April 1988 -February 1990
540
52.1 > 0.15 mg/L
0.15 mg/L (ppm)
1.2
10.4
0.8
1.7
0.12/tg/L(ppb)
6.4 (DCPA acid
metabolites)
7/126
atrazine
DCPA acid metabolites
dibromochloropropane
dinoseb
hexachlorobenzene
prometon
simazine
Yes
Yes
Yes
Rural Domestic Wells
Private, rural
Throughout nation
April 1988 -February 1990
752
57.0 > 0.15 mg/L
0.15 mg/L (ppm)
2.4
4.2
0.6
0.7
0.12/ig/L(ppb)
2.5 (DCPA acid
metabolites)
10/126
alachlor
atrazine
bentazon
DCPA acid metabolites
dibromochloropropane
ethylene dibromide
ethylene thiourea
gamma-HCH
prometon
simazine
Yes
Yes
Yes
Monsanto National
Alachlor Well
Water Survey
Private, rural
All counties where
alachlor reported sold
June 1988 - May 1989
1,430
> 50
Reported all values
5
Not reported
Not reported
12
Reported all values
Not analyzed
5/5
alachlor
atrazine
cyanazine
metolachlor
simazine
No
No
Yes
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 105
Drinking Water Well Surveys
Rhode Island
Private Well
Survey
Private, urban and
rural
Statewide
1986
463
81
0.1 mg/L (ppm)
1
11
0
Not reported
0.05 jig/L (ppb)
Not detected
(DCPA)
7/26
atrazine
aldicarb
butylate
carbaryl
carbofuran
dicamba
dinoseb
Yes
Yes
No
Minnesota Departments of Health (MDH)
and Agriculture (MDA) Surveys
MDA Study
Monitoring
(observation) wells1,
private drinking water
wells, and irrigation
wells
Agricultural regions
with vulnerable
ground water2
July 1985- June 1987
100
61% of the wells > 1
mg/L (ppm)
1 mg/L (ppm)
23
51
Not reported
47
0.01 A9/L (ppb)
Not analyzed
8/30
alachlor
aldicarb
atrazine
cyanazine
dicamba
metribuzin
pentachlorophenol
simazine
Yes
Yes
No
MDH Study
Public
Agricultural regions
with vulnerable
ground water2
July 1985 -June 1987
400
47.3
0.4 Mg/L (ppb)
7.1
28.5
Not reported
26.8
0.01 jtg/L (ppb)
Not analyzed
12/30
2,4-D
2,4,5-T
alachlor
atrazine
cyanazine
dicamba
EPIC
MCPA
metolachlor
metribuzin
picloram
propachlor
Yes
Yes
No
Iowa State-Wide
Rural Weil-Water
Survey
Private, rural
Statewide
April 1988 -June 1989
686
60% of wells < 3 mg/L
(ppm)
0.1 mg/L (ppm)
18.3
13.6
1.2
8
0.1 ^g/L (ppb)
0.4 (DCPA)
15/27 plus selected
metabolites
2,4-D
3-hy d roxy-carbof u ran
3-ketc-carbofuran
alachlor
atrazine
cyanazine
dacthal (DCPA)
deethyl-atrazine
desopropyl-atrazine
hydroxy-alachlor
metolachlor
metribuzin
pendamethalin
propachlor
trrfluralin
Yes
No
No
Wisconsin Grade A Dairy
Farm Well Water Quality
Survey
Dairy farm wells
Grade A dairy farms
statewide
August 1988 - February 1989
534
65
0.5 mg/L (ppm)
10
13
Insufficient data to estimate
12
0.15fig/L(ppb)
Not detected (DCPA)
4/44
alachlor
atrazine
metolachlor
metribuzin
No
No
Yes
1 Wells were originally installed and monitored by the Minnesota Department of Natural Resources or the United States Geological Survey
2 Agricultural regions of the state and, within those regions, areas where local or regional soils and hydrogeologic conditions make the ground water susceptible to pesticide
contamination.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 106
4.1 Major Results
This section reports the results discussed in Chapter 3, organized to address the five major factors
expected to affect contamination. The results are evaluated with respect to the quality of the data that support
them; their consistency with the results obtained by other studies, particularly other surveys of pesticides or
nitrate in ground water; and their consistency with major existing theories concerning the presence of
pesticides in well water.
4.1.1 Results for Sensitivity Factor
Sensitivity was defined, for this review of the analytic results, as the intrinsic susceptibility of an
aquifer to contamination. Sensitivity addresses the hydrogeologic characteristics of the aquifer and the
overlying soil and geologic materials, and is unrelated to agricultural practices or pesticide characteristics.
Aquifer vulnerability now is considered by EPA to be a more comprehensive term than sensitivity. It
encompasses the susceptibility of an aquifer to contamination from the combined effects of its sensitivity,
agricultural practices, and pesticide characteristics.3 However, during the design (stratification) and
implementation of the NFS, the term vulnerabiity and its derivatives had approximately the same meaning as
sensitivity now has been given.
These measures of sensitivity or vulnerability apply to the ground water lying directly under the area
of study. Sensitivity at any particular well drawing water from the aquifer is not measured. Other measures,
including the flow of ground water in the area, are important in determining a particular well's sensitivity.
Wells in a highly sensitive area, however, would generally be more sensitive to contamination than wells
drawing water from a less sensitive aquifer.
Analytic results for the sensitivity factor were available from several different sources. They came,
first, from the analysis of DRASTIC scoring. DRASTIC results assessed sensitivity factors at the county and
the sub-county level. In addition, sensitivity factors were addressed in NFS questionnaire data, where they
provided data at a well-specific level.
All of the ground-water sensitivity variables found to be associated with pesticide or nitrate detections
at or below the 0.05 level of significance are listed in Exhibit 4-2.
Overall, the results reported in Exhibit 4-2 do not present a strong pattern of relationships between
aquifer sensitivity variables and detections in individual drinking water wells. DRASTIC scores and subscores
for underlying ground waters in the NFS did not perform well as indicators of pesticide or nitrate detections
in individual drinking water wells. Although there is evidence that some DRASTIC factors for underlying
aquifers are related to pesticide or nitrate detections in either CWS or rural domestic wells, the evidence is
not recurring or consistent across both categories of wells. In general, the DRASTIC analyses reported
contain more occurrences of evidence opposing, rather than concurring with, the hypothesis that aquifer
sensitivity, as measured by DRASTIC, would correlate with detections in individual drinking water wells. The
lack of a consistent pattern linking contamination of wells with aquifer sensitivity as measured by DRASTIC
is also confirmed by the analysis of stratification variables presented in Section 3.1.
Comparison of analyses of county-level DRASTIC scores for underlying ground water with detections
in CWS wells and rural domestic wells do not yield similar results for pesticide and nitrate detections. For
example, depth to water appears to be associated with pesticide detections in CWS wells, but there is no
evidence of a similar relationship in rural domestic wells. Aquifer media is significantly related to nitrate
detections for both CWS wells and rural domestic wells at the county level, but in the opposite manner from
3 A Review of Methods for Assessing the Sensitivity of Aquifers to Pesticide Contamination. Report prepared for USEPA
by Geraghty & Miller, Inc. and ICF Incorporated, March 1991.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 107
(0 C
0
0
•o
o
1
Q.
i
z
u
0 75
0 5
cc
t.
fS (0
Ji
c E
!J
0
I
o
0 Q)
2
3
DC
1»
0
« «
5 0
II
Variable
0
§
I
gtj
0 UJ
5-8
3
(D
1
c „
Is
Q 0
0
^
*•
O ^
0 UJ
Q 0
0
T
""
IS
3£
5-5
of Measurement
rlable Name
!5
CM
O
O
+
^
o 5.
d z
1
§1 g §
0 d ° °
1 + 1 1
in
8 § § g ^ ^ 8
d d d d z z d
V
• • • + 1 1 +
c
(D ^
«liSst lit i
I'?!! 1 1 II ! I
J^CD-jj^COWC .S^ C *~
c^QZ^J~3Cw®^wIto "^
8 1 1 EJ 18-
o •£.
ts
ro
T3
2
.22
cu
CO
•s ra
CO »j
12
to S
11
QJ 3
b
.£ I
S 8
1 o
•: te
t Q
5 S
co o
« •§
™— W
S =*
1
0)
2
8
0
O)
CO
2
u
CD
1
O
f
CD
T3
CD
1 8
E 3
m o
U. Q
I 1
e §
3 «
2 13
CD
CO
O
CO
O
a «
CO JZ
2 S.
O) CD
.52 CD
o> ^
$ 8
to
*
° i_ *
<3 W e:
81
§2
«
c .3:
c fc = CD
1 8 If
| | If
O
tM
CD >?>•'=•
C DC CL
CO O Q)
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 108
that expected. No relationship was identified for rural domestic well detections and aquifer media. Overall,
there are very few occurrences of sufficient evidence to suggest a relationship supporting the DRASTIC
hypothesis.
Although DRASTIC at the sub-county level was more locally measured than DRASTIC at the first
stage, there is insufficient evidence based on the DRASTIC analyses for rural domestic wells at both stages
to suggest that DRASTIC measures of ground-water sensitivity at the sub-county level performed better in
predicting pesticide or nitrate detections in individual wells than DRASTIC at the county level. There is some
evidence that impact of vadose zone is an indicator of pesticide detections at the sub-county level, but no other
DRASTIC factors were shown to be associated with pesticide detections at either the county or sub-county
levels in rural domestic wells. As an indicator of nitrate detections, topography is the only DRASTIC variable
that appears to perform well at the county level for rural domestic wells and no DRASTIC factors at the sub-
county level are indicators of nitrate detections in individual wells. DRASTIC at the sub-county level does
appear to be an indicator of well nitrate concentrations, but this finding is not supported by analysis of county-
level DRASTIC factors with respect to nitrate concentrations, and it appears contrary to not finding a
relationship with nitrate detections.
County-Level DRASTIC
At the county level, as Exhibit 4-2 shows, the overall DRASTIC score either was not associated with
detections in individual wells, or in the case of nitrate detections for CWS wells was associated in a manner
contrary to the design of DRASTIC. If DRASTIC had functioned as expected, detections were more likely
to be associated with higher DRASTIC scores. DRASTIC was designed to address areas larger than 100 acres
in size.4 Although no upper limit for effective application of DRASTIC is defined, application to areas the
size of counties may significantly exceed the optimum feasible area, since a county-sized area of approximately
1,000 square miles equals about 640,000 acres. Prior to first stage stratification, the time necessary to map
a county of 1,000 square miles was estimated to be between 200 and 300 person hours and the time necessary
to classify all counties in the United States using the full DRASTIC system was estimated at between 300 and
450 person-years of effort.5 For purposes of first-stage stratification, however, the NFS designers concluded
some misclassification could be tolerated, if it was not extensive. Therefore, a modified DRASTIC system was
applied in the first-stage coding that could be scored using on average only a few person-hours of effort.6
The distribution of scores was found to coincide "reasonably well" with known hydrogeologic conditions in
certain ground-water regions. Because 60 percent of the scores were distributed in the moderate vulnerability
category, the likelihood of misclassification of a county between the high and low categories was reduced.
Independent rescoring of two subsamples of counties indicated that there was appreciable measurement error
in county-level scores, but the error was "not large enough to obscure actual differences in vulnerability among
counties."7 Average variations in scores were found to be probably less than 13 percent of a given mean
score.8 The DRASTIC scores therefore were found to meet the stratification needs of the NFS.
4 DRASTIC: A Standardized System, pp. 1, 11, 43-44.
5 Alexander, WJ. and S.K.LiddIe, 1986, Ground Water Vulnerability Assessment in Support of the First Stage of the
National Pesticide Survey. Proceedings of the Agricultural Impacts on Ground Water Conference, Omaha, Nebraska, pp.
77-87, National Water Well Association, Dublin, Ohio.
6 Alexander, W., S. Liddle, R. Mason, and W. Yeager, Ground-Water Vulnerability Assessment in Support of the First-
Stage of the National Pesticide Survey. February 14, 1986, pp. 1-2 (hereafter Ground-Water Vulnerability Assessment).
7 Ground-Water Vulnerability Assessment, p. 3.
8 Ground-Water Vulnerability Assessment, p. 4.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 109
Second-stage DRASTIC scoring, to identify vulnerable sub-county areas in the 90 counties chosen for
sampling of rural domestic wells, was carried out at a level of effort much closer to the estimate for county
scoring. Approximately 150 to 200 person hours were expended per county for the definition of the "cropped
and vulnerable" areas.
The DRASTIC system is designed to culminate in a single numeric value - the DRASTIC index or
overall score ~ for any hydrogeologic setting. Although individual subscores may dominate or override the
general index they generally are not interpreted separately in the DRASTIC system.9 The Phase II analysis,
however, did test for associations between detections and individual DRASTIC subscores as well as the overall
DRASTIC index. DRASTIC subscores for county-level scoring (depth to water, net recharge, aquifer media,
topography, and hydraulic conductivity) were found to be associated with detections, but primarily in the wrong
direction. These results do not indicate that the subscores are good indicators of contamination. Relatively
few associations were identified, considering the number of tests that were performed, and the associations
that were identified were not consistent over both CWS wells and rural domestic wells or over both pesticides
and nitrate. A larger number of associations were identified between DRASTIC subscores and nitrate
detections, but this is due to the larger number of nitrate detections.
For four of the seven statistically significant results reported in Exhibit 4-2 at the county level, the
possible effect was contrary to that expected from the underlying hypothesis of the DRASTIC system.10 The
results of tests of association between detections and net recharge suggest that a greater number of detections
can be expected with smaller quantities of water per land area penetrating the ground surface and reaching
the water table. DRASTIC scoring, in contrast, is based on the theory that higher rates of recharge create a
greater potential for ground-water contamination. (This result, however, is consistent with the significant
associations found between higher levels of precipitation and fewer nitrate detections discussed in Section 4.1.3
of this chapter.) The result for aquifer media, which was negatively associated with detections of nitrate in
both CWS wells and rural domestic wells, suggests that lower values for this DRASTIC subscore are associated
with detections of nitrate. DRASTIC, however, gives higher scores to aquifer media with larger grain sizes
and more numerous fractures or openings, which are expected to lead to greater permeability and less
attenuation through sorption, reactivity, and dispersion. Thus, the negative association is contrary to the
expected relationship.
The results of the analysis suggest that higher DRASTIC scores for topography are associated with
a greater likelihood of pesticide detections. Because DRASTIC gives higher scores to terrain of more gradual
(0 to 6 percent) slope, on the consideration that steeper topography will cause faster runoff and reduced
infiltration, this result supports the theory underlying DRASTIC. NFS questionnaire data for topography show
that approximately 32.5 percent of CWS wells were reported as located on a "flat valley," with about 20 percent
located on a hillside and about 21.7 percent in an unspecified topographical setting. About 34.6 percent of
rural domestic wells were reported on a "flat valley," with 33.5 percent on a hillside and 16.7 percent on a
hilltop. Analysis of topography as a questionnaire variable, however, did not reveal a significant association
paralleling the result for the analysis of the DRASTIC topography subscore.
Higher scores for hydraulic conductivity were found to be related to a greater likelihood of nitrate
detections in CWS wells. Because DRASTIC is designed to give higher scores to aquifer materials with a
greater ability to transmit water, and ground-water flow is considered to control the rate of contaminant
movement, this result corresponds to the underlying .basis upon which DRASTIC was designed. NFS
questionnaires did not collect data on hydraulic conductivity; instead, default-values were used.
9 DRASTIC: A Standardized System, pp. 80-83.
10 DRASTIC: A Standardized System, pp. 17-67.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 110
Well owners/operators of community water systems were asked if the aquifer from which water was
drawn was a confined aquifer.11 Approximately 43.5 percent of CWS wells were reported to draw from a
confined aquifer. Equivalent data were not obtained for rural domestic wells. Analysis of the
confined/unconfined aquifer questionnaire data showed that detections are associated with unconfined aquifers,
paralleling the result for the analysis of the DRASTIC hydraulic conductivity variable.
One county-level DRASTIC subscore, depth to water, was associated with pesticide detections in CWS
wells only. DRASTIC scoring gives lower ratings to situations in which the depth to water is greater, since
attenuation is presumed to be more likely as the depth to water increases. The association between higher
scores (reflecting shallower depth to ground water) for this DRASTIC component and pesticide detections
therefore is consistent with the basis of DRASTIC'S design. This result also is consistent with the results for
the analysis of the well depth variable from NFS questionnaire data, discussed under well-level variables later
in this section. Results for the analysis of this variable also are consistent with the association between
pesticide detections in CWS wells and the county-level DRASTIC subscore for depth to water. Strong
negative associations were identified between nitrate detections in both categories of wells and pesticide
detections in CWS wells and shallow wells. The association between shallow CWS wells and pesticide
detections also is consistent with findings of the Iowa State-Wide Rural Well-Water Survey. That survey,
conducted in 1988-89, reported more frequent pesticide detections and higher concentrations in wells less than
50 feet deep than in deeper wells.12
Sub-County Level DRASTIC
Sub-county DRASTIC scoring was carried out only for rural domestic well stratification. Therefore,
no tests of association are possible between detections in CWS wells and sub-county level DRASTIC scores.
No overall sub-county area DRASTIC score for underlying aquifers was found to be associated with
detections in individual wells, and only three sub-county subscores - aquifer media, impact of vadose zone,
and hydraulic conductivity - were related to detections. Of these, scores for aquifer media and hydraulic
conductivity were related to nitrate detections in the opposite manner to that expected. The sub-county
findings for the association of nitrate detections in rural domestic wells with aquifer media paralleled the
findings for the county-level analysis, but in both cases the direction of association (at almost the same
significance level) was contrary to that expected. The direction of the relation for hydraulic conductivity was
contrary to that obtained when the variable was analyzed for the county-level scoring. The result was for
nitrate detections in both cases, with hydraulic conductivity when measured at the county level associated with
detections of nitrate in CWS wells in the expected manner (e.g., a greater number of detections are associated
with higher DRASTIC scores for hydraulic conductivity) and associated with detections of nitrate in rural
domestic wells in the opposite manner when measured at the sub-county level. Thus, the finding for the
hydraulic conductivity subscore of DRASTIC is not consistent between county-level scoring and sub-county-
level scoring.
Finally, higher DRASTIC subscores for impact of the vadose zone were found to reflect a greater
likelihood of pesticide detections in rural domestic wells. The vadose zone media are presumed in DRASTIC
scoring to determine the attenuation characteristics of the material above the water table. Because higher
scores are given to materials with less attenuation potential, an association between higher scores and a greater
likelihood of ground-water contamination corresponds to the theory underlying the DRASTIC model.
11 The question was asked only for CWS wells (CWS Main Questionnaire, Question A25).
12 Iowa State-Wide Rural Well-Water Survey: Summary of State-Wide Results and Study Design,
p. 4; Perspectives of the SWRL Results, p. 1.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 111
In summary, results concerning sensitivity variables are not consistent. At the time the NPS was being
designed, the National Water Well Association cautioned that DRASTIC had not been designed to yield a
single numeric vulnerability score for a county and the problem would be magnified in larger counties. The
1985 Scientific Advisory Panel subpanel also noted that some DRASTIC parameters overlapped or were not
strictly related to actual vulnerability, and that DRASTIC pertained to shallow water vulnerability.13 The
Phase II results suggest that DRASTIC did not perform well as applied in the NPS. The poor performance
of DRASTIC measures of underlying aquifers and detections in individual wells may be due to several factors,
including measurement error, caused by measuring the DRASTIC variables on too gross a scale relative to
well sites. Site-specific factors for the individual wells sampled, such as recharge and flow of ground water,
may prevent the well from being representative of the county or sub-county vulnerability. Although wells draw
water from an aquifer that may be affected by a large recharge area, county and sub-county averages are
probably too large to produce accurate data concerning ground-water vulnerability at a particular well site.
The NAWWS also used a county-level DRASTIC score as a measure of relative ground-water
vulnerability. NAWWS constructed three well-specific hydrogeologic measures for each sample well. The first,
well-specific vulnerability, was based on a classification of the aquifer tapped by the well as surficial (most
vulnerable), confined (least vulnerable) and intermediate (moderate vulnerability). The second estimated
relative water level (low, moderate, or high) at the time of sampling and identified water level with recharge.
The third was a DRASTIC score for the aquifer tapped by the sample well.14 NAWWS concluded that
alachlor detections were associated with "highly vulnerable" wells and that the best general measure of
vulnerability was the well-specific vulnerability measure based on the type of aquifer tapped by the well. In
contrast, NAWWS, like the NPS, found that DRASTIC did not work successfully to identify vulnerable
drinking water wells.15
Well-Specific Variables
The NPS obtained strong results indicating an association between both nitrate and pesticide
detections and shallower wells. These results were consistent for nitrate in both CWS wells and rural domestic
wells and for pesticides in CWS wells.
Well-specific variables pertaining to sensitivity were obtained from NPS questionnaires rather than
DRASTIC scoring. Data were collected about the distance from the ground surface to the water surface in
the well. Respondents appeared to be aware of the depth of their wells; only about 5.5 percent of the
respondents to the CWS main questionnaire answered that they did not know the depth of the well and about
5.4 percent of respondents for rural domestic wells did not know the depth of the well. The extent of
respondent error concerning estimates of well depth, however, cannot be determined. Interviewer debriefing
found that only a small number of rural domestic well respondents (about 5%) had records from which they
13 NPS Phase I Report, Appendix A, pp. A-9 to A-10. See also Banton, O. and J.P. Villenueve, 1989, Evaluation of
Groundwater Vulnerability to Pesticides: A Comparison Between the Pesticide Drastic Index and the PRZM Leaching
Quantities. Journal of Contaminant Hydrology, 4, pp. 285-296.
