RESEARCH TRIANGLE INSTITUTE
RH/5100/17-02F
March 1992
NATIONAL HOME AND GARDEN PESTICIDE USE SURVEY
FINAL REPORT, VOLUME II:
Survey Design, Implementation, and Analysis Methods
Prepared by:
Roy W. Whltmore
Janice E. Kelly
Pamela L. Reading
Prepared for:
U.S. Environmental Protection Agency
Office of Pesticides and Toxic Substances
Biological and Economic Analysis Branch
Contract No. 68-WO-0032
RTI Work Assignment Leader:
Roy W. Whltmore
EPA Work Assignment Manager:
Edward Brandt
POST OFFICE BOX 12194 RESEARCH TRIANGLE PARK, NORTH CAROLINA 27709-21 94
vy'.. Printed on Recycled Paper
-------
CONTENTS - VOLUME I (Bound Separately)
Page
TABLES v
FIGURES v111
ACKNOWLEDGMENTS 1x
1. EXECUTIVE SUMMARY 1
1.1 Background 1
1.2 Study Objectives and Target Population 1
1.3 Summary Description of the Sampling Design 2
1.4 Overview of Results 3
1.4.1 Population Characteristics 3
1.4.2 Storage of Pesticide Products 6
1.4.3 Difficulty Opening Containers 13
1.4.4 Disposal of Pesticides 13
1.4.5 Severity of Pest Problems 16
1.4.6 Consumer Satisfaction 19
1.4.7 Use of Pest Control Services 21
2. STATISTICAL ANALYSIS RESULTS 23
2.1 Household-Level Analyses 24
2.1.1 Severity of Pest Problems 24
2.1.2 Continuous-Use Products 37
2.1.3 Liquid Chlorine Bleach 40
2.1.4 Disposal 40
2.1.5 Pest Control or Lawn Care Service 46
2.2 Product-Level Analyses 49
2.2.1 Use of the Master Product Label File 50
2.2.2 Number of Products 1n Storage 51
2.2.3 Container Characteristics 57
2.2.4 Security of Storage 57
2.2.5 Child Resistant Packaging 71
2.2.6 Length of Storage 78
2.2.7 When Last Used 85
2.2.8 Age-Sex Distribution of Users 96
2.2.9 Frequency of Application 107
2.2.10 Safety Precautions 119
2.2.11 Consumer Satisfaction 119
2.2.12 Pest Treated by Site of Application 124
2.2.13 Active Ingredient Tabulations 125
3. RECOMMENDATIONS FOR A FUTURE NHGPUS 129
3.1 Sampling Design 129
3.2 Recruiting Participants 129
-------
CONTENTS - VOLUME I (cont.)
3.3 Data Collection Instruments 130
3.3.1 Control Form 130
3.3.2 Study Questionnaire ... 130
3.3.3 Card A Pest List 135
3.3.4 Card B Site of Application List 135
3.3.5 Card C Pesticide Application Method List 136
3.3.6 Card D Safety Precaution List 136
3.3.7 Testing Survey Instruments 137
REFERENCES 139
Appendix A. Study Control Form, Questionnaire, and Cards A-D A-l
Appendix B. Distribution of Sites Treated for Each Pest B-l
Appendix C. Distribution of Pests Treated for Each Site of
Application C-l
Appendix D. Distribution of Sites Treated for Each of the 77 Active
Ingredients That Occurred 25 Times or More in the NHGPUS
Data Base D-l
Appendix E. Distribution of Active Ingredients Applied to Each Site E-l
Appendix F. Distribution of Treated Pests for the 63 Combinations of
Active Ingredients That Occurred 25 Times or More in the
NHGPUS Data Base F-l
Appendix G. Number of Products and Frequency of Use by Active
Ingredient G-l
ii
-------
CONTENTS VOLUME II
Page
TABLES v
EXHIBITS v
ACKNOWLEDGMENTS v11
4. SAMPLING DESIGN 1
4.1 First-Stage Sample of Counties 1
4.2 Second-Stage Sample of Subcounty Areas 3
4.3 Third-Stage Sample of Housing Units 7
5. DEVELOPING SURVEY INSTRUMENTS 13
5.1 Study Questionnaire 13
5.2 Notebook of Pest Sketches 16
5.3 Lead Letter and Study Brochure 16
5.4 Field Interviewer's Manuals 17
5.5 Training Materials 18
6. FIELD OPERATIONS 21
6.1 Counting and Listing Activities 21
6.1.1 Recruiting Field Supervisors 21
6.1.2 Recruiting Field Interviewers 21
6.1.3 Training 22
6.1.4 Counting and Listing 22
6.2 Primary Data Collection Activities 25
6.2.1 Recruiting Field Supervisors 25
6.2.2 Recruiting Field Interviewers 25
6.2.3 Training 25
6.2.4 Data Collection Activities 27
6.2.5 Validation Interviews 33
6.2.6 RTI Protection of Human Subjects Committee Review 36
7. DATA PROCESSING 41
7.1 Manual Editing 41
7.2 Data Entry 41
7.3 Computerized Editing 42
7.4 Data Processing Management 43
ill
-------
CONTENTS VOLUME II (cont.)
8. SAMPLING WEIGHTS 47
8.1 Weights Based on the Sampling Design 48
8.J.1 First-Stage Sample of Counties 48
8..1.2 Second-Stage Sample of Subcounty Areas 49
8.1.3 Third-Stage Sample of Housing Units 50
8.2 We1ght1ng-Class Adjustment for Unit Nonresponse 51
8.3 Compensating for Item Nonresponse 53
8.4 Quality Assurance Procedures 54
9. STATISTICAL ANALYSIS METHODS 57
9.1 Estimating Totals and Associated Variances 57
9.2 Estimating Means, Proportions, and Associated Variances 59
9.3 Suppression Rule 61
9.4 Statistical Inferences 62
9.4.1 Confidence Interval Estimates 62
9.4.2 Testing for Significant Differences 1n Proportions 62
REFERENCES 65
Appendix H. First-Stage Sampling Design for the 1981 National
Household Pesticide Usage Survey H-l
Appendix I. Survey Pest Notebook 1-1
Appendix J. Advance Mailing Materials J-l
1 v
-------
TABLES
Number Page
4.1 Characteristics of the NHGPUS Sample by County 10
6.1 Characteristics of Field Interviewers 26
6.2 Distribution of Final Interview Results 35
7.1 Standard RTI Consistency Codes for Survey Data Bases 44
8.1 Response Rates by Weighting Classes 52
EXHIBITS
Number Page
6.1 Training Agenda for Counting and Listing 23
6.2 Field Supervisor Training Agenda 28
6.3 General Field Interviewing Training Agenda 30
6.4 Training Agenda - August 1990 31
6.5 Refusal Conversion Letter 34
6.6 Field Interview Validation Form 37
6.7 Validation Letter 38
v
-------
THIS PAGE LEFT BLANK INTENTIONALLY
vl
-------
ACKNOWLEDGMENTS
The RTI Project Director for the National Home and Garden Pesticide Use
Survey was Dr. Roy W. Whltmore. Dr. Whltmore also served as task leader
for sample selection, sample weighting, and statistical analysis. He was
assisted by three task leaders: Ms. Nancy C. Monroe, Dr. Pamela L. Reading,
and Ms. Janice E. Kelly. Ms. Monroe served as task leader for developing
the survey instruments. Dr. Reading served as task leader for data
processing. Ms. Kelly served as task leader for survey operations.
Dr. Whltmore was assisted primarily by Mr. Shelton M. Jones,
Ms. Franclne A. Burt, Ms. Sally L. Branson, and Ms. Brenda K. Porter.
Mr. Jones provided computing support for sample selection. Ms. Burt
prepared maps and segment kits for all sample areas. Ms. Branson provided
computing support for sample weighting and statistical analysis.
Ms. Porter provided secretarial support.
Ms. Monroe developed the study questionnaire and other data collection
Instruments 1n close consultation with the RTI project director. She was
assisted primarily by Ms. Andrea F. Garner, who formatted the text of the
survey instruments for printing.
Dr. Reading was assisted primarily by Ms. L1111e B. Barber and Ms.
Jean C. Service. She was also assisted by the RTI data editing and data
entry staff, headed by Ms. Cherlylen Ford and Ms. Patricia M. Best,
respectively. Ms. Barber wrote the data entry programs, which contained
some rather complex logic. Ms. Service provided computing support for the
survey control system, preparation of analysis data files, and automated
data editing.
Ms. Kelly was assisted primarily by Mr. Michael A. Morgan, Ms. Marti L.
Dunn, Ms. Vivian E. Adklns, Mr. Randall S. Keesling, Ms. Karl A. Lambrlght,
and Ms. Jane L. Grant. Mr. Morgan assisted with training the interviewers.
He also received weekly field reports from the Field Supervisors during
data collection and provided summaries of these reports to the project
leadership. Ms. Dunn and Ms. Adklns provided Invaluable logistical
assistance with preparation of materials for mailing to the Interviewers
and with coordination of rooms and activities at the Interviewer training
sessions. Mr. Keesling conducted the general interviewer training for new
vii
-------
Interviewers. Ms. Lambrlght and Ms. Grant provided word processing
support.
Other RTI staff who provided Important support Include Mr. Donald W.
Jackson, Ms. Sophie K. Burkhelmer, and Ms. Michelle M. Hoffman.
Mr. Jackson attended several of the Initial project planning sessions and
assisted 1n securing the necessary support staff so that the project would
run smoothly. Ms. Burkhelmer directed compilation and printing of the pest
sketch notebook used to assist the survey participants with Identifying the
pest category that best represented their pest problem. Ms. Hoffman
assisted with conducting two rounds of pilot testing for draft
questionnaires.
The EPA Project Officers, Mr. Edward Brandt and Mr. Thomas C. Harris,
also provided invaluable assistance. They are commended for developing
table shells at the beginning of the project that served as the basis for
developing the questionnaire and the statistical analysis plan. They
consistently provided timely response to issues raised by the RTI project
staff, especially regarding questionnaire development, statistical analysis
plans, and data editing. They attended the field interviewer training
sessions, which helped emphasize the Importance of the study to the field
interviewers. Mr. Harris assumed the lead role for developing the notebook
of pest sketches; his assistance 1n this regard was Invaluable.
Mr. Craig VMssman, Consumer Products Division of Chevron Chemical
Company, provided samples of pesticide packaging for use 1n training the
field interviewers. His assistance 1s gratefully acknowledged.
v111
-------
4. SAMPLING DESIGN
The NHGPUS was a one-time, cross-sectional survey of the use of
pesticides In and around homes 1n the United States. The sampling design
can be described as a stratified, three-stage probability sampling design.
A probability sample 1s one In which all units In the population have a
calculable, positive probability of being Included 1n the sample.
Probability sampling procedures were necessary for the NHGPUS because of
the need to extend Inferences from the sample to the target population.
Probability sampling requires the existence of sampling frames, or
lists, from which elements of the target population can be selected Into
the sample. A multistage probability sampling procedure was used to select
the NHGPUS sample for reasons of convenience and cost efficiency. Thus,
counties were selected from a list of all counties and the District of
Columbia at the first stage of sampling. Subcounty areas were selected at
the second stage of sampling from lists of areas that completely covered
the entire land area of the sample counties. Sample housing units were
then selected from lists prepared by field staff that provided coverage of
all housing units currently located 1n the second-stage sample areas.
The first- and second-stage samples were stratified by variables that
are potentially related to household use of pesticides. Stratification
refers to partitioning the sampling frame Into disjoint subsets, called
strata, and Independently selecting a sample to represent each stratum. If
the strata are related to the analysis variables, stratification can
Improve the precision of survey statistics. If they are unrelated, loss of
precision due to stratification 1s virtually Impossible (Cochran, 1977,
Section 5.6). Moreover, selecting a sample from each stratum guarantees
that the population subsets represented by the strata are all appropriately
represented In the sample.
4.1 First-Stage Sample of Counties
A first-stage sample of counties was selected for the planned National
Household Pesticide Usage Survey 1n 1981, but the survey was never
Implemented. Considerable effort was expended at that time to stratify the
sampling frame by Census Division, an urbanization code, average annual
precipitation, average annual temperature, and ethnic composition of the
1
-------
population (as measured by percent black population).1 The original
sampling frame was not available for selecting the NHGPUS sample, but the
sample of 180 county selections was available. To make efficient use of
the previous stratification, a subsample of 60 selections was chosen for
the current NHGPUS.
The sampling design for the first-stage sample selected 1n 1981 1s
presented 1n Appendix H. The remainder of this section discusses the
process of selecting the subsample of 60 county selections for the 1990
NHGPUS sample.
The 1981 sample was stratified Implicitly by sorting the sampling frame
by the stratification variables and sequentially selecting counties using a
probability minimum replacement (pmr) sampling algorithm (Chromy, 1979).
This Implicit stratification was preserved 1n the subsample by using the
same pmr sampling algorithm to select the subsample from a frame consisting
of the previous 180 county selections sorted 1n the order 1n which they
were originally selected.
Counties were selected for the 1981 sample with probabilities
proportional to the estimated number of non-farm housing units 1n each
county because the target population at that time Included only non-farm
housing units. Because the current target population Includes both farm
and non-farm households, counties were selected for the subsample with
probabilities proportional to the ratio of "the estimated number of
occupied housing units 1n each county 1n December 1988" (obtained from
Market Statistics, Inc.) divided by "the size measure used for the 1981
sample," which yields overall probabilities of selection proportional to
the current estimates of total occupied housing units as shown below.
The 60 selections Included two counties that, because of their large
size, were selected twice: Los Angeles County, California and Cook County,
Illinois. Thus, 58 distinct counties were selected. These 58 sample
counties are located 1n 29 different States.
Letting 1=1,2,...,Ni Index the first-stage sampling units (FSUs)
(counties on the original sampling frame), we define the following
Percent black population 1n the county was used at the fifth level of
nested stratification for sampling counties (see Appendix A) and therefore
had little actual effect. Race was not used to stratify the second-stage
sample of subcounty areas (see Section 4.2).
2
-------
notation.
S(1) = the original size measure for the 1-th FSU (estimated number of
non-farm housing units 1n 1981).
Mi(1) = the new size measure for'the 1-th FSU (estimated number of
occupied housing units 1n December 1988).
n(1) = number of selections of the 1-th FSU 1n the 1981 sample.
mj(1) = number of selections of the 1-th FSU for the current NHGPUS
sample.
Then, the expected number of selections of the 1-th FSU for the 1981 sample
1 s
N1
E[n(1)] = 180 S(1) / E1 S(1) (4-1)
1=1
because there were 180 county selections at the first stage of sampling.
Since the 60 subsample FSUs were selected for the NHGPUS with
probabilities proportional to the ratio of the new to the old size
measures, the conditional expected number of selections for the 1-th FSU,
given that 1t was selected for the 1981 sample, S\, 1s
60 M.(1) / S(1)
E[m1(1) |1cSi] =lj * (4-2)
E1 [Mj(1) / S(1)] Is (1)
1=1 1 *1
where I<- 1s a (0,l)-1nd1cator of Inclusion 1n the 1981 sample.
^1
Therefore, the unconditional expected frequency of selection of the
1-th FSU (county) selected Into the NHGPUS sample 1s the product of (4-1)
and (4-2), namely
60 M,(i)
EfajO)] = (4-3)
[ E1 S(1)] { E1 I- (1) [Mj(1) / S(1)] / 180}
1=1 1=1 51 1
As previously stated, the expected frequency of selection 1s proportional
to the new size measure, Mi(1).
4.2 Second-Stage Sample of Subcounty Areas
Five subcounty areas were selected at the second stage of sampling for
each county selection at the first stage. The subcounty areas were defined
3
-------
by Census blocks and enumeration districts (EDs) because they provide
complete coverage of each sample county, and they are the smallest
geographic areas for which Census data are available. Data were extracted
from the 1980 Census Summary Tape F1.le 1-A (STF 1-A) for all blocks and EDs
1n each sample county to construct a stratified sampling frame of subcounty
areas.
Second-stage sampling units (SSUs) were created by combining blocks and
EDs, as necessary, to form sampling units that had a minimum 1980 Census
count of 35 occupied housing units. The block and ED records were combined
1n such a manner that the units combined were usually geographically
proximate to each other to minimize field travel costs. The combining
procedure was also designed to minimize the occurrence of large sampling
units and units that cross Census block group boundaries and minor civil
division (MCD) boundaries. Every block and ED 1n each sample county,
Including those that had no population 1n 1980, was Included 1n an SSU to
provide complete geographic coverage of each sample county at the second
stage of sampling.
The SSUs had to be relatively large to protect against selecting areas
that did not currently contain any occupied housing units and to facilitate
selection of noncompact clusters of sample housing units at the third stage
of sampling. Conversely, listing all current housing units 1s more costly
for large sampling units. Therefore, the sampling units had to be kept as
small as practically possible. After Investigating the distributions of
sampling unit sizes resulting from minima of 35, 50, and 65 occupied
housing units (based on the 1980 Census counts), 35 housing units was
selected as the most appropriate minimum size for the SSUs.
The second-stage sample of subcounty areas selected for each sample
county was stratified by urbanlclty, socioeconomic status, and proportion
of multiple-family housing units. These stratification variables were
thought to be potentially related to occurrence of pests and use of
pesticides. To the extent that they are related, the stratification will
result 1n more precise survey estimates.
The urbanlclty code used for stratification was defined to have two
levels: urban and rural. SSUs were coded as urban 1f most of the housing
units in the SSU were In a Place (incorporated or Census-defined city or
4
-------
town) or In a Census-defined urbanized area (the urban area surrounding a
metropolitan area). All other SSUs were coded as rural.
The socioeconomic status of an SSU was measured by the average
appraised value of the dwellings 1n the SSU, combining the 1980 Census data
for owned and rented housing units. For rented dwellings, the appraised
value was estimated as 100 times the monthly rent. The distribution of
housing unit values was examined separately for urban and rural SSUs within
each sample county to define high and low socioeconomic strata within each
urbanlclty stratum.
