SELECTED DATA ANALYSES RELATING TO
STUDIES OF PERSONAL CARBON MONOXIDE
EXPOSURE IN DENVER AND WASHINGTON, D.C.
PE ASSOCIA1 ES
-------
SELECTED DATA ANALYSES RELATING TO
STUDIES OF PERSONAL CARBON MONOXIDE
EXPOSURE IN DENVER AND WASHINGTON, D.C.
by
Ted Johnson, Jim Cape!, and Luke Wijnberg
PEL Associates, Inc.
505 South Duke Street, Suite 503
Durham, North Carolina 27701-3196
Contract No. 68-02-3496
PN 3550
Task Manager
Gerald Akland
U.S. ENVIRONMENTAL PROTECTION AGENCY
OFFICE OF RESEARCH AND DEVELOPMENT
ENVIRONMENTAL MONITORING SYSTEMS LABORATORY
RESEARCH TRIANGLE PARK, NORTH CAROLINA 27711
February 1986
-------
DISCLAIMER
This report has been reviewed by the Environmental Monitoring Systems
Laboratory, U.S. Environmental Protection Agency, and approved for publica-
tion. Approval does not signify that the contents necessarily reflect the
vi.ews and policies of the U.S. Environmental Protection Agency., Mention of
trade names or commercial products does not constitute endorsement or recom-
mendation for use.
-------
FOREWORD
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 environ-
mental protection efforts through the monitoring of long-term trends. The
Environmental Monitoring Systems Laboratory, Research Triangle Park, North
Carolina, has the responsibility for assessment of environmental monitoring
technology 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 selected analyses of personal carbon
monoxide (CO) exposure data obtained in a human exposure study performed in
Denver, Colorado, and Washington, DC, during the winter of 1982-83. This
report discusses relationships between personal exposure to CO and human
activity patterns, ambient aerometric variables, indoor sources, and other
factors.
Thomas R'. Hauser
Director
Environmental Monitoring
Systems Laboratory
111
-------
ABSTRACT
Under EPA Contract 68-02-3496, PEI Associates, Inc. conducted a study of
personal exposure to carbon monoxide (CO) in Denver, Colorado. The target
population for the study included all noninstitutionalized, nonsmoking resi-
dents of the urbanized portion of the metropolitan area who were between 18
and 70 years of age at the time of the study. A total of 454 study participants
were obtained through the use of a screening questionnaire administered to
several thousand households in the study area. Each participant was asked to
carry a personal exposure monitor (PEM) and an activity diary for two con-
secutive 24-hour sampling periods and to provide a breath sample at the end
of each sampling period. Each participant also completed a detailed background
questionnaire. A similar study was conducted in Washington, D.C., by
Research Triangle Institute. Analyses of the Denver fixed-site data suggest
that ambient CO levels decrease with increasing windspeed. Five monitors
reported daily maximum 8-hour concentrations exceeding 15; all were located
in the central business district. Linear regression analyses relating PEM
values to Washington fixed-site readings yielded R2 values exceeding 0.15 for
eight microenvironments: indoors-hospital (0.66), indoors-church (0.60),
indoors-garage (0.19), outdoors-park (0.15), train/subway (0.61), jogging
(0.30), truck (0.27), and bicycle (0.16). Daily maximum 8"-hour exposures in
Denver were found to be higher on days when fixed-site daily maximum 8-hour
values exceeded 9 ppm. Microenvironments found to be associated with daily
maximum 8-hour exposure above 9 ppm include service stations, public garages,
restaurants, outdoor locations within 10 yards of roads in areas of high
ambient CO, and trucks when the trip begins or ends in an area of high ambient
CO. Occupations involving proximity to running motor vehicles or internal
combustion engines in a closed space are strongly associated with high daily
maximum 8-hour exposures. Analyses of Denver in-transit exposures suggest
exposures are higher when inside motor vehicles than when walking. In-vehicle
exposures are higher during rush-hour periods. Smoking does not significantly
increase invehicle exposure. Analyses of indoor exposure data for Denver
identified 10 factors which significantly affected exposure. Exposures were
higher in homes with gas cooking stoves, with gas clothes dryers, with unvented
gas furnaces, with unvented space heaters, and with storm windows, storm doors,
or special dampers. A model was developed which explained 34 percent of the
variation in Denver PEM values. The daily maximum 8-hour exposure values
reported on consecutive days by Denver subjects were not highly correlated
(R2 = 0.16). The PEM's used in the Denver study were found to experience
zero-span problems more frequently on cold days and to experience lock-up
more frequently on warm days.
-------
CONTENTS
Foreword iii
Abstract iv
Figures vii
Tables ..... viii
Abbreviations xvi
Acknowledgment xviii
1. Introduction 1
Organization of report 2
References 3
2. Fixed-Site Monitors 4
Summary statistics for Denver sites 4
Relationship between Denver fixed-site readings and
selected meteorological parameters 7
Comparison of Denver and Washington composite site
summary statistics 13
References 13
3. Relationships Between Exposures and Fixed-Site Readings. ... 15
Review of previous analyses using Denver data 15
PEM values versus reported by optimized group of Denver
fixed-site monitors 17
PEM values versus values reported by nearest Washington
fixed-site monitor 19
Adjusted PEM values versus values reported by nearest
Denver fixed-site monitor 27
Daily maximum exposures versus values reported by Denver
fixed-site monitors 30
References 39
4. Relationships Between Exposures and Selected Explanatory
Variables 40
In-transit exposures 62
Indoor exposures 76
References 95
5. Models for Predicting Exposure in Denver 96
General models 1 through 4 96
General models 5 through 14 104
Aggregation of indoor microenvironments 123
Reference 134
6. Comparison of Consecutive Daily Maximum Exposures 135
Distributions of A and B values 135
Distribution of differences between A and B values. . . . 140
-------
CONTENTS (continued)
Correlation between A and B values 142
Distributions of daily maximum 8-hour exposures by day of
week 142
Distribution of differences between A and B values by day
of week 143
Reference 147
7. Time Spent in Selected Microenvironments 149
8. Factors Associated with Instrument Failure . . 157
Personal exposure monitor 157
Data base . . . . 159
Model used for statistical analysis 160
Test statistics 161
Discussion of results 162
Conclusions 164
Reference 164
9. Summary of Results 165
Fixed-site monitors 165
Relationships between exposures and fixed-site readings . 165
Relationship between exposures and selected exploratory
variables 166
Models for predicting exposure in Denver 168
Comparison of consecutive daily maximum exposures .... 169
Time spent in selected microenvironments 170
Factors associated with instrument failure 170
Appendices
A. A study of personal exposure to carbon monoxide in Denver,
Colorado 172
B. Site descriptions and summary statistics for Washington
fixed-site monitors 188
VI
-------
FIGURES
Number Page
3-1 Histograms of daily maximum 8-hour exposures to carbon
monoxide , 36
3-2 Histograms of logarithms of daily maximum 8-hour exposures
to carbon monoxide 38
5-1 Output of pairwise comparison test program (step one) 130
6-1 Histograms of daily maximum 8-hour exposures to carbon
monoxide 137
6-2 Histograms of logarithms of daily maximum 8-hour exposures
to carbon monoxide 139
8-1 Personal exposure monitor. 158
-------
TABLES
Number Page
2-1 Fixed-Site Monitors Operating in the Denver Metropolitan
Area During Study 5
2-2 Summary Statistics for Hourly Average Carbon Monoxide Values
Reported by Denver Monitoring Sites Between November 1,
1982, and February 28, 1983 6
2-3 Date and Time of Maximum Hourly Average Carbon Monoxide Value 7
2-4 Summary Statistics for Daily Maximum 1-hour Carbon Monoxide
Values Reported by Denver Monitoring Sites Between
November 1, 1982, and February 28, 1983 8
2-5 Summary Statistics for Daily Maximum 8-hour Carbon Monoxide
Values Reported by Denver Monitoring Sites Between
November 1, 1982, and February 28, 1983 9
2-6 Coefficients of Determination and F-to-Remove Values for
Best-Fit Models Determined by Stepwise Linear Regression. . 11
2-7 Summary Statistics for Daily Maximum 1-hour and 8-hour
Carbon Monoxide Concentrations Reported by the Composite
Sites in Denver and Washington 14
3-1 Denver Fixed-Site Monitors and Results of Weighted Linear
Regression Analyses With PEM Value as Dependent Variable
and Simultaneous Fixed-Site Value as Independent Variable . 18
3-2 Results of Step-Wise Multiple Linear Regression Analysis
(Weighted) 20
3-3 Results of Step-Wise Multiple Linear Regression Analysis
(Unweighted) 20
3-4 Format of Files (Washington Only) Listing PEM, Activity
Diary, and Fixed-Site Data for Nontransit and In-Transit
Microenvironments 22
viii
-------
TABLES (continued)
Number Page
3-5 Results of Weighted Linear Regression Analysis (Washington
Only) With Nontransit PEM Value as a Dependent Variable
and Simultaneous Value at Nearest Fixed-Site as
Independent Variable 23
3-6 Results of Weighted Linear Regression Analyses (Washington
Only) With In-Transit PEM Value as Dependent Variable and
Simultaneous Value From Composite Data Set as
Independent Variable 24
3-7 Results of Weighted Linear Regression Analyses (Washington
Only) With In-Transit PEM Value as Dependent Variable and
Simultaneous Value at Fixed-Site Nearest Start Address as
Independent Variable 25
3-8 Results of Weighted Linear Regression Analyses (Washington
Only) With In-Transit PEM Value as Dependent Variable and