14 NAWWS Project Summary, pp. 13-14.
15 NAWWS Project Summary, p. 17.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 112
could obtain a precise well depth, but interviewers concluded that about 75% of the respondents could
estimate a depth range to the nearest 50 feet.16
The results for well depth are consistent with the studies of pesticides in wells conducted by the
Minnesota Departments of Health (MDH) and Agriculture (MDA) and the results of the Iowa State-Wide
Rural Well-Water Survey. In the Minnesota studies, pesticides were detected more frequently in observation
and private drinking water wells than in public drinking water wells. MDH and MDA suggest that this
difference is most likely attributable to the shallower depths of many of the observation and private drinking
water wells.17 MDH and MDA also attribute the result to the proximity of observation and private wells
to fields receiving pesticide applications, and do not specify which association is stronger. The Iowa study
detected contaminants, including pesticides and nitrate, more frequently and at higher concentrations in wells
less than 50 feet deep than in wells deeper than 50 feet.18
4.1.2 Results for Pesticide and Fertilizer Use Factor
Use was defined for the NFS Phase II analysis as the scope and timing of applications of pesticides
and nitrogen fertilizers, and the scope and timing of activities, such as agricultural practices, that could
contribute to the presence of pesticides or nitrate in places where they could be transported into wells.
Analytic results addressing the use factor were available from a broad range of data sources. First,
a study was done to determine if the NFS data indicated any temporal effect due to disproportionate sampling
in seasons of particularly high or low pesticide and nitrate use. Data relating to use variables were then
analyzed from several sources:
• Questionnaire data pertaining to areas near the well, within 1/2 mile of the well, and
in the county, both from well owners and from county agricultural extension agents;
• County-level pesticide data available to EPA from sources outside the NFS; and
National Estimates for
Rural Domestic Wells
16 Summary of NFS Domestic Well Survey Field Interviewer Debriefing on March 9. 1990. Memorandum from Leslie
Athey to David Marker, August 16, 1990, p. 5 (Hereafter Debriefing Summary).
Data on well depth were distributed as follows:
National Estimates
for CWS Wells
Well depth:
< 20 feet
20 to 50 feet
51 to 100 feet
101 to 200 feet
201 to 500 feet
> 500 feet
Don't Know
for the Legislative Commission on Minnesota Resources, February 1988,
p. ix. Medians and ranges for well depth are summarized at pp. 78 to 88.
18 Iowa State-Wide Rural Well-Water Survey, Perspectives of the SWRL Results, p. 1. Statewide, 27.9 percent of private
drinking water wells are less than 50 feet deep, but in the western and southern portions of the state over 50 percent of the
wells are less than 50 feet deep.
Number
260
4,410
14,900
16,600
33,400
19,800
5,150
Percent
0.3
4.7
15.8
17.6
35.3
20.9
5.5
2. Schneider, Pesticides and
Well depth:
< 20 feet
20 to 50 feet
51 to 100 feet
101 to 200 feet
201 to 500 feet
> 500 feet
Don't Know
Groundwater: Surveys
Number
385,000
1,160,000
2,170,000
2,860,000
1,930,000
424,000
1,580,000
Percent
3.7
11.0
20.7
27.2
18.4
4.0
15.1
of Selected Minnesota Wells,
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 113
• County-level data on nitrogen sales and farming practices from sources outside the
NFS.
A summary of all the use variables found to be associated with pesticide or nitrate detections is
presented in Exhibit 4-3.
The evaluation of use variables led to mixed results. The investigation of pesticide use patterns
reported by NFS participants did not identify a relationship between pesticide use, as reported on NFS
questionnaires, and pesticide detections. In a number of cases, pesticides were detected in wells when neither
the well owner nor the county extension agent had reported use of that pesticide near the well.
In contrast, analysis of pesticide and nitrate use data reported from other sources found several strong
associations between use and pesticide or nitrate detections. These associations are based on county-level
pesticide and fertilizer use measures. They therefore are subject to an averaging effect, which is discussed in
Chapter 3.
NPS Pesticide and Fertilizer Use Data
Pesticides
Questionnaire responses concerning farming activities and pesticide use on the property or within one-
half mile of the well may help to explain the failure to identify a relationship between agricultural pesticide
use and pesticide detections based on questionnaire data. A large proportion of the respondents did not
indicate that any pesticides had been used within the specified time periods and proximities of the sampled
wells. Only approximately 7.4 percent of respondents reported farming on the CWS property during the past
five years, and of these only 34 percent reported pesticide use on the property. About 40.3 percent reported
crops farmed within one-half mile and only about 19.8 percent reported non-agricultural pesticide use within
one-half mile in the previous three years. For rural domestic wells, only about 11.7 percent of respondents
reported that the property upon which the well was located was farmed. That is, the wells were located in
rural areas, as defined in the NPS, but were not situated on farmed land. Of the farmed properties, about 83.6
reported use of pesticides in the past five years. Approximately 66.2 percent of rural domestic well
respondents did report that crops were farmed within one-half mile of the well in the past three years.
Whenever possible, questionnaire responses were cross-checked. In addition, interviewers for the rural
domestic well questionnaires were debriefed following data collection. The debriefings indicated that
considerable probing was necessary to obtain complete lists of pesticides used, stored, and disposed on the
property. Interviewers reported that the information obtained did reflect the respondents' best knowledge of
pesticide use around the well. However, respondents frequently were not able to estimate distances accurately
between storage and disposal sites and the well. Many respondents used considerable amounts of pesticides
for situations not explicitly covered in the questionnaire, or for livestock that did not earn at least $1,000 per
year, the reporting cutoff. Respondents' reports of disposal of pesticides and pesticide containers on the
property may underestimate the actual incidence of disposal. Interviewers also reported that the information
provided by county agricultural extension agents did not represent first-hand knowledge of pesticide use for
a particular site. County agricultural extension agents were knowledgeable about what crops were grown in
different parts of then- county, and could provide lists of common pesticides for those crops but their responses
were usually not specific to the sampled well. Finally, respondents were sometimes uncertain whether their
activities constituted use, storage, or disposal of "agricultural" pesticides.19
19 Debriefing Summary, pp. 3-5.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 114
o
ti
T3
0)
'o
o
•
a
.-e
UJ
_w
0>
<:
=
T3
(0
0)
TJ
*U
«
0)
a.
a
co
0 &
"5 n
c E
ii
I*
o
0
«
u
3 0>
*
ariable
0>
§ts
is
Q> UJ
5-S
o
1
IS
5-5
o>
II
IIB
§«
is
0> UJ
5"S
CM
s
O
+
8
CO
q
o
CO
q
o
CM
o
o
o
o
o"
o o
+1
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 115
EPA's experience in the evaluation of farming practices and pesticide application suggests five reasons
that reports of pesticide use may be inaccurate:
(1) Chemical applications may be performed in the absence of the landowner, and are
ordinarily applied in the absence of county agricultural extension agents, and may
differ from the planned or recommended application;
(2) Application equipment may be malfunctioning or poorly calibrated;
(3) Actual application rates may vary significantly even with properly calibrated
equipment;
(4) Recordkeeping can be poor, or good records may not be consulted; and
(5) Some conscious misreporting may have occurred.
The NFS results for pesticide use contrast strongly with the results obtained by the NAWWS survey.
That survey found that about 64 percent of all wells are within 300 feet of a cropped area, field, or pasture.
NAWWS estimated (for the alachlor use area) that 85 percent of the wells had row crops grown within a half
mile and 20 percent of the wells are in intense row cropping areas. NAWWS also estimated that 59 percent
of the wells in the target population had alachlor use within a half mile during the past 5 years and 50 percent
had alachlor use within the past year. Atrazine was estimated to have been used within a half mile of 55
percent of the wells during the past 5 years, and 22 percent of the wells were located in intense fertilizer
application areas.20
Nitrate
Information obtained from questionnaires on sources of nitrate also was subject to variability
stemming from several causes. Respondents were knowledgeable about the use of nitrogen fertilizer, but
frequently reported multiple applications and different rates of application on different parts of the property,
making recording of the data more difficult. When fertilizer services were used, respondents sometimes could
not provide exact information about application rates. Respondents also sometimes answered "yes" to the
presence of both a septic system and a cesspool when they had only one system.21
A significant association was identified in the analysis of NFS questionnaire data between the presence
of a pesticide retail outlet within one-half mile of a rural domestic well and nitrate detections.
Interviewers noted in connection with the questions that elicited information about pesticide retail outlets that
respondents who were concerned about the location of certain facilities near wells sought to make certain that
the interviewers recorded them. The result for pesticide retail outlets is based on a relatively small number
of positive responses. Approximately 7.3 percent of the well owners reported the presence of a pesticide retail
outlet near the well.22 Although pesticide retail outlets also frequently may be nitrogen fertilizer outlets,
20 NAWWS Project Summary, pp. 14-15.
21 Debriefing Summary, p. 4.
22 Results for this variable were not reported in the NFS Phase I Report, Appendix D. For those respondents who
answered the question with a response other than "Don't Know," the results (for Local Area Domestic Well Questionnaire,
Question lln) are as follows:
Is there a pesticide retail outlet
within one-half mile of the well:
Yes 7.3%
No 92.7%
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 116
the result must be considered in light of the relatively small number wells sampled that are near pesticide
retail outlets.
The second significant use variable identified from questionnaire data is for land used for pasture
within one-half mile of rural domestic wells. Itt is associated with a greater likelihood of nitrate detections in
rural domestic wells. Approximately 65.4 percent of respondents reported pasture within one-half mile, with
a "don't know" response rate of only about 1.9 percent.
Surrogate Measures of Pesticide and Fertilizer Use
Several variables derived from data on agricultural activities and fertilizer sales were strongly
associated with nitrate and pesticide detections. In particular, the market value of crops, which may be
considered a surrogate for pesticide use, is associated with pesticide detections in rural domestic wells. The
market value of livestock, a possible surrogate for the generation of nitrate by animals, is strongly associated
with nitrate detections in CWS wells. These variables are based on data collected by Resources for the Future
and by the Division of Resources Management at West Virginia University in conjunction with the National
Fertilizer and Environmental Research Center (NFERC).
These results represent county-wide averages and are not direct measures of application rates of
fertilizers or pesticides at the well level. The estimates of fertilizer expenditures by farmers used to develop
the county estimates are generated by allocating the NFERC state total among counties in proportion to
reported expenditures on all fertilizers, as reported by the 1987 Census of Agriculture. Expenditures on
fertilizer do not necessarily correspond exactly to tonnage acquired, but the correlation between these variables
should be high. The estimated nitrogen sales totals are weighted average state totals, where the weights
correspond to expenditures on fertilizers. Because the estimated nitrogen sales data are constructed
consistently for all counties, these data were used instead of the state-supplied NFERC data. In 1989, 30
states submitted county-level data to NFERC. Correlation analysis for states that do submit county
information indicates that the estimated values are highly correlated with the state data.
NFERC's estimates are accompanied by several caveats. The data reflect total sales of fertilizer
without regard to whether the fertilizer was purchased for agricultural or other uses; sales do not indicate
where or when the fertilizer was used; and the estimates do not include manure.23
The analysis found evidence of a relationship between nitrogen fertilizer sales data and nitrate
concentrations. Higher amounts of nitrogen sold per county acre was found to be associated with higher
concentrations of nitrate in both CWS and rural domestic wells. In addition, several Census of Agriculture
variables, particularly value of crops and livestock and acres of cropland fertilized, were found to be associated
with nitrate concentrations. The analysis did not identify relationships between the NFERC variables and
nitrate detections. Only very weak evidence for an association between pesticide detections and nitrogen
fertilizer sales was identified.
These results fall between those reported in the Minnesota and NAWWS studies. The Minnesota
survey sought to determine if there was a relationship between nitrate and pesticide occurrence and
concentration in ground water and whether nitrate testing might be a surrogate for pesticide testing. A clear
relationship between pesticide and nitrate occurrence was not observed. When pesticides were detected,
nitrate detections were also likely; but wells with detectable nitrate did not contain detectable concentrations
23 Documentation for NFERC/EPA Fertilizer Sales Data.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 117
of pesticides in 41.6 percent of the cases. Based on the results, nitrates were not found to be a reliable
indicator of pesticide occurrence or a quantitative predictor of pesticide concentration.24
In contrast, NAWWS found highly significant associations of nitrate levels exceeding 10 mg/L with
alachlor, atrazine, and simazine detections. In wells in which nitrate was detected at or above 10 mg/L,
atrazine was detected at a frequency of 43 percent, simazine at a frequency of 9 percent, and alachlor at a
frequency of 2 percent.25
4.1.3 Results for Transport Factor
Transport is defined as factors that contribute, either directly or indirectly, to the movements of
pesticide and nitrate, including precipitation, as well as factors that could inhibit the movement into well
water, such as well sheds, pads, or other protective measures for wells.
Results for the transport factor came from several sources. NFS questionnaires gathered data on well
construction and protection, water bodies and practices such as irrigation located near wells, and water bodies
in sub-county areas surrounding the well. Precipitation and drought data were collected from sources outside
the NFS.
A summary of the transport variables found to be related to pesticide and nitrate detections is
presented in Exhibit 4-4.
Transport variables, particularly those involving precipitation and the presence of water bodies near
sampled wells, show an interesting inverse relationship to detections, particularly strong with respect to nitrate
detections. Detections of nitrate in CWS wells are strongly associated with county-level precipitation variables;
the probability of detecting nitrate above the NFS minimum reporting limit of 0.15 mg/L in CWS wells is less
in counties that record high precipitation. This finding is supported for both short and long term
precipitation. Conversely, a lack of precipitation is associated with a higher probability of nitrate detections.
The results for nitrate concentrations in CWS wells provide further evidence of the same relationship.
The analysis performed using the Palmer Drought Index also shows that for CWS wells nitrate
detections and concentrations are greater in drought areas than in moist areas. Similar results, however, were
not obtained for the analysis of rural domestic wells, possibly because rural domestic well sampling occurred
in only 90 counties.
Detection of pesticides in CWS wells and rural domestic wells generally is not found to be related to
the precipitation and drought variables considered. The percent of detected pesticides in well water under
drought conditions is not dissimilar to that for near normal and moist conditions, implying that the presence
of drought is not associated with pesticide detections. Both short and long term average precipitation factors
do not appear to be related to pesticide detections in CWS or rural domestic wells. The effective sample size
for this analysis, however, limits the extension of this result beyond survey data. There is some evidence that
occurrence of intense precipitation at some point in the last five years is negatively associated with pesticide
detections, but this result is not corroborated by the other highly correlated precipitation variables.
The precipitation variables, including average precipitation occurring close to the time of sampling
and precipitation occurring one, two, and five years prior to sampling, all indicated that higher levels of
precipitation were associated with fewer nitrate detections. The result, however, was largely confined to nitrate
in CWS wells, although higher precipitation in the two years before sampling also was an indicator of fewer
24 Klaseus, T., and J. Hines, Pesticides and Groundwater: A Survey of Selected Private Wells in Minnesota. August 1989,
pp. xi, 20-21.
25 NAWWS Project Summary, p. 18.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 118
co
o
•B
0)
S
o
.•e
•a
S*
tg
X ~ c
UJ
(0
I
5
(0
|
(0
s
•o
1
V
Q.
sszz
w
as
o
| 75
05>
"5 «9
DC
1 «
5 "0
IE
§|
Eto
f*
o
ssssssss
i
_o
Q) ^
o *
Q
"5
C
o>
co £
IE
§ i
Et»
c >.
|w
O
Variable
§
a
i
^ «
£ UJ
5 "5
o
To
i
is
ss
5*5
El^S^SSS
0
"5
1
c r
O g
l£
5 o
0
3
"5
T
O ^*
•^ o
0 UJ
o'S
0
E 0
of Measure
riable Nam
r
5
CM CD ^
8 85
d do
i + i
S 55 8 B S s
o o o P P P
d do* o o o
i ' i i +i
in in u)
o o o o S ? ?- M ?-
o^^ ^ o oooo
o'dd d ci do do
V V V
iii i i 111 +
0^0)
tiii HIM 1 1
i 11 i H ! f *! Ȥ
1 ? ? ? i is 1 i 5 i if
?••§•§•§ 1 1 * i § ° si
.ET3T3 -D ^ "-OOO^S -E fe
C C C C Q . 2 .t± .ti « CD .t± 16 «
OOO O Q.Q"™"O^"'?H5 "r w_
|ff|§f§f§lffi § f 1 1 1 1
0 OT $1
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 119
pesticide detections in CWS wells. One precipitation variable, five-year precipitation, was associated with
fewer nitrate detections in rural domestic wells. Thus, the precipitation variables, all of which were measured
at the county level, consistently showed the same general tendency.
The evidence from other studies concerning pesticide detections and precipitation is mixed. Ground-
water monitoring in central South Dakota on a monthly basis from 1984 to 1989 shows a higher rate of
detections in 1988 and 1989, years in which annual precipitation was lower. No correlation was observed
between total annual precipitation and the number of pesticide detections per year.26 In contrast, the Iowa
State-Wide Rural Well-Water Survey experienced a lower frequency of pesticide detections and lower
concentrations of nitrate detections in rural private wells during 1988 and 1989, the two driest consecutive
years in the state's recorded history.27 Monitoring in Minnesota between 1986 and 1989 showed detections
at high concentrations in 1986, a wet year, and a decrease in the number and frequency of detections in 1988,
an extremely dry year in the state, suggesting that higher concentrations and frequency of occurrence are
associated with increased precipitation.
These results may suggest that factors that may lead to the frequent and extensive infiltration or runoff
of water near the well are associated with a reduced likelihood of detections, particularly for nitrate detections
in CWS wells. The evidence is insufficient to suggest the same conclusion for pesticides, and the results of
other surveys are not fully corroborative. A possible reconciliation of the NFS results for precipitation and
the results reported from other surveys is that the association of recharge and ground-water contamination
holds except for situations of extremely low or extremely high recharge. In the former case, under very dry
conditions, insufficient recharge may occur to mobilize contaminants. In the latter case, very high amounts
of recharge may lead to dilution of contaminants and a reduction in the potential for ground-water
contamination. The Minnesota Department of Agriculture is currently studying climatic influences on
pesticide concentration and detection, in part to investigate whether pesticides are immobilized during drought
conditions by material in the enlarged vadose zone and by the absence of infiltrating water to transport
pesticides.30
It is important to bear in mind, however, that several processes could be occurring. The NFS did not
collect information seeking to define the recharge area of sampled wells, and consequently cannot assess the
impact of the recharge area on results. Local rainfall may not occur in the aquifer recharge area or local
precipitation may have a smaller impact on the aquifer than cropping and pesticide use and other land use
in the recharge area. Secondly, the impact of precipitation runoff cannot be evaluated, nor can the impact
of antecedent soil moisture.
NFS results suggest that fewer pesticide and nitrate detections can be expected if surface water (rivers,
canals, bays, springs, and ponds) or drainage ditches are located within one-half mile of either CWS wells or
rural domestic wells. Similarly, fewer nitrate detections can be expected for CWS wells with drainage ditches
within 300 feet. Approximately 83 percent of CWS wells were reported to have surface water within one-half
mile, while approximately 93.8 percent of rural domestic wells were within one-half mile of such surface water
26 Kimball, C., and J. Goodman, Non-Point Source Pesticide Contamination of Shallow Ground Water. 1989 International
Winter Meeting, American Society of Agricultural Engineers, pp. 7-9.
27 Iowa State-Wide Rural Well-Water Survey: Summary of State-Wide Results and Study Design,
p. 2.
28 Minnesota Department of Natural Resources, Drought of 1988. January 1989; Klaseus, T., and J. Hines, Pesticides
and Groundwater. A Survey of Selected Private Wells in Minnesota, Minnesota Department of Health, Report Prepared for
the United States Environmental Protection Agency, August 1989.
29 DRASTIC: A Standardized System, pp. 47 and 64. The DRASTIC ranges and associated ratings do not reflect any
dilution factor.
*' Communication between Minnesota Department of Agriculture and Director, National Pesticide Survey, July 1991.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 120
bodies. Surface water within 300 feet of CWS wells and rural domestic wells also was associated with a
reduced likelihood of pesticide and nitrate detections, respectively.
A number of results at the well level also reflected the influence of transport-related factors. Rood
irrigation within one-half mile of CWS wells was associated with a higher number of nitrate detections,
although fewer than 25 percent of CWS wells reported such irrigation nearby. Leaching of nitrate to ground
water from irrigated fields has been documented.31 Presence of another operating well within 500 feet also
was associated with an increased likelihood of nitrate detections in CWS wells and of pesticide detections in
rural domestic wells.
EPA evaluated the differences in detections between small community well systems, defined as systems
with one or two wells, and large community well systems, defined as systems with more than two wells. For
the "any pesticide" variable, the difference is significant; the proportion of detections for large community well
systems is 15.0 percent and for small is 5.0 percent, with a
p-value for the difference in proportions of 0.002. For nitrate, in contrast, the difference is not significant;
the proportion of detections is 54.8 percent for large and 48.3 percent for small community well systems at
a p-value of 0.196.