Likewise, the distribution of the proportion of multiple-family
dwellings was examined for each socioeconomic stratum of each sample county
and used to define strata with high and low proportions of multiple-family
dwellings.
Having constructed the stratified second-stage sampling frame, five
subcounty areas were selected for each first-stage county selection. Ten
SSUs were selected for Los Angeles County, California and for Cook County,
Illinois because each county received two Independent selections at the
first stage of sampling.
Within each sample county, the second-stage sampling units were
selected with probabilities proportional to size using the same sequential
probability minimum replacement (pmr) sampling algorithm that was used to
select the first-stage sample of counties (Chromy, 1979). The size measure
for each area was the 1980 Census count of occupied housing units for the
area. The sampling frame for each county was sorted 1n a serpentine manner
by the following variables:
(1) Type of community (urban or rural),
(2) Socioeconomic status (high or low),
(3) Proportion of multiple-family dwellings (high or low), and
(4) Size (1980 Census count of occupied housing units).
Sequential selection from the serpentine sorted sampling frame ensured
proportional representation of the strata formed by the first three
variables, as discussed by Williams and Chromy (1980). The final sort by
size of the SSUs helped ensure that both large and small SSUs would be
represented 1n the sample.
Five SSUs were selected for each fist-stage selection. A single sample
of 10 SSUs was selected for each of the double-hit FSUs (Los Angeles
5
-------
County, CA and Cook County, IL) to ensue that no SSUs 1n these counties
would be selected more than once. One SSU (a large ED 1n Lipscomb County,
TX) received two selections. Since there were 60 FSU selections at the
first stage of sampling, the total, number of distinct SSUs selected was
299. These sample SSUs are referred to as sample segments.
Letting J=1,2,...N2(1) Index the second-stage sampling units (SSUs) In
the 1-th county, we define the following notation.
M2O1J) = the size measure for the j-th SSU In county "1" (1980 Census
count of occupied housing units).
nfcO.j) = number of selections of the J-th SSU In county "1."
Then, since 5 SSUs were selected with probabilities proportional to size
for each FSU selection, the conditional expected frequency of selection for
the J-th SSU, given that county "1" is 1n the NHGPUS first-stage sample,
Si, 1s2
ECmgd.J) |1cS*] = 5 M2(1,J) / E*(i) M2(1,J) (4-4)
The unconditional expected frequency of selection for the j-th SSU 1n
county "I" Is then given by the product of (4-3) and (4-4).
In May of 1990, approximately three months before beginning NHGPUS
field data collection, field staff were sent to the sample segments (SSUs)
to prepare lists of the current housing units to enable selection of the
third and final stage of the sample. Their first task 1n each area was to
make a quick count of the total housing units 1n each sample segment. If
there were too many housing units present to efficiently 11st all of them
(approximately 200 or more), the sample segment was divided Into
subsegments, a quick count of the current housing units was obtained for
each subsegment, and one subsegment was selected with probability
proportional to the quick count. In this case, only the housing units 1n
the sample subsegment were listed (Instead of all housing units 1n the
sample SSU) for the third stage of sampling.
^Technically, a single sample of 10 SSUs was selected for each of the
double-hit FSUs (Los Angeles County, CA and Cook County, IL). In this
case, the second-stage sample size 1s a random variable [5 mi(1)j, but the
unconditional expected frequency of selection 1s still given by the product
of (4-3) and (4-4).
6
-------
Letting k=l,2,...,N3(1,j) Index the subsegments created for the J-th
segment 1n county "1," define M3(1,J,k) to be the quick count of housing
units In the (1,j,k)-th subsegment. Then, the conditional probability of
selecting the (1,j,k)-th subsegment,.given that the j-th segment was In the
second-stage sample, S2, 1s3
M3(1.J.k)
P3(1J.k|jeS2) =
N ^ ^ 1f the (1,J)-th segment was
E3 MlJ.k) ^segmented
k=l J
(4-5)
. 1 1f the (1,j)-th segment was not subsegmented.
The unconditional probability of selection for the (1,j,k)-th subsegment 1s
then given by the product of (4-3), (4-4), and (4-5).
4.3 Third-Stage Sample of Housing Units
Having located the sample segment or subsegment, the field staff listed
all potential housing units 1n each selected area. The sample housing
units were then selected from these lists at the third and final stage of
sampling.4 Two area segments did not currently contain any housing units,
so lists of housing units were prepared for 298 sample segments.
When all housing units 1n the sample segments (or subsegments) had been
listed, the sample was allocated to the segments to achieve approximately
equal probabilities of selection for all housing units In the target
population. This resulted 1n larger sample allocations to the areas that
had grown the most since the 1980 Census. However, the sample allocation
was constrained to be no more than 55 sample housing units for most sample
counties because we believed that this was the maximum work load that could
be completed by a single interviewer during the approximately six-week
period of field data collection.
3A single sample of two subsegments was selected from the Lipscomb County,
TX, segment that received two selections at the second stage. In this
case, the number of subsegments selected 1s a random variable equal to the
number of second-stage selections. The unconditional probability of
selection of the subsegments 1s, nevertheless, given by the product of (4-
3), (4-4), and (4-5).
^Technically, this was the fourth stage of sampling for area segments that
were subsegmented.
-------
The sample was designed to yield complete data for 2,000 responding
households. Assuming that 88 percent of the listed dwellings would be
occupied housing units (based on recent RTI experience In national surveys)
and that 85 percent of the households would respond, a primary sample of
2,674 sample housing units was selected. The average allocation to the 298
sample segments was nine sample housing units. Although the allocation to
most segments was nine housing units, the allocation to the Individual
segments ranged from 2 to 22 housing units.
One additional sample line (potential housing unit) was selected from
each of the 298 area segments as a reserve or "hold" sample. Ten percent
of the sample lines 1n each segment were selected for validation Interviews
as a quality assurance measure. Two segments 1n each county, were randomly
selected as an early report sample to be worked first. The field status of
the early report sample was monitored regularly to determine 1f any lines
from the hold sample should be worked. In fact, none of the hold sample
lines were ever released for field data collection.
Letting £,=1,2,...,N4(1,j,k) Index the potential housing units listed
for the (1,j,k)-th sample segment (or subsegment), define 1114(1,J,k) to be
the final sample allocation to the (1,j,k)-th segment. Given the final
sample allocation, an equal probability sample of housing units was
selected from those listed for each segment using a sequential probability
minimum replacement (pmr) selection algorithm (Chromy, 1979). Therefore,
the conditional probability of selecting the (1,j,k,£.)-th listed housing
unit, given that the k-th subsegment was selected Into the sample S3, 1s
P4(1»j»k,£.|keS3) = m4(1,j,k) / N4(1,J,k). (4-6)
A "missed housing unit procedure" was employed when sample housing
units were Identified 1n the field to ensure that all housing units that
could be Identified at the time of field data collection had a positive
probability of being Included In the sample. This procedure Included 1n
the sample not only the housing units listed on the selected sample lines,
but also
any non-11sted housing units located within the selected sample
housing units, and
any non-listed housing units located between a selected sample
housing unit and the next listed housing unit.
8
-------
The probability of selection for each "added housing unit" Is the same as
that for the listed sample housing unit that resulted 1n Its inclusion 1n
the sample.
Characteristics of the final NHGPUS sample are presented 1n Table 4.1
by county. The Census Region and Census Division to which each county
belongs Is identified. In addition, the number of eligible and
participating sample households 1s presented for each county. This table
Illustrates the geographic diversity of the NHGPUS sample. At the same
time, 1t makes clear that the NHGPUS sample size Is too small to make
statistically defensible estimates for Individual States or Census Regions.
The survey was designed to support national-level statistical Inferences.
9
-------
Table 4.1 Characteristics of the NHGPUS Sample by County
Census Census
No. Area
Sam
pie Households3
Region Division
State
County
Segments
No. Eligible
b No. Participating
Northeast New England
Maine
Kennebec
5
42
37
New Hampshire
Rockingham
5
49
45
Connecticut
New Haven
5
48
37
Middle Atlantic
New York
Kings
5
32
30
Monroe
5
39
33
New York
5
36
29
Queens
5
42
36
New Jersey
Bergen
5
46
38
Mercer
4C
28
20
Morris
5
41
30
Pennsylvania
Lackawanna
5
41
41
Lawrence
5
42
41
North East North Central
Ohio
Lake
5
36
33
Central
Summit
5
43
37
Indiana
Hendricks
5
34
31
La "Porte
5
38
34
Marlon
5
38
26
Illinois
Cook
10
83
71
Michigan
Livingston
5
45
37
Wayne
5
40
27
W1sconsIn
Wood
5
41
39
West North Central
Minnesota
Anoka
5
53
50
Missouri
Boone
5
45
42
St. Louis
5
41
37
North Dakota
Foster
5
43
39
Nebraska
Madison
5
48
42
(continued)
-------
Table 4.1 Characteristics of the NHGPUS Sample by County (cont.)
Census
Region
Census
Division
State
County
No. Area
Segments
Sample Households8
No. Eligible" No. Participating
South
South Atlantic
West
Mountain
Maryland
Montgomery
5
36
26
Prince George's
5
37
25
South Carolina
Chester
5
41
41
Georgia
Catoosa
5
33
29
Fulton
5
44
42
Treutlen
5
31
30
Florida
Oade
4c
30
27
Hernando
5
49
43
Palm Beach
5
25
21
Volusia
5
39
34
Kentucky
Jefferson
5
54
48
A1abama
Dallas
5
39
37
Jefferson
5
45
41
Arkansas
Jefferson
5
38
31
Louisiana
De Soto
5
34
31
Texas
Harrls
5
46
lid
Lipscomb
5
38
31
Matagorda
5
30
29
McLennan
5
50
46
Randal 1
5
43
38
Colorado
Denver
5
42
37
Arizona
Yavapai
5
26
23
Utah
Cache
5
44
42
Salt Lake
5
39
37
(continued)
-------
Table 4.1 Characteristics of the NHGPUS Sample by County (cont.)
Census
Region
Census
Division
State
County
No. Area
Segments
Sample Households3
No. Eligible*1
No. Participating
West
(cont.)
TOTAL
Pacific
Oregon
California
Multnomah
5
47
37
Washington
5
44
38
Contra Costa
5
35
29
Los Angeles
10
87
66
Orange
5
37
32
San Diego
5
55
50
San Francisco
5
44
24
Santa Barbara
5
49
40
298
2,447
2,078
a Also see Table 8.1.
b All housing units occupied as permanent residences were eligible for the survey.
c One of the five area segments selected for this county did not contain any housing units at the time of
the survey (Aug-Sept 1990).
d 27 completed Interviews for Harris County, Texas, were lost 1n the mall when the Interviewer mailed all
of them to RTI 1n a single envelope, contrary to the established NHGPUS procedures. RTI verified by
telephone that Interviews were conducted for at least some of these sample households.
-------
5. DEVELOPING SURVEY INSTRUMENTS
The primary survey Instrument that had to be developed for the NHGPUS
was the study questionnaire, which ts discussed 1n Section 5.1. A notebook
of pest sketches was also developed to assist the survey respondents with
selecting the pest categories that represented the pests they were treating
or with which they were having a problem. Development of the pest notebook
1s discussed 1n Section 5.2. In addition, a lead letter and study brochure
were developed for an advance mailing to sample homes 1n an effort to
achieve the highest possible response rate. Development of these materials
1s discussed 1n Section 5.3. Manuals for training Interviewers and for
Interviewer reference during the study are discussed 1n Section 5.4.
5.1 Study Questionnaire
Development of the NHGPUS questionnaire began shortly after the project
klckoff meeting late 1n March 1989. The Agency developed table shells for
analyses that they would like to be able to conduct at the conclusion of
the survey. RTI developed a 11st of data Items from the table shells and a
flow chart that attempted to put the data Items 1n a logical sequence for
data collection. The Initial draft questionnaire was developed from that
flow chart.
We commend the Agency for making a concerted effort to specify the
analyses that they would like to be able to conduct. The table shells were
effectively an analysis plan that gave focus to the questionnaire
development process and provided a clear rationalization for each
questionnaire Item.
To the extent possible, data Items were borrowed from tested
questionnaires. Two questionnaires that were used as the primary resources
were the one pilot-tested 1n 1981 for a National Household Pesticide Usage
Survey (Berman, 1981) and the one used for the Agency's Nonoccupational
Pesticide Exposure Survey (NOPES) from 1986 through 1988 (Immerman and
Schaum, 1990). At the end of the NHGPUS questionnaire, a Bureau of the
Census question was used to record the type of structure for each sample
housing unit. The "new standard environmental inventory questionnaire" was
also reviewed for input to the NHGPUS questionnaire (Lebowltz et al.,
1989).
13
-------
During the first few months following the project klckoff meeting, the
draft analysis plans and questionnaire Items were Iteratlvely reviewed and
revised through frequent consultation between RTI and the Agency. A first
draft questionnaire and Information Collection Request (ICR) supporting
statement were ready to be circulated for peer review 1n July 1989. This
package was extensively reviewed by Agency staff 1n several divisions of
the Office of Pesticide Programs, by the Office of Research and
Development, and by the Statistical Policy Branch of the Office of Policy,
Planning and Evaluation. Additional review was also provided by the U.S.
Department of Agriculture's Economic Research Service. The reviewers
provided extensive comments. All their comments were discussed 1n detail
to determine the revisions that should be Implemented, consistent with the
study objectives and budget.
Extensive revisions based on the reviewer comments resulted 1n a second
draft questionnaire that was ready for pilot testing and further peer
review 1n September 1989. This questionnaire was sent for review to the
National Pest Control Association, the Professional Lawn Care Association
of America, the Chemical Manufacturers Association, and the Chemical
Specialty Manufacturers Association. Their comments were mostly positive.
This questionnaire was also pilot tested 1n October 1989 1n nine homes
In the Raleigh/Durham area of North Carolina. Because we expected that
different types of residences might reveal different difficulties with the
Instrument, the pilot test purposely Included a variety of dwelling types:
two farms, two mobile homes, one apartment, and four single-family detached
dwellings. We also purposely Included some homes with pets. The pilot
test revealed that major revisions were needed. One major problem was that
the questionnaire was awkward for the interviewer; 1t required too much
paper shuffling. Another problem was that many questions regarding the
pesticide products on-hand were not applicable for pesticides that either
were 1n continuous use or had not been used 1n the past year.
The questionnaire used In the pilot test began the Inventory of
pesticide products on hand with a series of yes/no questions regarding the
presence of pesticides to treat each pest 1n a rather long 11st, Intended
to be all-1nclus1ve. The pilot test revealed that these questions were a
burden to the respondents and would not guarantee that all pesticides would
14
-------
be listed. Therefore, the final questionnaire begins the Inventory of
pesticides on-hand after a brief definition of what the EPA considers to be
a pesticide. At the end of the Inventory section, a few questions are
asked to ensure that pesticides that are not normally thought of as such
(e.g., repellents) are included 1n the Inventory.
The pilot-tested questionnaire contained several questions about pest
problems that occurred 1n the past year, Irrespective of whether or not
pesticides for treating those products were currently on hand. These
questions were a burden to respondents because of perceived duplication
with the questions asked for all products currently on hand. Therefore,
this section was greatly shortened to ask only about the pest problems
treated 1n the past year by pesticides that are no longer on hand and about
the severity of the pest problems experienced 1n the past year.
This revised questionnaire was then pilot tested in five additional
homes 1n the Raleigh/Durham area 1n November 1989. The types of residences
selected for this second pilot test were two farms and three single-family
detached residences. We purposely interviewed a few households with a
relatively large number of pesticide products 1n this second pilot test.
All these Interviews proceeded smoothly.
Of course, a 1-year recall 1s too long for respondents to accurately
remember all pest problems, all applications of pesticides, etc., but
funding limitations resulted 1n a one-time survey with a 1-year recall
period Instead of a longitudinal study with multiple interviews throughout
the year. Because most pest problems and pesticide use occur in the Spring
and Summer, field data collection was scheduled for August and September.
Thus, many of the events of interest 1n the 1-year recall period actually
occurred 1n the months Immediately preceding the interview.
The questionnaire and Information Collection Request (ICR) supporting
statement were submitted to the Agency's Office of Policy, Planning and
Evaluation for subsequent submittal to the Office of Management and Budget
(0MB) 1n December 1989. 0MB provided RTI with comments from the Chemical
Specialties Manufacturers Association. 0MB approved the ICR 1n May 1990
conditional on deletion of one question and revision of two others. The
final questionnaire 1s shown 1n Appendix A of Volume I.
15
-------
5.2 Notebook of Pest Sketches
The first pilot test of the NHGPUS questionnaire revealed that the
general public could not easily classify their pest problems Into the pest
categories used 1n the questionnaire. Therefore, a notebook of pest
sketches was developed for use with the questionnaire. The notebook was
organized by the pest categories used 1n the questionnaire. For most pest
categories it Included several examples of pests belonging to the category.
For most pests, it provided a black-and-white sketch of the pest, a short
description of the pest and the damage 1t causes, and the size of the pest.
The book was not Intended to be a definitive Identification tool. It was
provided simply to help participants understand the pest categories being
used and to determine the proper pest category when they only knew what
their pest looked like.
RTI and the Agency collaborated closely on development of the pest
notebook. RTI prepared an Initial draft using materials provided by the
Agency and other materials obtained from local libraries. The RTI draft
contained only those pests expected to be difficult for the general public
to Identify. The Agency subsequently decided that the notebook should
cover all pest categories contained 1n the questionnaire. They then
prepared a much more extensive notebook using sketch materials and pest
experts available at the Agency. The draft developed by the Agency was
electronically transferred to RTI for final editing and printing late 1n
July 1989. The final pest notebook Is shown 1n Appendix I.
5.3 Lead Letter and Study Brochure
Achieving the highest possible response rate 1s Important for all
survey research because the only sure protection against nonresponse bias
1s to achieve a high response rate. In order to enhance the NHGPUS
response rate, we attempted to mall a lead letter and a brochure explaining
the study to every sample housing unit. Survey research has shown that
such lead mailings usually Increase the survey response rate (see Groves,
1989, Section 5.2.1).