Simultaneous Value at Fixed-Site Nearest End Address as
Independent Variable 26
3-9 Weighted Mean Adjusted PEM Concentration (Microenvironments
Ordered According to Table V Listing) 28
3-10 Results of Weighted Linear Regression Analyses With
Nontransit Adjusted PEM Value as Dependent Variable and
Simultaneous Value at Nearest Fixed-Site as Independent
Variable 29
3-11 Results of Weighted Linear Regression Analyses With
In-Transit Adjusted PEM Value as Dependent Variable and
Simultaneous Value From Composite Data Set as Independent
Variable 30
3-12 Percentage of Daily Maximum 8-hour Values Reported by „
Denver Monitoring Sites Exceeding Selected Carbon Monoxide
Concentrations Between November 1, 1982, and February 28,
1983 33
3-13 Days for Which One or More Denver Fixed-Site Monitors
Reported Daily Maximum 8-hour Values Exceeding 9 ppm. ... 34
3-14 Summary Statistics for Daily Maximum 8-hour Carbon Monoxide
Exposures 35
ix
-------
TABLES (continued)
Number Page
3-15 Summary Statistics for Logarithms of Daily Maximum
8-hpur Carbon Monoxide Exposures 37
3-16 Results of Statistical Tests Comparing the Logarithms of
Group L and Group H 38
4-1 Occupancy Period Statistics by Microenvironment and Exposure
Group 41
4-2 Assignment of Denver Fixed-Site Monitors to Site Groups I,
II, and III 44
4-3 Number of PEM Values Reported for Indicated Combinations of
Microenvironment, Exposure Group, and Site Group (Includes
Only PEM Values Recorded in Census Tracts Containing
a Fixed-Site Monitor) 45
4-4 Number of PEM Values Reported for Indicated Combinations of
Microenvironment, Exposure Group, and Site Group (PEM
Values are Matched to Nearest Fixed-Site Monitor 47
4-5 Combinations of Microenvironments and Site Groups With
Sample Sizes Exceeding Five and Significant Observed-to-
Expected Ratios in Table 4-4 Exceeding 1.00 48
4-6 Number of "Start" In-Transit PEM Values Reported for
Indicated Combinations of Microenvironment, Exposure Group,
and Site Group (PEM Values are Matched to Nearest Fixed-
Site Monitor) 49
4-7 Number of "End" In-Transit PEM Values Reported for
Indicated Combinations of Microenvironment, Exposure Group,
and Site Group (PEM Values are Matched to Nearest Fixed-
Site Monitor) 50
4-8 Number of Daily Maximum 8-hour Values Occurring on Indicated
Date by Exposure Group 52
4-9 Number of Person-Days in Occupational Categories Used by
U.S. Bureau of Census by Exposure Group 55
4-10 Number of Person-Days in Selected Aggregate Occupation
Categories by Exposure Group 60
4-11 Variables Considered in Analyses of Variance and Covariance . 62
-------
TABLES (continued)
Number Page
4-12 Analysis of Variance Table for Denver Personal Exposure
Data (Cell Weight = 1, Covariable = None) 67
4-13 Analysis of Variance Table for Denver Personal Exposure
Data (Cell Weight = 1, Covariable = DR) 67
4-14 Analysis of Variance Table for Denver Personal Exposure
Data (Cell Weight = DR, Covariable = None) 68
4-15 Analysis of Variance Table for Denver Personal Exposure
Data (Conditioned on TM) . . . 68
4-16 Analysis of Variance Table for Denver Personal Exposure
Data (Conditioned on MDTR) 69
4-17 Analysis of Variance Table for Denver Personal Exposure
Data (.Conditioned on SM) 69
4-18 Analysis of Variance Table for Washington, D.C. Personal
Exposure Data (Cell Weight = 1, Covariable = None) 70
4-19 Analysis of Variance Table for Washington, D.C. Personal
Exposure Data (Cell Weight = 1, Covariable = DR) 70
4-20 Analysis of Variance Table for Washington, D.C. Personal
Exposure Data (Cell Weight = DR, Covariable = None) .... 71
4-21 Cross Tabulated Exposure Means and Standard Deviations for
Denver MDTR Versus SM Conditioned on TM. 71
4-22 Cross Tabulated Exposure Means and Standard Deviation for
Denver TM Versus SM Conditioned on MDTR 72
4-23 Cross Tabulated Exposure Means and Standard Deviation for
Denver TM Versus MDTR Conditioned on SM 72
4-24 Summary Statistics for Carbon Monoxide Concentration Values
Recorded by Personal Exposure Monitors in Indoor
Microenvironments 76
4-25 Summary Statistics for Carbon Monoxide Concentration Values
Recorded by Personal Exposure Monitors in Indoor
Microenvironments After Transformation 77
-------
TABLES (continued)
Number Page
4-26 Format of Computer File Listing Indoor PEM Values,
Activity Diary Entries, and Selected Background
Questionnaire Responses 79
4-27 Results of Analysis of Variance of Indoor Exposures at Home
Versus Area of Living Quarters 81
4-28 Results of Analysis of Variance of Indoor Exposures at Home
Versus Number of Cigarette Packs Smoked Per Week by Other
Household Members 81
4-29 Results of Analyses of Variance of Indoor Exposures Versus
Response to Selected Questions Concerning Combustion Soures
in Living Quarters 82
4-30 Results of Analyses of Variance of Indoor Exposure Versus
Response to Selected Questions Concerning Energy-Saving
Devices in Living Quarters 83
4-31 Results of Analyses of Variance of Indoor Exposures Versus
Main Heating System in Living Quarters. . . 84
4-32 Results of Analyses of Variance of Indoor Exposures Versus
Main Heating System in Workplace 85
4-33 Results of Pairwise Comparisons of Indoor Exposures
Associated With Combustion Sources in Living Quarters ... 86
4-34 Results of Analysis of Variance of Indoor Exposures at
Home Versus Area of Living Quarters and Cigarette Packs
Smoked Per Week by Other Household Members 89
4-35 Results of Analysis of Variance of Indoor Exposures at
Home Versus Area of Living Quarters and Status of Gas
Furnace 89
4-36 Results of Analysis of Variance of Indoor Exposures at
Home Versus Area of Living Quarters and Status of Gas
Cooking Stove 90
4-37 Results of Analysis of Variance of Indoor Exposures at
Home Versus Area of Living Quarters and Status of Gas
Clothes Dryer 90
xii
-------
TABLES (continued)
Number Page
4-38 Results of Analysis of Variance of Indoor Exposures at
Home Versis Area of Living Quarters and Status of Gas
or Kerosene Space Heater 91
4-39 Results of Analysis of Variance of Indoor Exposures at
Home Versus Area of Living Quarters and Identity of Main
Heating System. 91
4-40 Results of Analysis of Variance of Indoor Exposures Versus
Status of Gas or Kerosene Space Heater and Status of. Gas
5-1
5-2
5-3
5-4
5-5
5-6
5-7
5-8
5-9
5-10
5-11
5-12
5-13
5-14
Group Codes Used in
Resul
Resul
Resul
Resul
ts of
ts of
ts of
ts of
Stepwise
Stepwise
Stepwise
Stepwise
Stepwise Linear Regression Analysis . . .
Li
Li
Li
Li
near
near
near
near
Candidate Exposure Factors
Regression Analyses Invol
Results of
Abrid
Gen
Resul
Resul
Resul
Resul
Resul
Terms
Stepwise
ged Results of
era! Model 6 .
ts of Stepwise
ts of
ts of
ts of
ts of
Stepwise
Stepwise
Stepwise
Stepwise
Selected From
Li
near
Regression
Regression
Regression
Regression
Using
Using
Using
Using
General
General
General
General
Model
Model
Model
Model
Used in Stepwise Linear
ving General Models 5 Through
Regression
Using
General
Model
1 .
2 .
3 .
4 .
14 .
5 .
Stepwise Linear Regression Using
Li-
Li
Li
Li
Li
near
near
near
near
near
General
Regression
Regression
Regression
Regression
Regression
Models 5 1
Using
Using
Using
Using
Using
fhrougl
General
General
General
General
General
i 11 for
Model
Model
Model
Model
Model
7 .
8 .
9 .
10.
11.
^ *r
96
100
101
103
104
106
109
111
113
114
115
117
118
Inclusion in General Models 12, 13, and 14 120
5-15 Results of Stepwise Linear Regression Using General Model 12. 121
xiii
-------
TABLES (continued)
Number Page
5-16 Results of Stepwise Linear Regression Using General Model 13. 122
5-17 Results of Stepwise Linear Regression Using General Model 14. 124
5-18 Summary Statistics for Carbon Monoxide Concentration Values
Recorded by Personal Exposure Monitors in Indoor
Microenvironments 126
5-19 Summary Statistics for Carbon Monoxide Concentration Values
Recorded by Personal Exposure Monitors in Indoor
Microenvironments After Transformation 128
5-20 Microenvironments Listed in Descending Order of Mean and
Median Values Based on Untransformed and Transformed Carbon
Monoxide Values 129
5-21 Results of Pairwise Comparison Tests 133
5-22 M1croenv1ronment Groups Suggested by Pairwise Comparisons
Analysis 134
6-1 Summary Statistics for Daily Maximum 8-hour Carbon Monoxide
Exposures " 136
6-2 Summary Statistics for Logarithms of Daily Maximum 8-hour
Carbon Monoxide Exposures 138
6-3 Results of Statistical Tests Comparing the Logarithms of
A Values and B Values 140
6-4 Summary Statistics for Difference Values 141
6-5 Summary Statistics for Daily Maximum 8-hour Carbon Monoxide
Exposures 144
6-6 Summary Statistics for C Difference Values 145
6-7 Summary Statistics for D Difference Values 146
6-8 Results of Nonparametric Test of Null Hypothesis That Median
C or D Value is not Greater Than Zero 148
xiv
-------
TABLES (continued)
Number Page
7-1 Weighted Summary Statistics for CO Exposures by Microenviron-
ment Considering Only Person-Days With Nonzero Durations. . 150
7-2 Aggregation of Microenvironments Defined for Denver Carbon
Monoxide Study Into Microenvironments Defined for NEM
Analyses of Carbon Monoxide Exposure 152
7-3 Unweighted Summary Statistics for Time Spent Per Day by
Denver Subjects in Microenvironments Used in NEM Analyses . 154
7-4 Unweighted Summary Statistics for Time Spent Per Day by
Denver Subjects in Microenvironments Used in NEM Analyses
Statistics for Each Microenvironment Omit Person-Days
With Zero Time Spent in the Microenvironment 155
8-1 PEM Failure Modes Analyzed by Logistic Regression Model . . . 160
8-2 Results of Fitting Logistic Regression Model to Failure Data. 163
xv
-------
FREQUENTLY-USED ABBREVIATIONS'
ABBREVIATIONS
X
A
ANOVA
b
B
B
BG
BMDP
c
CPEM
C
CMR
CO
d
D
D3
D.F.
