svEPA
United States Environmental Monitoring System!; EPA-600/4-84-031
Environmental Protection Laboratory April 1984
Agency Research Triangle Park NC 27711
Research and Development
Study of Carbon
Monoxide Exposure of
Residents of
Washington, DC and
Denver, Colorado
-------
-------
EPA-600/4-84-031
August 1984
STUDY OF CARBON MONOXIDE EXPOSURE OF RESIDENTS
OF WASHINGTON, DC AND DENVER, COLORADO
By
T.D. Hartwell, C.A. Clayton, R.M. Ritchie, R.W.! Whitmore,
H;S. Zelon, S.M. Jones, and D.A. Whitehurst
Contract No. 68-02-3679
Project Officer:
Gerald G. Akland
Environmental Monitoring System Laboratory
U.S. Environmental Protection Agency
Research Triangle Park, North Carolina 27711
ENVIRONMENTAL MONITORING SYSTEMS LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT ' •
U.S. ENVIRONMENTAL PROTECTION AGENCY
Research Tri.angle Park, NC 27711
-------
NOTICE
This document has been reviewed in accordance with
U.S. Environmental Protection Agency policy and•'
approved for publication. Mention of trade names
or commercial products does not constitute endorse-
ment or recommendation for use.
-------
iii
FOREWARD
Measurement and monitoring research efforts are designed to anticipate
potential environmental problems, to support regulatory actions by develop-
ing an in-depth understanding of the nature and processes that impact health
and the ecology, to provide innovative means of monitoring compliance with
regulations and to evaluate the effectiveness of health and environmental
protection efforts through the monitoring of long-term, trends. The Environ-
mental Monitoring Systems Laboratory, Research Triangle Park, North Carolina,
has the responsibility for: assessment of environmental monitoring tech-
nology and systems; implementation of agency-wide quality assurance programs
for air pollution measurement systems; and supplying technical support to
other groups in the Agency including the Office of Air, Noise and Radiation,
the Office of Toxic Substances and the Office of Enforcement.
-
This document is a report of the development and application of personal
exposure methodology for carbon monoxide to the residents of Washington, D.C.,
..during the winter of 1982-83. This report discusses the methodology used in
the study and the results of applying this methodology,.
Thomas &. Hauser ;.
Director |
Environmental Monitoring
Systems Laboratory
-------
iv
TABLE OF CONTENTS
Page
Disclaimer ' ii
• • • ,' /
Forward * iii
.._••:!:
List of Tables '...'. viii
List of Figures ^.. ' xii
.-> ---.,,
List of Exhibits xiv
Acknowledgements ...'........ xv
Abstract .................... xvi
1. INTRODUCTION '." 1
2. SUMMARY OF STUDY DESIGN AND PROCEDURES 4
3. SUMMARY OF STUDY RESULTS AND CONCLUSIONS 10
4. RECOMMENDATIONS 16
4.1 Design Recommendations 16
4.2 Recommendations Concerning Field Operations
and Data Collection 17
4.3 Recommendations for Further Statistical Analysis ... 21
5. METHODS AND PROCEDURES 23
5.1 Survey Design 23
5.1.1 Selection of First-Stage Sampling Units
(FSUs) 25
5.1.2 Selection of Second-Stage Sampling Units
(SSUs) 29
5.1.2.1 Selection of SSUs Within FSUs
With Donnelley Listings 29
5.1.2.2 Selection of SSUs Within FSUs
With No Donnelley Listings 35
5.1.3 Screening Response 36
5.1.4 Selection of the Third Stage Sample 41
5.1.5 Third Stage Response 46
5.1.6 Variance Estimation and Screener Analysis ... 46
-------
Table of Contents (continued)
5.2 Survey Activities 48
5.2.1 Public Relations Efforts in Denver 48
5.2.2 Data Collection Instrument Development
and Approval 49
5.2.3 Phase I - Household Screening Survey 49
5.2.3.1 Computer Assisted Telephone
Interviewing (CATI) System 50
5.2.3.2 Telephone Interviewers 50
5.2.3.3 Interviewing 51
5.2.4 Phase II - Washington, DC Area Survey 54
5.2.4.1 Telephone Interviewing 54
5.2.4.2 Final Document Preparation 57
5.2.4.3 Protection of Human Subjects 57
5.2.4.4 Field Staff Recruitment 57
5.2.4.5 Training the Field Staff; 58
5.2.4.6 Field Office 58
5.2.4.7 Special Field Studies 58
5.2.4.7.1 Missed Housing Units 59
5.2.4.7.2 Segments With No
Donnelley Listing 59
5.2.4.7.3 No Previous Telephone
Contact 59
5.2.4.8 Regular Field Assignments 59
5.2.4.9 Breath Sampling .| 61
5.3 Field Measurements and Quality Assurance! 62
5.3.1 Description of the Ambient Monitors 62
5.3.2 Description and Verification of the Field
Standards -...'... 64
5.3.3 Preparation of CO Monitors for the
Acquisition of a CO Exposure Sample 65
5.3.4 Analysis Method for Carbon Monoxide
in Breath .' 73
5.3.4.1 Description of Method ..!.. 73
5.3.4.2 Instrument Noise ;.. 74
5.3.4.3 Instrumental Response Time 74
5.3.4.4 Sample Bags - Recovery Study 77
5.3.4.5 Sample Contamination from CBH Bags . 77
5.3.4.6 Effect of Various Parameters on
Breath CO Measurement . i 79
5.3.4.7 Interference Due to Plastic
Mouthpiece .. ,, 83
5.3.4.8 Effect of Concentrated Organic
Compounds on Monitor and Prefilter
Performance 83
5.3.4.9 Method Precision 84
5.3.4.10 Analysis Procedure Used During
Field Sampling 84
-------
vi
Table of Contents (continued)
5.4 Data File Creation and Descriptions 86
5.4.1 Descriptions of Raw Data Files 86
5.4.2 Creation of Analysis Files 97
5.4.2.1 Creation of the Basic Analysis
File (BAF) 97
5.4.2.2 Creation of the Activity Analysis
File (AAF) 102
5.4.2.3 Creation of the Duplicate
Measurement File (DMF) 110
6. RESULTS AND DISCUSSION HI
6.1 Survey Design Results Ill
6.1.1 Household Screener Statistical Analysis ..... Ill
6.1.2 Personal Item Statistical Analysis 112
6.1.3 Introduction to Sample Design Results 125
6.1.4 Use of Geographically Classified Telephone
Directory Listings in Association With
Standard Area Household Sampling
Techniques 125
6.1.5 Lead Letter Results . 134
6.1.6 Sampling Person-Days 134
6.2 Field Survey Activities 136
6.2.1 Survey Post-Field Activities 136
6.2.2 Post Data Collection Discussions 139
6.3 Field Measurements and Quality Assurance 141
6.3.1 Field Measurement Activities 141
6.3.1.1 Personal Exposure Sampling 141
6.3.1.2 Analysis of CO Levels in
Respondent Breath Samples 144
6.3.1.3 Fixed Site CO Data 144
6.3.2 Problems With Monitors 146
6.3.2.1 The COED-1 (GE/Magus) Monitor . . 146
6.3.2.2 The GE/HP Monitor 158
6.3.3 Quality Assurance Activities 161
6.3.3.1 Quality Assurance Project Plan ..... 161
6.3.3.2 External (EPA-Conducted) QA
Systems Audits 161
6.3.3.3 Internal (RTI-Conducted) QC Audit .. 161
6.3.3.4 Multipoint Calibrations to Assess
Monitor Linearity „ 161
6.3.3.5 Monitor Stability Over the Course
of the Study 162
6.3.3.6 Assessment of Measurement
Precision and Accuracy 166
-------
VX1
Table of Contents (continued)
6.4 Results of Statistical Analysis . 175
6.4.1 Analysis of Hourly CO Exposure Data 178
6.4.2 Analysis of CO Breath Measurements 193
6.4.3 Analysis of Activities and Associated CO
Exposures 193
6.4.3.1 Activity and Location Patterns 195
6.4.3.2 Carbon Monoxide Exposures 197
6.4.4 Analysis of Measurement Variability 204
- • I
7. REFERENCES , 209
APPENDIX A: Maps of Target Areas !
APPENDIX B: Phase II Computer Model Input Questionnaire
APPENDIX C: Field Interviewer's Manual
APPENDIX D: Table of Contents of the OMB Package
APPENDIX E: Telephone Survey Unit Specifications ;
APPENDIX F: Phase II Telephone Interviewer's Manual
i
APPENDIX G: Materials on Protection of Human Subjects
i
,
APPENDIX H: Standard Operating Procedure for Collecting and
, Sampling Alveolar Carbon Monoxide j
APPENDIX I: Quality Assurance Plan
APPENDIX J: Results of High and Low CO Exposure Days
APPENDIX K: High Occupational Exposure Categories
i
APPENDIX L: Post-Field Work Questionnaire
I
APPENDIX M: Comparison of COED-1 and Fixed Site Monitoring Data
-------
VX11
LIST OF TABLES
Number Title Page
5.1.1 Distribution of Initial Telephone Screening
Final Result Codes 32
5.1.2 Distribution of Final Result Codes for Denver
Screening Sample 37
5.1.3 Distribution of Final Result Codes for
Washington Screening Sample „, 39
5,1.4 Screening Response Rates 40
5.1.5 Third Stage Sample Allocation for the Denver
Sample 42
5.1.6 Third Stage Sample Allocation for the
Washington Sample , 43
5.1.7 Distribution of Final Result Codes for
Individuals Selected for CO Monitoring in ......
. Washington , 47
5.3.1 Instrument Response Times in Minutes ,.. ,76
5.3.2 Loss of CO From Fenwal Sampling Bags 78
5.3.3 Loss of CO From CHB Sampling Bags 78
5.3.4 Effect of Filter on Measured CO 80
5.3.5 Effect of Humid Air on Measured CO 80
5.3.6 Effect of Storage in Sampling Bags on CO
Measurements at 3 ppm CO 80
5.3.7 Effect of Storage in Sampling Bags on CO
Measurements at 7 ppm CO 81
5.3.8 Effect of Storage in Sampling Bags on CO
Measurements at 15 ppm CO 81
5.3.9 The Effect of Ethanol on Measured CO 82
5.3.10 Breath Measurements for Non-Smoking Subjects 85
5.4.1 Number of Routine Samples With Valid Hourly
CO Values, By Hour of Day 99
-------
List of Tables (cont'd)
Number Title i Page
5.4.2
5.4.3
6.1.1
6.1.2
6'.1.3
6.1.4
6.1.5
6.1.6
6.1.7
6.1.8
6.1.9
6.1.10
6.1.11
6.1.12
6.1.13
6.1.14
6.1.15
6.1.16
Distribution of the Number of Hourly CO
Values Per Sample '.
Distribution of Sampling Dates, by Month and
Estimated Number of Households Using a Fireplace . . .
Estimated Number of Households Using a Wood Stove . .
Estimated Number of Households Using a Gas Furnace .
Estimated Number of Households Using a Gas or
Estimated Number of Households Using a G,as
Estimated Number of Households Using a Gas Hot
Water Heater
Estimated Number of Households Using a Gas
Clothes Dryer
Estimated Number of Households Using Other Gas
Appliances
Estimated Number of Households Having an Attached
Garage or Sharing a Multi-Family Garage
Estimated Sex Distribution .;
The Sex Distribution According to the 1980 Census . .
Estimated Age Distribution - Categorized
According to the 1980 Census Definitions
Age Distribution According to the 1980 Census
Estimated Distribution of Relationship to Head
Estimated Distribution of Persons 13 Years and
Estimated Distribution of Persons 13 Ye^rs or
Older Who Work Either Full or Part Time ... '
100
100
113
113
114
114
115
115
116
116
117
119
119
120
121
122
123
126
-------
List of Tables (cont'd)
Number
Title
6.1.17 Estimated Distribution of Persons 13 Years or
Older Who Travel Anywhere at Least 3 Times Per
Week 126
6.1.18 Estimated Distribution of Amount of Time Spent
Traveling One Way at Least 3 Times Per Week for
Persons 13 Years or Older 127
6.1.19 Results of Missed HU Checks 129
6.3.1 Statistics on Ambient Sampling Waves 143
6.3.2 Results of Respondent Breath Analyses (ppm) 145
6.3.3 Site Characteristics of Washington Carbon
Monoxide Monitors Operating During Study 147
6.3.4 Summary Statistics for Hourly Average Carbon
Monoxide Values Reported by Washington Monitoring
Sites Between November 8, 1982 and February 25,
1983 150
6.3.5 Date and Time of Maximum Hourly Average Carbon
Monoxide Value 151
6.3.6 Summary Statistics for Daily Maximum 1-Hour
Carbon Monoxide Values Reported by Washington
Monitoring Sites Between November 8, 1982 and
February 25, 1983 152
6.3.7 Summary Statistics for Daily Maximum 8-Hour
Carbon Monoxide Values Reported by Washington
Monitoring Sites Between November 8, 1982 and
February 25, 1983 153
6.4.1 Estimates of Mean Population CO Exposure Levels
(ppm) — Diurnal Patterns by Time of Week and
Type of Day „ 179
6.4.2 Summary of Maximum Hourly CO Concentration Data .... 182
6.4.3 Summary of Maximum 8-Hour CO Concentration Data .... 187
6.4.4 Summary of Mean Hourly CO Concentration Data ....... 192
6.4.5 Summary of Breath CO Concentration Data ............ 194
-------
XX
List of Tables (cont'd)
Number Title
6.4.6 Summary of Population Estimates Relating to
, the Number of Individuals Involved In, and
Amount of Time Spent In, Various Types of
Activities . .................. • ........ ! ............. 196
6.4.7 Summary of Population Estimates Relating to
the Number of Individuals Exposed To, and
Amount of Time Spent In, Various Types of
Environments .......................... • • • ..........
6.4.8 Summary of CO Exposure Levels, by Type .'^of
Activity ......... * ----- ................ .............. 199
6.4.9 Summary of CO Exposure Levels, By Type . of
Activity — Ranked According to Mean CO Level ...... 200
6.4.10 Summary of CO Exposure Levels, by Type of
Environment . •
-------
xii
LIST OF FIGURES
Number
Title
Page
5.1.1 Selection of First-Stage Units and the Initial
Sample of Donnelley Listings 28
5.1.2 Denver CO Sample Protocol 33
5.1.3 Washington CO Sample Protocol 34
5.2.1 Table of Contents of the Telephone Interviewer's
Manual (CO Exposure Study) 52
5.2.2 Final Telephone Interviewing Status Report -
Phase I Screening (Washington, DC and Denver,
Colorado) 53
5.2.3 Telephone Response Rates 56
5.3.1 COED-1 Monitor Status Sheet 69
5.3.2 GE/HP Monitor Status Sheet .' 70
5.3.3 Field Data Sheet, Side 1 71
5.3.4 Field Data Sheet, Side 2 72
5.3.5 Instrument Noise 75
5.3.6 Breath Sample Data Sheet, Side 1 , 87
5.3.7 Breath Sample Data Sheet, Side 2 88
6.2.1 Carbon Monoxide Exposure Batch Header Sheet 138
6.3.1 Locations of Fixed-Site Monitors 149
6.3.2 Response Levels 163
6.3.3 Monitor Battery Voltages (volts) 164
6.3.4 Flow Rate 165
6.3.5 Washington, D.C. Personal CO Exposure Project
PEM vs. Fixed Site Monitor (FSM) Comparison ........ 168
6.3.6 Washington, D.C. Personal CO Exposure Project
PEM vs. Fixed Site Monitor (FSM) Comparison:
Plot of PEMCONC*FSM 170
-------
xiii
List of Figures (cont'd)
Number Title ; ; Page
" "" n'nuHjin — ^ . | .
6.3.7 Washington, B.C. Personal CO Exposure Project
PEM vs. Fixed Site Monitor (FSM) Comparison:
PID=7000417 171
i
6.3.8 Washington, B.C. Personal CO Exposure Project
PEM vs. Fixed Site Monitor (FSM) Comparison:
PID=7000524 ! 172
6.3.9 Washington, B.C. Personal CO Exposure Project
PEM vs. Fixed Site Monitor (FSM) Comparison
for Concentration Bata £ 1.0 ppm „ 174
6.4.1 Average CO Exposure Levels, By Hour of Day 180
6.4.2 Maximum Hourly CO Concentrations by Occupational
Exposure, Washington, B.C 183
6.4.3 Maximum Hourly CO Concentrations for Weekdays
and Weekend Bays, Washington, B. C i. 184
6.4.4 Maximum Hourly CO Concentrations for Selected
Commuting Statuses, Washington, D.C. 185
6.4.5 Maximum 8-Hour CO Concentrations by Occupational
Exposure, Washington, B.C 188
.
6.4.6 Maximum 8-Hour CO Concentrations for Weekdays
and Weekend Bays, Washington, B.C 189
i
6.4.7 Maximum 8-Hour CO Concentrations for Selected
Commuting Statuses, Washington, B.C. ..j 190
6.4.8 Plot of STBCONC * MEANCONC ^ 205
6.4.9 Bistribution of Standard Beviations of
Replicate Observations 206
-------
xiv
LIST OF EXHIBITS
Number
5.4.1
5.4.2
5.4.3
5.4.4
Title
Contents of File A (PEM Data)
Description of Codes for Variables Appearing
In File B (Activity Diary Data)
Contents of File D (Questionnaire Data)
Contents of Basic Analysis File
Page
. . 90
92
94
103
-------
XV
ACKNOWLEDGEMENTS
The authors would like to sincerely thank the following individuals
for their assistance in carrying out the Carbon Monoxide Study: Dan
Ward, John Sokash, and Jon Lodge of RTI who ran the field laboratory in
Washington, D.C. and did an excellent job in maintenance and repair of
the CO PEMs under severe time constraints; Velma Brock, Field Supervisor
for the Washington, D.C. data collection effort; Martin Rosenzweig,
Cliff Decker, Tom Virag, and Barbara Alexander of RTI and Robert Jungers
of EPA who helped in many phases of the study including study design,
data analysis, quality control, and project management; Lanny Piper and
Janice Whelan who programmed RTI's computer assisted telephone inter-
viewing system for the study; and Linda Sheldon who managed the experi-
ments used to establish the breath analysis procedure,. Without the aid
of these individuals it would not have been possible to successfully
carry out the project and we certainly appreciate their time and effort.
Finally, we wish to express our appreciation to Ms. Carol Johnson for
her excellent work in typing the manuscript. •
-------
XVI
ABSTRACT
This report describes a study funded by the EPA and conducted by
the Research Triangle Institute in 1982 and 1983 to evaluate methodology
for collecting representative personal exposure monitoring (PEM) CO and
corresponding activity data in an urbanized area. This involved tele-
phone screening of households and sample selection of respondents in the
metropolitan areas in and around Denver, Colorado and Washington, D.C.
Data on CO breath levels were also collected in Washington, D.C. (PEDCo
Environmental conducted the field work in Denver.) The target popula-
tion in both cities consisted of the non-institutionalized, non-smoking
adults (ages 18 to 70) of these metropolitan areas. The data collected
in the field were edited and appropriately weighted to produce CO
exposure estimates for the target population. These estimates included
average maximum hourly and 8-hour CO levels, and average CO levels for
various activities and locations.
Based on the experience gained during the study, the methodology
developed, with some modifications, may be used effectively in other
areas of the country for collecting PEM data. These modifications
should make the methodology more cost effective, improve the response
rate, and lead to more accuracy activity information.
Estimates of CO exposure for the winter of 1982-83 in Washington,
D.C. were obtained using the data base constructed from the raw CO
levels by activity data which consisted of hourly CO values on 712
respondents, activity patterns and corresponding CO levels on 705
respondents, and CO breath measurements corresponding to the PEM CO data
on 659 respondents. The size of the target population was estimated to
be 1.22 million individuals.
The weighted average maximum hourljr PEM CO level in Washington,
D.C. was 6.74 ppm. The average maximum 8-hour CO level was 2.79 ppm.
The percentage of the population with maximum hourly CO values over the
35 ppm CO standard was estimated to be 1.28 percent while the percentage
with an 8-hour maximum over the 9 ppm standard was 3.9 percent.
Estimates were also made for subgroups of the population. Persons
in high-exposure occupations (about 4.6% of the total population)
generally exhibited higher CO exposure levels: it was estimated that
about 24% of this high-exposure group had 1-hour CO exposures above the
35 ppm standard and that about 28% exceeded the 8-hour standard. It was
also shown that CO levels were generally higher for commuters, especially
for those with larger amounts of travel.
Breath CO levels (taken at the end of the sampling periods, usually
in the respondents' homes) for the adult non-smoking population in
Washington averaged 5.12 ppm. Slightly higher levels were observed for
persons with high occupational exposures and for persons with large
amounts of travel.
By combining PEM data with data from individuals' diaries, estimates
of both CO levels and time durations for various activities and personal
-------
XV13.
environments were made. For example, the activities "in parking garage
or parking lot" and "travel, transit" had the highest average CO concen-
trations (6.93 ppm, and 4.51 ppm, respectively) while "sleeping" had an
estimated CO concentration of only .85 ppm. Among the environments
considered, the three with the highest average CO concentrations were
"indoor parking garage", "outdoor parking area", and '"in transit-car".
The average levels for these environments were 10.36, 4.67, and 5.05
ppm, respectively. !
-
Variation from duplicate hourly PEM measurements under field
condition were also analyzed. An analysis of variance of this data
which considered person-to-person, hour-to-hour, and measurement varia-
tion indicated that about 5 to 6 percent of total variation among the
hourly duplicate readings was due to deviation in measurements made by
two PEMs at the same hour for the same person.
-------
-------
1. INTRODUCTION
As the control of emissions increases, the burden of proof on EPA
to show that a particular level of emission control is justified also
increases. It has become more and more important to show that a given
level of control is justified for each air pollutant, with the relative
risk of public health approximately comparable for each pollutant
controlled.
I
A critical factor in determining the degree of risk to the popula-
tion is the exposure of members of the population. In the past, moni-
toring of airborne pollutants has necessarily been based on the assump-
tion that fixed-site monitoring is representative; of concentrations
surrounding the site, since monitoring techniques were generally not
developed for determining personal exposures. Then to obtain estimates
of population exposure, techniques such as computer simulation or
overlaying isopleths of pollution concentrations meiasured at fixed sites
on population density maps have been used. For some pollutants, these
techniques may be reasonable approximations; however, recent work has
i
shown that many pollutant concentrations are not homogeneous and that
activity patterns play an important role in an individual's actual
exposure. Therefore, data from ambient fixed sites often differ signif-
icantly from the concentrations with which people actually come into
contact. I
As EPA engages in modifications — or "fine tuning" — of emission
standards, it becomes necessary to focus more attention on those compo-
nents for which data are most lacking. The ultimate goal of the present
research program was to develop a methodology to determine the public's
exposure to air pollutants with known precision and accuracy.
A wide variety of air pollutants could have been selected for
'
study. For example in the gas phase, carbon monoxide1., nitrogen oxides,
and hydrocarbons have been of concern either because they cause adverse
health effects or because they are precursors in the formation of air
-------
pollutants that cause adverse health effects (e.g., ozone). In the
solid phase, particulates and their associated organics have been
targeted for control.
Carbon monoxide was selected for primary emphasis in the current
study because:
- Accurate and portable field tested instruments now are avail-
able for CO (Wallace & Ott, 1982).
- Most of the CO to which the public is exposed can be attri-
buted to motor vehicles.
- It appears that CO is a good "indicator" (i.e., surrogate)
pollutant for estimating exposures to several other motor
vehicle pollutants of interest.
Because CO is a nonreactive air pollutant, it is simpler to
treat analytically.
- The health effects of CO are reasonably well documented, and
NAAQS based on these effects have been promulgated.
- Considerable data exist showing that CO varies spatially and
that many locations in cities have concentrations that differ
from those reported at fixed air monitoring stations.
Thus, RTI and EPA formulated a study plan to develop and field test
a population exposure methodology using CO while making sure that the
methodology was broad enough to accommodate other pollutants of concern.
The specific objectives of this study were the following:
- To develop a methodology for measuring the distribution of
carbon monoxide (CO) exposures of a representative population
of an urban area for assessment of the risk to the population.
- To test, evaluate, and validate this methodology by employing
it in the execution of pilot field studies in Denver,
Colorado, and in Washington, DC.
- To -obtain an activity-pattern data base related to CO
exposures.
The study was carried out in Washington, DC and Denver, Colorado
during the winter of 1982-83 (the period of the year with maximum
ambient CO concentrations). The population exposure profile was deter-
mined by direct measurement of CO with personal exposure monitors (PEMs)
-2-
-------
through the use of statistical inference from the^statistically drawn
I
sample. The study provided sufficient data to determine exposure as a
function of concentrations within significant micrpenvironments (home,
in-transit, work, and leisure) and individual activity patterns.
The following report describes in detail the activities involved in
this study and presents the results and recommendations evolving from
the study. It is extremely important to note that! the study not only
developed and tested methodology for measuring the distribution of CO in
an urban area but also produced direct estimates of CO exposure that
apply to two large metropolitan areas. In addition, a very important
product of this work is a unique and valuable datji base on individual
exposures to CO and the corresponding activities that led to these
exposures.
-3-
-------
SUMMARY OF STUDY DESIGN AND PROCEDURES
The study conducted by RTI was to evaluate methodology for collect-
ing representative personal exposure monitoring (PEM) CO and correspond-
ing activity data in an urbanized area. This involved telephone screen-
ing of households and sample selection of respondents in the metropolitan
areas in and around Denver,, Colorado and Washington, DC, and collection
of PEM CO and activity data from respondents in Washington, DC. Data on
CO breath levels were also collected in Washington, DC. (PEDCo Environ-
mental conducted the field work in Denver.) The target population in
both cities consisted of the non-institutionalized, non-smoking adults
(ages 18 to 70) of these metropolitan areas. The data collected in the
field were edited and appropriately weighted to produce CO exposure
estimates for the target population. These estimates included average
maximum hourly and 8-hour CO levels, and average CO levels for various
activities and locations.
A probability sample of the target population was selected in both
cities. This sample was a stratified, three-stage, probability-based
design. Area sample segments defined by Census geographic variables
were selected at the first stage of sampling. Households were selected
at the second stage, and a household member was administered a short
screening interview covering all household members to identify indivi-
duals with characteristics believed to be postively correlated with CO
exposure. Thus, household members with these characteristics could be
oversampled in the third stage. Donnelley Market Corporation listings
were used to help select households for the screening interview. The
third stage sample was a stratified sample of screened eligible indi-
viduals (i.e., non-smoking, aged 18 to 70). The individuals in the
third stage sample were administered a Computer Model Input Question-
naire and were asked to carry a personal CO monitor and an Activity
Diary for 24 or 48 hours (for Washington and Denver, respectively). A
breath sample was also requested from these individuals and they were
-------
asked to fill out a Household (Study) Questionnaire. The third stage
sample design also allocated individuals to specific days within the
sampling period. A detailed discussion of the sample design is given in
Section 5.1 and in Whitmore, et al. [1983a]. :
To carry out the sample design, RTI developed the data collection
instruments and worked with EPA in obtaining OMB approval for the study.
An initial telephone screening was carried out in both Denver and
Washington, DC by using RTI's Computer Assisted Telephone Interviewing
(CATI) system. This initial screening was supplemented by limited field
screening in both sites. Specific information collected during this
interview included: time spent in regular commuting and smoking status
of each household member, as well as presence of Jgas appliances and
attached garages in their residences. After the initial screening and
the initial selection of potential participants,;another telephone
interview was conducted. The purpose of this call was to contact the
selected individual to further explain the study and attempt to enroll
him(her) into the study. If the individual agreed to be part of the
study, an appointment was established for a field [interview. In addi-
tion, during this call, a Computer Model Input Questionnaire was admin-
istered which collected additional data on commuting patterns, demogra-
phics of household members, and household characteristics.
Finally, participating individuals were met at their home or other
convenient location and given all study materials. These participants
carried both a PEM for the 24 hours of their participation and an
Activity Diary in which to record a description of their activities.
Participants were requested to push a button on their PEM every time
they changed activities and to record descriptions of the new activities
in their diaries. In addition, for a small sample of participants, a
GE/HP PEM was used which allowed the participant to also enter an
activity code into the monitor (see Section 5.3). Participants were
also asked to complete a self-administered Household Questionnaire which
,
provided information on themselves and on their home and work environ-
ments. The telephone screening and sample selection of individuals for
both Denver and Washington were carried out by RTI as was the field work
in Washington.
-5-
-------
The results of the telephone screening and field activities for the
study are described in detail in Sections 5.1 through 5.3. Briefly,
8643 household screenings were attempted by RTI in Washington, DC and
4987 were attempted in Denver, Colorado. The successful screening rates
were 75.8 percent in Washington and 70.4 percent in Denver. From these
telephone and field screenings, 5418 eligible respondents were
identified in Washington and 2232 in Denver. From this population of
eligibles, 1987 individuals were selected for participation (i.e., to
carry a PEM) in Washington and 1139 in Denver. Of these selected
individuals, 58 percent actually scheduled appointments to carry a PEM
in Washington. Finally, 35.8 percent of the individuals in Washington
selected to participate contributed usable CO monitor data. This
represented 712 sample respondents. Instrument failure was one of the
major reasons for the low response rate. Specifically, CO data was not
collected or was unusable for analysis purposes for 232 respondents (22%
loss rate) due to monitor failure or malfunction. Usable CO breath data
corresponding to the usable CO monitor data was collected on 659 sample
respondents.
In order to successfully implement the study in Washington, DC, a
field office/laboratory was established in the offices of the Metropoli-
tan DC Council of Governments. This office was used for several purposes
including supervision of field staff, storage of supplies, maintenance
of records, allocation of field assignments, and maintenance and repair
of the PEMs. This office was visited twice nightly by all interviewers
to receive PEMs and data collection forms for that evening and for
return of completed study materials including the PEMs used the previous
24 hours. All calibrations of the PEMs during the study were carried
out in this field laboratory. In addition to the field supervisor for
the interviewers, the field laboratory was staffed with two full-time
technicians working seven days per week throughout the study. A detail-
ed description of the PEMs (COED-ls and GE/HPs) used in this study and
the extensive daily technical support that they required is given in
Section 5.3.
As mentioned above, breath samples were collected from respondents
during the study. This required RTI to evaluate a method for collecting
-6-
-------
and measuring alveolar CO. This evaluation is described in detail in
Section 5.3.4. The method essentially required each respondent to blow
into a sample bag at the end of his 24 hour sampling period. This
Sealed bag was;then returned to the field laboratory for CO analysis.
Throughout the field work, a quality control and assurance program
was maintained for the sampling and analysis procedures employed (see
Section 5.3 and 6.3). This included using field standards to calibrate
all the CO monitors. The monitors were subject to calibration (two-
point, zero/span) before they were put in the field and 24 hours later
when they were returned from the field. The comparison of the two
calibration curves was used to assign validity codes to the PEM data.
Other quality control procedures employed were: a ten percent check of
data transcribed from monitor memory to field data sheets; monitoring
control charts on each monitor describing the course of differences
between pre-sample and post-sample span, zero, battery voltage, and flow
rate values; collecting duplicate colocated samples for the purpose of
characterizing monitor precision; performing external and internal QA
and QC audits; performing multipoint calibrations! to assess monitor
linearity during the study; and obtaining duplicate breath samples from
respondents. The results of these extensive quality control and quality
assurance procedures are given in Section 6.3.
After the field work was completed, the data were returned to RTI
where detailed editing of the data was carried out by RTI editors. The
data were then entered into computer files using RTI's mini-computer
data base entry system. All data were keyed and then 100 percent
key-verified. Extensive machine editing was carried out which resulted
in identifying many computer records which required further manual
editing. The process of editing the computer files took extensive staff
time. In particular, checking the consistency of the PEM data with the
diary data was a time consuming process. i
Sampling weights were computed according to prescribed formulas
(see Section 5.1). This involved extensive computations so that the
weights could be used to draw inferences to the target populations. The
sampling weights were then put on a computer file so that they could be
merged with the corresponding field data. i
-7-
-------
Estimates were computed of household and individual characteristics
in Washington and Denver using results from the Household Screening
Questionnaire. When possible, these estimates were compared with
corresponding Census estimates. Household estimates computed for the
two areas included proportion of homes using a fireplace, a wood stove,
a gas furnace, a gas cooking stove, and having an attached garage.
Estimates computed for individuals' characteristics included age and sex
distributions, proportion of smokers, proportion who work, and propor-
tion who travel at least 3 times per week.
Detailed statistical analyses were carried out using computer data
files with PEM CO and activity diary data (Section 6.4). Estimates
computed during this analysis were weighted estimates for the population
of inference - adult non-smokers in the Washington, DC metropolitan
area. Standard errors of estimates were produced by using specially
written software designed for analysis of data from complex sample
surveys (see Shah [1981]).
In particular, analyses were first produced for hourly CO exposure
data. These analyses included computing statistics describing diurnal
patterns, maximum hourly CO concentrations, maximum 8-hour CO concentra-
tions, and mean hourly CO concentrations. Statistics included means,
standard errors, and percentages of the population exceeding certain
specified CO levels. Estimates of these statistics were computed for
all days, week and weekend days, and low and high CO days (as indicated
by fixed site monitors). In addition, CO hourly level comparisons were
also made for 3 occupational groups; 6 commuter group (i.e., non-commut-
ers; commuters who traveled up to 5 hours/week; etc.); and 4 categories
describing the use of gas stoves.
Estimates were also produced for CO exposure levels for various
activities (e.g., in transit) and locations (e.g., indoors-at residence),
Statistics computed for each activity and location included mean CO
level, the estimated standard error, and estimates of the proportion of
the population having CO levels above specified levels. The distribu-
tion of times spent in the various activities and locations were also
computed.
—8—
-------
Breath measurements taken at the end of each individuals' monitor-
ing period were used to produce estimates of the distribution of CO
breath levels in the Washington, DC area. Finally, using the duplicate
CO monitor data, estimates were computed to assess variation in PEM
measurements under field conditions.
-9-
-------
3. SUMMARY OF STUDY RESULTS AND CONCLUSIONS
Based on the experience gained during the Washington, DC and Denver
PEM CO studies, the methodology developed, with some modifications (see
Section 4), may be used effectively in other areas of the country for
collecting PEM data. Experience gained during this initial study will
improve the execution of such similar studies. Modifications that are
suggested include a different sampling design using the classified
telephone directory listings, improvements in the CO monitors, and
additional refinement of the method used to collect activity data.
These modifications should make the methodology more cost effective,
improve the response rate, and lead to more accurate activity informa-
tion. However, it is important to realize that the response rates for a
study of this complexity will always be relatively low as compared to
studies where only a questionnaire is administered to a respondent. In
particular, for the current study in Washington, DC, 58% of the indivi-
duals selected and interviewed over the telephone agreed to supply PEM
CO data, and 35.8% actually gave analyzable CO data (see Sections 5.1
and 5.4 for additional details).
Important new information was learned for each of three sampling
methodology studies of the project: (1) It was found that geographical-
ly classified telephone directory listings can be used in a cost-effec-
tive manner in association with standard area household sampling tech-
niques for personal monitoring studies like the current CO study. The
sampling design for the cost-effective use of these telephone directory
listings differs substantially from the design used for the CO study,
however (see required procedure in Sections 4 arid 6.1). (2) Sending
lead letters to individuals who were selected for personal monitoring
prior to calling to schedule an appointment was found to be an effective
strategy. (3) The need for person-day sampling for studies that monitor
personal exposure to airborne pollutants is apparent. The CO study
gained valuable experience with this technique. Further study, possibly
•10-
-------
even another methodological study, is needed to refine this technique
(see Section 6.1).
Based on experience derived during this project, two important
conclusions were reached concerning the use of the COED-1 and GE/HP
monitors for monitoring personal CO exposure:
The COED monitors exhibited a less than desirable reliability
during this study producing a final successful sample comple-
tion rate of only 78 percent. Since most, of the lost samples
can be attributed to unreliability of the monitor electronics,
the battery packs, or the sample pump (169 of the 232 samples
lost due to monitor malfunction), these monitors will probably
become acceptable for future projects of this type providing
that the recommendations discussed in Sections 4 and 6.3.2.1
are successfully incorporated into the monitor design.
Excessive calibration drift accounted for the remaining 63 of
the 232 samples lost due to monitor malfunction (approximately
6 percent of the samples attempted). The monitors exhibited
high linearity (calibration r2 s 0.9997), acceptable stability
(86 percent within ± 10 percent of initial response levels
after 24 hours), and reasonable precision (median standard
deviation of duplicate measurements = 0. ,25 ppm) during field
monitoring.
The GE/HP monitors will probably be acceptable for such
monitoring following perfection of the design and incorpora-
tion of the recommendations suggested in Sections 4 and
•
6.3.2.2. The full user-programmability of these monitors will
add desirable flexibility, not achieveable with the COED-1, to
future monitoring projects. On-board micro-processor monitor-
ing of, and compensation for, parameters such as cell tempera-
ture and battery voltage may increase monitor stability and
precision.
Concerning the monitoring of alveolar carbon monoxide by the method
utilized during this project, the following conclusions were reached:
The proposed method performed well, producing a mean differ-
ence between duplicate samples of 0.11 p*pm ± 0.13 ppm at the
-11-
-------
95 percent confidence level and an estimated accuracy of ± 0.3
ppm at 3.5 ppm and ± 1.0 ppm at 40 ppm. The proposed modifi-
cation to the procedure concerning use of humidified zero and
calibration matrices is, however, deemed necessary for proce-
dural stability. The method is highly reliable (97.5 percent
successful sample completion rate).
The field wo.rk for the study also indicated, in addition to the
several suggested improvements in the present CO monitor and continued
development of the autolog to record activities on the PEM, that other
modifications can be made to enhance the reliability of the data collect-
ed and to ease the interviewer's work load for future PEM studies.
These include: (1) devising a sampling scheme that will allow for
down-time during data collection to permit instrument repair, enhance
rescheduling of appointments, and provide regular time-off for field
staff; and (2) allowing field staff earlier and fuller involvement in
the development of logistical support mechanisms — including monitor
evaluation and testing, field laboratory space, and publicity (again,
recommendations are given in Section 4).
Using the data collected in the Washington, DC and Denver metropol-
itan areas with the Household Screening Questionnaire, weighted esti-
mates were computed of population characteristics (Section 6.1). These
estimates were based on screening interviews in 4394 households in
Washington and 2128 households in Denver. In particular, the population
estimate for the number of households in the two areas was 953,714 for
Washington and 345,163 for Denver. Population estimates of percentages
of households with various characteristics were as follows:
Washington Denver
33%
Use Fireplace
Use Wood Stove 4%
Use Gas Furnace 56%
Use Gas Stove 64%
Use Gas Hot Water 57%
Have Attached Garage 22%
or Multi-Family Garage
30%
6%
71%
25%
78%
35%
-12-
-------
In addition to household characteristics, several estimates were
also obtained for individuals' characteristics in the two areas. For
example,
Washington Denver
Male 48% , 47%
-
Smokers (13 years i
• ' . or older) 33% | 38%
Work (13 years or older) 70% 72%
Travel £ 3 times/week 84% , 82%
It is important to note the distribxition of people selected for
actual monitoring necessarily differs from the above in several ways.
First, smokers were not sampled because they were ineligible for par-
ticipation in the survey. Second, other population subgroups thought to
be at risk for high CO exposure were oversampled,j As discussed in
Section 5.1.4, individuals with a usual daily commuting time of 30
minutes or more were oversampled. Individuals with, a gas stove or space
hteater in the home were also oversampled, as were tihose with an attached
garage. Thus, other population subgroups were relatively undersampled.
The purpose of the oversampling was to insure representation of the
population subgroups most likely to be exposed to high CO levels. This
'
oversampling is compensated for in analysis of the CO data by use of the
sampling weights. The sampling weights are inversely proportional to
the probability of selection. Thus, the member of subgroups that were
oversampled receive smaller sampling weights. As a result, weighted
analyses produce unbiased estimates even when the isubgroup sample sizes
.
are not proportional to the number of population members in the sub-
groups.
Regarding estimates of CO exposure for the winter of 1982-83 in
Washington, DC, a data base was constructed from the raw CO levels by
activity data which consisted of hourly CO values on 712 respondents,
activity patterns and corresponding CO levels on 705 respondents, and CO
breath measurements corresponding to the PEM CO data on 659 respondents.
These data were used to obtain estimates of CO exposure for the popula-
tion of inference ~ the adult (18 to 70 years old), non-smokers in the
urbanized portion of the Washington, DC SMSA. The| size of this popula-
-13-
-------
tion was estimated to be 1.22 million individuals. The results presented
below are weighted estimates which apply to this population. Before
analysis of the data could be undertaken, extensive editing was required
of the raw data collected in the field (see Section 5.4).
The weighted average maximum hourly PEM CO level in Washington, DC
was 6.74 ppm (this was computed as the weighted average of the maximum
hourly CO value for each individual in the sample). The average maximum
8-hour CO level was 2.79 ppm. The percentage of the population with
maximum hourly CO values over the 35 _ppm CO standard was estimated to be
1.28 percent while, the percentage with an 8-hour maximum over the 9 ppm
standard was 3.9 percent.
Estimates were also made for subgroups of the population. Persons
in high-exposure occupations (about 4.6% of the total population)
generally exhibited higher CO exposure levels: it was estimated that
about 24% of this high-exposure group had 1-hour CO exposures above the
35-ppm standard and that about 28% exceeded the 8-hour standard. It was
also shown that CO levels were generally higher for commuters, especially
for those with larger amounts of travel. For example, 8% of the commut-
ers indicating 16 or more hours of travel per week were estimated to
have maximum 8-hour CO concentrations over 9 ppm, whereas less than 1%
i,
of the non-commuters were estimated to have such levels.
Breath CO levels (taken at the end of the sampling periods, usually
in the respondents' homes) for the adult non-smoking population in
Washington averaged 5.12 ppm. Slightly higher levels were observed for
persons with high occupational exposures and for persons with large
amounts of travel.