Mailing labels were generated from the address lists prepared for the
third stage of sampling (see Section 4.3). A few sample housing units did
not have mailable addresses.
The lead letter (see Appendix J) provided legitimacy for the study by
Introducing the sponsoring agency and the purpose of the study. Other
16
-------
Important features of the lead letter are that 1t:
let the household know that an Interviewer would be coming to the
home,
presented the survey burden, (about a 45 minute Interview) and the
survey Incentive ($5 and a copy of the EPA brochure "Citizen's
Guide to Pesticide Use"), and
ensured the sample subject that participation was voluntary and
that all responses would be kept confidential.
Because the mailings could not be personalized (only addresses were
available, not names), we prepared special envelopes designed to
distinguish our mall from "junk" mall. The lead mailing was sent 1n an
envelope with prepaid postage and a combined RTI/EPA return address (see
Appendix J).
The study brochure was prepared 1n a question and answer format and
printed 1n blue (see Appendix J). It explained the study and the agencies
Involved (both EPA and RTI) 1n more detail. It also provided names and
telephone numbers of people who could be contacted to answer any other
questions regarding the study. It also provided the toll-free telephone
number of the EPA's National Pesticide Telecommunications Network to be
contacted 1n case of pesticide poisoning, as a public service.
5.4 Field Interviewer's Manuals
Interviewers hired for counting and listing activities were provided
with a copy of the RTI Counting and Listing General Manual. This manual
describes how to conduct the task of counting and listing the housing units
1n the area segments selected at the second stage of sampling. Specific
topics Include: applicable definitions such as area, segment, household,
living quarters, and group quarters; descriptions of the field sampling
materials, Including the segment sketches, maps, and lists of housing
units; step-by-step procedures for counting the housing units within the
segments; specifications for when to proceed with the listing and when to
call RTI's sampling staff; step-by-step procedures for listing the
segments; procedures for documenting unusual circumstances and referring
them to the appropriate Individual for resolution; reporting procedures;
quality control procedures; and disposition of completed work.
RTI developed the NHGPUS Field Interviewer's Project Manual to direct
and guide the field staff in collecting data from participants 1n the
17
-------
study. Topics covered 1n the manual Included: background of the survey;
overview of the assignment; confidentiality of data collection; locating
housing units; checking for missed housing units; contacting and screening
sample households; explaining the. survey and obtaining cooperation;
Information on pests, pesticides, and safety precautions; questionnaire
administration; quality control procedures; and general administrative
procedures. Appendices to the document Included examples of product
labels, study showcards, and a glossary of terms commonly used 1n the
manual.
In addition to the manual prepared especially for the survey, a copy of
the RTI Field Interviewer's General Manual was given to each interviewer.
The manual provides basic Information applicable to Interviewer fleldwork
for all RTI surveys and eliminated the need to provide Intensive coverage
of these topics 1n the project's field manual.
5.5 Training Materials
Development of training materials was a critical aspect of the study
since multiple training sessions were being conducted simultaneously by
different trainers. The training materials were designed to Insure that
the training of field staff, and ultimately, the collection of NHGPUS data,
were conducted 1n a uniform and standard manner.
The Training Guide for Counting and Listing and the NHGPUS Training
Guide were used by our trainers to prepare their training classes. These
guides include: detailed Instructions on presenting key training
components through the use of verbatim lectures on special topics;
Instructions for using special training aids; and procedures for conducting
demonstration mock survey interviews, classroom exercises, and written
tests.
Some of the training components covered by the Training Guide for
Counting and Listing are:
welcoming the trainees and Introducing the trainers and trainees;
explaining the purposes, design, and terminology of the survey;
reviewing responsibilities of Field Interviewers and Field
Supervisors;
reviewing, discussing, and conducting question and answer sessions
on locating segments;
18
-------
reviewing, discussing, and conducting classroom exercises on
counting and listing procedures;
reviewing administrative duties.
Some additional training components covered by the NHGPUS Training
Guide were:
reviewing the agenda and rules for the training session;
demonstrating how to make an Initial contact, conduct screening,
and perform data collection, including discussion of the
demonstration;
reviewing, discussing, and conducting exercises on locating,
contacting, and screening sample housing units and looking for
missed housing units;
discussing terminology related to pests, pesticides, and safety
precautions;
reviewing, discussing, and conducting round-robin practice, and
paired practice of techniques for completing the questionnaire and
paying the incentive; and
reviewing quality control and administrative procedures.
The guide also Included examples and exercises on identifying pesticide
containers and child resistant packaging.
A pretralnlng Home Study Exercise was developed to reinforce key points
made In the interviewer's manual. The exercise was to be completed after
reviewing the manual and prior to attending the training session. The
exercise was reviewed during training and Interviewers received individual
feedback on problem areas.
Finally, a video tape was developed that showed various product
packaging. We were concerned that Interviewers would have trouble
recording the appropriate container description and Identifying child
resistant packaging (Questions 24 and 25 of the study questionnaire).
While this was addressed through examples shown at training, the video tape
was prepared and distributed to each field Interviewer at training to be
used as a reference 1f they had problems classifying a container during
data collection.
19
-------
THIS PAGE LEFT BLANK INTENTIONALLY
20
-------
6. FIELD OPERATIONS
Field operations for the NHGPUS were comprised of two distinct
processes. The first process was counting and listing all potential
housing units In the 300 sample areas, called area segments, selected at
the second stage of sampling. These counting and listing activities are
discussed In the first section of this chapter. The second field process
was household Interviewing or field data collection, which Is discussed In
the second section of the chapter.
6.1 Counting and Listing Activities
Counting and listing activities took place from April through June
1990. The following sections describe the activities associated with
counting and listing.
6.1.1 Recruiting Field Supervisors
The field supervisors recruited for the study were used for both
counting and listing and for data collection. As soon as the counties were
selected for the study, their geographic distribution was studied and the
number of field supervisors to be recruited was determined. The study area
was divided Into four regions, and a field supervisor (FS) was recruited
for each region. FSs were recruited from RTI's active 11st of over 60
experienced FSs. FSs were selected based on their performance on previous
studies, experience 1n counting and listing, availability during the data
collection period, and geographic location.
6.1.2 Recruiting Field Interviewers
The field supervisors (FSs) were responsible for recruiting field
Interviewers (FIs) 1n each of their assigned counties. To assist the FSs
1n the recruiting activity, RTI's National Interviewer File was searched
for persons 1n the sample counties who had previous survey data collection
experience with RTI. A 11st of names for each county was generated
containing the last known address, telephone number, and performance rating
for each listed person. When the National Interviewer File listed no names
for a county, the FS was required to check with other research companies,
other RTI FSs who may have had previous experience or contacts in the area,
and local employment agencies to identify and recruit FIs for that county.
If this failed, the FS ran advertisements in the local newspapers to
Identify potential interviewers.
21
-------
Our primary goal was to hire FIs who had good to excellent ratings and
who lived 1n the county where they would be working. In counties where an
experienced FI was not available, the FS tried to recruit an experienced FI
who lived within one hour's driving time of the area. The guiding
principle was that the additional cost of having an experienced FI drive
into the study area was more than offset by the quality and expected
efficiency of their work. A total of 47 field interviewers were hired to
conduct the counting and listing.
6.1.3 Training
Four field supervisors were trained on April 23 and 24, 1990. Most of
their field staff needed for counting and listing had been hired prior to
that time. Prior to training, the FSs received a copy of the Counting and
Listing Manual for review.
The training methodology included instructor demonstration, group
discussion and Interaction, and classroom exercises. The Training Guide
for Counting and Listing described 1n Section 5.5 was used. A copy of the
training agenda 1s shown 1n Exhibit 6.1. To standardize the training of
field Interviewers, each FS was provided with a copy of the training guide
that outlined all the material they needed to cover when training their
field Interviewers.
In the two to three weeks following the training session, the FSs
trained their staff Individually and assigned the segments to be counted
and listed to them. Most Interviewers with previous counting and listing
experience were trained 1n telephone conferences with their FS. The FSs
traveled to the areas being worked by Inexperienced interviewers to train
each one 1n person.
6.1.4 Counting and Listing
Counting and listing Is the process of enumerating the housing units in
a well-defined geographic area selected through scientific sampling
procedures. The product of counting and listing In each defined area
(called a segment) 1s a complete 11st describing all housing units located
within the segment boundaries. Counting and listing was completed for 300
segments 1n 58 counties for this study. A total of 47 field interviewers
were used to conduct the counting and listing. The field supervisors did
the counting and listing In some areas themselves. Counting and listing
began 1n May 1990 and was completed by mid-June 1990.
22
-------
Exhibit 6.1
TRAINING AGENDA FOR COUNTING AND LISTING
National Home and Garden Pesticide Use Survey
WELCOME AND INTRODUCTIONS
OVERVIEW OF SURVEY
Sponsored by EPA
Purpose and Background
Data Collection Activities
Schedule
FIELD INTERVIEWER AND SUPERVISOR RESPONSIBILITIES
Field Interviewer
Field Supervisor
TERMINOLOGY
NHGPUS
Segment
Subsegment
Field Counting
Listing
COUNTING AND LISTING
Overview
Materials
Locating Segments
Counting Procedures
Subsegmentlng
Listing Procedures
Changes to the Counting and Listing Manual
ADMINISTRATIVE DUTIES
Disposition of Completed Materials
Segment Checklist
Field Reporting and Expense Reporting
Log of Counting and Listing Hours
PT&E Reporting
23
-------
FSs assigned FIs to work specific segments. Usually one FI was
assigned to work all the segments 1n a set of nearby counties. FIs were
requested to begin counting and listing Immediately after receiving their
assignments. FIs were required to. send their first completed segment to
their FS to be checked. If the FS determined that the FI's work was
satisfactory, the FI was allowed to send the remainder of his/her completed
work directly to RTI.
The FIs were instructed to work the segments with the largest number of
estimated housing units first. (The estimated housing unit count was
provided for each segment from 1980 Census data.) These segments were the
most likely candidates for a subsegmentlng procedure designed to Improve
the efficiency of the counting and listing process. Segments In rural
areas with large numbers of housing units (200 or more) are generally very
time-consuming to 11st. The subsegmentlng process divides the original
segment area Into smaller subareas called subsegments using housing unit
count Information provided by the FI. Thus, for segments containing 200 or
more actual housing units that could not be listed 1n a reasonable amount
of time (as determined by the FI), the FI recorded housing unit counts
along each street on the segment map and returned these segments with the
housing unit counts to RTI for subsegmentlng. Because this procedure
delayed the listing of large segments until they had been sent to RTI,
subsegmented, and returned to the FI, the FIs were Instructed to work these
larger segments first.
Segments received at RTI to be subsegmented were first divided Into
subsegments that generally contained about 30 to 50 housing units (HUs),
using actual surface features to form new boundaries. A sampling worksheet
provided guidance for appropriate subsegment sizes and was completed to
determine which subsegment was selected to be listed. A segment kit was
then assembled for the selected subsegment and sent back to the field
Interviewer to be listed.
When a counted and listed segment was received at RTI, 1t was logged In
and edited. An edit checklist was completed for all segments. For those
found to be 1n error, a copy of the checklist was sent to the FI, their FS,
and to the RTI data collection supervisor. Minor errors were corrected 1n-
house and were discussed with the FI. Any segment with errors that could
not be corrected in-house was sent back to the FI for correction.
24
-------
6.2 Primary Data Collection Activities
Data collection activities took place from mid-August through the first
week of October 1990. The following sections describe the activities
associated with data collection.
6.2.1 Recruiting Field Supervisors
The four field supervisors used for counting and listing were available
for the primary data collection, therefore, no additional effort was
required to recruit field supervisors.
6.2.2 Recruiting Field Interviewers
Most of the field interviewers (FIs) used for counting and listing were
also available for primary data collection. Our goal was to hire one FI
for each of the 60 PSUs. The same procedures used In recruiting FIs for
counting and listing were followed 1n recruiting FIs for primary data
collection. We recruited five bilingual Interviewers to help reduce the
number of interviews lost to language barriers. Table 6.1 provides a
breakdown of the demographic characteristics of the 60 field interviewers
used for data collection.
6.2.3 Training
Separate 2-day Field Interviewer Training sessions were conducted by
RTI staff in Raleigh, NC, on August 7-8, 1990 and 1n Dallas, TX, on August
16-17, 1990. Approximately half of the FSs and their FI staff attended the
Raleigh session and the other half attended the Dallas session. Prior to
training, all field staff received a copy of the NHGPUS Field Interviewer's
Project Manual, the RTI General Field Interviewers Manual, home-study
exercises, and a copy of the EPA Citizen's Guide to Pesticides. Each
trainee was required to review these materials prior to training and to
complete the set of home-study exercises.
The day before Field Interviewer Training (August 6 and August 16,
respectively), field supervisors were trained 1n the use of the NHGPUS
Training Guide and on FS responsibilities.
Additionally, a 1-day training session was held on the day prior to the
NHGPUS FI training sessions for Interviewers with less than one year of
experience or who had never worked for RTI. This session was designed to
train interviewers 1n basic procedures and techniques employed by RTI 1n
conducting surveys, and to Instruct them on RTI administrative procedures
and forms.
25
-------
Table 6.1
CHARACTERISTICS OF FIELD INTERVIEWERS
National Home and Garden Pesticide Use Study
RACE & SEX
INTERVIEWER AGE
TOTAL
18-34
35-44
45-54
55+
White Females
Black Females
TOTAL FEMALES
6
0
6
11
1
12
6
2
8
24
2
26
47
5
52
White Males
Black Males
TOTAL MALES
1
0
1
3
0
3
3
0
3
1
0
1
8
0
8
TOTAL
7
15
11
27
60
26
-------
The training methodology Included Instructor demonstration, group
discussion and Interaction, video demonstration, visual aids, round robin
mock Interviews, paired practice Interviews, and classroom exercises. The
NHGPUS Training Guide described 1n Section 5.5 was used by each trainer to
help ensure that training was conducted 1n a uniform manner. Additionally,
the RTI General Field Interviewing Training Guide was used at the 1-day
session on basic Interviewing, again to standardize the training of field
Interviewers.
The training agenda for Field Supervisor Training, General Field
Interviewing Training, and NHGPUS Field Interviewer Training are shown 1n
Exhibits 6.2, 6.3, and 6.4 respectively.
The Field Supervisor Training and the General Field Interviewing
Training were conducted by RTI survey specialists. The NHGPUS Field
Interviewer Training was conducted by the Field Supervisors and assisted by
RTI project staff. Generally, each FS trained his or her own staff of FIs.
The training sessions were conducted by the FSs to better establish their
authority with their staff, to better determine their FIs' weaknesses and
strengths, and to give each FS a better understanding of the project
materials. An RTI project staff member was always present to answer
questions and clarify material for the FSs. The RTI project director
participated in the training by presenting an overview of the NHGPUS sample
and a demonstration of examples of product packaging, emphasizing child
resistant packaging. A representative from EPA observed each of the
training sessions and presented background information on the NHGPUS.
6.2.4 Data Collection Activities
Field Interviewers were instructed to begin work on their assignments
immediately following training. There were 298 segments assigned to the
FIs (2 of the original 300 segments were found to have no housing units at
the time of counting and listing). Of the 298 segments, 116 were
designated as early report segments. Information on the progress of data
collection in these segments allowed project staff to determine that the
target response rates would be achieved without releasing a supplemental
sample of housing units.
While the FIs attended training, a lead letter and a glossy study
brochure (see Appendix J) were mailed from RTI to all cases in the early
27
-------
Exhibit 6.2
FIELD SUPERVISOR TRAINING AGENDA
National Home and Garden Pesticide Use Study
OVERVIEW
a. Review background of NHGPUS
b. Review project schedule and production goals
REVIEW FS RESPONSIBILITIES
a. FI training
Schedule/agenda/format
FS role - Lead Trainer
Home study
b. Weekly conference with FIs
Set up Reporting Schedule
Take reports using ACF and FSSR
Monitor Early Report Segments Closely
Discuss problem cases
Discuss editing problems
Discuss verification problems
Discuss work plans for upcoming week
Discuss concerns from latest PT&E charge
c. Weekly conference with FM
Must call at appointed time
Critical to monitoring field progress
d. Editing completed work
General edits
Completeness of package
e. Resolving Problems
Followup action on pending cases
Use of Law Enforcement Letter
Controlled Access Letter
28
-------
Exhibit 6.2 (continued)
ADMINISTRATIVE PROCEDURES
a. Reviewing PT&Es
Review points made 1n manual
Participant Incentive Receipts Attached
Use of log - must send regularly
Send weekly via Fed Ex to RTI
b. Authorization Forms
c. FS Travel
d. Sending documents to RTI
Review groupings
Send weekly
e. Quality Control
Edit reports from RTI
Verification reports from RTI
REVIEW OF TRAINING MATERIALS
Role of FS - lead trainer
Agenda
Guide - note corrections
Training materials
Hand out transparencies
Locating and contacting
Handling Pesticides
CRP examples
Q x Qs
Practice conducting practices
Administrative procedures
29
-------
Exhibit 6.3
GENERAL FIELD INTERVIEWING TRAINING AGENDA
National Home and Garden Pesticide Use Survey
Welcome and Introduction
Video
Overview of Survey Research Operations
What 1s a Sample Survey?
Examples of Sample Surveys
FI Procedures and Responsibilities
Professional Ethics and Respondents's Rights
BREAK
What Does a Field Interviewer Do?