4,1.4 Results for Chemical Characteristics Factor
Chemical characteristics are defined as characteristics of pesticides, such as organic partition
coefficients (K^) and half-life, or characteristics of the soil or ground-water, such as temperature, pH, and
electrical conductivity that could affect the behavior of pesticides or nitrate. A summary of the results for this
factor is presented in Fjdiibit 4-5.
Well Water Temperature
Low well water temperature is associated with detections of both pesticides and nitrate in CWS wells.
For pesticides, this result is consistent with the hypothesis that the rate of pesticide degradation will be slower
at low temperatures. The measured well water temperatures ranged from a low of 1 degree centigrade to a
high of 37 degrees centigrade, with a mean temperature of approximately 17 degrees centigrade. A recent
study of the effect of temperature and other factors on DCPA degradation found that the dissipation rate is
largely dependent on soil conditions, including soil temperature. The rate of DCPA degradation increased
as soil temperature was increased up to a maximum at 25-30 °C, after which it decreased. Soil temperature
was shown to contribute more to the rate of DCPA degradation than soil moisture content.32 The NPS
measured ground-water temperature, not soil temperature. Soil temperature varies with depth and time (daily
and seasonally). Well water temperature after purging was considered to reflect aquifer temperature.
Association of nitrate detections with lower temperature is more difficult to evaluate. Because the
process that controls the oxidation of the ammonium ion is enzymatically limited, cooler soils could be
expected to result in less nitrification of ammonium fertilizers to produce nitrate. Without consideration of
other factors, the optimum temperature for microbial activity is considered to be between approximately 30
and 35 degrees centigrade.33 Factors such as the effect of lower temperatures on plant uptake of nitrate and
31 Powers, J., and J. Schepers, 1989, Nitrate Contamination of Ground Water in North America. 26 Agriculture,
Ecosystems, and Environment, pp. 165-187.
32 Choi, J., T. Fermanian, D. Wehner, and L. Spomer, 1988, Effect of Temperature. Moisture, and Soil Texture on DCPA
Degradation. 80 Agronomy Journal 108-113.
33 Alexander, M., Nitrification, in Bartholomew, W.V. and Clark, F.E, Soil Nitrogen, American Society of Agronomy, pp.
326-327, 1965.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 121
in
(0
§
I
Q
i
«
13 £
i!
•SO
SI c
O
•5
(0
E
2
1
Q.
1
Z
o
|l
0 5
««
55
OC
b
Jl
II
1*
o
0
ts
5
_o
0 7)
o *
o
2
g
cc
V.
0
X£
ii
|£
o
0
a
S
ii
0
3
CO
I
c
° u
i!
5 "5
S
"5
1
es
0) ULI
5*
0
a
!
c c
7 °
s 15
^5 ^™
Q o
o
3
*
||
j» ^
1
E 0
i §
4) d)
i|
r
M*J
i 1 §
V
i i +
m
§ 2 ^
d d Z
V
<
1 z
o in i-
T- 1- O <
000 $
odd z
^
5
CD
C
1
Q) >, ^
1 | £
s 3-5
Q. "O CO
1 ^ § » ^
2 Q. o .2 en
i_ i_ i_ i- £
1 1 1 If S
id l ! 1 ill
I
co
.c
•o
CD
_D
O
O
I
O
i
.
CO
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 122
the effects of season or temperature on the time or rate of application of nitrogen fertilizers also could affect
the relationship.
Well Water pH
The analysis identified an association between between greater numbers of nitrate and pesticide
detections and low pH of well water.
For nitrate results this result was contrary to expectation, since nitrification is known to proceed more
slowly in acid soil. The range of pH values for samples of well water was pH 4.7 to 9.6, with a mean of 7.4.
The optimum range of pH conditions for the Nitrosomonas and Nitrobacter strains of bacteria that are
ammonium and nitrite oxidizers is between pH 7 and 9.34. Ammonium fertilizers are known to lower the
pH of soil through the nitrification process. However, tendencies toward lower pH might be expected to be
affected by the buffering capacity of most soils, and thus not to affect the pH of well water. Redox conditions
were not measured or evaluated.
Pesticide occurrence in community system wells was found to be associated with low pH; however,
domestic wells did not show a reliable association. Low pH in ground water is considered to be a condition
that enhances the persistence of pesticides in ground water because hydrolysis, a principal method of
degradation, is retarded. In addition, the acid metabolites of DCPA were the most frequently detected
chemical.
Well Water Electrical Conductivity
The analysis identified an association between nitrate detections and lower electrical conductivity in
CWS well water. This is contrary to the expected result, since nitrate in solution would be likely to increase
the conductivity of well water. Measured conductivity values ranged from 0 to 2,500 ppm total dissolved solids
(TDS) with a mean of approximately 230 ppm TDS. The survey did not collect sufficient data to explain or
analyze this result.
Persistence
The analysis indicated that half-life of pesticides in soil tends to be higher for detected pesticides than
for pesticides that were not detected. The finding of a greater likelihood of detecting pesticides with relatively
longer half-lives is consistent with the expectation that the longer the time necessary for a pesticide to break
down, the greater the possibility that the chemical can migrate below the root zone and leach to ground water.
The analysis could be performed for only 64 of the 126 pesticide analytes because accepted half-lives could
not be identified for all of the pesticide analytes. Other qualifications are described in Chapter 3.
The EPA also tested a two-variable model, the groundwater ubiquity score or GUS model, which has
been proposed as a screening tool to determine if particular pesticides are likely to leach to ground water.35
The GUS index makes use of pesticide organic partition coefficients (K^ for which no significant results were
obtained singly in the Phase II analysis) and measures of half life to calculate an index of leaching potential.
34 Alexander, pp. 327-328.
35 Gustafson, D.I., Groundwater Ubiquity Score: A Single Method for Assessing Pesticide Leachabiljty, 8 Environmental
Toxicology and Chemistry 339, 1989.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 123
The GUS index was calculated for each of the 64 chemicals in the NFS for which both half-life and
information is available.36 The predictive capabilities of the GUS index were then measured using a
logistic regression approach to determine if the GUS index could predict the probability of detection of an
NFS chemical (in either CWS or rural domestic well surveys). The p-value associated with this testing
procedure was approximately 0.02, providing evidence that the GUS index is useful for predicting pesticide
occurrence in well water samples (according to NFS data).
The mean GUS index for pesticides detected in the NFS was 3.6 (standard error of 0.7), and for non-
detected chemicals was 2.2 (standard error of 0.2). These results show comparable performance of the GUS
model and the model involving half-life only presented in Exhibit 3-8. Results of the logistic regression
procedure are presented in Exhibit 4-6.
Exhibit 4-6
Estimated NFS Logistic Regression Models
for the GUS Index
Independent
Variables
Model 1 :
GUS Index
Intercept
-2.95
Standard
Error
0.79
Beta-
Coefficient
0.52
Standard
Error
0.23
p-value
for Beta
0.023
EPA compared the GUS index analysis with the results of the analysis of half-life and K^ individually.
The GUS index does not provide more useful results. The individual analyses suggest that while half-life
predicts pesticide occurrence in well water samples, the same is not true of K^. Although the preferred model
includes a half-life term only, there is some evidence that an additive model including both half-life and
factors also provides reasonable predictions. Both models are presented in Exhibit 4-7 (the log scale is
base 10 for direct comparison with the GUS index analysis):
36 The GUS model is:
or:
GUS = \o^half-life) x (4-(Iog10(A:oc)))
GUS = 4 Iog10(/w//-/i/e) - \ogw(half-life) x log (Kx).
37 The data used for the GUS model are the same as the data used for the results presented in Section 3.8 of the NFS
Phase II Report. The logistic regression models presented in Exhibits 4-7 and 4-8 may be written, in order, as follows:
Logistic (Detection) = a + &GUS
Logistic (Detection) = a + 0Loglo(Half-life)
Logistic(Detection) = a + p,Log,0(Half-life) +
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 124
Exhibit 4-7
Estimated NFS Logistic Regression Models
for Half-Life and Koc
Independent
Variables
Model 1 :
Log (half-life)
Model 2:
Log(half-life)
Log(KJ
Intercept
-3.96
-2.61
Standard
Error
1.19
1.34
Beta-
Coefficients
1.47
1.97
-0.99
Standard
Error
0.64
0.74
0.47
p-value
for Betas
0.021
0.008
0.034
Models including interaction terms did not fit as well as either of these models. Both of these models fit the
data better than the model for the GUS index.
4.1.5 Physical Characteristics of Wells
A summary of the variables pertaining to the physical characteristics of wells that were found to be
related to pesticide and nitrate detections is presented in Exhibit 4-8. An association was found between
detections and two measures of well condition, well age and the presence of a well house over the well. Older
wells may be in worse condition than newer wells and therefore more susceptible to contamination; because
of the passage of time they may have had a greater likelihood of becoming contaminated even if they are in
good condition. About 12.7 percent of the CWS respondents and 14.9 percent of the rural domestic well
respondents did not know the age of their well.38 Interviewers estimated rural domestic well respondents'
accuracy at about +_ 5 years.39 NAWWS calculated that at least 30 percent and possibly up to 50 percent
of wells in its sampled population were over 20 years old. NAWWS did not report an association between
well age and a higher likelihood of detection.40
38 The data are distributed as follows:
Age of Well:
< 5 years
5 to 10 years
11 to 20 years
> 20 years
Don't Know
National Estimates
for CWS Wells
Number
5,900
16,300
26,700
33,600
12,000
Percent
6.2
17.2
28.2
35.6
12.7
National Estimates
for Rural Domestic
Wells
Number
1,640,000
1,710,000
3,070,000
2,520,000
1,570,000
Percent
15.6
16.3
29.3
24.0
14.9
39 Debriefing Summary, p. 4.
40 NAWWS Project Summary, p. 15.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 125
T3
Q)
J
IS
^ c
II
i- '
5
o
I
a.
2
73
CO
o>
Q.
i
h»
2
.0
Q o>
II
DC
IM
0)
« £
5 To
II
I*
o
o>
5
^^
_o
TS ^
o *
o
S
E
0)
!!
a cu
E9
o ^
O
z
CO
CO
>
0>
3
75
i
II
0 UJ
5-5
o
.2
§
^
|s
5 "5
UJ
5 "5
1
E Q)
£ E
3 (0
gz
S e
S 5
*s "S
»>
i
§^
o
C> 0
+
in
o ^~
*^ tJ
° °
i +
§
^j
0
V
1
!
c
m
o>
s
a
£
-------
Chapter Four: Evaluation of Results 126
A final, puzzling, result is the association between protection of the wellhead by a wellhouse and
pesticide detections at CWS wells. The result is based on positive questionnaire responses representing
approximately 56 percent of CWS wells. Although well sampling teams occasionally reported the presence
of stored pesticides in well houses (where well owners may keep them because they are close to a water source
for mixing) no comments by field interviewers reporting such an observation were found on any questionnaire
for a well where pesticides were detected.
4.2 Multivariate Analyses
EPA carried out multivariate regression analyses of numerous factors that were significantly associated
with detections of analytes.41 Section 2.2.4 explains the statistical procedures used to analyze the models
presented in this section. The purpose of these analyses was to investigate the relative importance of the
numerous associations and identify a combination of indicators for well contamination. Regressions were
performed for the dependent variables pesticide detections, nitrate detections, and nitrate concentrations in
both CWS and rural domestic wells. The 13 models presented represent the strongest results for
approximately 100 variables that were considered based on results of the univariate analyses presented in
Chapter 3 and Section 4.1. Results of the multivariate analyses are presented for each of the six dependent
variables in the following order: Pesticide detections in CWS wells; pesticide detections in rural domestic
wells; nitrate detections in CWS wells; nitrate detections in rural domestic wells; nitrate concentrations in
CWS wells; and nitrate concentrations in rural domestic wells. Significance levels and standard errors are
presented for estimated regression coefficients in each of the 13 models. The p-values for the coefficients
indicate the relative significance of the independent variable's effect on the dependent variable, accounting for
the effects of the other variables included in the model.
4.2.1 Pesticide Detections in CWS Wells
Exhibits 4-9 and 4-10 describe models for pesticide detections in CWS wells.
Exhibit 4-9
Estimated Logistic Regression Model for the Probability of
Detecting at Least One Pesticide in Community Water System Wells
Independent
Terms
Intercept
PASTFERT
0n(Well Depth)
Estimated
Coefficients
0.90
-64.26
-0.51
Standard
Error
1.02
27.07
0.20
p-value for
Estimated
Coefficients
0.376
0.018
0.001
The variable PASTFERT measures the proportion of acres within a county that are pasture or range
land on which fertilizers are used. The number of acres comes from the 1987 Census of Agriculture; county
size information was gathered from the 1986 County Handbook to account appropriately for the effect of large
or small counties. This is a county level measurement applied to well level data. Measurements of fertilizer
use at the well sites did not produce significant predictor variables.
41 The multivariate analyses were performed using version 5.41 of the SUDAAN software package.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 127
The second variable in the estimated model comes from the NFS Main Questionnaire item pertaining
to well depth. A logarithmic transformation was necessary to reduce the effect of a few samples from very
deep wells. Well depth is measured in feet.
The questionnaire item corresponding to presence or absence of other operating wells within 300 feet
of the sampled wells can be used in place of the well depth variable without substantially reducing the model
performance. That model is presented in Exhibit 4-10.
Exhibit 4-10
Alternative Estimated Logistic Regression Model for the Probability of
Detecting at least One Pesticide in Community Water System Wells
independent Terms
Intercept
PASTFERT
Presence of Other Operating Well
Estimated
Coefficients
-2.24
-71.05
0.89
Standard
Error
0.31
27.14
0.37
p-value for
Estimated
Coefficients
0.000
0.009
0.017
4.2.2 Pesticide Detections in Rural Domestic Wells
Exhibits 4-11 and 4-12 describe models for pesticide detections in rural domestic wells.
Exhibit 4-11
Estimated Logistic Regression Model for the Probability of
Detecting at Least One Pesticide in Rural Domestic Wells
Independent
Terms
Intercept
CROP_VAL
Estimated
Coefficients
-1.14
0.58
Standard
Error
0.60
0.19
p-value for
Estimated
Coefficients
0.061
0.002
The variable CROP_VAL measures sales of crops (including greenhouse and nursery crops) in $1,000,
prorated by county size. The sales information comes from the 1987 Census of Agriculture. County size
information was gathered from the 1986 County Handbook to account appropriately for the effect of large or
small counties. This is a county level measurement applied to well level data.
Models including more predictor variables did not perform as well in general as the model containing
only the CROP_VAL variable. Inclusion of a variable measuring the number of beef cows per county acre
(again from the 1987 Census of Agriculture) results in a model that performs almost as well as the one
presented (Exhibit 4-12). However, the BEEFCOW variable is highly negatively correlated with CROP_VAL
and enters the model with marginal significance only.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 128
Exhibit 4-12
Alternative Estimated Logistic Regression Model for the Probability of
Detecting at Least One Pesticide in Rural Domestic Wells
Independent
Terms
Intercept
CROP_VAL
BEEFCOW
Estimated
Coefficients
-3.53
0.61
-0.50
Standard
Error
1.36
0.21
0.23
p-value for
Estimated
Coefficients
0.011
0.004
0.034
4.2.3 Nitrate Detections in CWS Wells
Exhibits 4-13 and 4-14 describe models for nitrate detections in CWS wells.
Exhibit 4-13
Estimated Logistic Regression Model for the Probability of
Detecting Nitrate in Community Water System Wells
Independent Terms
Intercept
Average Monthly Precipitation
PH
PASTFERT
Estimated
Coefficients
5.67
-0.45
-0.55
-21.51
Standard
Error
1.35
0.10
0.17
7.47
p-value for
Estimated
Coefficients
< 0.00005
< 0.00005
0.001
0.004
Measures of precipitation, physical properties of the well water, and pesticide use are related to nitrate
detections in CWS wells. Average monthly precipitation is measured in inches and covers the 12 months prior
to sampling for each well. PASTFERT was described in Section 4.2.1, and pH measures the relative acidity
or basicity of the well water.
The CWS Main Questionnaire item Cl pertaining to farming on the property on which a CWS well
is located can be included in an alternate multivariate model (Exhibit 4-14). However, the variable's statistical
significance is comparatively low and it is highly negatively correlated with PASTFERT, suggesting that the
information it contains is already partially included in the model presented in Exhibit 4-12.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 129
Exhibit 4-14
Alternative Estimated Logistic Regression Model for the Probability of
Detecting Nitrate in Community Water System Wells
Independent Terms
Intercept
Average Monthly Precipitation
PASTFERT
PH
Farming on the Well Property
Estimated
Coefficients
5.78
-0.44
-22.66
-0.58
1.19
Standard
Error
1.40
0.10
8.03
0.18
0.53
p-value for
Estimated
Coefficients
0.0001
< 0.00005
0.005
0.001 1
0.025
4.2.4 Nitrate Detections in Rural Domestic Wells
Exhibits 4-15 and 4-16 describe models for nitrate detections in rural domestic wells only.
Exhibit 4-15
Estimated Logistic Regression Model for the Probability of
Detecting Nitrate in Rural Domestic Wells
Independent Terms
Intercept
Unlined Drainage Ditch < 1/2 mile
PH
Maximum Monthly Precipitation
Age of Well
Estimated
Coefficients
11.25
-1.43
-1.15
-0.22
0.03
Standard
Error
1.82
0.31
0.20
0.06
0.01
p-value for
Estimated
Coefficients
< 0.00005
< 0.00005
< 0.00005
0.0004
0.001
Data on the presence or absence of unlined drainage ditches near the sampled wells was obtained
using the Local Area Questionnaire (item 12d). Age of the wells was obtained using the Main Questionnaire
information on year of construction. Maximum monthly precipitation measures the greatest precipitation in
any one month during the five years prior to sampling.
The variable PASTFERT can be included in this model to provide marginal improvement in
performance (p-values for the other four variables do not change substantially) (Exhibit 4-16).
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 130
Exhibit 4-16
Alternative Estimated Logistic Regression Model for the Probability of
Detecting Nitrate in Rural Domestic Wells
Independent Terms
Intercept
Maximum Monthly Precipitation
Unlined Drainage Ditch < 1/2 mile
PH
Age of Well
PASTFERT
Estimated
Coefficients
13.41
-0.27
-1.53
-1.15
0.03
0.33
Standard
Error
2.03
0.07
0.33
0.15
0.01
0.15
p-value for
Estimated
Coefficients
< 0.00005
0.00002
< 0.00005
< 0.00005
0.0003
0.031
4.2.5 Nitrate Concentrations in CWS Wells
Exhibits 4-17 and 4-18 describe models for nitrate concentrations in CWS wells. Analyses were
conducted for wells with nitrate detections.
Exhibit 4-17
Estimated Linear Regression Model for the Logarithm of
Nitrate Concentrations in Community Water System Wells
Independent Terms
Intercept
Maximum Monthly Precipitation
Electrical Conductivity
Nitrogen Sales
Cn(Well Depth)
Moist (PDI)
Drought (PDI)
Estimated
Coefficients
0.59
-0.07
0.0008
0.12
-0.12
-0.76
-0.26
Standard
Error
0.33
0.02
0.0002
0.03
0.03
0.20
0.16
p-value for
Estimated
Coefficients
0.0725
< 0.00005
< 0.00005
< 0.00005
0.0002
0.0001
0.1073
Maximum monthly precipitation represents the amount of precipitation in inches per month and
covers the five years prior to sampling for each well. Electrical conductivity measures the total dissolved solids
(parts per million) in the well water. Nitrogen sales data conies from the 1987 Tennessee Valley Authority
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 131
NFERC/EPA database and measures the total sales in tons of nitrogen from July 1, 1986, to June 30, 1987,
for the county in which each well was sampled. The sales data are prorated by county size to account
appropriately for the effect of large or small counties. The well depth variable comes from the NFS Main
Questionnaire item pertaining to well depth. A logarithmic transformation was necessary to counter the effect
of a few extreme observations. Well depth is measured in feet.
The Palmer Drought Index (PDI) measures the extent to which counties have experienced unusual
drought or moist conditions in the past year. The variable is categorized for this analysis into three categories
indicating drought, normal, and moist conditions. The third condition is calculable from the variable in the
model presented above due to the constraint that the sum of the three PDI coefficents must be zero (i.e., the
moist indicator has a coefficient of -0.76, the drought indicator has a coefficient of -0.26, and the normal
category has a coefficient of 1.02).
The variable CROP_VAL can be used in place of the nitrogen sales variable without substantially
reducing the performance of the model (Exhibit 4-18).
Exhibit 4-18
Alternative Estimated Linear Regression Model for the Logarithm of
Nitrate Concentrations in Community Water System Wells
Independent Terms
Intercept
Maximum Monthly Precipitation
Conductivity
CROPVAL
?n(Well Depth)
Moist (PDI)
Drought (PDI)
Estimated
Coefficients
1.20
-0.07
0.0009
3.23
-0.11
-0.76
-0.32
Standard
Error
0.25
0.02
0.0002
0.70
0.03
0.20
0.16
p-value for
Estimated
Coefficients
< 0.00005
< 0.00005
< 0.00005
< 0.00005
0.0008
0.0002
0.0463
4.2.6 Nitrate Concentrations in Rural Domestic Wells
Exhibits 4-19, 4-20, and 4-21 provide models for nitrate concentrations in rural domestic wells.