By combining PEM data with data from individuals' diaries, estimates
of both CO levels and time durations for various activities and personal
environments were made. In general, these results were consistent with
a priori expectations. For example, the activities "in parking garage
or parking lot" and "travel, transit" had the highest average CO concen-
trations (6.93 ppm, and 4.51 ppm, respectively) while "sleeping" had an
estimated CO concentration of only .85 ppm. Among the environments
considered, the three with the highest average CO concentrations, were
"indoor parking garage", "outdoor parking area", and "in transit-car".
-14-
-------
The average levels for these environments were 10.36, 4.67, and 5.05
ppm, respectively.
Variation from duplicate hourly PEM measurements under field
conditions were also analyzed. An analysis of variance of this data
which considered person-to-person, hour-to-hour, and measurement vari-
ation indicated that about 5 to 6 percent of total variation among the
hourly duplicate readings was due to deviations in measurements made by
two PEMs at the same hour for the same person.
-15-
-------
4. RECOMMENDATIONS
The recommendations presented in this section suggest (1) improve-
ments in the sample design and associated sampling procedures (e.g., the
PEM); (2) changes in the logistics and methods of data collection for
improving data quality and response rates; and (3) performing additional
statistical analyses.
4.1 Design Recommendations
The suggested improvements in the sample design are described in
detail in Section 6.1.4. Briefly, from RTI's experience using the
geographically-classified telephone directory listings in the current CO
study, it appears that such listings can best be utilized with a dual
frame sampling procedure. In this approach, two independent samples of
first stage units would be selected from the (complete) area frame: (1)
one sample would be a standard area sample with sample clusters identi-
fied from field listings of all housing units in the selected area
segments, (2) the other sample would use the commercial listings to
identify sample clusters in the selected area segments. The commercial
listing sample would only be used to generate telephone interviews based
upon the telephone directory listing while the standard area frame
sample would be used to compensate for the bias resulting from the
telephone interviews generated by the commercial listing sample (see
Section 6.1.4 for details on how the bias would be compensated for).
The lead letter methodology study (Section 5.1.4) indicated that
these letters appeared to have had a positive effect upon response
rates. Accordingly, in future studies, RTI would recommend that lead
letters be sent to all individuals selected for monitoring. In this
regard, it is important that the entire data collection methodology be
further reviewed to determine other methods that might be used to
increase response rates.
Some form of person-day sampling is necessary for future studies
which monitor personal exposure during some time period. However, the
-16-
-------
procedure used in the current study (Section 5.1.4J) was somewhat awk-
ward. Therefore, a better procedure needs to be designed. One possible
methodology would be to select six days within the study period for each
person selected for monitoring. These could be three consecutive days
in one week and the same three days in the next week. Priorities from
one to six could then be assigned for each person.> When the person is
asked to participate, he would be asked to participate on his first
priority day. If he could not participate then, the interviewer would
proceed to the second priority day, etc. This methodology, or an
alternative method, should be explored in future studies.
4.2 Recommendations Concerning Field Operations and Data Collection
As suggested in Section 3, there are several ways in which the
field aspects of this study could be improved. These improvements would
affect the willingness of individuals to participate, reduce the burden
placed on respondents, and make the interviewer's jpb less difficult and
time consuming. In general, the recommendations for change or improve-
ment fall into three areas. These include: (1) the Carbon Monoxide
(CO) Monitor and general data collection, (2) the sampling and field
data collection process, and (3) logistical support.
Data Collection. Further development work must be done on both the
CO monitors and the survey instruments used to collect data. When a
respondent has agreed to participate and an appointment has been sche-
duled (sometimes with great difficulty), the loss of data due to instru-
i
ment failure is quite distressing to the respondent and the field staff;
moreover, the loss of data at an unpredictable and variable rate makes
it extremely difficult to schedule field operations! (e.g., to determine
completion dates). The reliability of the CO monitor is thus extremely
important to the results of a study of this type; therefore, further
work must be done to improve the COEDs before they are used again in a
full-scale field study.
Some specific problems with the two types of piortitors used in this
study, the COED-1 monitor and the GE/HP monitor, that need to be resolved
before either monitor is used in the field for future studies are
addressed below: i
-17-
-------
COED-1 Monitor
(1) The electronic problems with the Magus data unit of the COED-1
monitor must be corrected. Problems to be corrected are
"lock-up", "mode shift", and susceptability to static dis-
charge .
(2) Alkaline batteries should be considered as a substitute for
the nickel-cadmium (Ni-Cd) batteries currently powering the
GE-CO unit. Many of the battery related monitor failures were
due to charging difficulties and reliability problems with the
Ni-Cd batteries. Brief field tests indicated that six alkaline
batteries will power the data unit for up to seven 24-hour
sampling periods before replacement and four alkaline batteries
will adequately power the CO monitoring unit for up to four
24-hour periods.
(3) A more durable sample pump, still compatible with monitor
specifications, is necessary. The service life of the current
pump may be as low as 900 hours.
(4) The configuration of the sample flow path should be modified
so the flow through the prescrubber is up with respect to
gravity to minimize the deposition of prescrubber material
fines in the pump. If this is not possible, the scrubber
should be horizontal. An efficient filter between the pre-
scrubber cartridge and the pump may solve the problem if the
filter can be easily replaced. Research-has shown the filter
will be quickly contaminated.
(5) The unit should have a sample pump on/off switch inaccessible
to the respondent but available to the interviewer to reduce
the load on the monitoring unit batteries.
(6) The electrical connection to the sample pump should be modi-
fied to facilitate removal and replacement of the pump because
of the field requirements for frequent pump service. The
connection currently requires soldering a piece of printed
circuit tape to the pump motor terminals.
-18-
-------
GE/HP Monitor :
(1) The electronic design of the GE/HP monitor should be carefully
examined and modified to eliminate the llogic faults experi-
enced during this project.
(2) The compatibility of the lead-acid gel cell batteries with the
GE/HP unit should be investigated. Indications are that the
batteries may not be capable of powering ;the currently design-
ed unit for the necessary 28-32 hours. i
(3) After the battery capacity question is decided, clear and
complete instructions for charging the batteries must be
written and charger /charging circuits olf the appropriate
capacity should be assembled and supplied to future users.
(4) The packaging of the GE/HP unit should', be redesigned to
combine the current two-component package!.
(5) If batteries are to be removed from the; GE/HP monitor for
recharging, they should be made easier to access and remove.
i
Also, battery connections should be polarized to prevent
accidental reversal of polarity.
(6) The redesigned monitors should be thoroughly evaluated in the
, -laboratory and in the field before they ;are used in another
sampling project. i
The rationale for these specific recommendations is further elaborated
in Section 6.3.2.
In addition, further development and refinement of the capability
to enter activities directly into the GE/HP monitor's data storage would
yield benefits in at least two ways. The first wo;uld be the reduction
in the number of steps involved in preparing data for analysis. The
second would be higher quality information on activities, since more
consistency from respondent-to-respondent would be attained. Since
f
activity and exposure data would exist on the saute data string, the
problems associated with matching of data points by time, dealing with
missing or inappropriate entries, and preparing analysis files would be
ameliorated. Although the experience gained using the Hewlett Packard
(HP) based COEDs was limited in time and numbers, the response of the
persons who used the devices was very favorable. If the size of the
-19-
-------
keys and the printing of activities were enlarged, the instrument and
the process of logging activities would be totally acceptable to respon-
dents. In order to have more flexibility as to the number and types of
permissible activities, a large detailed set of disjoint activities
could be developed (based on data from this and other studies). Two-
digit activity codes could then be assigned, and the HP programming
could be modified to store and use the two-digit entry for such activi-
ties. After further development and testing, such a device might allow
the use of the activity diary to be dropped or highly modified.
Review of the diaries indicated problems with legibility of respond-
ents' entries and varying degrees of conformation to the specifications
for entering activity and location information. In particular, diaries
for later studies should be structured to force a standardized format
for address entries. Further explanation by interviewers can address
these problems, but a cost in time spent with respondents in a tightly
scheduled field activity has .definite impacts on the study. While
modification of the diary format, instructions, and explanations may
help, the best solution is the further development of the automatic
entry of activity codes. A diary should be retained to provide quality
control checks for the automatic logger, to test other concepts of data
collection, and to provide detailed location (address) information.
Sampling and Field Data Processing. It is very difficult to
maintain a field data collection process of the intensity of the CO
study on a seven-day-per-week basis. The logistics required to start
and finish individual respondents' data collection periods created
extreme scheduling problems for both interviewers and respondents. It
was very difficult to permit changes in respondents' appointments, even
within a three day window, and it was exceedingly difficult to schedule
time off for interviewers without compromizing the entire data collec-
tion process. Undoubtedly, a larger interviewing staff would aid the
problem, but of course there would be an associated increase in costs.
A review of the process of selecting respondents in terms of location
and appointment availability might also yield some benefits by cluster-
ing work to provide increased efficiency and reduced expense.
-20-
-------
Logistical Support. For future studies, it is recommended that the
field staff who are to work in the laboratory and field office space be
involved in the selection of the area to be used, i For example, for a
study similar in size and duration to the CO study, using similar
monitoring equipment, approximately 500 square feet of office space
would be needed. About 325 square feet would be used for the field
technicians as a laboratory. The remaining 175 square feet would be
used by the field interviewers for office and storage space. This space
should be centrally located in the interviewing area with easy entry for
staff members. Without this amount of space, the field staff and
technicians will be hindered in performing their duties.
Another major concern is publicity. With the current climate of
nonresponse to personal and telephone interview contacts, field data
collections are increasingly difficult. Early and continued publicity
on a project, with increased intensity at the beginning of each phase
will help reduce non-participation by increasing awareness of the
legitimacy of the study. A multi-media publicity drive before the
telephone screening and before the appointment-scheduling telephone
calls should be incorporated into further studies of this type. Use of
lead letters before the final round of calls increased the response rate
in the sample tested. Inclusion of a letter from the project sponsor
would also be of benefit.
4.3 Recommendations for Further Statistical Analysis
Additional statistical analysis of the CO data over and above that
presented in Section 6.4 should be undertaken. The data base developed
by RTI and described in Section 5.4 is extremely rich and allows infer-
ences to be made to a large urban area. In particular, possible addi-
tional .analyses include the following:
(1) Statistical testing of differences between CO levels in
various activities and environments using appropriate statis-
tical software.
(2) Modeling maximum hourly and maximum 8-rho'iir CO levels as
functions of activities, environments, and questionnaire data
(for these analyses, it may be desirable to control for
individual characteristics, e.g., type of occupation).
-21- !
-------
(3) Comparing fixed-site and PEM values.
(4) Computing time-weighted CO levels (to take into account the
time spent in an activity; e.g., the maximum 8-hour CO level
is not highly affected by 15 minutes in a parking garage).
(5) Performing additional analysis of questionnaire data to
determine the usefulness of specific questions.
(6) Examining in more detail relationships between CO levels and
diary and questionnaire information (e.g., if occupationally-
exposed individuals were actually working, etc.).
(7) Determining whether meteorological data are useful in predict-
ing CO levels (analyses described in Section 6.4 only used
high- and low-CO days; temperature, wind speed and direction,
relative humidity, atmospheric stability, mixing height, and
precipitation are also available).
(8) Correlating breath and PEM levels.
(9) Analyzing childrens' breath levels.
(10) Investigating between-person versus within-person variation.
(11) Analyzing data from the Computer Model Input Questionnaire.
(12) Presenting standard errors of estimated quantities (e.g., the
proportions shown in Section 6.4).
-22-
-------
5. METHODS AND PROCEDURES
5.1 Survey Design
The purpose of the carbon monoxide (CO) study was to develop
methodology and to monitor personal exposure to carbon monoxide for
residents of the metropolitan areas surrounding Denver, Colorado and
Washington, D.C. The target populations consisted of the non-institu-
tionalized, non-smoking residents, aged,18 to 70, of these metropolitan
areas. For the purpose of sample selection, the Washington, D.C.
metropolitan area was precisely defined to be all areas simultaneously
t
inside the Washington, D.C. SMSA and inside the Washington, D.C. Urba--
nized Area as defined by the 1980 Census (see Appendix A). The Denver
metropolitan area was defined to be the following places in Colorado as
defined by the 1980 Census: Denver, Englewood, Arvada, Aurora, and
Commerce City. These areas are also in the Urbanized Area of the Denver
SMSA (see Appendix A). |
Among the objectives of the CO study was to make inferences con-
cerning the personal CO exposure for all members of the Denver and
Washington metropolitan areas. The only statistically valid procedure
that is widely accepted for making such inferences^ is to select a
probability sample from the target populations. Hence, the sample
design for the CO study is a stratified, three stage probability-based
design. Area sample segments defined by Census geographic variables
were selected at the first stage of sampling. Households were selected
at the second stage, and all household members were administered a short
screening interview. The purpose of this intervie^r was to identify
'
individuals with characteristics believed to be positively correlated
|
with CO exposure so that they could be oversampled '. in the third stage
sample. The third stage sample was a stratified sample of screened
eligible individuals. The individuals in the third stage sample were
administered a Computer Model Input Questionnaire jby telephone (see
Appendix B) and asked to carry a personal CO monitor for 24 or 48 hours
-23-
-------
for Washington or Denver, respectively. A breath sample was also
requested from the individuals who were monitored, and they were asked
to fill out a household questionnaire (see Appendix C).
Whenever probability sampling techniques are used in sample sur-
veys, sampling weights that reflect the procedures used for sample
selection must be used when analyzing the data. The weight of a sample
unit can be viewed as the number of units in the target population that
the sample unit represents. The initial sampling weight for a unit is
usually calculated as the reciprocal of either the probability of
selection of the unit or the expected frequency of selection when
sampling with replacement. This initial sampling weight is often
adjusted in later steps to reduce the bias caused by nonresponse and
undercoverage of the target population. The adjusted sampling weights
serve to differentially xreight the sample data to reflect the level of
disproportionality in the final sample relative to the population of
interest.
If the sampling weights were all equal, the weights could be
ignored in constructing survey estimates. Otherwise, the weights must
be used in order to obtain unbiased population estimates. Even when the
sample has been designed to affect exact proportional representation of
the target population, the differential impact of nonresponse and
undercoverage leads to a distortion in the sample which requires the
construction of differentially adjusted sample weights. Thus, it would
be quite unusual for sample suirvey data to yield unbiased estimates
without the use of sampling weights in the analysis.
Since probability sampling techniques were used to select the
individuals to be monitored, the validity of inferences for this survey
is based upon the statistical theory of sample surveys (see, e.g.,
Cochran [1963] or Raj [1968]). Probability sampling affords unbiased
inferences to the target population when sampling weights are used in
the analyses.
To the extent that respondents and nonrespondents are alike with
respect to probability of responding and/or response values within non-
response weighting classes, the use of adjusted weights for the analyses
-24-
-------
reduces the bias due to nonresponse. This topic is discussed further in
Whitmore, et al. [1983a]. |
Section 6.1.2 contains analyses of the screening data, including
comparisons to 1980 Census data on age, race, and sfex for the target
areas. The Census data do not reflect changes in the populations of the
target areas between April, 1980, and the Fall of 1982, when the CO
study was performed. They do, however, provide a useful benchmark for
comparison. The results presented in Section 6.1.2 show that the sample
i
estimates and Census values are indeed comparable. jNo such benchmarks
i
are readily available for the persons selected for GO monitoring since
these individuals were required to be nonsmokers between 18 and 70 years
of age. These benchmark comparisons are not necessary, however, due to
the solid probability foundation of the sample selection methodology.
The probability structure of the sample provides the basis for valid
inferences to the target population. '
The purpose of this section is to document the sample design and
construction of sampling weights for the CO study. |(Additional discus-
sion of the sampling weights is also presented in Whitmore, et al.
[1983a]). Initial, unadjusted sampling weights were computed for all
individuals selected into the monitoring sample. These sampling weights
are simply the reciprocal of the overall probability of selection for
each individual. Two sets of adjusted weights have :also been computed.
The first set of adjusted person-level weights is adjusted only for
household-level nonresponse to the screening interviews. The second set
I
of adjusted weights is also adjusted for person-level nonresponse to the
CO monitorings. Considerations related to use of these analysis weights
are presented in Sections 5.1.5 and 5.1.6 and in '/Jhitmore, et al.
[1983a]. j. ' .
5.1.1 Selection of First-Stage Sampling Units (FSUs)
The first step in selecting the samples|was to extract all
block group (BG) and enumeration district (ED) records for each target
area from the 1980 Census Summary Tape File 1A (STF--1A) data tapes. In
both target areas, all Census records were found to be block group
records. However, there is sometimes more than one;record for a single
-25-
-------
block group. Hence, the next step was to produce a data set of records
for each site such that there was one and only one record for each block
group.
Since some block groups contain few, if any, occupied housing units
as reported by the 1980 Census, it was necessary to combine block groups
to form a sampling frame of first-stage units (FSUs). The block group
records for each site were first ordered by the following Census geo-
graphic variables:
State, County, Tract Basic, Tract Suffix, Block Group.
The block groups were then combined within tract to form FSUs with a
target size of 40 or more occupied housing units. The BG-level records
were processed sequentially and records with a small size measure were
combined with the following records until either the last BG record in a
tract was reached or the frame unit contained 40 or more occupied
housing units. The FSUs. were not allowed to cross tract boundaries
because of a desire to use sequential selection by tract number to
assure geographic dispersion of the sample. In retrospect, the FSUs
could have been allowed to cross tract boundaries without affecting the
geographic dispersion of the sample. Each FSU would then contain at
least 40 occupied housing units, and undercoverage of tracts with no
1980 Census occupied housing units would be prevented. None of the
small frame units were actually selected into the samples, however.
In order to achieve approximately equal probabilities for the
second-stage sampling units (SSUs) for each site, the first-stage
sampling units were selected with probabilities proportional to size as
measured by the 1980 Census counts of occupied housing units. Equal-
sized random samples of SSUs within each FSU would then result in
approximately equal probabilities for the sample of SSUs.
A sequential, minimum probability replacement (MPR) sampling
procedure developed at RTI was used to select the sample of FSUs (See
Chromy [1979] and Williams and Chromy [1980]). The frame for each site
was first ordered in a serpentine fashion by the following variables:
-26-
-------
State, County, Tract, TOTHUS,
where TOTHUS is the number of 1980 Census occupied housing units for the
FSU. The FSUs were then selected with probabilities proportional to
size as measured by TOTHUS from the ordered sampling:frame. The sample
sizes were 250 FSUs for Washington and 100 for Denver. Since the
sampling was performed with replacement, multiple selections of the same
FSU were possible. One FSU in the Denver sample was:selected twice, so
the Denver sample contains 99 distinct FSUs or area: sample segments.
The sample for Washington consists of 250 distinct FSUs or area segments.
In addition, the ordering of the sampling frames resulted in a controlled
allocation (proportional allocation) to the implicit geographic strata
?
formed by crossing the sorting variables. This control results from the
sequential nature of the sample selection and is analogous to selection
of a systematic sample from an ordered frame.
After the sample FSUs were selected, a computer tape listing the
block groups (BGs) in the sample FSUs was sent to Donnelley Marketing
Corporation. The tape was returned to RTI with name, address, and tele-
phone number listings for each sample BG. The first-stage sample
selection and selection of the initial sample of Donnelley listings is
summarized by the flow chart in Figure 5.1.1. !
The Donnelley listings had been compiled from two sources: tele-
phone directory listings and vehicle registration records. Ideally, the
number of Donnelley listings for a BG should be comparable to the 1980
Census of occupied housing units for the BG. The Donnelley list count
and the Census count were comparable for most BGs. Unfortunately,
however, three FSUs in the Washington sample had no Donnelley listings.
These three FSUs were in an area of Maryland for which the Donnelley
Corporation had no telephone listings, and Maryland
does not allow the
Donnelley Corporation access to its vehicle registration records. For
all other FSUs, the Donnelley lists were used as the second-stage
sampling frames.
-27-
-------
Extract all block group and enumeration
district records for the target area from
the 1980 Census STF-1A data tape. I/
Combine records whenever necessary to pro-
duce one record for each block group.
Form first stage units (FSUs) by combining
block groups as necessary within Census
Tract so that each FSU contains at least 40
Census occupied housing units.
Select 250 FSUs for Washington and 100 FSUs
for Denver with probability proportional to
size as measured by the Census occupied
housing units.
Send the list of block groups comprising
sample FSUs to Donnelley Marketing Corpora-
tion and receive their lists of block group
residents .
Select 50 Donnelley listings from each
Denver FSU and 40 listings from each Wash-
ington FSU, or select all listings from a
FSU that has fewer than the required number.
I/ All extracted records were block group records.
Figure 5.1.1 Selection of First-Stage Units and the Initial Sample of
Donnelley Listings.
-28-
-------
5.1.2 Selection of Second-Stage Sampling Units (SSUs)
, A second-stage .sample of housing units, as defined for the
1980 Census, was selected within each sample FSU. ;.For the FSUs with
Donnelley listings, the Donnelley listings were used to select the
I
second-stage sample. The listings were used as a second-stage sampling
frame in lieu of the traditional field listing of all housing units in
the sample segments. For the three Washington FSUs: with no Donnelley
listings, field procedures were used to select second-stage samples of
housing units. ,.
'
5.1.2.1 Selection of SSUs Within FSUs With Donnelley
Listings
For each FSU with Donnelley listings, a simple
random sample of listings was selected without replacement. A random
sample of 50 listings was selected within each Denver FSU, and 40 within
each Washington FSU. For any FSU that did not contain this many list-
ings, all listings were selected. A sample of 9,876 listings was
selected for Washington, and 4,987 listings were selected for the Denver
sample. When it became apparent that a smaller sample size would
produce sufficient screening data for Washington, the Washington sample
was randomly subsampled. The utlimate sample size for Washington was
8,643 listings. Whitmore, et al. [1983a] discusses the computation of
the first stage sample weights using the above scheme.
In order to obtain complete coverage of the target population, the
i
sample of Donnelley listings was regarded as a sample of housing units
identified by the name and address in each Donnelley listing. There
were many Donnelley listings for which the address alone was not suffi-
cient to identify a specific housing unit. This was particularly true
of housing units in apartment complexes. Since most Donnelley listings
come from telephone directory listings, the address;shown for apartment
residents was often no more than the street address of the apartment
complex. Whenever the address was not sufficient to identify a parti-
cular housing unit, the individual's name was also> used. The sample
housing unit was then defined to be the housing unit at the sample
address in which the named individual either currently or previously
resided. i
-29-
-------
A single housing unit also may be linked to more than one Donnelley
listing. This can occur whenever the Donnelley Corporation listed
multiple telephone subscribers and/or vehicle registrants for a single
housing unit. These frame multiplicities can be associated with current
residents of the housing unit and/or previous residents of the housing
unit. Since the name and address must sometimes be used simultaneously
to identify a sample housing unit, it is virtually impossible to accu-
rately identify all of these frame multiplicities. Consequently, no
attempt was made to remove these multiplicities from the sampling frame
prior to sample selection. It also does not appear to be cost-effective
to pursue any multiplicity adjustments in analysis of this sample.
Hence, the sampling weights have been computed treating each sample
housing unit as if it was linked to only one Donnelley listing.
All sample Donnelley listings were initially assigned to the
telephone mode for administration of the screening interview. About 75
percent of the sampled Denver listings and about 88 percent of the
sampled Washington listings had phone numbers. The difference seems to
be that vehicle registration records were not available to Donnelley
Corporation for the Maryland portion of the Washington sample, but were
available for all of Colorado. RTI's telephone interviewing staff
attempted to obtain telephone numbers for the sample listings with no
phone number. Phone numbers were obtained for about ten percent of
these listings.
For each call made by the telephone interviewing staff, the tele-
phone number was first verified with the individual who answered. If
the correct telephone number had been reached, the address was also
verified. Since the sample was regarded as an address sample, not a
telephone number sample, the interview was terminated if the indivi-
dual's address was not the address shown for the sample Donnelley
listing. These addresses were placed in a pool to be subsampled for
field interviewing. Otherwise, all individuals living in the housing
unit (1980 Census definition) were screened by a telephone interview.
The name shown for the Donnelley listing was never verified. It was
implicitly assumed that the correct housing unit had been accessed if
the telephone number and address were correct.
-30-
-------
Table 5.1.1 presents the final result codes generated by the
telephone screening attempts. It also identifies a pool of result codes
for which the listings were treated as not covered by the telephone
screenings, e.g., those for which no phone number could be obtained.
[
Some assumptions are inherent in this categorization. For example, it
has been assumed that listings with a final result of "ring-no-answer"
are listings for which the address would have been verified as correct
if someone had been contacted. The distribution of the telephone result
codes is also presented in Table 5.1.1. ;
i
It was not feasible with the project's time and [money constraints
to attempt field screenings for all sample Donnelley listings not
covered by the telephone screenings. Hence, a subsample of these
listings was selected as illustrated by the subsample of n« Donnelley
listings shown in Figures 5.1.2 and 5.1.3. Given the sampling rates
'
shown in Figures 5.1.2 and 5.1.3, the loss in precision due to this
subsampling is considered acceptable. A field interviewer was sent to
each housing unit represented by this subsample of Donnelley listings.
The field interviewer administered the screening interview if a reliable
respondent was available on the first attempt. Otherwise, the inter-
viewer attempted to get the correct telephone number from a neighbor or
information operator. When the field interviewer was able to get a
telephone number for the sample housing unit, subsequent screening
attempts were made by telephone. When the first attempt did not yield
either a completed interview or a telephone number, additional field
attempts were made for the Denver sample. The wide geographic disper-
sion of the Washington sample made additional field [attempts prohibi-
tively expensive for the Washington sample. j
As shown in Figure 5.1.2, a total of 1,825 sample Donnelley list-
ings were not covered by the initial telephone screenings for Denver. A
simple random sample of 242 of these listings was selected without
replacement. A field screening attempt was made for each listing in
this subsample. Similarly, as shown in Figure 5.1.3,, a simple random
sample of 353 listings was selected without replacement from 2,396
sample listings not covered by the initial telephone screenings for
Washington.
-31-
-------
Table 5.1.1 Distribution of Initial Telephone Screening Final Result
Codes
Result
Code
01
51
52
53 -1
64 -l
71^
72
73
74^
75
76 -/
81 - -
82
83
84
95
Total
Frequency
Interpretation
Completed Interview
Refusal or Breakoff, Conversion
Attempt Failed
Refusal or Breakoff, Not Reached for
Conversion Attempt
Circumstantial Non-Interview
Partial Interview
Nonworking Number
Nonresidential Number
Entire Household Moving
Address Doesn't Match Donnelley Listing
No Reliable Respondent (3 attempts)
No Phone Number
Final Phone Problem (confirmed by
operator)
Ring-No-Answer (6 attempts)
Answering Machine (6 attempts)
Final Busy (10 attempts)
Other (both eligible & ineligible cases)
Denver
1,997
469
135
6
13
436
91
92
350
21
1,032
7
207
22
4
105
4,987
Washington
4,245
673
93
10
32
997
187
187
609
36
782
8
503
35
4
242
8,643
I/ An eligible housing unit was contacted, but circumstances prevented
a screening interview.
2j Breakoff after Question 7.
3/ Result Codes for the pool of Donnelley listings not covered by the
initial telephone screening.
4/ Branching, complete silence, fast-busy or other problem confirmed
by an Operator.
-32-
-------
Initial Sample of n.
Listings
= 4987 Donnelley
4987 Telephone
Screenings Attempted
T
1165 Refusals; Non-
interviews ; Unable
to Contact Reliable
Respondents; Non-
Residential Phones,
Etc.
Stratified Sample of
n. = 1000 Eligib3.es
Refusals;
Ring-No-Answers
Etc.
Refusals; Not-
At Homes, Etc.
1
1825 Donnelley
Listings Not
Covered by Tele-
phone Screening
Attempts
Sulisample of
n '=• J50 Listings
to .Check for
Hifised HUs
Refusal; Mon-
Tnterview;
Unable, to Con-
tatt Reliable
Respondent, Etc.
Figure .5.1.2 Denver CO Sample Protocol
-33-
-------
Initial Sample of "h. '
Listings
8643 Donnelley
8643 Telephone
Screenings Attempted
2002 Refusals; Non-
interviews; Unable
to Contact Reliable
Respondents; Non-
Residential Phones,
Etc.
2396 Donnelley
Listings Not
Covered by Tele-
phone Screening
Attempts
Stratified Sample
of n. - 1779
Eliglbles
Sample of n = 96
HUs from Segments
With no Donnelley
Listings
1
Refusals;
Ring-No-Answers
Etc.
Refusals;
Not-At Homes,
Etc.
Subsample of n_
= 300 Listings
to Check for
Missed HUs
Refusal; Non-
Interview;
Unah.1 e. to Con-
tact Reliable
Respondent, Etc.
1 A total of 1217 interviews were scheduled between all Washington samples.
2 Usable CO monitor data were obtained for 712 individuals in the Washington samples.
Figure 5.1.3 Washington CO Sample Protocol
-34-
-------
The second-stage sample of Housing units also addressed the problem
of undercoverage by the frame of Donnelley listings,; In a traditional
listing of all HUs in the sample segments, the interval from each sample
housing unit (HU) to the next listed HU is checked for missed HUs. If
unlisted HUs are found, they are included in the sample. The direct
extension of this procedure to the sample of Donnelley listings would
require a check for missed HUs for each of the 8,643 Washington sample
HUs and each of the 4,987 Denver sample HUs. This|procedure was not
feasible since most of the screening interviews were conducted by
telephone. Instead, a subsample of Donnelley listings to check for
missed HUs was selected for each site. The subsample of n. HUs, using
the notation from Figure 5.1.2, was selected in two stages. First, a
subsample of FSUs was selected, 15 for Denver and 30 for Washington.
Then a subsample of ten Donnelley listings from the initial sample of nn
U
Donnelley listings was selected by simple random sampling without
replacement from each of these area segments. The F,SUs were selected as
a stratified simple random sample without replacement. This stratifi-
cation was designed to guarantee that some of the missed HU sample would
fall within the segments for which the number of 1980 Census occupied
housing units was 50 percent or more greater than the number of Donnelley
listings. |
The sample of Donnelley listings for which a. missed HU check was
performed is considered to be minimally adequate. A. larger sample was
not selected because the expense of the missed HU checking depended upon
the unknown quality of the Donnelley listings. The missed HU field
procedure was designed to also detect and document misclassifications of
HUs into block groups. Hence, the Missed HU Sample, was important for
assessing the usefulness of the Donnelley listings as a sampling frame.
As a result, it was decided that a thorough check in a few FSUs was the
best approach. The results of these checks were generally favorable and
are discussed in Section 6.1.4.
5.1.2.2 Selection of SSUs Within FSUs With No Donnelley
Listings
As discussed earlier, there were three FSUs in the
Washington sample for which there were no Donnellejr listings. Sample
-35-
-------
housing units were selected from these FSUs even though no funds were
budgeted for field listing of housing units in area segments. Hence, a
minimal cost procedure that affected complete coverage of these FSUs was
adopted. Block statistics from Summary Tape File IB (STF-1B) were used
to obtain the 1980 Census count of occupied housing units for each block
within these FSUs. The blocks were then combined into subsegments and
one subsegment was selected from within each FSU. The subsegments were
selected with probabilities proportional to their 1980 Census count of
occupied housing units. Thus, approximately equal sized samples within
the subsegments were selected to obtain an approximately equally weight-
ed sample.
Since the budget for field screenings was very limited, the field
samples were selected without listing all housing units. Instead, every
k HU was selected into the sample, with a random start from 1 to k.
The value of k was chosen to yield approximately 35 sample housing units
in each subsegment based upon the 1980 Census data. The only compromise
resulting from this procedure is that direct quality control of the
listing became impractical. However, as an alternative, these samples
were listed by one of RTI's most experienced and reliable field staff
members. We feel that the quality of the listing was probably superior
to the qualilty usually achieved with the standard procedures.
If a reliable respondent was available on the first pass through
each subsegment, a screening interview was conducted. If the field
interviewer could get the home telephone number, subsequent screening
attempts were made by telephone from RTI. Otherwise, the budget per-
mitted no further interview attempts. Further discussion of the special
subsample is presented in Whitmore, et al. [1983a],
5.1.3 Screening Response
The distribution of final result codes for the Denver and
Washington screening samples is shown in Tables 5.1.2 and 5.1.3. The
overall response rates for the screening phase, shown in Table 5.1.4,
were 70.4 percent for Denver and 75.8 percent for Washington. The
sample design proposed in Section 6.1 is expected to raise these re-
sponse rates to near 80 percent for future studies of this type.
-36-
-------
Table 5.1.2 Distribution of Final Result Codes for Denver Screening
Sample ;
Result
Code
01
02
03
05
11
Interpretation
Primary
Phone
Screenings
Completed Interview/
Telephone Number
Obtained
Completed Interview/
Telephone Number Not •
Obtained
Completed Interview/
No Telephone Available
Entire Household
Moving Based on
Field Contact
No Age Eligible During
Field Contact/Telephone
Number Obtained
1,997
No
Previous Missed
Contact HUs Total
109
27 2,133
!o
j_*-t ix \j WLI<=> LIVJIUG ISIJ.JL o-i-ig
Field Contact /Telephone
Number Not Obtained 0
17 Language Barrier/
Telephone Number
Obtained 0
18 Language Barrier/
Telephone Number
Not Obtained 0
20 Field Refusal/
Breakoff 0 2
25 No Contact 0 2
30 Not a Residence 0
31 Vacant 0
•». 32 Demolished /Condemned 0
33 No Such Address 0 1
40 Other Field Result 0 4
-37-
00 0
4 1 5
1 0 1
2 8 30
9 12 41
8 ,0 8
6 0 6
3 0 3
1 0 11
3 0 43
continued
1
-------
Table 5.1.2 continued
Result
Code
51
52
53
64
71
72
73
74
75
82
83
84
95
TOTAL
Primary
Phone
Interpretation Screenings
Refusal or Breakoff
Conversion Attempt
Failed
Refusal or Breakoff
Not Reached for Con-
version Attempt
Circumstantial
Non-Interview
Partial Interview
Nonworking Number
Nonresidential Number
Entire Household Moving
Address Doesn't Match
Donnelley Listing
No Reliable Respondent
(3 attempts)
Ring-No-Answer
(6 attempts)
Answering Machine
(6 attempts)
Final Busy (10 attempts)
Other (both eligible
and ineligible cases)
469
135
6
13
0
91
92
0
21
207
22
4
105
3,162
No
Previous
Contact
0
0
0
0
0
0
0
0
0
0
0
0
0
242
Missed
HUs
0
0
0
0
0
0
0
0
0
0
0
0
0
48
Total
469
135
6
13
0
91
92
0
21
207
22
4
105
3,452
-38-
-------
Table 5.1.3
Distribution of Final Result Codes for Washington
Screening Sample
I/
Result
Code
01
02
03
05
11
14
18
20
25
30
31
32
33
40
51
52
53
64
71
72
73
74
75
82
83
84
95
Total
Primary Phone
Screenings
4,245
0
0
0
0
0
0
0
0
0
0
0
0
0
673
93
10
32
0<
187
187
0
36
503
35
4
242
6,247
No Previous Missed No Donnelley
Contact HUs Listings Total
85 15
2 1
0 0
2 0
0 1
9 1
1 0
29 3
49 17
8 0
19 0
5 0
7 0
38 6
0 0
0 0
0 0
3 0
5 0
1 0
0 1
27 0
1 0
2 1
1 0
0 0
59 0
353 46
56 4,401
2 ,5
2 2
0 2
0 1
0 10
0 1
10 42
15 81
0 8
1 20
0 5
0 7
0 44
1 674
0 93
0 10
5 40
0 5
0 188
0 188
3 30
0 37
0 506
0 36
0 4
1 302
96 6,742
I/ See Table 5.1.2 for result code description.
-39-
-------
Table 5.1.4 Screening Response Rates _!/
Sample
initial Telephone Screenings
No Previous Contact Sample
Missed HUs
No Donnelley Listings
Overall
Denver
0.720
0.524
0.563
NA
0.704
Washinjgtoja
0.791
0.282 21
0.364 2J
0.632 2J
0.758
Ij Response rate calculation in terms of final result codes from
Table 1.7:
Rate
01 + 02 + 03
Total - 05-30-31-32-33-72-73-82
2j Field screening response rates are low for Washington because
the field effort was minimized to control costs. Due to the
wide geographic dispersion of the Washington field sample, a
more exhaustive field effort was considered to be prohibitively
expensive.
-40-
-------
Nonresponsc adjustment to the sampling weights are discussed in Whit-
more, et al. [1983a] including the formation of weighting classes.
5.1.4 Selection of the Third Stage Sample i ^
As discussed in Section 5.1, the eligible individuals for
the CO study were the nonsmoking residents of the tiarget areas between
18 and 70 years of age. The field screenings yielded 139 eligible
individuals in responding households for the Denver samples and 214
eligible individuals for the Washington samples. The field screenings
were generated by the No Previous Contact Sample ,knd the Missed HU
Sample for Denver. They also included the sample tor the three area
segments with no Donnelley listings for Washington. All eligible
individuals from the Denver field samples were selected into the third
stage sample for CO monitoring. For Washington, jthe 208 eligible
individuals with a known telephone number were all selected for CO
monitoring.
j
The initial telephone screenings yielded 2093 eligible individuals
for Denver and 5209 eligible individuals for Washington. For each site,
a stratified simple random sample of eligible indiyidxials was selected
for CO monitoring as shown in Tables 5.1.5 and 5.1.6. The purpose of
the stratification was to oversample those individuals who appeared
likely to be exposed to the highest CO levels based upon their screening
data. Since the sample sizes defined by Tables 5.1; 5 and 5.1.6 were not
explicitly allocated to first-stage sampling units,;the number of sample
individuals selected from each first-stage unit is actually a random
variable. Hence, the sample design is not strictly a nested design,
which presents some problems for estimation of precision, as discussed
in Whitmore, et al. [1983a]. I
I-
The sample stratification for the initial telephone screenings is
shown in Tables 5.1.5 and 5.1.6. The strata are defined in terms of
four stratification variables. Based upon discussions with EPA staff
members, it was decided that commuting time was tihe most important
stratification variable. Respondents who had indicated a usual daily
one-way commuting time of 30 minutes or more were defined as belonging
to the "long commuting time" strata. All other respondents were defined
to belong to the "short commuting time" strata. It was decided that the
-41-
-------
Table 5.1.5 Third Stage Sample Allocation for the Denver Sample
Stratum
Number
L
2
3
4
5
6
7
8
9
TOTALS
Commuting
Time
Short
Short
Short
Short
Short
Short
Long
Long
Long
Gas
Appliance
Space
Heater
Space
Heater
Gas _!/
Stove
Gas Stove
Other
Other
Space
Heater or
Gas Stove
Other
Other
Attached
Garage
Yes
No
Yes
No
Yes
No
—
Yes
No
Screened
Eligibles
48
43
70
342
529
577
143
182
159
2,093
Sample
Size
48
43
56
148
229
250
74
78
74
1,000
I/ Gas stove, but not a space heater.
-42-
-------
0)
t-l
Cu
e
CO
Cfl
e.
o
4-1
60
fi
•H
CO
cfl
^
0)
4J
0
4-1
o
•H
cfl
O
0
r-H
<3
0)
p.
6
CO
en
01
60
CO
4U
Cfl
T3
•H
H
vO
•
i — i
m
cu
i-H
.0
CO
H
*o o)
CO 4J
01 4-1
t-l 01
. .J
CU
rH CU
CU H
CO CO
CO
CM CM 00 VO
CM CM -vf — 1
r— 1
vO vO CO vO
vO vO -d" < K &
cu
o
d
CO CO
CO -H
O tH
Cu
0) 0)
^ ^
^^«- o o
M i-
CO CO CO O CO CO
CU 0) CO 4J Cfl CO
CO 33 O CO O O
60
C
•H
4->
s
O *H
O H
4J 4-1 4J 4-1
O O 00
en co co co
e
s t->
4J CU
cc T£)
ij £«
rH CM CO «*
00 •— 1 CM CM — < 1X3 CM -• la i >H a iz >H S3
'
0) O) 0)
o o o
J-* 4-J 4-J 4J
>-t t-i oioicncoitn vi j-i
cu cu cj 4J 01 cu
rfirC COCO CO. CO CO ^S ^3
4J 4-1 CUOI CO CO Cfl 4-1 4-1
O CO CO33 O CJ O O O
4J 4J i
i-4 M 60 60601606060
00(3 CCCCG
^s . ,a o ooooo
co co i-j i_) >_] p4 ,-j ,_]
en
in vor— oo ONO i— i CM
-------
second most important stratification variable was a gas applicance
variable. This variable was defined to have the following levels in
terms of the gas appliances used at the respondent's residence:
(a) a space heater is used,
(b) a gas stove is used, but not a space heater, or
(c) neither space heater nor gas stove is used.
The screening questionnaire had probed for use of several types of gas
appliances in the- residence. It was decided that all types of gas
appliances other than space heaters and gas stoves were usually vented.
Hence, space heaters and gas stoves were considered the major sources of
CO generation within the home and were used for stratification. Rare
groups based upon these two stratification variables were oversampled,
and the remainder of the sample was proportionally allocated to the
remaining strata for each site. Some of the larger strata that received
a proportional allocation of the remaining sample were divided into
substrata. Since the substrata received a proportional allocation, the
additional stratification simply added control over distributional
characteristics of the sample.
Two additional variables were used to define substrata: Presence
of an attached garage and use of tobacco in the home. Individuals in
households that indicated an attached garage were assigned to one level
of the attached garage variable, and all other individuals were assigned
to a second level. The individuals from a residence in which someone
was identified as a tobacco user were assigned to one level of the
smoking or tobacco-use variable. The individuals in all other house-
holds were assigned to a second level of this variable. The attached
garage variable was considered to be the more important of these two
substratification variables. The smoking variable was not used for the
Denver stratification due to the smaller stratum sizes.