Video
Techniques
Greeting and Introduction Examples
Obtaining Cooperation and Overcoming
Objections
Role Play Exercises
Refusals
Conducting an Interview
Editing Requirements
Interviewer Efficiency and Performance
LUNCH
Questionnaire Administration
Basic Interviewing Skills
Bias Exercise (Written)
Trust the Instrument
Focusing the Respondent
Probing
Verbal and Written Probing Exercises
Recording Responses (Examples)
BREAK
Administrative Procedures
Employment with Powerforce
Supplies
PT&E
PT&E Exercises
Payment
Advances
Practice Round Robin Mock Interview
Summary Exercise
30
-------
Exhibit 6.4
TRAINING AGENDA - August 1990
National Home and Garden Pesticide Use Survey
Day 1
BREAKFAST
Registration
Welcome and Introductions
The NHGPUS
Overview of the NHGPUS Sample
Training Session Protocol
Confidentiality and Data Collection Agreement
BREAK
Demonstration Interview
Locating Sample Housing Units
LUNCH
Contacting and Screening Sample Households
Contacting and Screening Practice Exercise
Pests and Pesticides
BREAK
Child Resistant Packaging Examples and Exercises
Take Pictures for ID Badges and Distribute Assignments
Day 2
BREAKFAST
Announcements, Questions, and Answers
Review Home Study Exercise
Questionnaire Administration
BREAK
Quest1on-by-Quest1on Review
Round Robin Practice Interview
LUNCH
Paired Practice Interviews
Quality Control Procedures
BREAK
Administrative Procedures
Wrap-Up & Distribute ID Badge, Paycheck, & Incentive Advance
31
-------
report segments. These Items explained the study, asked for cooperation,
Informed of pending contact by an RTI Interviewer, and explained that an
Incentive of $5.00 and a copy of the EPA brochure "Citizen's Guide to
Pesticide Use" would be given for participation In the study.
Methodological studies have shown that such small Incentives can produce 5
to 10 percent Improvements 1n response rates (Groves, 1989, Section 5.2.3).
For the remaining cases, FIs were provided with a supply of
preaddressed envelopes, containing the lead letter and glossy study
brochure, and Instructed to mall the materials to sample housing units
approximately 3-5 days prior to working the segment. This methodology
increases the likelihood that the household member will remember receiving
the letter when the Interviewer arrives.
During the initial contact at the sampled housing unit, a lead letter
and glossy study brochure were given to the respondent 1f they did not
remember receiving one. The Control Form (Appendix A of Volume I) was used
to identify the case, to document Its status, and to screen the household
for eligibility. To be eligible for the survey, the housing unit had to be
a permanent residence, not a vacation home, and not a group quarters.
When screening resulted 1n an eligible household, the Interviewer
Immediately attempted to administer the questionnaire. The household
members were rostered and the person(s) most knowledgeable about the
pesticides and cleaning products used at the residence were Identified. If
more than one knowledgeable person was Identified the Interviewer tried to
schedule the Interview when all the knowledgeable Individuals were
available.
The Interviewer then continued to administer the questionnaire, and
asked to see all the pesticide products used in and around the home. At
the conclusion of the Interview, the FI completed an observation section
Identifying the principal respondent, describing the structure and location
of the residence and Indicating 1f continuation pages were Included with
the questionnaire. The FI then paid the $5.00 Incentive, had the
respondent sign an incentive receipt, left a copy of the EPA "Citizen's
Guide to Pesticides," completed the record of calls section of the Control
Form, informed the respondent that an RTI staff member might call them to
verify that the Interview had been conducted, and thanked the respondent
for their participation. Upon returning home, the interviewers edited the
32
-------
study Instruments and sent them to their FS or to RTI. A minimum of the
first five cases and the first two completed Interviews were sent to the FS
for review. If the FS determined that the FI's work was satisfactory, the
FI was allowed to send the remainder of his/her completed work directly to
RTI.
If the household member refused to participate 1n the Interview, the FI
attempted to overcome the objection and documented the results so a FS
could evaluate the situation and recommend follow-up action. In most cases
a refusal conversion letter (see Exhibit 6.5) was sent to the person
requesting that they reconsider and informing them that they would be
contacted again 1n the near future.
During the course of the data collection, neighbors were used as a
source of Information for sample housing units (SHUs) where the FI could
not find anyone at home. The FI asked neighbors 1f they knew 1f the house
was occupied and the best time to find the residents at home or to
determine 1f the residents were away for an extended period. The FI was
then required to enter the name and telephone number of the neighbor who
provided this Information 1n Part D of the Control Form.
Data collection began on August 9 and was completed on October 7, 1990.
Table 6.2 presents the distribution of final Interview results. Nine
Interviewers had no refusals. Seven Interviewers achieved a completed
Interview for over 90 percent of their cases. An additional 22
interviewers achieved a completed Interview for over 80 percent of their
cases.
6.2.5 Validation Interviews
Ten percent of the SHUs 1n each segment were selected for validation.
During each screening Interview, the FI requested the respondent's
telephone number and Informed the respondent that they could be contacted
later to verify the FI's work. Of the 267 cases selected for validation,
210 resulted 1n a completed Interview and were eligible for validation.
There were a few special non-1nterv1ew cases that were validated as well,
but generally, no attempt was made to validate non-Interview cases.
RTI's Telephone Survey Unit conducted the validations. Their many
years of validation interview experience made them well suited for this
task. The validation telephone interviewers (TIs) were given the Control
Form and questionnaire. The TI abstracted the housing unit ID number, FI
33
-------
Exhibit 6.5
REFUSAL CONVERSION LETTER
UNITED STATES ENVIRONMENTAL PROTECTION AGENCY
WASHINGTON, O.C. 20460
OFFICE OF
Dear Respondent:
Recently one of our workers, , came to your home and asked you to
take part in the National Home and Garden Pesticide Use Survey. At that time,
you were unwilling to participate in the survey. Please allow me to give you
some background on the survey.
The survey provides the Federal Government with important information on
the public's exposure to pesticides and on what pests are problems. The
Environmental Protection Agency is mandated by congress (P.L. 92-516, as amended
by P.L. 94-140, as amended by P.L. 95-396) to register pesticides used in the
United States on the basis of a scientific evaluation of both the risks and the
benefits that would result from the use of the product. Information on products
used in and around the home is needed to support this congressional mandate.
Respondent's answers and also the areas that are selected to be surveyed are
kept strictly confidential, and data are presented only as summary statistics
for the U.S.
We have selected a limited number of households and the participation of
each and every person is important. Of course, participation by selected
respondents is completely voluntary. However, high participation rates are
necessary for valid survey results.
We have enclosed a question and answer brochure with more information about
the survey. Please reconsider. We will be contacting you again in the near
future. Your cooperation would be greatly appreciated. If you have any
questions or would like to discuss the survey further, please feel free to call
me at (703) 308-8050.
Sincerely yours,
Edward Brandt
Project Officer
National Home & Garden Pesticide Use Survey
34
-------
Table 6.2
Distribution of Final Interview Results
Final Result Number Percent
Eligible sample housing unit
2,447
90.87
Completed Interview
2,078
77.16
Refusal
209
7.76
Breakoff
3
0.11
No eligible respondent home
88
3.27
Language barrier
10
0.37
Physically/mentally Incompetent
9
0.33
Other eligible
50
1.86
Ineligible sample housing unit
246
9.13
Vacant housing unit
167
6.20
Not a housing unit
42
1.56
Vacation/second home
33
1.23
Other Ineligible
4
0.15
TOTAL
2,693
100.00
35
-------
ID number, and Interview date from the Control Form and entered 1t on the
FI Validation Form (see Exhibit 6.6). They then looked at the
questionnaire to determine 1f any pesticide products were reported and
recorded either "yes" or "no" as appropriate on the validation form. They
also abstracted the response to Question 5a of the questionnaire ("During
the past year, did your household raise any crops or livestock for sale?")
and entered 1t on the validation form.
For cases where phone numbers were available, the TI attempted the
validation call. The first step was to determine 1f the respondent
remembered being Interviewed and If they had been asked about their use of
pesticide products. The TI then validated two Items from the questionnaire
and verified that the field Interviewer had looked at the cleaning products
to see 1f any qualified as pesticides. The TI then asked 1f the field
Interviewer had been courteous and thanked the respondent. During the
validation Interview, 1f a respondent gave Information that was different
from the questionnaire data, the Interviewer was trained to probe the
respondent to clarify the discrepancy. This often resulted 1n a resolution
of the problem.
In cases where an SHU selected for the validation did not have a
telephone number or the wrong number was recorded on the Control Form, a
validation letter (see Exhibit 6.7) was mailed to the address. The
validation package Included a postage paid envelope for return of the
letter to RTI. Cases where the letter was not returned were not validated.
The results of the Interview validation are as follows. Of the 210
completed Interviews Included In the validation sample, 30 (14 percent)
either did not have a telephone, refused to provide a telephone number,
gave us the wrong telephone number, or could not be contacted to validate
the case after several attempts. We were able to successfully validate 176
(98 percent) of the remaining 180 cases, 162 by telephone and 14 by mall.
The four cases 1n which the validation questions were not confirmed were
scattered across four different Interviewers and usually had a plausible
reason for the discrepancy; most were elderly respondents who did not
remember the study.
6.2.6 RTI Protection of Human Subjects Committee Review
Every project conducted by RTI requiring data collection from people
must receive approval of the RTI Committee for the Protection of Human
36
-------
Exhibit 6.6
FIELD INTERVIEW VALIDATION FORM
National Home and Garden Pesticide Use Survey
Verified By:
A: ASSIGNMENT INFORMATION
1. HU ID Number:
2. FI ID Number:
Date Verified:
/ /
Interview Date:
- 90
3. Look at the top of page 4 of the questionnaire. If the circled number at the
top of the page 1s more than ZERO, enter YES on the line provided 1n Item 4
below. If the circled number 1s ZERO, enter NO on the line provided 1n Item 4
below.
4. Record the answer to Question 5a on page 1 of the questionnaire on the line
provided 1n Item 6 below (YES, NO, or DK).
B: VALIDATION
1. Name of Person Contacted (required only 1f
you have a problem flagged?)
READ THE FOLLOWING INTRODUCTION WHEN YOU CONTACT ELIGIBLE RESPONDENT TO CONDUCT
VALIDATION.
"Hello, my name 1s from the Research Triangle
Institute In North Carolina. I am calling to verify that one of our staff
recently contacted you about a survey we are conducting for the
Environmental Protection Agency."
YES
2. Do you remember someone from Research Triangle
Institute coming to your home? 01
3. Did the person ask you questions about your use of
pesticide products? 01
4. Did you have any stored pesticide products for home
use at the time the Interview was conducted? 01
5. Did the Interviewer look at your cleaning products
to see 1f any of them qualified as pesticides? 01
6. During the past year, did your household raise
any crops or livestock for sale? (Q5a) 01
7. Was the Interviewer courteous? 01
8. THANK THE RESPONDENT.
9. FLAG/PULL PROBLEMS: B4 response not equal to QUEX response on line.
B6 response not equal to QUEX response on line for Q5a.
B7 = no.
NO
02>PR0BE
OR STOP!
02
02 QUEX
02
02 QUEX
02
37
-------
Exhibit 6.7
RESEARCH
VALIDATION LETTER
National Home and Garden Pesticide Use Study
TRIANGLE INSTITUTE
K-- "
" i
/ L V I L
*r~ , * I VV
HU ID #
-m
FI ID #
October 19, 1990
Dear Respondent,
During the past month, the Research Triangle Institute has been
conducting a nationwide survey on the use of pesticides sponsored by the
United States Environmental Protection Agency. Our records Indicate that you
were Interviewed. We would appreciate 1t 1f you would take a moment to
complete the questions listed below and return them In the enclosed pre-
addressed postage paid envelope. This Information helps us to verify our
records, and the quality of our Interviewer's performance.
Please answer the following questions by circling the appropriate
response.
1. Do you remember someone coming to your home
collecting Information on the use of pesticides
In your home7 Yes No
2. Did you have any stored pesticide products
for home use at the time the Interview was
conducted? Yes No
3. Did the Interviewer look at your cleaning
products to see if any of them qualified as
pesticides? Yes No
4. During the past year, did your household raise
any crops or livestock for sale? Yes No
5. Was the Interviewer courteous? Yes No
6. Thank you for your participation 1n our survey.
We look forward to your response.
If No,
Go to
Question 6.
Sincerely,
Janice Kelly
Post Office Box 12194 Research Triangle Park, North Carolina 27709-2194 Telephone 919-541-6000
38
-------
Subjects before any data collection can begin. This committee reviews the
entire study protocol, all data collection Instruments and forms, and all
data collection procedures. In June 1990, we presented final drafts of the
lead letter, informational brochure, questionnaire, control form, and
showcards to the committee along with the complete study protocol. The
committee expressed greatest concern over interviewers not being trained to
detect potentially dangerous use or storage of pesticides. This concern
was overcome by pointing out that each respondent would receive a copy of
the EPA "Citizen's Guide to Pesticides," which covers alternatives to
chemical pesticides, tips for handling pesticides, correct storage and
disposal of pesticides, reducing exposure to pesticides, what to do 1n a
pesticide emergency, and other topics. On July 19, 1990, the committee
approved Implementation of the study.
39
-------
THIS PAGE LEFT BLANK INTENTIONALLY
40
-------
7. DATA PROCESSING
All data collection Instruments used 1n this study that contained
respondent data were submitted to pianual editing and data entry. These
processes are discussed 1n the first two sections of this chapter.
Additional computerized data editing was also performed for the two primary
data collection Instruments (the questionnaire and Card A) as discussed 1n
Section 7.3. The final section of this chapter discusses a computerized
survey control system, which was used to ensure that all data processing
steps were executed, and executed 1n the proper order, for every data
collection Instrument for every household.
7.1 Manual Editing
Every data collection Instrument that contained respondent-provided
data was manually edited after 1t was received at RTI. Six editors and one
supervisor were trained and began manual editing on September 5, 1990.
This manual edit was conducted using edit specifications developed for each
Instrument. General edits Included a check of the legibility of entries, a
check of the housing unit Identification Information to ensure Its
consistency for all associated survey materials, a check for completeness
of designated key Items 1n the Instrument, and a check for proper use of
skip patterns 1n the Instrument. The number of pesticide products was
counted and verified, and Table B continuation sheets were stapled to the
questionnaire to facilitate data entry. Any Items that failed to meet the
edit criteria were documented. Problems that could not be resolved by
project staff using other data sources from the project were resolved via a
followup telephone call from RTI project staff to the data collector
responsible for the completed Instrument.
The Instruments that were manually edited at RTI were the Control Form,
Card A, and the questionnaire. Data collectors were also Instructed to
edit all these Instruments before mailing them to RTI. Interviewer editing
and Field Supervisor review are discussed 1n Section 6.2.4.
7.2 Data Entry
Data entry programs were developed for each of the three separate
documents: the Control Form, the household questionnaire, and Card A. The
product Information 1n the household questionnaire was developed as a
41
-------
"repeating screen" for data entry, which resulted 1n variable length data
files, depending on the number of products reported.
Data entry programs were developed 1n Easy Entry, which allowed
Imbedded quality-control checks 1n the data-entry process. The data-entry
programs performed the following quality-control checks:
checked all Items for permissible ranges and codes
verified all digits on the critical IDs to ensure consistent entry
verified that any Items defined as critical were not blank.
In order to produce data of high quality, all data were keyed twice.
Any discrepancies between the two keylngs were detected by the software and
corrected by the second data entry operator.
Because the data-entry files are organized 1n the same format as the
data-collection Instruments to simplify data entry, the files were
reformatted and combined to produce raw analysis files. These raw files
were then read Into SAS data sets. During this SAS Input step, product-
level data 1n the repeating screen was removed from the questionnaire data
file and placed 1n a separate file along with appropriate Identifying data.
7.3 Computerized Editing
The NHGPUS data were analyzed for automated editing as four separate
files: 1) Control Form Items, 2) household-level questionnaire items
(Questions 1-16, 34, and 40-54), 3) Card A data (Questions 35-39), and 4)
product-level Items (Questions 17-33 for each pesticide product found 1n
storage at the home). Because the Control Form was not a primary data
collection Instrument, no computerized editing was necessary for that form.
The data bases for all other forms were subjected to extensive computerized
editing.
The primary purposes of computerized data editing were: 1) to find and
correct Inconsistencies and errors, and 2) to replace missing data with
"consistency codes" to explain why the data were missing. The computerized
editing process was begun by generating unweighted frequency tabulations of
the data Items. For most variables, multi-way tabulations were examined to
simultaneously check for: 1) Illegal codes, 2) Inconsistencies, and 3) skip
pattern violations. Virtually all of the problem situations were checked
against the hard-copy questionnaires to determine the best possible
42
-------
resolution. Telephone calls to the respondents to resolve problems were
not attempted primarily because the majority of the data collected
concerned the pesticide products 1n storage at the time of the Interview.
Problems with these data generally could not be resolved by recontactlng
the respondents.
After resolving as many Inconsistencies as possible, missing data
fields were replaced with consistency codes that explained the reasons why
the data were missing. Table 7.1 explains the consistency codes. For
example, for a two-digit numeric field, a code of "94" represents a
response of "I don't know;" a code of "98" means that the item was left
blank but should have been completed; and a code of "99" means that the
Item was legitimately skipped based on the responses to previous questions.
The consistency code for "not applicable" (93 for numeric fields and NA
for alphabetic fields) was used for blank entries 1n fields that allowed
multiple responses and had a non-blank entry for at least one of the
potential response fields. For example, this code was used for the name,
age, and sex variables that extended beyond the number of members In the
household. Another example 1s Question 22. The unclrcled responses to
Question 22, type of product, were coded 93 when at least one of the type
of product responses had been circled.
7.4 Data Processing Management
During data collection, an RTI-developed control program on a personal
computer system maintained records that indicated the current status of
each sample household. The record for each sample member contained
location and sampling Information as well as codes Indicating each action
or event that occurred for the sample household member. Information
recorded by the system included:
receipt by RTI of completed forms
final result code from each screening form
status of editing and coding operations
data-entry status
Programs were developed on RTI's VAX system to allow Data Preparation
staff to quickly enter (using bar-coded ID labels) all forms received and
edited. The transaction files produced by these programs were then loaded
43
-------
Table 7.1
STANDARD RTI CONSISTENCY CODES
FOR SURVEY DATA BASES
Alpha Field
Code3
Numeric Field
Codeb
Description
NK
91
Never know. Respondent doesn't know now and
never will; therefore, do not attempt to
update the data later.