DQC
DR
EMSL
EPA
EV
EWA
FSU
GE
H
Ho
HU
L
LCD
LRM
MAGUS
MDTR
n
N
NAAQS
n.e.c.
NEM
NU
OER
P
PEI
or C
PEM
coefficient for Box-Cox transformation
Section 6 only: first of two consecutive sampling periods
analysis of variance
blank
location code from activity diary (Section 3)
Section 6 only: second of two consecutive sampling periods
block group
package of computer programs for statistical analysis
mean CO concentration, ppm, measured during occupancy
period (Section 4.1)
PEM CO reading, ppm
Section 6 only: C = B value - A value
Colorado Market Research
carbon monoxide
mean duration of occupancy period (Section 4.1)
Section 6 only: D = In (B value) - In (A value)
transit mode code from activity diary
degrees of freedom
data quality code
duration, minutes (Section 4.2)
Environmental Monitoring Systems Laboratory
U.S. Environmental Protection Agency
explanatory variable (Section 8.3)
enclosed work area
first-stage sampling unit
General Electric
differing definitions of Group H can be found in Sections
3.5 and 4.1
null hypothesis
housing unit
differing definitions of Group L can be found in Sections
3.5 and 4.1
liquid crystal display
logistic regression model (Section 8.3)
Magus Group, Incorporated
mode of transit (Section 4.2)
number of values or entries
number of pairwise comparisons (Section 5.3) or number of
positive values (Section (6.4)
National Ambient Air Quality Standard
not elsewhere coded
NAAQS Exposure Model
not used (Section 4.3)
observed-to-expected ratio (Section 4.1)
probability
PEI Associates, Inc.
xvi
-------
ABBREVIATIONS
PEM
PID
ppm
PREC
PROPN
R2
RDEN
ROM
RTI
SAMPLE-DATA
SAROAD
SASD
s.d.
s.e.
SM
SSU
std. dev.
TAVG
TDEN
TM
TMAX
TMIN
UNV
UV
WIND
-- personal exposure monitor
— personal identification number
-- parts per million
-- total precipitation for day, inches (Sections 2.2 and 8.3)
-- proportion of PEM's associated with a given temperature and
a given failure mode (Section 8.3)
— coefficient of determination
— residential density (Section 2)
-- read-only memory
-- Research Triangle Institute
-- computer file containing personal exposure data
-- Storage and Retrieval of Aerometric Data (EPA'data bank)
--Strategies and Air Standards Division
-- standard deviation
— standard error
— smoking (Section 4.2)
-- second-stage sampling unit
-- standard deviation
-- daily mean temperature, F (Sections 2.2 and 8.3)
-- trip density (Section 2)
-- time of day (Section 4.2)
-- daily maximum temperature, F (Sections 2.2 and 8.3)
-- daily maximum temperature, F (Sections 2.2 and 8.3)
-- used and not vented (Section 4.3)
-- used and vented (Section 4.3)
— mean windspeed for day, mi/h (Sections 2.2 and 8.3)
Additional abbreviations are listed in Tables 3-1, 3-4, 5-1, 5-6, and 8-1.
xv ii
-------
ACKNOWLEDGMENT
The analyses described in this report were performed by Ted Johnson, Jim
Capel, and Luke Wijnberg of PEI Associates, Inc. Luke Wijnberg is the primary
author of Sections 4.2 and 8; Ted Johnson wrote the remaining sections. Jim
Capel and Luke Wijnberg created most of the data files used in these analyses
and contributed all necessary computer programs. Dave Dunbar was the Project
Director for PEI, and Ted Johnson served as the Project Manager.
The Project Officer for the U.S. Environmental Protection Agency (EPA)
was Gerry Akland. Dave Mage served as Assistant Project Officer. Tom Lawless
supplied data files containing the carbon monoxide values reported by the
fixed sites operating in Denver and Washington during the study period. This
work was supported by the Environmental Monitoring Systems Laboratory of EPA
through Contract No. 68-02-3496 and Contract No. 68-02-3755.
The majority of data analyzed in this report were obtained by PEI from
a study of human exposure to carbon monoxide conducted in Denver, Colorado.
The following persons contributed to the success of that study.
PEI Associates, Inc.
Dave Dunbar - Project Director
Ted Johnson - Project Manager, data validation, and statistical analysis
Tom Wey - instrument calibration and repair; coordination of field
activities; data validation; Sections 3.4, 3.7, 4.5, and
6.3 of this report
Jim Capel - computer programming and data validation
John Schoettelkotte - computer programming
Luke Wijnberg - statistical analysis and nonresponse adjustments to
sample weights
Colorado Market Research
Mitch Veeder - coordination of field activities
Gary Graff - solicitation of participants and administration of tele-
phone questionnaires
Evan Cole - instrument calibration and data reduction
Mike Hughes - instrument calibration and data reduction
Jeff Kicig - coding
Frank Kunc - field screening, interviewer
Mike Dulacki - field screening
Bob Devita - field screening
Brenda Harris - interviewer
xviii
-------
Lorena Jones - interviewer
Nick Marchese - interviewer
Mirza Mirza - interviewer
Jim Roberts - interviewer, field screening
John Russell - interviewer
Glenda Smith - interviewer, field screening
Mitchell Zahn - interviewer
Bob Artman - interviewer
Jack DeVita - interviewer
Pat Kanieski - interviewer
Research Triangle Institute
Roy Whitmore - study protocol, sample selection, and sample weights
Harvey Zelon - administration of telephone screening questionnaires,
development of field procedures
Denver Mayor's Office
Jack Green - Mayor's representative
Cooper Wayman - coordinator for the city, county, and State governments
Rick Young - field coordinator and site operator
State of Colorado - Department of Health
Steve Arnold - fixed-site monitoring and data processing manager
Gordon MacRae - assistant manager, fixed-site monitoring and data
processing
Wayne May - fixed-site monitor operator and field coordinator
Rick Kramer - fixed-site monitor operator and field coordinator
Environmental Protection Agency - Region 8
Jim Lehr - Federal, State, county, city liaison
Charles Stevens - regional representative
Marshall Payne - validation of fixed-site monitoring data
Bill Basbagill - fixed-site monitor calibration and data transfer
Keith Tipton - fixed-site monitor calibration and data transfer
Environmental Protection Agency - Research Triangle Park
Gerry Akland - Project Officer, precision estimates
Robert Jungers - program management
Harold Sauls - program planning
Ron Drago - instrument design
Charles Rodes - instrument design
Tom Hartlage - COED-1 procurement
Ray Ballard - instrument service and repair
Bill Barnard - field audits
Jack Bowen - field audits
Mike Beard - quality assurance and acceptance testing
Larry Purdue - quality assurance and acceptance testing
xix
-------
University of North Carolina - Biometrics Laboratory
William Kalsbeek - evaluation of sampling protocol and proposed
sample stratification
xx
-------
SECTION 1
INTRODUCTION
The National Ambient Air Quality Standard (NAAQS) for carbon monoxide (CO)
states that 1-hour CO concentrations shall not exceed 35 ppm more than once
per year and that 8-hour CO concentrations shall not exceed 9 ppm more than
once per year. Compliance with these standards is usually determined by fixed-
site monitoring data. However, fixed-site monitoring data may not provide an
accurate indication of-personal exposure within an urban population, which is
a function of both geographic location (e.g., downtown versus suburbia) and
immediate physical surroundings (e.g., indoors versus outdoors). Better esti-
mates of personal exposure can be developed by equipping a large number of
subjects with portable monitors and activity diaries. If the subjects are
properly selected, their exposures can be extrapolated to a larger "target"
population.
Two such studies were conducted during the winter of-1982-83 in Denver,
Colorado, and Washington, D.C. Both studies were sponsored by the Environmental
Monitoring Systems Laboratory (EMSL) of the U.S. Environmental Protection Agency
(EPA). In the Denver study, PEI Associates, Inc. (PEI), asked each of 454
subjects to carry a personal exposure monitor (PEM) and an activity diary for
two consecutive 24-hour sampling periods and to provide a breath sample at the
end of each sampling period. Each participant also completed a detailed back-
ground questionnaire. The questionnaire results and approximately 900 subject-
days of PEM and activity diary data collected between November 1, 1982, and
February 28, 1983, were analyzed to determine if factors such as microenviron-
ment and the presence of indoor CO sources significantly affect personal CO
exposure. In addition, the exposure of a defined target population was extra-
polated from exposures recorded by the study participants. PEI also compared
CO levels recorded by fixed-site monitors to levels recorded simultaneously
by PEM's. Detailed descriptions of the Denver study design and data collec-
tion procedures, together with results of initial data analyses, are available
in a report by Johnson.