Another aspect of the third stage sample design was the allocation
of individuals to specific days within the sample period. A major
purpose of the CO study was to estimate the distribution of personal CO
exposures for the study populations during the study season. Of parti-
cular interest was the maximum personal CO exposure. Individual CO
exposure is heavily dependent upon several factors including: weather,
-44-
-------
location, and activity patterns. Since weather is such an important
factor, it was necessary to field as large a sample, as possible on each
day during the study period. Otherwise, there could be no one monitored
on the days with weather patterns producing the highest CO levels.
Since activity patterns are important, the sample participants could not
be allowed complete freedom of choice in selecting a day to be monitored.
The sample subjects could introduce a bias by selecting mostly days when
they plan to be inactive or stay at home. The strong influence of
weather and activity patterns upon CO exposure suggests a specific day
I
should be randomly selected for each individal to be monitored. How-
ever, it was anticipated that the response rate would be so poor as to
j
invalidate the study if only one day was offered fcjr each sample subject
to participate. Hence, the sample for Washington was randomly allocated
to non-overlapping three-day interview periods. likewise, the sample
for Denver was allocated to four-day interview periods. Four-day
interview periods were used for the Denver sample because each indivi-
dual selected for Denver was asked to participate for two consecutive
days. Each individual in the Washington sample participated for only
one day. ,
The allocation of individuals in the Washington sample to specific
three-day interview periods had a greater negative impact on the re-
sponse rate than had been expected. Some individuals indicated that
they were willing to participate, but not within the selected time
period. These individuals were given one additional opportunity to
participate by randomly reallocating them to one new three-day period.
A total of 550 individuals were reallocated in this manner for the
Washington sample. Reallocation to new time periods was also allowed
for the Denver sample. However, the method of reallocation was somewhat
different.
The third stage sample for Washington also incorporated a lead
letter methodology study. A sample of 596 individuals was selected to
receive a lead letter. A random subsample was selected from each of the
I
strata shown in Table 5.1.6. The lead letter informed the individual
that he or she had been selected for monitoring and that a telephone
interviewer would be calling soon. The lead letters appear to have had
-45-
-------
a positive effect upon the response rate. The overall response rate for
individuals selected into the Washington sample was about 58 percent,
but the response rate for individuals in the lead letter sample was
approximately 63 percent. (These response rates are calculated as the
number of individuals who agreed to schedule a monitoring appointment
divided by the number of individuals selected.) Hence, a person-level
response rate of about 65 percent may be possible for future studies of
this type. Third stage response a.nd sample weights are also discussed
in detail in Whitmore, et al. [1983a].
5.1.5 Third Stage Response
The distribution of final result codes for all individuals
selected for CO monitoring in Washington is shown in Table 5.1.7. It is
easily seen from the table that appointments were scheduled for about 58
percent of the individuals selected into the sample. However, due to
various factors, usable CO monitor data were obtained for only about 36
percent of the individuals sampled. Instrument failure and refusal to
carry the monitor were two of the major reasons for the low response
rate.
Thus, as shown in Table 5.1.4, approximately 76 percent of the
eligible households in the Washington screening sample responded. And,
from Table 5.1.7, usable CO monitoring data were obtained for about 36
percent of the individuals selected. These response rates could be
improved in the future by sending letters to all individuals selected
for monitoring, by reduction of monitor failure, and by making the
monitors less intrusive. Like most personal monitoring studies, the CO
study achieved a relatively low overall response rate. However, it may
be very plausible to presume that the CO exposures of respondents and
nonrespondents are alike within weighting classes. If so, the low
response rate is not as much of a problem as it might be in some other
type of study, e.g., a study of people's attitudes and opinions.
5.1.6 Variance Estimation and Screener Analysis
The sampling design for the CO study is a stratified, three
stage design. Area segments defined by 1980 Census block groups are
selected at the first s.tage. Donnelley listings are selected at the
second stage. However, the second-stage sample is a multi-phase sample.
-46-
-------
Table 5.
Result
Code
02
03
04
12
14
15
21
22
23
24
25
.26
30
99
1.7 Distribution of Final Result Codes for Individuals
Selected for CO Monitoring in Washington
Interpretation ;
No Contact After Appointment Scheduled
Need to be Rescheduled
Refused to Keep Appointment
Appointment Not Kept - Will Not
Reschedule
Refused After Field Contact
Snowstorm Forced Cancellation
Completed Data Collection; Unusable CO
Data
Partial Data Collected
Wrong Person Monitored
Monitor Malfunctioned
Other Result , Eligible for Monitoring
Ineligible (e.g., smoking or illiterate)
Usable CO Data
Unable to Schedule an Appointment
TOTAL
Frequency
N %
16 0.8
1 0.1
40 2.0
76 3.8
51 2.6
9 0.5
132 6.6
7 0.4
2 0.1
68 3.4
36 1.8
11 0.6
712 35.8
826 41.6
1987 100.1
-47- . '
-------
Each sample Donnelley listing is initially assigned to the telephone
phase. A subsample of the listings not covered by the telephone phase,
e.g., listings with no telephone number, is then selected for a second
phase. In particular, the No Previous Contact Sample is a field inter-
view phase of the second stage sample. Moreover, a third phase of the
second stage sample is selection of a subsample of listings for the
Missed HU Sample. Finally, the third stage sample of people for moni-
toring is not completely nested within the first-stage sampling units.
As a result of these design complexities, exact formulas for estimation
of variances and standard errors are complex, if not intractable.
Approximate formulas are generally used to obtain appropriate estimates
of standard errors with designs of this type. See Whitmore, et al.
[1983a] for additional details on this topic. In addition, Whitmore
also discusses estimating totals and proportions for the CO screening
sample as well as their associated standard errors.
5.2 Survey Activities
This section describes the field survey activities for the project.
As described in Section 5.1, survey activities occurred in two phases.
Phase I was an initial telephone screen in both Denver, Colorado and
Washington, D.C. Phase II involved another telephone interview to
identify a specific respondent and to set up an appointment for a field
interview. The field interview for the collection of personal exposure
monitoring (PEM) data and breath samples followed. RTI performed Phase
II in Washington, D.C. only.
5.2.1 Public Relations Efforts in Denver
On August 9, 1982, the survey task leader and other project
personnel met in Denver with various city, state, and EPA regional
officials. All aspects of the study were discussed. Substantive issues
discussed included data requirements, and placement of fixed site
monitors in relation to segments selected for personal monitoring.
Recommendations were made for placement of additional fixed site moni-
tors. Discussions also centered on the types of local support needed to
complete the project and included the need for public relations activi-
ties prior to each phase of telephone interviewing.
-48-
-------
5.2.2 Data Collection Instrument Development and Approval
Four data collection instruments were developed and reviewed
by internal project staff. After revision, they were sent for review by
the sponsor. Two of the forms were administered by the Computer Assist-
ed Telephone Interviewing (CATI) system and the sponsor reviewed them
for substance only. Those forms were the Household Screening Question-
naire and the Computer Model Input Questionnaire. The remaining forms,
the Activity Diary and the Study Questionnaire, were completed directly
by the respondents. All forms were finalized and: put into the OMB
package for approval. (Copies of the forms appear in the Field Inter-
viewer's Manual in Appendix C.) .
In early July a draft OMB package was prepared and submitted to EPA
for internal review. Based on this review, revisions were made and
copies were prepared and sent to EPA on July 27, 1982.
A final version of the OMB package was prepared and submitted on
August 18, 1982. (See Appendix D for a Table of Contents of the OMB
Package.) The revisions were based on comments supplied by EPA as well
as RTI reviews of the earlier draft submissions. The required copies of
the form were delivered to the project officer at EPA-RTP. A memo from
the EPA project officer denoting interim approval to proceed with the
telephone screening phase was received on August 19, 1982. Based on
this memo, all activities were continued according to the previously
prepared schedule. Formal OMB approval was received later in the month.
The OMB number (2080-0003) and expiration date (September 1983) were
placed on the data collection instruments.
5.2.3 Phase I - Household Screening Survey
' As described above, the first phase of the study was a
screening of selected households in Denver and Washington, DC. Infor-
mation dealing specifically with length of time spent in regular commut-
ing as well as demographies were collected about all members of the
household. Specific information collected about the housing unit
included presence of gas applicances and presence of an attached garage.
A Computer Assisted Telephone Interviewing (CATI) system was used
to perform this task. Local, experienced telephone interviewers were
hired and trained and interviewing was begun on August 24, 1982.
-49-
-------
5.2.3.1 Computer Assisted Telephone Interviewing (CATI)
System
The Computer Assisted Telephone Interviewing system
at RTI was used to administer the Household Screening Questionnaire and
the Computer Model Input Questionnaire used during Phase II. The CATI
system involves programming a computer so that questions are presented
on a screen in front of a telephone interviewer. The interviewer enters
answers from the respondent directly into the computer data base. The
answers entered then interact with the program to select the next
question to be presented on the screen and asked of the respondent. In
addition, this system provides immediate access to answers for analysis
and reduces clerical error introduced in multi-stage data handling
operations.
Programming of the Household Questionnaire for the CATI system was
started in July 1982. Final testing of the CATI system and the screener
were completed a week before the screening phase was implemented. The
speed of the machine was somewhat less than desirable, but did not have
an effect on the respondents' willingness to participate. After work
was completed on the development and testing of the screening question-
naire, initial development of the CATI programming for the Computer
Model Input Questionnaire was begun. Testing and refinement was done in
an iterative manner. RTI personnel acted as interviewers and respond-
ents during test phases and provided knowledgeable immediate inputs for
modifications of the system.
5.2.3.2 Telephone Interviewers
During July 1982, recruiting and hiring of tele-
phone interviewers for the household screening phase was begun by the
Telephone Survey Unit. In response to advertisements in local news-
papers, over 700 persons requested information about the interviewing
positions. All were sent applications and mock interview forms. Those
persons whose applications seemed promising were called and asked to
administer the mock questionnaire. From those deemed acceptable, the
Telephone Survey Unit (TSU) supervisor selected candidates for personal
interview and made offers to nineteen persons who accepted positions.
All nineteen were trained on Monday, August 23, 1982 and started inter-
-50-
-------
viewing after training. One of the nineteen was asked to move into a
task leader slot, and two others who were doing unacceptable work could
I
not be successfully retrained and left the project. Replacements were
recruited and trained when hired. The Table of Contents of the Tele-
phone Interviewer's Manual is given in Figure 5.2.1.
5.2.3.3 Interviewing i
Specifications for the execution of the Computer
Assisted Telephone Interview (CATI) data acquisition were provided to
the Telephone Survey Unit (TSU) for use during hiring of staff as well
as during project operations. A copy is provided In Appendix E.
Interviewing for the Household Screening Questionnaire was imple-
mented in full on Tuesday, August 24th. No major problems occurred and
the speed of the interview process increased as the staff gained famil-
iarity with the CATI system and the screener, and as they gained confi-
dence in their ability to use the system.
Telephone interviewing continued through September 21, with the
last seven days devoted to conversion attempts of cases which were
refusals or breakoffs. Approximately 30% of the recontacted cases were
converted to completed interviews. This conversion rate compared
favorably with in-person refusal conversions, the method considered most
effective. After all conversions were completed, a series of clean-up
procedures were applied to the cases. All cases coded as "other" were
'
categorized based on comments in the data file. All cases showing
pending codes were reviewed and placed into appropriate final codes.
Based on the final cleaned data, a final telephone response rate was
calculated by dividing the number of complete screening interviews
(6243) by the sum of the completes (6243), the refusals (1142), break-
offs (228), the partial interviews (45), and the number of final others
(347) [see Figure 5.2.2]. This final telephone response rate was 78.0%.
•l
Negative publicity in Denver, due to the report of an apparently
bogus survey asking highly sensitive questions, led to some problems
during the telephone interview. By delaying most of the Denver cases
until after a local EPA press conference, most of this negative influ-
i
ence was overcome. In general, the delay in the public announcement of
the study presented some obstacles which had to be overcome during the
•-51-
-------
Figure 5.2.1
TABLE OF CONTENTS OF THE
TELEPHONE INTERVIEWER'S MANUAL
(CO Exposure Study)
Page
I. Introduction I-1
A. Research Triangle Institute 1-1
B. Background and Purpose of the Study 1-1
II. Computer Assisted Telephone Interviewing (CATI) H-l
A. Introduction II-l
B. CATI Screen H-l
C. CATI Keyboard H-3
D. CATI Input H-4
E. CATI Control Features II-6
F. Error Messages * II-7
III. Administering the Questionniare III-l
A. Overview III-l
B. Reaching an Eligible Respondent III-l
C. Question-By-Question Specifications Ill-3
IV. Administrative Procedures IV-1
A. Terms of Employment IV-1
B. Confidentiality IV-1
C. Proj ect Interviewing Schedule IV-3
D. Control Cards IV-3
E. Scheduling Calls IV-6
F. Result Codes IV-7
G.
Accounting for Control Cards IV-11
-52-
-------
Figure 5.2.2
Final Telephone Interviewing Status Report - Phase I
Screening (Washington, DC and Denver, Colorado)
ListJng of Project Codes
FINAL CODES;
01 - Completed Interview
51 - Final Refusal or Breakoff [
52 - Refusal or Breakoff, Not Reached for Conversion
53 - Circumstantial NI
64 - Partial Interview ;
.
71 - Nonworking Number
72 - Nonresidential Number
73 - Entire Household Moving
74 - Wrong Address
75 - No Reliable Respondent (3 attempts)
76 - No Listed Phone Number
81 - Final Phone Problem (confirmed by operator)
82 - Final No Answer (6 attempts)
83 - Answering Machine (6 attempts)
84 - Final Busy (10 attempts)
91 and 95 - Other
NUMBER
6242
1142
228
16
45
1433
278
279
959
57
1814
15
710
57
8
347
-53-
-------
interviews. The use of RTI's toll-free number as a contact point and
later the provision of the EPA public affairs number helped allay the
fears of the respondents in both Denver and Washington, DC. After the
major press releases in the two cities, cooperation and participation
were more easily obtained.
5.2.4 Phase II - Washington, DC Area Survey •
The second phase of the study developed as two distinct
tasks. The first task was an additional round of telephone interviews.
The purpose of this round was twofold. The primary reason was to
contact a specific member of the household, who was the sample respond-
ent selected, and to further explain the study in an attempt to enroll
the selected respondent into the study. Establishing an appointment for
a field interviewer to bring the study materials and CO monitor to the
respondent was the successful endpoint to this telephone call. The
second purpose was to conduct a brief (approximately 25 minute) inter-
view to obtain information about each member (up to the eldest six) of
the household. The information sought was for use in one of the extant
computer models used to calculate carbon monoxide exposures. This round
of telephone interviews was done by RTI for the Washington, DC area
sample only.
The second task of Phase II was the actual field sampling. The
selected respondents were met at their home or at another convenient
location and given all materials. Each respondent carried a personal
exposure monitor for the twenty-four hours of their participation. In
addition, they carried an Activity Diary to record a description of all
their activities and they were asked to complete a self-administered
Study Questionnaire by providing information on themselves and their
home and work environments. Each time the respondent recorded an
activity in the diary, he/she had to push a button on the monitor to
record the corresponding CO value for that activity.
5.2.4.1 Telephone Interviewing
On October 25, a two-hour training session was held
at RTI for the telephone interviewing staff who worked on the second
phase of the project. The staff consisted for six interviewers and one
supervisor, all of whom had worked on Phase I. After training, work on
-54-
-------
the first wave (15 day sampling period - see Section 5.2.4.8) of inter-
.
views of respondents who would participate in the main field study
began. Plans were to arrange up to twenty appointments per day, but a
delay in the delivery of the monitors caused a reduction to a maximum of
ten per day for the first two week period. At the time of this decision,
151 appointments had been made. No further new appointments were made
for this time.period, but schedules were adjusted in an attempt to yield
ten appointments per day. A decision was also made to limit the second
wave of interviewing appointments (November 28 through December 12,
1982) to fifteen per day. •
.1 , . • -
Telephone interviewing continued through February with participants
from Waves 2 and 3 contacted during November, from Wave 4 contacted
during December, from Waves 5 and 6 during January, and from Wave 7
during February. !
After the seventh and final wave of telephone interviewing, the
following results were obtained. Out of 1987 cases assigned, 1126 had
been interviewed by telephone and had scheduled appointments. This is a
completion rate of 56.7% (1126/1987). A more accurate response rate can
be calculated by removing from the denominator those cases which ended
in a status no longer eligible for inclusion in the study. These
include those respondents who are physically or mentally incapable (13),
those respondents for whom there was a language barrier (20), those
cases where the respondent or the entire household had. moved (128), and
a share of those cases in the 'other' categories (30 cases where a
written comment indicates ineligibility). By removing from the denomi-
nator these 191 cases and 11 cases not worked because the sampling
quotas were completed, the response rate becomes 63^1% (1126/1785). The
refusal rate was 225/1785 = 12.6%. When the number of telephone inter-
views without established appointments is included in the calculations
by adding 153 cases to the 1126, a completion rate of 64.4% (1279/1987)
and a response rate of 71.7% (1279/1785) are obtained. Figure 5.2.3
displays all of these figures and their relationships:. Final rates,
which include the field screened cases as well as the telephone screened
cases described here, are presented in Section 5.1.5.
-55-
-------
Figure 5.2.3 Telephone Response Rates
1,987
Cases Assigned
11
Non-Worked
Cases
191
Ineligible
Cases
1,785 (89.8%)
Eligible Cases
225 (12.6%)
Refusals
1,279 (71.7%)
Telephone Interviews
281 (15.7%)
Other
Non-Interviews
1,126 (61.2%)
Interview and
Appointment
153 (8.3%)
Interview Only
-56-
-------
5.2.4.2 Final Document Preparation
During October, all materials for use in the field
were prepared, reviewed, and finalized. Copies of all documents were
delivered to the EPA project officer and/or to PEDCo for use in Denver.
Documents delivered included the second round Telephone Interviewer
Manual, a hardcopy version of the Computer Model Input Questionnaire
with instructions, the Study Questionnaire, the Activity Diary and
instructions, Worksheets and Screeners for the special field activities,
and the Field Interviewer Manual. RTI's Duplicating Department repro-
duced these documents and a Consent/Incentive Receipt Form for use in
the field. The Field Interviewer's Manual (Appendix C) contains copies
of the forms used in the field. A copy of the Phase II Telephone Inter-
viewer's Manual and the hardcopy version of the questionnaire are given
i
in Appendices B and F.
5.2.4.3 Protection of Human Subjects |
A research protocol was submitted to RTI's Commit-
tee for the Protection of Human Subjects during October. Permission to
proceed was received on October 21, 1982, before fieldwork began. The
protocol provides sufficient information for the committee to attest all
requirements for the Protection of Human Subjects jare being met within
the design of the project. The protocol was reviewed by the Coorporate
Vice-President. A copy of the protocol submitted and the review letter
are attached as Appendix G.
5.2.4.4 Field Staff Recruitment !
|
Potential interviewers were recruited from RTI's
National Interviewer File, from recommended interviewers from recent
studies conducted in the Washington area, and from responses to news-
paper advertisements. A newspaper advertisement for field interviewers
for the main field study was run in the Washington Post and local
suburban shoppers newspapers. The suburban weeklys generated the most
responses. The responses were screened, qualified applicants were
called, and appointments for personal interviews in DC were established.
Offers were made to those deemed suitable and a staff, of fifteen was
retained. Based on performance at training and during the early part of
the field study, staff adjustments were made. As the project proceeded,
-57-
-------
the length was extended and additional recruiting took place. This
recruiting involved an additional newspaper advertisement and personal
recruiting by the current staff. Training was done for new hires by the
field supervisor and her assistant.
5.2.4.5 Training the Field Staff
On November 1 and 2, all day training sessions were
held in Washington, DC for the field staff which consisted of thirteen
interviewers, a field supervisor, and an assistant supervisor. Training
covered all aspects of the study including special assignments, regular
assignments, problem resolution, and reporting. All staff members left
the training with a good grasp of the activities required. Each was'
given an initial special assignment on Tuesday afternoon (November 2)
and was asked to report to the field office on Thursday to have their
initial work checked and to receive additional special assignments and
their Wave 1 field assignment. With a single exception, no problems
were noted after careful review of the first special assignments. All
work was done according to specifications. The one exception was an
interviewer who had some specific problems and questions. An additional
one-half hour of training and some close follow-up of this individual by
the supervisor eliminated the problem.
5.2.4.6 Field Office
A field office/laboratory was established in the
offices of the Metropolitan DC Council of Governments. One office was
allocated for RTI use. The field office served several purposes. It
was the supervisor's office and was used to store supplies, maintain
records, create assignments, and supervise staff. It served as the
location to which all interviewers reported nightly to receive monitors
and data collection forms to take to respondents and to which completed
materials were returned.
5.2.4.7 Special Field Studies
As described in Section 5.1, three special field
studies were undertaken during November to assure complete coverage of
the target population. (See Section 5.1 for the rationale for each of
these studies.)
-58-
-------
5.2.4.7.1 Missed Housing Units .
Thirty segments were selected and ten
addresses were chosen in each. Each address was located and used as the
Start of an interval to be checked for missed unitss. The housing unit
to the left of the start point was identified and its address checked
against the Donnelley listing for the segment. If it was found on the
listing, then the process was complete. If not, the interviewer com-
pleted a Household Screening Questionnaire, or obtained a phone number
for the housing unit. The interviewer proceeded tip the next unit (to
the left) and repeated the process. Data from this activity showed a
low yield of missed units which is encouraging for future use of the
Donnelley lists.
5.2.4.7.2 -Segments With No Donnelley Listing
Three segments in the Washington, DC
area had no listed housing units on the Donnelley lists. These segments
were counted and listed using standard procedures. In two of the
segments every house was selected for inclusion, prhile in the third
segment every other house was selected. Members ;from each selected
house were given a chance to complete a Household Screener or to provide
a telephone number for later screening from RTI. i
5.2.4.7.3 No Previous Telephone Contact
A sample of cases where there ,was no
telephone contact during screening was selected. These cases were
worked by the field staff. Respondents were asked,to complete a screen-
er or to provide a correct telephone number for interview by phone from
RTI. , . I
All three special field studies were completed with information
returne.d to 'RTI for data entry or subsequent telephone interviewing.
During December, the telephone interviewers completed the special field
sample activities by entering data obtained in the field or by using the
phone numbers obtained to call identified sample units and complete the
screening interview. Both processes were done using the CATI system.
5.2.4.8 Regular Field Assignments j
CO data were collected in th4 field in Washington,
D.C. from November 8, 1982 through February 25, 1SJ83 with the exceptions
-59-
-------
of November 24 - 27, 1982 (Thanksgiving weekend), December 22, 1982
through January 3, 1983 (Christmas week), and February 10 - 14, 1983
(heavy snowfall). Cases were assigned to the telephone unit in waves.
Original plans called for twenty cases to be assigned to the field per
day for each of the fifteen days in a wave. The use of waves of assign-
ments was done in order to keep the time between the telephone contact
and the in-person appointment reduced to a workable amount. People are
most often reluctant to make definite appointments too far in advance,
and if they make the appointment, may forget and make other plans. The
use of waves -reduced the interval to a maximum of 17 days between
telephone call and data collection appointment. The telephone staff was
allowed to schedule appointments on any of the three days indicated for
each case (see Section 5.1), thus allowing some flexibility for the
respondent. After all cases in a wave were worked, those cases with
scheduled appointments were sent to the field. The field supervisor
divided the control cards into assignments for the interviewers.
Assignments were created with an attempt to minimize driving time. Each
field interviewer was required to make a reminder telephone call to the
respondent at least twenty-four hours before the appointment. This call
reconfirmed the appointment and was used to get specific directions to
the house.
Prior to the appointment, the interviewer reported to the central
office and was issued a monitor and all required data collection forms.
The interviewer went to the respondent's home, further explained the
study, obtained informed consent, and made an appointment to return in
twenty-four hours to retrieve all materials and to pay the incentive.
Cases were first assigned to the interviewing field staff on
November 4. This allowed sufficient time for the required reminder
calls to be made before the initial PEM delivery appointments which
started on November 8. In general, no major problems were encountered
in the field. However, the amount of driving involved in getting to and
from the respondents' homes created a logistics problem as did creating
an interviewer's work assignment which permitted time off. The addition
of staff reduced these problems. Problems with the monitors created
some lost data situations, but, by the end of November, this was being
-60-
-------
reduced. There were some problems with rescheduling; broken appointments.
An immediate attempt was made to reschedule within the three-day sampling
period. When unsuccessful, the case was returned to RTI for rescheduling
in a different wave. This process enabled RTI to maintain a high rate
of participation. All activities ran until December 22 when work was
stopped for a holiday break. Other than one day of field activity lost
due .to inclement weather (heavy snowfall), all activities proceeded as
planned.
The field response rate was somewhat lower than planned during
January. This was attributed to continued monitor failures and to a
somewhat higher than expected refusal rate. Broken appointments or
requested reschedules also decreased the completion rate.
Three snow days in Washington delayed field sampling during early
February. The delay caused the rescheduling of the three days' appoint-
i
ments to three days appended at the end of the scheduled wave. The
rescheduling of appointments was handled by the field interviewers,
since all RTI telephone operations had been terminated. During
I
February, the newly developed monitoring package, involving the
Hewlett-Packard calculator and interface was made available. From the
units available, equipment problems reduced the number of opportunities
to place the monitor with respondents, and minimal data were collected.
i
Most respondents liked the new device, and, in several instances, a
different member of the household had carried the original PEM, allowing
for good comparative data. The problems reported by the respondents
included the size of the keys (too small to push easily) and the size of
the lettering of the labels, also too small.
All data and forms were received from the field by the end of
1
February and batched and logged in according to instructions prepared
for this phase of the study. Counts of each document were used to
account for all cases and were used to determine final response and
refusal rates.
5.2.4.9 Breath Sampling
Each respondent was asked to provide a breath
-
sample at the end of the 24-hour monitoring period,. Following a stand-
ard protocol, which the interviewer read to the respondent, the respond-
-61-
-------
ent was asked to take a deep breath and hold it, and then expel it
fully. Another deep breath was taken and held. After half of the
second breath was expelled, the remaining volume was collected in the
sample bag through a disposable mouthpiece. The sample bag was sealed,
labelled with the respondents unique study number, and returned to the
lab for analysis. After analysis was completed, the bags were recycled
into the system. The repetative use caused some problems due to leakage
of sampling bags and loss of samples (i.e., some bags sent back to the
field developed leaks after being used several times). This loss of
samples was the only major field problem encountered. There .appeared to
be no major difficulty involved with the actual collection of the
samples.
At the request of EPA, RTI also collected breath samples from
children ages 5-7 in households of respondents. Breath sampling of
children started slowly as few children in the age range of 5-7 years
were found. Field staff were instructed to expand the age range to 4-8
during December. The acceptable age range was further expanded during
January.
Acquisition of breath samples from children continued slowly. Even
with the increased age range and increased interviewer awareness and
diligence, only 12 samples were collected.
5.3 Field Measurements and Quality Assurance
The field measurement of personal exposure levels of carbon monox-
ide was accomplished using battery-powered, portable CO monitors. Two
configurations of the CO monitor were evaluated in Washington, DC.
However, virtually all of the ambient data were acquired from one
configuration. The use of such monitors required extensive, daily
technical support on-site to keep the monitors functioning properly.
This support was provided from an on-site, field laboratory staffed with
two full-time technicians working seven days per week throughout the
study.
5.3.1 Description of the Ambient Monitors
The monitoring of ambient levels of CO was accomplished
using two configurations of portable, battery-operated monitors which
were assembled from commercially-available subassemblies by Rockwell
-62-
-------
International for EPA. Each of the monitor configurations was composed
of two major subsections — the CO monitoring subseiction and the data
acquisition subsection. In each configuration, the CO monitoring
Subsection consisted of a specially modified General Electric CO-3
carbon monoxide monitor. This monitor operated on the; principle of a
reversed fuel cell. A small, diaphragm pump within the monitoring
i
subsection drew a sample of ambient air into the monitor through a
prefilter designed to remove reducing species (e.g., alcohols, alde-
hydes, ketones, etc.) within the sample. The sample was then passed
through the detector cell where it came into contact with a moist,
polymeric membrane containing a wiring grid which conducted a small,
constant electric current. In the presence of the el€>ctrical current,
the CO in the sample reacted with the water in the membrane to form C0_
and H_. This reaction resulted in a slight alteration in the grid
current. The current alteration was sensed by the monitor's electronics
which transformed the signal into an analog output voltage proportional
to the CO concentration in the sample. The electronics package within
the monitor contained provisions for adjusting the zero- and span-level
responses as well as for electrically adjusting flow rate by controlling
pump speed.
The two monitor configurations differed in their data acquisition
subassemblies. One configuration, the model COED-1, utilized a data
acquisition package supplied by Magus, Inc. The second configuration,
the so-called GE/HP model, utilized a Hewlett-Packard HP-41CV program-
mable calculator and a HP-IL interface loop and converter as the data
acquisition unit. In both configurations, the continuous analog voltage
from the GE monitoring unit was fed into the data lunit where it was
formatted into time-period averages which were stored in on-board memory
for later manual acquisition. The definition of the time-period averages
was accomplished either by an on-board clock (i.e,time periods which
'" • I
were hourly averages) or by userr-initiated "activity signals" (i.e.,
averages based on periods of specific activities). The output from the
Magus data unit consisted of time/average pairs where the average datum
represented the true arithmetic average of the CO concentration for a
time period and the time datum represented the actual time, in 24-hour
-63-
-------
clock, that the pair was stored (i.e., the time that the time or activ-
ity period was completed). The scanning rate for this data unit was
about 6 per minute. The average was computed by accumulating the sum of
the individual scan values obtained during an individual time period and
dividing it by the number of scans within the period at the end of the
period. The Magus unit was capable of storing about 120 time/average
pairs, but was capable of defining only one type of activity. The
program operating the Magus was primarily "hard wired" (i.e., unaccess-
ible to the user), although some limited operating parameters were
user-selectable.
The HP data unit also output time/average pairs, but was capable of
being programmed to retain in memory (and, subsequently output) addi-
tional data with each pair. Examples of such additional data included
the minimum and maximum scan value within a time period, the standard
i
deviation of period data, the number of observations (i.e, scans) within
a period, etc. Since the HP system was based on a continuous-memory,
fully-programmable, scientific function calculator, any statistic
desired could be computed and retained for each time period subject only
to the time constraints of the data acquisition cycle and the available
memory of the calculator. The memory of the unit, as configured for
this project, approached 4,800 bytes thereby allowing the user a fair
amount of flexibility concerning what data and how much data was acquir-
ed during a sample. The HP units were programmed for this project to
differentiate between ten different activity designations. The activity
designation was stored with the time/average pair. The scanning rate
was programmed at 4 per minute. The unit output data to a small-format,
thermal printer, but could be configured to off-load stored data to
magnetic tape or to a mini-computer.
5-3.2 Description and Verification of the Field Standards
The field standards used for calibrating the CO monitors
were generated from multiple cylinder gases at pre-selected concentra-
tion levels rather than from dynamic dilution of a single, high concen-
tration standard. This alternative was chosen to minimize the time
required for standard preparation in the field. An additional benefit
derived from the use of fixed concentration standards was the elimina-
-64-
-------
tion of the day-to-day variation in standard concentration levels.
Cylinders of carbon monoxide in air were obtained from Airco Industrial
Gases, RTP, NC at concentration levels of 10, 50, 100, 150, and 200 ppm
(v/v). A certificate of analysis was required with each cylinder
ordered. Additionally, prior to use in the field, each cylinder was
verified in the RTI Environmental Standards Laboratory against NBS SRM
or CRM standards or against RTI GMPS (Gas Manuf acturjer's Primary Stand-
ard) cylinders. '
Since the Airco standards were composed of a synthetic air matrix
.
(i.e., pure oxygen and pure nitrogen blended to approximate the compo-
sition of air), the oxygen content of each cylinder brdered was verified
to be 20.8 ± 0.5 percent. Since EPA ruggedness testing of the monitors
prior to the project had revealed no carbon dioxide or water vapor
interference, these gases were omitted from the standard gas matrices.
Airco 0.1 Grade Zero Air was utilized as the i?ero-level matrix.
This gas also was dry and contained no CO.. It was certified by Airco
as containing less than 0.1 ppm CO and was verified|at RTI at the same
level. i
.
5.3.3 Preparation of CO Monitors for_the Acquisition of a CO
Exposure Sample !
The following procedure was performed on a daily basis on
every monitor assigned to a project sample. In the early afternoon,
prior to delivery to the field interviewers, the sampler was provided
with fresh batteries (if its batteries had not been charged overnight
the previous night), powered-on, and allowed to operate on a maintenance
charger for approximately 1 hour. During this warm-up period, the
monitor's clock was set (if necessary), the user-selectable program
options were set, and the pre-scrubber and cell water reservoir were
checked and serviced, if necessary. The pre-scrubber was replaced or
refilled with 1/16" Purafil® (potassium permanganate coated on silica)
spheres when its normal pink color had changed to brown halfway down the
column. The water reservoir was refilled with deionized water whenever
it was found less than half full.
Following the warm-up period, the monitor was moved to the calibra-
tion manifold and, again, connected to a maintenance charger. Monitor
-65-
-------
flow rate was measured and adjusted to 70 ± 10 seem, if necessary. If
this flow specification could not be met, the monitor was removed from
service until the problem could be diagnosed and corrected. The monitor
was subjected to a two-point, zero/span check at CO concentration levels
of 0.0 and approximately 50 ppm. The two-point check was deemed adequate
due to the high degree of linearity (0.9993 £ r2 ^ 0.9999) demonstrated
by pre-study, post-study, and within study multipoint calibrations
covering a 0 - 200 ppm range. Zero and span responses were monitored
both at the analog voltage output from the GE monitoring section and at
the digitized and integrated (5 minute integration periods) output from
the data unit. The zero response of the monitor was adjusted to a
nominal level of 1.0 ppm whenever response to the zero concentration
matrix fell outside of the 0.5 to 2.5 ppm range as determined at the GE
monitor analog output. A nominal zero setting of 1.0 ppm, rather than
0.0 ppm, was chosen to avoid the likelihood of negative responses to
near-zero concentration levels due to monitor drift. This was necessary
because the Magus data unit interpreted all incoming negative data as
the absolute of the data, thus leading to possibly large errors in
accumulated averages. The span response was adjusted whenever the
response to the span matrix varied by more than ± 5 percent from the
nominal value as determined at the GE monitor analog voltage output.
After all adjustments were completed, the responses to the zero and span
matrices were redetermined and the slope and intercept of the response
curve were computed based on the 5 minute integrated data output from
the data unit.
Following the zero/span operation, the monitor was removed from the
maintenance charger and allowed to operate for 5-10 minutes. During
this time the sample operating parameters were established (e.g., memory
clear, auto-log mode on, time-display mode selected, and data logging
enabled). After 5-10 minutes had elapsed, the battery voltage for each
of the battery packs was measured at the pack. The GE CO monitor
battery voltage was expected to be 5.65 ± 0.10 volts, the Magus voltage
was to be 8.40 ± 0.10 volts, and the GE/HP voltage was to be 5.2 - 6.4
volts. If these voltage specifications were not met, the battery pack
was replaced with a freshly charged one. If the specification was still
-66-
-------
not met, the monitor was removed from active status until the problem
was diagnosed and corrected. Following the battery check, the monitor
cover was installed and secured and the monitor was stored (with the
sample pump running) on a maintenance charger until \ delivery to the
field interviewer (FI). The monitor was delivered to the FI within 0-2
hours of calibration, generally. '
i
Following the sample period, the monitor cover wass removed, the
monitor was inspected for obvious problems (e.g., dead batteries,
depleted prescrubber, physical damage, etc.), the sample pump was turned
off, and the data unit was placed in the "display" mode for data re-
covery. The data were transcribed from monitor memory to the field data
sheet by the field interviewer. Ten percent of the incoming monitors on
any given day were subjected to a QC data reread where the monitor-to-
data-sheet transcription was checked point-by-point. Discrepancies were
resolved immediately and the FI was notified of the findings and resolu-
tions.
I
The battery voltage(s), the water reservoir levjel, and the pre-
scrubber condition were, again, checked and noted. The zero/span check
was repeated, however, no adjustments were made. This post-sample check
was performed with the monitor operating on its internal batteries only
and not connected to a maintenance charger. Again,. the slope and
intercept of the response curve were computed based on integrated output
from the data unit. The post-sample response curve was compared with
1
the pre-sample curve and an appropriate data validity code was assigned
to the sample. All computations connected with the pre- and post-sample
zero/span operations including calculation of percent variation, slope,
and intercept and assignment of validity code were performed by a
Hewlett-Packard Model 41-C programmable calculator operating under the
control of a program prepared by RTI field personnel ^specially for this
•'
project.
The following specifications defined the range of each validity
code: i
-67-
-------
Code
1
2
3
4
where:
Slope
|AM[ < 5%
5% 15%
Intercept
JAbj < 1.0 ppm
1.0 ppm __ 2.0 ppm
AM
M,
x 100
Ab
and:
M. = post-sample slope,
MI s pre-sample slope,
b» = post-sample intercept, and
b, - pre-sample intercept.
Whenever the codes for the slope and intercept differed, the most
conservative one (i.e., the code of higher number) was chosen.
The computation of applicable slope and intercept for ambient data
reduction was based on the validity code assigned to the data. If the
assigned code was 1, then the applicable slope and intercept were equal
to the pre-sample zero/span slope and intercept. If the assigned code
was 2, 3, or 4, then the applicable slope and intercept were the averages
of the pre- and post-sample slopes and intercepts, respectively.
Monitors which failed to complete a post-sample zero/span check
were assigned a validity code of 5 and all ambient data from the sample
were flagged as invalid. This situation usually occurred because of
insufficient battery capacity, but may also have been caused by data
unit malfunctions such as "lock-up" or "mode shift".
All data derived during zero/span operations were recorded on a
"Monitor Status Sheet" such as those depicted in Figures 5.3.1 and
5.3.2. Ambient data recovered from monitor memory were transcribed onto
the "Field Data Sheet" depicted in Figures 5.3.3 and 5.3.4. Zero/span
data were also transferred to control charts describing the course of
-68-
-------
CO EXPOSURE STUDY, WASHINGTON, DC
MONITOR STATUS SHEET
MONITOR EPA NO.
PID #
Sample Date
PARAMETER
Date
Time
Analyst
Barometric Pressure
Laboratory Temp.
SHEET NO.
NOMINAL
RANGE
750-770 mm
20-30*C
COND. CODE'
CO STO ID
CO STD Cone
PRE-SAMPLE
VALUE
POST-SAMPLE
VALUE
Zero Response (Unadj) 0 +_ 2 ppm
Span Response (Unadj) '"SO ppm
Span Variation (Unadj) -_+ 5%
Zero Response (Adj )
Span Response (Adj)
Span Variation (Adj)
0 +_ 2 ppm
*> 50 ppm
+ 5%
xxxxx
xxxxx
xxxxx
Integrator Zero Resp.
Integrator Span Resp.
0 j^ 2 ppm
w 50 ppm
Zero Intercept
Slope
0 +_ 2 ppm
1.00 + 0.05
Flow Rate (Unadj)
Flow Rate (Adj)
70 _+ 10 SCOT
70 + 10 seem
xxxxx
Battery Voltage
CO-3 Unit
Integrator
5.65 +_ 0.05 volts
8.40 * 0.05 volts
4.95
7.75
O.OSv
0.05v
Water Level
Pre-Scrubber level
1/2-3/4 Full
1/2-1/1 Pink
COMMENTS:
Figure 5.3.1. COED-1 Monitor Status Sheet;
-69-
-------
MONITOR EPA NO.
PID #
Sample Date
CO EXPOSURE STUDY, WASHINGTON, DC
6E/HP MONITOR STATUS SHEET
PARAMETER
Date
Analyst
Barometric Pressure
Laboratory Temp.
SHEET NO.
NOMINAL
RANGE
750-770 mm
20-30*C
COND. CODE
CO STD ID
CO STD Cone
PRE-SAMPLE POST-SAMPLE
VALUE VALUE
Zero Response (Unadj) 0 _+ 2 ppm
Span Response (Unadj) 50 ppm
Span Variation (Unadj) +; S%
Zero Response (Adj) 0 i 2 ppm
Span Response (Adj) 50 ppm
Span Variation (Adj) £ 5%
xxxxx
xxxxx
xxxxx
Integrator Zero Resp.
Integrator Span Resp.
0 +; 2 ppm
50 ppm
Zero Intercept
Slope
0 j* 2 ppm
1.00 + 0.05
Flow Rate (Unadj)
Flow Rate (Adj)
70 £ 10 seem
70 + 10 seem
xxxxx
Voltages
Pump
Batteries
6.4 to 5.2 volts
Cell Temperature
Water Level
Pre-Scrubber level
20 to 30 deg
1/2-3/4 Full
1/2-1/1 Pink
COMMENTS:
Figure 5.3.2. GE/HP Monitor Status Sheet
-70-
-------
CO EXPOSURE STUDY, WASHINGTON, DC
FIELD DATA SHEET
Monitor ID:
Status Sheet No.
PID f:
Read by: | 1 1
Verified by: | | 1
Approved by: ] [ [ ]
NOTE: 1 - Use back of form
2 - Record all times
Seq.
No. Time Value (ppm)
01 1 | | 1
02 I I I
03 | | |
04 | |
OS | |
06 I I
07 I | |
08 1 II
09 | |
10 (
11 LZ
12 1 —
13 1 1
14 I I
15 IZ
16 1 ' "
17 Q
18 I I
19 Q
20 | '""
21 1 1
22 | | ,
23 CI
24 LZ
1 !TI 1
1 1,1 1
1 ITI 1
LZ
LZ
1
EZ
1
1 1 1
Z3.O
Date Sampled: j j
Date Read: 1 1
Date Verified: | |
• Data Approved: | |
to describe any problems.
in 24-hr clock.
Seq.