IL
92
Illegible. This code 1s used only for those
questions 1n which the response could not be
determined.
NA
93
Not applicable.
DK
94
Don't know. This code Indicates a written
response by the interviewer Indicating that
the respondent did not know the answer.
BD
95
Out-of-ranqe response. This code Is used
when the response or transcription exceeds
the specified field width or allowable value
range (e.g., cannot have a month = 13).
MR
96
Multiple response. This code 1s used when
the respondent gave more than one answer to a
question that called for only one response,
and the multiple response could not be
resolved.
RE
97
Refusal. This code 1s used when the
respondent refused to answer the Item.
BL
98
Blank or nonresponse. This code 1s used for
all cases In which there was no response for
an item, other than legitimate nonresponse
(see below).
LS
99
Legitimate nonresponse. This code 1s used
when the respondent should not have answered
the question (I.e., was routed around the
1tem).
aAll alphabetic data, Including consistency codes, are left-justified and
filled with rightmost blanks.
t»Al1 numeric consistency codes are left-filled with 9s for field widths
greater than two.
44
-------
Into the PC system and processed. The update process checked each
transaction according to pre-determlned logical rules before updating the
system. Therefore, for example, an edit event would not be allowed before
a receipt event. These checks prevented keying errors and other mistakes
from producing inconsistent status codes 1n the control system.
The control system was used to track progress 1n receiving forms from
the field and to monitor production 1n editing and keying operations.
Reports were run at least weekly and delivered to the project director and
the data collection task manager. The system was also used to locate
particular forms when questions arose.
45
-------
THIS PAGE LEFT BLANK INTENTIONALLY
46
-------
8. SAMPLING WEIGHTS
A properly designed sample survey 1s based on sampling units selected
with known probabilities of selection. Design-unbiased estimates of linear
statistics are then achieved by weighting the observations for each
sampling unit inversely to the probabilities of selection (Cochran, 1977;
K1sh, 1965). Analytical expressions for the NHGPUS sampling weights are
presented 1n this chapter.
The NHGPUS sampling design can be described briefly as a three-stage
probability sampling design. As discussed In Chapter 4, counties were
selected at the first stage of sampling, subcounty areas were selected
within the sample counties at the second stage, and Individual housing
units were selected from the sample areas at the third stage. The first
section of this chapter discusses the sampling weights based on the
probabilities of selection at the three stages of sampling.
The sampling weights based on the probabilities of selection would
enable unbiased estimation of population characteristics If data were
successfully collected for all units selected Into the sample. In
practice, however, virtually all surveys experience some level of
nonresponse (e.g., some randomly selected sample subjects refuse to
participate). When nonresponse occurs, the sampling weights enable
unbiased estimation of linear statistics only for the population of units
that would have responded to a census (a sample 1n which all members of the
population were surveyed). Therefore, statistical nonresponse adjustments
are generally performed to extend inferences from the respondents to the
entire survey population.
Survey nonresponse can be dichotomized as unit-level nonresponse and
Item-level nonresponse. The former occurs when no data are collected for a
sampling unit (e.g., when a person refuses to participate), and the latter
occurs when a participating sample member falls to provide data for an
individual survey Item (e.g., when the participant doesn't know how often a
pesticide product was used during the past year). Weight adjustment
procedures are generally used to compensate for unit-level nonresponse.
Adjustments for Item-level nonresponse Include both weight adjustments and
imputations (substitution of estimates for the missing data). The
47
-------
nonresponse adjustment procedures are generally designed to compensate, to
the extent possible, for the potential bias that could occur because of
differences between the responding and nonrespondlng members of the
population. An overview of survey nonresponse adjustment procedures 1s
provided by Madow et al. (1983).
The weight adjustment procedures used to compensate for unit
nonresponse in the NHGPUS are discussed 1n the second section of this
chapter. The third section discusses the nonresponse adjustment procedures
employed to compensate for Item nonresponse.
8.1 Weights Based on the Sampling Design
This section discusses the NHGPUS sampling weights based on the three
stages of the probability sampling design described 1n Chapter 4.
8.1.1 First-Stage Sample of Counties
When first-stage sampling units (FSUs) are selected with probabilities
proportional to size, as they were for the NHGPUS design, the weight
component for each sample FSU selection 1s normally the reciprocal of the
expected frequency of selection. Moreover, the sum of the products of
these weights times the size measures 1s normally the estimated total
number of units 1n the target population based on the sampling frame size
measures. For example, using the expected frequency of selection (4-1) for
the 1981 sample,
N,
¦ CM
£
leS
E1 S(1')
r=i
180 S(1)
S(1) = E1 S(l') . (8-1)
r=i
However, for the NHGPUS sample, the expected frequency of selection for
the 1-th FSU (county) 1s given by (4-3). The denominator of (4-3) 1s an
estimate of the December 1988 household population computed as 1f this
Information were available only for the 180 counties 1n the 1981 sample.
Because the December 1988 projections from Market Statistics, Inc. were
available for afH counties 1n the target population, the sampling weights
given by the reciprocal of (4-3) were poststratlfled by multiplying them by
the following ratio-adjustment factor,
48
-------
N1
i1 Mjd)
R(t) = jj ^-jj (8-2)
[ E1 S(1)] { E1 Is (1) CM1(1) / S(1) / 180}
1=1 1=1 h 1
Therefore, the first-stage weight component for the 1-th FSU (county) 1n
the NHGPUS sample Is
Ni
W,(1) = E1 FUl) / 60 M, (1) . (8-3)
1 1=1 1 1
The sum of the products of these weights times the county size measures,
Hi(1), estimates the total number of households 1n the NHGPUS population 1n
December 1988, I.e.,
N1
E * W.(1) M.(1) - E1 M.(1) . (8-4)
1eSj 11 1=1 1
8.1.2 Second-Stage Sample of Subcounty Areas
An Implicitly stratified sample of five area segments was selected at
the second stage of sampling for each first-stage FSU selection. Thus, the
weight component for the second stage of sampling is the reciprocal of the
conditional expected frequency of selection given by (4-4). Therefore, the
conditional weight component for the j-th second-stage sampling unit In the
1-th county 1s
Mi)
W2(1J) = E M2(1,j) / 5 M2(1J) . (8-5)
J ^
For large area segments (generally those containing 200 or more housing
units), an additional stage of sampling, called subsegmentlng, was imbedded
Into the second-stage sample selection process. The weight component for
this stage of sampling is the reciprocal of the probability of selection of
the subsegment, given by (4-5). Therefore, the conditional weight
component for the (1J,k)-th subsegment is
49
-------
M3(1J,k)
1f the (1,J)-th segment was (8-6)
subsegmented
W3(1,J,k) =
. 1 1f the (1,j)-th segment was not subsegmented.
When the sample segment contained many more housing units than expected
based on the second-stage size measure, M2OJ), this weight component
compensates for the unanticipated growth.
8.1.3 Third-Stage Sample of Housing Units
An equal probability sample of housing units was selected from those
listed for the (1,J,k)-th area segment (or subsegment). The weight
component for this stage of sampling 1s then the reciprocal of the third-
stage probability of selection (4-6). Therefore, the conditional third-
stage weight component for the (1,j,k,£.)-th housing unit 1s
Given the weight components for all stages of sampling, the final
sampling weight for the (1,j,k,£.)-th housing unit based on the sampling
design 1s the product of the weight components for all stages of sampling.
Therefore, the final design-based sampling weight for the (1,j,k,£.)-th
housing unit 1s
Because sampling units were selected with probabilities proportional to
size at the first and second stages of sampling, and the third-stage
allocation was designed to yield approximately equal probabilities, the
final sampling weights (8-8) are approximately equal for most sample
housing units. They are not Identically equal primarily because the size
measures used for the second stage of sampling (1980 Census counts of
housing units) were somewhat Inaccurate due to being out of date. They
were, however, the best size measures available at the time that the sample
was selected.
W4(1,j,k,£.) = N4(1J,k) / m4(1,J,k).
(8-7)
W*(1J,k,£) = Wx(1) W2(1,j) W3(1,j,k) W4(1,j,k,£).
(8-8)
50
-------
8.2 We1qht1nq-CIass Adjustment for Unit Nonresponse
Nonresponse Inevitably occurs 1n all sample surveys. The NHGPUS was
designed to collect data from all sample housing units that were occupied
as permanent residences. Failure to collect data for any eligible sample
housing units (e.g., because no one was found at home or because of
refusal) results 1n the possibility of bias due to differences between the
respondents and nonrespondents.
The best protection against nonresponse bias 1s a high survey response
rate, generally 80 percent or better. Since the NHGPUS achieved an 84.9
percent response rate, the potential for nonresponse bias 1s low.
Nevertheless, compensation for unit nonresponse 1s necessary to enable
estimation of population totals and to compensate for differential rates of
nonresponse.
We1ght1ng-class weight adjustment procedures were used to compensate
for unit nonresponse. These procedures categorize the sample housing units
Into categories called weighting classes that are defined so that
respondents and nonrespondents are more alike within classes with respect
to their survey responses and/or their propensity to respond than between
classes. The sampling weights of the respondents are then ratio-adjusted
to the sum of the sampling weights for all eligible sampling units within
each weighting class. For estimation of linear statistics (e.g.,
population totals), this weight adjustment procedure 1s equivalent to
substituting the mean response of the respondents for the missing
observations within each weighting class. Therefore, each weighting class
1s generally required to contain at least 20 to 30 respondents.
Survey response rates were examined with respect to several potential
weighting-class variables: Census Division, State, county, county-level
and ED/block-level urbanization variables used for stratifying the sample,
and the average dwelling value and percent multl-family dwelling variables
used to stratify the second-stage sample. Weighting classes based on
Census Divisions and the urbanization variable used to stratify the first-
stage sample of counties were determined to be the most effective
weighting-class variables. The NHGPUS response rates are presented by
weighting classes 1n Table 8.1. In most Census Divisions, higher response
rates were achieved 1n predominantly rural counties than 1n predominantly
51
-------
Table 8.1 Response Rates by Weighting Classes
County
Census Division3 Urban1c1ty
No. Eligible
Households^
No. Responding
Households
Response
Rate
1. New England
Urban
97
82
84.5%
Rural
42
37
88.1
2. Middle Atlantic
Urban
305
257
84.3
Rural
42
41
97.6
3. East North Central
Urban
319
262
82.1
Rural
81
73
90.1
4. West North Central
Urban
139
129
92.8
Rural
91
81
89.0
5. South Atlantic
Urban
244
204
83.6
Rural
121
114
94.2
6. East South Central
Urban
99
89
89.9
Rural
39
37
94.9
7. West South Central
Urban
177
126°
71.2C
Rural
102
91
89.2
8. Mountain
Urban
81
74
91.4
Rural
70
65
92.9
9. Pacific
Urban
398
316
79.4
Rural
_d
_d
_d
Total
2,447
2,078
84.9
aSee Table 1.2 of Volume I for definition of the Census Divisions.
&A11 housing units occupied as permanent residences were eligible for the
survey.
cThe forms for all 36 sample housing units worked by one Interviewer In
Harris County, Texas were lost 1n the mall. The Interviewer claimed to
have completed 27 interviews. RTI verified by telephone that Interviews
were conducted for at least some of these sample households. Another four
completed Interviews were lost 1n the mall for three other sample areas.
dNo counties classified as rural were selected for the Pacific Division.
52
-------
urban counties. Although Census Divisions are fairly broad categories,
they were sufficient to capture some major response rate differences. More
narrowly-defined weighting classes could potentially remove more
nonresponse bias, but they could, also reduce precision by Increasing
unequal weighting.
The weighting-class weight adjustment factor for the (1 ,J,k,£.)-th
sample housing unit was computed as follows:
lc wt(1 ,j,k,£.) IE(1J,k(fc)
W5(1,j,k,£) = (8-9)
Ec W4(1J,k,£.) IR(1,j,k,e.)
where Ec denotes summation over all sample housing units that belong to the
same weighting class "c" as the (i,J,k,£)-th housing unit, If 1s a (0,1)-
1nd1cator of eligibility for the (1,j,k,£)-th sample housing unit, and Ir
1s a (0,l)-1nd1cator of response for that housing unit. Therefore, the
final analysis weight for the (1,j,k,£.)-th housing unit 1s given by
W6(1 ,j,k,£.) = wjd.j.k,^ * W5(1(j,k,e) * IR(1,j,k,£.). (8-10)
The sum of these analysis weights over all households 1n the NHGPUS
sample 1s 84,572,672, which 1s the survey estimate of the size of the
NHGPUS target population of households at the time that the survey was
conducted (August and September 1990). Considering differences 1n
definitions of the survey populations, this estimate 1s consistent with the
Census Bureau's estimate of 94,596,000s occupied housing units 1n the
United States in the third quarter of 1990 based on the Current Population
Survey. The Census Bureau estimate Includes the following domains that do
not belong to the NHGPUS survey population: the states of Alaska and
Hawaii, vacation homes (non-permanent residences), homes on military
reservations, and homes on Indian reservations.
8.3 Compensating for Item Nonresponse
Item nonresponse occurs when the survey respondent does not provide a
response for some Individual data Items, such as the number of times a
product was used 1n the past year. The strategies available to compensate
for Item nonresponse include statistical Imputation and weight adjustments.
5Personal communication with Bob CalUs, (301) 763-8165, at the U.S. Bureau
of the Census on October 31, 1990.
53
-------
Both Imputation and weight adjustment procedures begin by partitioning the
sample Into classes so that respondents and nonrespondents are more alike
within classes than between classes, as discussed previously for weighting-
class adjustments for unit nonresponse.
Imputation procedures replace missing Item data values with data values
selected from members of the same Imputation class who responded to the
item. Imputations simplify analyses because the analyses can proceed as 1f
complete data were obtained. However, sampling variances tend to be
underestimated when the Imputed values are treated as actual responses
(Rubin, 1987). Moreover, relational analyses based on Imputation-completed
data can be misleading because the relationships between variables may be
affected 1n unexpected ways (Lepkowskl et al., 1984). Imputations were not
used for any of the NHGPUS analyses.
Weighting-class weight adjustment procedures were used when adjustments
were considered necessary for Item nonresponse 1n the NHGPUS analyses
(e.g., for estimating the total number of single-family and mult1-fam1ly
households 1n the target population). The nonresponse adjustment factors
were computed exactly as described 1n Section 8.2 for unit nonresponse,
except that only the households for which data were available for the
1tem(s) 1n the specific analysis were treated as the respondents. The
weighting classes established to compensate for unit nonresponse were also
used for Item-level weight adjustments for all estimates of population
totals. New weighting classes were not created for the item nonresponse
adjustments to save time and expense and because the rate of occurrence of
Item nonresponse was usually low.
8.4 Quality Assurance Procedures
The correctness of sampling and analysis weights is of such fundamental
Importance for correct statistical analyses that RTI routinely performs
quality assurance checks for all weight files. For example, the products
of the first-stage weights and size measures are summed to verify that they
sum to the total size measure for the sampling frame. When weighting-class
weight adjustments are performed, the Initial weights for all ellglbles and
the final weights for all respondents are summed to verify that they are
identical for every weighting class.
Weight checks were Implemented for the NHGPUS for every stage of
sampling and every stage of weight adjustment. The weight checks were
54
-------
reviewed and certified by a senior sampling statistician not directly
Involved 1n the project.
55
-------
THIS PAGE LEFT BLANK INTENTIONALLY
56
-------
9. STATISTICAL ANALYSIS METHODS
Because a stratified, multistage sampling design was used to select an
efficient sample for the NHGPUS, analysis procedures that account for the
complex sampling design must be used to properly analyze the survey data.
The sampling weights discussed 1n Chapter 6 must be used to compute design-
unbiased point estimates of population parameters. The sampling variances
of survey statistics must also account for the stratification and
multistage sampling (Wolter, 1985). Therefore, special-purpose software
developed by RTI over the past 15 years for analysis of complex sample
survey data was used to analyze the NHGPUS data base (Shah et al., 1989).
RTI's software has been tested and reviewed by many independent researchers
and found to produce accurate results efficiently (Francis and Sedransk,
1979; Kaplan et al., 1983; Cohen et al., 1986).
Estimation of population totals and their variances provides the basis
for variance estimation for other population parameters such as means and
proportions. Therefore, this chapter begins with a discussion of the
NHGPUS estimation procedures for population totals.
9.1 Estimating Totals and Associated Variances
The sampling distribution of a statistic based on a probability sample
from a finite population 1s the distribution Induced by the sampling
design: the strata, clusters, and probabilities of selection. Therefore,
the choice of appropriate estimators of sampling variance 1s linked
directly to the sampling design.
The sample design for the NHGPUS was a stratified, three-stage design.
The 60 first-stage sampling units were selected with probabilities
proportional to size (pps) using a sequential probability minimum
replacement (pmr) algorithm (Chromy, 1979). The size measure for each FSU
was a current estimate of the number of occupied housing units In the
county.
Variance estimation procedures for the sequential pmr sampling
algorithm are discussed by Chromy (1981). He recommends any of three
alternative estimators: an assumed replacement estimator, a successive
difference estimator, and a collapsed stratum estimator. The successive
difference and collapsed stratum estimators make use of the Implicit
57
-------
stratification that results from sequential sampling from an ordered 11st.
The NHGPUS variance estimates were computed using a collapsed stratum
estimator that assumes sampling with replacement at the first stage of
sampling.
When the 60 FSUs were selected at the first stage of sampling, the
sequential pmr algorithm created 60 sampling zones or strata of equal size
(equal expected numbers of housing units) and selected one FSU to represent
each zone. Stratification of the sample resulted from sorting the sampling
frame by the stratification variables discussed 1n Appendix A. A sample of
58 distinct counties was selected, two large counties having been selected
to represent two zones each. Collapsed strata were defined for variance
estimation by first sorting the 58 sample counties 1n exactly the same
order as used for sample selection (I.e., by the first-stage sampling
strata). The first pair of sample counties was then assigned to Analysis
Stratum 1, the next pair to Analysis Stratum 2, and so on through Analysis
Stratum 29 for the last pair of sample counties.