The first two variables included in this model are described in previous sections. The "body of water"
variable consists of a composite measure from several questions in the Local Area Questionnaire (questions
12a,b,d,g,h,i,j,k,l) and pertains to unlined, rather than lined, water bodies. The DRASTIC topography factor
was measured at the second stage of the rural domestic well survey and pertains to the slope of the land in
the hydrogeologic settings in which the sampled wells reside.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 132
Exhibit 4-19
Estimated Linear Regression Model for the Logarithm of
Nitrate Concentrations in Rural Domestic Wells
Independent Terms
Intercept
0n(Well Depth)
CROP_VAL
Body of Water within 1/2 mile
DRASTIC Topography Factor
Estimated
Coefficients
0.54
-0.12
0.21
0.48
0.03
Standard
Error
0.34
0.03
0.06
0.16
0.01
p-value for
Estimated
Coefficients
0.117
0.0001
0.0002
0.003
0.004
The presence of bodies of water within 300 feet of the well (information obtained from the Well
Observation Record question 6b) can be added to this model to provide marginal improvement in
performance. (Exhibit 4-20) Also, the nitrogen sales variable described in the previous section can be used
in place of the CROP_VAL variable without substantially altering model performance
(Exhibit 4-21).
Exhibit 4-20
Alternative Estimated Linear Regression Model for the Logarithm of
Nitrate Concentrations in Rural Domestic Wells
Independent Terms
Intercept
fri(Well Depth)
CROP_VAL
DRASTIC topography factor
Body of Water within 1/2 mile
Body of Water within 300 feet
Estimated
Coefficients
0.51
-0.12
0.21
0.03
0.51
-0.004
Standard
Error
0.35
0.03
0.01
0.06
0.17
0.001
p-value for
Estimated
Coefficients
0.155
0.0001
0.0004
0.005
0.003
0.020
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 133
Exhibit 4-21
Alternative Estimated Linear Regression Model for the Logarithm of
Nitrate Concentrations in Rural Domestic Wells
Independent Terms
Intercept
fn(Well Depth)
DRASTIC topography factor
Body of Water within 1/2 mile
Body of Water within 300 feet
Nitrogen Sales
Estimated
Coefficients
-1.38
-0.13
0.024
0.45
-0.009
0.17
Standard
Error
0.49
0.03
0.009
0.15
0.001
0.06
p-value for
Estimated
Coefficients
0.006
< 0.00005
0.008
0.003
< 0.00005
0.007
Results for the multivariate regression models identified in the Phase II analysis can be compared with
multivariate models identified in the NAWWS. That study identified a model for predicting the occurrence
of nitrate at or above concentrations of 3 mg/L. That model included season, the DRASTIC soil media
component, well-specific vulnerability (measured with respect to the aquifer as high for surficial, low for
confined, and moderate), and the percent of land within one-half mile with nitrogen fertilizer applied.
NAWWS achieved similar results using the measured concentration of nitrate as the dependent variable in
a multiple linear regression. In contrast to the models presented in this section, multivariate models identified
in the NAWWS generally include both a use and a vulnerability component. NAWWS also concluded that
no adequate linear combination of the seven DRASTIC components suitable for prediction of pesticide or
nitrate occurrence could be identified. Some DRASTIC components were associated with detections.42
For pesticides, NAWWS sought variables that best predict the occurrence of detectable levels of
herbicides. The best models contained variables for vulnerability and use.
4.3 Potentially Limiting or Confounding Factors
This section reports two analyses that were conducted to investigate phenomena that might have
limited or confounded the results of the statistical studies. First, in order to evaluate whether the detections
of pesticides and nitrate in the NFS were related to point sources of contamination, a review was conducted
on the data pertaining to each well in which a detection occurred. The review found that detections could not
be attributed to point sources. Second, to assess the constraints placed on the analysis by the effective number
of detections, an analysis of the relationship of MRLs and detections was carried out, and concentration
distributions for detected analytes were developed with which to estimate the presence of pesticides and nitrate
at concentrations below their respective MRLs.
42 NAWWS Project Summary, pp. 19-20.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 134
4.3.1 Review of Well-Specific Data
In addition to the statistical analysis of NFS questionnaire items, EPA reviewed the records for each
of the wells at which a detection occurred. Each questionnaire was examined for every well with a pesticide
detection. The primary objective of the review was to explore the idea that site-specific criteria such as
pesticide spills or poor well construction were related to the detection. Point sources of contamination for
the purpose of this review were defined as controllable factors that are affected by human participation (e.g.,
spills, mismanagement of chemicals) other than the normal application of pesticides within recommended label
rates for their intended use.
Answers to selected individual questions, well diagrams and comments placed in the margin of
questionnaires by interviewers, and sketches of the area within 300 feet of the well were examined for
information that might lead to insights into the causes of contamination at that particular well. Interviewers'
comments on well construction and descriptions of unusual situations near the well, such as materials stored
near the well, were analyzed for data that were not part of the items specifically asked for in the
questionnaires. Questionnaire items were crosschecked and property sketches reviewed to determine the
distances of land features around the well (drainage ditches, septic tanks, farmland, cropland, wooded areas,
buildings, other wells) that might explain site-specific well detections.
The review did not identify any situations in which a detection could be traced clearly to a point
source of contamination. In one case, a pesticide manufacturing facility was identified within one-half mile
of a well and in another pesticide storage was identified within 300 feet of the well pump. In both cases,
however, the specific pesticides detected in the wells and the pesticides associated with the identified activities
did not match.
4.3.2 Review of Effects of Detection Limits
Statistical analysis of Survey data was limited by the number of detections that were included in the
analysis. In particular, a modest increase in the number of pesticide detections available for analysis might
have substantially affected the findings and conclusions. The number of detections available for analysis was
affected by a number of elements of the Survey design, including the number of samples collected for the
survey, and the use of minimum reporting limits to achieve a high level of confidence in those detections and
concentrations that were reported. The MRLs were intended to eliminate false positives, but also ensured that
the number of positive detections available for analysis in Phase II was smaller than it would otherwise have
been.43 For example, the proportion of atrazine detections nationally was estimated to be 1.7% and 0.7%
respectively for CWS wells and rural domestic wells. These rates of detection were too low to allow
reasonable statistical analysis similar to analyses presented in this report for nitrate and the "any pesticide"
group. Other studies such as the Monsanto National Alachlor Well Water Survey used much lower detection
limits for atrazine and consequently reported higher detection rates.44 It is clear that atrazine probably
occurs in some wells at levels lower than its NPS MRL of 0.12 /ig/L, and use of a lower detection limit would
43 Support for the use of MRLs and descriptions of the procedures to be followed for establishing detection and reporting
limits are provided in Glaser, J.A, D.L. Foerst, G.D. McKee, S.A Quare, W.L. Budde, December 1981, Trace Analysis for
Wastewaters. Environmental Science and Technology, 15(12), pp. 1426-1435; Keith, L.H., W. Crummett, J. Deegan, R.
Libbey, J. Taylor, G. Wentler, Principles of Environmental Analysis. 55 Analytical Chemistry, 1983, pp. 2210-2218; Kirchner,
CJ., Quality Control in Water Analysis. 17(4) Environmental Science and Technology, 1983, pp. 174A-181A; and Long, G.L.
and J.D. Winefordner, Limit of Detection. A Closer Look at the ITJPAC Definition. 55(7) Analytical Chemistry, June 1983,
pp. 712A-724A Arguments for reporting all data, with documentation of their limitations, are found in Klein, L.H., Report
Results Right! Part 1. June 1991, CHEMTECH, pp. 352-356; Klein, L.H., Report Results Right! Part 2. August 1991,
CHEMTECH, pp. 486-489; Porter, P.S., R.C. Ward, H.F. Bell, The Detection Limit. 22(8) Environmental Science and
Technology, 1988, pp. 856-861; and Lambert, D., B. Peterson, and I. Terpenning, Non-Detects. Detection Limits, and the
Probabilities of Detection. 86 Journal of the American Statistical Association, June 1991, pp. 266-277.
44 For the Monsanto National Alachlor Well Water essentially all positive values were reported as detections; the rate
of detection of atrazine was estimated to be approximately 12% for the alachlor use area as defined by Monsanto.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 135
permit more data analysis than is currently possible. EPA carried out two studies to estimate the influence
of MRLs on detections. The first, reported in Section 4.3.2.1, addressed the association between pesticide and
nitrate detections and MRLs. The second, described in Section 4.3.2.2, involved the preparation of estimated
concentration distributions for nitrate and DCPA acid metabolites, and compared the MRLs to those
concentration distributions to estimate the likelihood that pesticides were present in wells at concentration
levels below the MRLs for those analytes.
4.3.2.1
Analysis of Association of Detections and MRLs
EPA's analysis to examine the relationship between MRLs and detections of pesticides was carried
out for the 111 quantifiable analytes. Because the analysis was chemical-specific, the Survey weights were not
required. The distribution of the MRL data is heavily skewed so a logarithmic transformation was used.
Exhibit 4-22 provides summary statistics for the MRLs for the detected and non-detected groups of
pesticides respectively. Although the mean MRL is greater for non-detected pesticides, the difference between
the means for detected and non-detected pesticides is not statistically significant. That is, the analysis indicated
that there was no significant difference between the MRLs of analytes that were detected and those that were
not. The descriptive significance level associated with tests of difference between the means, transformed to
the logarithmic scale, is approximately 0.1.
Exhibit 4-22
Summary Statistics for NFS MRLs by Detected and
Non-Detected NFS Pesticides
Pesticide Group
Detected
Non-Detected
Mean (|ig/L)
1.23
2.09
Standard Error
0.73
0.72
Mean of the Logarithm
-1.21
-0.54
Standard Error
0.52
0.13
4.3.2.2 Analysis of Relationship of MRLs to Concentration
Distributions
In order to examine further the effect of MRLs on pesticide and nitrate detections in wells, EPA
developed estimated concentration distributions.
The procedures followed are described in Section 2.2.4.7. Results for the concentration estimates,
reported in Section 4.4.2, suggest that approximately 0.9% of rural domestic wells (95,000 wells) contain DCPA
acid metabolites at concentrations below the MRL of. 10 ng/L; and approximately 7.3% (770,000 wells) contain
nitrate at concentrations below the MRL of .15 mg/L. Results for CWS wells suggest that 4.3% (4,000 wells)
contain nitrate at concentrations below the MRL. The statistical procedure could not generate reasonable
estimates for the number of CWS wells with DCPA acid metabolites below the MRL.
A comparison of the national estimates of wells containing nitrate, DCPA acid metabolites from Phase
I and Phase II is presented in Exhibit 4-23. Phase I estimates are based on detections above the MRL only,
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 136
Exhibit 4-23
Estimated Number and Percent of Wells* Containing
Nitrate or DCPA Acid Metabolites
Analyte
Nitrate
DCPA Acid
Metabolites
Community Water System Wells
Phase I*
49,300 52.1%
6,010 6.4%
Phase II**
53,000 56.1%
— —
Rural Domestic Wells
Phase 1*
5,990,000 57.0%
264,000 2.5%
Phase II**
6,720,000 64.1%
315,000 3.2%
(Based on estimated 94,600 CWS wells and 10.5 million rural domestic wells)
* Phase I estimates are based on detections above the MRL.
** Phase II estimates are based on estimated concentration distributions including concentrations below
the MRL.
whereas Phase II estimates are based on the modeled distributions including concentrations below the MRLs.
The modeled distributions can also be used to estimate the number of wells containing chemicals above the
MRL; these estimates should be very close to the Phase I estimates but may not be exactly the same due to
the modeling procedure used (see Section 2.2.4.7). Exhibit 4-23 indicates that many wells tested during the
survey could contain specific pesticides or nitrate below levels the analytical laboratories could reliably
quantify. It is also possible that many of the analytes not detected in the survey may be present in
concentrations below their respective reporting limits.
4.4 Exposure and Risk Estimates
This section describes the results for national estimates of adult exposures to concentrations of DCPA
acid metabolites and nitrate, and infant exposures to nitrate, through consumption of these compounds in
drinking water drawn from community water systems and rural domestic wells sampled as part of the National
Pesticide Survey (NFS). Estimates have also been made of the populations exposed to at least one pesticide
and to at least one pesticide above health-based levels. These estimates are based on procedures similar to
those used to produce estimates for the NPS Phase I Report.
The results of the exposure and risk analyses are presented in the following sections. Section 4.4.1
presents estimates of the concentration distributions for nitrate and DCPA acid metabolites for rural domestic
wells and for community water system wells. Section 4.4.2 presents the corresponding population exposure
and risk estimates for the analytes in each survey. The major findings of these analyses are summarized in
Section 4.4.3.
As a result of Survey design constraints and the limitations of the available NPS data, the exposure
and risk estimates developed in this exercise must be viewed with caution. The NPS was designed to be a
national, statistical study of the occurrence of pesticides and nitrate in public and private drinking water wells.
The Survey was not designed to provide estimates of population exposure and risk. The development of
exposure and risk estimates is further complicated by the small number of detections for all NPS analytes
except nitrate. The two analytes included in the development of concentration distributions and resultant
exposure and risk estimates are those with the highest frequency of detection in the NPS. A large number
of assumptions had to be made in the development of these estimates. The reader is strongly encouraged to
carefully review these assumptions, which are summarized in Sections 2.2.4.7 and 2.2.4.8, and note that proper
interpretation of the results presented in this section can only be made in light of these caveats.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 137
4.4.1 Concentration Distributions
Estimates of the distribution of concentration of nitrate and DCPAacid metabolites for rural domestic
wells and for community water systems are presented in this section. The estimated distributions are
summarized by the following three parameters (see Section 2.2.4.7 for a detailed discussion of statistical
procedures):
JT The probability of non-occurrence;
\i The mean of the foi(concentrations) given occurrence; and
a The standard deviation of the ^(concentrations) given occurrence.
Assuming that the concentrations of analytes in contaminated wells follow a lognormal (p,,a) distribution, the
dn(concentrations) follow a normal distribution with mean /z, and standard deviation a. In the following
exhibits, estimates of the three defining parameters of the estimated concentration distributions, n, p., and a,
are presented along with estimates for medians and means of the distributions. Confidence intervals are also
presented.
Exhibit 4-24 presents a national concentration distribution of nitrate in rural domestic wells.
Exhibit 4-24
Estimates of the Nitrate Concentration Distribution
for Rural Domestic Wells
Parameter
it
H
o
median concentration
mean concentration
95th percentile
99th percentile
median given occurrence
mean given occurrence
Estimate
35.9%
0.04
1.49
0.31 mg/L
2.00 mg/L
8.47 mg/L
26.00 mg/L
1.04 mg/L
3.09 mg/L
95% Confidence Interval
Lower Bound
30.7%
-0.30
1.31
0.18 mg/L
1.60 mg/L
6.87 mg/L
19.85 mg/L
0.74 mg/L
2.53 mg/L
Upper Bound
43.5%
0.28
1.68
0.48 mg/L
2.56 mg/L
10.72 mg/L
34.73 mg/L
1.32 mg/L
3.91 mg/L
As Exhibit 4-24 demonstrates:
• Approximately 64% of rural domestic wells contain nitrate (the parameter n represents the
probability of non-occurrence and shows that approximately 36% do not contain nitrate);
• Approximately 7.3% of rural domestic wells (770,000 wells) are estimated to contain nitrate
at concentrations below the minimum reporting limit.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 138
• The median concentration of nitrate in all rural domestic wells is approximately 0.3 mg/L.
The mean concentration is approximately 2.0 mg/L;
• The median concentration of nitrate in rural domestic wells in which nitrate occurs is
approximately 1.0 mg/L. The mean concentration in those wells is approximately 3.1 mg/L;
and
• In the NFS Phase I Report, EPA estimated that approximately 2.4% of rural domestic wells
(254,000 wells) contain nitrate above the MCL of 10 mg/L. The corresponding estimate
from the estimated lognormal model used in the Phase II analysis is 4.0% (with 95%
confidence bounds of 3.0% and 5.4%).
In addition to these results for well concentrations, the estimated proportion of wells containing
nitrate at concentrations greater than the minimum reporting limit of 0.15 mg/L is 56.8% compared to an
estimated 57.0% developed in Phase I.46
The estimated concentration distribution of nitrate in rural domestic wells is presented graphically in
Exhibit 4-25. The proportion corresponding to non-occurrence is not included on the graph; consequently
the probability area encompassed by the graph is approximately 64.1% (i.e., I-JT). The graph shows that the
mode of the distribution, given occurrence, is close to the MRL for nitrate of 0.15 mg/L.
Exhibit 4-26 presents a national concentration distribution of DCPA acid metabolites in rural domestic
wells. As Exhibit 4-26 demonstrates:
• Approximately 3% of rural domestic wells contain DCPA acid metabolites;47
• Approximately 0.6% of rural domestic wells (63,000 wells) are estimated to contain
DCPA acid metabolites at concentrations below the minimum reporting limit of 0.1
The median concentration of DCPA acid metabolites in rural domestic wells in which
it occurs is approximately 0.29 /ig/L. The mean concentration of DCPA acid
metabolites in these wells is approximately 0.43 /*g/L.
45 National Survey of Pesticides in Drinking Water Wells. Phase I Report. EPA 570/9-90-015, November 1990, p. 66.
The 95% confidence interval bounds are 1.2% and 4.4%.
* The 95% confidence interval lower and upper bounds of the 56.8% estimate are 51.3% and 62.8%. The 95%
confidence interval bounds of the 57.0% estimate are 50.3% and 63.8%.
47 The 40 re-samples used to develop this estimate yielded two distinct sets of results. The first set, consisting of 38 of
the 40 re-samples, estimates the probability of non-occurrence to be approximately 97%; the second set estimates the same
parameter to be 0%. The problem for the second set is that the re-samples contain positive data that does not allow the
maximum likelihood estimation procedure to converge to reasonable estimates. For these two re-samples, the maximum
likelihood procedure is unable to distinguish between non-occurrence and occurrence at very low levels (e.g., 10'10 pg/L).
Reporting the median, as opposed to the mean, of the re-sample estimates reduces the effect of these two re-sample values
on the reported estimates, but the lower confidence bound is largely based on the re-sample values. Although the maximum
likelihood estimation procedure is unable to distinguish between non-occurrence and occurrence at very low levels, the effect
on estimates of percentiles in the right tail of the distribution (i.e., higher concentrations) is not as pronounced.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 139
Exhibit 4-25
Estimated Distribution of Nitrate Concentration
in Rural Domestic Drinking Water Wells*
0.6 H
468
Concentration (mg/L)
10
12
The estimated distribution is the conditional probability for wells containing nitrate only. The estimated
proportion of all rural domestic wells in which nitrate occurs is approximately 64.1% (corresponding
approximately to 6,740,000 rural domestic wells). The MCL for nitrate is 10 mg/L. Nitrate occurs above
this level in approximately 425,000 rural domestic wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 140
Exhibit 4-26
Estimates of the DCPA Acid Metabolites Concentration Distribution
for Rural Domestic Wells*
Parameter
i
ii
a
median concentration
mean concentration
95th percentile
99th percentile
median given occurrence
mean given occurrence
Estimate
97.0%
-1.23
1.24
o/ig/L
0.017 itQ/L
OnQ/L
0.407 ng/L
0.29 uglL
0.53 uglL
Lower Bound
0.00%
-7.90
0.38
OMg/L
0.007 tiQ/L
OnQlL
0.143 /tg/L
0.00 jig/L
0.02 M9/L
Upper Bound
99.2%
-0.19
2.80
0.000 /tg/L
0.034 nQ/L
0.04 Mg/L
0.82 ugll
0.83 Mg/L
1.13/xg/L
* Concentrations entered as 0 imply non-occurrence. Entries of 0.00 imply
occurrence at very low levels.
The estimated proportion of wells containing DCPA acid metabolites at concentrations greater than
the minimum reporting limit of 0.1 /ig/L is 2.4%, compared to an estimate of 2.5% generated in Phase I.48
The estimated concentration distribution of DCPA acid metabolites in rural domestic wells is
presented graphically in Exhibit 4-27. The probability area encompassed by the graph is approximately 3%,
corresponding to the estimated proportion of occurrence of DCPA acid metabolites in rural domestic wells
nationally. This graph shows that the mode of the distribution, given occurrence, is below the MRL of 0.1
/ig/L for DCPA acid metabolites.
Exhibit 4-28 presents a national concentration distribution of nitrate in CWS wells. As Exhibit 4-28
demonstrates:
• Approximately 56% of CWS wells contain nitrate;
• Approximately 4.4% of CWS wells (4,000 wells) are estimated to contain nitrate at
concentrations below the minimum reporting limit of 0.15 mg/L.
• The median concentration of nitrate in CWS wells is approximately 0.2 mg/L. The mean
concentration is approximately 1.5 mg/L; and
• The median concentration of nitrate in CWS wells in which nitrate occurs is approximatley
1.0 mg/L. The mean concentration in those wells is approximately 2.6 mg/L.
48 The 95% confidence interval lower and upper bounds for the Phase II estimates are 0.7% and 3.4% respectively. The
95% bounds for the Phase I estimates are 1.2% and 4.5%.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 141
Exhibit 4-27
Estimated Distribution of DCPA Acid Metabolites Concentration
in Rural Domestic Drinking Water Wells*
0.09-n
0.5
1 1.5 2
Concentration (|ig/L)
The estimated distribution is the conditional probability for wells containing DCPA acid metabolites only.