No. Time
25 | | |
26 | | |
21 \ 1 II
28 I I I I
Z9 LI I 1
Z3.EH 30 i i ii
zi.a
zi.a
zi.a
|TL7I
zi.a
zLa
zi.a
zi.n
zi.a
zi.a
zi.a
zi.a
zi.a
31 1__L__J_J
32 [" "][ |
33 I I ||
34 L_L-__U
35 I I II
36 r i ii
37. i i l l
38 | | ||
39 1 1 1 1 1
40 | | I I I
41 | 1 1 1 1
42 I I I I I
43 |""| | '-| |
| ^ j 44 | | j | |
zi.a
in .a
Z3.O
zi.a
45 I | | I |
46 I I I I |
4/ I 1 1 1 1
48 I I . | I I
Month Day Year
ca co EZO
CO r~n
co ca
ca ca
Value (ppm)
aza.a
aza.a
aza.a
azo.a
azo.a
aza.a
Lzcia.a
azo.a
aza.a
aza.a
aza.a
aza.a
izizz:.a
aznn.izi
aza.a
aznn.a
azo.a
If data continued on back of page, check here
Data Validity Code: | |
Applicable Slope: f~1 . | | ) | Applicable Intercept: f~| . FT") I
Figure 5.3.3.
Field Data Sheet, Side 1
-71-
-------
PID *:
Page 2
Seq.
No. Time
49 1 1 1 1
50 | | | |
51 1 1 1 1
52 I I I I
53 | | | |
54 | | | |
55 | | | |
56 | | 1 i
57 | | | |
. 58 | | | |
59 1 1 1 1
60 | | | |
61 1 II 1
62 1 1 1 1
63 | | 1 1
64 1" | i |
65 | | | |
66 | | ||
67 I I I I
68 | | 1 1
69 1 1 1 1
70 | | 1 1
71 1 1 1 1
7H I I I I
Comments :
Value (ppra)
1 1 1 1 IT| 1
1 1 1 1 l.l 1
1 1 1 1 LI 1
1 1 1 1 I.I 1
1 1 1 1 I.I 1
| 1 1 1 LI 1
1 1 1 1 LI 1
1 1 1 1 ITI 1
| 1 1 1 I.I 1
1 1 1 1 I.I 1
1 1 1 1 I.I 1
| 1 1 1 I.I 1
| 1 1 1 I.I .1
1 1 1 1 l,ljrl
1 1 1 1 I.I 1
i i i i U,l
1 1 1 1 I.I 1
till 1,171
1 1 1 1 I.I _.l
) 1 1 1 LI 1
J 1 1 1 I.I 1
J 1 1 1 LI 1
J 1 1 1 1,1 1
1 1 1 1. Ij 1
Seq.
No.
73 ~
74 ~
75 ~
76 ~
77 £
78 ~
79 -
80 ~
81 ~
82 ~
83 ~
84 ~
85 ~
86 £
87 '
88 "
89 "
90 '
91 "
92 [
93 £
94 [
95 [
96 [
Time Value (ppra)
1 1 1 1 1 LI 1
1 1 1 1 1 I.I 1
1 1 1 1 I.I 1
1 1 1 1 1 I.I 1
1 1 1 1 L! .1
L ill ••! 1,1 I
1 1 1 1 1 1 I.I 1
1 1 1 1 1 I.I 1
1 1 1 1 1 I.I 1
1 1 1 1 LI 1
1 1 1 1 1 IJ 1
1 1 1 1 1 LI 1
1 1 1 1 1 I.I 1
i i i i i i_r~i
II 1 1 I.I 1
1 1 1 1 1 1 I.I 1
1 1 1 1 1 I.I 1
II 1 1 1 1 I.I 1
1 1 1 1 1 1 1 I.I 1
1 1 1 1 1. 1 1 I.I 1
1 1 1 1 1 1 1 I.I 1
1 1 1 1 1 1 1 LI 1
1 1 1 1 1 1 1 I.I 1
1 1 1 1 1 1 1 IJ 1
Figure 5.3.4. Field Data Sheet, Side 2
-72-
-------
differences between pre-sample and post-sample span, zero, battery
voltage, and flow rate values. Examples of these charts are presented
and discussed in Section 6.3.3.5 of this report. Complete files of the
i • -
data sheets and control charts, organized on a monitor-by-monitor basis,
will be transferred to EPA at the conclusion of the project.
5.3.4 Analysis Method for Carbon Monoxide in Breath
A study was undertaken to develop and evaluate a method for
i
collecting and measuring alveolar carbon monoxide prior to the field
study. The effects of sampling bag, sample storage, measurement time,
and instrument interferences on measured CO were investigated using a
single General Electric CO-3 monitor identical to the ones used for
breath analysis in the field. Calibration curves w'ere generated under
several conditions to assess both precision and accuracy of the measure-
ment method. Finally, the precision of the final method was tested by
collecting and measuring four breath samples from each of eight nonsmok-
ing subjects. Since EPA loaned RTI only one CO-3 monitor and the field
study started within two weeks after the instrument was received, a very
limited amount of time was available to conduct tjhesse experiments.
Accordingly, the number of determinations for each jexperiment were not
statistically designed. This led to an unequal number of determinations
i
in the various experiments.
5.3.4.1 Description of Method
Alveolar CO was measured using a General Electric
CO-3 monitor equipped with an on-line activated charcoal/Purafil® pre-
filter to remove potential interferences from breath samples. The
filter was prepared by filling a 10 mL disposable pipette with 9.0 cm of
Purafil® (potassium permanganate coated silica spheres) and 9.5 cm
activated cocoanut charcoal. The Purafil® consisted of 1/16" spheres
and the charcoal was 6-14 mesh. The filter was attached to the sample
inlet of the CO monitor at the end containing Purafil®. A strip chart
recorder was attached to the monitor for output signal recording.
The following technique was used to acquire a field sample. The
participant was instructed to take a deep breath then expel all air from
his lungs. He took a second deep breath and held this breath for 20
seconds. He then expelled the first part of his breath into the room
-73-
-------
and the last portion into a sample bag. The bag was sealed by clamping
the inlet tube and was transported back to the field laboratory for
breath analysis. For analysis, the bag was attached to the GE monitor
With the prefilter in line and the measurement was taken after the strip
chart trace had stabilized. The amount of CO in the sample was deter-
mined from the recorder trace by transforming it to a concentration
value according to the monitor's calibration curve.
5.3.4.2 Instrument Noise
The General Electric CO-3 monitor was tested for
output signal noise while monitoring zero gas (0.0 ppm CO) from a gas
mixing manifold. Instrumental noise was characterized as one half the
peak to peak variation in the signal output. Figure 5.3.5 shows the
recorder tracing obtained during testing. Noise was determined to be at
±0.2 ppm for this instrument.
5.3.4.3 Instrumental Response Time
Instrumental response time was determined for the
GE CO-3 monitor using CO concentrations ranging from 5 to 50 ppm. The
strip chart recorder was used to monitor instrument output over time.
Measurements were made with the monitor attached directly to the CO
source with and without the charcoal/Purafil® prefilter in-line. Under
both conditions, gas flow through the monitor was adjusted to 75 mL/min.
The monitor response was considered stable once the recorder had
maintained the same reading for at least 30 seconds. Stabilization
times for all conditions are listed in Table 5.3,. 1. Stabilization times
varied from 1.5 to 7.5 minutes with the longest times required for the
highest CO concentrations. In addition, the times required for the
monitor to reach 90 and 95 percent of the final stabilized reading were
measured and have been listed in Table 5.3.1. These times were also
dependent on concentrations, but in all cases, values were less than 2.5
minutes. The increased stabilization time for those samples employing
the in-line filter was due to the air volume contained within the
filter. These data demonstrated that for the CO concentrations expected
in nonsmoking subjects (10 ppm or less), the instrument with the pre-
filter in-line will reach 95 percent of the actual CO concentration in
less than 2 minutes and will reach stabilization after 4.5 minutes. At
-74-
-------
5.0 —
Time (minutes)
Figure 5.3.5 Instrument Noise
-75-
-------
Table 5.3.1 Instrument Pesponse Times in Minutes
CO
Concentration
(ppm)
5.0
5.0 4/
8.0
8.0 4/
10.0
10.0 4/
31.0
31.0 4Y
50.0
50.0 47
90%
0.3
1.0
0.5
1.4
0.4
0.6
0.5
1.3
0.7
1.6
Percent Stabilization
I/ 95% 21
0.4
1.4
0.7
1.7
0.5
0.8
0.9
1.7
1.7
2.3
100% 3/
1.5
2.8
4.0
4.5
1.5
2.0
4.5
6.5
5.5
7.5
JY Time at which the instrument reached 90% of the final
stabilized reading.
2j Time at which the instrument reached 95% of the final
stabilized reading.
3j Measured after instrument had given a stable reading for
30 seconds.
4/ Measured with prefilter on-line.
-76-
-------
an air flow of 75 mL/min, these times correspond to,a sample volume of
150 and 420 mL, respectively. This information verified that a 600 mL
sample bag could provide an adequate sample volume for accurate read-
ings, i
As well as determining the stabilization time for the monitor to
respond to the presence of CO, the time required fbr the monitor to
reequilibrate at a zero level after exposure to CO was studied. It was
determined the the GE monitor would equilibrate at zero within two
minutes after exposure to CO levels as high as 15 ppm.
5.3.4.4 Sample Bags - Recovery Study l
Experiments were performed to determine the amounts
.
of CO lost from spiked air loaded into two types of sampling bags: 600
mL blood transfer bags (Fenwal, Inc.) and 1 L carboxyhemoglobin bags
(Energetics Science Division, Becton-Dickenson). During testing, CO
|
concentrations were first measured by attaching the CO monitor directly
to the gas mixing manifold. Finally, CO concentrations in the bag air
were measured immediately after filling the bags, ; Experiments were
performed using CO concentrations ranging from 2.5 t:o 15.5 ppm with both
dry and humid air (75% relative humidity). A new sampling bag was used
for each measurement. A single measurement was performed for each
condition. Results are given in Tables 5.3.2 and 5.3.3. The data show
that a reproducible loss of CO occurred for each bag regardless of CO
concentration. The Fenwal bags showed an average loss of 1.7 ± 0.5 ppm
CO while the carboxyhemoglobin (CBH) bags showed a;loss of only 0.3 ±
0.2 ppm CO.
With both bags, there appeared to be greater CO losses when dry,
rather than humid, air was used. Due to high CO losses, the Fenwal bags
were considered unacceptable. All further testing;was performed using
only the carboxyhemoglobin bags.
5.3.4.5 Sample Contamination from CBH Bags
Contamination of samples resulting from their
storage in the CBH bags was examined. Humidified zero air was placed
into three bags and analyzed immediately. Next, humidified zero air was
placed into three bags and stored for approximately 20 hours. No
elevation of the zero CO response level was observed for either set of
samples.
-77-
-------
Table 5.3.2 Loss of CO From Fenwal Sampling Bags
Air Type
humid
dry
humid
humid
dry
dry
humid
dry
dry
humid
dry
Measured
Manifold
15.5
15.5
12.3
8.8
8.8
8.0
5.8
5.8
2.8
2.5
2.5
Average Humid ± S.D.
Average Dry ± S.D.
Average Both ± S.D.
[CO]
Bags
13.8
12.8
10.3
7.8
7.0
5.8
4.3.
3.8
1.4
1.3
.8
(ppm)
CO Lost
1.7
2.7
2.0
1.0
1.8.
2.2
1.5
2.0
1.4
1.2
1.7
1.5 ± 0.
2.0 ± 0.
1.7 ± 0.
4
5
5
Table 5.3.3 Loss of CO From CHB Sampling Bags
Air Type
humid
dry
humid
dry
dry
Measured [CO] (ppm)
Manifold Bags GO Lost
7.0 6.8 0.2
7.0 6.8 0.2
3.0 2.8 0.2
3.0 2.4 0.6
2.8 2.4 0.4
Average Humid ± S.D. 0.2 ± 0.0
Average Dry ±S.D. 0.4±0.2
Average Both ±S.D. 0.3±0.2
-78-
-------
5.3.4.6 Effect of Various Parameters cm Breath CO
Measurement
Air spiked with CO at known concentrations was
measured under several conditions. Measurements were taken with the
monitor attached directly to the CO source, to the'CO source with the
prefilter in-line using both dry and humid air, and to the bag with the
prefilter in-line using humid air spiked with ethanpl at a concentration
of 5 ug/L. Measurements from the CHB bags were maiple with and without
storage.
i
Table 5.3.4 shows measured CO concentrations with the monitor
attached directly to the gas mixing manifold with and without the
prefilter on-line. Table 5.3.5 shows measured CO concentrations with
the monitor attached to the gas mixing manifold w:ith the prefilter
in-line using dry and humid air. A single measurement was made for each
condition. The instrument was allowed to rezero using ambient air after
each measurement. Results show no significant differences in measured
CO under the conditions tested.
Tables 5.3.6 through 5.3.8 show comparisons of. spiked air sampled
directly from the gas mixing manifold to measurements for air taken from
I
the same source, but loaded into CHB sampling bags prior to measurement.
Experiments were performed at three CO concentrations (3, 7, and 15
ppm) . The prefilter was in-line for all measurements. Bags were first
|.
flushed and then filled with air to minimize carry-over effects.
Results show some loss from sampling bags. Losses appear to be greatest
for dry air with a storage period, but the data are not conclusive.
During storage, a mean loss of 0.8 ± 0.4 ppm was observed for dry air
samples over all tested concentrations. Samples using humid air showed
a loss of 0.6 ± 0.3 ppm CO from the bags.
Data in Table 5.3.9 show measured CO values for air spiked with
ethanol sampled under a variety of conditions. All measurements were
taken with the prefilter on-line. Ethanol does not appear to affect the
CO measurement. The general trend of some sample loss from the bag over
time appears to have also occurred during this experiment.
-79-
-------
Table 5.3.4 Effect of Filter on Measured CO
CO Air
Source Type Measured [CO]
manifold - dry 5.0 8.0
no filter
manifold - dry 5.0 8.0
filter
on-line
(ppm)
31.5
31.5
Table 5.3.5 Effect of Humid Air on Measured CO
CO Air
Source T7'Pe Measured [CO]
manifold - dry 2.8 7.0
filter
manifold - humid 3.0 7.0
filter
Table 5.3.6 Effect of Storage in Sampling Bags on CO
3 ppm CO
CO Air Storage
Source I/ Type Time (h) Measured
manifold dry 0 " 2.9 ±
manifold humid 0 3.0
CHB bag dry 0 2.4 ±
CHB bag humid 0 • 2.8
CHB bag dry 18 2.5 ±
CHB bag humid 18 2.7 ±
(ppm)
15.5
15.5
Measurement at
[CO] (ppm)
0.1 21
I/
0.2 4/
I/
0.3 2J
0.1 2J
_!/ Measured with filter on-line.
21 Triplicate determinations ± S.D.
3f Single determination.
j4/ Duplicate determination ± mean deviation.
't
-80-
-------
Table 5.3.7 Effect of Storage in Sampling Bags on CO Measurements at
7 ppm CO
CO
Source I/
manifold
manifold
CHB bag
CHB bag
CHB bag
CHB bag
Air
Type
dry
humid
dry
humid
dry
humid
Storage
Time (h)
0
0
0
0
23
23
Measured [CO] (ppm)
.7.0 _2_/
, 7 JO 21
6.8 2J
6,8 2J
6.3 t 0..3 3/
6.3 ± 0.3 37
I/ Measured with filter on-line.
2/ Single determination.
"3~/ Triplicate determinations ± S.D.
Table 5.3.8 Effect of Storage in Sampling Bags on CO Measurements at
15 ppm CO !
CO Air Storage ;
Source I/ Type Time (h) Measured [CO] (ppm)
manifold dry 0 14.8 2J
manifold humid 0 14.8 2j
CHB bag dry 20 13.1 ± 0.5 3j
CHB bag humid 20 13.9 ± 0.1 _3/
\J Measured with filter on-line.
2j Single determination.
3/ Triplicate determinations ± S.D.
-81-
-------
Table 5.3.9 The Effect of Ethanol on Measured CO
GO
.Source I/
manifold
CHB bag
manifold
CHB bag
manifold
CHB bag
CHB bag
manifold
CHB bag
Air
Type
humid
humid
humid
humid
humid
humid
humid
humid
humid
Storage
Time (h)
0
0
0
0
0
0
18
0
0
Measured [CO] (ppm)
15.5 2J
15.2 ± 0.1 I/
7.3 2J
6.8 ± 0.0 J7
3.3 21 •
3.1 ± 0.3 I/
2.7 ± 0.1 _!/
-0.3 2J
-0.3 ± 0.0 If
ll Triplicate determinations ± S.D.
2j Single determination.
-82-
-------
5.3.4.7 Interference Due to Plastic Mouthpiece
Interferences or losses of CO due to the use of a
disposable mouthpiece were determined. No noticeable CO interferences
were observed when using the mouthpieces. An average loss of 0.7 ± 0.2
ppm CO occurred when using humid air spiked with 6.3 ppm CO loaded into
and measured from a CHB sampling bag with the mouthpiece in place. This
compares to a loss of 0.5 ppm CO observed when using the bag without the
mouthpiece. All samples were loaded and analyzed with less than ten
minutes storage time.
5.3.4.8 Effect of Concentrated Organic Compounds on Monitor
and Prefilter Performance ' i
Experiments were performed to determine possible
interferences with CO measurements using compounds which might be found
in breath samples. In an initial experiment, a 10 imL volume of neat
(i.e., not in solution, undiluted) ethanol, acetone, methyl ethyl
ketone, or propionaldehyde was placed in a small beaker. Headspace of
the beaker was sampled using the GE monitor adjusted to a flow rate of
100 mL/min without the charcoal/Purafil® prefilter. i The monitor did not
respond to acetone and methyl ethyl ketone, but gave an immediate high
response to ethanol and propionaldehyde. For these flatter two compounds,
the small, integral Purafil® filter within the GE monitor had turned a
noticeable brown color. ; .
The experiment was repeated with ethanol, propionaldehyde, and
acetaldehyde using a charcoal/Purafil® prefilter. The breakthrough time
and sample volume for each compound were determined for the prefilter
using the GE monitor attached to a strip chart recorder. For ethanol, a
baseline deflection or monitor response became noticeable after approxi-
mately 60 minutes, which is equal to a 6 L breakthrough volume. The
Purafil® in the prefilter had started to turn brown1 before any detector
response was recorded. Since this was the case, a color change in the
Purafil® could be used as the criterion for replacing the filter during
•
field sampling. , !
Propionaldehyde appeared to instantaneously [breakthrough the
charcoal/Purafil® prefilter. However, the Purafil® did not discolor
during this experiment. Since previous experiments with propionaldehyde
-83-
-------
had shown a brown discoloration of the Purafil® and aldehydes are
unstable and tend to break down into other chemicals, the breakthrough
response was attributed to a contaminate in the propionaldehyde.
Because of these spurious results acetaldehyde was tested. The filter
appeared to retain acetaldehyde for approximately 20 minutes before the
Purafil® discolored and the monitor gave an off-scale response.
These experiments were performed under "worst case" conditions by
loading the prefilter with high concentrations of organic compounds
where breakthrough should result from saturating the filter rather than
from a chromatographic effect. Even under these conditions, the filter
was effective and should be adequate for field testing of breath samples.
5.3.4.9 Method Precision
In a final experiment, the precision of the method
was evaluated using eight nonsmoking subjects who gave four breath
samples each. All samples were collected in the CBH bags using dispos-
able mouthpieces and analyzed on a GE CO monitor, Model 3. The CO
levels analyzed ranged from 1.9 to 3.8 ppm with coefficients of varia-
tion from 0 to 16 percent. Except for one individual, all CVs were 5%
or lower. The final results are listed in Table 5.3.10.
The range of concentrations examined during the laboratory evalua-
tion of method precision were necessarily low due to use of samples from
non-smoking subjects. A measure of the method precision at higher
concentration levels can be inferred from the analysis of laboratory and
field control samples performed during the field sampling phase of the
project. The means of analyses of laboratory and field control samples
at 9.98 ppm were 9.46 ± 0.45 ppm (± one standard deviation) and 9.51 ±
0.50 ppm, respectively. The means for analyses of samples at 39.6 ppm
were 39.3 ± 0.35 ppm and 39.4 ± 0.58 ppm, respectively.
5.3*.4.10 Analysis Procedure Used During Field Sampling
Based on the result of the method development and
evaluation, a Standard Operating Procedure entitled "Collecting and
Sampling Alveolar Carbon Monoxide" was written and has been included
herein as Appendix H. This SOP was used as the analysis procedure
during field monitoring with one exception concerning the preparation of
standard atmospheres for monitor calibration. Analysis data were
-84-
-------
Table 5.3.10 Breath Measurements for Non-Smoking Subjects
Subj ect
1
2
3
4
5
6
7
8
Measured [CO] ppm ± S.D. (C.V.)
1.9 ± .1 ( 5)
2.7 ± .1 ( 14)
2.7 ± .1
1.9 ± .1
2.1 ± .1
2.4 ± .4
(4)
(15)
1.9 ± .1 ( !5)
(5)
3.8 ± 0 ( 0)
(16)
-85-
-------
recorded on the data sheet depicted in Figures 5.3.6 and 5.3.7. The
exception to the SOP is discussed below.
The calibration standards for the breath analysis consisted of CO
in synthetic air cylinder gases at CO levels of 0.0, 3.59, 9.98, and
39.6 ppm. Like the standards for the ambient analyses, these gases
contained no carbon dioxide or water vapor. They were analyzed for CO
and oxygen content by the manufacturer and verified for CO content
against NBS-tracable standards at RTI before being used in the field.
At the beginning of the field analyses, however, it was demonstrated
that the routine transition from "wet" breath samples (i.e., samples at
essentially 100 percent relative humidity) to "dry" zero or calibration
matrices (or vise-versa) induced a nonreproduceable zero-level response
from the CO-3 monitor. These phenomena were attributed to retention and
subsequent release of water vapor by the prescrubber column during
sample/standard transitions. This led to the imposition of a water
vapor gradient on the monitor which lasted for 5-10 minutes and produced
the nonreproduceable zero responses. It was decided that wet zero and
calibration matrices would be used, for the analyses. The humidification
of the zero air was accomplished by placing an impinger containing
deionized water between the zero air supply and the zero air manifold.
The humidified standards were prepared by filling sampling bags to which
1-2 drops of water had been added with the various calibration matrices.
These bags were set aside for approximately 30 minutes while the added
water evaporated into the calibration gas. The use of the wet zero and
calibration matrices eliminated the nonreproduceable zero response and
it is recommended that this modification be added to the SOP before it
is again utilized in field sampling.
5.4 Data File Creation and Descriptions
5.4.1 Descriptions of Raw Data Files
Data for the Washington, D.C. carbon monoxide exposure study
consisted of four basic types (exclusive of the sampling information,
described in Section 5.1). These four data files are briefly described
below.
File A; Personal Exposure Monitor (PEM) Data. CO exposure levels
from the PEM were obtained for four kinds of samples:
-86-
-------
CO EXPOSURE STUDY, WASHINGTON, DC
BREATH SAMPLE DATA SHEET
SAMPLE DATE ANALYSIS DATE ANALYST
P.I.D. # CO LEVEL (PPM)|
'
i i i "i.r~i
i.O.
1.0
1.0.
i i
i i i i i uD
n.o
iii i i i uO
if data continued on back: | I
Figure 5.3.6 Breath Sample Data Sheet, Side 1
•• i -
-87-
-------
BREATH SAMPLE DATA SHEET (Continued)
P.I.O. # CO LEVEL (PPM)
I I I I I I I I
I I I I I I I I
3.LZI
ii.n
.LZI
.n
Figure 5.3.7 Breath Sample Data Sheet, Side 2
-88-
-------
(1) routine samples — 774 data records — for CO measurements on
persons selected into the sample,
(2) duplicate samples — 60 data records covering 28 person-days
— for CO measurements on interviewers carrying multiple PEMs,
(3) colocated samples — 10 data records — for PEMs monitoring
•
CO levels in the vicinity of a fixed-site monitoring station,
(4) EPA audit samples — 9 data records.
This file thus consisted of 853 data records. j
In addition to sample-identifying information, each record
indicated the time and value of a series of PEM measurements. The
CO measurements were generated as automatic recordings on the hour
or as manual readings when activities of sample members changed.
The measurements represented the average (integrated) CO level
during the preceding time interval (i.e., since the beginning of
the hour or since the last manual entry). On the raw data file,
the CO data were recorded in raw units. However, the slope and
intercept, as well as a validity check code for; the straight-line
calibration curve, were available on each record so that CO levels
could be converted to parts per million (ppm). The contents of
this file are shown in Exhibit 5.4.1. For the routine samples, the
time period during which the PEM was in the presence of the sample
member was identifiable only through the diary data.
File B; Activity Diary Data. Diary data consisted of (a) identi-
fication data (person and monitor ID numbers, and starting and
ending times and dates); and (b) information on each activity. The
latter included:
- activity code
- location code
— address information
mode of travel (if in transit)
indicator for whether a garage was attached to building (if
indoors) j
- indicator for whether a gas stove was in use (if indoors)
j
- indicator for whether smoker(s) were present^
-89-
-------
Exhibit 5.4.1 Contents of File A (PEM Data)
Variable
BATCH
PID
MONID
SAMPDATE
VALCODE
APPINTRC
APPSLOPE
THR1-THR96
TMIN1-TMIN96
C1-C96
Description
Batch Number
Person or sample identification number
Monitor ID number
Date of sample (month/day/year)
Validity check code for calibration curve
Intercept of calibration curve
Slope of calibration curve
Hour of PEM reading (up to 96 readings)
Minute of PEM reading (up to 96 readings)
CO level from PEM (up to 96 readings)
-90-
-------
Exhibit 5.4.2 shows the data codes used for the above variables.
The raw data file of diary data contained 16,820 data records
(i.e., activity segments) for 917 persons.
File C; Breath CO-Level Data. The "breath" CO data file, in its
initial form, consisted of 1,390 data records,! which were distri-
buted as follows:
(1) 793 routine samples from sample members
(2) 110 duplicate routine samples (some of these represent the
only usable breath measurement for some individuals, even
though they were coded as duplicates)
(3) 12 children's breath samples (from children in households of
study participants) |
(4) 14 suspected smoker's breath samples
(5) 61 quality control samples (blank, field control, and labora-
tory air samples).
The information on each record consisted of the (30 level (in ppm),
as well as sample identification data and dates of sampling and
chemical analysis.
File D; Questionnaire Data. In addition to providing PEM and
breath measurements and the Activity Diary data, study participants
were asked to furnish information on their homes, typical commuting
activities, etc. through a Study Questionnaire (see Appendix C).
Data from this questionnaire were incorporated into a data file;
the file contained data for 916 persons. [ .
I
Exhibit 5.4.3 shows the variables on this file. Only a few of
the pertinent variables in these data records were utilized in the
analyses reported herein due to time and cost constraints (addi-
t
tional use of the questionnaire data are recommended for future
analyses). These included the following: i
Questions 20 and 29A — variables 13 and 69 — were utilized
to classify persons into three occupational exposure cate-
gories: |
(1) "doesn't work outside home"; '
(2) "works outside home — low exposure"; or
-91-
-------
Exhibit 5.4.2 Description of Codes for Variables Appearing in File B
(Activity Diary Data)
Activity Codes;
Act Description __________
1 transit, travel
2 work, business meeting
3 cooking
4 laundry
5 inside house - chores
6 outside house - chores
7 errands, shopping, etc.
8 personal activities
9 leisure activities
11 sleeping
12 school, study
13 eating, drinking
14 sports and exercise
15 church, political meetings, etc.
16 inside house - misc.
17 in parking garage or lot
18 outside, not otherwise specified
19 doctor or dentist office
87 start diary
88 end diary
89 any other activity
90 no activity entry
91 activity entry not legible
92 uncertain of applicable activity code
Location of Activity
Loc Description
0100 in transit
0200 indoors - residence
0300 indoors - office
0400 indoors - store
0500 indoors - restaurant
0661 indoors - garage
0662 indoors - auditorium, sports arena, etc.
0663 indoors - church
0664 indoors - shopping mall, theater at mall
0665 indoors - school, school gym
0666 indoors - hospitals
0667 indoors - laboratories
0668 indoors - not specified
0669 any other indoor location
0700 outdoors - within 10 yards of road or street
0881 outdoors - garage, parking lot
0882 outdoors - construction site
0883 outdoors - residential area
0884 outdoors - park, sports arena, playground
0885 outdoors - gas station
0888 outdoors - not specified
0889 any other outdoor location
0900 uncertain
9800 missing
-92-
-------
Exhibit 5.4.2 (cont'd)
Mode of Travel
Modetrav Description
0100 walking
0200 car
0300 bus
0400 truck
0500 train/subway
0661 jogging
0662 bicycle
0663 motorcycle
0664 van
9500 bad data
9600 multiple response
9800 missing
Smokers Present?
Smokers Description
• missing
1 yes
2 no
3 uncertain
Garage Attached to Building?
Garage Description
• missing
1 . yes
2 no
3 uncertain
Gas Stove in Use?
Gasstove Description
« missing
1 yes
.2 no
3 uncertain
-93-
-------
Exhibit 5.4.3 Contents of File D (Questionnaire Data)
# Variable Label
45
74
75
76
39
67
1
47
108
52
106
105
107
72
50
71
70
35
4
25
8
9
10
60
62
64
61
63
65
46
29
27
30
28
101
103
48
19
84
13
87
89
91
93
95
88
AIRCONDT
AREAAIRC
AREAFANS
AREASMOK
ATTICFAN
AUTOYEAR
BATCH
BUSGARGE
COMMENTS
COMMJT3X
DATEDA
DATEMO
DATEYR
DESCSPAC
ELECPLNT
ENCLSFAC
ENCLWORK
EXTRAINS
FINUMBER
FIREPLAC
FRMINUT1
FRMINUT2
FRMINUT3
FRMODES1
FRMODES2
FRMODES3
FRSMOKR1
FRSMOKR2
FRSMOKR3
GARAGEAT
GASDRYER
GASFRUNC
GASKERHT
GASRANGE
HODMATER
HOUSTYPE
HWEHDEP
INGARMIN
KINDBUSN
KINDWORK
LEISACT1
LEISACT2
LEISACT3
LEISACT4
LEISACT5
LEISPLA1
Q8 Air Cond used in living quarters
Q24 Enclosed area air-conditioned
Q25 Fans used in enclosed area
Q26 Smokers present in enclosed area
Q6A Attic fan used in living quarters
018 year of auto most.used in normal week
Q10A Living quarters within 3 blocks bus garage
Comments
Qll Commute to work, school* etc., 3X/week
Day
Month
Year
Q220th Describe other enclosed area
Q10D Living quarters near electric or steam
plant
Q22 Enclosed area at work most of time
Q21 Some part worktime in enclosed area
Q5C Extra insulation
FI Number
Q4A Fireplace used in living quarters
Q14B1 Minutes traveling for commute
Q14B2 Minutes traveling for commute
Q14B3 Minutes traveling for commute
Q14A1 Mode transportation for commute
Q14A2 Mode transportation for commute
Q14A3 Mode transportation for commute
Q14C1 Smokers present during commute
Q14C2 Smokers present during commute
Q14C3 Smokers present during commute
Q9 Garage attached or in structure
Q4F Gas clothes dryer in living quarters
Q4C Gas furnace used in living quarters
Q4G Gas or kero. space heater in living
quarters
Q4D Gas cookstove used in living quarters
FIUSA housing construction material
FIUSB type of housing structure
Q10B Living quarters near heavy vehicle depot
Q34 Minutes spent in indoor garage
Q29B Kind of business employed in
Q29A Occupation
Q32A1 Leisure activity
Q32A2 Leisure activity
Q32A3 Leisure activity
Q32A4 Leisure activity
Q32A5 Leisure activity
Q32B1 Place of leisure activity
-94-
-------
Exhibit 5.4.3 (continued) :
90 LEISPLA2 Q32B2 Place of leisure activity
92 LEISPLA3 Q32B3 Place of leisure activity
94 LEISPLA4 Q32B4 Place of leisure activity
96 LEISPLA5 Q32B5 Place of leisure activity
14 LEISTIM1 Q32C1 Time spent in leisure activity
15 LEISTIM2 Q32G2 Time spent in leisure, activity
16 LEISTIM3 Q32C3 Time spent in leisure activity
17 LEISTIM4 Q32C4 Time spent in leisure activity
18 LEISTIM5 Q32C5 Time spent in leisures activity
21 LIVQALLP Q3A Number of people in living quarters
23 LIVQCIGT Q3C Cigarette smokers in living quarters
20 LIVQKIND Ql Living quarters description
24 LIVQSMOK Q3D Number cigarette packs used in week
3 LIVQSQFT Q2 Square feet in living quarters
22 LIVQTOBA Q3B Tobacco smokers in living quarters
43 MAINHEAT Q7 Main type of heating system
44 MAINHTSP Q7SP Specify other type main heat
79 NRBUSGAR Q28A Work 3 blocks from bus garage
82 NRELPLNT Q28D Work 3 blocks from elec. op. steam plant
80 NPHVDEPO Q28B Work 3 blocks from heavy vehicle depot
81 NROPBURN Q28C Work 3 blocks from open burn site
83 NRSMPLNT Q28E Work 3 blocks from heavy smoke emitter
49 OPENBURN Q10C Living Quarters near site of open burning
41 OTHERFAN Q6C Other type fan used in living quarters
37 OTHERGAD Q5E Other energy-saving device
31 OTHERGAS Q4H Other gas appliance
42 OTHFANSP Q6CSP Specify other type fan
38 OTHGADSP Q5ESP Specify other energy-saving device
32 OTHGASSP Q4HSP Specify other gas appliance
102 OTHMATER FIUSAOTH Describe other housing material
104 OTHSTYPE FIUSBOTH Describe other housing structure
86 OUTDWKWK Q31 Hours work outdoors in week
69 OUTSIJOB Q20 Full- or part-time job outside home
66 PASSGHRS Q17 Passenger for how many hours weekly
2 PID
99 RESP_AGE Q35B Respondent's age
100 RESP_EDU Q35C Respondent's education level
98 RESP_SEX Q35A Respondent's sex
59 SAMEMODE Ql.3 Return home same travel mode
97 SHOPPARK Q33 Indoor parking on shopping trips
73 SIZESPAC Q23 Size of enclosed area £t work
51 SMOKPLNT Q10E Living quarters near heavy smoke emitter
36 SPDAMPER Q5D Special dampers stove/fireplace
34 STORMDOR Q5B Storm doors
33 STORMWIN Q5A Storm windows
11 TIMEARRI Q15 Time arrive at destination
12 TIMEDEPA Q16 Time depart for home
5 TOMINUT1 Q12B1 Minutes traveling for commute
6 TOMINUT2 Q12B2 Minutes traveling for commute
7 TOMINUT3 Q12B3 Minutes traveling for commute
-95-
-------
Exhibit 5.4.3 (continued)
53 TOMODES1 Q12A1 Mode transportation for commute
55 TOMODES2 Q12A2 Mode transportation for commute
57 TOMODES3 Q12A3 Mode transportation for commute
54 TOSMOKR1 Q12C1 Smokers present during commute
56 TOSMOKR2 Q12C2 Smokers present during commute
58 TOSMOKR3 Q12C3 Smokers present during commute
68 WEEKLHRS Q19 Hours spent in auto in average week
40 WINDOFAN Q6B Window fan used in living quarters
78 WKHEATSP Q27SP Specify other type heating system
26 WOODSTOV Q4B Woodstove used in living quarters
77 WORKHEAT Q27 Main heating system at work place
85 WORKWEEK Q30 Number hours in normal work week
-96-
-------
(3) "works outside home —high exposure". (The occupations
i
identified as having high potential exposure were:
crane, derrick, or hoist operator; automobile mechanic;
garage or gas station worker; machines operator; bus
driver; taxi driver/chauffeur; truck driver; construction
laborer; warehousmen; cook; airline host/hostess; firemen;
police/detective - see Appendix K.)
Question 4D —• variable 28 —- was used to classify persons
according to the type of stove used in their home: (1) vented
I
gas stove, (2) unvented gas stove, or (3) other (non-gas).
Questions 11 and 17 — variables 52 and 66 — were utilized to
classify individuals according to commuting status and amount
of travel: (1) non-commuter, or (2) commuter with total travel
of 0-5 hours per week, 6-10 hours per week, 11-15 hours per
week, or more than 15 hours per week. !
5.4.2 Creation of Analysis Files I
Three analysis files were created: (1) the basic analy-
sis file (BAF); (2) the activity analysis file (AAF); and (3) the
duplicate measurement file (DMF)» :
These files and their construction are described, respectively, in
subsections 5.4.2.1, 5.4.2.2, and 5.4.2.3.
I
5.4.2.1 Creation of the Basic Analysis File (BAF).
A first step in creating the BAF involved the exami-
'
nation of the sampling dates and times as reported in the Activity Diary
(i.e., File B). Where necessary, corrections were made to these start
and stop times (e.g., by comparison to the PEM data) and dates (e.g., by
correcting the year from 1982 to 1983 in some cases).
I
The next step in creating the BAF was to perform edit checks on the
PEM data (File A). The 774 routine samples were first examined to
determine those individuals with unusable or insufficient data. This
involved eliminating 41 samples for which the calibration data were
questionable (validity codes = 4); eliminating 15 samples with less than
an 18-hour monitoring period (it was felt, since 8-hour CO maximums were
to be computed, that at least 18 hours of data should be available for
i
each sample respondent); eliminating 5 samples for which no diary data
-97-
-------
(and hence no start or stop times) were available; and eliminating one
sample due to misassignment of the PEM (i.e., the wrong person had been
monitored). This left 712 sample members whose PEM information, after
further editing, was considered acceptable.
The editing of the PEM data involved identifying and resolving the
following types of potential problems:
(1) out-of-range hour and/or minute values ,«
(2) times not in the proper sequence
(3) missing times and/or missing CO values *
i,j|
(4) large CO values (e.g., over 20 ppm).
The PEM information was listed and examined manually for any individual
whose data exhibited any of the above. When deemed necessary, the
hard-copy field data and/or diary data were consulted in order to make
the appropriate resolutions in the time and/or CO values. '
After this editing, the PEM CO data values for each of the 712
individuals were time weighted to produce hourly CO values. The CO
values were converted to ppm units by using the slope and intercept
values for the calibration curve. Any CO level less than .05 ppm was
set equal to .05 ppm. The hourly values were constructed only for those
hours for which the PEM data indicated coverage of the entire hour.
This hourly data file contained from 18 to 26 hourly CO values, depend-
ing upon the particular start and stop times of the particular sample
(or in some cases, the time at which a monitor failure occurred). The
number of samples, by hour of day, is shown in Table 5.4.1. As these
results indicate, the hourly data do not fully cover a 24-hour time
period for all 712 sample members. Coverage was especially lacking
during the 6-9 p.m. time period when sampling was begun or terminated.
The distributions of the number of hourly CO values per sample and of
the day of sampling are shown, respectively, in Tables 5.4.2 and 5.4.3.
After editing of the.routine-sample PEM data and construction of
the hourly CO values, several additional variables were constructed and
ji.
augmented onto each record of the file. These were the following: :|
-* ;!
(1) the mean hourly CO concentration (ppm);
(2) the maximum hourly CO concentration (ppm);
(3) the maximum 8-hour CO concentration (ppm); and ;
-98-
-------
Table 5.4.1
Number of Routine Samples With Valid Hourly CO Values,
By Hour of Day
Hour (ending)
01
02
03
04
05
06
07
08
09
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
Day 1*
„
— .
__
—
—
—
—
—
—
—
—
—
—
—
—
- —
—
5
28
158
469
664
708
712
Day 2 Day si or Day 2
711 711
712 ! '712
712 712
712 712
712 , 712
712 712
712 712
712 712
711 711
712 712
712 712
712 712
712 712
712 712
708 708
707 707
702 : 702
666 670
548 570
284 426
85 540
9 668
708
712
Day 1 is the day that .sampling began (in the evening); day 2
is the following day. ... '
-99-
-------
Table 5.4.2 Distribution of the Number of Hourly CO Values Per Sample
Number of
Hourly CO
Values Per
Sample
18
19
20
21
22
23
24
25
26
Total
No. of
Samples
4
3
6
27
96
374
170
23
9
712
Table 5.4.3 Distribution of Sampling Dates, by Month and Day of Week
Sampling Days Nov. 1982 Dec. 1982 Jan. 1983 Feb. 1983 Total
Sun - Mon
Mon - Tues
Tues - Wed
Wed - Thurs
Thurs - Fri
Fri - Sat
Sat - Sun
26
36
13
12
13
6
13
19
37
30
26
29
27
14
40
47
35
29
41
39
24
15
20
37
22
24
23
15
100
140
115
89
107
95
66
119
182
255
156
712
-100-
-------
(4) the beginning hour (index) of that 8-hour period for which the
maximum 8-hour CO concentration occurred, i
The mean (maximum) hourly CO concentration for a particular sample
individual was based upon averaging (maximizing) over all of the hourly
CO values that were available. The third variable was determined in the
following manner. Let f and % denote the hour index of the first and
last hourly CO values available for a particular sample. The following
8-hour averages were computed:
. a+7 :
Y - £ 3C.I./n a-f.(f-H), .'.., (4-7)
3L r"* X J- 3. ;
i=a
i
i._
where I. =0 if the hourly CO value for hour index i is missing
= 1 otherwise; [
X = the hourly CO value in ppm for hour index i:, and
a+7 |
n = V" I. = number of non-missing hourly CO values in the
a * 8-hour interval beginning at hour "a".
i=a '
The maximum 8-hour CO concentration was then determined as the maximum
Y value for which n fi 6 (i.e., at least six hourly values were requir-
a a - . | •
ed during a consecutive eight-hour period in order for a Y value to be
! • **
considered as an "8-hour average"). The fourth variable indicated the
beginning hour index of the maximum 8-hour CO concentration period —
i.e., the value of "a".