1
-------
The Washington study, which was performed by Research Triangle Institute,
2 3
has been described in detail by Hartwell et al., Settergren et al., and
A
Clayton et al. It differs from the Denver study in that 1) twice as many
subjects were used in the Washington study and 2) each subject carried a PEM
and a diary for a single 24-hour period.
The present report describes various statistical analyses related to the
Denver and Washington studies which were performed by PEI subsequent to the
reports by Johnson and Hartwell et al. Most of the analyses are exploratory
in nature with the general goal being the development of a model for predicting
CO exposure (as indicated by PEM's) using data recorded at the fixed monitor-
ing sites, in the activity diaries, and in the background questionnaires.
Some of the analyses were performed to answer specific questions posed by EMSL
and the Strategies and Air Standards Division (SASD) of EPA. As noted in the
text, the results of many of the analyses suggested new questions and the
need for additional analysis.
1.1 ORGANIZATION OF REPORT
This report is organized as follows. Section 2 provides summary statistics
for data reported by the fixed-site monitors operating dur~ing the Denver and
Washington studies and discusses relationships between the Denver fixed-site
data and selected meteorological parameters. Sections 3 and 4 discuss relation-
ships between CO exposures as measured by PEM's and selected explanatory vari-
ables based on fixed-site data, activity diary entries, and background question-
naire responses. In Section 5, candidate models for predicting CO exposure are
constructed using the most promising explanatory variables and are then opti-
mized using stepwise regression techniques. Section 5 also discusses how micro-
environments with similar exposure characteristics can be grouped into aggregate
microenvironments. Section 6 describes analyses performed to determine if
daily maximum exposures experienced by participants on consecutive days were
statistically related. Data concerning the average time spent by participants
in various microenvironments and aggregate microenvironments are provided in
Section 7. Factors associated with the failure of PEM's in the field are dis-
cussed in Section 8. Section 9 presents a summary of analytical results and
major conclusions. Note that all analyses are based on Denver data unless
otherwise indicated.
-------
Appendix A provides an overview of the Denver study and summarizes the
statistical analyses described in the earlier report by Johnson. It is
recommended that the reader review Appendix A before proceeding to Section 2.
Site descriptions and summary statistics for the fixed-site CO monitors operat-
ing during the Washington study are provided in Appendix B. Other background
information concerning the Washington study will be provided in the text where
necessary.
1.2 REFERENCES
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983.
2. Hartwell, T. D., C. A. Clayton, R. M. Michie, R. W. Whitmore, H. S.
Zelon, S. M. Jones, and D. A. Whitehurst. Study of Carbon Monoxide
Exposure of Residents of Washington, D. C. and Denver, Colorado.
Prepared for the U.S. Environmental Protection Agency. Research
Triangle Institute, Research Triangle Park, North Carolina. 1984.
3. Settergren, S. K., T. D. Hartwell, and C. A. Clayton. Study of
Carbon Monoxide Exposure of Residents of Washington, D. C.--Addi-
tional Analyses. Prepared for the U.S. Environmental Protection
Agency. Research Triangle Institute, Research Triangle Park, North
Carolina. 1984.
4. Clayton, C. A., S. B. White, and S. K. Settergren. Carbon Monoxide
Exposure of Residents of Washington, D. C.: Comparative Analysis.
Prepared for the U.S. Environmental Protection Agency. Research
Triangle Institute, Research Triangle Park, North Carolina. 1984.
-------
SECTION 2
FIXED-SITE MONITORS
Fifteen fixed-site CO monitors operated in Denver during the period of
the study. Figure 3 in Appendix A shows the locations of these 15 monitors;
Table 2-1 provides the corresponding site characteristics. Nine monitors
were temporary and were discontinued at the conclusion of the study. All of
the monitors reported hourly average CO data and operated continuously.
The quantities RDEN and TDEN relate to the average residential and traf-
fic densities of the census tract containing each monitor and are explained
in Section 5.1 of Reference 1. The land use designation pertains to the
neighborhood in the immediate vicinity of each monitor. Appendix G of
Reference 1 contains more detailed descriptions of the area surrounding each
site. Site selection, data acquisition, and quality assurance activities
are described in Reference 2.
2.1 SUMMARY STATISTICS FOR DENVER SITES
A computer file was created which provides the hourly average values
reported by all 15 fixed-sites between November 1, 1982, and February 28,
1983, on an hour-by-hour basis. From these data an additional variable was
created—the hour-by-hour arithmetic mean of the values reported by the 15
sites. This synthethic data set is denoted by the three-digit code "AVG" and
is referred to as the "composite" data set in the discussion that follows.
Table 2-2 summarizes the results of analyzing the hourly average values
using BMDP program P2D. Table 2-3 lists the date and time of the maximum
value reported by each site. Ten of the 15 maximum values occurred during
either the morning or the evening high traffic periods (8:00, 17:00, or 18:00).
Four of the 15 maximum values occurred on January 27, 1983. The maximum
value in the composite data set was 15.8 ppm and occurred at 8:00 on December 17,
1982.
A supplementary file was created that contained daily maximum 1-hour and
8-hour values for all 16 data sets on a day-by-day basis. Daily maximum
-------
TABLE 2-1. FIXED-SITE MONITORS OPERATING IN THE DENVER METROPOLITAN AREA DURING STUDY
en
Map
code
A
B
C
D
E
F
G
H
I
J
K
L
H
N
0
District
or town
Denver
Denver
Denver
Denver
Denver
Denver
Denver
Greenwood
Village
Denver
Denver
Aurora
Arvada
Highlands
Englewood
Hontbello
Address
2105 Broadway
2325 Irving
14th & Albion
208 Grant St.
1821 S. Yates
3635 Qulvas St.
3509 S. Glencoe
6060 South Quebec
3620 Franklin St.
Speer & Lawrence
50 S. Peoria
5701 Garrison
8100 S. University
3600 S. Elatl
4845 Oakland
SAROAD code
060580002F01
060580014F01
060580013F01
06208082 1F05
062080822F05
062080820F05
062080823F05
062080825F05
0620808 18F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
06208081 7F05
Building
Special
Special
Special
School
School
School
School
Trailer
School
School
Special
Trailer
Trailer
Municipal
building
School
Land use
Central business,
high traffic
Residential
Strip commercial,
high traffic
Downtown
residential
Residential,
near commercial
Residential.
near expressway
Residential, on
suburban artery
Office, light
business
Residential
Campus near
high traffic
Suburban golf
course
Residential near
shopping district
Vacant land at
edge of residen-
tial area
Light commercial
near major shop-
ping center
Offices and small
warehouses
(suburban
comnercial)
Census
tract
a
5.02
43.01
28.02
46.01
11.02
40.04
67.02
36.01
19.00
77.02
103.08
56.15
60.00
41.05
RDEN
b
18. 35
23.34
30.77
11.79
27.95
8.51
7.97
26.75
67.64
8.51
8.89
9.12
13.43
0.0
TOEN
c
5.91
35.27
19.03
5.72
7.90
5.53
2.78
5.06
8.56
4.96
6.19
2.59
6.23
3.60
1981
violations
33
16
d
e
e
e
e
e
e
e
0
3
0
e
e
'16.00 or 25.00.
1981 data.
65.13 or 56.09. C9.97 or 13.51.
Permanent site with no 1981 data. eTemporary site with no
-------
TABLE 2-2. SUMMARY STATISTICS FOR HOURLY AVERAGE CARBON MONOXIDE VALUES REPORTED
BY DENVER MONITORING SITES BETWEEN NOVEMBER 1, 1982, AND FEBRUARY 28, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Composite
Number
of
hourly
values
2865
2621
2579
2693
2788
2777
2708
2744
2724
2618
2842
2846
2827
2724
2716
2880
Hourly average carbon monoxide concentration, ppm
Minimum
0.0
0.0
0.0
0.0
0.0
0.0
0.2
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.2
Maximum
44.1
26.5
25.6
26.7
21.0
29.4
13.9
14.0
24.7
26.8
16.2
23.8
9.6
24.7
15.9
15.8
Mean
4.75
3.38
3.87
2.97
2.28
2.94
1.94
1.53
2.86
2.94
1.83
3.04
0.93
2.86
t
1.72-
2.64
Std. dev.
4.40
3.75
3.54
3.06
2.69
3.07
1.69
1.54
3.03
3.13
1.63
2.85
1.05
3.04
1.76
2.24
Percentlles
10
1.0
0.4
0.9
0.5
0.3
0.5
0.5
0.2
0.5
0.5
0.4
0.7
0.0
0.5
0.3
0.7
25
1.8
0.9
1.4
1.0
0.6
1.0
0.8
0.5
1.0
1.0
0.8
1.1
0.3
0.9
0.5
1.1
50
3.5
2.0
2.6
2.0
1.3
1.9
1.3
1.0
1.8
2.0
1.4
2.1
0.6
1.8
1.1
1.9
75
6.2
4.5
5.1
3.8
2.9
3.8
2.3
2.0
3.8
3.9
2.3
4.0
1.2
3.8
2.2
3.4
90
9.8
8.4
8.8
6.6
5.6
6.9
4.1
3.5
6.6
6.6
3.8
6.6
2.2
14.6
4.1
5.6
cr>
-------
TABLE 2-3. DATE AND TIME OF MAXIMUM HOURLY AVERAGE CARBON MONOXIDE VALUE
Map code
A
B
C
D
E
F
.6
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Maximum
hourly avg., ppm
44.1
26.5
25.6
26.7
21.0
29.4
13.9
14.0
23.8
26.8
16.2
23.8
9.6
24.7
15.9
Date
12-16-82
1-27-83
1-27-83
12-26-82
1-15-83 '
1-27-83
11-05-82
1-03-83
12-17-82
1-27-83
11-10-82
1-03-83
12-09-82
12-17-82
1-11-83
Time
17:00
19:00
17:00
17:00
21:00
19:00
10:00
17:00
8:00
18:00
8:00
8:00
13:00
8:00
17:00
values were not determined for days with less than 18 hours out of the possible
24. Table 2-4 summarizes the results of analyzing the daily maximum 1-hour
values using BMDP program P2D. Table 2-5 provides similar results for daily
maximum 8-hour values. These tables supersede Tables 6-15 and 6-16, respectively,
in Reference 1.