The estimated proportion of all rural domestic wells in which DCPA acid metabolites occur is
approximately 3.3% (corresponding approximately to 338,000 rural domestic wells). The MCL for DCPA
acid metabolites is 4,000 pg/L. DCPA acid metabolites were found only at much lower levels in rural
domestic wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 142
Exhibit 4-28
Estimates of the Nitrate Concentration Distribution
for Community Water System Wells
Parameter
n
n
o
median concentration
mean concentration
95th percentile
99th percentile
median given occurrence
mean given occurrence
Estimate
43.6%
0.02
1.35
0.21 mg/L
1.46 mg/L
6.48 mg/L
17.57 mg/L
1.02 mg/L
2.61 mg/L
95% Confidence Interval
Lower Bound
36.8%
-0.25
1.13
0.11 mg/L
1.20 mg/L
5.28 mg/L
13.26 mg/L
0.78 mg/L
2.13 mg/L
Upper Bound
49.8%
0.36
1.52
0.33 mg/L
1.75 mg/L
7.63 mg/L
22.09 mg/L
1.44 mg/L
3.13 mg/L
The estimated proportion of CWS wells containing nitrate at concentrations greater than the minimum
reporting limit of 0.15 mg/L is 52.0%, which parallels an estimate of 52.1% obtained in Phase I.49
The estimated distribution for nitrate concentrations in CWS wells is presented graphically in Exhibit
4-29. The proportion corresponding to non-occurrence of nitrate in CWS wells (approximately 43.6%) is not
included in this representation. The graph shows that the mode of the concentration distribution, given
occurrence, is approximately 0.3 mg/L.
Exhibit 4-30 presents the estimated distribution of DCPA acid metabolite concentrations in CWS
wells. As Exhibit 4-30 demonstrates, the estimation procedure failed to distinguish between non-occurrence
and occurrence at very low levels (e.g., 10"10 /zg/L). Consequently, the estimated proportion of wells that
contain DCPA acid metabolites is estimated to be 100%, but CWS wells are expected to contain DCPA acid
metabolites at extremely low levels as indicated by the median value of 0.003 /zg/L. Although the method used
does not appear to provide a good model of DCPA acid metabolites concentrations near zero, it does provide
more reasonable estimates for the mean and median of the distribution and for the higher quantiles presented.
For example, the estimated proportion of wells containing DCPA acid metabolites above its MRL of 0.1
is 9.1% compared to a value of 6.4% determined in Phase I.51
49 The 95% confidence interval lower and upper bounds for the Phase II estimate are 47.8% and 56.1% respectively.
For the Phase I estimate the bounds are 48.8% and 56.3%.
50 The 40 re-samples used to develop the estimates presented in Exhibit 4-29 yielded 37 estimated models for which the
probability of occurrence was 100%. For eight of the 37 re-samples the maximum likelihood procedure did not converge
fully. The estimated models for these 8 resamples are based on partial convergence only. The effect of including these 8
estimated models to produce bootstrap estimates is not likely to be pronounced because of their similarity to most of the
other re-sample estimated models.
51 The 95% confidence interval lower and upper bounds for the Phase II estimates are 6.3% and 14.5%. For the Phase
I estimates the bounds are 3.4% and 9.3%.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 143
Exhibit 4-29
Estimated Distribution of Nitrate Concentration
in Community Water System Wells*
0.45
0-
468
Concentration (mg/L)
10
12
The estimated distribution is the conditional probability for wells containing nitrate only. The estimated
proportion of all CWS wells in which nitrate occurs is approximately 56.4% (corresponding approximately
to 53,300 CWS wells). The MCL for nitrate is 10 mg/L. Nitrate occurs above this level in approximately
2,500 CWS wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 144
Exhibit 4-30
Estimates of the DCPA Acid Metabolites Concentration
Distribution for Community Water System Wells
Parameter
T
n
a
median concentration
mean concentration
95th percentile
99th percentile
median given occurrence
mean given occurrence
Estimate
0.0%
-5.69
2.65
0.003 nQ/L
0.11 uglL
0.25 /ig/L
1.6209/1
0.003 nQ/L
0.12 uglL
Lower Bound
0.00%
-6.66
1.75
0/ig/L
0.05 /ig/L
0.16 /tg/L
0.65 M9/L
0.001 nQlL
0.05 M9/L
Upper Bound
88.7%
-1.48
3.20
0.016 jig/L
0.23 iig/L
0.34 jig/L
2.85 ng/L
0.23 ng/L
1.08 jig/L
The estimated distribution of .concentrations of DCPA acid metabolites for CWS wells is presented
graphically in Exhibit 4-31. As the concentration approaches zero, the probability density function value
becomes very large. The graph portrays the entire probability distribution (except for the right tail which
contains close to zero probability) since the estimated probability of non-occurrence of DCPA acid metabolites
in CWS wells is 0%. Exhibit 4-30 demonstrates the mathematical procedure's inability to distinguish between
non-occurrence and occurrence at very low concentrations.
4.4.2 Goodness-of-Fit of the Estimated Concentration Distributions
Goodness-of-fit analysis of the estimated concentration distributions was accomplished by comparing
quantiles of the estimated distribution to quantiles of the empirical (observed) distribution. Two approaches
were taken for this comparison. The first was to calculate both sets of quantiles52 from the observed data,
and to plot the two sets of quantiles against one another. If the estimated distribution provides an exact fit
for the data then the points of the quantile-quantile plot should fall on a straight line with slope 1. The extent
to which these points do not fall on a straight line indicates how poorly the data are fit by the estimated
distribution.
The results of the quantile-quantile plots were promising, but they provide too little information about
the tail of the estimated distribution (high concentrations), because no matter how discrepant the estimated
distribution may be from the data, quantiles near 100% will fall close to the straight line of slope 1. The
percentage change in quantile values may be a more relevant measure of goodness-of-fit for extreme
concentrations.
52 The term quantile is used in this section as opposed to the term percentile used in other sections of this chapter.
Quantile is the term more applicable when exact percentiles are not being used (e.g. the 95th percentile is the 0.95 quantile,
but the 0.951 quantile is not often referred to as the 95.1 percentile).
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 145
Exhibit 4-31
Estimated Distribution of DCPA Acid Metabolites Concentration
in Community Water System Wells*
16-,
14-
1
o
A 12
O
o 10-
o
8-
6-
,0
>.
C
0>
Q
>>
*•<
•—• 4 —
cc
JQ
2 2J
0
0.1
0.2
0.3 0.4 0.5
Concentration
0.6
0.7
0.8
0.9
According to the estimated distribution, DCPA acid metabolites occur in all CWS wells, although the
concentration is extremely small (less than 0.05 /jg/L) in most CWS wells. The MCL for DCPA acid
metabolites is 4,000 /jg/L. DCPA acid metabolites were found only at much lower levels in CWS wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 146
The second approach was adopted to better assess goodness-of-fit for high concentrations. The
observed concentrations were plotted against predicted concentrations, where predicted concentrations were
calculated through a two step process involving both sets of quantiles. Initially quantiles were calculated for
each data point (positive concentration) from the empirical distribution using the survey weights. Then the
concentration corresponding to the same quantile of the estimated distribution was calculated. Plots of the
observed versus predicted values determined in this way should also fall on a straight line if the estimated
distribution models the data perfectly. The advantage of this method is that the range of the plot is not
limited for high concentrations, which makes an assessment of goodness-of-fit for these values more viable.
Exhibit 4-32 shows the quantile-based, predicted versus observed data plot for nitrate concentrations
in rural domestic wells. Whereas the estimated distribution models the data well at low concentrations (where
much of the data is contained), it does not model the data well at higher concentrations. Exhibit 4-32 is
typical of the predicted versus observed data plots generated for concentration distributions in the Phase II
analysis. The estimated distributions model comparatively low concentrations well, but do not model high
concentrations, where there is far less information or data, very well. Consequently, estimates presented in
this chapter concerning high concentrations at high quantiles should be treated with more caution than
estimates at more likely concentration levels. Although estimates at extremely high concentrations are less
accurate, the estimates of the number of wells containing nitrate above the MCL (10 mg/L) are reliable.
4.4.3 Population Exposure and Risk Estimates
In this section, results are presented for estimates of national population exposure, and resultant
health risks, due to nitrate and DCPA acid metabolites for rural domestic wells and for community water
systems. Estimates are provided of the populations corresponding to quantiles of general interest (e.g., 95th
and 99th percentiles), and of the number of individuals exposed above health-based levels. Percentiles and
concentration values corresponding to health levels of concern are presented in Exhibits 4-33 through 4-37.
Confidence intervals are also presented.
The results presented in this section are based on the following population estimates:
• Approximately 30,300,000 people drink ground water from rural domestic wells
nationally. Approximately 450,000 of these people are infants under the age of one
year;
• Approximately 136,000,000 people drink ground water from CWS wells nationally.
Approximately 2,000,000 of these people are children under the age of one year;
• There are approximately 10,500,000 rural domestic wells nationally, 9,900,000 of
which provide ground water for drinking. On average, each rural domestic drinking
water well that provides ground water for drinking serves approximately 3.1 people;
and
• There are approximately 94,600 wells in 35,800 community water systems that
provide ground water for drinking. On average, each system consists of
approximately 2.6 wells.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 147
CO
.o
75
§
c
o
O
Exhibit 4-32
Plot of Predicted Versus Observed Nitrate Concentrations
in Rural Domestic Wells
60
80
100
120
140
Observed Concentrations (mg/L)
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 148
Exhibits 4-33 and 4-34 present estimates at various concentration levels of the number of people
exposed to nitrate from rural domestic wells.
• More than one and a half million people are estimated to drink water from rural
domestic wells that contain at least 10 mg/L nitrate (with an upper bound 95%
confidence interval of 11.8 million persons and a lower bound estimate of 8.02
million). Of the people exposed to nitrate above this level, approximately 22,500 are
expected to be infants. The health risk posed by nitrate at concentrations at or
above 10 mg/L, particularly if the water is also contaminated with bacteria, is a
condition known as methemoglobinemia. Infants are at most risk of developing this
condition. The Maximum Contaminant Level (MCL) for nitrate is 10 mg/L;
• The median nitrate concentration to which people who drink ground water from
rural domestic wells are exposed is approximately 0.32 mg/L; and
• Approximately 19.2 million people drink water from rural domestic wells that contain
nitrate.
Exhibit 4-33
Estimates of Population Exposed to Nitrate
in Rural Domestic Wells by Distribution Percentile
Percentile
Median
95
99
People Exposed
15,100,000
1,510,000
303,000
Concentration
(mg/L)
0.32
10.1
31.5
95% Confidence Interval
Lower Bound
(mg/L)
0.20
8.02
22.3
Upper Bound
(mg/L)
0.47
11.8
40.9
Exhibit 4-34
Estimates of Population Exposed to Nitrate
by Concentration in Rural Domestic Wells
Concentration
(mg/L)
All concentrations > 0
>. 10
Population
Exposed
19,200,000
1 ,530,000
95% Confidence Interval
Lower Bound
17,800,000
1,140,000
Upper Bound
21,800,000
1,820,000
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 149
Exhibits 4-35 and 4-36 present estimates of the number of people exposed to nitrate in CWS wells.
• Approximately 3 million people drink water from CWS wells that contain nitrate at
a concentration of at least 10 mg/L, the maximum contaminant level (MCL) for
nitrate. Of the people exposed to nitrate above this level, approximately 43,500 are
expected to be infants at possible risk of developing methemoglobinemia, particularly
if the water is also contaminated with bacteria.
Exhibit 4-35
Estimates of Population Exposed to Nitrate
in Community Water System Wells by Distribution Percentile
Percentile
Median
95
99
People Exposed
68,000,000
6,800,000
1,360,000
Concentration
(mg/L)
0.63
6.52
14.2
95% Confidence Interval
Lower Bound
(mg/L)
0.45
5.34
10.6
Upper Bound
(mg/L)
0.95
7.60
17.7
Exhibit 4-36
Estimates of Population Exposed to Nitrate
by Concentration in Community Water System Wells
Concentration
(mg/L)
All concentrations > 0
> 10
Population Exposed
85,300,000
2,980,000
95% Confidence Interval
Lower Bound
78,100,000
1,600,000
Upper Bound
98,900,000
4,260,000
• Approximately 85 million people drink water from community water system wells
that contain nitrate; and
• The median nitrate concentration to which people in the United States are exposed
is approximately 0.63 mg/L.
Exhibit 4-37 shows the estimated concentration of DCPA acid metabolites corresponding to the 99th
percentile of the estimated concentration distribution for people who drink well water from rural domestic
wells. The exhibit indicates that one percent of the population (303,000 people) served by rural domestic wells
are exposed to DCPA acid metabolites at concentrations above 0.33 /ig/L. Other percentiles, such as the
median and 95th percentile, are not shown because they correspond to non-occurrence of DCPA acid
metabolites in rural domestic wells. (Consequently, the estimate of the percent of people exposed to DCPA
acid metabolites from rural domestic wells is less than 5%.) Exhibit 4-37 shows the 99th percentile of the
concentration for DCPA acid metabolites. The number of persons exposed at lower concentrations are not
shown (see footnote 43).
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 150
Exhibit 4-37
Estimate of Population Exposed to DCPA Acid Metabolites
in Rural Domestic Wells at 99th Percentile
Percentile
99
People Exposed
303,000
Concentration
(H9/L)
0.33
95% Confidence Interval
Lower Bound
(H9/L)
0.11
Upper Bound
(H9/L)
0.78
NFS estimates that no individuals served by rural domestic wells are exposed to DCPA acid
metabolites at concentrations above those corresponding to the lifetime Health Advisory Level of 4,000 ng/L.
Therefore, no adverse health effects are expected in this population as a result of exposure to DCPA acid
metabolites in drinking water. As Exhibit 4-38 indicates, approximately 1.2 million people are estimated to
drink water from rural domestic wells that contain DCPA acid metabolites (with an upper 95% confidence
bound of 30.3 million and a lower bound of 0.4 million).53
Exhibit 4-38
Estimates of Population Exposed to DCPA Acid Metabolites
by Concentration in Rural Domestic Wells
Concentration
(H9/L)
All concentrations > 0
>. 4,000
Population
Exposed
1,230,000
0
95% Confidence Interval
Lower Bound
460,000
0
Upper Bound
30,300,000
0
Exhibits 4-39 and 4-40 present estimates of the number of people exposed to DCPA acid metabolites
in CWS wells.
• No individuals served by community water systems are exposed to DCPA acid
metabolites at levels above those corresponding to the lifetime Health Advisory
Level of 4,000 /ig/L. Therefore, no adverse health effects are expected in this
population as a result of exposure to DCPA acid metabolites in drinking water; and
• EPA estimates that approximately 9.4 million people drink water from community
water system wells that contain DCPA acid metabolites.
53 The upper confidence limit is affected by the 5 re-samples that resulted in an estimate of the proportion of people
exposed to be close to 100%. Two of the 5 re-samples provided data for which the maximum likelihood estimation procedure
did not converge. These two estimated models may provide inaccurate estimates, although their effect on the upper
confidence intervals is likely to be more pronounced than their effect on the maximum likelihood estimates presented. The
lack of convergence for any of the re-samples brings into question the appropriateness of the modeling procedure. Possible
explanations are that the lognormal/binomial mixture model is not appropriate, or that the number of detections is too few
to be able to properly estimate the model.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 151
Exhibit 4-39
Estimates of Population Exposed to DCPA Acid Metabolites
in Community Water System Wells by Distribution Percentile
Percentile
95
99
People Exposed
6,800,000
1,360,000
Concentration
(H9/L)
0.23
0.99
95% Confidence Interval
Lower Bound
(H9/L)
0.00
0.48
Upper Bound
(ng/L)
0.42
1.80
Exhibit 4-40
Estimates of Population Exposed to DCPA Acid Metabolites
by Concentration in Community Water System Wells
Concentration
(H9/L)
All concentrations > 0
>. 4,000
Population Exposed
9,430,000
0
95% Confidence Interval
Lower Bound
5,880,000
0
Upper Bound
12,500,000
0
Exhibits 4-41 and 4-42 present estimates of numbers of wells containing at least one pesticide above
health-based levels and estimates of populations exposed to at least one pesticide above the NFS MRLs and
to at least one pesticide above health-based levels. These estimates are not based on concentration models,
but on the estimates of populations served and the number of sampled wells containing pesticides. The
estimates and confidence intervals result from the application of the sample design through the sample weights
to appropriate sample statistics. The estimates are the weighted sum of the sample statistics. The formulas
for calculating the estimates and confidence intervals are presented in the NFS Phase I Report/'
54
As Exhibit 4-41 shows, EPA estimates, based on the upper 95% confidence bounds for the estimates
of wells containing pesticides at concentrations exceeding health-based limits, that a maximum of 7.3 percent
of CWS wells and 28.3 percent of rural domestic wells that contain detected pesticides exceed the maximum
contaminant level (MCL) or Lifetime Health Advisory Level (HAL) for those chemicals for which an MCL
or HAL has been established. These estimates are influenced by the following critical factors: (1) less than
50% of NFS analytes have a health-based standard (see the NFS Phase I Report for a more detailed discussion
of this issue); and (2) some of the pesticides and pesticide degradates may not have been detected because
their MRLs were relatively high.
54 NFS Phase I Report. Appendix B, pp. B-52 to B-62.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 152
Exhibit 4-41
Estimates of Percentages of Wells and Number of Wells
Containing At Least One Pesticide Above Health-Based Levels
Percentage and
Number of Wells
with
Concentration
Above Hearth-
Based Standard
% of Wells
Number of Wells
CWS Wells
Mean
Proportion
0%
0%
95% Confidence
Interval
Lower
Bound
0%
0
Upper
bound
7.3%
700
Rural Domestic Wells
Mean
Proportion
13.7%
60,000
95% Confidence
Interval
Lower
Bound
3.4%
15,000
Upper
bound
28.3%
125,000
Exhibit 4-42
Estimates of Population Exposed to Pesticide
Above Reporting and Health-Based Levels
Concentration
Above MRL
Above Health-Based
Standard
Population Exposed
CWS
Wells
14.0 million
0
95% Confidence
Interval
Lower
Bound
7.4 million
0
Upper
Bound
22.4 million
31.5 million
Rural
Domestic
Wells
1.27 million
150,000
95% Confidence
Interval
Lower
Bound
690,000
12,000
Upper
Bound
2.01 million
450,000
4.4.4 Major Findings
The following are the major findings of this analysis of exposure and risk for populations served by
rural domestic wells and community water systems:
• About 1.5 million people served by rural domestic wells are estimated to be exposed
to nitrate above the Maximum Contaminant Level/Lifetime Health Advisory Level
of 10 mg/L. Included in this group are approximately 22,500 infants under one year
old exposed to a concentration of greater than 10 mg/L. Persons with high levels of
nitrate in their private wells should consult their pediatricians and may wish to
obtain water from alternate sources that have less than 10 mg/L of nitrate to help
protect infants from this risk. Physicians are usually well informed about the risks
to infants of high levels of nitrate in drinking water and are able to provide medical
treatment.
National Pesticide Survey: Phase II Report
-------
Chapter Four: Evaluation of Results 153
About 3 million people served by CWS wells are estimated to be exposed to nitrate
above 10 mg/L (ppm). Included in this group are approximately 43,500 infants under
one year old exposed to a concentration of greater than 10 mg/L. Public water
supplies that violate the Maximum Contaminant Level of 10 mg/L for nitrate are
required to notify their customers about the violation, and the adverse health effects
caused by nitrate. (40 CFR 141.32) Local and state health authorities are the best
source for information concerning alternate sources of drinking water for infants.
Systems that apply for variances or exemptions while in violation of the standard may
be required by the state to provide bottled water or point-of-use or point-of-entry
devices to avoid unreasonable risks to health. (40 CFR 141.62(f))
About 1.2 million people are estimated to be exposed to DCPA acid metabolites
through consumption of drinking water obtained from rural domestic wells. About
9.4 million people are estimated to be exposed to DCPA acid metabolites through
consumption of drinking water obtained from CWS wells. However, the
concentrations to which these individuals are estimated to be exposed are much
lower than health-based levels. Therefore, no adverse health impacts are expected as
a result of this exposure, based on currently available information on the health
effect from exposure to DCPA and its acid metabolites.
An estimated 1.27 million people served by rural domestic wells are exposed to at
least one pesticide above NPS minimum reporting limits (MRL). An estimated
150,000 people served by rural domestic wells are exposed to at least one pesticide
above health-based levels.
An estimated 14 million people served by CWS wells are exposed to at least one
pesticide above a minimum reporting limit (MRL). However, the concentrations to
which these individuals are estimated to be exposed are lower than health-based
levels.
The use of MRLs in the NPS had a significant effect on the estimates of frequency
of occurrence for some analytes. Environmental contaminants tend to occur with
higher frequencies at lower concentrations. Estimates of frequency of occurrence in
rural domestic wells increased from those presented in the NPS Phase I Report,
which were based only on concentrations above the MRL, for DCPA acid
metabolites (from 2.4% to 3.0%) and nitrate (from 57% to 64%) as a result of an
analysis that involved estimation of the frequency of occurrence at all concentrations,
including those below the MRL. A similar increase was noted for nitrate in CWS
wells (52% to 56%).