Using the collapsed strata and assuming replacement sampling at the
first stage results 1n estimates of sampling variances with small positive
bias. The bias results from collapsing strata and from Ignoring the
covarlances Induced by sampling without replacement from a finite
population. Such estimates are conservative 1n the sense that Interval
estimates based upon them (e.g., for means or proportions) will be slightly
wider In expectation (or, on the average, over all possible samples) than
intervals based on an unbiased estimator. Estimates of population totals
are of Interest for the NHGPUS primarily because estimates of population
proportions are actually ratios of estimated totals. For example, the
estimated proportion of households that disposed of a pesticide container
with the regular household trash 1n the past year 1s the ratio of the
estimated number of such households divided by the estimated number of
households 1n the NHGPUS target population.
If we let y represent any household characteristic for which the
population total 1s of Interest (e.g., a (0,l)-1nd1cator of whether or not
the household disposed of a pesticide container with the regular household
trash), the total number of households 1n the survey population with that
characteristic can be estimated as
58
-------
29 2 *
Y = £ r Y(r,1)
r=l 1=1
(9-1)
where Y(r,1) is the estimated total of the characteristic y over all
households In the 1-th sample county of analysis stratum "r." This county
A
total, Y(r,1), can be estimated as
Y(r,1) = I weOJ.k.fc) yd.j.k.t), (9-2)
where the summation, E, 1s over all responding sample housing units In the
1-th sample county and W6 is the final analysis weight discussed 1n Section
8.2
The sampling variance of the estimated population total, Equation
(9-1), was then calculated for NHGPUS estimates as follows:
2
29 2
V(Y) = 2 E E
r=l 1=1
Y(r,1) - ?(r)
(9-3)
where
Y(r) = [Y(r,1) + Y(r,2)] / 2 . (9-4)
As noted previously, the Importance of the procedures for estimating NHGPUS
population totals and their sampling variances 1s primarily the role they
play 1n estimation of population proportions, which 1s explicitly
formulated 1n the next section.
9.2 Estimating Means, Proportions, and Associated Variances
The parameters of primary interest for the NAWWS are population
proportions (e.g., the proportion of households that disposed of a
pesticide container with the regular household trash In the past year).
Both means and proportions 1n the survey population can be expressed as the
ratio of two population totals. In general, the population ratios can be
expressed as
R = Yn / Yd , (9-5)
where Yn and Y
-------
would be the total number of households that used such a service, and yn
and yd would be (0,l)-1nd1cators of these conditions for each household 1n
the population.
A consistent estimator for R (i.e., one that converges to R as the
sample and population sizes become Infinite) 1s
where Yn and Yd are the estimated population totals as calculated from
Equation (9-1) for the numerator and denominator varlates yn and yd,
respectively. R 1s a slightly biased estimator for R, but the bias 1s on
the order of 1/n, where n 1s the number of sample households that
contribute to the estimated domain total Yd. Hence, the bias 1s generally
negligible for reasonably large analysis domains (e.g., with 30 or more
sample observations).
Ratio estimates are nonlinear functions of the observations. The
sampling variance Induced by the sampling design cannot generally be
expressed 1n closed form for nonlinear statistics. A frequently used
approximation for the variance of an estimated ratio 1s based on the first
term of a Taylor series expansion of the ratio (Cochran, 1977). A
computationally convenient way to express this variance calculation is to
first define a new "linearized" variable, as follows:
The estimated variance of the ratio statistic, R, can then be expressed as
where Z is the estimated population total, given by Equation (9-1) for the
linearized statistic, z, and V(Z) 1s calculated from Equation (9-3).
Cochran (1977) discusses the accuracy of Equation (9-8) for estimating
the variance of the ratio statistic, Equation (9-6). He notes that
omitting the second and higher order terms of the Taylor series expansion
generally results 1n Equation (9-8) being an underestimate of the true
variance. A guideline suggested by Cochran 1s that Equation (9-8) will
generally yield satisfactory results 1f the ultimate sample size exceeds 30
units and the coefficients of variation (CVs) for the estimates of 7n and
(9-6)
* bn(U.k.l) - R ydO.J.k.O] / *d
(9-7)
V(R) = V(Z)
(9-8)
60
-------
?d are both less than 10 percent. In practice, If the ratio estimate Is a
domain mean or proportion, the number of domain members 1n the sample
contributing to Yn and Y(j should be at least 30.
9.3 Suppression Rule
The survey estimates presented 1n the next chapter are footnoted as
having poor precision whenever the relative standard error of the estimate
exceeds 50 percent. The relative standard error, RSE, of an estimated
a
population proportion, P, can be represented as
RSE(P) = Jv (P) / P , (9-9)
*S A A
where P 1s calculated from Equation (7-6) and V(P) 1s calculated from
Equation (9-8).
However, as noted 1n Section 9.2, the linearization method of variance
estimation for ratio statistics such as proportions can produce
underestimates. This occurs primarily for very small domains and for very
small proportions. To guard against reporting unreasonably small standard
errors, the variance expected using a simple random sample of the same size
was substituted 1n the tables and 1n Equation (9-9) for the design-based
variance calculated from Equation (9-8) whenever the simple random sampling
variance was larger. For estimates of population proportions, the simple
random sampling variance was calculated as
VSRS (P) = P (1 - P) / n , (9-10)
where n 1s the number of sample units contributing to the denominator of
the proportion. This variance estimate 1s a logical upper bound for the
A
variance of P under the survey design because the clustering Involved 1n
multistage sampling almost always results in positive Intracluster
correlations.
Many estimates 1n the analysis tables have RSEs that are approximately
100 percent. These estimates are usually based on only a single
observation. This observation could represent a rare event that was
observed only once 1n the survey or could be the result of interviewer
error, data entry error, etc. Such estimates should be regarded with
considerable skepticism. Generally, estimates with RSEs less than 30
percent are quite reliable; estimates with RSEs between 30 and 50 percent
61
-------
are acceptable; estimates with RSEs between 50 and 100 percent are quite
unreliable; and estimates with RSEs of 100 percent or more are totally
unreliable.
9.4 Statistical Inferences
Two statistical Inference procedures were used 1n analysis of the
NHGPUS data base: confidence Interval estimation and testing for
significant differences between population proportions. These Inference
procedures are discussed 1n the context of the complex probability sampling
design utilized for the NHGPUS 1n the subsections that follow.
9.4.1 Confidence Interval Estimates
The proportion of the NHGPUS target population, or of any analysis
domain (e.g., households that used a commercial lawn care service 1n the
past year), that had a given characteristic (e.g., received written safety
precautions) 1s estimated as a ratio statistic computed using Equation
(9-6). So long as the denominator of the ratio or proportion 1s based on a
reasonably large sample size (e.g., 50 or more observations), the sampling
distribution of the estimated population proportion 1s approximately the
normal probability distribution. Therefore, an approximate 95 percent
confidence Interval estimate of a population proportion, P, Is given by
where P and V(P) are the point estimate of P and Its sampling variance
computed from Equations (9-6) and (9-8), respectively.
9.4.2 Testing for Significant Differences 1n Proportions
Letting Pj and P2 represent the ratio estimates calculated from
Equation (7-6) for two NHGPUS population proportions Pi and P2,
respectively, the difference, D, between the population proportions 1s
estimated by
The sampling variance for the difference in proportions 1s computed using
the "linearized" difference, zq, defined as
P + JV(P)
(9-11)
(9-12)
zD = 2j(1 J,k,£) - z2(1,j,k,£)
(9-13)
62
-------
where z\ and zz are the linearized variables computed from Equation (9-7)
A A
for the ratio statistics Pi and P2, respectively. The sampling variance of
/v
the estimated difference, D, 1s then computed as
5(B) = V(ZD) , (9-14)
a
where Zq 1s the estimated population total calculated using Equation (9-1)
A
for the linearized difference, zq, and V(Zq) 1s calculated from Equation
(9-3).
A test statistic, Tjt for the null hypothesis of no difference 1n the
population proportions Pi and ?z can then be calculated as
This test statistic 1s a classical "t-stat1st1c" as used for traditional
statistical sampling inferences except that the sampling variance 1n the
denominator 1s based on the sampling distribution induced by the stratified,
multistage NHGPUS sampling design. Thus, as 1s true for traditional
statistical Inferences, 1f the null hypothesis of no difference between the
population proportions 1s true, the sampling distribution of Ti 1s
approximately the standard normal probability distribution (see Section 2.4 of
Skinner et al., 1989). The null hypothesis 1s rejected 1f the ratio, T1, 1s
large relative to that probability distribution. The normal probability
distribution 1s the appropriate reference distribution, rather than the
Student's t distribution, because of the large size of the NHGPUS sample.
63
-------
THIS PAGE LEFT BLANK INTENTIONALLY
64
-------
REFERENCES
Berman, P. L. (1981). National Survey of Household Pesticide Usage Pilot
Study Final Report] RTI/1864/18-01F. Research Triangle Institute,
Research Triangle Park, NC.
Chromy, J. R. (1979). Sequential Sample Selection Methods. Proceedings of
the American Statistical Association Section on Survey Research Methods.
401-406.
Chromy, J. R. (1981). Variance Estimators for a Sequential Sample Selection
Procedure. In: Current Topics 1n Survey Sampling, D. Krewski, R. Platek,
and J.N.K. Rao, eds., Academic Press, New York, NY, 329-347.
Cochran, W. G. (1977). Sampling Techniques. 3rd ed. Wiley, New York, NY.
Cohen, S. B., Burt, V. L., and Jones, G. K. (1986). Efficiencies 1n Variance
Estimation for Complex Survey Data. The American Statistician 40:2, 157-
163.
Francis, I., and Sedransk, J. (1979). A Comparison of Software for Processing
and Analyzing Survey Data. Bulletin of the International Statistical
Institute 48, 1-31.
Groves, Robert M. (1989). Survey Errors and Survey Costs. W1lev, New York.
NY.
Immerman, F. W. and Schaum, J. L. (1990). Nonoccupational Pesticide Exposure
Study (NOPES) Final Report. EPA/600/3-90/003. U.S. Environmental
Protection Agency, Washington, D.C.
Kaplan, B., Francis, I., and Sedransk, J. (1983). A Comparison of Methods and
Programs for Computing Variances of Estimators from Complex Sample
Surveys. Proceedings of the American Statistical Association Section on
Survey Research Metnods. 97-100.
K1sh, L. (1965). Survey Sampling. Wiley, New York, NY.
Lebowltz, M. D., Quackenboss, J. J., Kollander, M., Soczek, M. L., and Colome,
S. (1989). The New Standard Environmental Inventory Questionnaire for
Estimation of Indoor Concentrations. Journal of the Air and Waste
Management Association 39:11, 1411-1419.
Lepkowskl, J. M., Stehouwer, S. A., and Landls, J. R. (1984). Strategies for
the Analysis of Imputed Data 1n a Sample Survey. Proceedings of the
American Statistical Association Section on Survey Research Methods. 622-
627.
Lynch, J. T., Drummond, D. J., and Waddell, R. (1981). National Household
Pesticide Usage Survey California Pretest Final Report. RTI/1864/20-04F.
Research Triangle Institute, Research Triangle Park, NC.
65
-------
Madow, W. G., Olkln, I., Rubin, D. B., eds. (1983). Incomplete Data 1n Sample
Surveys. Volume 2: Theory and Bibliographies. Academic Press, New York,
NY.
Rubin, D. A. (1987). Multiple Imputation for Nonresponse 1n Surveys. Wiley,
New York, NY.
Shah, B. V., LaVange, L. M., Barnwell, B. G., K11linger, J. E., and Wheeless,
S. C. (1989). SUDAAN: Procedures for Descriptive Statistics, User's
Guide. Research Triangle Institute, Research Triangle Park, NC.
Williams, R. L. and Chromy, J. R. (1980). SAS Sample Selection MACROS.
Proceedings of the Fifth Annual SAS Users Group International Conference.
392-396.
Wolter, K. M. (1985). Introduction to Variance Estimation. Sprlnger-Verlag,
New York, NY.
66
-------
APPENDIX H
FIRST-STAGE SAMPLING DESIGN FOR THE 1981 NATIONAL HOUSEHOLD
PESTICIDE USAGE SURVEY
-------
H.l The Primary Sampling Frame
The primary sampling frame Included all of the county units (counties,
parishes, and Independent cities) 1n the 48 coterminous States and the
District of Columbia. Size-measures assigned were 1980 estimated numbers
of non-farm housing units, computed by subtracting estimated 1978 numbers
of farms from 1980 U.S. Census total numbers of housing units. Numbers of
farms were estimated from 1978 U.S. Census of Agriculture data; for each
county unit the number of farms that reported 1n the mall survey was
adjusted by adding a proportionate share of the State estimate of non-
reporting farms, derived from the area segment survey. It was assumed that
numbers of farms had not changed appreciably from 1978 to 1980, and that
only a very small proportion of farms would have more than one farm housing
unit.
So that large enough samples could be selected at the second stage,
small county units were combined so that all sampling units had a size
measure of 600 or more non-farm housing units. In every case the units
Involved were contiguous rural counties 1n the same State climatic
division. As a result, the primary sampling frame consisted of 3,062
sampling units, of which 3,017 were Individual counties, 43 were made up of
two counties, and two Included three counties, thereby Including all of the
3,109 primary governmental units outside of Hawaii and Alaska.
H.l.l. Stratification of the Primary Frame
In preparation for selecting the sample, the 3,062 sampling units were
grouped or stratified Into 83 primary strata. Stratification Is the
grouping together of units to Improve the precision of population estimates
(make sampling errors smaller) and to ensure a representative spread of the
H-l
-------
sample across the range of values for the stratification variables. For
the pesticide use sample, census geographic division codes are an example
of a stratification variable. Ideally, from a statistical viewpoint,
counties that have about the same use of pesticides should be Included In
the same stratum; however this 1s Impractical to achieve 1n the actual
design of the sample. Consequently, stratification variables are selected
that are thought to produce strata that are relatively homogeneous In
pesticide use.
Stratification of the primary sampling frame was carried out on the
basis of geographical area, urbanization, precipitation, temperature, and
ethnic composition of population. First, It appeared obvious that 1n
different parts of the United States there are various kinds of Insects,
and It also seemed logical to assume that different kinds of pesticides are
used to combat some of the same pests In different areas. The use of a
particular pesticide might be legal 1n some States but not 1n others. For
the first level of stratification 1t was therefore decided to assign to
each primary sampling unit Its census geographic division (CGD) code. The
makeup of the CGDs 1s shown 1n Table H.l.
Secondly, It was considered that In rural areas there would be more
pests per household than 1n urban locations, some of them attracted by
livestock, crops, and marshy areas. In addition,less community spraying
would be expected, so that more of the responsibility for pest control
would fall on the residents themselves. For the second level of
stratification an "urbanization code" (UC) was therefore assigned to each
unit. If a county unit either (a) was part of a standard metropolitan
H-2
-------
Table H.l. The 48 Coterminous States and the District of Columbia
by Census Geographical Divisions
Code Division States
1 New England Maine, New Hampshire, Vermont, Massachusetts
Rhode Island, Connecticut
2 Middle Atlantic New York, New Jersey, Pennsylvania
3 East North Central Ohio, Indiana, Illinois, Michigan, Wisconsin
4 West North Central Minnesota, Iowa, Missouri, North Dakota,
South Dakota, Nebraska, Kansas
5 South Atlantic Delaware, Maryland, District of Columbia,
Virginia, West Virginia, North Carolina,
South Carolina, Georgia, Florida
6 East South Central Kentucky, Tennessee, Alabama, Mississippi
7 West South Central Arkansas, Louisiana, Oklahoma, Texas
8 Mountain Montana, Idaho, Wyoming, Colorado, New Mexico
Arizona, Utah, Nevada
9 Pacific Washington, Oregon, California
-------
statistical area (SMSA)1 with a central city of 200,000 or more population
or (b) contained all or part of a city of 25,000 or more, 1t was assigned
the value "2"; otherwise the value was "1".
It also appeared logical that numbers of pests are related to average
temperature and annual precipitation, and It was known that within some of
the census divisions rather wide ranges of one or both occur. Accordingly,
a code for the third level of stratification was assigned to reflect
differences 1n long-term average temperature, average annual precipitation
or both. It was not possible to obtain data for Individual county units,
but average temperature and precipitation data for the period 1949-70 were
obtained for State climatic divisions, made up of contiguous areas having
fairly uniform conditions.2 In most of the States the boundaries were found
to follow county lines, and each county unit was assigned the data for Its
division; 1n the remaining States the data for the division containing the
largest share of the county unit were applied. Average Farenhelt
temperatures were converted to codes representing two-degree Intervals, and
average annual precipitation data to codes representing two Inch Intervals,
and, using size-measures as weights, a frequency distribution of the
1 Except 1n New England, a standard metropolitan statistical area 1s defined
by the Office of Management and Budget publication Standard Metropolitan
Statistical Areas: 1967. U.S. Government Printing Office, Washington, DC,
20402, as ua county or group of contiguous counties which contains at least
one city of 50,000 Inhabitants or more, or 'twin cities' with a combined
population of at least 50,000 ... contiguous counties are Included If they
are socially and economically Integrated with the central city." In New
England the units are cities and towns instead of counties.
2Monthly Averages of Temperature and Precipitation for State Climatic
Divisions 1941-70. CIImatography of the United States No. 85 (By State).
U.S. Department oT Commerce, National Oceanic an3 Atmospherlc
Administration Environmental Data Service, National Climatic Center,
Ashevllle, NC, July 1973.
H-4
-------
primary listing units was run by census division code, urbanization code,
temperature code, and precipitation code. The pattern for each of the CGD-
UC combinations was carefully examined, and on the basis of the ranges and
the distribution, the units were assigned codes for two or three strata
based on one or both of the climatic factors. The ranges of the
distribution are shown in Table H.2; the values used 1n defining the strata
are shown 1n Table H.3.