2.2 RELATIONSHIP BETWEEN DENVER FIXED-SITE READINGS AND SELECTED
METEOROLOGICAL PARAMETERS
Two of the permanent monitors (Map Codes A and B) and three of the
temporary monitors (Map Codes D, F, and J) reported daily maximum 8-hour
-------
TABLE 2-4. SUMMARY STATISTICS FOR DAILY MAXIMUM 1-HOUR CARBON MONOXIDE VALUES REPORTED
BY DENVER MONITORING SITES BETWEEN NOVEMBER 1, 1982, AND FEBRUARY 28, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Composite
Number
of
daily max.
values
120
108
108
113
120
118
115
118
114
110
118
119
116
118
114
120
Daily maximum 1-hour carbon monoxide concentration, ppm
Minimum
1.5
0.6
2.0
1.0
0.5
0.8
0.8
0.7
0.8
0.8
1.3
0.9
0.4
0.8
0.2
0.7
Maximum
44.1
26.5
25.6
26.7
21.0
29.4
13.9
14.0
23. 8a
26.8
16.2
23.8
9.6
24.7
15.9
15.8
Mean
12.87
10.33
11.22
8.89
7.92
8.71
5.30
4.53
8.64
8.72
5.09
8.42
2.98
,8.80
5.23
6.62
Std. dev.
7.41
5.58
5.26
5.11
4.60
4.75
2.71
2.47
5.12
5.21
2.81
4.69
1.57
5.16
2.69
3.16
Percentiles
10
5.6
2.8
4.4
3.6
2.6
3.2
2.3
2.0
3.4
3.5
2.0
2.5
1.3
3.0
1.7
2.8
25
8.0
5.9
7.4
5.6
4.7
5.6
3.3
2.5
5.0
5.0
3.1
5.0
1.9
5.0
3.4
4.5
50
10.5
9.9
10.6
7.9
7.1
8.0
4.8
4.0
7.3
7.4
4.5
7.5
2.8
7.8
5.1
6.0
75
16.1
14.2
14.3
10.7
10.5
11.1
7.0
6.2
12.3
10.8
6.3
11.1
3.9
12.1
6.6
8.2
90
23.7
17.3
19.1
16.8
14.1
15.0
9.5
7.9
15.6
16.1
9.2
14.8
5.3
15.1
8.3
10.6
00
The maximum 1-hour value of 24.7 listed in Table 6-13 is not listed here because it occurred on a day
which did not meet data completeness criteria discussed in Section 6.6.
-------
TABLE 2-5. SUMMARY STATISTICS FOR DAILY MAXIMUM 8-HOUR CARBON MONOXIDE'VALUES REPORTED
BY DENVER MONITORING SITES BETWEEN NOVEMBER 1, 1982, AND FEBRUARY 28, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Composite
Number
of
daily max.
values
120
108
107
113
118
118
114
113
112
108
118
118
116
116
114
120
Daily maximum 8-hour carbon monoxide concentration, ppm
Minimum
1.3
0.4
1.4
0.8
0.4
0.5
0.7
0.6
0.7
0.5
1.0
0.7
0.2
1.2
0.1
0.6
Maximum
20.7
18.5
13.1
15.2
14.1
15.1
7.8
7.2
13.6
15.2
9.3
13.2
5.8
13.5
8.6
10.3
Mean
7.66
6.11
6.31
5.13
4.14
5.16
3.05
2.48
4.97
5.00
2.98
4.86
1.57
i
5.01
3.01
4.16
Std. dev.
3.97
3.60
2.86
2.85
2.45
2.84
1.51
1.40
2.93
2.96
1.51
2.40
1.08
2.79
1.58
2.01
Percent! les
10
3.4
1.9
2.5
1.9
1.5
1.9
1.6
1.1
1.8
2.1
1.5
2.0
0.5
2.2
1.3
1.7
25
4.9
3.3
3.9
3.1
2.4
3.0
2.0
1.4
2.8
3.1
1.9
2.8
0.9
2.7
1..9
2.9
50
6.8
5.9
5.7
4.5
3.4
4.9
2.6
2.1
4.2
4.2
2.7
4.5
1.3
4.4
2.7
3.7
75
9.6
7.9
8.5
6.2
5.6
6.6
4.0
3.2
6.4
6.2
3.7
6.5
1.8
6.4
3.8
5.3
90
13.8
10.9
10.3
9.5
7.5
9.0
5.1
4.1
9.3
8.9
5.0
8.1
3.3
9.5
5.2
7.0
-------
values exceeding 15 ppm. These five monitors were all located in the central
business district of Denver, an area of high traffic density. A detailed
analysis of fixed-site data and associated meteorological conditions by the
State of Colorado concluded that the critical meteorological conditions pro-
ducing 8-hour CO concentrations in excess of 15 ppm were 1) wind speed < 6
mph, 2) morning temperature inversion > 4°C/100 m, 3) inversion depth between
1000 and 1500 feet, and 4) high pressure with light winds at 700 mb and
500 mb levels.2
In a supplemental analysis, PEI attempted to relate fixed-site readings
with the following meteorological variables as reported at Stapleton Inter-
national Airport:
TMAX = daily maximum temperature, °F
TMIN = daily minimum temperature, °F
TAVG = daily mean temperature, °F
' PREC = total precipitation for day, inches
WIND = mean windspeed for day, mi/h.
Note that there is only one value per day for each of the_variables. At the
time of this analysis, meteorological data reported at 3-hour intervals (the
standard interval for the National Weather Service) were unavailable for
analysis. Figure 3 in Appendix A shows the location of Stapleton International
Airport with respect to the 15 fixed-site monitors operating during the study.
PEI conducted stepwise linear regression analyses using the general
model
CFS = a + (BjMTMAX) + (62)(TMIN) + (B3)(TAVG)
+ (S4)(PREC) + (35)(WIND) + (agMWIND)"1
+ (3?)(WIND)'2 + (B8)[ln(WIND)] (2-1)
where Crr = estimated hourly average fixed-site CO concentration, ppm.
2
Table 2-6 lists the overall R value of each "best-fit" model suggested by
the stepwise regression analyses and the F-to-remove value associated with
10
-------
TABLE 2-6. COEFFICIENTS OF DETERMINATION AND F-TO-REMOVE VALUES FOR BEST-FIT
MODELS DETERMINED BY STEPWISE LINEAR REGRESSION
Map
code
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Composite
R2
0.072
0.133
0.071.
0.072
0.077
0.114
0.093
0.089
0.110
0.087
0.068
0.097
0.100
0.063
0.098
0.122
F to remove3
TMAX
88
199
133
115
b
153
b
b
219
134
b
87
b
70
70
163
TMIN
b
b
b
b
b
b
b
b
b
b
b
126
b
b
82
78
TAVG
b
144
b
b
b
91
b
b
137
b
b
b
b
b
b
b
PREC
b
b
b
b
b
b
b
b
b
b
b
b
b
b
b
b
WIND
b
b
b
b
b
b
b
b
b
b
b
b
b
b
b
b
WIND"1
b
b
127
b
b
231
276
b
131
. b
208
b
b
158
b
285
WIND"2
b
238
b
b
b
b
b
268
b
. b
b
b
313
b
b
b
In(WIND)
176
b
b
141
232
b
b
b
b
174
b
151
b
b
181
b
Largest F value is underlined.
'Variable not retained in model.
11
-------
each variable retained by the best-fit model. The relative contribution a
variable makes to the best-fit model increases with the F-to-remove value.
Note that the largest F-to-remove value for each best-fit model is under-
lined in Table 2-6. In most cases, the largest F-to-remove value is
associated with WIND"1, WIND"2, or In(WIND). Although TMAX has the largest
F-to-remove value in only two cases, it is retained in most of the best-fit
models. TMIN and TAVE are retained in less than half of the best-fit models;
PREC and WIND do not appear in any of the best-fit models. The best-fit
model for the composite site is
cps = -1.041 + (0.060)(TMAX) - (0.051)(TMIN) + (13.8)(WINb)"; (2-2)
2
R = 0.122. Note that the estimated CO concentration increases with maximum
daily temperature, decreases with minimum daily temperature, and decreases
with windspeed.
2
None of the R values in Table 2-6 exceeds 0.133. Consequently, the
meteorological variables included in the general model (Equation 2-1) appear
to be poor predictors of the 1-hbur values measured by the fixed-site moni-
tors. The two best predictors are the natural logarithm of windspeed and the
reciprocal of wind speed. TMAX is the best temperature-related predictor.
To determine if WIND raised to a power could provide a better fit
than any of the variables considered in Equation 2-1, PEI evaluated the model
CFS = (a)(WIND)b (2-3)
by conducting linear regression analysis using the equivalent expression
ln(cFS) = ln(a) + b[ln(WIND)] (2-4)
where CFS was the 1-hour CO value at the composite site. The regression
analysis yields ln(a) = 2.293, b = -0.835, and a slightly larger R2 (0.138).
The best-fit model is thus
cps = (9.908)(WIND)"0'835. (2-5)
As would be expected, CO concentration decreases as windspeed increases.
12
-------
As previously indicated, the meteorological variables considered in the
general models represented by Equations 2-1 and 2-3 consist of daily values.