Although the Survey did not identify any CWS wells with pesticides at concentrations
exceeding health based limits, estimates were developed, using the same procedures
as were followed to generate similar estimates for the NPS Phase I Report. Using
the upper bound of the 95% confidence interval, these estimates indicate that a
maximum of 7.3 percent of CWS wells (corresponding to about 700 wells) containing
pesticides exceed the maximum contaminant level (MCL) or Lifetime Health
Advisory Level (HAL) for those chemicals for which an MCL or HAL has been
established. (The mean proportion was estimated to be zero.) Using the upper
bound of the 95% confidence interval, a maximum of 28.3 percent of rural domestic
wells (approximately 125,000 wells) containing pesticides exceed the maximum
contaminant level (MCL) or Lifetime Health Advisory Level (HAL) for those
chemicals for when an MCL or HAL has been established. (The lower bound of the
95% confidence interval was estimated to be 3.4%, corresponding to approximately
15,000 wells, and the mean proportion was estimated to be 13.7%, corresponding to
approximately 60,900 wells.)
National Pesticide Survey: Phase II Report
-------
-------
Chapter Five: Conclusions and Recommendations
EPA's understanding of pesticides and nitrate in drinking water wells has changed as a result of the
National Pesticide Survey. The Survey provided some support for many existing theories concerning well-water
contamination, and a broader national picture of the problem emerges from the analyses conducted in Phase
II.
Many of the statistical tests and analyses conducted for Phase II did not provide significant
correlations among many of the factors that were expected to be correlated. In addition, there was some
inconsistency in the significant correlations found for community water system wells and rural domestic wells.
While there are plausible explanations for the inconsistencies, the results must be interpreted with caution as
either a confirmation or a refutation of any model of the physical mechanisms or causation without further
investigation, hypothesis testing, or model development. Chapters 2 and 3 provide extensive, detailed
discussions of the limits and constraints in the NPS database. These conclusions reflect the careful
deliberations and evaluations of the limitations of the Survey data. It is, however, useful to summarize why
the Survey was unable to provide definitive relationships between pesticide use and occurrence.
As noted throughout the Phase II report, EPA's ability to investigate thoroughly the relationship of
specific pesticide use and occurrence in drinking water wells was severely limited by the small number of
detections (i.e., sample size) for most of the pesticides tested in the Survey. The limited sample size prevents
an extensive analysis of the association of pesticide occurrence with expected contributing factors and limits
the "power" of the Survey to detect the associations or correlations that may exist. In addition, using survey
methods and secondary data to measure possible factors related to pesticide occurrence is a less precise
method of assessing the potential influence of many of the factors that determine pesticide contamination of
drinking water wells. The EPA recognized, early in the design of the Survey, the inherent problem of using
respondent recall or expert opinions to develop measures of ground water characteristics and agronomic
activity. This was the only cost-effective method of measurement and data collection for a national survey.
Such data do, however, include measurement error that cannot be corrected and this further constrains the
Survey's ability to uncover correlations that may be present.
Before establishing new policies to protect drinking water wells, readers should review these results
in the context of data from other sources. These comparisons may offer important perspectives that the NPS
does not fully explore. For example, the analyses were designed to identify factors that could be potentially
useful topics for additional detailed research. The conclusions are based on a one-time "snapshot" of the
nation's community water system and rural domestic drinking water wells. The analyses cannot definitively
address factors affecting the presence of chemicals in drinking water wells that may change over time.
Furthermore, the Survey's statistical design was chosen to provide national estimates of wells containing
pesticides or nitrate. The results cannot address regional trends. The results of the statistical analyses carried
out in Phase II may provide evidence of association between two variables, but such associations do not
necessarily imply causation. The statistical analyses were guided by hypotheses that expressed plausible causal
mechanisms. The Phase II results should not be interpreted to imply unambiguous attribution of causation.
As an observational survey, the NPS cannot control all of the major factors that could influence the occurrence
of pesticides and nitrate in drinking water wells. Finally, the choice of wells to be sampled and the number
of water samples obtained by the Survey were determined primarily to estimate the number of wells
contaminated by pesticides or nitrate (reported in Phase I). This choice was not ideally suited to carry out
all the statistical analyses performed in Phase II. Thus, the conclusions reported in this chapter apply to the
findings of EPA's National Survey of Pesticides in Drinking Water Wells. They should prove useful, in
combination with the results of other studies, to inform the development of improved policies for drinking
water well protection.
5.1 Conclusions
The NPS results show associations between many different factors and the contamination of drinking
water wells. The conclusions in this chapter are based on the evaluation of data quality, strength and
consistency of statistical associations, and other factors discussed in Chapter 4. Although the Survey did not
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 156
identify associations between reported uses of pesticides within 300 feet, 500 feet, or one-half mile of the
sampled wells and detections in those wells, the Survey did identify associations between detections in both
public and private drinking water wells and measures of agronomic activity (such as crop value and livestock
production) that may be indirect indicators of agricultural use of pesticides and fertilizers. Similar analyses
also identified relationships between non-agricultural use of DCPA and detections of its acid metabolites.
Nitrate's frequent presence in drinking water wells, already noted by a number of State and local
surveys, is shown by the NFS to be a national situation. Furthermore, Phase II results show that nitrate's
presence is associated with sales of fertilizers and with other sources of nitrate, particularly livestock
production within the county.
Several other factors whose relationship to occurrences of pesticides in wells was already suspected
also were identified in the Phase II analysis. Pesticide persistence, one of the criteria for determining whether
particular pesticides were included as analytes in the NPS, was confirmed to be related to detections. Evidence
suggests that more persistent pesticides occur in wells more frequently than less persistent pesticides.
Shallower wells and, to a lesser extent, older wells were both shown to be associated with increased likelihood
of detections of both pesticides and nitrate.
Other expected associations did not materialize in the NPS results. County-level measures of ground-
water vulnerability did not correlate with detections in individual wells and pesticide mobility was not found
to be associated with a greater incidence of detections.
When the significant single variables are grouped by the five broad categories of factors considered
to be related to the presence of pesticides and nitrate in well water - ground-water sensitivity, pesticide and
nitrate use, transport mechanisms, chemical characteristics of pesticides, and the physical condition of wells -
- a number of conclusions can be drawn. Each is highlighted in the following section along with a discussion
of its significance.
(1) The Survey did not Identify a useful measure of ground-water sensitivity from Its
county-level and sub-county level assessments for predicting the location of
individual wells containing pesticides or nitrate.
The Agricultural DRASTIC system, used to generate measures of ground-water vulnerability (the term
by which sensitivity was known at the beginning of the Survey) for purposes of stratification, did not function
effectively when measured at the county level to locate drinking water wells containing pesticides or nitrate.
DRASTIC is designed to assess the ground-water pollution potential of hydrogeological zones. A higher
overall score, and consequently higher subscores, is expected to identify areas of higher potential ground-water
vulnerability. Higher DRASTIC scores were not consistently associated with more frequent detections (the
expected direction of association) or with less frequent detections. For example, at the county level, the
average DRASTIC score and the subscores for net recharge and aquifer media were found to be related to
detections, but in a manner contrary to that expected. These county-level DRASTIC results indicated
associations between detections and lower overall scores or subscores. When a particular DRASTIC factor
was associated with detections in CWS wells, the same factor was only infrequently associated with detections
in rural domestic wells. For example, county-level results for topography and depth to water table indicated
that a larger number of detections could be expected in areas of relatively low slope and shallow water table,
but these results were not corroborated by appearing for both CWS and rural domestic wells.
Sub-county DRASTIC results showed similar inconclusive tendencies. The score for impact of the
vadose zone indicated that more pesticide detections were associated with vadose zones in which relatively less
attenuation occurred before chemicals reached ground water. Scores for aquifer media and the total
DRASTIC score at the sub-county level were associated with nitrate concentrations.
Overall DRASTIC scores and subscores did not provide clear and unambiguous indicators for locating
contaminated wells. A number of factors may have affected the ability of DRASTIC to identify vulnerable
drinking water wells. The degree of misclassification, although considered acceptable for stratification, was
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 157
recognized as an obstacle to analysis. The scale of application of DRASTIC may have obscured intra-county
variability. Some DRASTIC factors partially overlapped and all factors affecting vulnerability are not included.
(2) The Survey's measures of pesticide use based on questionnaire data were not good
indicators for locating drinking water wells containing pesticides.
The NFS was designed in part to test the proposition that pesticide use is related to pesticide
detections in well water. Three direct sources of data were used. First, a proprietary source of pesticide
marketing data was used to develop one of the two stratification variables. Second, the NFS devoted a great
deal of attention in its data collection from well owners or operators to inquiries about the agricultural and
non-agricultural use of pesticides on or near the property surrounding the sampled well. The areas of interest
were defined as the property on which the well was situated or areas within 300 feet, 500 feet, or one-half mile
of the well, depending on the particular question and respondent. Inquiries were made about farm and non-
farm use, storage, and disposal of pesticides. Inquiries were also made about spills of pesticides and other
mishaps or misuses. Although individual questionnaires were reviewed for internal consistency, questionnaire
data were not validated from other sources. Finally, the Survey collected data on pesticide use from county
agricultural extension agents, who were expected to be particularly knowledgeable about use in their areas.
The pesticide use estimates developed from marketing data that were used for stratification showed
no correlation with a higher likelihood of locating wells containing pesticides. Detections in CWS wells
occurred at almost the same frequency, irrespective of whether the well was located in the high, moderate, low,
or uncommon pesticide use strata. The highest proportion of detections in rural domestic wells occurred in
the low and moderate use strata.
Neither questionnaire data on the agricultural use of pesticides from well owners or operators nor
questionnaire data from county agricultural extension agents was found to be associated with pesticide
detections. Pesticides were detected where they were not reported as used, and reported as used where they
were not detected. Four explanations, singly or in combination are possible: (1) that the three or five year
period about which inquiries were made was not long enough to include the time when the pesticides were
used or could migrate to nearby drinking water wells; (2) that respondents mistakenly or willfully
underreported pesticide use and other activities or events, such as spills, that might have led to the presence
of pesticides; (3) that the geographic area about which information was sought did not include the recharge
areas of a significant number of wells; and (4) that pesticides were not detected in areas where they were
reported as used because of detection limits, diffusion, dilution, sorption, degradation or other causes.
A "cropped and vulnerable" stratum, consisting of a combined score for ground-water vulnerability and
cropping intensity, did not distinguish rural domestic wells in which pesticide detections were more frequent
from other rural domestic wells. Although a slightly higher frequency of occurrence for pesticides was
associated with rural domestic wells in the "cropped and vulnerable" stratum, the result was not statistically
significant.
(3) Measures of pesticide use and of agronomic activity, such as data on crop values and
livestock values collected from other sources and analyzed by the Survey, did show
associations with pesticide detections In drinking water wells.
The NFS obtained data from several external sources for use in the analysis. The sources and
principal results include the following:
(a) A database compiled for EPA by Resources for the Future. Inc.. consisting of estimates of
pesticide use by crop for selected herbicides and insecticides primarily used on farm crops; use estimates for
10 selected pesticides used by urban applicators: and use estimates for 10 selected pesticides used on golf
courses. County-level estimates were calculated from regional totals for pesticides applied to specific crops,
using estimates of the state's or region's total crop acreage. Analysis of the relationship between pesticide use,
measured through the RFF database, and pesticide detections could be carried out for only one pesticide. The
analysis showed an association between the amount of urban and golf course use of DCPA and greater
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 158
likelihood of detection of DCPA acid metabolites in both CWS and rural domestic wells. No conclusive
evidence was found that occurrences of DCPA acid metabolites in well water are associated with the rate of
use of DCPA in agriculture as measured by the RFF data.
(b) A database consisting of county-level measures of agronomic activity compiled by the National
Fertilizer and Environmental Research Center (NFERQ at the Tennessee Valley Authority from the 1987
Census of Agriculture, including general crop acreage, value of crop sales and value of livestock sales, specific
crop acreage and production, and animal counts. Analysis of the relationship between these measures of
agronomic activity and pesticide detections indicated a positive correlation between one of the five measures
tested, market value of crops, and pesticide occurrence in rural domestic wells. No correlation was noted for
market value of crops and CWS wells. Two other measures of agronomic activity that vary inversely with
market value of crops, acres of pasture and rangeland fertilized and numbers of beefcows, showed a negative
correlation with pesticide detections. The data show that acres in cropping decrease as pasture and rangeland
and the number of cattle per county increase. If the market value of crops and acres of fertilized rangeland
complement each other as indicators of pesticide use, then overall pesticide use decreases as cropping
decreases. The results support the prior hypotheses.
(4) Measures of agronomic activity, such as data on fertilizer sales, crop values, fertilized
acres, and market value of livestock, were found to be associated with concentrations
of nitrate in drinking water wells, and also found (although less frequently) to be
associated with nitrate detections.
(a) A database of state-level nitrogen fertilizer sales data compiled by NFERC. State-level
estimates of fertilizer sales, prepared at West Virginia University using the NFERC data, were allocated to
counties using factors developed from the 1987 Census of Agriculture. Significant correlations were found
between sales of nitrogen fertilizer and nitrate concentrations in both CWS wells and rural domestic wells.
Analysis of the relationship between nitrogen fertilizer sales, measured through the NFERC database, and
detections of nitrate did not indicate an association between these variables. The reasons for this apparent
inconsistency cannot be determined using NPS data.
(b) Using the same NFERC database of county-level agronomic activity described in (3)(b),
measures of agronomic activity were tested for associations with nitrate detections and with nitrate
concentrations. Crop value, acreage of cropland fertilized, and market value of livestock were all shown to
be associated with nitrate concentrations above 0.15 mg/L in both CWS wells and rural domestic wells. A
correlation between market value of livestock and nitrate detections was also observed.
The results described in conclusions (3) and (4) for the external databases are subject to several
limitations. Some of the data reflect regional, state, or county-level measures of use of pesticides and sales
of nitrogen fertilizers. Assumptions used in the process of extrapolating from regional or state level data to
county-level data and then to well-level data could affect the results associated with well-specific detections
and concentrations. Measures of sales indirectly reflect use, and the sales data may not accurately identify
counties in which use actually occurs. In some cases, chemicals purchased in one location may not be used,
or may be used in another location. Finally, comparisons of different databases indicate some discrepancies
or inconsistencies in their estimates of pesticide use.
Despite these limitations, the results contain some suggestions that agricultural activity and livestock
operations are associated with greater numbers of pesticide and nitrate detections. Market value of crops was
associated with detections of pesticides in rural domestic wells. Sales of nitrogen fertilizers, crop value,
acreage of cropland fertilized, and market value of livestock were associated with nitrate concentrations in both
CWS wells and rural domestic wells.
Multivariate analyses, which tended to control for some of the confounding factors listed above,
further support the findings concerning nitrate and pesticide use. The results of the multivariate analyses show
that nitrate concentrations are affected by fertilizer use and agronomic activity. Two surrogate measures of
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 159
agronomic activity that strongly predicted pesticide and nitrate detections were the market value of crops and
acres of fertilized rangeland.
(5) Analysis of variables relating to transport of chemicals to ground water, including
several precipitation measures, Indicated that Increased precipitation was inversely
related to detections and concentrations of nitrate.
Data obtained from the National Oceanic and Atmospheric Administration concerning precipitation
measured by weather stations in the counties in which detections occurred were analyzed for associations
between rainfall variables and detections. Higher amounts of rainfall were found to be inversely correlated
with nitrate detections. Several measures of precipitation for periods ranging from the month prior to
sampling to five years prior to sampling all indicated that fewer nitrate detections and lower nitrate
concentrations in CWS wells were associated with increased rainfall. Some evidence was obtained that intense
precipitation over an extended period was associated with a reduced likelihood of pesticide detections in CWS
wells. The probability of detecting pesticides or nitrate in rural domestic wells was generally not shown to be
related to precipitation, although there is some evidence of an inverse relationship between nitrate detections
and intense precipitation during the previous five years. The rainfall variable was measured in those counties
nationwide where sampling occurred and provides a good estimate of average rainfall for the period being
evaluated. The result suggests that after rainfall, pesticides and nitrate may run off before entering ground
water or their concentrations may be reduced to the point where detections are less likely.
Some DRASTIC subscores related to water transport also were associated with detections, consistent
both with the precipitation findings and with the design of DRASTIC. These results were not recurring in
both the CWS and rural domestic well surveys. The subscores that showed an association were depth to water
table and hydraulic conductivity. Parallel results were obtained for well-level variables from NFS questionnaire
data. Shallow well depth was found to be associated with detections, while detections were less likely in wells
that tapped confined aquifers.
The Survey investigated the presence and severity of drought, as measured by the Palmer Drought
Index, in counties where sampling occurred. Analysis of these data showed that there was no discernable effect
of drought on the Survey results. Sampling occurred in areas with conditions ranging from "extreme drought"
to "extremely moist," as measured by Palmer Drought Index. No statistically significant different frequencies
of occurrence of detections were shown among the categories. Some evidence, paralleling the results for
precipitation, was obtained that nitrate detections in CWS wells were associated with moist conditions.
Finally, analysis of Survey questionnaire data on the presence of surface water near wells indicated
that nitrate was less likely to be detected in either CWS wells or rural domestic wells located with such water
bodies nearby. Presence of a body of water within 300 feet of the well also was negatively associated with
pesticide detections in CWS wells. Rural domestic wells located within 500 feet of other operating wells and
CWS wells located within one-half mile of flood irrigation were both more frequently associated with
detections. The use of ground water for irrigation within one-half mile was associated with a lesser likelihood
of detections. These results do not provide a consistent pattern of associations between detections and the
nearby presence of surface water.
(6) The Survey Identified well depth and to a lesser extent well age as factors related to
detections of contamination.
Based on the results from Survey questionnaires, an association was found between detections and
shallower wells, although NPS interviewers reported that well owners frequently lacked precise knowledge of
the depth of their wells. A relationship between detections and wells drawing water from unconfined aquifers
(consistent with the well depth results) was identified for nitrate detections in CWS wells. Weaker evidence
also was obtained for association between detections and older wells.
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 160
(7) Phase II results suggest that several factors contributed to a smaller number of
pesticide and pesticide degradate products being detected than would otherwise
have occurred. The number of detections reduced the Survey's power and ability to
identify possible relationships between well contamination and other factors.
The effective number of pesticide detections considered in the analysis was 44 in CWS wells and 17
in rural domestic wells. The effective number of nitrate detections considered in the analysis was 220 in CWS
wells and 232 in rural domestic wells. The substantially greater number of times that nitrate was detected by
the Survey supplied additional data points for nitrate occurrence that increased the ability to detect
associations that may have been present. Accordingly, the Phase II analyses identified a greater number of
associations involving nitrate detections or concentrations than involving any combination of pesticide
occurrence data.
The number of pesticide detections available for inclusion in the Phase II analysis was affected by
several factors. First, the Survey's statistical design was chosen to provide national estimates of wells
containing pesticides or nitrate. The choice of wells to be sampled and the number of water samples obtained
were not designed specifically to supply sufficient data to ensure that all the possible Phase II statistical
analyses could be performed successfully. A substantially larger number of samples would have made it likely
that additional analyses of pesticide occurrences would have been possible. Second, the stratified design did
not, in fact, identify areas in which a higher proportion of wells containing pesticides were located. Third, the
specification of minimum reporting limits designed to strictly control false positives meant that concentrations
below the MRL were reported as nondetections. Concentration distributions were determined for DCPA acid
metabolites and nitrate through extrapolation below the MRL and above the highest detected occurrence. The
estimated occurrence frequency rose above that estimated using the MRLs. DCPA increased from 2.4% to
3.0% and nitrate from 57% to 64% in rural domestic wells. Additional analyses involving other analytes
would have been possible at higher frequencies of occurrence if more samples had been gathered and/or if
MRLs had been lower.
(8) Multivariate analysis produced models that suggest factors that may prove to be
useful screening tools to Identify drinking water wells likely to contain pesticides or
nitrate.
Multivariate analysis was used in an attempt to identify a set of variables that best predicts the
occurrence of pesticides and nitrate in drinking water wells. The Phase II analysis carried out multivariate
regression analyses to determine if combinations of variables would be particularly strong predictors of
pesticide and nitrate contamination of drinking water wells. Because of the relatively small number of
pesticide detections and substantial commonality among variables, however, the predictive value of many of
the factors considered in the analysis was difficult to evaluate. Very few variables appeared in more than one
model. The best models were identified on the basis of how well they fit the data. The appearance of
different variables in equally successful models does not necessarily reflect physical or theoretical
interchangeability of the variables.
For pesticide detections, two variables - fertilized pasture and rangeland and well depth - provided
the best model for detections of a pesticide in CWS wells. The presence of other operating wells near the
sampled wells can be used in place of the well depth variable without substantially reducing model
performance. An indirect measure of agricultural activity - market value of crops in thousands of dollars --
by itself was the best model for pesticide detections in rural domestic wells. A variable measuring the number
of beef cattle per acre could also be included. The small number of variables included the regression models
largely reflects the effective sample size and number of detections for rural domestic wells, and should not be
interpreted as showing that the factors under investigation are not related to pesticide occurrence in rural
drinking water wells.
This Phase II multivariate analysis results parallel in part the logistic regression model identified by
the NAWWS survey, which also contained a use variable. The NAWWS model, however, also reported a
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 161
vulnerability variable. The variable identified by NAWWS, aquifer water level, bears some resemblance to
factors identified in the NFS analysis, since it reflects the effects of precipitation and recharge.