The above four variables were derived exclusively from the hourly
data. In addition, variables from other sources were incorporated into
the BAF. These variables included the following:
(1) the set of variables described for File D:(questionnaire data)
in Section 5.4.1
(2) the breath CO level (averaged over the duplicate readings in
those cases where an individual provided two breath samples)
I
(3) a variable indicating those dates for which fixed-site moni-
tors in the DC area showed high CO levels ('based on informa-
tion furnished by EPA), and
-101-
-------
(4) the pertinent sample design variables and adjusted sampling
weights.
The sample design variables included the person-level weighting class
indicator, and the "analysis stratum" and "analysis PSU" variables
described in Section 5.1. Three sampling weights were constructed:
(a) the adjusted sampling weight (see Section 5.1), which is
appropriate for analyses in which all 712 individuals provide
data on the relevant analysis variable(s);
(b) the diary-activity adjusted sampling weight, which is appro-
priate for analysis of diary activity data (available for only
705 of the 712 sample members); and
(c) the breath-level adjusted sampling weight, which is appro-
priate for analysis of the breath CO concentration data
(available for only 659 of the 712 sample members).
The total of the adjusted sampling weight for the person-level weighting
class was also included on the file so that additional weight adjust-
ments could be readily made. Such adjustments are needed, for instance,
when estimating the mean CO•level for some specific hours of the day
since not all 712 individuals furnished such data (see Table 5.4.1).
Exhibit 5.4.4 shows the contents of the Basic Analysis File.
5.4.2.2 Creation of the Activity Analysis File (AAF)
The creation of the AAF first required that an
extensive amount of editing be performed on File B (see Section 5.4.1).
This involved the following:
(1) deletion of all activity segments associated with nonrespond-
ents — defined as those sample members not among the 712 who
furnished sufficiently complete and valid PEM data (205
individuals and 3345 activity segments — recall that monitor
failure caused a substantial amount of CO data to be lost from
sample members);
(2) deletion of all activity segments for seven additional indi-
viduals — namely those having obviously incomplete diaries (7
individuals with a total of 28 activity segments);
(3) deletion of specific activity segments — duplicate segments,
segments with missing/invalid times, activities, etc. (50
segments);
-102-
-------
Exhibit 5.4.4 Contents of Basic Analysis File
f Variable Label , -,
43 APSU Analysis PSU
44 ASTRATUM Analysis stratum
7 AVC18 CO concentration for day 1, hour 17-18
8 AVC19 CO concentration for day ls hour 18-19
9 AVC20 CO concentration for day 1, hour 19-20
10 AVC21 CO concentration for day I, hour 20-21
11 AVC22 CO concentration for day 1, hour 21-22
12 AVC23 CO concentration for day 1, hour 22-23
13 AVC24 CO concentration for day 1, hour 23-24
14 AVC25 CO concentration for day 2, hour 00-01
15 AVC26 CO concentration for day 2, hour 01-02
16 AVG27 CO concentration for day 2, hour 02-03
17 AVC28 CO concentration for day 2, hour 03-04
18 AVC29 CO concentration for day 2, hour 04-05
19 AVC30 CO concentration for day 2, hour 05-06
20 AVC31 CO concentration for day 2, hour 06-07
21 AVC32 CO concentration for day \2S hour 07-08
22 AVC33 CO concentration for day \2, hour 08-09
23 AVC34 CO concentration for day 2, hour 09-10
24 AVC35 CO concentration for day 2, hour 10-11
25 AVC36 CO concentration for day 2, hour 11-12
26 AVC37 CO concentration for day |2S hour 12-13
27 AVC38 CO concentration for day 2, hour 13-14
28 AVC39 CO concentration for day 2, hour 14-15
29 AVC40 CO concentration for day 2, hour 15-16
30 AVC41 CO concentration for day 2, hour 16-17
31 AVC42 CO concentration for day|2, hour 17-18
32 AVC43 CO concentration for day 2, hour 18-19
33 AVC44 CO concentration for day;2, hour 19-20
34 AVC45 CO concentration for day 2, hour 20-21
35 AVC46 CO concentration for day 2, hour 21-22
45 BRCO Breath CO concentration
48 BWEIGHT Breath analysis weight !
55 CMUTXHRS Variable derived from Qll and Q17
52 COMMUT3X Qll Commute work, school* etc., 3X/week
56 DAYOFWK Day of week '
47 DWEIGHT Diary analysis weight
46 FWEIGHT Field data analysis weight
49 FWTT Total of FWEIGHT within PWTCLASS
51 GASRANGE Q4D gas cookstove used in living quarters
6 HICODAY High CO day indicator
50 KINDWORK Q29A Occupation |
37 MAXHRC Maximum hourly CO concentration
39 MAX8HC Maximum 8-hour CO concentration
38 MEANHRC Mean hourly CO concentration
36 NHRC Number of hourly CO valu'es
53 PASSGHRS Q17 Passenger for how ma.ny hours weekly
-103-
-------
Exhibit 5.4.4 (continued)
1 PID Check Digit: ID
42 PWTCLASS Person-level weight class
2 SAMPDATE Date of sample (day 1, mmddyy)
57 TIMEWEEK Weekday, weekend indicator
4 TSTART Start time for sample (from diary)
5 TSTOP Stop time for sample (from diary)
54 TYPEXPOS Type of exposure
40 T8HSTRAT First hour of 8-hour maximum CO
3 VALCODE Validity code for claibration data
41 WTCLASS2 Final weighting class
-104-
-------
(4) corrections of sampling dates (15 individuals);
(5) corrections of start or stop times of the sampling period (4
individuals); [
(6) revisions in activity codes (282 segments);
(7) revisions in location codes (77 segments;); and
(8) revisions in activity start times originally appearing as
missing, partially missing, out-of-range or out-of-sequence
(90 segments). !
With the exception of item (1), all of the above items required at least
some manual examination of the data, including iri some cases review of
the actual hard-copy diary data and/or of the PEM time values. This
process took considerable staff time and lasted for several weeks.
The need for the manual examination of the data was largely due to
potential inconsistencies in the location and activity codes. Some of
these inconsistencies were resolved on a case-by-case basis, whereas
some were resolved by establishing (and subsequently programming) a set
of "consistency rules". In both cases, the location codes were general-
ly regarded as being more accurate than the activity codes, since the
I
activity codes were developed from a totally open-ended question (i.e.,
apart from instructions given to the study participants, they were
permitted to provide activity descriptions in th«ir own words and from
their own perspective — see Appendix C, Figures; 4 and 5). A major
I
exception to this was when both the activity ccide and the address
information indicated that the activity was "in transit". The programmed
"consistency rules", which in some cases made use of the address informa-
tion provided in the diary, are shown below:
-105-
-------
Original Codes*
Revised Codes**
Activity
5
5
all
all
90-92
90-92
7
1
4,16
5
7
15
16
Location
700
800-889
694-698
100
200-669
700-889
missing
all
883
300
200
700
881
Address
Indicators***
all
all
all
#(1,0,0)
all
all
(1,0,0)
(0,1,1)
all
all
all
all
all
Activity
6
6
U
1
16
18
U
U
U
U
U
18
U
Location
U
U
668
U
U
U
400
100
668
668
668-
U
661
* See Exhibit 5.4.2 for definitions of activity and location
codes.
** U - unchanged (i.e., same as original code).
*** As illustrated in Figures 4 and 5 of Appendix C, the first
address field should have been completed only for non-transit
activities. Both the second and third address fields should
have been completed if the activity was "in transit". The three
address indicators are defined as (X, ,X_,X») where X=0 if
**
the i
address is blank, and as X. = l, otherwise.
A second major reason for needing a manual examination of a large
portion of the diary information was the recognition of the fact that
participants' omissions of diary entries (i.e., changes in activities)
could potentially lead to serious biases in the study results. That
such situations actually occurred in the raw diary data are clearly
illustrated by the fact that only 463 of the 705 diary-respondents
entered "eating, drinking" as an activity during their sampling period
(at least 18 hours) . Based upon the manual examination of a large
number of individuals' diary data, it was clear that many of these types
of data anomalies occurred due to the open-ended nature of the activity
descriptions (i.e., respondents were simply allowed to describe their
activities) . Some respondents provided detailed activity descriptions
whereas others furnished vague or general descriptions that may have
encompassed several "activities" (e.g., for some respondents, the
"eating, drinking" activity may have been subsumed under the activity
"inside house - miscellaneous") .
-106-
-------
At the data editing/processing stage, little could be done to
circumvent or ameliorate the problem indicated in the paragraph above.
However, those individuals exhibiting extremely "long" activities were
scrutinized in more detail. ("Long" depends, of course, upon the
particular activity.) This detailed manual examination of the diary
data led to (a) the decision to delete all data segments for those seven
individuals who had been extremely vague in providing activity descrip-
tions or who had been extremely negligent in providing a complete
activity pattern (i.e., those with major omissions in their diary
information); and (b) the decision that all activity segments that were
likely to have included a sleep period should be r&coded to reflect this
"fact". (It should be noted that some of the descriptions of activities
in the diary actually included such multiple activities — e.g., "studied
and slept.") \
The receding of those activities containing a suspected sleep
period was made possible by the fact that sleeping is generally a
long-duration activity that occurs at roughly the same time for most
individuals. Such duration and regularity does not occur for other
activities (e.g., eating) so that additional recedings of this type were
not feasible. ;
At the beginning of this subsection it was indicated that 282
activity recedes had been made; of these, 167 arose from the manual
recedes and activity/location consistency checks previously described,
and 115 were the result of recodings of activities! to include suspected
sleep periods. It should be emphasized that the latter receding was
carried out in such a manner that identification qf the original activ-
ity code was still possible. In particular, if AJrepresents the origi-
nal activity codes (see Exhibit 5.4..2), then the receded value, A*, was
determined as
A* = A + 20, if 1 S A S 19
A* = 77 if A = 87.
I
This permitted these segments to be utilized in various ways during the
analysis phase.
-107-
-------
At this point, the AAF contained 13,398 activity segments (for 705
sample members). The next step in creating the final AAF involved a
machine comparison of the activity times entered in the diaries with the
times entered on the PEMs. The following results were obtained:
(a) Of the 13,398 segments, exact matches between the diary and
PEM times were found for 11,356 segments.
(b) Of the remaining 2042 segments, a near match in the two times
(within 2 minutes) was found for 1022 segments.
(c) Of the remaining 1020 segments, 214 cases were found for which
the same number of PEM and diary entries occurred within an
hour, even though no PEM time could be found that matched the
diary time (within 2 minutes).
(d) Of the remaining 806 non-matched cases, 68 occurred due to
lack of PEM data during the period of interest.
Based upon the above comparisons between the PEM and diary times,
two rules were adopted and implemented for updating the diary activity
times. These rules were:
1. If the diary and PEM data indicated the same number of activ-
ities within a given hour, then the times from the diary were
replaced by the times from the CO monitor.
2. If rule 1 did not apply, but a diary time and a PEM time
matched within 2 minutes, then the diary time was replaced
with the time from the monitor.
The results of the above-described updating of the diary times are
summarized in the table below, which shows the number of activity
segments falling into various categories:
-108-
-------
Character-
istic of
Activity
Start
Time
Titne
Match
Exact -
No Need
to Update
Total
No. of
Segments
11,356
Degen-
erate
Time
Interval
650
Segme
No
CO Data
Available
0
nt Chara.cte
Partial
CO Data
Available
32 :
rxstic
Full
CO Data
Available
10,674
Partial
or Full
CO Data
Available
10,706
Time 1,022
Match w/in
2 min. -
Time Updated
(Rule 2)
Time Not 214
Matched
w/in 2 min. -
Time Updated
(Rule 1)
Time Not 806
Matched
w/in 2 min. -
Time Not
Updated
50
970
972
210
211
68
12
717
729
13,398
712*
68
47
12,571
12,618
* Time intervals of 0 minutes duration; 607 of the 712 cases
represent "start-diary" or "end-diary" activities (i.e.,
activity codes 87 or 88). '
The last column in the above table indicates those activity segments for
which a corresponding CO concentration level could be constructed from
the PEM data.
The final step in creating the AAF was to determine these CO levels
and to augment them onto the file. These were computed by time-weight-
ing the PEM CO measurements over the time period associated with the
given activity segment (or, in the 47 cases having only partial CO data,
over that portion of the activity time interval for which the CO data
were present).
The final AAF, unlike the BAF described in the previous subsection,
cannot be used directly for most statistical analyses — i.e., additional
processing prior to the analysis is required (in order to augment
-109-
-------
sampling weights, revise codes, aggregate over an individual's segments,
etc.). Since the additional data processing on the AAF that must "be
carried out prior to statistical analysis depends upon the particular
analysis, this processing is considered a part of the analysis and is
therefore not discussed further in this subsection.
5.4.2.3 Creation of the Duplicate Measurement File (DMF)
File A, as described in Section 5.4.1, provided the
source data for the DMF. The 60 data records in this file corresponding
to the duplicate samples (from interviewers with multiple PEMs) were
first extracted. A printout of these data was produced and manually
screened for timing errors. Several of the readings were discarded and
several time corrections were made. The CO values were then time-
weighted to produce hourly CO values. A file containing these hourly
values — the DMF — was then created; a record was generated for each
hour and PEM. The file contained information on the following variables:
PID sample ID number (identifies person and PEM)
HOUR hour index
CONG hourly CO concentration
PERSON code identifying interviewer/sample date
REP =*! for first PEM; =2 for second PEM; etc.
The DMF contained a total of 1539 hourly CO values which covered 28
interview-"days" and a total of 724 unique interviewer-hours. At least
two hourly "readings" were available for 689 of the 724 hours.
-110-
-------
6. RESULTS AND DISCUSSION
6.1 Survey Design Results
6.1.1 Household Screener Statistical Analysis
Gas household appliances and certain sources of home heating
are considered producers of carbon monoxide. The1 following related
questions of interest were asked of a knowledgable 'household member, and
analyzed for each household:
Is there
a. a fire place which is used?
b. a wood stove?
c. a gas furnace? \
I
d. a gas cooking stove?
e. a gas hot water heater? ;
f. a gas clothes dryer? '
g. a gas or kerosene space heater?j
h. any other gas appliances?
i. an attached garage, or a shared, multi-family
garage? ! . .
Estimates from these items are presented in Tables 6.1.1 through
6.1.8. These results are based on successful interviews from 4394
households in the Washington, D.C. metropolitan area, and 2128 house-
I
holds from the Denver metropolitan area. The population estimate for
the number of households in the two sites for the time the survey was
conducted is 953,714 for Washington and 345,163 for Denver. Due to the
fact that no item-level nonresponse adjustments or 'item imputations were
I
made, these estimates are underestimates of the true total (i.e., in the
tables there is a "Not Known" category). However, the difference
between the estimates and the true totals are believed to be small.
For each data item, a separate category (namely, "not known") was
created to represent those households for which the respondent or
I
interviewer was uncertain whether a particular appliance, etc. existed.
-Ill-
-------
As an example, the study determined that there are an estimated 467
households in Washington, B.C. (Table 6.1.1) for which the screening
questionnaire did not ascertain whether a fireplace was being used.
In terms of home heating, about 33 percent of the Washington area
households have fireplaces in use and roughly 67 percent do not (Table
6.1.1). Fewer than 30 percent of Denver households have fireplaces
(Table 6.1.1).
As expected, there are fewer homes utilizing wood stoves - an esti-
mated 4 percent for the Washington area and 6 percent for Denver (Table
6.1.2). Among the four means of home heating, the gas furnace is
clearly the most common heating source. This is indicated in Table
6.1.3: 56 percent or 532,347 households use gas furnaces in the Wash-
ington area and 71 percent or 245,902 households use gas furnaces in the
Denver area. The percent of households using gas or kerosene space
heaters is similar to those using wood stoves, about 3 percent for
Washington and 5 percent for Denver (Table 6.1.4).
Statistical results for the usage of gas appliances are presented
in Tables 6.1.5 through 6.1.8. An estimated 64 percent or 609,029
households in Washington, DC, and 25 percent or 85,542 households in
Denver use gas cooking stoves (Table 6.1.5). Gas hot water heaters
serve 542,855 or 57 percent and 269,810 or 78 percent of households from
the respective sites, Washington and Denver (Table 6.1.6). An estimated
191,803 households or 20 percent and 57,402 households or 17 percent use
gas clothes dryers in their homes (Table 6.1.7). Other gas appliances
are seldom used. This is evident in Table 6.1.8: approximately 2
percent and 0.2 percent of Washington and Denver metropolitan area
households use other gas appliances, respectively.
According to Table 6.1.9, a combined estimate of 207,719 or 22
percent of Washington area households have an attached garage or share a
multi-family garage. This compares to a combined estimate of 120,460 or
35 percent for Denver area households.
6.1.2 Personal Item Statistical Analysis
As with most studies which involve a sample of households,
there is some interest in specific attributes of the household members.
In the CO Study a respondent was requested to answer several questions
-112-
-------
Table 6.1.1 Estimated Number of Households Using 21 Fireplace
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
Washington, DC
Metropolitan Area
Total
316,925
29,538
636,322
59,462
467
193
953,714
4,394
Proportion
Denver
Metropolitan Area
Total
0.3323 i03,211
0.0301 16,030
0.6672 241,629
0.0302 • 18, 608
0.0004
0.0002
227
141
Proportion
0.2991
0.0362
0.7002
0.0361
0.0006
0.0004
,345,068
2,128
Table 6.1.2 Estimated Number of Households Using a Wood Stove
Washington
Metropolitan
, DC Denver
Area Metropolitan
Area
Total Proportion Total Proportion
Yes
Standard Error
No .
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
37,721
6,858
915,382
64,623
611
378
953,714
4,394
0.0395 20,314
0.0075 2,841
0.9598 324,453
0.0075 24,594
0.0006 396
0.0003 184
325,163
2,128
0.0588
0.0089
0.9400
0.0089
0.0011
0.0005
-113-
-------
Table 6.1.3 Estimated Number of Households Using a Gas Furnace
Washington
Metropolitan
, DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
532,347
40,943
391,130
52,169
30,237
13,517
953,714
4,394
0.5581
0.0355
0.4101
0.0365
0.0317
0.0146
Denver
Metropolitan
Area
Total Proportion
245,902
22,231
94,964
14,608
4,297
1,246
345,163
2,128
0.7124
0.0379
0.2751
0.0376
0.0124
0.0036
Table 6.1.4 Estimated Number of Households Using a Gas or Kerosene
Space Heater
Washington
Metropolitan
, DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
30,530
5,621
921,536
61,817
1,649
444
953,714
4,394
0.0320
0.0050
0.9662
0.0050
0.0017
0.0004
Denver
Metropolitan
Area
Total Proportion1
18,352
4,480
326,152
22,301
659
265
345,163
2,128
0.0531
0.0111
0.9449
0.0111
0.0019
0.0007
-114-
-------
Table 6.1.5 Estimated Number of Households Using a Gas Cooking Stove
Washington,
Metropolitan
DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
609,029
49,353
344,329
41,033
357
213
953,714
4,394
0.6385
0.0329
0.3610
0.0329
0.0003
0.0002
Denver
Metropolitan
Area
Total Proportion
85,542
7,308
259,365
24,195
256
160
345,068
2,128
0.2478
0.0245
0.7514
0.0245
0.0007
0.0004
Table 6.1.6 Estimated Number of Households Using a Gas Hot Water
Heater
Washington
Metropolitan
, DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
542,855
37,564
361,036
50,113
49,823
13,628
953,714
4,394
0.5692
0.0308
0.3785
0.0346
0.0522
0.0152
Denver
Metropolitan
Area
Total Proportion
269,810
20,930
67,425
13,256
7,928
1,175
345,163
2,128
0.7816
0.0332
0.1953
0.0331
0.0229
0.0037
-115-
-------
Table 6.1.7 Estimated Number of Households Using a Gas Clothes Dryer
Washington
Metropolitan
, DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
191,803
17,316
737,472
64,529
24,284
13,259
953,560
4,393
0.2011
0.0195
0.7733
0.0217
0.0254
0.0142
Denver
Metropolitan
Total Prc
57,402
8,018
285,830
20,381
1,931
724
345,163
2,128
Area
^portion
0.1663
0.0179
0.8281
0.0178
0.0055
0.0021
Table 6.1.8 Estimated Number of Households Using Other Gas Appliances
Washington
Metropolitan
, DC
Area
Total Proportion
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
16,036
3,362
934,713
64,821
2,966
704
953,714 '
4,394
0.0168
0.0037
0.9800
0.0038
0.0031
0.0007
Denver
Metropolitan
Area
Total Proportion
782
293
344,019
24,659
362
172
345,163
2,128
0.0022
0.0008
0.9966
0.0009
0.0010
0.0005
-116-
-------
Table 6.1.9
Estimated Number of Households Having an Attached Garage
or Sharing a Multi-Family Garage
Attached Garage
Standard Error
Multi-Family Garage
Standard Error
Washington, DC
Metropolitan Area
Total
165,919
22,525
41,800
13,845
Proportion _;
0.1740
0.0191
0.0438
0.0153
Denver
Metropolitan Area
Total Proportion
108,934
11,188
11,526
3,919
0.3156
0.0284
0.0333
0.0099
Neither
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
744,976
58,749
862
426
953,557
4,393
0.7812
0.0250
0.0009
0.0004
222,199
18,870
2,504
2,469
345,163
2,128
0.6437
0.0298
0.0072
0.0070
-111-
-------
on behalf of each individual living in the household. Below is the
information asked for each individual:
a. sex
b. age
c. individual's relationship to head of household
d. whether individual presently smokes or uses tobacco in any
form "i,
e. whether individual is employed either full or part time '
f. whether individual travels to and.from work, school, or any
other place at least 3 times a week '
g. amount of time spent traveling one way when going any place at
least 3 times per week.
Statistical results to follow are based on varying sample sizes due
to missing data for individual questionnaire items. The number of
responses was highest for the sex item: 11,545 for Washington, DC,and
5,142 for Denver. No item-level nonresponse adjustments or imputations •„
were made for two reasons. First, the item response rate was high for
most items. Second, analysis of the screening data was considered to be
less important than analysis of monitoring data. Instead of weight
adjustments or imputations, the category "not known" is presented to
.
represent those individuals for whom the respondent or interviewer could
not determine a correct entry for any individual's attributes. Included ;
in this category is a population estimate for the data item of interest.
The statistical results of all personal item data are presented in '
Tables 6.1.10 through 6.1.16.
According to Table 6.1.10, an estimated 1.29 million people or 48 ;T
percent of the Washington metropolitan population are male. This
compares to the 1980 Census which reports 1.33 million males or 48 '
percent in the same area (Table 6.1.11). Also, 1.38 million females or
52 percent are estimated from the CO Study results versus 1.44 million w ;
females or 52 percent from the 1980 Census. Neither of these statistics ;
are expected to be exactly equal, primarily because of a population * :
change from 1980 to 1983 and sampling variability. They are, however,
approximately the same. When sex is disregarded, the CO Study results
estimate the 1983 Washington metropolitan population to be 2.67 million
-118-
-------
Table 6.1.10 Estimated Sex Distribution
Washington, DC
Metropolitan Area
Male
Standard Error
Female
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
Total
1,286,056
133,431
1,379,812
116,697
1,329
483
2,667,197
11,545
Proportion
0.4821
0.0094
0.5173
0.0094
0.0004
0.0001
Denver
Metropolitan Area
Total Proportion
! 415,730
' 30,762
. .
460,836
39,728
i 556
402
! 877,122
; 5,142
0.4739
0.0091
0.5253
0.0091
0.0006
0.0004
Table 6.1.11 The Sex Distribution According to the 1980 Census
Male
Female
Population Total
Washington, DC
Metropolitan Area
Total
1,327,797
1,435,308
2,763,105
Proportion
0.4805
0.5195
Denver
Metropolitan Area
Total
380,479
401,305
781,784
Proportion
0.4867
0.5133
-119-
-------
Table 6.1.12 Estimated Age Distribution - Categorized According to the
1980 Census Definitions
Under 18 Years
Standard Error
Between 18 to 64 Years
Standard Error
65 Years and Older
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
Washington, DC
Metropolitan Area
Total
699,556
108,889
1,720,330
139,452
153,628
10,816
39,405
7,320
2,612,919
11,188
Proportion
0.2677
0.0197
0.6583
0.0150
0.0587
0.0055
0.0150
0.0024
Denver
Metropolitan Area
Total
227,112
27,009
564,554
46,137
67,091
5,959
5,939
822
864,695
5,015
Proportion
0.2626
0.0155
0.6528
0.0142
0.0775
0.0092
0.0068
0.0010
-120-
-------
Washington
Metropolitan
, DC
Area
Total Proportion
Under 18 Years
Between 18 to 64 Years
65 Years and Older
Population Total
721,170^
1,826~^12-/
215, 523-^
2,763,105
0.261
0.661
0.077
Denver
Metropolitan Area
Total Proportion
197, 17 l-( 0.2522-/
506 ,955^ 0.6484-'
77,658^ 0.0994-/
781,784-/
\J Estimated from proportions published in the 1980 General Population
Characteristics for District of Columbia.
2j Estimated from proportions published in the 1980 General Population
Characteristics for Colorado. i
-121-
-------
Table 6.1.14 Estimated Distribution of Relationship to Head of
Household
Washington
Metropolitan
, DC
Area
Total Proportion
Head of Household
Standard Error
Spouse
Standard Error
Child of Head
Standard Error
Other Relation
Standard Error
No Relation
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
949,796
63,594
554,922
49,274
891,611
151,940
126,729
20,789
147,567
26,472
5,069
2,533
2,675,694
11,543
0.3549
0.0126
0.2073
0.0180
0.3332
0.0293
0.0473
0.0088
0.0551
0.0078
0.0018
0.0009
Denver
Metropolitan
Area
Total Proportion
346,637
24,712
191,165
14,866
229,522
16,388
30,882
3,969
41,584
8,666
36,991
21,100
876,781
5,138
0.3953
0.0099
0.2180
0.0105
0.2617
0.0144
0.0352
0.0053
0.0474
0.0087
0.0421
0.0214
-122-
-------
Table 6.1.15 Estimated Distribution of Persons 13 Tears and Older Who
Smoke or Use Tobacco in Any Form
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
Washington, DC
Metropolitan Area
Total
639,739
69,002
1,287,619
85,016
633
324
1,927,991
8,791
0.3318
0.0180
0.6678
0.0180
0.0003
0.0001
Denver
IMetropolitan Area
Proportion
Total
244,884
'29,266
i
400,079
! 23,164
254
i 124
lU5,217
I
i 3,953
Proportion
0.3795
0.0234
0.6200
0.0234
0.0003
0.0001
-123-
-------
versus 2.76 million for the 1980 Census. Denver metropolitan statistics
by sex can be seen in Tables 6.1.10 and 6.1.11. The proportions are
quite comparable; CO study estimates are 47 percent males and 53 percent
females compared to 1980 Census statistics of 49 percent males and 51
percent females.
Ages have been grouped for the sake of complying with Census group
definitions. These age groups are:
(1) under 18 year of age
(2) between 18 and 64 years of age, and
(3) age 65 and older.
Estimated age group totals and proportions for both sites, compare quite
well with Census statistics. Also, the age distributions for the two
sites are nearly identical. CO Study estimates for the age groups are
presented in Table 6.1.12 and corresponding 1980 Census results are in
Table 6.1.13. As can be seen, the percent of individuals under 18
years, between 18 and 64 years, are respectively, about 26 percent and
65 percent (ignoring site and reference source). The comparison for
persons 65 years and older is not quite as good. Census data are not
available for other items of interest.
The CO Study results for relationships of the respondent to the
head of the household screened are categorized in Table 6.1.14. As
anticipated, more head of households were screened than any other
category, 35 percent and 40 percent for Washington and Denver metro-
politan areas.
For the remaining person-level data items, statistical analysis was
conducted only for those individuals age 13 and older. Also, estimated
totals from data items to follow are far more affected by bias due to
item nonresponse than previous data items. The nonresponse is indicated
by a reduction in sample size. The point should be emphasized that
these proportions are probably reliable, since proportions are not
affected as much by nonresponse bias (i.e., if the likelihood of obtain-
ing a response is a random variable). The Washington and Denver study
site findings indicate an estimated 33 percent and 38 percent of the
respective populations smoke or use tobacco in some form (Table 6.1.15).
Between 70 percent and 72 percent of the individuals in these sites work
-124-
-------
either full or part time (Table 6.1.16). Persons who travel anywhere at
least 3 times per week are estimated to be between 82 percent and
84 percent (Table 6.1.17) for the study sites. The* majority of persons
I
traveling (anywhere at least 3 times per week) ares estimated to take
short trips, with most trips actually taking less than. 15 minutes (Table
6.1.18). This is true for both sites.
6.1.3 Introduction to Sample Design Results ,
The sample design for the CO monitoring project incorporated
three methodology studies:
,(1) The use of telephone directory listings classified by Census
geographic variables in association with standard area house-
hold sampling techniques to identify sample housing units.
(2) The use of a lead letter mailed to sample subjects stating
j'
that they will be called to schedule an appointment for
monitoring. j
(3) Sampling person-days for monitoring, rather than simply
selecting persons and letting each person choose a day to be
monitored. i
The purpose of the remainder of Section 6.1'is to report the results of
i .
these methodological studies.
: ' ' 'i
6.1.4 Use of Geographically Classified Telephone Directory
Listings in Association with Standard Area Household
Sampling Techniques '. ,
As discussed in Section 5.1, the sample design for the CO
study was a stratified, three stage design. The EPA purposively select-
ed the metropolitan areas surrounding Washington, B.C., and Denver,
Colorado, as the study sites. Area sample segments defined by Census
block groups were selected at the first stage of sampling. A computer
tape listing the selected block .groups was then jsent to Donnelley
Marketing Corporation. The tape was returned with 'computerized listings
!,
of names, addresses, and telephone numbers for the selected block
groups. A sample of listings was then selected for each first stage
sampling unit to identify the sample housing units. The screening
interviews were conducted by telephone for all sample listings with a
telephone number. Since the target population included households other
than those with listed telephone numbers, field screening interviews
-125-
-------
Table 6.1.16 Estimated Distribution of Persons 13 Years or Older Who
Work Either Full or Part Time
Washington, DC
Metropolitan Area
Yes
Standard Error
No
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
Total
1,333,061
96,555
561,831
42,112
4,435
1,055
1,899,327
8,778
Proportion
0.7018
0.0140
0.2958
0.0139
0.0023
0.0005
Denver
Metropolitan Area
Total
464,960
38,133
179,528
13,535
602
258
645,090
3,951
Proportion
0.7207
0.0159
0.2783
0.0158
0.0009
0.0004
Table 6.1.17 Estimated Distribution of Persons 13 Years or Older Who
Travel Anywhere at Least 3 Times Per Week
Washington, DC
Metropolitan Area
Yes
Standard Error
No
Standard Error
Not Known
Standard Error .
Grand Total
Sample Size
Total
1,606,757
110,928
297,010
28,404
7,639
2,785
1,911,406
8,787
Proportion
0.8406
0.0085
0.1553
0.0084
0.0039
0.0015
Denver
Metropolitan Area
Total
530,420
41,049
110,533
9,833
4,189
2,571
645 , 143
3,952
Proportion
0.8221
0.0130
0.1713
0.0127
0.0064
0.0039
-126-
-------
Table 6.1.18 Estimated Distribution of Amount of Time Spent Traveling
One Way at Least 3 Times Per Week for Persons 13 Years or
Older ' ; /
Washington, DC Denver
Metropolitan Area Metropolitan Area
Total Proportion Total
Less Than 15 Minutes
Standard Error
16-30 Minutes
Standard Error
31-45 Minutes
Standard Error
Above 45 Minutes
Standard Error
Not Known
Standard Error
Grand Total
Sample Size
1,002,041
84,829
686,847
66,360
216,547
21,000
186,342
22,119
18,380
2,015
2,110,157
9,268
0.4748 357,480
0.0119 20,647
0.3254 257,484
0.0121 29,920
0.1026
0.0056
0.0883
0.0067
0.0087
0.0011
28,013
4,445
33,724
7,775
7,314
3,433
Proportion
0.5226
0.0239
0.3764
0.0228
0.0409
0.0066
0.0493
0.0097
0.0106
0.0051
684,014
4,152
-127-
-------
were also necessary. A subsample of the listings for which a telephone
interview was not possible (see Table 5.1.1) was selected for field
screening. Also, a subsample of listings was selected for a missed
housing unit (missed HU) check.
The missed HU subsample consisted of 150 listings for Denver and
300 listings for Washington. In each case, the FSUs which had listed
1980 Census occupied housing units 50 percent or more greater in number
than those listed by the commercial listings were deliberately over-
sampled. The missed HU check was implemented by using standard field
listing protocol to produce a unique geographic ordering on Census maps
for each FSU. Each listing in the missed HU sample was located in the
field. The interviewer then proceeded to the next housing unit as
identified by the geographic ordering and checked to see if that housing
unit was on the commercial list for the FSU. If not, a screening
interview was attempted and the check was continued at the next housing
unit. When the next housing unit was found to be on the commercial
list, the missed HU check was complete. Technically, a screening
interview should not have been conducted if the missed HU was on the
complete frame of commercial listings and was simply misclassified with
regard to Census block group. Interviews were conducted for all missed
HUs, regarding them as not simply misclassified, partly to check the
completeness of the listings. In most cases, missed HUs occurred in
groups of one or two. In one instance, an entire block face of five HUs
was missed (See Table 6.1.19). These five missed HUs were regarded as
misclassified, and their data were disregarded for analyses and selec-
tion of participants for CO monitoring.
The results of the missed housing unit checks for Denver and
Washington are summarized in Table 6.1.19. For each study site, approxi
.
mately two percent of the listings were found to not belong to the FSU,
or area segment, to which they had been classified by Donnelley Corpora-
tion. Although a unique geographic ordering was not possible for
listings outside the assigned area segment, a missed HU check was
attempted for these listings. The purpose of this check was mainly to
investigate the completeness of the commercial listings. The results in
Table 6.1.19 for start addresses outside the segment would seem to
-128-
-------
Table 6.1.19 Results of Missed HU Checks
&estllt
A. Start Address Inside Area Segment
1. Completed missed HU check and
found no missed HUs
2. Completed missed HU check and
found one or more missed HUs
3. Invalid start address
4. Could not locate start address due
to incomplete Donnelley listing
5. Could not identify the apartment
at which to begin the missed HU
check
6. Found one or more missed HUs but
not able to complete missed HU
check (unable to match names in
the Donnelley listings to apart-
ment numbers)
7. Start address and next address
were both office complexes;
missed HU check was aborted
8. Start address was inside an old
age or convalescent home; missed
HU check aborted as group quarters
were ineligible
continued
Denver
No. i 1.
12
I/
Washington
No. %
108 72.0 203 67.7
18 \L2.0 22 7.3
'
8 ' 5o3 2 0.7
0 0,,0 1 0.3
8,.0 39 13.0
0,7
0.0 8
1.3
2.7
0 . 0.0 16 5.3
-129-
-------
Table 6.1.19 (continued)
Result
B. Start Address Outside Area Segment
1. Completed missed HU check and
found no missed HUs
2. Completed missed HU check and
found exactly one missed HU
3. Could not identify apartment at
which to begin the missed HU
check
4. Aborted missed HU check after
traveling one mile, to first
corner, or listing nine missed
HUs
TOTAL
21 I/
Denver Washington
No. % No. %
0 0.0
0 0.0
0.7 0
0.7
0.3
0.0
4/ 47
2 1.3 2 0.7
150 100.0 300 100.0
\J Field work done by Research Triangle Institute.
2/ Field work done by PEDCo Environmental, Inc. under a separate
contract.
3j In one case an entire block face of five HUs was missed. The data
for these five HUs was disregarded. These missed HUs were regarded
as misclassified.
tjj The data for these missed HUs was disregarded. These missed HUs
were regarded as misclassified.
-130-
-------
Indicate that clusters of HUs, e.g., block faces, tend to be tnisclassi-
fled occassionally and that random misclassification of individual HUs
also occurrs. i
When the missed HU start address was inside an apartment complex,
implementation of the missed HU check was sometimels quite difficult.
Table 6.1.19 shows that it was not possible to complete the missed HU
check in apartment complexes for about 15 percent: of the listings
selected for the missed HU check. Since the listings did not generally
include apartment numbers, it was necessary to get apartment numbers
from mail boxes, apartment managers, or apartment residents. Sometimes
these sources proved fruitless. Many of the instances in which a missed
HU check could not be begun occurred in restricted-access apartments.
Missed HUs seemed to occur more frequently in apartment complexes than
in other areas when the check could be implemented. This may be due to
the more transient nature of apartment dwellers. Only one missed HU
check identified an entire block face (of five HUs) that had been missed
i
by the commercial listings within the selected area segments. The
general impression was that the commercial listings provided a reason-
ably complete listing of housing units.
Most of the listings were found to be correctly classified accord-
ing to block group. Donnelley Corporation claims that their listings
are about 95 percent complete. Our experience is not inconsistent with
this claim. However, we found that the undercoverage does not seem to
occur at random. Instead, there were small geographic areas for which
there were no listings whatsoever. The telephone directories for these
areas simply had not been used in compiling the listings. Standard
field procedures were used to list all housing units and select clusters
of sample housing units for these block groups.
The major problem encountered in using the listings to identify
sample housing units was that it was often difficult to locate the
housing units corresponding to the sample listings) in the field. The
addresses generally came from telephone directory listings. Hence, most
P
residents of apartment complexes all had the same address, namely the
street address of the apartment complex. This presented some problem
for location of the sample housing units. But, more importantly, it
-131-
-------
made the check for missed housing units very difficult to implement
correctly. Because of these problems and other more subtle problems
with the operational definition of missed housing units, there seems to
bfe no completely satisfactory way to perform the check for missed
housing units for a sample from the commercial listings.
Based upon RTI's cited experience using geographically-classified
telephone directory listings, it appears that the best way to use such
listings is to select two independent samples. Standard area sampling
procedures are used for one sample, and commercial listings are used
with the other sample. In particular:
(1) One sample is a standard area sample with sample clusters
identified from field listings of all housing units in the
selected area segments.
(2) The other sample uses the commercial listings to identify
sample clusters in the selected area segments.
It is recommended that the commercial listing sample be used only to
generate telephone interviews based upon the telephone directory list-
ings. Using this methodology, the standard area frame sample is used to
compensate for the bias resulting from the telephone interviews generat-
ed by the commercial listing sample. In order to compensate for this
bias, it is necessary to determine whether or not each household in the
standard area frame sample is included on the commercial listing frame.
This is easily done for commercial listings that come directly from the
current telephone directory. A single questionnaire item can determine
whether or not the household is served by a residential telephone number
that is listed in the current telephone directory. It is not so easy to
determine telephone coverage with respect to commercial listings based
upon vehicle registrations. .Hence, it is recommended that the vehicle
registration records be disregarded. See Whitmore, et al. [1983a] for
further discussion of these design recommendations.
Use of only the telephone directory listings for the commercial
listing sample makes implementation of the dual frame methodology very
straightforward. States with and without vehicle registration records
in the commercial listings are handled in exactly the same way. For
every sample household, one or two questionnaire items can be used to
-132-
-------
determine the number of residential telephone numbetrs listed in the
current telephone directory for the household. This information is
sufficient to facilitate unbiased estimation for linear statistics using
either'multiple frame multiplicity estimators, such; as those discussed
by Casady and Sirken [1980], or difference estimators,, such as those
discussed by Konijn [1973]. The difference estimators may be preferable
since they address the bias correction more directly.
The CO study found that there were some area segments with no
commercial listings whatsoever. Hence, a determination of whether or
not telephone directory listings are available is needed for,each area
segment in the standard area household sample. If telephone directory
listings are not available for some area segments in the standard area
household sample, the households in these area segments must be treated
for estimation as not represented in the frame of telephone directory
listings. Otherwise, all households with a currently listed telephone
I
number are treated as being present on the frame of telephone directory
listings. i
For monitoring studies, such as the CO study, . it is recommended
|
that half of the screening interviews be generated by the standard area
frame sample, and half by using the commercial listings sample whenever
this procedure is less costly than obtaining the same number of inter-
views from a standard area frame sample alone. Some savings will be
achieved by using the commercial lists instead of lists of housing units
i
produced by field staff to identify sample clusters. The use of tele-
phone interviews instead of field interviews may produce some additional
savings. These cost savings will more than compensate for the costs
associated with selecting and analyzing two independent samples for
studies that require a large number of screening interviews.
i
This dual frame approach (utilizing two independent samples) could,
of course, also be used for a field half-sample and1 a random-digit-dial
(HDD) telephone half-sample. Some advantages of using the geographi-
cally-classified telephone directory listings instead of random digit
dialing for one-half sample are the following: \
(1) Census geographic variables can be used to oversample subpopu-
lations of interest at the first stage. i
-133-
-------
(2) The proportion of telephone numbers called that are working
residential numbers will be much higher for the telephone
directory sample.
(3) When- field follow-up interviews are necessary, such as for
personal monitoring studies, the geographic clustering will
reduce subsequent field interview costs.
Of course, there is some loss in precision due to clustering and due to
use of the incomplete telephone directory frame. These losses will
generally be compensated by decreased cost for the sample survey. Thus,
the proposed design is expected to be cost effective for monitoring
studies.
6.1.5 Lead Letter Results
The sample design for Washington incorporated a lead letter
methodology study, as described in Section 5.1.4. A random sample of
596 of the individuals selected for monitoring was sent a lead letter.
The purpose of this lead letter was to inform the sample subject that he
or she had been selected for monitoring and that an interviewer would be
calling to schedule an appointment for monitoring. The individual was
thanked for participating and the importance of the study was stressed.
The lead letters appear to have had a positive effect upon response
rate. The overall response rate for individuals selected into the
Washington sample was about 58 percent, but the response rate for
individuals in the lead letter sample was approximately 63 percent.