2
Higher R values for best-fit models may have resulted if three-hour meteo-
rological variables had been included in the general model.
2.3 COMPARISON OF DENVER AND WASHINGTON COMPOSITE SITE SUMMARY STATISTICS
Site descriptions and summary statistics for the 11 fixed-site monitors
operating in Washington are presented in Appendix B. PEI prepared a data set
for a "composite" monitor, the data set consisting of the hour-by-hour
arithmetic means of the values reported by the 11 Washington sites. Table
2-7 presents daily maximum summary statistics for the Washington composite
site together with those of the Denver composite site discussed in Section
2.1. Comparison of the summary statistics for the two composite sites reveals
that Denver experienced much higher ambient CO levels during the study period
than did Washington. With respect to composite daily maximum 1-hour CO con-
centrations, Denver has a mean of 6.6 ppm--more than twice Washington's mean
of 3.2 ppm.
2.4 REFERENCES
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983.
2. The Denver Carbon Monoxide Study: Fixed Station Siting, Data
Acquisition, and Quality Assurance. Air Pollution Control Division,
Department of Health, State of Colorado. September 1983.
3. Dixon, W. J., ed. BMDP Statistical Software 1981. University of
California Press, Berkeley, California. 1981.
13
-------
TABLE 2-7. SUMMARY STATISTICS FOR DAILY MAXIMUM 1-HOUR AND 8-HOUR CARBON
MONOXIDE CONCENTRATIONS REPORTED BY THE COMPOSITE SITES IN DENVER AND
WASHINGTON
Statistic
Minimum
Maximum
Mean
Standard deviation
10th percentile
25th percentile
50th percentile
75th percentile
90th percentile
95th percentile
98th percentile
99th percentile
Daily maximum CO concentration, ppm
Denver3
1-h values
0.7
15.8
6.6
3.2
2.8
4.5
6.0
8.2
10.6
13.0
14.7
15.3
8-h values
0.6
10.3
4.2
2.0
1.7
2.9
3.7
5.3
7.0
8.2
9.2
9.4
Washington
1-h values
0.8
8.6
3.2
1.8
.1-4
1.9
2.8
4.1
5.9
6.9
7.9
8.4
8-h values
0.7
6.4
2.3
1.2
1.1
1.4
1.9
2.7
4.1
4.8
5.4
5.5
120 values.
110 values.
14
-------
SECTION 3
RELATIONSHIPS BETWEEN EXPOSURES AND FIXED-SITE READINGS
PEI has performed a series of regression analyses to test whether any
strong, linear relationships exist between personal exposures, as indicated by
PEM readings, and CO levels reported simultaneously by fixed-site monitors.
The analyses are described here in the order in which they were performed.
Section 3.1 reviews linear regression analyses of Denver data preciously re-
ported by Johnson and summarized in Appendix A. In these analyses, the
dependent variable is PEM value (grouped by microenvironment) and the inde-
pendent variable is hourly-average CO value reported simultaneously by the
nearest fixed site or by the "composite" site. Section 3.2 describes step-
wise linear regression analyses in which PEM value is the dependent variable
and the simultaneous values reported by all 15 Denver fixed sites and the
composite site are 16 independent variables. PEM values are not grouped by
microenvironment in the Section 3.2 analyses. Section 3.3* describes linear
regression analyses performed by PEI using Washington data. The analyses
are similar to those described in Section 3.1 in that the dependent variable
is PEM value grouped by microenvironment and the independent variable is
simultaneous value reported by the nearest fixed site or by the composite
site. Section 3.4 repeats the analyses described in Section 3.1 using
"adjusted" PEM values from the Denver study as the dependent variable. Section
3.5 compares daily maximum 1-h and 8-h exposures determined from PEM data
with daily maximum 1-h and 8-h CO values reported by the Denver fixed-site
monitors.
3.1 REVIEW OF PREVIOUS ANALYSES USING DENVER DATA
In the absence of personal monitoring data, estimates of population
exposure are often based on fixed-site monitoring data. In some applications
of the NAAQS Exposure Model (NEM), for example, the air quality in a particu-
lar microenvironment is estimated using the equation
15
-------
= a
Vt m
where x_ . is the estimated pollutant concentration in microenvironment m at
HI 9 U
time t; am is an additive factor related to pollutant sources in the micro-
environment (e.g., gas stoves in the residential microenvironment); b is a
multiplicative factor; and x^ . 1s the air pollutant concentration reported
by a particular fixed-site monitor at time t. Equation 3-1 Implies that a
strong, linear relationship exists between pollutant levels in certain micro-
environments and simultaneous pollutant levels measured at fixed-site moni-
tors. This assumption can be examined in the case of CO by performing linear
regression analyses that use PEM values grouped by microenvironment as the
dependent variable and simultaneously-recorded fixed-site values as the in-
dependent varialbe.
To perform these analyses, each PEM value must be paired with a value
reported by a single fixed-site monitor. Since the census tract of each non-
transit PEM value is known, it seems reasonable to assign a single fixed-site
monitor to each census tract in the study area. Whenever a PEM value is
reported for a given census tract, it is paired with the simultaneous value
of the fixed-site monitor assigned to that census tract. -
One possible method of assigning fixed-site monitors to census tracts is
to use the monitor located nearest to the geographic centroid of the census
tract. An implicit assumption of this method is that the correlation between
ambient CO measurements taken at two locations increases as the separation
distance decreases. As a test of this assumption, the correlations between
all pairs of fixed-site monitors in Denver were calculated using BMDP pro-
gram PSD. As discussed in Section 6.9 of Reference 2, correlation was found
to decrease as separation distance increased.
This analysis suggested that a linear regression analysis that pairs
each nontransit PEM value with the simultaneous value reported at the nearest
fixed site might be appropriate for the Denver study data. A computer program
was written that determined the fixed-site monitor nearest to each census
tract centroid. Weighted linear regression analyses were then performed with
the data grouped by selected microenvironment codes (i.e., B + D3). Results
for nontransit microenvironments for which n > 10 are listed in Table V of
2 2"
Appendix A in order of R value. Values of R range from 0.00 to 0.46. As
16
-------
2
might be expected, many of the microenvironments with small R values are
associated with local CO sources that tend to reduce the correlation between
PEM value and nearest fixed-site value; however, other microenvironments
2
associated with local CO sources have relatively large R values.
Table V does not list any in-transit microenvironments because of the
difficulty in pairing in-transit PEM values with a "nearest" fixed-site moni-
tor value. In the SAMPLE-DATA file, each in-transit PEM value has two census
tract listings, one associated with the start address and the other with the
end address. Neither may be a good indicator of the CO conditions encountered
during the trip. One alternative procedure is to pair the intransit PEM
values with simultaneous values from the composite data set described in
Section 2.1. As discussed in Section 6.9 of Reference 2, the composite data
set shows relatively high correlations with most of the fixed-site data sets.
It also provides an indication of the average CO level in the study area.
Table VI in Appendix A lists the results of linear regression analyses pairing
in-transit PEM values with simultaneous values from the composite data set.
2
Values of R . range from 0.04 (car) to 0.58 (motorcycle).
3.2 PEM VALUES VERSUS VALUES REPORTED BY OPTIMIZED GROUP-OF DENVER FIXED-
SITE MONITORS
The analyses described in Section 3.1 suggest that the correlation be-
tween PEM values and nearest fixed-site (or composite-site) 1-hour CO values
is weak for most microenvironments. PEI also investigated whether the 1-hour
CO values reported by a particular fixed-site monitor or "optimized" group
of fixed-site monitors were better correlated with PEM values. In an exploratory
analysis, PEI performed weighted linear regression analyses with PEM value as
dependent variable and simultaneous fixed-site value at a particular monitor
as the independent variable. Table 3-1 lists the results for each of the 15
2
Denver monitors and for the "composite" site. Note that R values are quite
low; they range from 0.010 to 0.049. Site 4 (2105 Broadway) has the largest
R value. Other sites with R >_ 0.037 include Site 16 (Composite), Site 9,
Site 10, and Site 11.
PEI also performed step-wise multiple linear regression analyses with PEM
value (cpFM) as dependent variable and the simultaneous values reported by all
15 fixed-site monitors and by the "composite site" as 16 independent variables
17
-------
TABLE 3-1. DENVER FIXED-SITE MONITORS AND RESULTS OF WEIGHTED LINEAR
REGRESSION ANALYSES WITH PEM VALUE AS DEPENDENT VARIABLE AND
SIMULTANEOUS FIXED-SITE VALUE AS INDEPENDENT VARIABLE
Site,
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Site code
3-digit
HIG
ARV
AUR
002
013
014
817
818
819
820
821
822
823
824
825
AVE
Map3
M
L
K
A
C
B
0
I
J
F
D
E
G
N
H
SAROAD
060080002F01
060120002F01
060140002F01
060580002F01
060580013F01
060580014F01
062080817F05
062080818F05
062080819F05
062080820F05
06208082 1F05
062080822F05
062080823F05
062080824F05
062080825F05
Composite
Linear regression
Intercept
2.66
2.19
2.02
1.71
2.02
2.43
2.20
2.35
2.05
2.11
2.05
2.40
1.97
2.22
2.07
1.73
Slope
0.624 .
0.364
0.632
0.322
0.321
0.263
0.631
0.355
0.385
"0.382
0.381
0.367
0.621
0.355
0.711
0.589
RZ
0.010
0.026
0.026
0.049
0.032
0.022
0.028
0.028
0.039
0.037
0.037
0.025
0.028
0.029
0.027
0.044
See Figure 3 of Appendix A.
18
-------
(c.j, i = 1,2,...,16). The step-wise regression was performed by adding or
subtracting one site at a time and testing for an improvement in model fit.