For nitrate detections, a three-variable model, composed of fertilized pasture and rangeland, average
monthly precipitation, and well-water pH, provided the strongest results for nitrate detections in CWS wells.
Farming on the property where the well is located also can be included in the model. A four-variable model,
composed of well age, maximum monthly precipitation in the five years prior to sampling, well water pH, and
the presence of an unlined drainage ditch within less than one-half mile, created the best model for nitrate
detections in rural domestic wells. A variable for fertilized pasture and rangeland could also be included.
For nitrate concentrations in CWS wells, five variables - maximum monthly precipitation in the past
five years, well water electrical conductivity, total nitrogen sales by county for counties containing wells in
which nitrate was detected, well depth, and a categorical variable that reflects the Palmer Drought Index score
for the year prior to sampling - create the best model. Crop value can be used in place of total nitrogen sales
without substantially reducing model performance. For nitrate concentrations in rural domestic wells, four
variables are included in the best model - well depth, market value of crops, presence of an unlined body of
water within one-half mile, and the Agricultural DRASTIC subscore for topography measured at the sub-
county level. Total nitrogen sales can be used in place of crop value. A variable for the presence of a body
of water within 300 feet of the well could also be included.
(9) Estimates of the concentration distributions for DCPA acid metabolites and nitrate
were prepared. The estimates indicate that over 10 million persons are exposed to
DCPA acid metabolites in drinking water wells. Very few are expected to be exposed
to levels above the health advisory or MCL. Approximately 1.5 million persons could
be exposed to nitrate, above the maximum contaminant level of 10 mg/L.
Estimates of the concentration distributions for nitrate and DCPA acid metabolites were prepared for
both CWS and rural domestic wells. These estimates are calculated from detections above the MRL and on
maximum likelihood estimates of concentrations below the MRL. They indicate that the frequency of
occurrence of those pesticides is somewhat greater than indicated by the estimates reported in the NPS Phase
I Report, which were calculated from concentrations that exceeded minimum reporting levels. Approximately
64% of rural domestic wells (6,720,000 wells) and 56% of CWS wells (53,000 wells) are estimated to contain
nitrate. Approximately 3.0% of rural domestic wells (315,000 wells) are estimated to contain DCPA acid
metabolites, and about 100% of CWS wells are estimated to contain extremely low levels of DCPA acid
metabolites.
EPA estimates that about 10.4 percent of CWS wells and 4.2 percent of rural domestic wells contain
detectable levels of one or more pesticides. The Phase II study estimated the chance that a well that contains
one or more pesticides also exceeds a maximum contaminant level or health advisory. EPA estimates that no
more than 7.3 percent of the 10.4 percent of CWS wells that contain one or more pesticides could exceed an
MCL or health advisory. Similarly, no more than 28.3 percent of the 4.2 percent of rural domestic wells that
contain detectable levels of one or more pesticides are also expected to exceed a health based limit. In
summary, about 1 percent of all drinking water wells in the U.S. are estimated to exceed a health based limit.
EPA concluded that the overall chance of a given well exceeding a level of concern for a pesticide is low. If
a well contains a detectable amount of one or more pesticides, it has a slightly higher risk of also exceeding
a health based limit. EPA recommends that well owners that know or suspect that their well is affected by
pesticides have the water tested to ensure that any pesticides are present at levels below the MCLs or health
advisories.
Quantitative exposure and risk estimates prepared in Phase II for the populations served by wells
containing DCPA acid metabolites indicate that, as indicated in Phase I, the current potential health effects
are low. Approximately 1.2 million people are estimated to drink water from rural domestic wells and 9.4
million drink water from CWS wells that contain DCPA acid metabolites, but none are exposed at
concentrations above the Lifetime Health Advisory Level of 4,000 /jg/L.
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 162
Approximately 19 million people are estimated to be exposed to nitrate in rural drinking water wells,
with about 1.5 million exposed to levels of nitrate over the Maximum Contaminant Level of 10 mg/L (ppm),
and approximately 22,500 infants under 1 year old exposed to a concentration of greater than 10 mg/L.
Persons with high levels of nitrate in their private wells should consult their pediatricians and may wish to
obtain water from alternate sources that have less than 10 mg/L of nitrate to help protect infants from the risk
of methemoglobinemia (blue-baby syndrome). Physicians are usually well informed about the risks to infants
of high levels of nitrate in drinking water and are able to provide medical treatment. Approximately 85
million people are estimated to drink water from CWS wells that contain nitrate, with about 3 million exposed
to levels of nitrate over 10 mg/L (ppm), and approximately 43,500 infants under 1 year old exposed to a
concentration of greater than 10 mg/L. Public water supplies that violate the Maximum Contaminant Level
of 10 mg/L for nitrate are required to notify their customers about the violation, and the adverse health effects
caused by nitrate. (40 CFR 141.32) Local and state health authorities are the best source for information
concerning alternate sources of drinking water for infants. Systems that apply for variances or exemptions
while in violation of the standard may be required by the state to provide bottled water or point-of-use or
point-of-entry devices to avoid unreasonable risks to health. (40 CFR 141.62(f))
5.2 Recommendations
This section presents several recommendations for planning and design of future surveys and other
studies. They offer ideas on how to expand the results of this Survey and of future similar studies. The
recommendations offer ideas on ways to study the presence of pesticides and nitrate in drinking water, new
questions, and subjects that need to be further investigated.
Three aspects of survey design had particularly strong effects on the quality of results achieved. They
are:
• Stratification of the country by anticipated probability of contamination;
• Methods of data collection; and
• The number of wells to be sampled.
The Survey's results suggest ways in which each of them could be improved.
Stratification. Oversampling of strata should be undertaken only if the criteria used can be measured
with sufficient accuracy to improve the survey estimates and precision. This requires an alternative to the
county-level DRASTIC scores as implemented in the NFS. The NPS spent considerable energy and resources
on attempting to identify those areas of the country most likely to be contaminated by pesticides. Survey
designers expected that by stratifying and oversampling such areas the Survey would be more likely to obtain
a representative picture of the presence of pesticides and nitrate in drinking wells across the country. All
eligible wells for both the rural domestic well and community water system surveys were stratified according
to the county in which they were located. A twelve category stratification was used, based on surrogates for
pesticide usage and ground-water vulnerability. The former was based upon pesticide sales and agricultural
information that was only available at the county level. The latter was based on county-level DRASTIC
measures. For the rural domestic well survey, wells were further stratified at the second stage (sub-county),
based on whether they were located in cropped and vulnerable parts of the county.
In order for improved precision to result from oversampling of strata, the stratification variables
should have an effect on what is being measured and they must be measured with reasonable accuracy.
Furthermore, oversampling, such as the oversampling in cropped and vulnerable areas, is advantageous when
a very high percentage of contaminated wells are known to be in the oversampled strata. Because stratification
did not effectively identify high risk areas, oversampling did not increase the precision of the Survey's
estimates. The Phase II results indicate that neither the overall DRASTIC score nor the individual subscores
at the county level functioned effectively, with both the design of the DRASTIC model and the
implementation of the scoring responsible in part for the ineffectiveness of county-level stratification. With
respect to the sub-county stratification based on a definition of "cropped and vulnerable," little evidence was
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 163
obtained to indicate that the sub-county stratification performed better than the county-level stratification.
The ineffectiveness of second-stage stratification in turn limited the effectiveness of oversampling.
Stratify by pesticide use only if a good local measure of such use is available. The second weakness
in the stratification was that the pesticide use data used for stratification only dealt with agricultural pesticides.
The pesticide most frequently identified in wells by the NFS, DCPA acid metabolites, is frequently used in
non-agricultural settings. Even for agricultural pesticides, the county level data and reports by well owners
do not appear to have identified wells more likely to contain pesticides.
Ensure that the data used for stratification include accurate data about wells. A third potential
weakness in the stratification used for the NFS that may have diminished its effectiveness was that the frame
of community water systems, the Federal Reporting Data Systems (FRDS), did not identify the location of
individual wells. It also did not indicate the number of wells used by a CWS. If FRDS is to be used as a
frame for future surveys it should first be evaluated for its coverage and accuracy of reported data on well
location and system size. When determining the location of the well with respect to the oversampled counties,
it was assumed that all wells in a system were located in the same county as the system's mailing address.
While this assumption is likely to be true in most cases, it is known to be false in at least a few cases.
In summary, these judgments concerning the effectiveness of stratification variables imply that future
surveys should not heavily oversample, since it is not known in advance which areas are most likely to be
contaminated. For the same reason all areas of the country should continue to be given a chance of inclusion
in pesticide studies. Even if it is accurately known that all the pesticides being studied are not sold or used
in a particular part of the country, such areas must be included in order to obtain national estimates. Thus
a typical two-stage design might stratify the country geographically at the first stage to ensure
representativeness but not do any oversampling. Counties might still be the unit selected at the first stage to
reduce travel costs. At the second stage it might be desirable to oversample areas thought most likely to be
contaminated. Such oversampling should be light unless there is great confidence in the predictive power of
the stratification variables.
Methods of Collecting Data. The NFS provides good information on the quality of data that can be
collected from both laboratory analyses and questionnaires and useful guidance on how such methods might
be improved.
The new cost-effective multi-residue methods that were developed for use in the NFS should continue
to be of great use for other environmental studies. These methods enable analysts to search for large numbers
of analytes simultaneously while limiting the costs of the analysis. One drawback to the use of multi-residue
methods, however, is that they tend to be less precise for individual analytes and therefore to have higher
detection limits than methods concentrating on a single analyte. Whatever methods used, all concentration
data obtained should be reported.
In addition to reporting data from chemical analyses for which both qualitative and quantitative
accuracy is assured, also report additional results (with less than the specified levels of precision) of chemical
analysis while maintaining a high degree of quality assurance through confirmation of the presence and
identity of the analytes. The identification of all pesticide and pesticide degradation products detected in the
NFS was both qualitatively and quantitatively confirmed. Confirmations were conducted by reanalyses of all
sample extracts using a second capillary gas chromatographic (GC) or high performance liquid
chromatographic (HPLC) column. These analyses provided both a preliminary qualitative confirmation for
all GC determinations and the final confirmation for HPLC analyses. The analytes detected using a GC-based
method were also qualitatively confirmed using gas chromatography/mass spectrometry (GC/MS).
In the future, in addition to reporting data that have satisfied both qualitative confirmation and
quantitative measurement (equivalent to data exceeding specified MRLs), a second tier of data should be
reported that have been qualitatively confirmed but which have not been quantitatively measured at the
specified levels of precision. Such results should be reported with sufficient information to judge the reliability
of the reported concentrations, assuming that all gas chromatographic identifications have been subject to at
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 164
least one confirmation step. This two-tiered system will maintain data quality requirements for reporting data
associated with specific sampling sites (such as the notice given to well owners and operators of confirmed
detections above the MRL) while increasing the number of data points available for statistical analysis using
data in which only the identity of the analyte has been assured. In particular, statistical analyses of categorical
variables, which is less sensitive to precise measurement of detections, would be enhanced. Broader reporting
of data, along with evaluations of its precision and accuracy, would allow a wider variety of analyses of the type
included in this Phase II report without endangering the accuracy of the national estimates that were included
in the NFS Phase I report. Analysts would have to continue to be very careful of the sensitivity of their
statistical findings to the quality of the laboratory data. This approach would be consistent with the suggested
new American Chemical Society guidelines, which are currently under development, and advocate reporting
of all analytic data, including data below the level at which quantitative results may be obtained with a
specified degree of accuracy, if it is completely documented with respect to problems and limitations.1
Questionnaire data may be limited in quality for topics that are sensitive to respondents such as
pesticide spills and disposal. The NPS results show that it is possible to collect useful information related to
pesticide contamination of wells using questionnaires administered to homeowners, renters, well operators,
and agricultural extension agents. Data collection by telephone also worked effectively. But respondents do
not appear to have provided good data on several of the key factors associated with the presence of pesticides
or nitrates in drinking water wells, including pesticide use, spills, and disposal.
There are at least four possible explanations of why NPS respondents did not report the use of
pesticides near wells where those pesticides were detected: the pesticides had migrated through the ground
water and had not been used near the well; the respondents could not (or did not wish to) recall using the
pesticide; the respondent only remembered a generic type of pesticide (e.g., crab grass killer) for which it was
impossible to determine the active ingredient; or the number of wells with pesticide detections was too small
to find such a relationship. There is little that can be done to modify the first three causes of this limit of the
utility of the questionnaire data. The last cause can be reduced by increasing the size of future surveys or by
dropping or relaxing the use of MRLs.
It is important that future surveys ask about non-agricultural fertilizer use in addition to agricultural
use. When the NPS was designed its emphasis was on detecting agricultural pesticides. The NPS asked about
both agricultural and non-agricultural pesticide use but asked only about agricultural use of fertilizers. Some
of the nitrate contaminating ground water may occur as a result of lawn applications.
Number of Drinking Water Wells to be Sampled. The NPS had two main goals: to estimate the
number of drinking water wells in the United States contaminated by pesticides and to examine relationships
between contamination and a variety of explanatory variables. The determination of the number of wells to
be sampled was only based on achieving the first of these two goals. In fact, the national estimates reported
in the Phase I report had a greater level of accuracy than was anticipated when the survey was designed
because more wells contaminated by pesticides were found in the sample than had been anticipated in the
design. Extensive statistical analysis of survey data in Phase II was limited by the number of detections that
could be included in such analyses. In particular, a modest increase in the number of pesticide detections
available for analysis might have substantially increased the number and explanatory power of the significant
findings and conclusions.
When determining the sample size for the National Pesticide Survey EPA decided to design the study
to meet certain accuracy requirements for the Phase I analysis, and to conduct whatever limited Phase II
analyses would be possible. This is what has been done. If future studies wish to both produce national
estimates and conduct in-depth relational analyses they may require significantly larger sample sizes than used
for this survey.
1 Keith, L.H., Report Results Right!. Part 1. June 1991, CHEMTECH, pp. 352-356, and Report Results Right!. Part
2, August 1991, CHEMTECH, pp. 486-489.
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 165
Type of Study. The NFS obtained the first national estimates of the presence of pesticides and nitrate
in drinking water wells. For that purpose, a national survey was indispensable. No other method will provide
good national estimates for a variety of analytes. The amount of data sufficient to develop accurate national
estimates, however, was not sufficient to perform statistical analyses in Phase II for all of the chemicals and
all of the issues identified. Furthermore, the Phase II analyses could identify associations between variables,
but they could not address questions of causation. Selective use of hot spot studies and case control studies
is one method of achieving the different sample sizes required for different analytic purposes. Each of these
types of studies can be appropriate for specific situations.
Hot spot studies can provide detailed information on causes and relationships between contamination
by individual analytes and other variables. These studies are relatively inexpensive, can provide excellent
insights, and can be useful in developing hypotheses. Since hot spots are by definition not representative of
the nation such studies cannot be used to test hypotheses for application to the nation as a whole. They might
be used, for example, to examine the unexpected findings on the incidence of DCPA acid metabolite
contamination.
Case control studies with standardized methods can be used for comparing hot spot wells with other
wells in similar conditions to test broad hypotheses. Two potential drawbacks to case control studies are the
fact that hot spot wells may not be representative of all contaminated wells, and the set of comparisons that
can be made are limited by the size and scope of the study.
Additional Research. The Survey analysis identified a number of topics that could be useful areas for
future study:
• Analysis of links between surface and ground-water contamination. The Phase II
analysis identified associations between rainfall, the presence of nearby water bodies,
irrigation, and detections.
• Analysis of recharge and ground-water flow. The NFS did not collect data on the
direction of ground-water flow or to calculate recharge for sampled wells. Additional
evaluation of potential sources of contamination could be conducted with such data.
• Studies of seasonal and temporal effects on contamination. The NFS was not
designed to examine temporal variations or seasonal effects on contamination.
Continuing study of those questions is necessary to better understand how detections
and concentrations vary over time.
• Evaluation of site-specific data on soil characteristics. The NFS did not collect data
on soil profiles near sampled wells. Additional evaluation of pesticide persistence
and mobility could be conducted with such data.
• Investigation of the greater prevalence of contaminants in CWS wells compared to
rural domestic wells. The results supply some inconclusive evidence that CWS wells
may be more vulnerable to contamination than rural domestic wells.Greater
drawdown by large system wells, recharge areas for CWS wells, and non-agricultural
use of pesticides could be investigated.
• Development of statistical procedures for examining small samples of weighted
survey data. Currently existing statistical procedures for analyzing small sample sizes
do not allow the inclusion of weights such as those designed for the NFS.
• Analysis of non-agricultural uses of pesticides and their impacts on water quality.
NFS results involving DCPA acid metabolites obtained from data on non-farm use
of pesticides suggest that non-agricultural sources are strongly associated with
detections.
National Pesticide Survey: Phase II Report
-------
Chapter Five: Conclusions and Recommendations 166
• Analysis of pesticide degradates. The frequency of detection of DCPA acid
metabolites also confirms the importance of considering pesticide metabolites in the
design of studies.
• Development of improved data on pesticide use. Survey experience shows that better
data concerning pesticide use and application are required. Analysis of NFS
pesticide use data has included data from several sources; data used to construct
stratification variables, data from NFS questionnaires, RFF data, and manufacturer's
data. The problems encountered with analysis of stratification variables and pesticide
use data from questionnaires, and the differences between RFF and manufacturer's
data suggested that an accurate picture of pesticide use nationally is not available.
Reconciliation of these pesticide use data or provision of accurate data on a farm
level basis would allow development of a more accurate and useful pesticide use
database. Greater availability of accurate, locally precise pesticide use and
distribution data in the public domain would help to ensure that assessments of the
vulnerability of drinking water wells to contamination can be improved and made
more reliable.
The NFS Phase II results show that pesticides could be present at low levels of concentration in a
greater number of wells than originally estimated in Phase I. Several specific indicators are related to the
contamination of wells across a broad spectrum of site-specific conditions. Due to lack of sufficient detections
for statistical analysis, lack of findings of relationship between detections and possible factors indicating a
likelihood of detection should not be interpreted as showing that there is no relationship between detections
and these factors. There are, at present, no simple inexpensive methods to identify vulnerable wells likely to
experience contamination. NPS data cannot develop models that apply to site-specific conditions or account
for every conceivable factor that determines the fate and transport of chemicals in ground water. The Phase
II results show that it is difficult to identify a simple set of factors that may always reduce the presence of
pesticides in wells. The Phase II results show that many factors and practices contribute to the problem of
pesticide and nitrate contamination. Protection of drinking water quality is best accomplished using a
comprehensive integrated approach that emphasizes prevention of contamination to account for a wide variety
of factors.
National Pesticide Survey: Phase II Report
-------
National Pesticide Survey
Appendix A: Results of Univariate Chi Square (X2)
Tests with Significance Level (P-Value)
Between 0.05 And 0.1
-------
Appendix A A-1
S
•o
75
CO
t>
o>
«
Op
•D in
CD O
!§ °
8A
§ §
«5
« -1
58
|§
Is
£<2,
0) fc
ii
T3 D)
u
II
O £
II
OC
i*
u
S
§ P
o o
o o
£
e|
-1
ii
oc
o
To
5-S
o
II
5-5
CO
&
odd
S
odd
d d d d o d o
I + +
I I + + + + +
(0
o
S
Sz
So
8
U)
d
A
III! I
m
d
c
I
m
I 2 ° E
*•* eo o A
l;li
111
IS"
S i= C
I
National Pesticide Survey: Phase II Report
-------
Appendix A A-2
2J
•s
o
0
Q.
0
k>
Z
0
« *
CO •"
H
75 1
55
oc
0
75 «
5 0
c E
H
§«
0
_o
0 75
Q 5
— *j
OC
^
0
fl
11
o ^^
o
0
5
a
1
0
J3
75
I
§t>
1 £
0 *"
5-8
0
3
I
11
0 UJ
5-5
0
3
CO
>
1
c
~ 0
0 UJ
5-5
0
_D
I
ii
0) UJ
k. ^
5 o
rement
me
3 a
10
|1
>
0
J
i i i
0 0 C)
^ .
Z
o ci do
i i + +
£
ci
1
u?
o
® ^ o ^
5 S a> 5
« o E «
(0 *" S= *"
i "8 "S S "5
>•» ® ffi cr S
* E ff | S
i 111 1 c
i g> | i s if
| i s * g, •* s
E -n TJ -c to £ •=
>, ;. o " = « S
? ! fc § § S €2
S B Z c gj ' o §>S
Z E eOS'lo^x:!^ f3
^ I— ol CO *-Q3*'^QO
85 §
v|
a>
National Pesticide Survey: Phase II Report
-------
Appendix A A-3
_o
Q o>
a «
55
8 DC
•o
o
0 i-
°- •§ »
Ji
§ 0
E t5
g*
O
o
0 "S
0 v
2|
DC
0
e
0
fl
||
0
Variable
0
75
i
c •g
t>
fi -M| i
s ? &| «
-> 1 s* §1 =
B | 1 g, ^ S i
8|sS?a8*
s s 1 1 Ml i *
>-3£|5-=!lio
i « i S. i E^sSs.
1^5^53 laic?