(These response rates are calculated as the number of individuals who
agreed to schedule a monitoring appointment divided by the number of
individuals selected.) Hence, a person-level response rate of about 65
percent may be possible for future studies of this type using lead
letters for all individuals selected for monitoring.
6.1.6 Sampling Person-Days
Individual exposure to carbon monoxide is heavily dependent
upon both weather patterns and personal activity patterns, as discussed
in Section 5.1.4. Since weather is such an important factor, it was
necessary to field as large a sample as possible on each day during the
study period. Otherwise, there would be no one monitored on the days
with weather patterns producing the highest CO levels. Since personal
-134-
-------
activity patterns are important, the sample participants could not be
allowed complete freedom of choice in selecting a day to be monitored.
The sample subjects could introduce a bias by selecting mostly days when
they plan to be inactive or stay at home. The strong influences of both
weather patterns and personal activity patterns upon CO exposure sug-
gest that a specific day should be randomly selected for each individual
to be monitored. However, such a procedure is totally impractical. . The
response rate would surely be so poor as to invalidate the study if only
one day was offered for each sample subject to participate. Hence, the
sample for Washington was randomly allocated to non-overlapping three-day
interview periods. This procedure had a greater negative impact on the
response rate than was at first anticipated. Some sample members
indicated that they were willing to participate, btlt not within the
selected time period. These individuals were given one additional
opportunity to participate by randomly reallocating them to one new
three-day period. A total of 550 individuals were reallocated in this
manner for the Washington sample.
The methodology for person-day sampling used fo'r the Washington CO
study is somewhat awkward. The reassignment to a new 3-day interview
period required constant interaction between field jstaff and sampling
staff during data collection. It also required continual updating of
sampling files. Hence, the sampling task was much more expensive than
that of a typical sample survey. A better procedure needs to be found
for future air monitoring studies. The procedure must control alloca-
tion of persons to days and still be somewhat flexible with regard to
the allocation to days. A methodology study to explore alternative
methods of person-day sampling for studies monitoring personal exposure
to airborne pollutants may be needed.
One methodology for person-day sampling could! be the following.
Suppose six days were randomly selected within the skudy period for each
person selected for monitoring. These could be three consecutive days
in one week and the same three days in the next week. Suppose further
that priorities from one to six were assigned to the days selected for
each person. Each person selected for monitoring icould then be told
something like the following:
-135-
-------
You recently participated in an EPA-sponsored study by
responding to a short questionnaire. You were told that
someone might be calling on you to participate in a per-
sonal monitoring follow-up study. You have been selected
for participation in the follow-up study. Due to the in-
fluence on exposure by weather patterns and personal ac-
tivity patterns, it is necessary to monitor a representa-
tive sample of people of each day in the study period.
Hence, I have an ordered list of six days that have been
selected for you. You are asked to choose the first of
these days on which you can possibly participatedOther-
wise, your data may be discounted in the analyses and not
have as much impact as it would if you had participated on
the first available day. Hence, would it be possible for
you to participate on (Day 1)?
If the person could not participate on Day 1, the interview would
proceed to Day 2,. etc. A short reminder that the person needs to choose
the first day on which participation is possible might be appropriate
between offering Day 1 and Day 2.
Some type of weight adjustment procedure could then be used to
compensate for the bias due to self-selection of days. For example, a
weight standardization could be performed using a covariance model.
This type of procedure could be used to adjust each day's sample to a
standardized population, based, for example, on age, race, sex, and
occupation. This might require a fairly large sample, e.g., enough to
produce more than 25 respondents each day.
6.2 Field Survey Activities
6.2.1 Survey Post-Field Activities
When the field interviewer returned to the respondent's
residence at the end of the 24-hour monitoring period, they began a long
chain of post-field activities. These activities included editing
documents, providing numeric codes for certain alphanumeric fields,
entering the data into the computer, and preparing the data for the
final step which was the analysis.
As each interviewer retrieved a set of documents from a respondent,
the documents were quickly reviewed for completeness and legibility.
Any obvious problems were to be addressed and resolved while at the
respondent's home. Documents were then returned to the field lab and
logged in on the case control card. All materials for each case were
-136-
-------
maintained separately and material was returned to RTI on a regular
basis, transported by the returning members of the field laboratory
staff.
When the documents were received at RTI, each case was handled by
opening the storage envelope, logging-in on the control card all mate- .
rials found, and then separating the four documents for the case. All
control cards were filed in chronological order. All consent form/-
incentive-receipts were batched by wave and hand delivered to the survey
task leader for secure storage. The study questionnaires and activity
diaries were put into batches (maximum size of 30) for further process-
ing. Batch header sheets (Figure 6.2.1) were created for use in tracing
the progress of any batch through the data processing activities.
Batches were first given to editors and coders who were told to
review all documents for completeness, to attempt to resolve apparent
discrepancies, to attempt to decipher illegible entries, and to code
alphanumeric fields, using pre-prepared_code specifications. Batches
were quality-controlled by the data preparation supervisor who reviewed
100% of the first two batches of each editor/coder -, and then 10% of all
subsequent batches. Discovery of any systematic errors led to retrain-
ing.
After batches were completely edited and coded, they were sent to
data entry. Using RTI's mini-computer system, the ;documents were keyed.
A 100% rekey provided complete verification of all data. Any discrepan-
cies between the two keyings caused the keyboard at the data entry
station to lock up and required problem resolution before proceeding.
Data tapes were prepared on a regular basis with output files checked
against hard copy to assure the correctness of the files. Outputs of
the mini-computer data entry system were used to create the data tapes,
which were then available to the statisticians for use in analysis.
During the first phase of analysis, unusual] values, unexpected
values, and outliers were identified. These data points were then
checked against the hard copy of the data (see Section 5.4). After the
problems were resolved, the original hard copy of the data was boxed and
stored in data vaults.
-137-
-------
Type of Document
Study Questionnaire
Activity Diary
Field Data Sheet
Action Taken
Figure 6.2.1
Project 2390
Carbon Monoxide Exposure
BATCH HEADER SHEET
Quantity
SOC Scan Edit ....
Edit and Code
SOC •»• Data Entry .
Data Entry
Data Entry •>• SRDC
Stored
Batch Information *
Batch No.
Batched By
Date Batched
By Whom
Date
Study Numbers of Enclosed Documents
* Xerox copy of Batch Header Sheet to H. Zelon, 300 Park as soon as created.
-138-
-------
6.2.2 Post Data Collection Discussions
After the field data had been collected, two additional
interviews took place. One was a validation of a sample of respondents.
Several key data points were reviewed and an attempt was made by the
supervisor conducting the interview to ascertain data on the performance
of the interviewers.' All data points checked yielded appropriate
responses. Interviewers were given uniformly good comments in terms of
being personable, well prepared, and helpful to the1respondents. All
interviewers were said to have explained the study thoroughly and
appropriately. One question was asked about the incentive. Many
respondents sampled stated that they would have participated without any
compensation, but most were pleased to have received it. Respondents
were also asked if carrying the monitor affected th|eir daily routine.
The general response was that people maintained their normal schedule of
activities.
Each of the field interviewers was also sent a !brief questionnaire
which collected data on the interviewers experiences during this study.
Responses were received and tabulated. In general, the responses
received were useful only to provide written documentation of comments
already provided by the field staff. Several of the comments from the
field staff were useful and are integrated into the study recommenda-
tions .
In particular, fifteen responses were received from the field
interviewers who were sent the Post-Fieldwork Questionnaire (Appendix
L). These responses are representative of those interviewers who worked
on the study for the major portion of the effort. The answers to the
questions are summarized in the following paragraphs.
Question One asked the interviewers about problems the respondents
had understanding the operation of the PEM. The only problems mentioned
by the interviewers involved the size and "hassle" !of carrying the PEM,
confusion with the "on-off" button, and definition of an activity. No
one mentioned actual operating problems. i
Question Two dealt with the reluctance of some respondents to carry,
the monitor. Several interviewers reported refusals directly related to
the PEM. Some interviewers reported reluctance, later overcome, related
-139-
-------
to size and potential job conflicts, possibly related to dangerous
situations.
Question Three discussed the incentive payment. The comments
returned yielded no clear consensus. The only common fact was that less
than 40% of the respondents stated that they would have participated if
there was no incentive. The remaining comments covered the spectrum of
larger, the same, and smaller incentives working equally as well. The
single comment that.a larger incentive, if big enough, can convert most
refusals is a well-known concept but in general is only anecdotally
documented.
Question Four involved respondents' difficulties in completing the
questionnaire and diary. Problems reported included the difficulty
defining and documenting leisure activities, the inconvenience of the
diary, some problems with the diary layout, some problems with multi-
part questions, the tediousness of maintaining the diary, and the
reluctance to record activities. In most cases, good interviewers
removed difficulties with good explanations and reviews of data require-
ments .
Question Five asked what additional information could be given to
the respondents by phone to reduce fieldwork. Comments included a need
for better explanation of the size of the PEM, reduced replication of
questions, a need to stress the necessity of the availability of the
respondent for two appointments with the interviewer, further explana-
tion of how the respondent was selected, a need to allow respondents to
refuse - no oversell by phone, and a stated problem of inadequate
screening since some smokers were selected.
Questions Six discussed problems interviewers had in collecting the
breath samples. The only problems mentioned were the sample bags
leaking, a few concerns over sterility/disposability of the stems, a
need to tell respondents about the breath sampling at an earlier point
in the study, and a single problem of collecting the sample from a
respondent with only one lung.
Question Seven was a branch point to allow those interviewers who
had the new H-P unit to continue.
-140-
-------
Question Eight discussed respondent reactions to the new monitor.
Problems included the existence of too many buttons with poorly defined
activities and the lack of reliability of the monitor. One interviewer
reported a situation where one family member used the initial PEM and a
second member got to use the new version. The new unit was highly
"preferred". . j '
Question .Nine asked if, after the new model was made more reliable,
it could be integrated into this type of study, and if, by thus removing
the Activity Diary burden, the response rate would increase. The
reactions were that the unit would be useful, but needs to be made
smaller with larger buttons. The interviewers felt: that the Diary
should be maintained as part of the study. It serves as a reminder to
the interviewer to be sure the respondent understands the study. The
diary is also more easily corrected than the PEM.
6•3 Field Measurements and Quality Assurance ,
6.3.1 Field Measurement Activities |
, , Field measurement activities took place between June 1, 1982
I
and April 1, 1983 with the actual acquisition of field data occurring
between November 7, 1982 and February 24, 1983. These activities
included preparation of the standard procedures for the analysis of
breath and ambient CO levels; acquisition, verification, and calibration
of field standards and equipment; acquisition of actual field data; and
validation, reduction, and preliminary analysis of said data.
6.3.1.1 Personal Exposure Sampling
As .mentioned previously, ambient sampling was
conducted in seven sampling waves between November 8, 1982 and February
25, 1983. Each wave lasted 2-3 weeks. The target sampling rate varied
by wave, with a rate of 10 per day proposed for Wave 1:, 15 per day for
wave 2, and 20 per day for Waves 3-6. However, because of not being
able to schedule enough appointments, sample cancellation, rescheduling
at the request of respondents, and COED-1 monitor malfunctions, the rate
of successful completion averaged between 8.9 and 12.3 for Waves 1-6.
Since this lower sampling rate lowered the total number of samples
acquired to an unacceptable level, sampling Wave 7 was instituted to
recover some of those samples lost to scheduling problems, monitor
-141-
-------
malfunctions, etc. The rate of successful sample completion during this
wave was 10.3 per day.
Some statistics for the sampling waves are presented in Table
6.3.1. The percentages for samples completed, cancelled, and lost are
based on the number of samples attempted. The percentages for validity
code data and for the last 3 rows of the table are based on the numbers
of samples completed. It should be noted that an average sample contain-
ed approximately 50 time/value pairs. Therefore, the number of mistakes
detected represented only a small "portion of the total data reviewed
(approximately 0.1 percent). Duplicate, colocated samples were collect-
ed by lab personnel for the purpose of characterizing monitor precision.
The duplicate sampling was discontinued in Waves 2 and 3 due to the
shortage of monitors created by the reliability problems coupled with
the increased daily sampling rate. Duplicate sampling was diminished
during Waves 5 and 6 for the same reasons.
Data in the Table 6.3.1 column marked "Total" represent the status
as of the end of the field monitoring phase of the project. At that
time, 1051 samples had been attempted in the field (i.e., a field inter-
viewer had left the laboratory with an operating, calibrated monitor).
Of those attempted, 814 samples had been completed (i.e., the monitor
had been returned to the laboratory in an operating condition and
containing apparently valid ambient data). Samples which were cancelled
or rescheduled at the request of the respondent accounted for 107 of the
1051 attempted. The remaining 130 samples were lost to monitor malfunc-
tions during the sampling period (i.e., the monitor was returned to the
laboratory in a nonfunctioning or malfunctioning condition). During the
data validation phase and prior to entry of the data into the study data
base, 40 of the 814 completed samples were invalidated for reasons
related to the quality of the ambient CO data. Twenty-two samples
having a validity code of 4 we.re invalidated because of differences
between pre- and post-sample slopes or intercepts of more than 20
percent or more than 2.5 ppm, respectively. An additional 18 samples
having validity codes of 1 - 3 were invalidated due to unreasonable
anomalies in the data. Examples of such anomalies include large blocks
of missing time and/or average data within a sample and gross, unre-
-142-
-------
0]
00
c
•H
03
CO
g
•H
g
m
u
•H
4J
CD
•H
•U
ctf
4J
rt
H
1-1
f4
b
U
3
£
Q)
01
"
0
91
g
i^
in
o
twl
91
in
CM
en
N
N
C»
VO
SM
r*.
^^
-»
00
04
a-s
o
CM
**
»<
CM
C<4
i-^
Csl
94
00
^
00
^4
>4
•^
^4
N
94
-»
oo
r*i
94
w4
r-
Ml
i
Samples Cancelled —
94
CM
8
-
94
3
•»
*H
94
ov
.
«
—t
•^
s
94
cn
r»
en
94
O\
^
1
3
5
§
2
cs|
4J
a
o
J
a
45
i-
a
en
«
s
g
m
94
R
p^
p-
94
00
vO
-»
*
w
p»
sO
»^
-3-
tN
*>^
94
a\
54
«J
p-
«
vO
94
P-
p»
«M
a\
^
94
en
r—
-j
00
en|
»4
O
•o
a
o
X
u
•H
•o
•rt
^4
g
94
-»
•^
•^
»4
94
r»
r~
94
CM
••*
oo
94
CO
M)
**s
en
en
»4
sr
e»\
M4
94
vO
94
vO
s^
*+
»H
94
m
p*
94
••*
—4
.-4
-4
X
in
VO
9<
CO
•a
o
CJ
>,
4J
•H
•o
•H
*-*
•5
94
r«.
-H
•-'
en
en
*H
>4
O
~4
O
•M
94
-»
p«4
CT\
94
4
en
«M^
co
M4
94
CM
CM
94
cn
CM
m
en
8
•*
p^
-»
kl
Samples Reread -
i
i
M
en
en
•^
-4
sr
I
i
Mistakes Detected —
94
•£>
f*4
in
94
•»
-Cf
t— (
94
en
CM
94
sr
^^
CO
94
O
%•*
sr
^H
94
O '
0
**N
00
o
>*l
«
>*H
CM
i«4
Duplicate Samples —
4J
S
0)
•o
§
o.
«
2
J
4J
U4
0
4J
09
0)
3-
01
h
4J
4J
cd
•o
01
f-i
3
1
"o
a
0)
14
14
O
•a
01
i-4
i-i.
0)
o
!
B
•H
14
0
4J
to
5
s
91
3
cd
4J
ed
•0
01
J3
u
j.
o
•H
'*
O
U-l
a
0)
r-l
i-
a
to
•O
0)
U
01
1-1
eu
B
3
N|
to
to
C
cd
0)
B
»4
0)
T3
0
CJ
CO
4J
a.
0)
u
M
0)
.u
C
•H
•a
c
S
to
9)
a.
o
f-4
a
§
•H
JJ
a
14
5
t-t
3
9)
i-l
0.
S
ed
CO
1
«j
a
i.
•a
B
ed
t
2
a.
a)
Ij
0
«J
•H
C
91
A
U
B
9>
§
U
^Q
O.
•H
to
C
o
Ti
u
Cd
1-4
• a
M
01
J=
4J
0)
J-l
cd
u
1-4
•a
B
•H
CQ
9)
•a
tS
en"!
Q.
S
U
91
U
B
T-t
1
ed
8.
o
1-4
ta
91
i-4.
!•
id
to
0)
u
a
S
u
J3
4J
•H
5
*
91
U
00
ed
4-1
a.
8
I-I
91
4J
•H
•9
B
cd
91
a.
o
rf
09
B
O
t4
u
cd
H
^
•rt
i-4
cd
u
91
i-)
§
!
4J
00
a
91
TJ
14
a
a
01
u
h
91
4J
c
.•H
•a
cd
O. 91
o. o.
m
i— i
+i
0
4J
-M
+1
U-l
0
•o
B
td
91
n.
0
r-4
09
91
X
4J
U
O
14-1
94
O
M4
0
M
94
m
+1
IH
O
91
U
B
91
V4
9)
IH
(H
•H
•0
cd
to
9)
4J
Cd
u
•H
1
•H
o
B.CM
01
4j
J3
U
•H
J=
4
9)
r-4
O.
a
cd
to
TJ
01
4J
91
i-4
U
>.
•-1
i-l
3
U-l
to
to
0)
o
o
3
0)
01
"o
U
_>.
9)
U
U
O
I-l
to
sr
0)
T3
O
U
4J
O.
91
O
M
91
4J
B
•rt
5
4J
1-1
,0
tH
'B
a
o.
CM
•H
O
4J
m
^4
•H
•a
s
'Cd
9)
a,
o
1-1
to
91
a
u
u
o
1U
>tt
sn
^4
o
4J
3
.-4
+1
O
a
91
u
IB
91
'•H
9) '*4
0.<
4J
1-1
t-H
cd
3
0-
H
O
VH
&
s
ed
u
ed
T3
V4
o
4J
^
B
§
tu
0
I
•O
cd
9)
M
91
U
ed
O
1-1
^4
0.
•3
o
4J
"0
91
4J
a
01
i-i
1
CO
01
•-4
e-
s
cd
tn
5|
IH
O
u
3
O
«k
CM
91
>
S
o
U-l
M
oi
^•*
oo
5
•a
91
9)
U
cd
CJ
•rt
r-4.
B*
3
•o
&
S
•u
9
•H
V4
3
•a
9)
iH
CO
cd
5
.£
44
•rt
3
•a
CJ
4-J
CJ
91
4J
01
•O
a
4J
at
•n
01
3
i-4
cd
>
V4
O
91
5
4J
•H
ed
3
•a
-H
>
•H
^
5
to
.3
m
u
CO
s
«l
^
x^
•O
O
VH
91
H
91
3
ca
h
O
U
14
91
^-
5-
B
O
£
U
ed
01
to
H
•rl
ed
CU
3
1-1
1
0)
i
•H
4J
o
m
3?
91
4J
cd
4
X
o
M
O,
o.
Cd
J5
«J
•H
•s
9)
M
91
14
to
4J
I1
a
to
to
9)
1
3
•«
0)
u
U
o
u
«
u
(4
U
•H
O-
^1
n
0)
f-H
a*
Q
§
to
M
0)
U
(U
1-1
t
u
B
O
•o
91
a)
cd
CO
91
00
cd
u
0)
u
01
P*4
-143-
-------
solvable differences between activity diary and ambient sample data
point times. Invalidation of these 40 samples left 774 valid ambient
samples available for inclusion in the data base. As has already been
described in Section 5.4.2.1, during the creation of the ambient data
file, the remaining 41 samples having a validity code of 4 were invali-
dated along with 21 other samples eliminated for reasons wholly or
partially relating to monitor malfunction. This brought the total
number of samples lost due to monitor malfunction to 232. This repre-
sented an overall monitor malfunction rate of 22 percent, considerably
greater than the 10 percent rate targeted at the beginning of the
project. Of these failures, approximately 73 percent were attributable
to outright failure while 27 percent were attributable to instability in
monitor calibration.
The frequency data listed for the occurrence of each of the four
validity codes in Table 6.3.1 indicate that these monitors were general-
ly stable over a 24-28 hour sampling period. Across the entire project,
post-sample monitor performance was within ±5 percent of pre-sample
performance for the slope of the calibration curve (and within ±1 ppm
for the intercept) 72 percent of the time. The slope and intercept were
within ±10 percent and ±1.5 ppm, respectively, 86 percent of the time.
6.3.1.2 Analysis of CO Levels in Respondent Breath Samples
Samples of respondent alveolar air (breath samples)
were collected throughout the field monitoring phase of the project.
Field interviewers collected the samples at the conclusion of the
24-hour ambient sampling period. The samples were analyzed in the field
laboratory within 24 hours of collection. Table 6.3.2 presents the
results of the breath analysis by ambient sampling wave. Successfully
completed samples averaged 5.4 ppm (v/v) with a standard deviation of
5.2 ppm. The geometric mean was 4.4 ppm. Sample values ranged from 1.2
to 54.7 ppm. Successfully completed samples numbered 870. The number
of breath samples with valid exposure data was 659.
6.3.1.3 Fixed Site CO Data
During the period of the field study, from November
8, 1982 through February 25, 1983, EPA collected CO data at eleven fixed
sites in the Washington, D.C. area. These data were used to classify
-144-
-------
&
co
in
en
.
^^
oo m in en -4 en in
•••4 sr in sr NO m ••*
x1 sr
NO
tfm^
en in NO en r*» •* CM
— 4 sr in en in en -4
"^^ sr
sr
-
B1^
en co oo m -H en in
—4 o m sr NO en —*
— ' sr
M
o o o sr *^ NO r^
•^ sr m sr sr NO «" »
>-< CM
m
. "
a
•« 0
0) i-l
01 -4 1 CO
>H a -H
fu o) >
o a
•r( 13
m a u ^ 0 g
NO
^•'
00
1 -
1
,^
i,."A
1 i~4
cn
i H
i 1-1
1 1
1 05
O
sr
CM
-*
en
CM
in
CM
sr
m
CM
-*
CM
en
CM
NO
—4
CM
in
eo
pj
§
o
o
en
o
sr
CM
-1
en
^
sr
CO
O
en
o
NO
C9
O
NO
O
00
o
c
5
4J
01
a
•o
M
10
"B
a
en
in
I
fs.
o
en
in
1 '
00
m
en
I
r-
"?
in
sr
1
sr
en
en
1
.
0
en
:en
1
CM
en
en
1
en
a\
CM
t
^
O
'
0>
to
00
eg
•3
CD
3
a)
U
01
,0.
eg
O
u
M
O
•H
kl
O.
,
u
3
I
$
a
3"
i
IS
CO O ,K
01 O
1-1 k4 J3
cu 01 a
•a •§ ^
CO
oj a co
U O it
o) tn
r-l t) >,
a. oi - ,;-(
-145-
-------
days as low CO days and high CO days for later analysis. Table 6.3.3
describes the site characteristics of these monitors and Figure 6.3.1
shows the approximate location of each site on a map of the Washington,
D.C. area.
Table 6.3.4 summarizes the results of an analysis of the hourly
average values during the period of interest. In this and succeeding
tables, the "composite site" data were created by taking the hour-by-
hour mean of the hourly values reported by the eleven fixed sites. None
of the sites reported hourly average values exceeding the standard of 35
ppm. Table 6.3.5 lists the date and time of the maximum value reported
at each site. Eight of the eleven maximum values occurred during either
the morning or the evening high traffic periods. Three days (Nov. 8,
1982, Feb. 15, 1983, and Feb. 22, 1983) account for all but one of the
maximum values. The maximum value at the composite site was 8.6 ppm and
occurred at 18:00 on Feb. 15, 1983.
A file was also created containing the daily maximum 1-hour and
8-hour CO values. Tables 6.3-6 and 6.3.7 summarize the results of an
analysis of this data. As indicated in Table 6.3.7, two sites had daily
maximum 8-hour values exceeding 9 ppm. Site 090020023102 reported one
exceedance; site 210220001F01 reported five exceedances. None of the
sites had daily maximum 8-hour values exceeding 15 ppm.
6.3.2 Problems With Monitor^
COED-1 monitors were used throughout the field monitoring
phase of this project. GE/HP units were field-evaluated during sampling
Wave 7 from February 15 through February 24, 1983. The following
problems were noted with each type. It is estimated that approximately
1/4 - 1/3 of the field man-power effort was expended on corrective
maintenance and repair activities for the monitors.
6.3.2.1 The COED-1 (GE/Magus) Monitor
A total of 49 COED-1 monitors were employed in the
field sampling in Washington. Of these, 6 to 15 monitors were out-of-
service on a daily basis for corrective maintenance or repair. This
out-of-service rate is attributed to three main causes — (1) failure of
the bias battery mounting system, (2) failure of the nickel-cadmium
(Ni-Cd) batteries, and (3) failure of the sample pump.'
-146-
-------
-a
p
t j
CO
60
S
O
oo
•H
4J '
cO
M
CU
P-i
O
CO
o
4-1
•H
C
0)
•H
X
O
PS
O
S
(S
o
CO
u
c
o
4-1
60
c
,C
CO
CO
0
CO
u
•H
4—*
CO
e Character!
4J
•H
CO
CO
CO
cu
rH
rQ
E-i
Immediate
Area
Land Use
CO
cu .>,
' rH Cfl
if c2
CD .
O rC <4-l
PU
CU
^> rH
co|ctf
co
Address
CU
13
O
CJ
1
CO
n
o
•H
4J
(0
O
O
rJ
cu
CO O
S u
Commercial
o o o
o o o
o m o
CM «* r*.
•— i CM
co o m
oo r^ «a-
CO
CO
CO
Z
^_^
5 &*
• M
CO ,0.
4-1 1-1
CO ,J
^'1
JS CO
4-1 CU
-a- £3
CM '— '
,_,
09002001710
"4-1
O
4J CO
0 -H
•H ,0.
5-1 6
4-1 9
CO rH
•H O
O 0
<1
Street
corridor
o o
o o
en r-<
O CO
CM rH
oo o
rH rH
CM
rH
i— 1
S
j.
4J \S
CM
CU CO
CU 4-1
S CO
•H rH
S -H Q
CO CO S
W
Light com-
mercial &
residential
0 0
O 0
« A
CO CO .
O 0
vO
CO
S3
•
4-1
CO
t?
CO 4J
rH
CO CU
PM
•
r-* a
in >— ' *
__i
480080009H01
„
cfl
•H
M
T3
C
CO
X
CU
rH
h'
Commercial
and
residential
o o o
o o o
m CM o
V V •>
vO
O C O
O O 00
CM CM rH
vO
SB
co
CU
SB
CO
<&
4-1
OO
rH •
CO
• 4-1
co co
•
480200020GO]
^
o
4-1
00
3^
o
Office
buildings
o
o
CM
V
O
O
V
CM
rH
53 .
CU
g
cu
4
cu
60
CO
O
o
o
rH
481040005G01
.
a
14-4
S-i
•H
CO
fn
ta
Light com-
mercial &
residential
O O c-
o m
O\ CM
•i ««
r«. oo
rH
O O O
CTI m oo
rH CM rH
CM
!_ 4
53
pH;
rH
cO
13
O
g
CU
co
m
vO PS
CM O
481060018G01
ft
—^
o
c
CU
S >
M
cu
I
4-1
C
o
u
-147-
-------
1
^
*c
••
4J
C
0
CJ
s^x
CO
•
en
*
vO
eu
r-{
W
"
0)
4J 01
cd co
•H Cd !3
"C 0)
OJ l-l "O
f4H
pi
(U
--. r-t
CO
CO
to
0)
M
t3
0)
O
O
P
O
pcj
*gj
CO
C
0
•H
cd
u
0
•J
0)
Cd O
S u
rH
1 cd
B * -H
O 4J
0 i-H C
cd 01
4J -H T3
,fl 0 -H
00 M W
•H QJ 0)
. _3 C4 Li
r^ H* ^*
O O\ O4
m co en
f^. r-t CO
•-I en o
en r-i
CO VO CO
\o r-i en
CM
r-l
55
*
s
rH
•H
CO
r-l
t~H
cd
PQ
f«^
en
•a-
rH
O
0
r-l
O
O
0
in
CO
rH
CO
•*
*Q
^>
*
rj
cd
(U
rH
U
,_
4J
C 1
Cd (3
4J -i
g 1
0 ilO
O 4J
4J ti
00 O
tf S
•H
i-l UH
rf O
<^
UH /^*s
O O 13
O 0 rl
•-I rl Cd
VO ^— ' |3
rH
0
O
N<
(
C
•H
cd
4J
}H
0)
o
c
=3
0
TJ CO
O rl
O 53
n cj
rl CJ
o cd
J3 <3
60 !H CU
•HO 4J
CO
23 S W
»^_ ^^
cd| ,a|
-148-
-------
-149-
-------
CO «3-
m
i
o ; jo co
60 1°* CM
C i
,C
ca co
nj £3 fl) in CO
P. -H r-<
« w
C O)
•a o o
0) Ti H O CM
H OO CO P-l <-<
O ON M
P.r-4 4-1
ai c
pe; n 0) I
in oj m P-
OT CM PM CM •
01 O, O
CO CO O)
n -H o m
0) rQ X •— 1 *
»O 01 O O
•H Pd R
X O
G C •
o co c >
S O flj CM
CM ,a p CM
C OO H , •
O CT\ Cg • r— 1
rO rH O T3
J-l 1 '
CO » (1) CO
U CO M
CO
01 M !-i
00 0) 0! C 00
r-4 S •<« Ol •
ai co S T-H
r*» i 3 6
-Tj g og^
HO P3 B "
3d) -H 1-1
0 & i X -<
*** ^i ' 5
M rn
o
o> G
M 4-1 3
0 T\ 6 •-<
w W Co
W -H
4-1 O
CO 4J H r>i W
•H 01 r-l 0) vO
b~, C ,Q iw H 3 CM
1-1 0 6 0 3 r-l VO
CO S 3 O CO CM
§ 53 ^2 >
g
CO —<
a) o
T3 H
-3- o P-
• O •-<
CO O
• £2 CO
VO << CM
O 0
01 pd O
r-l
O
pa
.*
m
i — i
CO
CM
CM
VO
rH
1— <
1—4
OO
o
^.
•— i
r-l
00
O
O
•—I
^
o
p^
r-~
CO
CM
CM
O
H
i— I
CO
O
0
CM
O
O
er\
o
O
en
00
0
-d-
cn
CT\
O
st
O
CM
O
CM
CM
CM
in
vO
sf
CM
CM
0
o
r~-
i— i
CO
CM
O
f»
r— 1
O
o
o
CM
CM
O
i— I
CM
Q
r-l
<•
VO
CM
CO
CM
P-
0
,Vf
O
r~.
en
O
ON
CO
r-l
o\
o
o
m
o
CM
O
fu
r-l
0
o
o.
vO
iri
r-<
l~l
CM
W
o
m
o
CO
m
O
.m ,
o
m
o
^
CM
r-l
00
CO
o
CM
in
oo
o
o
ON
r-4
VO
CM
.^
cO|
O
O
O
CM
0
0
o
CM
o
oo
«3"
O
in
sr
o
CM
m
o
o
rH
m
o
m
CO
o
CM
O
rH
0
I—I
in
o
o
o
r— 1
ON
-,- O
oo
ON
o
CO
co
1— 1
o
o
-3-
vO
m
CM
.
cO|
o
0
-3-
o
o
o
r-.
oo
CM
CO
4
•H
01 13
4J 01
•H 4-1
0} 5-1
0 O
rP-i
CD
0
CO
JJ
CO
O
^.
cfl
-150-
-------
Table 6.3.5 Date and Time of Maximum Hourly Average
Carbon Monoxide Value
Map Code
A
B
C
D
E
F
G
H
"; I
J
K
SAROAD Code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H01
480200020G01
481040005G01
481060018G01
481850001G01
482870004G01
Maximum
Hourly Avg.., ppm
11.2
16.0
10.0
22.4
9.1
14.0
8.5
8.0
17.0
9.5
13.0
Date
11-08-82
02-15-83
11-08-82
02-22-83
02-22-83
02-15-83
02-22-83
02-22-83
12-08-82
021-15-83
11-08-82
Time
20:00
19:00
17:00
7:00
7:00
19:00
8:00
8:00
8:00
18:00
9:00
-151-
-------
C
O
•u
00
c
•H
rC
(0
(t]
ts
>°
*XJ
0 ,
Ico
^
»2
rt C
S GJ
O 4J
CO
o bO
•H (3
4J -H
CO M
•H O
•U 4J
« -H
OO O
b
vO
CO
VO
a)
r-l
rO
CO
H
1 ^^
OS
o
OS
£3
p.
p, ro
i CD in
M i"^
0 4J
•H C
W (!)
re o
>•! H O
4J|
*fi iS
5rt
^
s 1
"rt "x
f"S
e
|
•H
•H
|y]
M CC
a) >-, d
,0 IW 1-4 3
g O -H i-
H T3 >
01
TJ
o
u
§
§
<
V.
a
p.t:
rt c
CO
OS
OS
vO
m
^j.
r^
CM
OO
r— 1 '
sr
i— i
oo
CM
CM
VO
m
CO
CM
1— 1
r-l
r^.
O
0
1—4
I— 4
090020017101
)
1
! •<
)
r-l
^j.
1—4
O
I-H
i-H
CO
OS
VO
vO
O
^P
•*
CM
O
CO
CO
-d-
r--
VO
o
vO
r-H
m
r-l
CO
O
i-H
090020023102
M
m
OS
Os
vO
co
CM e
-------
c
o
00
.g
1
£
****
*X3
CU
4-)
]»l
O
CU CO
cu oo
erf a\
«— 4
CO
Q) «
3 m
iH CM
cd
> >,
cu cd !
-0 3
•H H
0 "§
Cj Jj_|
0
S3 -a
c
C cd
ja CM
M 00
O -4
)-l »
3 00
0
ta *
I CU
CO ,0.
e cu
3 >
"d ^
to s
Sol
u*
, CU
iH W
•H CU
cd PQ
O
co
i-i cu
O 4J
m i-i
CO
U 00
•H C
4J -H
co )-i
•H O
4J ' 4J
to -H
4-* CJ
CO 0
•>.
1
i
3
CO
r-~
CO •
*
cu
r-l
o
td
00 vO
in
1 S N
! *
11 ''
O.1 CO
% 3 r-
•H -H r^
Q cd
*STj
S
3 ^D
"c o
•H
S
M co
cu !>, cu
rO «W rj 3 0
g O -H r-l —1
SB -a >
cu o
T3 M
, 0 l-~
O
9 o
00
o
*-*4
0
o
0
CM
CM
O
r— 1
CM
0
CO
,-j.
«*
CO
vO
CM
ON
1—4
CM
r-~
0
m
o
•
^^
CM
i— i
VO
m
CO
o
£
o
fn
. I— 1
o
o
o
m
i— i
i — t
CM
S3
^
vO
CM
PI
•ft
CM
vo
^
0(3
O
v«3
O
r«4
CM
;•
*"'
CPi
r~
r—t
VlS
VJ3
-------
The Magus data unit used in the COED-1 monitor was supplied origi-
nally with two "hearing aid" type batteries strapped to the circuit
boards. These batteries provided bias voltage to various electronic
components within the data unit circuitry. The batteries were strapped
into their respective circuits utilizing two, spring steel straps — one
soldered to the circuit board and the other screwed to the board. Even
before the field monitoring phase was initiated, it became obvious that
the battery mounting system was not reliable. If the screws retaining
the battery were tightened too greatly, the solder joint(s) on the
second strap would break. If the screws were not tightened enough,
proper electrical contact between the battery and the circuit could not
be maintained. Loss of the bias voltage could produce any of several
unpredictable effects. However, the two main effects noted were "lock-
up" of the data system logic and rapid discharge of monitor main bat-
teries.
Beginning in November and continuing throughout January, COED-1
units were returned to Magus in California in 6 unit batches for retro-
fitting of the bias battery system. The battery-and-strap system was
replaced with an "active" voltage supply system operating on the main
battery package for the data system. This retrofitting eliminated the
bias battery problem. However, problems involving a defective batch of
integrated circuit chips, erroneous installation of circuit components,
erronous wiring of circuit grounds, and the susceptibility of the data
unit to static discharge under conditions of low ambient humidity
emerged. Magus continued to provide excellent support in monitor repair
and problem resolution throughout the field monitoring. However, some
of the electric problems with the data unit, particularly the suscepti-
bility to static discharge, were never resolved.
As the number of monitor failures due to the bias battery system
and to general electronic defects began to decrease, failures due to the
power supply batteries and the sample pump began to increase. This
resulted in a relatively constant rate of monitor failures throughout
the project. The decrease in the reliability of the Ni-Cd main battery
packs was first noted in late December. The packs used to power both
the monitoring subsection and the data acquisition subsection began to
-154-
-------
exhibit increased resistance to accepting a full charge. Additionally,
the incidence of failure of single cells within a battery pack increas-
ed. This behavior developed dispite a regular program of completely
discharging all battery packs once every seven days. Although this
problem encompassed the batteries powering both the subsections, the
decreased battery performance seemed to be more critical with the
i
monitoring subsection..
The reliability of the GE unit sample pump also decreased during
the project. The decrease in performance was caused by several factors
including: i
(1) wear of the pump bearing surfaces; i
I
(2) loss of resilience in the pump diaphragmi? and flapper valves;
(3) mechanical failure of the pump diaphragms; and
(4) increased deposition of Purafil® (the prescrubber material)
fines in the pump chambers and passages.
The wear of the bearing surfaces, diaphragms, and valves could be
attributed to pump aging and could have been corrected by replacement of
pumps or diaphragms and valves had replacement parts been available
on-site. A question that must be addressed, however» is whether or not
such aging is to be expected after a usage periodiof approximately 900
hours. It may be prudent to seek a pump with a greater life expectancy'
for use in these monitors. ••> -
The increased deposition of Purafil® in the piump could probably be
attributed to aging of the prescrubber material aind filters. As the
active ingredient of the Purafil® (potassium permstnganate) was consumed
in the prescrubbing process, manganese dioxide was produced as a by-
product. Manganese dioxide was released as fines from the scrubber
support material by the pulsations inherent in the sample flow stream
and by the vibrations produced by physical movement of the monitor.
These fines eventually broke through the filters between the prescrubber
and the pump and were deposited in the pump chambers and passages,
thereby blocking the passages and interfering with the effective opera-
tion of the valves. One method of retarding the deposition process,
which was used successfully during field sampling,! was substitution of a
glass wool plug for the two foam rubber pads located at the downstream
-155-
-------
end of the prescrubber cartridge. However, frequent (once per week)
cleaning of the pump and filters coupled with frequent (once every two
weeks) changing of the prescrubber material proved to be the only sure
way of resolving the deposition problem.
Other, less significant, factors contributing to monitor failure
were broken solder joints, disconnection of the data unit from the *
monitoring unit, switching off of the pump by the respondent, etc. "
Broken solder joints and other connections were not considered a major *
problem in these monitors. The occurrance was rare and such a failure |
was easily repaired on-site. In at least two cases, spot welds holding
portions of the monitor's framework together broke. Again, this seemed
to be a minor occurrance, however, it was not easily remedied in the I
field. The accidental switching off of the .sample pump by the respond-
ent occurred several times near the beginning of the project. This
occurred when the respondent mistook the pump on/off button for the
activity button. The problem was solved by making the pump switch
inaccessible from outside the monitor's case. This, however, may have '
contributed to the problems with the monitoring unit's battery life.
Defeat of this switch resulted in an operating mode in which the sample
pump had to run from the time the monitor left the laboratory on the way
to the respondent until the time it was returned approximately one day-
later . Operating in this mode placed a non-sampling pump load of as
many as four hours on the monitor's batteries.
The recommendations for improving the COED-1 monitor are the
following:
(1) It is recommended that the electronic problems with the Magus
data unit be resolved before this monitor is utilized in
another study. The problems which must, be address are "lock-
up", "mode shift", and susceptatility to static discharge.
(2) It is recommended that alkaline batteries be considered as a
substitute for the nickel-cadmiun (Ni-Cd) batteries presently
powering the unit. Many of the battery-related monitor ' ' '.'
failures can be attributed to charging difficulties and other
reliability problems with the Ni-Cd batteries. The field J
staff briefly investigated the feasibility of using alkaline !
-156-
-------
batteries in these monitors during the project. Our experi-
ence indicated that six alkaline batteries would power the
data unit for up to seven 24-hour sampling periods before
replacement was necessary. Additionally, 4 alkaline batteries
adequately powered the GE-CO unit for up to four 24^-hour
periods.
(3) It is recommended that the availability of a more durable
sample pump, which is still compatible with monitor specifica-
tions, be researched. Field experience demonstrated that the
service life of the currently utilized pump may be as low as
900 hours.
(4) It is recommended that the configuration of the sample flow
path be modified such that the flow through the prescrubber is
up with respect to gravity. If this is not possible, the
scrubber should be oriented in a horizontal configuration.
This change will eliminate or, at least, minimize the deposi-
tion of prescrubber material fines in the pump. An efficient
filter between the prescrubber cartridge and the pump may
resolve the problem satisfactorily. However, such a filter
must be easily replaceable because experience has shown that
it will be contaminated quickly. '
(5) It is recommended that the unit be equipped with a sample pump
on/off switch which is inaccessible to the respondent, but
available to the interviewer. Availability of such a switch
will allow the sample pump to be turned off during periods
other than those of actual sampling. This will, in turn,
reduce the load on the monitoring unit batteries.
(6) It is recommended that the electrical connection to the sample
pump be modified to facilitate removal and replacement of the
pump. The connection is presently made Iby soldering a piece
of printed circuit tape to the pump motor| terminals. In light
of the requirement for frequent pump removal (for cleaning,
repair, or replacement) discovered during this project, the
present system is cumbersome and time-consuming.
-157-
-------
I'*-"
6.3.2.2 The GE/HP Monitor i!
During the seventh sampling wave, the field team
evaluated 10 units of a new version of the CO monitor. This new version
utilized the GE CO-3 CO monitor as the COED-1 had. However, the Magus
data unit had been replaced by an HP-41CV programmable calculator and an !'