The results are summarized in Table 3-2 for the weighted analysis and in
2
Table 3-3 for the unweighted analysis. Note that in each table the R value
2
for the best fit is small (<0.07) and that the R value for Site 16 (the
2
composite site) by itself is almost as large as the multiple R value for
the best fit. The regression equation corresponding to the best fit is
= 0.781 + (0.257)^) + (0.155)(c2) + (0.135)(c4)
+ (0.047)(c5) - (0.077)(c6) + (0.142)(c10)
- (0.075)(c12) + (0.312)(c13) (3-2)
for the weighted analysis (Table 3-2) and
cpEM = 0.777 + (0.252)(Cl) + (0.057)(c4) - (0.154)(c8)
- (0.138)(c12) + (0.175)(c13) + (0.764)(clg) (3-3)
for the unweighted analysis (Table 3-3). For a discussion of sample weights,
see Section 2.4 of Reference 2.
Overall, these analyses suggest that one-hour values reported by a
particular fixed-site monitor or "optimized" group of fixed-site monitors do
not provide a good means of predicting simultaneous PEM values.
3.3 PEM VALUES VERSUS VALUES REPORTED BY NEAREST WASHINGTON FIXED-SITE
MONITOR
As previously indicated, Research Triangle Institute conducted a -study
of personal exposure to carbon monoxide (CO) in Washington, D.C., during the
period November 8, 1982 - February 25, 1983. Appendix B provides summary
statistics for 1-hour and 8-hour CO values reported by the Washington fixed-
site monitors for this period. This section contains the results of linear
regression analyses conducted by PEI which relate PEM values provided by RTI
to 1-hour CO concentrations recorded simultaneously at the fixed sites.
Statistical analyses performed by RTI on the Washington data have been
34 5
described by Hartwell et al., by Settergren et al., and by Clayton et al.
19
-------
TABLE 3-2. RESULTS OF STEP-WISE MULTIPLE LINEAR REGRESSION
ANALYSIS (WEIGHTED)
Step
1
2
3
4
5
6
7
8
9
10
Operation
Add Site 16
Add Site 13
Add Site 4
Add Site 2
Remove Site 16
Add Site 1
Add Site 10
Add Site 6
Add Site 12
Add Site 5
Resulting R
0.0515
0.0544
0.0562
0.0572
0.0571
0.0577
0.0582
0.0585
0.0587
0.0589
2
Change in R
0.0515
0.0030
0.0018
0.0010
-0.0001
" 0.0007
0.0003
0.0003
0.0002
0.0003
TABLE 3-3. RESULTS OF STEP-WISE MULTIPLE LINEAR REGRESSION
ANALYSIS (UNWEIGHTED)
Step
1
2
3
4
5
6
Operation
Add Site 16
Add Site 1
Add Site 8
Add Site 12
Add Site 13
Add Site 4
2
Resulting R
0.0606
0.0638
0.0647
0.0656
0.0663
0.0667
2
Change in R
0.0606
0.0032
0.0009
0.0009
0.0007
0.0003
20
-------
Table 3-4 presents the format of two files developed by PEI which combine
PEM, activity diary, and fixed-site data. For purposes of this analysis, PEI
identified the microenvironment associated with each PEM value through the
four-digit variable LOG (columns 12-15). In-transit microenvironments were
further differentiated through the four-digit variable MODETRAV (columns 16-19),
given LOG = 0100 (i.e., in transit). In developing these files, PEI edited some
of the RTI data to ensure that the LOG and MODETRAV codes were consistent.
In the file, a single census tract (TRACT1) is listed for each non-
transit PEM value. Two census tracts are listed for each in-transit PEM value:
TRACT2 corresponds to the "start" address, and TRACTS corresponds to the "end"
address. Based on the census tract code TRACT1, each nontransit'PEM value is
paired with the value (COLEV1) reported by the nearest fixed-site monitor for
the hour during which the PEM value was measured. Three fixed-site values are
assigned to each in-transit PEM value. COLEV1 is the value for the "composite"
fixed site; COLEV2 is the value for the fixed site nearest TRACT2; and COLEV3
is the value for the fixed site nearest TRACTS.
Weighted linear regression analyses were performed with the data grouped
by selected microenvironment codes. Results for microenvironments for which
2
n ^ 10 are listed in Tables 3-5 through 3-8 in order of R- value. Table 3-5
provides the nontransit results where the independent variable is the fixed-site
value and the dependent variable is the PEM value. Tables 3-6, 3-7, and 3-8
provide in-transit regression results where the independent variables are the
2
composite, start, and end fixed-site values, respectively. Values of R range
from 0.00 to 0.66. As might be expected, many of the microenvironments with
2
small R values are associated with local CO sources that tend to reduce the
correlation between PEM value and nearest fixed-site value.
2
Only two nontransit microenvironments have R values exceeding 0.20:
hospital (R2 = 0.66) and church (R2 = 0.60). The R2 value for office is 0.06;
2
the R value for residence is 0.02. Excluding the subcategories "multiple
2
response" and "missing," the in-transit microenvironments which have an R
value exceeding 0.20 in Table 3-6, 3-7, or 3-8 are train/subway, jogging, and
2
truck. The R values for car range from 0.06 to 0.08 in the three tables.
It is interesting to note that for a given in-transit microenvironment, Tables
2
3-6 and 3-8 generally contain larger R values than Table 3-7. This suggests
that in-transit PEM values are better paired to fixed-site values reported by
21
-------
TABLE 3-4. FORMAT OF FILES (WASHINGTON ONLY) LISTING PEM, ACTIVITY DIARY,
AND FIXED-SITE DATA FOR NONTRANSIT AND IN-TRANSIT MICROENVIRONMENTS
Variable
PID
ACTNO
ACT
LOC
MODETRAV
GARAGE •
GASSTOVE
SMOKERS
BEGTIM
SAMPDATE
DUR
COLEV
LCAT
CLA
TRACT1
TRACT2
TRACTS
DWEIGHT
SITE1
COLEV 1
SITE23
COLEV2a
SITES3
COLEV3a
Length
7
2 '
2
4
4
2
2
2
7.4
6
7
8
2
2
6
6
6
12
6
5
6
5
6
5
Columns
1 -
8 -
10 -
12 -
16 -
20 -
22 -
24 -
26 -
33 -
39 -
46 -
57 -
59 -
61 -
67 -
73 -
79 -
91 -
97 -
102 -
108 -
113 -
119 -
7
9
11
15
19
21
23
25
32
38
45
53
58
60
66
72
78
90
96
101
107
112
118
123
Description
Person ID number
Activity sequence number
Activity code
Location of activity
Mode of travel
Garage attached to building?
Gas stove in use?
Smokers present?
CO interval start time (hours)
Date of sample (MMDDYY)
Duration of activity (minutes)
CO level from field data (PPM)
Major environment
Minor environment
Census tract for address 1
Census tract for address 2 (start)
Census tract for address 3 (end)
Diary analysis weight
Site code for address 1
CO concentration at nearest site for non-
transit microenvironment, at composite
site for in- transit microenvironment
Site code for address 2
Concentration at SITE2
Site code for address 3
Concentration at SITE3
Appears only in in-transit file.
22
-------
TABLE 3-5. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSIS (WASHINGTON ONLY)
WITH NONTRANSIT PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE AT
NEAREST FIXED SITE AS INDEPENDENT VARIABLE
Code
0660
0663
0661
0884
0667
0883
0300
0700
0400
0200
0881
0668
0665
0500
0669
Mi croenvi ronment3
Category"
Indoors
Indoors
Indoors
Outdoors
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Subcategory
Hospital
Church
Garage
Park, sports arena,
playground
Laboratories
Residential area
Office
Within 10 yards of
road or street
Store
Residence
Garage, parking
lot
Not specified
School , school gym
Restaurant
Other indoor
location
Linear regression
n
46
44
70
11
23
82
1741
224
178
14962
38
57
239
120
129
Intercept
-0.05
-0.04
4.02
0.06
0.30
0.53
0.94
1.33
1.25
1.21
5.05
3.52
1.01
2.88
5.07
Slope
0.63'
0.58
3.43
-0.01
0.26
0.52
0.45
0;50
.0.33
0.18
-0.42
-0.16
0.06
-0.03
. 0.09
R2
0.66
0.60
0.19
10.15
0.11
0.10
0.06
0.04
0.02
0.02
0.00
0.00
0.00
0.00
0.00
P
0.000
0.000
0.000
0.239
0.132
0.003
0.000
0.002
0.047
0.000
0.709
0.751
0.555
0.848
0.900
aListed in order of R value.
pProbability that slope = 0.
23
-------
TABLE 3-6. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSES (WASHINGTON ONLY)
WITH IN-TRANSIT'PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE FROM
COMPOSITE DATA SET AS INDEPENDENT VARIABLE
Code
0500
0661
9600
9800
0200
0400
0300
0100
0664
0662
In-transit
subcategory
Train/ subway
Jogging
Multiple response
Missing
Car
Truck
Bus
Walking
Van
Bicycle
Linear regression
n
38
11
20
22
2646
85
67
510
21
16
Intercept
0.05
0.43
-0.98
-0.21
1.51
2.16
1.01
1.21
1.91
3.62
Slope
1.09
0.67
2.58
1.83
1.74
2.00
2.45
"0.94
0.33
-0.08
R2
0.61
0.25
0.20
0.13
0.08
0.07
0.05
0.03
0.03
0.01
P
0.000
0.118
0.050
0.100
0.000
0.014
0.066
0.000
0.478
0.721
a 2
Listed in order of R value.
pProbability that slope = 0.