Finally, 1t appeared logical that within an area with fairly uniform
urbanization and climatic conditions different use of pesticides might be
made by population groups with different ethnic backgrounds and/or
different levels of Income. It was decided to carry out a fourth level of
stratification on the basis of 1980 Census percentage of the black
population because 1t could be expected to reflect both of those factors.
Accordingly, a weighted frequency distribution of the listing units was run
by census division code, urbanization code, climatic code, and percentage
black population, using five-percent intervals for the latter. Then 26 of
the 47 strata that had been developed by the use of the first three
stratification factors were each divided Into two or three final strata,
depending on the percentage range and the distribution of the size-
measures. The 83 strata are described 1n Table H.3.
H.2 Allocation and Selection of the Primary Sample
It would have been possible to allocate the desired 180 primary sample
selections explicitly to the 83 strata on the basis of their size-measures,
the estimated numbers of nonfarm housing units. That would, however, have
resulted 1n considerable variable from pps (probability proportional to
size) selection as a result of the necessary rounding of the allocations to
Integers. For that reason an alternative procedure was used, whereby it
H-5
-------
Table H.2 Minimum and Maximum Average Annual Precipitation
and Mean Temperature, by Census Divison and Urbanization Code
Census Urbanization Precipitation, 1n. Temperature, °F
Division Code M1n. Max. M1n. Max.
1
1
36.47
44.23
40.0
49.1
1
2
40.16
44.23
43.8
50.4
2
1
32.13
44.98
41.5
54.1
2
2
32.86
44.98
45.0
54.1
3
1
28.68
43.95
40.8
56.7
3
2
28.68
43.76
41.1
56.7
4
1
15.44
46.30
38.6
58.8
4
2
18.77
43.78
38.6
57.0
5
1
34.30
59.86
47.5
74.7
5
2
38.67
59.86
53.0
74.7
6
1
44.20
62.05
55.3
67.4
6
2
44.20
64.62
55.3
67.6
7
1
11.57
61.14
57.2
73.8
7
2
11.57
61.14
58.2
73.8
8
1
4.13
28.12
37.1
71.3
8
2
6.96
20.86
43.6
68.9
9
1
7.41
94.62
43.3
64.2
9
2
9.73
62.36
45.2
60.7
-------
Table H.3 Description of Primary Strata
Stratum
f
Stratification
Codesa
Average Annual
Precipitation (in.)
Average
Temperature (of)
Percent
Black
Estimated No.
of NonFarm
Housing Units
1
1111
less than 42
m
635,640
2
1121
42 or more
-
-
407,003
3
1211
below 48
-
376,352
4
1221
48 or above
less than 10
3,125,719
5
1222
48 or above
10 or more
276,724
6
2111
less than 38
-
-
503,528
7
2121
38 or more
below 50
-
799,916
8
2131
38 or more
50 or above
.
803,544
9
2211
less than 38
-
less than 10
451,274
10
2212
less than 38
-
10 or more
650,933
11
2221
38 or more
below 50
less than 10
1,230,653
12
2222
38 or more
below 50
10 to 34.99
316,469
13
2223
38 or more
below 50
35 or more
451,050
14
2231
38 or more
50 or above
less than 10
3,740,301
15
2232
38 or more
50 or above
10 to 34.99
3,287,633
16
2233
38 or more
50 or above
35 or more
1,883,800
17
3111
below 48
-
1,364,198
18
3121
-
48 to 51.9
-
1,367,181
19
3131
52 or above
-
1,047,132
20
3211
.
below 48
less than 15
753,530
21
3212
.
below 48
15 or more
455,327
22
3221
-
48 to 51.9
less than 15
4,145,102
23
3222
.
48 to 51.9
15 or more
4,158,234
24
3231
-
52 or above
less than 15
1,252,017
25
3232
52 or above
15 or more
973,722
26
4111
.
below 46
-
799,994
27
4121
_
46 to 51.9
-
944,315
28
4131
-
52 or above
-
1,163,236
29
4211
below 46
-
1,035,437
30
4221
.
46 to 51.9
less than 10
548,427
31
4222
.
46 to 51.9
10 or more
155,389
32
4231
.
52 or above
less than 10
733,391
33
4232
.
52 or above
10 or more
889,578
34
5111
-
below 58
less than 5
853,159
35
5112
below 58
5 to 14.99
373,234
36
5113
.
below 58
15 or more
429,448
37
5121
58 to 65.9
less than 20
602,164
38
5122
-
58 to 65.9
20 to 34.99
837,352
39
5123
58 to 65.9
35 or more
675,734
40
5131
66 or above
less than 15
428,978
41
5132
66 or above
15 or more
497,183
42
5211
-
below 58
less than 10
1,411,739
43
5212
-
below 58
10 to 39.99
819,266
44
5213
-
below 58
40 or more
693,254
45
5221
.
58 to 65.9
less than 20
920,784
46
5222
-
58 to 65.9
20 to 29.99
991,414
47
5223
.
58 to 65.9
30 or more
1,026,936
48
5231
_
66 or above
less than 10
963,486
49
5232
66 or above
10 to 14.99
1,488,609
50
5233
-
66 or above
15 or more
1,355,761
51
6111
.
below 60
less than 5
782,098
52
6112
_
below 60
5 or more
471,005
53
6121
.
60 or above
less than 30
568,614
54
6122
60 or above
30 or more
562,272
55
6211
_
below 60
less than 10
625,454
56
6212
-
below 60
10 to 19.9
530,133
57
6213
below 60
20 or more
502,489
-------
Table H.3 Description of Primary Strata (cont.)
Estimated No.
Stratum Stratification Average Annual Average Percent of NonFann
{ Codes1 Precipitation (in.) Temperature (°F) Black Housing Units
58
6221
.
60 or above
less than 30
59
6222
-
60 or above
30 or more
60
7111
less than 40
.
less than 5
61
7112
less than 40
-
5 or more
62
7121
40 or more
-
less than 15
63
7122
40 or more
-
15 or more
64
7211
less than 40
.
less than 10
65
7212
less than 40
-
10 to 14.99
66
7213
less than 40
-
15 or more
67
7221
40 or more
.
less than 10
68
7222
40 or more
-
10 to 14.99
69
7223
40 or more
.
15 or more
70
8111
.
below 46
.
71
8121
.
46 to 55.9
72
8131
.
56 or above
73
8211
.
below 48
less than 10
74
8212
.
below 48
10 or more
75
8221
-
48 to 59.9
.
76
8231
-
60 or above
77
9111
less than 26
.
78
9121
26 or more
.
79
9211
less than 28
less than 10
80
9212
less than 28
10 or more
81
9221
28 to 45.9
less than 5
82
9222
28 to 45.9
5 or more
83
9231
46 or more
.
567,534
570,681
784,984
282,775
625,340
892,770
1,781,144
805,821
812,675
580,883
1,181,241
1,107,375
748,181
613,183
330.408
617,809
227,458
1,042,671
825,729
658,002
985.409
4,348,673
3,614,110
1,002,231
591,653
700,981
»First digit - U.S. Census Division Code
Second digit Urbanization code (1 - rural, 2 ° urbanized)
Third digit - Precipitation and/or temperature code
Fourth digit - Percentage black code
-------
was possible to have all selections made so that (1) all selections of non-
certainty units would be made with probability exactly proportional to size
and (2) each certainty unit (I.e., unit with an expected number of
selection hits equal to or greater than unity) (a) would definitely be
selected the number of times corresponding to the Integer part of the
expected number and (b) would be selected an additional time with
probability proportional to the fractional part of the number. In order to
retain most of the Intended effect of the stratification, the listing units
were ordered serpentlnely by the four stratification codes and by size-
measure. First, ordering was done by census division code, 1n ascending
order. Then alternate divisions were ordered by urbanization code In
ascending and descending order. Next, within each CDC x UC group ordering
was done by climatic codes In alternating directions, followed by similar
ordering by black percentage code within each of the resulting three-way
groups. Finally, the units within the four-way groups were ordered by
size-measure In alternating directions. The sampling frame as so ordered
was then divided Into 180 "zones" of equal widths 1n terms of size-
measures, and one selection was made from each zone. As a result of these
procedures, all selections were made with the desired probabilities from
equal-sized strata, each of which tended to be homogeneous for most of the
ordering factors. The selections were made using the probability minimum
replacement sampling procedure (pmr) developed by Chromy (1979). As
desired, 180 selections were made; because several "certainty" units were
hit more than once, the number of distinct units selected was 166. The
combined use of hierarchical serpentine ordering and the sequential sample
selection method is described 1n Williams and Chromy (1980).
H-9
-------
APPENDIX I
SURVEY PEST NOTEBOOK
-------
National Home and Garden
Pesticide Use Survey
Pest Examples and Descriptions
for Card A Pest Categories
-------
National Home and Garden Pesticide Use Survey
INTRODUCTION
At various points throughout the survey the respondent is
asked to identify the categories of pests with which they have had
problems. Card A lists these categories along with examples of the
pests which fall into each category.
The pictures and descriptions in this book are meant to clarify
the terms used on Card A. Sizes given are approximate for full
grown pest unless noted otherwise. This information is intended
to be helpful when a respondent is familiar with the appearance of
a pest or the damage it causes but does not know the name of the
pest or uses a name different from the one used in the list.
CAUTION: This book should NOT be used as a definitive
identification tool. While the pictures shown are representative of
the pest named, there are often many varieties of the same pest,
each differing in appearance and habits. There are also many more
pests than those listed on Card A or in this book. Survey respond ents
should be referred to their county extension agents for positive pest
identification and suggestions for appropriate pest control methods.
Keep in mind that the goal is simply to record
the pest category. The specific pests are provided
only to guide you into the correct category. Do not
spend too much time identifying specific pests.
Note on terminology:
Various terms are used to describe the young
of insects including larvae, nymphs, grubs,
maggots, and juveniles.
Page i
-------
National Home and Garden Pesticide Use Survey
TABLE OF CONTENTS
Microorganisms
1. Mildew, Mold, Bacteria, Virus 1
2. Wood Decay or RoL 3
3. Plant Diseases 5
Insects
4. Cockroaches 7
5. Stored Food Insect Pests 9
6. Fabric Insect Pests 13
7. Termites 15
8. Other Wood Destroying Insects 17
9. Fire Ants 19
10. Any Other Ants 21
11. Scorpions 23
12. Bees, Hornets, Wasps 25
13. Mosquitoes 27
14. Flies, Gnats, Midges 29
15. Fleas 31
16. Ticks, Chiggers 33
17. Spiders, Crickets, Sowbugs/Pillbugs, Millipedes,
Centipedes 35
18. Soil-Dwelling Insects, Nematodes 37
19. Plant-Chewing Insects 39
20. Plant-Sucking Insects and Mites 45
Plants
23. Brush 49
24. Grass-like Weeds 51
25. Broadleaf Weeds 53
Animals
27. Slugs, Snails 57
28. Birds 59
29. Mice, Rats 61
30. Bats 63
31. Other Mammals 65
Sources for Illustrations 69
Page ui
-------
Pest 1
Mildew, Mold, Bacteria, Virus
Mildew, Mold
Caused by fungi; may result in decay and discoloration
Bacteria, Virus
May cause diseases
Page 1
-------
Pest 2
Wood Decay or Rot
Wood Decay or Rot
May result when wood is exposed to damp conditions; caused
by mold, fungi, or bacteria
Page 3
-------
Pest 3
Plant Diseases
Plant diseases are usually caused by bacterid, fungi, mycoplasma,
and viruses and tend to be associated with specific environmental
conditions (eg. temperature, humidity), insect infestations, or plant
stress. Diseases often produce a characteristic pattern of colors,
spots, growths, or rot. Treatment depends on the specific disease.
While exact diagnosis isbest left to experts, a few diseases are easily
recognized:
Powdery mildew
Very common, especially on squash, roses, zinnias, lilac; looks like
powdery white to gray covering on leaves
Black spot
Common, especially on roses; large, circular black spots on leaves
Brown patch
Turf disease; circular, small to large brown areas; darker at edges;
grass may regrow green from center of circle
Page 5
-------
Pest 4
Cockroaches
German Cockroach
Cockroaches (roaches, waterbugs)
Fast moving, flat, oval, usually brown; prefer dark, warm, moist
places
(size: l/8"-l 1/2" depending on species and stage of development)
American Cockroach
Page 7
-------
Pest 5
Stored Food Insect Pests
Confused flour beetle
Note: Use Pest 19 for frult/vlnegar fly found on fresh fruit
Grain/flour/meal beetles
Various small beetles (and their larvae) found in grain and
flour products; adults often brown, hard; larvae often light
colored, soft
(size: larva & adult l/10"-l/4")
Saw toothed beetle
Page 9
-------
Pest 5
Stored Food Insect Pests
Indian Meal Moth
Grain/flour/meal moths
Some caterpillars live inside kernels, others feed
in ground grain leaving web of silk threads;
adults tan, some with stripes or fringe on wings
(size: larvae 1/4" - 3/4", adults 2/3" - 3/4")
Angoumois Grain Moth
page 10
-------
Pest 5
Stored Food Insect Pests
Grain Weevil
Grain weevils
Characteristic weevil snout; adult brown; larvae of stored grain
weevils live inside grain kernels
(size: adult 1/8")
Yellow Mealworm
Mealworms
Larvae yellow or brown, "skin" is stiff, not soft; adults brown, slow
moving
(size: larva to 1"; beetles to 3/4")
Page 11
-------
Pest 6
Fabric Insect Pests
Webbing Cbthes Moth
Clothes Moths
Small, light colored caterpillars that eat fabric made of animal fibers
(wool, fur, silk, feathers) producing holes in fabric; adults are small
brown moths with fringed wings
(size: larva to 1/2"; adult 1/2")
Blade Carpet Bettle
Carpet beetles
Small, round, dark-colored beetles found in fabrics, carpeting,
upholstered furniture; larvae brownish, hairy
(size: adult l/10"-l/5", larvae to 1/3")
Silverfish
Silverfish and firebrats
Fast moving, silver colored, torpedo shaped; eat starch in glue,
fabric, paper
(size: to 1/2")
Page 13
-------
Pest 7
Termites
Termites
Wood eating insects; some live in soil, others in wood; soft bodied;
require moisture
(size: worker l/10"-l/2"; soldier and winged l/2"-l")
NOTE: Termites are not the same as flying ants. Certain life
stages and species of both termites and flying ants may have
wings and are often confused by the public.
Note: Record Flying ants under Pest 10
Termite
Thick waist
all four wings same length
wings much longer than body
antennae straight
Flying Ant
Very thin waist
front wings longer than hind wings
front wings about length of body
antennae elbowed
Termite
Flying Ant
Page 15
-------
Powderpost Beetle
Pest 8
Other Wood Destroying Insects
Powderpost beetles
Small red, brown, or black oblong beetles; damage visible in
wood as pinholes oozing sawdust
(size: adult l/8"-l/4")
Black Carpenter Ant
Carpenter ants
Large black (sometimes reddish) ants found both inside and
outside home; "sawdust" may indicate work area
(size: l/4"-l/2")
Carpenter Bee
Carpenter bees
Large; similar to bumble bee but with dark metallic abdomen;
bores 1/2" holes in unpainted wood
(size: to 3/4" long)
Page 17
-------
Pest 9
Fire Ants
Southern Fire Ant
Fire ants
Aggressive; red or black ants; painful sting; can build large
mounds; only found from Carolinas to Texas to Florida
(size: 1/4" or less)
Fire Ant Mound
Page 19
-------
Pest 10
Any Other Ants
Ant
Many species; live both indooors and outdoors; may become
pest when they invade houses or when they create mounds in
gardens or lawns
NOTE: Termites are not the same as flying ants. Certain life
stages and species of both termites and flying ants may have
wings and are often confused by the public.