8 3
m C
-------
Bibliography
Alexander, W.J. and S.K. Liddle. 1986. Ground Water Vulnerability Assessment in Support of the First Stage
of the National Pesticide Survey. Proceedings of the Agricultural Impacts on Ground Water
Conference. Omaha, Nebraska. 77-87.
Aller, L., T. Bennett, J.H. Lehr, R. Petty, and G. Hackett. April 1987. DRASTIC: A Standardized System
for Evaluating Ground Water Pollution Potential Using Hydrogeologic Settings. U.S. Environmental
Protection Agency. 600/2-87-035.
Baker, D.B. and R.P. Richards. 1991. Herbicides in Ohio's Drinking Water Supplies: A Quantitative
Exposure Assessment. Pesticides in the Next Decade: The Challenges Ahead. Proceedings of the
Third National Research Conference on Pesticides. Blacksburg, Virginia. 9-30.
Baker, D.B. and R.P. Richards. 1990. Herbicide Concentration Patterns in Rivers Draining Intensively
Cultivated Farmlands of Northwestern Ohio. Proceedings of a National Research Conference:
Pesticides in the terrestrial and aquatic environment. Blacksburg, Virginia. 103-120.
Banton, O. and J.P. Villenueve. 1989. Evaluation of Groundwater Vulnerability to Pesticides: A Comparison
Between the Pesticide DRASTIC Index and the PRZM Leaching Quantities. Journal of Contaminant
Hydrology. 4:285-296.
Barcelona, M.J. and T.G. Naymik. 1984. Dynamics of a Fertilizer Contaminant Plume in Groundwater.
Environmental Science and Technology. 18: 257-261.
Barlow, R.E., D.J. Bartholomew, J.M. Brenner, and H.D. Brunk. 1972. Statistical Inference Under Order
Restriction: The Theory and Application of Isotonic Regression. John Wiley and Sons. London.
Bartholomew, W.V. and F.E. Clark. 1965. Soil Nitrogen. American Society of Agronomy. Wisconsin.
Black, P.K., L. Johnson, and H. Lester. February 1991. National Pesticide Survey Data Quality Objectives:
Evaluation and Results. Proceedings of the Fourth Annual Ecological Quality Assurance Workshop.
Cincinnati, Ohio.
Burchfield, H.P. and E.P. Storrs. Dacthal. 67-77.
Cain, D., D.R. Helsel, and S.E. Ragone. March-April 1989. Preliminary Evaluations of Regional Ground-
Water Quality in Relation to Land Use. Ground Water. 27: (2) 230-244.
Cairns, T. and W.M. Rogers. January 1983. Acceptable Analytical Data For Trace Analysis. Analytical
Chemistry. 55: (1) 54-57.
Cassel, C, C.E. Sarndal, and J.H. Wretman. 1977. Foundations of Inference in Survey Sampling. John Wiley
and Sons. New York.
Choi, J.S., T.W. Fermanian, D.J. Wehner, and L.A. Spomer. February 1988. Effect of Temperature. Moisture.
and Soil Texture on DCPA Degradation. Agronomy Journal. 80:108-113.
Cohen, S.Z. February 1990. The Cape Cod Study. Golf Course Management. 58: (2) 26-44.
Cohen, S.Z., S.M. Creeger, R.F. Carsel, and C.G. Enfield. 1984. Potential for Pesticide Contamination of
Ground Water Resulting From Agricultural Use. Treatment and Disposal of Pesticide Wastes.
American Chemical Society Symposium Series 259. Washington, D.C. 2987-325.
National Pesticide Survey: Phase II Report
-------
Bibliography
Cohen, S.Z., S. Nickerson, R. Maxey, A. Dupuy, and J. Senita. A Ground Water Monitoring Study for
Pesticides and Nitrates Associated with Golf Courses on Cape Cod. 1-24.
Cooper, R.J. February 1990. Evaluating the Runoff and Leaching Potential of Turfgrass Pesticides. Golf
Course Management. 58: (2) 8-16.
Davoli, E., E. Benfenati, R. Bagnati, and R. Fenalli. 1987. Analysis of Atrazine in Underground Waters at
Part per Trillion Levels as an Early Warning Method for Contamination and for Soil Distribution
Studies. Chemosphere. 16: (7) 1425-1430.
Dippo, C.S., R.E. Fay, and D.H. Morganstein. August 1989. Computing Variances from Complex Samples
with Replicate Weights. Proceedings of the American Statistical Association, Section of Survey
Research Methods. Washington, D.C.
Efron, B. 1982. The Jackknife. the Bootstrap and other Resampling Plans. SLAM, CBMS-National Science
Foundation Monograph. 38.
Eisenreich, S.J., B.B. Looney, and J.D. Thornton. January 1981. Airborne organic contaminants in the Great
Lakes ecosystem. Environmental Science and Technology. 15: (1) 30-38.
Everitt, B.S. 1977. The Analysis of Contingency Tables: Monographs on Applied Probability and Statistics.
Chapman and Hall. London.
Exner, M.E. and R.F. Spalding. January - February 1984. Ground-Water Contamination and Well
Construction in Southeast Nebraska. Groundwater 23: (1) 26-34.
Exner, M.E. Winter 1990. Pesticide Contamination of Ground Water Artificially Recharged by Farmland
Runoff. Ground Water Monitoring Review. 147-159.
Final Descriptive Summary. 1986 Survey of Community Water Systems. October 1987. Research Triangle
Institute. (RTI/7805/02-02F).
Funari, E., G. Acquafresca, E. Area, M. Baldi, J. Bastianetti, A. Cappelli, A. Carniel, S. Chierici, A. Fanuzzi,
P. Ferraro, A Lopez, R. Mattioni, M. Narese, A Peretti, M. Salamana, and
G. Zapponi. 1989. Preliminary Report of the Atrazine and Molinate Water Supply Contamination
in Italy. Chemosphere. 18: (11/12) 2339-2343.
Gianessi, L.P. and C.A Puffer. The Use of Herbicides in the United States. Resources for the Future, Inc.
Washington, D.C.
Gianessi, L.P. and C.A Puffer. Estimation of County Pesticide Use on Golf Courses and by Urban
Applicators. Resources for the Future, Inc. Washington, D.C.
Gile, J.D. and J.W. Gillett. 1981. Transport and Fate of Organophosphate Insecticides in a Laboratory
Model Ecosystem. Journal of Agricultural Food and Chemicals. 29: 616-621.
Gilbert, M. and D.L. Lisk. 1978. Residues of Dacthal Herbicide in Carrots. Bulletin of Environmental
Contamination and Toxicology. 20: 180-183.
Glaser, J.A, D.L. Foerst, G.D. McKee, S.A Quare, and W.L. Budde. December 1981. Trace Analysis for
Wastewaters. Environmental Science and Technology. 15: (12) 1425-1435.
Goldberg, V.M. 1989. Groundwater Pollution by Nitrates from Livestock Wastes. Environmental Health
Prospectives. 83: 25-29.
National Pesticide Survey: Phase II Report
-------
Bibliography
Goolsby, D.A. and E.M. Thurman. 1990. Herbicides in Rivers and Streams of the Upper Midwestern United
States. Proceedings of the 46th Annual Meeting of the Upper Mississippi River Conservation
Committee. Bettendorf, Iowa.
Graham, J.A. June 1, 1991. Monitoring Groundwater And Well Water For Crop Protection Chemicals.
Analytical Chemistry. 63: (11) 613-622.
Gustafson, D.I. October 18, 1988. Groundwater Ubiquity Score: A Simple Method For Assessing Pesticide
Leachability. Environmental Toxicology and Chemistry. 8: 339-357.
Haas, C.N. and P.A Scheff. 1990. Estimation of Averages in Truncated Samples. Environmental Science and
Technology. 24: (6) 912-919.
Helsel, D.R. 1990. Less than obvious: Statistical treatment of data below the detection limit. Environmental
Science and Technology. 24: (12) 1767-1774.
Hileman, B. March 5, 1990. Alternative Agriculture. C&EN The American Chemical Society. Special
Report. 26-40.
Hines, J.W., W.L. Haugan, and D.B. Deluca. September 21, 1990. Minnesota Department of Agriculture
Water Quality Monitoring Biennial Report: MDA Monitoring Years 1988 and 1989. September 15,
1987 through September 14. 1989. Monitoring and Survey Unit, Technical Support Section,
Agronomy Services Division.
Hogmire, H.W., J.E. Weaver, and J.L. Brooks. 1989. Survey for Pesticides in Wells Associated with Apple
and Peach Orchards in West Virginia. Bulletin of Environmental Contamination and Toxicology.
44: 81-86.
Holden, L.R. and J.A Graham. February 1990. The National Alachlor Well Water Survey: Project
Summary. Monsanto Agricultural Company. St. Louis, Missouri. 1.
Hosmer, D.W. and S. Lemenshaw. 1989. Applied Logistic Regression. John Wiley and Sons. New York.
Huang, L.Q. and C.R. Frink. 1989. Distribution of Atrazine. Simazine. Alachlor. and Metolachlor in Soil
Profiles in Connecticut. Bulletin of Environmental Contamination and Technology. 43: 159-164.
Immerman, F.W. October 23, 1987. Final Descriptive Summary 1986 Survey of Community Water Systems.
Research Triangle Institute (RTI/7805/02-02F). Prepared for Office of Drinking Water, U.S.
Environmental Protection Agency.
Iowa State-Wide Rural Well-Water Survey: Perspectives of the SWRL Results. 1989. The University of
Iowa.
Iowa State-Wide Rural Well-Water Survey: Summary of State-Wide Results and Study Design. 1989. The
University of Iowa.
Iowa State-Wide Rural Well-Water Survey: Summary of Results: Pesticide Detections. 1989. The University
of Iowa.
Isensee, A.R., C.S. Helling, T.J. Gish, P.C. Kearney, C.B. Coffman, and W. Zhuang. 1988. Groundwater
Residues of Atrazine. Alachlor. and Cvanazine Under No-Tillage Practices. Chemosphere. 17: (1)
165-174.
Judkins, D.R. 1990. Fay's Method for Variance Estimation. Journal of Official Statistics. Sweden. 6: (3)
223-239.
National Pesticide Survey: Phase II Report
-------
Bibliography 4
Keith, L.H. June 1991. Report Results Right! Part 1. CHEMTECH. 352-356.
Keith, L.H. August 1991. Report Results Right! Part 2. CHEMTECH. 486-489.
Keith, L.H., W. Crummett, J. Deegan, R. Libbey, J. Taylor, and G. Wentler. 1983. Principles of
Environmental Analysis. Analytical Chemistry. 55. 2210-2218.
Kimball, C.G. and J. Goodman. 1989. Non-Point Source Pesticide Contamination of Shallow Ground Water.
ASAE Meeting Presentation. Paper No. 892529: 1-17.
Kirchmer, C.J. 1983. Quality control in water analyses: The definitions and principles underlying the practice
of quality control need to be critically evaluated. Environmental Science and Technology. 17: (4)
174A-181A.
Klaseus, T., G.C. Buzicky, and E.G. Schneider. February 1988. Pesticides and Groundwater: Surveys of
Selected Minnesota Wells. Minnesota Department of Health and Minnesota Department of
Agriculture.
Klaseus, T. and J.W. Hines. August 1989. Pesticides and Groundwater: A Survey of Selected Private Wells
in Minnesota. Minnesota Department of Health.
Lambert, D., B. Peterson, and I. Terpenning. June 1991. Non-Detects. Detection Limits, and the Probabilities
of Detection. Journal of American Statistical Association. 86: 266-277.
LeMasters, G. and D. Doyle. April 1989. Grade A Dairy Farm: Well Water Quality Survey. Wisconsin
Department of Agriculture, Trade and Consumer Protection. Wisconsin Agricultural Statistics
Service.
Long, G.L. and J.D. Winefordner. June 1983. Limit of Detection: A Closer Look at the IUPAC Definition.
Analytical Chemistry. 55: (7) 712A-724A.
Lorber, M.N., S.Z. Cohen, and G.D. DeBuchananne. Fall 1989. A National Evaluation of the Leaching
Potential of Aldicarb: Part 1. An Integrated Assessment Methodology. Ground Water Monitoring
Review. 109-125.
Lorber, M.N., S.Z. Cohen, S.E. Noren, and G.D. DeBuchananne. Winter 1990. A National Evaluation of
the Leaching Potential of Aldicarb: Part 2. An Evaluation of Ground Water Monitoring Data.
Ground Water Monitoring Review. 127-141.
Madison, R.J. and J.O. Brunett. 1984. Overview of the Occurrence of Nitrate in Ground Water of the United
States. National Water Summary 1984- Water Quality Issues. 93-105.
Mancino, C.F. February 1991. Nitrate and Ammonium Concentrations In Soil Leachate And N Leaching
Losses From Fertilizers Applied to Turfgrass. Golf Course Management. 68-72.
Marti, L.R., J. de Kanel, and R.C. Dougherty. 1984. Screening for Organic Contamination of Groundwater:
Ethvlene Dibromide in Georgia Irrigation Wells. Environmental Science and Technology. 18: 973-
974.
McCarty, P.L., M. Reinhard, and B.E. Rittmann. January 1981. Trace organics in groundwater: Processes
affecting their movement and fate in the subsurfaces environment, and research needs in this field are
discussed. Environmental Science and Technology. 15: (1) 40-49.
Miller, P.M. and E.D. Gomes. June 1974. Detection of DCPA Residues in Environmental Samples.
Pesticides Monitoring Journal. 8: (1) 53-58.
National Pesticide Survey: Phase II Report
-------
Bibliography
Miller, J.H., P.E. Keeley, R.J. Thullen, and C.H. Carter. January 1978. Persistence and Movement of Ten
Herbicides in Soil. Weed Science. 26: (1) 20-27.
Minnesota Department of Natural Resources. January 1989. Drought of 1988: Executive Summary.
Missedine, C. June 1989. Farm Bureau Releases Nitrate Study Results. Groundwater Education in Michigan
Program (GEM Notes). Supplement No. 6.
Mosteller, F. and J.W. Tukey. 1977. Data Analysis and Regression, a second course in statistics. Addison-
Wesley Publishing Company. Reading, Massachusetts.
Newman, N.C. and P.M. Dixon. April 1990. Uncensor: A Program to Estimate Means and Standard
Deviations for Data Sets With Below Detection Limits. AEL. 26-30.
Novick, N.J., R. Mukherjee, and M. Alexander. 1986. Metabolism of Alachor and Propachlor in Suspensions
of Pretreated Soils and in Samples from Ground Water Aquifers. Journal of Agriculture Food and
Chemistry. 34: 721-725.
O'Neil, H.J., T.L. Pollock, H.S. Bailey, P. Milburn, C. Gartley, and J. E. Richards. 1989. Dinoseb Presence
in Agricultural Subsurface Drainage from Potato Fields in Northwestern New Brunswick. Canada.
Bulletin of Environmental Contamination and Toxicology. 43: 935-940.
Palmer, W.C. February 1965. Meteorological Drought. U.S. Department of Commerce, Weather Bureau.
Research Paper 45. Washington, D.C.
Pankow, J.F., L.M. Isabelle, and W.E. Asher. 1984. Trace Organic Compounds in Rain. 1. Sampler Design
and Analysis by Adsorption/Thermal Desorption (ATP). Environmental Science and Technology.
18: (5) 310-318.
Pereira, W.E. and C.E. Rostad. 1990. Occurrence, Distributions and Transport of Herbicides and Their
Degradation Products in the Lower Mississippi River and Its Tributaries. Environmental Science and
Technology. 24: 1400-1406.
Pignatello, J.J. and S.Z. Cohen. 1990. Environmental Fate of Ethvlene Dibromide in Soil and Ground Water.
Review of Environmental Contaminant Toxicology. 112: 1-47.
Pionke, H.B. and J.B. Urban. January-February 1985. Effect of Agricultural Land Use on Ground-Water
Quality in a Small Pennsylvania Watershed. Groundwater. 23: (1) 67-80.
Porter, P.S., R.C. Ward, and H.F. Bell. 1988. The detection limit: Water quality monitoring data are plagued
with levels of chemicals that are too low to be measured precisely. Environmental Science and
Technology. 22: (8) 856-861.
Powers, J.F. and J.S. Schepers. 1989. Nitrate Contamination of Groundwater in North America. Agriculture,
Ecosystems and Environment. 26: 165-187.
Rathbun, R.E. and D.T. Tal. 1987. Volatilization of Ethvlene Dibromide from Water. Environmental
Science & Technology. 21: (3) 248-252.
Rhode Island Private Well Survey Final Report. May 1990. Ground Water Section, Department of
Environmental Management. Providence, Rhode Island.
Shah, B.V., L.M. LaVange, B.G. Barnwell, J.E. Killinger, and S.C. Wheeless. March 1989. SUDAAN:
Professional Software for SUrvev DAta ANalvsis. Research Triangle Institute. Research Triangle
Park, North Carolina.
National Pesticide Survey: Phase II Report
-------
Bibliography
Shoemaker, L.L., W.L. Magette, and A. Shirmohammadi. Winter 1990. Modeling Management Practice
Effects on Pesticide Movement to Ground Water. Ground Water Monitoring Review. 109-115.
Stukel, T.A, E.R. Greenberg, B.J. Dain, F.C. Reed, and N.J. Jacobs. 1990. A Longitudinal Study of Rainfall
and Coliform Contamination in Small Community Drinking Water Supplies. Environmental Science
and Technology. 24: 571-575.
Summary of NFS Domestic Well Survey Field Interviewer Debriefing on March 9. 1990. August 16, 1990.
Memorandum From Leslie Athey to David Marker.
Tierney, L. July 1988. XLISP-STAT : A Statistical Environment Based on the X-LISP Language Version
2.0. University of Minnesota, School of Statistics. Technical Report 528.
U.S. Congress. May 1990. Beneath the Bottom Line: Agricultural Approaches to Reduce Agrichemical
Contamination of Groundwater. Summary. Office of Technology Assessment. Washington, D.C.
OTA-F-417.
U.S. Department of Commerce. 1988. County and City Data Book. Bureau of the Census.
U.S. Environmental Protection Agency. July 1991. Methods for Assessing the Sensitivity of Aquifers to
Pesticides Contamination. Office of Ground-Water Protection. Washington, D.C. Prepared by
Geraghty & Miller, Inc. and ICF Inc.
U.S. Environmental Protection Agency. November 1990. National Pesticide Survey Phase I Report.
Washington, D.C. EPA 570/9-90-015.
U.S. Environmental Protection Agency. September 1990. County-Level Fertilizer Sales Data. Office of Policy
Planning and Evaluation. Washington, D.C. PM-221.
U.S. Environmental Protection Agency. 1989. Revisions to the Superfund Public Health Evaluation Manual
(1986). Environmental Criteria and Assessment Office. Washington, D.C.
U.S. Environmental Protection Agency. May 1986. Pesticides in Ground Water: Background Document.
Washington, D.C. WH-550G.
U.S. Environmental Protection Agency. 1971. Study of the Degradation of Dimethyl Tetrachlorterephthalate.
Interaction of Herbicides and Soil Microorganisms. 15-18.
Walker, M.J. and K.S. Porter. Winter 1990. Assessment of Pesticides in Upstate New York Ground Water:
Results of a 1985-1987 Sampling Survey. Ground Water Monitoring Review. 116-126.
Watschke, T.L. and R.O. Mumma. The Effect of Nutrients and Pesticides applied to Turf on the Quality of
Runoff and Percolating Water. Environmental Resources Research Institute (ER 8904) and
Departments of Agronomy and Entomology, The Pennsylvania State University.
Watschke, T.C. February 1990. The Environmental Fate Of Pesticides. Golf Course Management. 58: (2)
18-24.
Wauchope, R.D. February 20, 1991. Selected Values for Six Parameters from the SCS/ARS/CES Pesticide
Properties Database: A Brief Description.
Weaver, M.F., S.Z. Cohen, and J.J. Pignatello. 1988. Environmental Chemistry of Ethvlene Dibromide.
Proceedings of the Agricultural Impacts on Ground Water- A Conference. National Water
Association. Dublin, Ohio.
National Pesticide Survey: Phase II Report
-------
Bibliography
Wells, D. and E. Waldman. June 26, 1991. County Level Assessment of Aldicarb Leaching Potential.
Ground Water Technology Section, Environmental Fate and Ground Water Branch, Office of
Pesticide Programs, U.S. Environmental Protection Agency. 1-88.
Williams, D.T., E.R. Nestmann, G.L. LeBel, P.M. Benoit, and R. Otson. 1982. Determination of Mutagenic
Potential and Organic Contaminants of Great Lakes Drinking Water. Chemosphere. 11: (3) 263-276.
Williams, W.M., P.W. Holden, D.W. Parsons, and M.N. Lorber. 1988. Pesticides in Groundwater Database:
1988 Interim Report. U.S. Environmental Protection Agency.
Wilson, L.J. November 1989. Groundwater Monitoring in the Big Spring Basin 1984-1987: A Summary
Review. Iowa Department of National Resources. 1-68.
Wilson, M.P., E.P. Savage, D.D. Adrian, M.J. Aaronson, T.J. Keefe, D.H. Hamar, and J.T. Tessari. 1987.
Groundwater Transport of Herbicide. Atrazine. Weld County. Colorado. Bulletin of Environmental
Contamination and Toxicology. 39: 807-814.
*U.S. GOVERNMENT PRINTING OFFICE: 1992—617-003/47052
National Pesticide Survey: Phase II Report
------- |