HP-IL interface loop and converter acting as the data acquisition unit. J
Several problems, with both the HP-41C program and the monitor, ^
itself, prevented the acquisition of valid field data from these moni- >
tors. The unit was designed and assembled by Rockwell, who also design-
ed the original program for the HP-41CV calculator. This original
program, while performing the data acquisition and monitor controlling
tasks admirably, performed numerous data analysis tasks in the routine
acquisition loops. Because of these tasks, the logic in the acquisition
loop required approximately 8 seconds to execute. Since the loop was L
j"
programmed to execute every 10 seconds, only 2 seconds in ten were
available for respondent activity initiation. This length of time was •
deemed inadequate by both EPA and RTI. Therefore, the program was i
completely rewritten by EPA personnel. While the rewrite remedied the -
loop timing problem, it also deleted certain essential logic controlling
monitor power-up and power-down. Without this logic, the monitor
remained 100 percent powered-up at all times resulting in depletion of ;•:
the monitor's batteries after only 8 hours. Additionally, the rewritten
logic did not handle seldom arising situations such as a respondent :
initiating an activity exactly on the hour. The program was written for
a third time by RTI field personnel just prior to the initiation of
sampling Wave 7. The acquisition loop logic was shortened and the loop
was programmed to execute once every 15 seconds. The final configura-
tion allowed for from 8 to 12 seconds per loop for activity initiation [!».
depending on the characteristics of the individual HP microprocessors. |
The power-up/power-down logic was refined and included provisions for *
putting the monitor "to sleep" (i.e., putting the monitor in program "
controlled standby mode for battery conservation) whenever the sample *
pump was turned off. Logic was also included to shut down the entire ^
• is
data system if main battery failure occurred. This provision was
designed to protect any previously accumulated data from being lost due
-158-
-------
to the battery failure. This program performed the acquisition and
control tasks as well as the original program, while holding the acqui-
sition loop execution time to a minimum.
Unfortunately, the monitor problems which developed following the
preparation of the program prevented the acquisition of any significant
amount of ambient data. Rockwell provided EPA with a main battery
charging rig to charge the lead-acid gel cells used by these monitors.
Field personnel began charging batteries immediately after the arrival
of the monitor on-site in preparation for the initiation of sampling.
However, the charging rig was defective, producing an excessive charging
rate which destroyed the charger and partially damaged most of the
cells. The damage prevented the batteries from accepting a full charge
and, thus, substantially reduced the capacity of tihe cell. The reduced
capacity proved insufficient to operate the monitors for the required
24-28 hour sampling period. .
The battery charging problem was discussed with Rockwell personnel
who suggested several possible field repairs to keep the monitors in
operation. One repair, not available to the field team, was the replace-
ment of the damaged batteries. The battery manufacturer could not
supply replacements rapidly enough to be of any use to the project. The
EPA Project Officer was informed of the problems and repair options were
discussed. It was decided to attempt reasonable field repairs with the
objective of obtaining, at least, some performance data on the monitors.
A new battery charging rig was assembled in an effort to adequately
charge any batteries which were not damaged. Discharge curve experi-
ments were undertaken to determine the extent of battery damage. These
j
efforts did allow sampling to begin with the new monitors. However,
several new problems developed during the sampling periods which usually
resulted in the loss of most or all of the sample1 data. The problems
i
included:
(1) refusal of a monitor to "wake up" when the intereviewers
started the sample pump, |
i
(2) destruction of 1C chip U16 resulting in loss of sample pump
I
speed control, and
(3) spontaneous "going to sleep" by the monitor while the sample
pump continued to run. ;
-159-
-------
When the same monitors that had demonstrated the above described fail-
ures were operated in the laboratory on a maintenance charger, valid
24-hour samples were obtained without difficulty. These results indi-
cated that the problems were battery related rather than monitor design
or program design related. Sampling attempts continued throughout the
10 days of Wave 7 at a rate of 4-5 per day. However, only one complete,
valid sample was obtained. Three additional 1/2 - 3/4 complete samples
were obtained.
Recommendations for improving the GE/HP monitor are:
(1) It is recommended that the electronic design of this monitor
be carefully reexamined and modified as necessary to eliminate
the various logic faults experienced during this project.
(2) It is recommended that the compatibility of the .lead-acid gel
cell batteries with this unit be examined carefully. There
are indications that the batteries may not be capable of
powering the current design for the 28-32 hours generally
required.
(3) It is recommended that, after the battery capacity question is
settled, clear and complete instructions for charging said
batteries be written and that charger/charging circuits of
appropriate capacity be assembled and supplied to future
users.
(4) It is recommended that the packaging of the unit be redesign-
ed. The two-component package with its interconnecting wiring
is cumbersome and represents a reliability problem due to
broken and disconnected leads.
(5) It is recommended that, if the batteries are to be removed
from the monitor for charging, they should be made more easily
accessible and easier to remove. Additionally, the battery
connections should be polarized to prevent accidental reversal
of polarity.
(6) It is recommended that these monitors be thoroughly evaluated,
both in the laboratory and in the field, before being employed
in another sampling project.
-160-
-------
6.3.3 Quality Assurance Activities
6.3.3.1 Quality Assurance Project Plan
The Quality Assurance Project Plan was prepared
during the months of June and July, 1982. A copy of the approved plan
is contained in Appendix I.
.
6.3.3.2 External (EPA-Conducted) QA Systems Audits
In November/December 1982 and in January 1983, EPA
conducted external QA systems audits of the field laboratory. These
audits examined the opacity system and determined the performance of the
PEMs and the breath analyses. They also examined the general operation,
record keeping, data reporting, data custody, and QC activities of the
laboratory. Both audits found that the analyses were within the project-
ed ±10 percent tolerances, that the COED-1 monitors did not suffer
sensitivity decreases with time generally, and that the output from the
monitors exhibited excellent linearity from zero to full range.
6.3.3.3 Internal (RTI-Conducted) QC Audit
On January 20 an internal quality control systems
j
audit of the field laboratory was performed by the B.TI QA coordinator
for the project. The audit revealed that the field operation was being
conducted in accordance with the project QA plan and good QA practice.
.However, one deficiency was noted. Duplicates of the field sampling
data sheets were not being maintained either in the field laboratory or
at RTI. This deficiency was remedied immediately by initiating a
program of data sheet duplication in the field laboratory before the
sheets were transferred to RTI. This was deemed appropriate even though
all data sheets were being hand-carried from the laboratory to RTI.
Once the sheets reached RTI, they were duplicated ifor a second time and
the originals stored in a safe file in sealed packages. Additionally,
all data that had been transferred to RTI prior to the audit was immedi-
ately duplicated and the originals were placed in the. safe files.
6.3.3.4 Multipoint Calibrations to Assess Monitor Linearity
In early January (between Waves 3 and 4) and again
in early March (following Wave 7), all COED-1 monitors currently in
operating condition were subjected to a multipoint; calibration to assess
response linearity. Prior to delivering any monitors to RTI, EPA had
-161-
-------
completed such a calibration and reported that all monitors displayed
good linearity. These calibrations were based on monitor response to
atmospheres defining six upscale concentrations (3.5 - 200 ppm) of
carbon monoxide as well as monitor responses to a zero concentration
matrix. The data from the calibrations were reduced according to the
technique of least-squares linear regression using the atmosphere
concentrations as the independent variable and monitor responses as the
dependent variable. Within the two calibrations, a total of 78 monitors
were calibrated. The slopes of the regression lines ranged from 0.64 to
1.20 with an average of 0.98 and a standard deviation of 0.10; inter-
cepts ranged from 0.14 to 4.08 with a mean and standard deviation of
1.63 and 0.93, respectively. No zero or less-than-zero values were
expected for the intercepts of these lines due to the practice of
setting the monitor zero level at a nominal value of 1.0 ppm. This
value was chosen to avoid the likelihood of negative responses to
near-zero concentration levels due to monitor drift. This was necessary
because the Magus data acquisition system interpreted all incoming
negative data as the absolute of the data. The coefficient of determi-
nation (r2) was computed to describe the linear relationship between the
calibration atmospheres and the monitor responses. It ranged from
0.9993 to 0.9999 with an average of 0.9997. This indicated that the
linearity of response for all analyzers was well within acceptable
limits.
6.3.3.5 Monitor Stability Over the Course of the Study
In order to characterize the stability/variability
of the monitors with time in general, a series of five control charts
was maintained for each of the 49 COED-1 monitors used during the course
of the study. Another objective of the control chart series was that
they were useful as a tool for predicting when degrading monitor per-
formance would become unacceptable. An example of each of these charts
is presented in Figures 6.3.2 through 6.3.4. Figure 6.3.2 depicts the
variability in the differences between pre- and post-calibration zero
and span response over the course of the study. The event numbers on
the abscissae refers to the sequential number of times that the monitor
was selected for assignment to a sample. Event numbers with missing
-162-
-------
|.y;::..:4=2stf
'&.'.: . . "Jkx""^^
.. . . '. i-@^-—=:
Event Number
Figure 6.3.2. Response Levels
-163-
-------
Battery Voltage
COED-1 # if
Block *> Data Umt
Sed ** Monitor
15 20
Event Number
Figure 6.3.3. Monitor Battery Voltages (volts)
-164-
-------
rp — I'M; — i — , -i ,i .) M i ' i ' — ' i i j i i | i ! i i
,i j
1 * ,i i i . _
11 . _
* FFMilil 111' 1 ll \\v\\\\
1> J 1_ (_- -- -- -J- -- - -
*j j • • ,t- ...-j-i — __ _ — U -4-- -
« ' 1 _._ __ ^_ _C_ -- - -r-
OL \\\ — T —
1 f-n-ip i ii HL-4 ii 1 : :
90 l^iii i in ii| = : :: : :
80 1 II 1 [ "'-T ^--: 33-- -
70 ::i[S :it!i -::::::::::::::::::
- a>i i t i -«^--?| — 3J«'i^ •
j It Jy>y- ^-^r —
so ±~ . | j j | L :niii:ii:if?4 ii_i._
so :^::i
- M ~~l~~ III II IE I I II I I
0 5 10- 15
Ev
'-, .
i SAMPLE hLuw
— H f-- A — 1 COED-1 # 11*^^5^
11 I 1 1 [IT I i 1 M
: +:: :: i+g--..—.^^ -rfjj-Hr
~ " i" ii i mi iii-i-im imiiipi
I I II II I IIII III!-i-I"IEI IIIIII i t :
I _j_ I HI [T I | 1 | | 1 -1- '•*- - -l||| — i i i" | i i i i
20 25 30 3
snt Number
Figure 6.3.4. Flow Rate
-165-
-------
values indicate times that the monitor was assigned to a sample and was
returned to the laboratory in a nonfunctioning or malfunctioning condi-
tion. Figure 6.3.3 displays pre- and post-sample voltage levels of the
monitors main battery packs. The upper sequence of points at about
eight volts refers to the data unit battery pack; the lower sequence
refers to the CO analyzer battery pack. In point pairs joined by
vertical lines, the upper point is the pre-sample voltage; the lower
point is the post-sample voltage. Figure 6.3.4 depicts the variability
in sample flow rate. Again, in point pairs joined by a vertical line,
the upper point represents the pre-sample flow rate. Points within
double circles represent samples where the post-sample flow was the same
as the pre-sample flow. In some cases the flow rate actually was higher
after the sample than before it, although no such cases are depicted
here.
6.3.3.6 Assessment of Measurement Precision and Accuracy &
During the field monitoring phase of the project,
certain procedures were undertaken to assess the precision and, where
possible, the accuracy of the measurement process for both breath and
ambient CO levels.
Precision of PEM Values. The assessment of the precision of the
ambient measurement was performed by having a member of the project
field staff carry two or more randomly assigned COED-1 monitors for a
24-hour period. The staff member was instructed to carry the monitors
with him, wherever he went, throughout his daily activities. Thus, the
monitors were exposed to typical sampling conditions of changing tem-
perature, humidity, elevation, etc. as well as to the vibrations and
physical shocks inherent in transporting the instruments. Each instru-
ment was calibrated before and after the sampling period just as it
would have been had it been assigned to a regular sample. The field
staff member made no attempt to define activities during the sample, nor
did he keep an activity diary. If he did activate the activity button,
he was instructed to do so on a random basis and to activate the buttons
for all monitors assigned to him simultaneously. Twenty-eight 24-hour
samples were obtained. After the sampling period, the acquired data
were off-loaded from each monitor separately following the same proce-
-166-
-------
dures used for routine samples. The data were transferred to the study
data base, validated, and analyzed for variations among replicate
samples. The description and results of this analysis are presented in
Section 6.4.4 of this report.
Accuracy of PEM Values. The accuracy of the ambient measurement
was evaluated in two ways during the project — (1) EPA-conducted,
independent performance audits and (2) colocation of COED-1 monitors
with EPA-designated reference analyzers operated by the District of
Columbia air pollution agency. As has been previously mentioned, two,
independent, EPA audits found that the ambient analyses were within the
projected ±10 percent tolerances. Accuracy was also assessed by coloca-
tion of COED-1 monitors with District of Columbia fixed site monitors.
Eleven, 22-hour samples were acquired in this way from two different DC
fixed site stations ~ the West End Library site (SAROAD NO 090020017101)
and the C&P Telephone Building site (SAROAD NO 090020023102).
j«>
The COED-1 monitors which were randomly assigned to these colocated
samples, were calibrated, maintained, and utilized | as they would have
been for routine samples. At the conclusion of a sampling period, the
acquired data were off-loaded from the colocated monitors as with
regular, ambient samples. The COED-1 data were compared with the fixed
site data by obtaining the difference produced whan the fixed site
average for a particular hour at a particular site jwas subtracted from
I
the COED-1 datum for the same hour and site. The mean of the 242
differences so obtained was -0.515 ± 0.0557 ppm a:t the 95 percent
confidence level (i.e., the fixed site was, on the average, larger).
The differences ranged from -1.72 to +0.33 ppm; concentrations observed
at the fixed sites during the comparison ranged from 0.10 to 8.60 ppm.
The distribution of the differences obtained during the comparison is
presented in Figure 6.3,5. The figure clearly indicates that in general
the diff(PEM-FSM) was negative. I
The hourly averages determined by the COED-1 monitors were regress-
ed against the corresponding averages obtained frpm the fixed site
monitors. The regression analysis yielded a slope, intercept, and
coefficient of determination of 0.947 ± 0.034 ppm, -0.411 ± 0.084 ppm,
and 0.927, respectively. A plot of the COED-1 averages versus the fixed
-167-
-------
a>
u
o
o
01 C
M O
S CO
01 1-1
a 1-1
a
K (X
w m
8s
O O
P-i U
•H
• B
" 4J
C -H
O CO
4J
00 13
C V
•H X
.C -H
(0 t^
c-.
VO
01
M
(t)
£
O
13
M
a)
g-
o
•JJ
* 0)
6 o
CJ O
PU
U
0)
M
0)
P-.
II
S
J-l
<"*>
S
co
F5*
1
P-I
**^
„
S
04
PH
a
•H
S
CO
f»
S
c
I-*
M-i
££!
PS
P-I
o
0
0
Qifferen
S*,
o t*
0,1-1
•a Q
•H
P~ 0 0
O CM CM
CM vO VO
r** en O
C r-< 0
CM -* O
1 m , m m
_. -<
m 0 0
*
*
•X
•B
•X
*
•X
*
*
•X -X
•X -K
•X *
•X -X
•X -X
•B *
* -X
•K *
•X -X
•X *
o o o
o o o
o o o
VO U"» ' •*
1 I 1
CM cn
P 0
1*^ Cn
en r*^
00 O
O CM
l-» CM
f-H CM
CM in
•X
•X
•K
1 **&*
I*******
o o
0 0
o c
en CM
1 1
en
en
o
*— 4
-S'
CM
f-H
m
CM
en
! ******
o
o
0
1
VO
-3-
vr
f— 1
en
f— i
•*
in
CO
o
f— t
•K
*
•X
*
•K
•X
•X
•X
•X
•f.
•X
•X
V
1 *******
o
c
o •
o
f-H
CM
vr
cn
vo
<•
-3-
CM
f— H
•X
•X
•X
•x
•X
•X
•X
-X
•X
•X
*
-X
•X
•X
•K
•K
| *******
O
0
o
a\
o
16i
VO
co
VO
CM
-3-
r—
m
vc
oo
f-H
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
*
*
•X
•X
•X
•X
•X
•X
*
•X
•X
•X
-X
-X
•X
•X
•is
J *******
o
0
0
oo
c
1
00
vO
en
CM
en
co
-3-
CM
•X
•X
•X
•X
•X
•X
*
•X
•X
•K
•X
•X
•X
•X
^c
-'<
•X
•X
•X
-X
^x
•X
•X
*
•K
•K
•X
*
•X
•X
*
•X
•X
•X
*
•H
•X
•X
•X
•X
•X
1 *******
o
0
.0
r —
o
1
vO
VO
^*'
CM
C*
en
c-H
•a-
CM
•X
*
•X
-X
•X
*
•X
•X
•X
•X
•X
•X
•X
•X
*
^f
•X
•X
•X
ie
•X
is
•X
•X
•K
•X
•X
•X
•X
•X
•X
*
•X
•X
•X
•X
•X
•X
-X
•X
•X
1 *******
c
c
o
vO
o
1
en
m
m
co
vo
00
en
f-H
CM
•X
•X
•X
•X
•X
•X
•X
•X
-X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•K
•X
•K
*
•K
-X
•X
-X
•X
•X
•X
•X
^<
-X
•X
•X
•X
•X
•X
•X
•X
•X
•X
o
o
Q ,
m
o
1
o
-31
CM
VO
Cvl
o
r-
in
r-
»— t
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
•X
*;
•X
-X
-A
•X
•X
-X
•X
!*******
€
Q
p
en
CM
p~"
en
en
0
vo
in
CM
•X
•X
•X
•X
•X
•X
•X
•X
•X
*
X
•X
-f
•X
-<
*
•X
•X
•X
•X
*
•X
•X
•X
•X
*
•X
•X
•X
•X
•X
•X
•X
•X
•X
-X
•X
-X
•X
•ic '
•X
!*******
o
0
o
en
0
1
CM
co
f-H
cc
CJ\
o
&
00
CM
CM
•X
•X
-X
•X
•X
•X
•X
S
•X
•X
•X
•X
-X
*
•X
•X
•X
•K
•X
•X
•X
•K
•X
•K
-X
•X
•K
•X
41
•X
•X
•X
•X
4:
!*******
o
o
o
CM
O
1
* r "
o ob en en o i '
vo o vo cr> o >• "
r-~ o
-------
site averages is presented in Figure 6.3.6. The comparison data were
examined by site to determine if the observed differences were site-
dependent. No significant difference between the two sites was noted
(Library site mean: -0.528 ± 0.058 ppm; C&P site m€>.an:: -0.499 ± 0.101
ppm; £(240) = 0.50, not significant). Plots and frequency distributions
of the comparison data by site are presented in Appendix M.
Examination of Figure 6.3.6 revealed that a substantial number of
zero responses were obtained from the COED-1 units when fixed-site-
determined concentrations ranged from 0.1 to 1.4 ppm. Further exami-
nation of the comparison data on a sample-by-sample» COED-1 monitor-by-
monitor basis indicated that the minimum sensitivity of the COED-1
monitors varied on a monitor-by-monitor basis and ranged from approxi-
mately 0.2 ppm to approximately 2.0 ppm. Figures 6.3.7 and 6.3.8
present examples of this variability in minimum sensitivity for two
representative COED-1 monitors. Plots of COED-1-determined concentra-
tion versus fixed site monitor-determined concentration for the remain-
ing nine samples are presented in Appendix M.
The comparison data were examined at or above fixed site/PEM
concentrations of 1.0 ppm to determine whether or not the poor sensi-
tivity near zero concentration had unduly influenced the mean difference
between the COED-1 and fixed site monitors. No significant difference
was noted (mean difference for data above 1.0 ppm: -0,472 ± 0.091 ppm).
A plot and a frequency distribution of the data above 1.0 ppm are
presented in Appendix M. !
The relationship between the PEM-determined concentrations at or
above 1.0 ppm and fixed site-determined concentrations at or above 1.0
ppm was examined another way utilizing the statistic:
( (PEM-FSM) / FSM ) * 100, |
I
where: PEM = the PEM-determined concentration (ppm), and
FSM = the fixed site monitor-determined concentration (ppm) .
This statistic produces a concentration-normalized difference expressed
i
as a percentage of the, fixed site concentration at which the difference
was observed. The mean normalized difference so obtained was -15.0 ±
2.76 percent at the 95 percent confidence level. The interval into
which the "next observed value" of this statistic ±s expected to fall 95
-169-
-------
u
u
Ift
C2
d
c a.
« «
£. 0.
X
u o
ee. u
V)
U
a
Vtf.
ox
u »
xu. -•
o ce
oo
a
z
UJ
o
S~ o
X
v>
Xu
£«
(0
u
o a
>- o
o « z
s > o
v>u u
» a. a
o
_!
C.
U
3
•- s a. a. :
-170-
-------
>-» ^ a. G.
•171-
-------
u
»—
u
04
II
a.
00
bj I/}
O OS
B «t
0.0.
uo
§" j;
Mo 03
P SC O
XU. •-"
i.i ^,1
V 1C
u o m •<
a
CCU M
Uf- O.
2 •-> U.
I—
-------
percent of the time (i.e., the 95 percent prediction limit) is -15.0 ±
30.2 percent. A frequency distribution for this statistic over this
comparison is presented in Figure 6.3.9.
Three conclusions can be drawn from the analysis of the comparison
data from COED-1 and fixed site monitors. First, there appears to be a
consistent -0.5 ppm bias in the COED-1 data with respect to the fixed
site data. This bias is neither site-dependent nor absolute concentra-
tion dependent within the constraints of this analysis. Second, this
analysis indicates that PEM-determined concentrations will be within ±30
percent of the fixed site monitor-determined concentrations 95 percent '
I
of the time, once any consistant bias between the methods (-15 percent
in this analysis) is taken into consideration. Finally, the COED-1
monitors appear to exhibit varying minimum detectable sensitivities that
vary from monitor-to-monitor and range from 0.2 to 2.0 ppm.
Precision of Breath Values. The data derived to assess the preci-
sion of the respondent breath analyses were based on duplicate samples
obtained from respondents by having them inflate two separate breath
sample bags. The mean difference between duplicates analyzed during the
project was 0.11 ± 0.13 ppm at the 95 percent confidence level. Results
of paired difference tests performed on the duplicate analysis data at
the 95 percent confidence level indicated that the|mean difference was
i
not statistically different from zero. Additional statistical tests
performed on the results of analysis of laboratory and field blank
breath samples indicated that the blanks were not significantly diffe-
rent from zero nor were they significantly different from each other.
Results of analyses of control samples at levels of 3.5, 10, and 40 ppm
indicated that the controls generally did not vary; from their nominal
value a statistically significant amount. When the variation became
statistically significant, it was due to small standard deviations and
was not of practical significance. Additionally, Ifield and laboratory
control analyses were not significantly different from each other.
Accuracy of Breath Values. A straightforward assessment of the
accuracy of the breath analyses procedure was not possible since there
was no "reference value" with which to compare them. However, results
of EPA-conducted audits of the breath analysis procedure indicated
-173-
-------
3 O
3 M
O 0)
PM
CO
O
CO
O
CM
m
VO
m
rH co m CM vo
co in CM o CM
r^
co
COCOCMvOvOvO
co o
CM co
co co
-*
co
o
o
CO
VO
CM CM
in m
o
CM
CM
r-
CMvOCMI~~CMCOOCM O CO
O O rH CM CM
m vo vo
i Si
3 M
) to I
^H CO \O
O CO VO
CM CM CO
si
pfj I
rHOCMCOCOinvOCOCO
o
x
(A
to,
M
0
W
•H
0
O
S
at
4J
•H
M
•O
§
to
0)
M
O
1
O
cj
rH
td
a
0
as*
0) PM
PM
O
* *
O -H
•
O X
•> A
0
o td
1 1 1 1
60 Id
0 O
to o
td iH
r% *rJ
O>
•
4J
H
id
^5
M
cd
M 0
o
^ *rl
O 4J
C cd
0) M
3 *J
O4 0
(U
o
-------
attainment of an accuracy level within ±10 percentiof nominal. Addi-
tionally, since relatively extensive ruggedness testing was performed on
the method during its development phase and since that testing revealed
no significant interference under normally encountered sampling condi-
tions, a measure of accuracy may be inferred from the analyses of
control and blank samples. After allowing time for conditioning of new
sampling bags, the ability of the method to correctly analyze standard
atmospheres varied from ±0.3 ppm at 3.6 ppm to ±1,0 at 40 ppm. The
method analyzed zero-level samples to within ±0.1 ppm. During the
conditioning of the bags, the variation in the analysis of standards was
somewhat greater. j
6.4 Results of Statistical Analysis
Using the computer data files described in Section 5.4, a detailed
analysis of the data was undertaken. Results of 'this analysis are
presented here. The population of inference - adult non-smokers in the
B.C. area - is estimated to include about 1.22 million, individuals.
Unless specifically noted otherwise, all results shown in this section
apply to this population or to some specified subgroup of this popula-
tion. The results also apply only to the winter of 1982-83, as this was
the period of data collection.
Subgroups of the population deemed to be of particular relevance
that are used in the analyses of subsections 6.4.1, 6.4.2, and 6.4.3
were described in Section 5.4. These relate to three potential sources
of carbon monoxide exposure - occupational exposure, exposure through
travel/commuting, and exposure through gas cooking. Some of the results
in this section are also shown by type of day — days of high potential
CO exposure versus other days; and weekdays versusiweekend days (actual-
ly , Friday evening through Sunday evening). !
Subsections 6.4.1 and 6.4.2 present the results of statistical
analyses of the BAF file. Subsection 6.4.1 provides results from
analyses of the exposure levels from PEMs as indicated by hourly CO
concentrations, by mean and maximum hourly CO concentrations, arid by
maximum 8-hour concentrations. Subsection 6.4.2 presents the results of
the analysis of breath CO levels. Subsection 6.4.3 provides the analyt-
ical results relating to individuals' activities land environments and
-175-
-------
their associated CO exposures. Finally, subsection 6.4.4 investigates
the measurement variability of hourly CO exposures through analyses of
the duplicate-sample CO data.
Except for the duplicate sample analyses, all of the estimates are
representative of the population (or subpopulations) of adult non-smok-
ers in the Washington area, since weighted analyses were performed.
Standard errors of estimates were produced by using SESUDAAN, a SAS
(Statistical Analysis System) procedure developed by RTI for analysis of
data from complex sample surveys (see Shah, [1981]).
It should be recognized that time periods over which data were
available for different sample individuals are different in two re-
spects:
(1) Different days are involved, since only a single sampling
"day" was utilized for each person. (From a purely statist-
ical standpoint, a much longer sampling period , e.g., several
weeks, would have been used if data throughout such an extend-
ed period could have been anticipated, and if a much larger
number of PEMs had been available for the study).
(2) Some variations in the starting times and lengths of sampling
periods occur, due partly to variations in the times at which
interviewer/respondent contacts could be made, and due partly
to monitor malfunctions.
A principal impact of the first point is that estimates for subsets of
days are unreliable unless a large number of days are involved. Hence,
the results shown in this section apply to an "average" winter "day" in
1982-83 (or to an "average day" within some specified subset of days
involving a large number of samples — e.g., weekdays or weekend days).
No attempt was made to adjust for the unequal sampling weights and
sample sizes occurring for different types of days (e.g., in order to
achieve equal representation of the days of the week or the relative
representation of the number of days within months — see Table 5.4.3).
This approach was taken not only because of the types of estimated
parameters deemed to be of most importance, but also because of the
problems that would be encountered in making such adjustments, given
point (2) above. That is, with only 18 to 26 hours of monitoring/activ-
-176-
-------
ity information available for individuals, over varying sampling periods,
it was not really feasible to construct such day-specific variables.
The activity "sleep", for .instance, overlaps types of days for most
individuals. j
The net result of the above problems/decisions; is that analytical
results labeled and described in terms of a specific subset of days
! • . • -
(e.g., "weekend days") are actually estimated from the data of indivi-
duals who were sampled on specific days. For example, "estimates for
weekend days" is more precisely stated as "estimates for persons whose
monitoring periods began on Friday or Saturday evenings, assuming that a
census of the inference population had been conducted in the same manner
as actually employed (on a sample basis) in this study". True coverage
of both the person and time dimensions of the inference population was
not practical because of the small number of available monitors per day;
the reliability of the monitors; and the degree of nonresponse experi-
enced due to individuals' unavilability for, or unwillingness to partic-
ipate in, the various phases of the study. Becausfe of the emphasis of
the study — namely, information on personal activities, environments,
and associated CO exposure levels — the person ddmension was given
priority over the time dimension in the study design and during the
development of nonresponse adjustment strategies.
With regard to point (2) above — that durations and starting times
of individuals' sampling periods vary — several potential impacts
should be noted. First, in conjunction with the time-inference issues
'- ' l
and concomitant nonresponse adjustment decisions described in the
previous paragraph, it is clear that certain biases in the estimates may
be present relative to the time dimension. (Potential biases in the
person dimension can also occur whenever nonresponse is present. This
was addressed in Sections 5.1 and 6.1). For example, in estimating a
diurnal pattern for "weekend days", the timing of tihe sampling intervals
(see Table 5.4.1) suggests that Friday hours will lie ssomewhat overrepre-
sented relative to Sunday hours for some hours of the day (and vice-versa
for certain other hours of the day). j
Secondly, it is apparent that similar types of biases may occur
with regard to estimates of time durations and of CO exposures of
-177-
-------
particular activities and environments. In terms of the available data,
for instance, certain activities may have been curtailed when an inter-
viewer arrived at the respondent's home to either start or conclude the
data collection. Hence, estimates of time durations of activities
frequently occurring at times near the termination (or initiation) of
the sampling period (usually early evening) may be underestimated. To
the extent that the study induced respondents to be at home at the
prescribed starting or ending time of the sampling period, such esti-
mates may also suffer from so-called Hawthorne effects.
The above-described limitations (related to the time dimension of
the inferences) need to be considered carefully when interpreting the
results presented later in this section. However, as previously indi-
cated, the emphasis of the study relates to the person dimension, so
that they should not be considered as severe limitations on the study
nor as severe reservations concerning the basic inferences and conclu-
sions of the study. It should be emphasized, for instance, that many of
the potential biases described above are indirectly reflected (e.g., in
standard error estimates) as a part of the person dimension, in that the
sampling error includes day-to-day variation as well as person-to-person
variation.
6.4.1 Analysis of Hourly CO Exposure Data
Diurnal Patterns. Table 6.4.1 shows the mean diurnal
patterns of CO exposures estimated for the Washington area population of
inference. These (weighted) estimates, and their approximate standard
errors, are given for all days, for weekdays and weekend days, and for
days of low and high CO levels (as indicated by fixed site monitors).
The pattern for weekdays (and, hence, for all days) exhibits the
well-known effect of commuting traffic, with dramatic increases in CO
levels between 7-9 a.m. and 4-7 p.m. 'The lowest levels occur between
4-6 a.m. For weekend days, the increase in CO level during the morning
is much less pronounced than that for weekdays. The p.m. peak is of
about the same magnitude, however. The late night and early morning
hours also appear to have higher CO levels on weekend days, although
this difference cannot be declared as statistically significant. Figure
6.4.1 shows the estimated mean diurnal exposure patterns for weekdays
and for all days.
-178-
-------
CO
cfl
a
« -ig
i
P-i CO
o
CO <4-4
>
O cfl
US G
I—(<—lr-4OOO>—ICNCSCN»—1 i—I i—IrHr-lr-ICMCMCv|>—<•—I
ervoor— voioc^cNoo
OOOOOOCNIi— l
CMCNCNCN
OOO>— 'OOOOO'— '•— 'Oi— ii— i
i—JCvJCNi—ii—ir-i
Or- it— <^-i 1-4 »— I£N1>— it— 10
i—it—HOOOO-—'CMCNr—ti—l^-lt-Hr—Ii—Ii—. d
CO O
T3 ^3
CO
•rl CO
4J T3
CO
^ J=
O
M-4 <^x
CO
"O •
cu ro
C •
•H \0
iw
0) CU
T3 rH
>> CO
s-i cu
CO CU
4-> CO
CO
CU
CU O
& •
co0
f^ •>
co p:
T3 O
4-1
o oo
&0 co
•H CO
-179-
-------
Figure 6.4.1 Average CO Exposure .Levels, By Hour of Day
Avg.
CO
Level
(ppm)
3.0 -
2.8 -
2.6 -
2.4-
2.2 -
2.0
1.8 -
1.6 -
1.4 -
1.2
1.0
.8 -4
Avg. Day
Avg. Weekday
jilt:...
.6 -
^-*
I 1 i 1 I
0 2 4 6 8 10
Hour of
1
'12
1
14
Day (ending
1
16
hour)
till
18 20 ^22 24
-180-
-------
The days designated as high CO days generally exhibited higher PEM
exposures, as 20 of the 24 hourly values were highesr. The difference
between the hourly exposures on high and low CO days was greatest
v I .
between 8 and 10 a.m. and between 9 p.m. and 1 a.m. j
Maximum Hourly Concentrations. The maximum q'f the hourly CO
~" ..... ' I
concentrations for each individual (over 18 to 26 hourly values) was
I
determined for the 712 respondents. The analysis o,f this variable is
shown in Table 6.4.2. For each of several subgroups, this table shows
the sample size, the estimated number of individuals in the population
. i
subgroup, the estimated percentage of the total population represented
by the subgroup, the average maximum hourly CO concentration estimated
for the subgroup, and the approximate standard error of this estimated
.
mean. The table also characterizes the population distribution of the
maximum hourly CO values by providing estimates of the percentage of the
population (subgroup) having maximum hourly CO values that exceed
certain specified levels (1 ppm, 2 ppm, 4 ppm, 9 ppm, 25 ppm, and 35
For the overall population, the mean of the maximum hourly values
was estimated to be 6.74 ppm (during the winter 1982-83 data collection
period). As indicated in Table 6.4.2, this mean level varied by type of
day, with higher maximum hourly CO values, on average, occurring on
weekdays (7.35 ppm). The mean level also varied, in accordance with
a priori expectations, depending on individuals' occupational and
traveling characteristics. Persons working .outside the home, especially
those in occupations with potentially high exposures, exhibited higher
maximum hourly CO concentrations, on average, than those persons not
working outside the home. In fact, in the high-exposure occupational
category (an estimated 4.63% of the overall population) , about 24%
exhibited a level above the one hour standard for carbon monoxide (35
ppm), and over half of this subgroup was estimated to have hourly CO
exposures over 9 ppm. Figure 6.4.2 illustrates the differences in
maximum hourly CO levels for the 3 occupational groups examined and
Figure 6.4.3 illustrates the differences for weekdays and weekend days.
As shown in Figure 6.4.4, commuters, especially those with longer
traveling times, also showed higher CO exposures (based on their one-
-181-
-------
13
4-1
< CO <
4 •
03 > O
63 •< CJ
-U O
EO
CU M
S "*
CU IN
in
cn
«
X
CN
&
cn
X
sr
A
CM
X
1
X
si
° >s
gl
u
e 3
5 &
£,2
•s .
to 3
0 B.
B. 0
X CU
f 1
a
§
CC
A
3
1C
E
0
T!
U
B
1-1
3
c.
a
CM
CO
CM
—
s
I-*
cn
CM
CO
VO
r—
vO
cn
cn
CM
CC
c\
CM
cn
*
0
s-
VO*
8
|
3
in
in
CM
2
r.
a
a
B
r-t
f-4
CO
B
O
tc
U
01
CH
r-1
r-1
^
vo in
sr vo
-» O
cn cn
— 0
CM n
CM fs.
sr o*
1— C
vo en
VO CM
CO CM
in r-
cn — <
cn cn
— CO
CO CO
r» vo
st vo
CM sr
cn en
o in
VO CM
0 0
— < CM
r- CO
VO VO
§1
22
o^ a\
cn m
in in
in m
CM CM
C\ CM
•a- CM
CM|
•K
-X *
•K
00 ?•*
a c
is
J i
to to
B B
C O
to co
U u
Gi 01
CM C-
r-l i-<
ic; **C
in o in
en •-* cn
O O cn
CM
in CM —
cn r* »~*
O O co
CM
in vo en
vo co O
co cn r-
VB O vo
C cn co
sr co in
vo tn cn
co cn r-.
sr vo sr
vc sr cn
r*- co cn
sr r» o
CM «• • O
cn sr o
co cn o
vo n sr
en cn vo
O O in
CM sr -H
CM en — *
in vo CM
CM
~* sr cn
co in vo
cn in sr
CM vo
vo o sr
cn r* CM
CM VO VO
vo cn in
cn r-.
-i cn r-
CO O CM
-- in
^•t
01 U
M 3
3 CO
CU CO O
BOB.
O CL >C
== X K
bl
(U 1-4.
•O rH B
•rl CO B
to B 0
•" C -H
3 T4 4J
O U 13
e e-
cfl a. 3
B S U
•rl U O
_V* {J Q
M O
O ^
& Z ta
O -H
4J j as
33 & »S
to co co
EBB
O O O
CO CO CO
^ 1-, 1-
0) Cl O
Cu CU d.
CM O
CO 0
— i Q
en r-
CM O
e/> sr
-1 O\
-H O
en *a"
CM O •
VO vo
CO m
co en
sr co
CO p»
CM cn
t— CM
cn —•
r~. en
O 0
in c
cn cn
r*- tn
0 O
0 O
o o
0 0
O% Ov
cn cn
in in
in in
CM CM
•—i *••(
in ve
in «-i
He
to
* >*
•K a
to0
S"E
3J
01 01
££
C C
O O
to u
U 4)
^-t t— H
r* CM m
tn CM o \o cn oo
r«- VD CM oo ** o
en -H
»n o co o —* ^o
m «3" -fl- ^a* CM m
*•* CM r*. en co CM
r-. c\ tn vo ~* in
wo ON — < tn tn CM
^ CN ^r — —
in O cn -a- en r-
O oo cn o P*. -*
•"4 i-* CM rH
/_N
^•^ B^
iB. 3 •*-» M 3
*g m f-j j- tn
1- J= in J=
11) O —*
cu m —* i +
1 1 — * vO (fl
4) *J
iH O 0) 0) 0) *J
V^*« £-« H &^ ^
i-t t-i iH f-l *J
IB r-1 f-i i-I f-i 3
* i S- i s i
j-t o
d co w co to o
g p «Vi VJ U
E. C OJ CJ QJ C
cS s a "S S o
1 g E E E C
o o o c c c
5= O U CJ CJ p
O in
m «s
o o
S§
O CM
-1 CO
vo in
en — i
en cn
r- cn
sr vo
vo vo
ri vO
sr CM
a\ cn
CM cn
m CM
sr co
cn cn
in en
co r*-
0 O
o m
sr o
vo r-
in r-
o sr
cn CM
cn vo
0 CM
cn cn
t-. vo
CO f»
tu
c
TJ 01
01 S*
C 4J
O O
> z
1 1
1 1
sldence
sidence
0) Cl
ei K
u u
co eg
tu a
go"
J-i 4J
co cn
CC CO
cc c
u o
32
CM en
a> co
sr —•
CM en
r- t-.
CM CM
a o
o cn
O VO
VO CM
in t-.
-H sr
co cn
-» cn
r- 03
tn vo
in co
co sr
co cn
vo cn
o —
CM f.
co sr
VO vo
CO CO
sr CM
— • cn
sr
CM r-
CM CO
sr m
o cn
in
in sr
CM CM
cn
£J
§
CU 3.
01
•a o
•H >
to o
it u
f& CO
CO CO
u
0)
> VM
o o
4J
CC 01
« g.-
a
Ctl
3
•a1
a
ai
01
.c
counts
B
0
•rl
4J
CO
f^
3
O
CU
Jg
•X
a
co
,
CJ
c
c
o
00
c
•H
to
0)
M«
C
•H
CO
O
1
ca
***
CO
u
Cl
f:
u
o
CO
B
01
.c
• to
01 »> •
1- C5 E
3 *a B.
to B C.
B. CO O
01 4J "
U B
r-i O CO
C 01 4-
C H
•H 11
U CO CJ
C3 -H 4J
C.£ CO
3 4J (U
U H
CJ r. CO
0 C
OO-O O
•n V f
f B
VM «-| c
O O O
CO "° i-
s ^ cc
C rH C3
•rt T-t 01
W P v-i
•rl CC
B U W
•rl 4J CO
VM -rl
01 A -rj
o a ^
m r< cn
01 •
t*- 3 cn
K B vo"
•a co at
B "O r-1
o *a
B*O CO
Q.U H
pB 0)
01 CO 01
01 -rl tu
CO S — '
^^_ ^^
•-i jes(
lit
til.
-182-
-------
I
(U
13
•H
cn
4-J
O
oo
(3
•H
O
4J
O
to
g
cn
0) CD
< Fd
§
•H
4-J
n)
&
cj
O
O
g
cn
g
3
CO
O
g
•H
4-1
Cfl
a
o
O
o
to
g
W CO
p (J
0)
P-i
i
i
I
9
i
o
o
o
00
o
o
CN
O
6-S
-183-
-------
CJ
•
a
1
u
S3
CO
£
en
I
16
w
c
SB
<;
en
1
M
Ul
»
g
(^
CO
O
i
H
z:
w
o
2
8
8
01
ST
•
\O
W
O
w
h
(0
O
-------
a)
o.
03
I
•H,
^
0)
0)
j-i
a)
o.