Note: Record Termites under Pest 7
Flying ant
Very thin waist
front wings longer than hind wings
front wings about length of body
antennae elbowed
Termite
Thick waist
all four wings same length
wings much longer than body
antennae straight
Flying Ant
Termite
Page 21
-------
Scorpion
Pest 11
Scorpions
Scorpions
8 legs; painful sting; venom of some species can be dangerous;
common in South and Southwest
(size: to 3" long)
Page 23
-------
Pest 12
Bees, Hornets, Wasps
Bumble bees
Large; hairy; usually black & yellow
(size: adult to 1")"
Bumble Bee
Honey Bees
Yellowish brown; live in large colonies which may be problem
when built inside building walls; not likely to sting unless
handled
(size: worker 2/3")
Honey Bee
Hornets
Large wasps; dark colored; some with yellow-orange markings;
build large, oval paper nests hanging from trees
(size: adult to 11/2")
Bald-faced Hornet
Page 25
-------
Pest 12
Bees, Hornets, Wasps
i
Yellowjacket
Yellowjackets
Wasp; black with yellow stripes; attracted to soda and food
outdoors; often aggressive and may sting with little
provocation
(size: adult 1/2")
Paper Wasps
Paper wasps
Medium size wasps; dark colored often with yellow-orange
markings; build single layer paper combs hanging from
ceilings or rafters
(size: adult 1")
Mud Dauber Wasps
Mud dauber wasps
Build nests from mud on walls and rafters; usually avoid
humans
(size: adult 1")
page 26
-------
Mosquitoe and Larva
Pest 13
Mosquitoes
Mosquitoes
Numerous species; larvae (wigglers) develop in water; females
drink blood; some species can spread certain disease organisms
(size: 1/4")
Page 27
-------
House Fly
Deer fly
Pest 14
Flies, Gnats, Midges
Flies
Some produce painful bites (not stings) and may spread disease;
all flies only have 1 pair of wings (versus 2 pairs for bees); larvae
(maggots) have no legs; the term "gnat" refers to various tiny
flies
Note:Use Pest 19 (Plant-Chewing Insects) for Fruit/Vinegar
Files which are found on fruit, not on people or animals
Deer flies
Large flies; painful-bites; attack wild and domestic mammals,
man
(size: to 1/2")
House flies
Gray with stripes; live both indoors and outdoors; very common;
do not bite but may spread disease when they land on food
(size: adult 1/4")
Biting Midge
Black flies & biting midges (no-see-ums)
Small; painful bite; can be major nuisance in recreational areas
(size: adult to 1/8")
Page 29
-------
Pest15
Fleas
Fleas
Small, reddish brown to black; flattened sideways; strong
jumpers; drink blood
(size: 1/10")
Page 31
-------
Pest 16
Ticks, Chiggers
Tick
Ticks
8 legs; oval shaped; small head; drinks blood
(size varies from size of pin head to 1/4", larger when filled with
blood or eggs)
Chigger
Too small to see
Chiggers
Barely visible biting mites that cause severe itching in man and
animals
(size: 5/1000")
Page 33
-------
Pest 17
Spiders, Crickets, Sowbugs/Pillbugs, Millipedes, Centipedes
Spiders
8 legs; most produce silk but many do not spin webs; some bite,
a fewer are poisonous
(size: varies up to 1" or larger)
House Cricket
Crickets
Large hind legs; yellow- brown to black; may be noisy; often
invades houses, especially in Fall; may eat fabrics and fruits
(size: to 1")
Sowbug/Pillbug
Sowbugs/pillbugs
Gray; 7 pairs of legs; found in damp, dark places (often under
pots, rocks, in greenhouse); often roll into ball when disturbed
(size: 1/4"-3/4")
Page 35
-------
Pest 17
Spiders, Crickets, Sowbugs/Piilbugs, Millipedes, Centipedes
Millipedes
Millipedes
Up to 40-60 pairs of short legs (2 pairs per body segment);
slow moving; short antennae; usually found outside in dark,
damp places but may enter home
(size: 1/2" to 11/2" or larger)
Hm
Centipedes
Centipedes
Up to 15-30 pairs of long legs (1 pair per body segment); fast
moving; long antennae; prefers dark, damp places; may be
found in houses; may bite
(size: 1"-11/2" or larger)
page 36
-------
Pest 18
Soil-Dwelling Insects, Nematodes
Nematodes
Too small to See
Nematodes
Microscopic, worm-like; usually live in soil; feed on plant tissue
causing stress which may result in stunted and deformed growth
White grubs (grubworms)
General term that applies to larvae of various scarab beetles;
fleshy, whitish body with yellow to brown head; 6 true legs;
found in soil especially under turf
(size: to 1 1/2")
Use Pest 19 for Adults which are Beetles that Chew Foliage
Northern Mole Cricket
Mole crickets
Soil dwelling crickets with large paddles on front legs; light
brown to black; eats roots
(size: 1")
Page 37
-------
Pest 18
Soil-Dwelling Insects, Nematodes
Black Cutworm
Cutworms
Caterpillars that cut off young plant shoots at or below soil
surface; also climb to feed on foliage of larger plants; hide in
soil during day then emerge at night to feed
(size: larva 1" - 2")
Use Pest 19 for Caterpillars which are usually found on
Plants
Southern Corn Rootworm
Rootworms
Larva of various beetles (including cucumber beetles); feed on
roots or bore into stalks at soil line
(size: larva to 1/2")
Use Pest 19 for Adults which are Beetles that Chew
Foliage
Wireworm
Wireworms
Yellow to brown; "skin" is stiff, not soft; feed on roots of grass
and root vegetables; adults are click beetles (if placed on their
backs these beetles will snap body with a clicking sound to
pop into air and flip over)
(size: larva to 11/4")
Use Pest 19 for Adults which are Beetles that Chew
Foliage
page 38
-------
Pest 19
Plant Chewing Insects
Eastern Tent Caterpiller
Caterpillers
(such as tent caterpillar, tomato hornworm, sod webworm; also
gypsy moth, and (not shovvn)cabbage looper)
General term referring to the larva of butterflies and moths; 6
true legs at head end
Use Pest 18 for any Caterpillers that are Usually Found in
the Soil
Tent caterpillars
Hairy, black caterpillars with various blue, white, yellow, orange
markings; defoliate many trees, esp. apples and cherries; several
species spin tent in branches while others spin mat on trunk
(size: larva to 2", adult moth 1"-11/2")
Tomato Hornworm
Tomato Hornworm
Large caterpillar; green (usually) with white diagonal stripes and
black horn at rear
(size: larva to 4", adult moth 4"-5")
Sod Webworms
Sod webworms (lawn moths)
Larvae build silk tunnels above ground near base of turfgrass;
adults small tan to brown moths with wings held in pleated
peak over back
(size: larva to 3/4", adult moth 3/4"-l")
Page 39
-------
Pest 19
Plant Chewing Insects
Gypsy Moth Larva
Gypsy Moth
Hairy, dark caterpillars with blue and red spots; can occur in
huge numbers and defoliate trees over large areas; Northeast
US west to MI, south to VA and spreading
(size: larva to 2", adult 11/2"-2")
Squash Vine Borer
Borers
This term refers to two types of insects, both of which chew
plants:
A) the larva (caterpillars) of moths which feed inside stems,
trunks, or other plant parts of various plants (eg. squash,
peaches); may cause stem to wither, break, or may allow entry
for disease
(size: larva to 11/2", adult moth to 11/2")
Flatheaded Apple Tree Borer
B) the larva (grubs) of various beetles which feed on the inner
bark and sapwood of various trees and shrubs (eg. elm, apple,
palm)
(size: larva to 1", adult to 3/4")
page 40
-------
Pest 19
Plant Chewing Insects
Japanese Beetle
Colorado Potato Beetle
Cucumber Beetle
Beetles
(such as Japanese beetle, Colorado potato beetle, cucumber
beetle)
Large order of insects; beetles have a hard shell on their backs
covering a pair of wings underneath; beetles chew their food
Japanese beetles
Adults metallic green and bronze; feed on foliage, fruit,
flowers (esp. roses)
(size: adult 1/3'-1/2")
Use Pest 18 If Pest is Larval Stage (White Grub)
Colorado potato beetles
Adults striped yellow and black; larvae red or orange with
black spots on sides; both larva and adult eat foliage of
potatoes, eggplant, tomato
(size: adult 3/8", larva to 1/2")
Cucumber Beetle
both spotted and striped species; yellow and black; adults
attack melons, cucumbers, squash, and many ornamental
flowers
(size: adult l/4"-l/3")
Use Pest 18 If Pest is Larval Stage (Rootworm)
Page 41
-------
Pest 19
Plant Chewing Insects
Mediterranean Fruit Fly
Fruit flies
This term is often used for two different types of flies; both
are included in the plant chewing category:
A) Small; adult fly often dark with yellow markings; very
destructive; young maggots tunnel throughout flesh of
growing fruit; examples include Mediterranean fruit fly and
apple maggot
(size: larva & adult to 1/4")
Vinegar Fly
B) More properly called vinegar flies; tiny; common on
harvested fruit; reproduce quickly so often found in large
numbers
(size: adult 1/10")
Strawberry Weevil
Weevils (snout beetles, billbugs)
Family of beetles all with characteristic elongated snout;
attack flowers and fruit; larvae usually live in soil and feed on
roots
(size: l/10"-l/2")
Use Pest 18 if Pest Is Larval Stage In Soli
Use Pest 5 If Weevil is found in Stored Food
page 42
-------
Pest 19
Plant Chewing Insects
Leaf Mimer Damage
Leaf miners
Larvae of various flies, moths, beetles; feed between the upper
and lower surface of a leaf leaving light colored tunnels or
blotches visible on leaves
(size: larva tiny to 1/4")
Earwigs
Pinchers or forceps at tail; dark colored; bad odor when crushed;
usually outside but may wander into house; may bite or pinch
(size: to 1")
Grasshopper
Grasshoppers, locusts
Large hind legs; voracious plant eaters; many species
(size: to 2")
Page 43
-------
Pest 20
Plant-Sucking Insects and Mites
Squash Bug
True Bugs
In scientific use, this term refers to a specific order of insects;
bugs have wings which fold over their backs but do not have the
hard wing coverings seen in beetles; bugs use their piercing
mouthparts to suck juices from plant.
Squash Bugs
Adults dark brown; nymphs grayish; prefer squash and
pumpkins, will eat cucumbers, melons
(size: 3/4")
Chinch Bugs
Adults black and white with red legs and base of antennae;
young all red; attack turf and grains
(size: 1/8")
Chinch Bug
>
BoxelderBug
Boxelder Bug
Black and red; live outside but may enter house in large numbers
to hibernate; bad odor when crushed
(size: 1/2")
Page 45
-------
Pest 20
Plant-Sucking Insects and Mites
Aphids
Small, soft bodied, with or without wings; many species,
attack many plants; often seen on rose buds; may be green,
yellow, red, black; reproduce quickly, so often seen in large
numbers
(size: to 1/5")
Mites
Barely visible; 8 legs; often red but many other colors; damage
on leaves appears as tiny white or yellow dots; some produce
fine webs; common in hot, dry weather
(size: 1/100")
Scale Insects
Scale insects (scales, soft scales, armored
scales)
While some stages look like whiteflies, the stage most often
seen does not move and looks like a bump on a stem, twig, or
fruit; on some, a shell covers the back; colors include white,
gray, brown, black
(size: l/10"-l/4")
page 46
-------
Pest 20
Plant-Sucking Insects and Mites
Thrips
Barely visible; fringed wings; attack buds, flowers, fruit, foliage
(size: to 1/10")
Thrip
Psyllids (psylla)
Look like tiny cicadas; active
(size: adult to 1/8")
Pear Psyllids
Leafhoppers
Small; fast; characteristic triangular body shape; many species;
many colors, often with bright stripes or spots
(size: 1/8" -1/4")
Potato Leafhopper
Page 47
-------
Pest 20
Plant-Sucking Insects and Mites
Greenhouse Whitefly
Whiteflies
Small; winged adults fly; immatures attach to underside of
foliage; numerous; frequent pest of house and greenhouse
plants in North, also outside in South; discarded shells of
nymphs litter leaf surfaces
(size: to 1/8")
Mealybugs
Mealybugs
Small; sluggish; have powdery appearance, look like fluffs of
cotton
(size: to 1/4")
Green Stink Bug
Stink Bugs
Shield-shaped green or brown; attack peas, beans, fruit; bad
odor when crushed
(size: 1/2")
page 48
-------
Pest 23
Brush
Poison Ivy
Brush
General term referring to woody or semi-woody vines, shrubs,
small trees; usually perennial
Poison Ivy
Smooth, often shiny leaves in groups of 3; leaf edge smooth or
toothed; both bush and climbing forms
(size: leaf 3"; plant 6" to tall vine)
Honeysuckle
Honeysuckle
Woody vine with fragrant flowers in late spring to summer;
Mid-Atlantic to South only
(size: leaf 1"; plant large vine)
Kudzu
Kudzu
Large 3-lobed leaf; smothering vine; Southeast to Mid-Atlantic
only
(size: leaf 4"-6"; plant trails to 100')
Page 49
-------
Pest 23
Brush
Prairie Rose
Brier
General term referring to various thorny shrubs and vines
such as wild roses, blackberries, green brier
Mesquite
Mesquite
Shrub or small tree; pinnate compound leaf; Southwest only
(size: leaflet to 1/2", plant 3' - 36')
Smooth Sumac
Sumac
Shrub or small tree; pinnate compound leaf; fruit cluster
pyramid-shaped; fruit covered with reddish hairs
(size: leaflet to 4" long, plant 2* - 20')
page 50
-------
Pest 24
Grass-like Weeds
Crabgrass
Annual; characteristic finger-like flower head; forms clumps
(size: plant 6" - 12")
Yellow Foxtail
Foxtail
Characteristic bristly spike flower head (straight or drooping)
(size: plant 1' - 3')
Nutsedge
Nutsedge (nutgrass)
Stem triangular in cross-section; prefers moist to wet soil
(size: plant 1* - 2')
Page 51
-------
Pest 24
Grass-like Weeds
Wild Garlic
Wild garlic/onion
Tubular leaves; grows from bulb
(size: plant 1' - 3')
page 52
-------
Pest 25
Broadleaf Weeds
Dandelion
Dandelion
Toothed leaf in basal rosette; yellow flower; fluffy seedhead
(size: leaf 3" -18")
Broadleaf Plantain
Plantain
2 common forms each with characteristic flowerhead:
(A) Broad oval leaf
(leaf size: 4" wide by 8" long)
Buckhorn Plantain
(B) Long narrow leaf
(leaf size: 1" wide by 12" long)
Page 53
-------
Pest 25
Broadleaf Weeds
Clover
Clover
Leaves in groups of 3; globular flowerhead; many species
(size: plant 3" to 16")
Common Chickweed
Chickweed
Small pointed leaves; small white flowers; grows during cool
season
(size: leaf to 1/2"; plant sprawls 6" - 24")
Spotted Spurge
Spurge
Grows flat on ground; milky sap; small oval leaves often with
dark splotch in center
(size: leaf to 1/3"; plant sprawls to 12")
page 54
-------
Morningglory
Pest 25
Broadleaf Weeds
Morningglory
Leaf heart-shaped to ivy-like; showy, tubular flowers
(size: leaf 3"; plant vines to 15')
Thistle
Thistle
Spiny, toothed leaves; characteristic flower shape
(size: plant 1' -10')
Tumbleweed
Tumbleweed
Oval leaves; densely branched plant; when mature, stem breaks
at base and plant blows across ground often piling up at fences
or other obstructions
(size: leaf to 2 1/2", plant to 3')
Page 55
-------
Pest 25
Broadleaf Weeds
Henbit
Henbit
Square stem; leaves attached directly to upper stem; purple
flowers; grows during cool season
(size: leaf 1 /2"; plant 6"-12")
page 56
-------
Pest 27
Slugs, Snails
-$lugs arj^'Damage
Slugs and Snail and Damage
Slugs,Snails
Soft; slimy; leave silvery trails on sidewalks and foliage when
mucous dries; snails have shells, slugs do not have shells; eat
foliage
(size: 1/4" to several inches depending on species)
Page 57
-------
Pest 28
Birds
Birds
May be nuisance when they nest in or on buildings, eat seeds, or
eat crops
Blackbird
Starling
Page 59
-------
Pest 28
Birds
Sparrow
page 60
-------
House Mouse
Pest 29
Mice, Rats
Mice.rats
Rodents; found in fields and structures (especially where food
products are stored)
(size: 1"-12")
Page 61
-------
Pest 30
Bats
Bat
Bats
Nocturnal, winged mammals; eat insects, fruit; can be nuisance if
they nest in buildings
(size: body to 4" with wingspan to 14")
Page 63
-------
Pest 31
Other Mammals
Tree Squirrel
Squirrels
Both tree and ground dwelling species; tree types may nest in
buildings; holes of ground types are hazardous to livestock
(size: adult to 27" (tree) or 12" (ground))
Ground Squirrel
Eastern Mole
Moles
Smooth, short fur; large front claws; tunnel underground eating
insects and plant roots
(size: adult 5" to 8")
Page 65
-------
Pest 31
Other Mammals
Striped Skunk
Skunks
Nocturnal; usually eat insects and small rodents; nuisance
when eat poultry or vegetables, excavate under buildings, or
spray people or pets
(size: adult to 18")
NJEVy >
Black-tailed Prairie Dog
Prairie Dogs
Large, stocky rodents; coarse brown fur; eat vegetation;
burrows create rough ground surface
(size: adults to 18")
Woodchuck
Woodchucks (goundhogs.marmots)
Large, stock, short legs; eat plants; dig underground dens
(size: adult to 20")
page 66
-------
Pest 31
Other Mammals
Eastern Cottontail
Rabbits
Large ears; powerful hind legs; eat vegetation; two main groups:
cottontails and jackrabbits (hares)
(size: adult to 22")
Page 67
-------
Sources for Illustrations
Leaflet 21152,1980, Weed Control in the Home Landscape.
Cooperative Extension Service, Univ. of California, Berkeley,
CA.
Ball, Jeff, 1988, Rodale's Garden Problem Solver. Rodale Press,
Emmaus, PA.
Fichter, George S., et al, 1966, Insect Pests. Golden Press, NY.
Davidson, Ralph H. and William F. Lyon, 1979, Insect Pests. John
Wiley k Sons, NY.
ENT-29,1985, Household Insect Control. Cooperative Extension
Service, Univ. of Kentucky, Lexington, KY.
NA Swan, Lester A. and Charles S. Papp, 1972, The Common
Insects of North America. Harper & Row, NY.
Public Health Service Publication 1955,1969, Pictorial Keys to
Arthropods. Reptiles. Birds and Mammals of Public Health
Significance. US Public Health Service of the US Dept. of Health,
Education, and Welfare, Government Printing Office, Washing-
ton, DC
Prevention and Control of Wildlife Damage. 1983, Cooperative
Extension Service, Univ. of Nebraska, Lincoln, NE.
Bulletin 512 (Agdex 670), 1987, Pesticides for Household and
Structural Pests. Cooperative Extension Service, Ohio State
Univ., Columbus, OH.
Bennett, Gary W., et al, 1988. Pest Control Operations. Edgell
Communications, Duluth, MN
Training Manual for the Structural Pesticide Applicator. 1976,
Environmental Protection Agency/Office of Pesticide Programs,
Washington, DC.
Circular 718,1974, Weeds of the North Central States. Univ. of
Illinois, Urbana, IL
Hanson, A. A. and F. V. Juska, 1969, Turfgrass Science. American
Society of Agronomy, Madison, WI.
Page 69
-------
Sources for Illustrations
Fact Sheet 242,1988, The Gvpsv Moth and the Homeowner.
Cooperative Extension Service, Univ. of Maryland, College
Park, MD.
WL 2. Moles. Cooperative Extension Service, Univ. of Ver-
mont, Burlington, VT.
G3096- Bats: Information for Wisconsin Homeowners.
Cooperative Extension Service, Univ. of Wisconsin, Madison,
WL
Agricultural Handbook 366,1970, Selected Weed of the
United States. US Department of Agriculture, Government
Printing Office, Washington, DC.
page 70
-------
APPENDIX J
ADVANCE MAILING MATERIALS
-------
------- |