3
o
0)
OJ
tn
3
O
O
i-H +
I ^^
\O T-l
CQ «J
en s-J
01 H
41
J-l
3
en • CD
4) 01
4-1 4-1
3 3
O
U
w
PS
•a
M
r~n—r
T T
o
o
a
CT\
o
00
o
O
UO
o
CM
o
B~S
-185-
-------
hour maximum values) than non-commuters. Those commuters indicating
total travel of more than 15 hours per week, for instance, had a mean '
estimated maximum hourly concentration of 12.01 ppm, as compared to 4.94 f
ppm for non-commuters. Persons having unvented gas stoves also showed
slightly higher means than those persons not having gas stoves (an
average maximum hourly concentration of 7.05 ppm versus 6.82 ppm).
Maximum 8-Hour Concentrations. Table 6.4.3 shows the results of
the analysis of this exposure measure. For the overall population, the «
I
estimated mean 8-hour CO concentration was 2.79 ppm; it was estimated i
that about 4% of the overall population had levels exceeding 9 ppm, the
8-hour standard. These were primarily persons in high-exposure occupa- i
I
tions and/or persons with extensive amounts of motor vehicle travel.
The maximum 8-hour mean for the group with high occupational exposures
was 7.51 ppm; for commuters with 16 or more hours of total travel per
week, the mean was estimated to be 3.80 ppm. Persons with gas stoves in
"
their homes also appeared to have higher 8-hour levels, on average, than
persons without such stoves. In general, the only subgroup with a •
! I*. •
relatively large percentage over the 8-hour standard was, the high L
occupational exposure group (28.1% over 9 ppm). Figures 6.4.5, 6.4.6, i
and 6.4.7 present the percent of the populations with maximum 8-hour CO ;
levels greater than specified levels for various groups.
Tests Between Exposure Groups. Approximate pairwise tests of
—
significance were conducted to determine if the various population
subgroups were different in regard to their average maximum hourly and
average maximum 8-hour CO exposures. The results, shown below, indicate
no significant differences between low and high CO days or between the |
various categories of stove type. The non-significance between high and !
low CO days, as arbitrarily defined in this report (see Table 6.4.2), is |
i •
not surprising considering that the winter of 1982-83 in Washington,
D.C. was very warm and had only six days where any of the fixed stations ~
had a maximum 8-hour average greater than the 9 ppm CO standard.
Significant results are indicated for several other groups: persons not
working outside the home versus those that do; persons with low occupa-
tional exposure versus those with high occupational exposure; commuters
versus non-commuters; and commuters with less than 6 hours of total
-186-
-------
|
id
a
Q
g
1-4
to
U
m
OJ
u
c
c
u
g
w
£
s
s
J
CO
E
X
1
Summary of
CO
SO
cu
ft
a
e
o
•H
4J
CO
*3 X
O PH
0.
„*
0) A
(0
O £
0. C
q
U^ ^d
0 4J
C
u c
c u
tu c
c c
M U
01
0. O
CJ
•o
IS 2
•H
to
w
X 0
O
< to ,
^ <0
„«
!|
J S
a to
c c
o o
eo to
a) o
rH .-1
r-4 *-4
O 00 00
O — 03
00 —
o . en
so en r-
m in en
in os r*.
in os -31
•31 00 Os
1— r» Os
co m oo
o o o
in so —
en so in
CM CM r*»
— »4
in so
in —
"
•K
CO
* ^
* a
O
CO
>s-0
a e
•a o)
•s-s
gg
CO CO
§ §
CO CO
U H
a tu
r-l i-i
i-i t-4
000000
O O M? O G O
O O O O O O
c o — « m O o
c o «-• en o
C *•* *3" M3 CO O
o c*a o CM m -^
m m en m r^ \o
CM 1** f""» CO *tf O
•** «-* CM CM »-l
CO CM vO \D ~* 00
in m »n in \o r**
os os en oo m m
tn co CM en en vo
sssssss;
f^. co r- CM m m
o o o o o o
o m os CM o m
CM CM CM en eo CM
r^ so CM ao ^ O
en •* «3- CM CM CM
in O co O <-4 so
in *3" *sr ^T CM en
— CM r- en eo CM
r» en in so — • in
so os — m in CM .
in O en J? 1» *-s
cu 3 — co !5
o **» to w ^^
^ CO M JZ CO
' M •£ •" in £
eu o — 4
p* in'— 1 -*•
1 l — so eo
CO O so — — 3
tU 4-*
g _| ,-< —I r-l C9
en b M _u M no
s 3.3.33 ^
3 CO CO CO CO CJ
g 1- U l_ U
0 cj eu cu cu B
O 4J AJ 4J 4J 2
CJ 3 3 3 g 0
t E 5 E € B
C g S E g Ji
C 0 O C O B
Z.CJ CJ CJ CJ 3
o — o e;
m CM o e
o o o c
o os in t
in en •s' e
O O O <
in GTS oo e
Os in in •
—4 en in c
O CM CO i
so in -ir t
in en -^ i
i— i •-> —
os oo en »
r- m co
SO so -3" '
O oo eo
— r^ CM
00 O *3"
O r- CM
0 O 0
en co in
os o m
CM CO CM
O •» CO
en r*. •«•
O <• —
CO CM **
370703
300726
504262
r^ so in
CO r* CM
— *-* eo
•O
eu
S
•a eu
0 >
B 4-4
eu c
1 1 a
lesidence -
iesidence -
11- Roatilpnr
**« ^-« M
jj jj a
a a ;
c
eu a) *
5> > a
0 0
4-> JJ C
CO CO t
c.
to co
ffl C t
CJ) O 2
g
2
§
>
O
O
T
3
—
0
;
X)
*fc.
o
»
•o
o
2
CM
00
CM
en
-3-
00
en
eo
«a~
CM
1
S
0)
g
4J
CO
m
cS
14-1
O
O
u
1 B
i d>
tu
) h
1 0-
o
^
a
1
a.
rH
a
14
o
o
eu
Si
u
o
tu
to
o
•£>
O
w
e
o
M-l
e
o
o
4J
•e
o
CO
3
•o
td
c
tu
tu
0)
n
)n counts h
•H
4J
, «
3
a
o
•K
jxposure.
rM
«
e
c
•H
u
o
o-
3
o
o
o
•H
J=
0
to
c
o
4J
•H
c
•rt
0)
O
M
*H
T3
(—
S
a.
ex
cu
eu
CO
-187-
-------
!-fc
o
54
-188-
-------
cd
O
co
>. -a
to a
T3 (U
0)
QJ
(U
0)
CO
a
o
CO
M
0)
Pu
CO
a
o
CO
i 1
o
CX3
o
o
CO
U-l
O
-189-
-------
0)
I
-------
travel per week versus commuters with 6 or more hours per week. Week-
days versus weekend days are significantly different only for the one
i
hour maximum. The tests are only approximate; the asterisks indicate
statistical significance at the 5% level and no asterisk indicates the
test was not significant at the .05 level. !
Population Subgroup
Low CO days
High CO days
Weekdays
Weekend days
Gas stove at residence-vented
Gas stove at residence-unvented
Gas stove at residence-unvented
No gas stove at residence
Persons not working outside home
Persons working outside home
Persons with low occupational exposure
Persons with high occupational exposure
Non-commuters (less than 3 times /week)
Commuters
Commuters — all travel =0-5 hrs/wk
Commuters — all travel = 6+ hrs/wk
Average of
Maximum
1-Hour CO
Cone (ppra)
6.71
6.82
7.35*
5.30
6.40
7.0!J
7.0!5
6.82
5 . 22
7.38*
6 . 34
22.11*
I
4.94
7.06*
5.04
7.90*
Average of
Maximum
8-Hour CO
Cone (ppm)
2.72
3.03
2.94
2.43
2.99
3.03
3.03
2.55
2, 35
2.98*
2.66
7.51*
2.30
2.88*
2.35
3.09*
Mean Hourly Concentrations. Table 6.4.4 presents the results for
the mean hourly concentrations (over all available hours) in the same
i
format as for the previous two variables. The overall average was 1.61
i - - •
ppm. The patterns exhibited for the various subgroups are similar to
those evidenced for the maximum one-hour and maximum eight-hour vari-
ables. Again, the high-exposure occupation subgroup stood out as the
!
group with highest overall levels.
-191-
-------
e
4J
Q
O
B
O
•rt
e
S
B
U
u
B
s
O
U
*^
S
o
C
(3
ej
E
0
•&
en
2
O
J5
C
O
•rt
U
Is
CUCU
o
tux
t> A
CJ
CO B
O O
CUT*.
u 5
u- u
C B
01
u U
c c
O O
U ^
0) C
P* CJ
•o js
O JJ
gS
•S
a
Id
: O
o
CO.
SlS
• U4
§M
* C
4J *H
00
m
X
a\
jj;
CJ •
to 3
o a
a. o
w
B|
to cn
c
t
-J
g
c
•H
JJ
(t
r"
C
C
P.
NO
O
o
in
0
CO
CO
in
o
CM
pv.
CM
en
o
NO
r-
O
O
NO
••*
g
o
o
1215539
IS
Oj
i?
E>
r-l
^
to
0
CO
u
CJ
Cu
fj
00 O
0 0
o c
•T CO
cn cn
o o
P» O
NO -T
in co
NO t—
-a1 in
CM en
CM r-.
O ON
co r-
in NO
ON O
O *-*
O 0
in en
in co
o a
0 0
0 0
c o
r-4 «•*
1215539
1215539
r-4 r~*
as CM
^-
* *
CO
08
CJ
i-3 33
co to
B B
0 0
to to
rl U
CJ o
CU CU
r-( i-t
§O 0
« c
o o o
O O 00
o o so in
»-4 -^ O *-T CO ^^
O O O O O O
n o\ o^ e*J —• «T
cn ^ m co o m
m CM o vo CT* co
r*» \o CM co
B " B
cj a ^
^ !2 O
G
i i a j£
i t u s
B S
O 01 01
o u -a a)
B B *r! >
OI OI CO O
*rj *O 01 *J
•H -H D3 CO
Qi Q 4J t3
Bi K a ec
c;
4J 4-1 CO
a a > i«
o o
at 01 w
CO U
U U to E
CO CO CO C
m K u
m co o u
cj c z eu
B
C
•H
a
3
CU
O
CU
,C8
M
01 _
O
ai
JS
4J
those for
. exposure.
i™i
O CO
"1
o a
B 3
O O
u o
o
"•5
•S3
4J
CO *W
3 O
t3 03
a c
B^
Q) -H
•O C!
-------
6.4.2 Analysis of CO Breath Measurements
Breath samples taken at the end of eachiindividuals' moni-
toring period (late afternoon or early evening) were analyzed to deter-
mine CO concentrations. These measurements were available for 659
sample members. The results of the analysis of these data are shown in
Table 6.4.5. The overall mean CO concentration for the population was
estimated to be 5.12 ppm (with a standard error of .07 ppm). Among the
subgroup categories shown in the table, little variation in this mean
level is evidenced. About 95% of the overall population exhibited
breath CO levels in the range from 1 to 10 ppm. The lack of extreme
variation is perhaps due to the fact that (almost) all of the breath
measurements were made in the same type of environment (i.e., in the
respondents' homes rather than while commuting or on-the-job). Higher .
mean levels were, nevertheless, observed for persons1, with high exposure
occupations and with large amounts of travel.
6.4.3 Analysis of Activities and Associated CO Exposures
As indicated in Section 5.4.2.2, analysis of the Activity
Analysis File (AAF) data required additional processing prior to analy-
sis. Based upon frequency counts of the various activity and location
codes (see Exhibit 5.4.2), a set of five major and sixteen minor environ-
ments were developed. These were the following: '
-193-
-------
P.W
<3 co
C* p i
O P-4
•H
jj K^
CO
i-H A
3
P. C
O O
P-l *H
4->
n3 cd
01 M
CO 4J
0 (3
P4 CU
o
14H CJ
o
o
6-S CJ
• rfl
4-1 4-1
CO iH
w ts
•
u
s §
VO
O vo
fvl •
^^J •
H 0
X
m r-s
rH O
II •
X CM
O •*
1— < i-H
II •
X m
m
in CM
X co
CO
i-H
CM 00
X vo
ON
7j r.
1 °.
vO O
oo o
O 0
CM vo
CO CM
CM rH
«* CO
vo m
m co
CM ON
ON m
i-H PS,
CO CO
ON O
CO CM
VO CO
ON ON
ON VO
O CM
OvClO OOP's OCOCMOOO in OO -sf vo
o m o t-sco ON sr co o o o CMCOCMCO
rH O O OO «-H O O O O O rH O O -si-
CMCOvO rH i— i PS. I-H CM O in O -* -sT in vo
•— i r-< vo inm ooorHor^OrHi-Hstf-co
•st rH CM t-HCO 00 O rH O CM O CM rH CM stf-
in *s1" vO ON ON rH psk m oO rH O CTv ON CD ON
CM O\ in OCO COvOCOvOOO CMCMOOvO
-* si- -3- coo oo -H co PS. ,-1 o co -if vo m
rH rH r-H
vo rH oo ON oo vo vo ON rH ON in co c^ s^* cy\
rHO»O COOO OOCOPTinOCM rHO\CMvO
oo-sl-rH COCM sd-cocMvocor-- corocMin
CMCf)s3- COCO COCMCOCOssf -sts^-CM
rH in ON CO rH ps» CM CM s^* CO in CD CO in CO
cooco vovo ooooocoom vor»sp».co
VOP-.VO ps-ssT oominps-oco oooo-*
O
•H
2
4J
C
CU
U
C
O
CJ
o
CJ
1rj
JJ
ca
cu
m
14-1
o
r^
W
cd
B
g
3
CO
in
•
,
ca o
Q
0
O CJ
CJ
f£j
& OJ)
O -H
1 1
CO CO
f-J pj
O O
co co
M M
cu cu
PH P-i
i-H i-H
i-H i-H
ONOrH covo corHCSooom ssrvocoin
fss ^4 i-H f^ co ^If CO ON CO ^"*N CO s^p CO CO CO
sd-ifNvo mm mscTsa-mvoco mmvr-sf
CMCM-st VOCO COrHVOOOCM-sf COOOONON
PS- v£) CM CD in ON PS* i-H ON VO rH P*s SO ON rH
rHs3- inrH rHCM rHi-HCM
**t irj
CU & 3 CU
S ^s, £l |3 -s^, ^ w
OCU ,M!3-"s.CO& fj
tlC CU M |3 "^s» 5fl ^4 *-s^ rrj Q) pj
M 3 03 M *rj 03 CU ^> !5'
OJ ;3 co }-j ^ i M 4-1 o
'CJ 03 O CO Q) ^Cl lO ^ri{ pj 4-J pj
*rH C3 f^^ &*( £LI f^ ^^ fl) O 0} *^{
CQ £X X Ctj lO *~4 1 + CO t> _3 O C
4J X W O CO)|rHv03 1 1 (2 &
rJCs5 CO CUOVOrHi— 44J CUCUCU
o • >s.T3 B u u u u ca 0 o 13
OOCX3 13 CU 4-iCUCUCUCUCO CUCUCQO
C3 3 CJ t^ii .s^ ^ I> ^ ^ T3 ni) CU i *
iHOu cucu cocdcacdcdbo -H-HP^CO
rMOO CUCU V M M M M (2 COCO
MO 313 N-'iHHHHiH CUCU4JCO
O JS 4-> Pi pi CO CB
&CSM II COrHi-Hr-HrH3 O
CJ-H MrHrHr-jrHg 4J4JCU
4-1 ^3 ;ii coco 01 MH
0-~^ CC 4J||||0 00
3 EJ & oo 3 en en en w o cu 01 4-1
coco BMMMM >>cncu
en m en MM 6 cu cu cu cu C oo o
O & ti cucu o 4-1 j-i 4-1 4-1 g 4-1 4-1 en c
OOO PnP-i U3333O cncocdCU
COQ3CO lEBBBc* C5 co
MMM rHrH SES6@^ COCO CU
cuajcu rHi-H oooooc1 cacdOM
-194-
-------
Major Environments Minor Environment!?
In transit walking, jogging, bicycling
car
other travel \
Indoors - at residence sleeping '
cooking - gas stove
cooking - other or unknown
all other activities
Indoors - not at residence office
store ,
restaurant • j
parking garage i
other
Outdoors near road, constniiction site
or service station
parking area
other ;
Unknown ;
The time spent in each of these environments was determined by adding
times over the activity segments in the AAF for each sample member.
Similarly, by time-weighting the activity segments, the average CO
exposure level for each individual was determined for each environment.
In addition to the environments, times and CO jlevels by activity
i
type were determined in a similar manner. The "sleep" activity was
separated from the "suspected sleep" activity in this case, whereas for
the environments shown above, these two types of segments were treated
as one.
Section 6.4.3.1 below examines the activity patterns and environ-
ments in terms of time durations. The next subsection then deals with
the associated CO exposure levels. '
6.4.3.1 Activity and Location Patterns J
Table 6.4.6 characterizes the activity patterns of
the adult non-smoking population in the B.C. area for the winter of
1982-83. The table shows the estimated number of individuals who were
involved in a particular activity (i.e., the number of persons exposed
to CO through the given activity). Then, for this exposed population,
the amount of time involved in the activity is then characterized by the
mean time duration (and its standard error), and by selected percentiles
-195-
-------
4-1
O
•X3
0)
o
I
m
r-»
co
•a
M
U-)
O
0)
3
O
4-1
CO
5
4J
•H
00 >
c 4-i
•H 4J
j-i a
td <
r-H
CO 4-4
PS O
CO CO
d) (I)
4-1 CU
CO >N
HH
4-> CO
cq 3
W O
•H
C M
O CO
•H >
4J
CO «
iH 0
3 l-l
P4
O 4-1
PU C
0)
<4-4 P.
o en
n
IH
3 «W
CO O
vO
•
««•
0)
CO
4J W
gs,
^§
•H
CD
S
a
«w o
O PM
O CD
co
o
p.
^
4J VI
CO O
•H M-4
P
O
w
0) O /-^
M«H CQ
CO 4J M
(-4 CO fi
CD J-I ^^
^^
iH C
*CJ CO O
CO 4J iH
4J O 4-1
CO H CO
S t-j
•H >H 3
4J O p.
CO O
W 5-5 PH
•X3 S
i-r-~i— i
m
co in co
o o o"
r~comi~~oo cooooo cocxjo
CT* m coint^* omo *—4coi—4
omoooooo—i
r^r-iininocococvio
OCNOOOOOOO
ao r-~ —t o o
oo o oo in cri
r— vo o o o
—4 o o
in r- r--
ooo ooo
ooo
oo CN in
o co o
mi-.ooocovosf.3--3-
.-40JOi-li-4CNf-4OCO
m
m
CN i—I CM -^ o oo m o
CM O O O •—4
cooooomcMi-ir—
r--. t~- co oo CM
o oo co ca
to
p. 1-1
O 4-J
ca
3 co
i-l rH
CD
CO
•H el
60 O CO
CD iH O
M 4-1
4->
O
rH 0)
CO
O O &
CO M
i-l CU CU
e &P.C
i
-« c a) 1-1
3 -H
5-i O
cfl
o
y
4J
co
•H
CU CU 00 4-1
co o
•H'd'O
w
CU
•H
CO CO
n pi
G O
CO -rl 4-1
O 60
cu i cu c
CQT3CI&OOOC -H
0)
.5
3 •COp,OC4JCJCU'l3cO'HCUOO
H O O cfl ti
MtUdJ
rH O CO P
-------
— namely the 10th, 25th, 50th (median), 75th, and 90th percentiles.
The table indicates, for instance, that about 40% of>the total popula-
tion engaged in the activity "indoor chores". For these; 481,602 (est.)
individuals, the mean amount of time in this activity was 2.18 hours and
the median time duration was 1.65 hours. The remaining 60% of the total
population did not engage in this activity (i.e., had a time duration.of
zero hours). The table shows that the average time spent in "transit,
travel" was 2.07 hours (among those persons with this! activity during an
18-26 monitoring period), that "workers" averaged 6.63 hours at work,
i
etc.
Table 6.4.7 shows, in a similar format, the time duration analysis
with respect to the five major environments and the sixteen minor
environments. This table indicates that the average /time indoors at the
residence was 17.63 hours. The mean time in a parking garage (for those
so exposed) was 48 minutes; however, the median time was only 11 minutes.
This indicates, as might be expected, a highly skewed distribution.
6.4.3.2 Carbon Monoxide Exposures
Table 6.4.8 shows estimates of' the CO exposure
levels for the various activity'types. The estimates apply only to
i
those persons actually involved in the particular activity. The table
shows the mean CO exposure level and its estimated standard error, along
with estimates of the proportions of the exposed population having CO
exposure levels above specified levels (1, 2, 4, 9, 25, and 35 ppm).
Note that 28.2% of the population exposed to CO in parking garages
experienced CO levels above 9 ppm while in that environment.
Some of the key information in this table was extracted and refor-
matted to produce Table 6.4.9; in Table 6.4.9, the activities have been
reordered in accordance with the estimated mean CC levels (from highest
to lowest). Table 6.4.9 also furnishes population pjercentile estimates
for the 10th, 25th, 50th (median), 75,th, and 90th percentile points.
Two general observations concerning the results of Table 6.4.9 that
deserve mentioning are:
(1) The ranking of activities, with minor exceptions, is as one
i
would expect (e.g., in parking garage and in transit have, by
far, the highest averages). There is considerable overlap in
-197-
-------
4-1
0
3
O
^q
r— 1
0
CO
*
o
T3
CU
CO
o
k£
w
CO
rH
CO
3
•H
TJ
0
M
UH
o
M
O
•§
3
a
o co
•M 0
CU
O 6
•M 0
Ml M
0 tH
•H >
4J 0
CO W
rH
CU «-l
PS O
CO CO
0) CO
4J CU
co >>
e H
•H
4J CO
CO 3
W 0
•H
0 M
O CO
•H >
4J
CO •«
rH 0
M , *j
3 M
CU
O 4-1
PM 0
0)
1^4 £U
o w
^> o
M 0
CO tH
5 EH
gl
^ 1^4
CO O
fftt
*
sr
•
SC
CU
rH
f£3
CO
H
CO
0 /->
O CO
•H M
4J W
cO *— '
3 0
O O
tH
0) 4-1
S cd
H 3
P.
UH O
0 PH
O CU
tH CO
4J O
3 P.
•H W
M
CO O
*r4 '11
O
0
CO
CO
rH
•rl
0
CO
jj
CO
CM
0
4J
CO
rH
3
P.
•
•U
CO
.
j i
en
0
cu o
60 tH
CO 4->
Met
w
m M
•§ 3
rH
Td CO
CO 4-1
4J O
g H
tH <4-l
4J O
CO
W frS
'O
CU
• CO
Of\
\J
& p.
.A
4J
CO 0
fd tH
c
tH
O\
1^.
O
in
CM
o
O
a
^*v
CO
M
5
0
O
tH
4J
rH
3
P
O
PU
0
0
•H
CO
rH
3
P
O
PH
CO
rH
P
6
cn
4J
0
cu
o
}•{
tH
0
Pd
MH
O
CU
P.
H
*^" vo ^"^ ON co ON c^i f^ cr\ CM co ^**** rH rH co CM *"^* CM m
Or-»<±vo enr~u"ir^cT» incMoocn<-— i i —
vo o\ •— i m inoscvivocM voenenen r^ ^H «^* en ' en rM in
rH t i rH f^N [*^, QO *"•' rH VO ^-p p** f~^ f^*i {^5 ^"^ f*"^ f^ f^ (*™^
•— 1
cn in in CM CM oo in P^ co en c^i ON ^H o^ vo *<3" c^i r— t QO
o co r** cs vo o o** r-- vo. -d* oo * in *— * ^-< o CM o OO^-H o
i— t
co LO r- H vo CD CD to ON »™4 vo ^r *d" ON r*** co ^ r^* CD CD C*"i O1*! ON vO PO CD i^ r*^ "™"^ CO CO vO 00 CD
ON CM co »— i OOC^JCNON r^ co O O ON co r** o tO- — « ON LOONCM co
^* i^ \o vO lO to CO *sf lO lO '"^ *^* *vf *"™^ O4 vO CO CO CD CD
r-iinoN^r minmoNcn -*-*cMcnr«-en r-- rnooN i-^
•^TinCMrH rHi-HOr~»O CMCOP—'-HOOOO O C7NOOO rH
r-icnocM CMCM<-CMCM ONincnen en -3- CM ^
A A A A A
rH rH rH I— 1 rH
cMvocMo miAvocMrH cnoNinooNr«- en •* •* a\ • •*
VOCMONCO OOCOONO oo-*cMr--inin st vomr«
. .
"0 CU •
& O 00)
0-0 O 4J
0 CO -H tH
CU ^! T3 4J CO
CU 0 0 -rl CO
rH 0 3 CO 4J 0
U CO CU 0) CO O
>", "O > M Pi tH
U tH O O CU CO U
tH CO 4J 4-> 60 CO U
rd CUCQM CO CO &03CO
PA CO M M CU
- MX 4J 4-1 Cd « 4J M
60 4JCO-4J O 060 ISCQCd
O ,MCO MM ^CU 0
cOrHM^i MCUOO^C M 4-4 O CO M rC Ocd M,C! §
MCOC04J OrHOO4J O<4H4-ICUCd4J O CU COW O
H&OO OCOUOO OOCOMP.O T30 P.O 0
t3 "O W yX
000 30
M M --..• M O P
a
CO
J-l
•2
CO
g
rH
P.
cd
CO
CO
o
CO
. M
o
M-4
co
CU
•H
4J
tH
•H
O
cd
M
CO
fi
4J
o
U-4
o
0
o
•H
CO
1 ^J
1 0
' 0
•H
O
4J
i CU
: 3
T3
CO
CO
cd
•H
o
CU
,JQ
^1
1
CO
CO
4J
cd
g
•3
4J
CO
W
*
-198
-------
£
•rl
45
y
o
CJ
&
X
*
en
-H 3 «H E3 CM \O CM O CM <•*
< H Q w ^
X C
O *M U * /"J r*. en co en o -*•
^*o o SI — 4 ^-t —* \o ^3- CM
o-63 > o oJ o o o o o o
H
* Ol »•'* •**• O f* O O\
• • u £1 in — ^ tn r** co
to > o o 3 ^- «—* CM ^* — * o
w < u o <*1
1 cn co co in CM —*
^•t OA *d~ o*» o vo *^
u o w cu en o os -^* c?» »•*
co ool o> m W (Xt •-* 3 *C — * »T -»
w *«^
<=4
•rtSl cMcor^-srco**
m to o co i— r- !•»
«ste co| to sr co CM
-
JO w
e to u
•H Oh
•U I-t O
oi o .e
01 f* r_i
"' E ' CJ
•-I1 1
01 CO 1
f, m at
a a 01 ca
K e co 3
H "to o -
^ « 3 E ~
4-1 : 4J tft 00 >i 01
-4: t-t c ij o) -a
•H C J
0. -H
O *J
J= U
CO <1
•• r-i
CO 03
•O C
C O
to to
i-l 0)
Ed a*
sr
o
o
o
c
tO
sr
O
O
tO
tO
_i
CO
r-.
CM
in
in
r~
00
to
CO
CO
*-*
o
at
sr
t-4
tO
at
CM
CM
in
00
tO
00
CO
o
CM
in
m
01
•H
•rt
•H
U
01
U
3
CC
•H
G
O O
0 C
0 O
o o
0 0
o c
00 —
CO CM
-4 O
CO O
o o
r»>* r^
in —
o at
••4
to at
CO O
CO O
sr CM
to \c
o oo
f-4
CM -J
•H M*
O 0
o m
—* CO
— 0
r- r~
o at
in in
— • oo
to co
CM CO-
CM at
co sr
oo sr
1-4 O
CO r»
— • at
-^ m
c.
0)
01
t-H
CO
0)
JJ
u
01
CU
CO
3
CO
00
B
•H
•a
3
i-r
o
c
i
to
3
O
0)
c
ra oa
l-i T<
01 ci.
O CI
to o
c
c
c
o
o
o
o
1^
tO
o
2
CTs
en
-
o
to
sr
Ox
CM
00
CM
0
at
r-.
-1
CO
CO
CO
tO
CM
o
to
in
^,
•o
jj
CO
c
o
t!
CO
§
o
0
c
o
,
in
f-^
at
at
2
in
in
r~
o e
—4 01
at r
O -
at r
s
a a
— c
o a
O r
to tt
u
co r
sr c
,
u
Eti
C
to <
00 C
C f
•H (
JJ i-
• C r>
a «
E <
c
i-H f
a S
CJ
1-t
JJ
11 p
l-i C
o :
a> j
j: c
' u n
' 3 !
— C
; co
i at
J ' —
S CM
> CO
'"'
9 CO
9 CM
4 00
CM
4 m
9 to
1 O
i CM
1 CO
9 -4
- o
9 m
i C3
4 O
j . CO
4 0
••
9 sr
l r-
9 0
1 CO
i at
4 tO
1 —4
. to
> 0
'
9 CO
i O
» at
9 CM
1 -<
- CM
1 00
JJ
) OO
1 ££
\ CU
4
1 M
) ' 0
0
i , OI
ecj
. CC
J C
B
1 00
9 C
J W
4 Pu
a
: c
4 hi
o
o
o
o
o
o
o
CO
CM
^
00
CM
r~
r-*.
CM
r-*
-3-
00
o
_l
CM
0
^
r-
1-4
at
o
CO
CO
tO
o
at
in
f-
CO
13
01
•H
•H
U
a.
o
•rt
E
0)
O
JJ
c
2
-c
CO
4-1
3
O
o o
0 C
o o
O CM
o ^
c o
— < r-*
in m
in en
f-i CM
m crv
O' CM
CM cn
•^ tn
-3- cn
-------
,_)
cu
§
w
0
o
pj
1
o
CO
•H
*X3
M
0
o
CJ
na
CU
c
p5
|
i
4J
•H
•H
U
O
14-1
O
0)
ex
H
j^
PO
A
co
>
cu
0)
3
CO
o
ex
w
o
o
M-4
o
^>
M
(rt
R
R
3
CO
a\
**r
*
VO
0)
f— {
.0
H
*t3
CU
4J
CO
s
•rl
4J
CO
H
*^J
cu
i i
•H
4J
CO
w
CO
co e
CU 0
o cu
i-4 CU
W rH
U^^ -rl
\^ Tl
4-1 O W
B Cl S
•H O CJ
4J -H S-4
CO *J CU
W cd Pn
CU 3 na
I-H ex cu
•H O 4J
4-1 PH O
C CU
CU 13 rH
O CU CU
M CO CO
cu o
PH ex •
w 3
ex
vS
•o o
CU U
u-t co . 1-1
O O 3 J3 CU
ex ex *j >
&* « o iH cu
LM Py f^C |J
•
CU U
• CO O
•U M U
CO CU
w > o
*3J (^
p*
0) O
W) 3
r~cop>-cO'Hcoo\o\-*omr». vf oovooivo
OOJr-4r-CNsOr~CJ>'-l.-iOsOt-- COOS—tol
t-lvOCOCOOJCO'-tOJ^H^-ICMCOOJCNl'H'-l i-HO— •'-<
__l
co o*\ ^* *.r>-.r-.in-tcovo^
O O O i-t «
rH -H 00 i >>O -H 4J 00 4) rH W CO
Cdi-HtH -HC Wf^W W-H-HCOfX Ol
)-4 01 Cu 4J TH OT TH 13 -H 1 t-l > 3 Cu C ^ «l > 3 > > -H ! O CO 0) 4J 0)
oocdo a) c ai -H 4-i -H a) >ri4->cu^3-Ui-ica
60-UCO M*HO U3 OCOJJXCJCJOO
c 5-1 13 m co s-i co o cd ocuaexex^:
•H«« O 3 O 1 ft rH,C! COCO 60
f2sH ^J 01 &Q ** _£!) rt \U ^% QJ OtJ * r^ JI} •* C) C
\4 *tH T3 Cj ^*l toQ 3 r™H ^3 Q) ^ V^ Cl Q) W ^H 03 ?d 'O *rH
cdwpJ-HOtf «oo-HrO'OSOiTd4Ja> O»HCL.
P, pj (jj ^^ 4J *H f.Vi! £2 O CO *ni d CO CO •^^ MO M CO Cl)
cd M O O 4«J )*i ,*^i _c! W CO j3 *iH M CO O CO 13 4-* CU
CMnoocdoeoscicdcucuc 0.1-1 & 3 rH
H&HWOQWSPCQOM.J.JCl.HCQS. OOCO
-200-
-------
the definitions of activities, as described previously in
Section 5.4'. This, along with some bias probably introduced
by respondents' omissions of activities from their diaries,
probably accounts for the exceptions. The most notable
i
exceptions are the higher than expected ranking for "doctor or
dentist office" and the lower-than-expected ranking of "out-
side house - chores".
(2) The distributions of CO exposures appear to be skewed to the
right for all activities; this is indicated not only by the
estimated percentiles, but also by the fact that the mean
levels exceed the median levels.
Tables 6.4.10 and 6.4.11 provide summaries of the analysis of the
CO exposure levels by type of environment. Because the environments
were derived principally from the location codes
-------
•U
c
I
o
^H
>
£
O
1
*"*
&
a
e
5
*"*
O
3
01
i.
Q
t.
Q
W
cd
E
t3
3
OT
C
va
el
*-4
§
C
O
•rt
u
C
O.O4
O Cu,
C-
X
•a
0) A
CO
§.§
x 4J
B
IM 14
O 4J
B
4J 01
e a
0 G
28
CU O
•a
-S 3
< H £3
14 O
CUbl >
< w <
• •
a > O
U < u
4J a
a
U M
• a
4-1 IH
CO
U XX
1
in
en
X
in
CM
X
ON
X
*3"
x
CM
A
s
e'o
•3£
u «-»
d "s
O &
o <-*
• x^
O £
o &
-< .
2 0.
o o
H PW
•0
01 •
a a
O 0.
5. o
tS
M
*.5
^
4.
4.
c
en o vo o
CM o en o
O O C O
-* o en o
en o ON c
o o o o
vo CM in VO
r*> «a vo ON
CO M O ON
ON OV t- CM
f^ oc en ON
en CA r^ ^
,3. ^- T *3" oo cr»
KT K
•H 0 C
CO U
a tn n
PS 01
w ^
<: * d o
i i i
O. li
CO C Ji J•* en *-^
-H — i — in —
cn CM in co -a- c?v
ON -a in ON ON co
ON co r- r^ o — • -a- — <
cn in vo o »^ «•*
-a o CM ON •-> o«
*a- GO -4" -* co
-------
Table 6.4.11
Percentile Estimates of the Exposed Population CO
Levels (ppm) , By Type of Environment
Percentile Point
Environment
In Transit
walk, jog, bicycle
car
other
Indoors - At Residence
sleep *
cook - gas stove
cook - other/unknown
other
Indoors-Not At Residence
office
store
restaurant
parking garage
other
Outdoors
near road, gas station,
construction site
parking area
other
10
1.03
0.09
1.19
0.51
0.07
0.05 .
0.31
0.05
0.07
0.16
0.07
0.05
0.21
0.24
0.05
0.05
0.05
0.33
0.05
25
2.16
0.81
2.46
0.90
0.24
0.05
1.25
0.10
0.33
0.54
0.53
0.69
0.50
1.81
0.20
0.10
0.21
0.41
0.05
50
3.49
1.80
3.72
2.69
0.82
0.44
2.23
0.61
1.04
1
1.11
1.23
1.78
1.40
4.80
0.88
1.04
• i.
1.30
1.51
0.07
"
75".
6.20
3.43
6.42
4.79.
1.62
1 .29
3.56
1.85
2.01
2.47
2.09
3.29
2.71
13.52
1.75
2.89
3.10
3.37
0.78
90
8.42
4.62
9.81
8.65
2.68
1.95
4.93
4.21
3.54
4.04
3.69
4.77
4.04
23.28
4.43
6.19
6.03
17 .77
2.68
Estimates may be biased due to inclusion of other activities for
some sample members. ' j
-203-
-------
Average CO Level (ppm)
(1) indoors - at residence 1.19
(2) indoors - not at residence 2.04
(3) outdoors 2.62
(4) in transit 4.51
The average CO exposure level for the first of these environments was
significantly lower than that of the latter three, and the fourth
environment was significantly higher than any of the first three envi-
ronments. The approximate tests utilized a 5% level of significance.
The results shown in this subsection apply to the "average day" of
the data collection period. Additional results — for high and low CO
exposure days — are shown in Appendix J.
6.4.4 Analysis of Measurement Variability
The Duplicate Measurement File described in Section 5.4.2.3
was utilized for assessing variation in PEM measurements under field
conditions. Hourly observations; from two or more PEMs were available
for 689 hours. It was noted that an extreme deviation occurred between
a pair of PEMs at one hour.
After this outlier was removed, there were 688 hours with two or
more CO exposure measurements. Consequently, 688 standard deviations
were computed. A plot of the 688 standard deviations (STDCONC) versus
their corresponding means (MEANCONC) is given in Figure 6.4.8. Al-
though there is some indication that the standard deviations increase
with increasing mean levels, this tendency is not especially strong.
The distribution of the 688 standard deviations is shown in Figure
6.4.9. It should be noted that the vertical axis of this plot is given
in terms of interval midpoints. The median of the standard deviations
is .25 ppm, and their average is .39 ppm. A corresponding distribution
of the 688 coefficients of variation (CVs) had a median of 16.3% and a
mean of 30.6 %.
In order to compare the measurement component of variability with
person and hourly variations, a variance components model of the follow-
ing form was estimated:
(i)
-204-
-------
CD
§
§
6D u
CO pel
CD g
*rH *K
e o
| |
H
CO OT
O M-l
•H O
CB 4J
> O
!-l r-H
(U PL.
CO
"O 00
W
Pi
W O
H M
O P-i
-205-
-------
O O O O O
o o o o o
o in m
O rH rH
OCMrHOOrHOOOOOOOOOOOOOOOOO
m
rHCOvOCMrHrHCMrHOOOOOrHOrHrH
Percentage Bar Chart
H
2! O
13 •
O O
P-> O
H W
*******************************************
************************************************************************
*************************************************************
4£ * *
*******
*******
*******
0 H -
O O
*
*
*
*
*
*******
oo
O
*
*
*
cr»
O
*
*******
****
O rH
rH T~I
*
*
*
-------
where X , = observed hourly CO concentration for the k* PEM, j
fli
hour, and i interviewer
i -
U = overall mean . , I
P. = effect of i person (interviewer)
th i-h
H ,^ - effect of j hour (within the i interviewer)
ek(ii) = effect of ttie k PEM for person i at hour k. (The
variation associated with this component represents
the measurement variation, under field conditions.)
The results of this analysis is shown in Table .6.4.12.
These results indicate that about 5 to 6% of the total variation
among the hourly readings is due to deviations in the measurements made
by the two (or more) PEMs at the same hour for the same person. The
pooled estimate of the measurement variance is .292.
-207-
-------
Table 6.4.12 Analysis of Replicate Hourly CO Concentrations
Variance
Source
Total
Person
Hour
Error
Pnof^M,
Sum of
D.F. Squares
1537 8495.76
27 3562.01
696 4695.90
814 237.86
pni-. of Variation .
Mean
Squares
5.528
131.926
6.747
0.292
2.35
0.54
, . 0.23
Variance
Component
5.615
2.284
3.039
0.292
Percent
100.00
40.67
54.12
5.20
c -
is:
l
-208-
-------
7. REFERENCES
1. Casady, Robert J. and Sirken, Monroe G. [1980]. A Multiplicity
Estimator for Multiple Frame Sampling.* Proceedings of the American
Statistical Association Section on Survey Research Methods, 601-605.
2. Chromy, James R. [1979]. Sequential Sample Selection Methods.
Proceedings of the American Statistical Association Section on
Survey Research Methods, 401-406.
3. Cochran, W.G. [1963]. Sampling Techniques. 2ud_ed._, John Wiley and
Sons, New York, pp. 327-353.
i
4. Jones, S.M. and J.R. Chromy [1982]. Improve Variance Estimators.
Using Weighting Class Adjustments for Sample Survey Nonresponse.
American Statistical Association 1982 Proceedings of the Section on
Survey Research Methods, pp. 105-110. !
5. Konijn, H.S. [1973], Statistical Theory of Sample Survey Design
and Analysis. American Elsevier Publishing Company, New York,
126-132. i
I
6. McCarthy, Philip J. [1966], Replication; An Approach to the
Analysis of Data From Complex Surveys^ (NCHS Vital and Health
Statistics Series 2 Number 14). Public Health Service, Washington,
B.C.
7. McCarthy, Philip J. [1969]. Pseudoreplicatiou; Further Evaluation
and Application of the Balanced.Half-Sample Technique (Vital and
Health Statistics Series 2 Number 31). Public Health Service,
Washington, D.C.
i
8. Raj, Des [1968]. Sampling Theory. McGraw-Hill, New York, pp.
139-163.
i
9. Shah, B.V. [1981]. SESUDAAN: Standard Errors Program for
Computing of Standardized Rates From Sample Survey Data. Research
Triangle Institute, Research Triangle Park, N,,C.
10. U.S. Department of Commerce, Bureau of the Census. [1978]. The
Current Population Survey: Design and Methodology. Technical
Paper No-. 40. U.S. Government Printing Office, Washington, D.C.
11, Wallace, L.A. and W.R. Ott. "Personal Monitors: A State-of-the-
Art Survey," JAPCA, Vol. 32, 601.
12. Whitmore, R.W,, Jones, S.M., and Rosenzweig, M.S. [1983a], Final
Sampling Report for the Study of Personal CO Exposure. Prepared
for U.S. Environmental Protection Agency. Research Triangle
Institute, Research Triangle Park, N.C. |
-209-
-------
References (cont'd)
13. Whitmore, R.W., Mason, R.E., Hartwell, T.D., and M.S. Rosenzweig
[1983b]. Use of Geographically Classified Telephone Directory
Listings in Multi-Mode Surveys. To appear in: Proceedings of the
American Statistical Association, Survey Research Methods Section.
14. Williams, Rick L. and Chromy, James R. [1980]. SAS Sample Selec-
tion Macros. Proceedings of the Fifth Annual SAS Users Group
International Conference.
tf"
-210-
-------
-------
------- |