24
-------
TABLE 3-7. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSES (WASHINGTON ONLY)
WITH IN-TRANSIT PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE AT
FIXED SITE NEAREST START ADDRESS AS INDEPENDENT VARIABLE
Code
0500
9800
0661
0400
9600
0200
0100
0662
0664
0300
In-transit
subcategory
Train/ subway
Missing
Jogging
Truck
Multiple response
Car
Walking
Bicycle
Van
Bus
Linear regression
n
23
21
11
63
14
1748
355
11
16
36
Intercept
0.67
0.30
1.13
0.48
0.26
2.98
1.50
3.97
2.39
8.90
Slope
0.49
2.29
0.41
3.94
1.28
0.97
0.46
-0.06
-0.62
-0.65
R2
0.47
0.31
0.30
0.27
0.21
0.07
0.07
0.01
0.01
0.01
P
0.000
0.009
0.081
0.000
0.099
0.000
0.000
0.737
0.695
0.600
aListed in order of R value.
pProbability that slope = 0.
25
-------
TABLE 3-8. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSES (WASHINGTON ONLY)
WITH IN-TRANSIT PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE AT
FIXED SITE NEAREST END ADDRESS AS INDEPENDENT VARIABLE
Code
9800
0661
0662
0400
0200
9600
0500
0100
0664
0300
In-transit
subcategory
Missing
Jogging
Bicycle
Truck
Car
Multiple response
Train/ subway
Walking
Van
Bus
Linear regression
n
12
11
11
48
1708
11
13
323
16
45
Intercept
0.01
1.13
4.28
0.96
3.12
4.06
2.61
2.33
2.54
5.18
Slope
1.73
0.41
-0.18
3.10
0.83
-0.41
0.16
0.22
0.03
0.05
R2
0.53
0.30
0.16
0.11
0.06
0.01
0.01
0.00
0.00
0.00
P
0.007
0.081
0.231
0.024
0.000
0.727
0.729
0.235
0.920
0.946
aListed in order of R value.
pProbability that slope = 0.
the composite site or site nearest the end address than to those reported by
the site nearest the start address.
It is also worth noting that for a given in-transit microenvironment, n
will be larger in Table 3-6 than in Table 3-7 or Table 3-8. This is because
in-transit PEM values could always be paired with a composite fixed-site
value but not always with a "start" or "end" fixed-site value. In many cases,
no start or end census tract code was provided for a PEM value.
The potential for high CO exposures in indoor garages is evidenced by
the large slope (3.43) and intercept (4.02) values listed for this microen-
vironment in Table 3-5. No other nontransit microenvironment has a slope >1.
Two microenvironments have intercepts larger than 4.02: outdoors - garage/
parking lot (5.05) and indoors - other location (5.07). The finding that both
2
of these microenvironments have R values of 0.00 suggests that local sources
26
-------
(e.g., automobiles in the garage) may mask the traffic-oriented ambient levels
measured by the nearest fixed-site monitor.
3.4 ADJUSTED PEM VALUES VERSUS VALUES REPORTED BY NEAREST DENVER FIXED-SITE
MONITOR
Table III of Appendix A lists weighted means and standard deviations for
the PEM values recorded in various microenvironments during the Denver study.
At the request of EMSL, PEI developed a similar table based on "adjusted" PEM
values. Each adjusted PEM value is the reported PEM value minus the simulta-
neously reported CO value at the nearest fixed site (nontransit microenviron-
ments) or at the composite site (in-transit microenvironments). 'Table 3-9
lists these results. Note that indoors - public garage, motorcycle, indoors -
service station, and bus are the microenvironments with the largest means in
Table III and in Table 3-9. The adjustment process does not appear to nullify
the differences between the microenvironments.
Tables 3-10 and 3-11 are similar to Tables V and VI in Appendix A except
that adjusted PEM values were used as the dependent variables in the regres-
sion analyses. For 13 of the 32 microenvironments, the adjustment process
2
yields larger R values. These microenvironments include-
79 bb outdoors - park or golf course
78 bb outdoors - sports arena
61 bb indoors - other public building
59 bb indoors - health care facility
02 bb indoors - residence
60 bb indoors - school
03 bb indoors - office
76 bb outdoors - residential grounds
72 e outdoors - public garage
56 bb indoors - auditorium
53 bb indoors - manufacturing facility
51 bb indoors - residential garage
62 bb indoors - other location
2
Adjustment yields only reduced R values among the in-transit microenvironments,
These results are consistent with our expectations. The microenviron-
2
ments which yield higher R values after adjustment are those with low average
CO levels before adjustment. Thus the adjustment process subtracts relatively
large fixed-site values from relatively small PEM values to yield adjusted
values nearly equal to the fixed-site values but opposite in sign. Linear
27
-------
TABLE 3-9. WEIGHTED MEAN ADJUSTED PEM CONCENTRATION
(MICROENVIRONMENTS ORDERED ACCORDING TO TABLE V LISTING)
Code
B
52
01
54
01
72
01
71
62
01
55
58
51
07
01
bb
05
74
03
73
56
04
80
59
61
53
02
77
60
57
76
01
78
79
D3
a
93
bb
03
a
02
bb
bb
04
bb
bb
bb
c
01
bb
bb
bb
c
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
92
bb
bb
Microenvironment
Category
Indoors
In transit
Indoors
In transit
Outdoors
In transit
Outdoors
Indoors
In transit
Indoors
Indoors
Indoors
Outdoors
In transit
Not specified
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
In transit
Outdoors
Outdoors
Subcategory
Public garage
Motorcycl e
Service station or motor
vehicle repair facility
Bus
Public garage
Car
Residential garage or
carport
Other location
Truck
Other repair shop
Shopping mall
Residential garage
Within 10 yards of road
Walking
Not specified
Restaurant
Service station or motor
vehicle repair facility
Office
Parking lot
Auditorium, sports arena,
concert hall , etc.
Store
Other location
Health care facility
Other public building
Manufacturing facility
Residence
School grounds
School
Church
Residential grounds
Bicycle
Sports arena, amphitheater,
etc.
Park or golf course
n
110
22
112
76
29
3631
22
381
405
46
55
66
468
619
583
486
11
2090
51
94
675
115
336
111
41
20953
15
342
178
70
9
16
18
CO concentration,
ppm
Mean
8.24
7.32
6.23
5.88
1.70
5'. 44
4. '47
4.87
4.58
3.35
2.18
1.65
1.28
1.28
1.05
1.06
2.00
0.39
1.13
0.51
0.49
0.65
-0.51
-1.02
-1.42
-0.41
-0.11
-0.92
-0.53
-0.45
0.19
-0.12
-1.55
Std. dev.
18.21
6.22
8.56
6.05
3.66
9.72
8.79
19.33
9.39
7.20
5.52
7.44
4.82
5.90
6.47
3.85
3.48
4.45
4.21
5.01
5.38
4.13
4.39
3.29
3.27
4.19
2.61
3.18
3.05
2.47
3.39
3.85
1.35
Includes D3 = bb, 01, and 02.
'Blank.
Includes D3 = bb and 01.
Includes D3 = bb, 01, 02, and 03,
28
-------
TABLE 3-10. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSES WITH NONTRANSIT
ADJUSTED PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE
AT NEAREST FIXED SITE AS INDEPENDENT VARIABLE
Code
B
80
79
77
54
05
74
07
57
73
55
78
61
58
04
59
02
60
03
71
bb
76
72
52
56
53
51
62
D3
bb
bb
bb
bb
bb
bb
c
bb
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
e
e
bb
bb
bb
bb
Mi croenvi ronment
Category
Outdoors
Outdoors
Outdoors
Indoors
Indoors
Outdoors
Outdoors
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Not
specified
Outdoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Indoors
Subcategory
Other location
Park or golf course
School grounds
Service station or
motor vehicle
repair facility
Restaurant
Service station or
motor vehicle
repair facility
Within 10 yards of
road
Church
Parking lot
Other repair shop
Sports arena, amphi-
theater, etc.
Other public building
Shopping mall '
Store
Health care facility
Residence
School
Office
Residential garage
or carport
Not specified
Residential grounds
Public garage
Public garage
Auditorium, sports
arena, concert
hall , etc.
Manufacturing
facility
Residential garage
Other location
n
115
18
15
112
486
11
468
178
51
46
16
111
55
675
336
20953
342
2090
22
583
70
29
110
94
41
66
381
Linear regression,,
Intercept
0.35
-0.09
-0.37
4.18
1.69
1.61
1.58
0.09
2.26
3.69
3.05
0.74
1.24
1.67
0.97 -
1.00
0.97
2.53
5.67
2.07
0.84
3.02
9.41
2.25
1.41
3.98
7.94
Slope
0.11
-0.61
0.15
0.68
-0.24
0.21
-0.11
-0.30
-0.40
-0.12
-2.76
-0.58
0.43
-0.44
-0.55
-0.57
-0.68
-0.66
-0.39
-0.37
-0.70
-0.20
-0.22
-0.62
-0.82
-0.86
-0.93
R"
0.01
0.65
0.01
0.06
0.03
0.01
0.00
0.05
0.11
0.00
0.31
0.24
0.01
0.06
0.12
0.13
0.26
0.18
0.02
0.02
0.18
0.08
0.00
0.10
0.41
0.10
0.03
P
0.325
0.000
0.780
0.012
0.000
0.780
0.166
0.004
0.019
0.659
0.024
0.000
0.389
0.000
0.000
0.000
0.000
0.000
0.504
0.001
0.000
0.143
0.567
0.002
0.000
0.009
0.000
Listed in order of Table V.
Includes D3 = bb and 01.
^Includes D3 = bb, 01, and 02.
"Blank.
Includes D3 = bb, 01, 02, and 03.
pProbability that slope = 0.
29
-------
TABLE 3-11. RESULTS OF WEIGHTED LINEAR REGRESSION ANALYSES WITH IN-TRANSIT
ADJUSTED PEM VALUE AS DEPENDENT VARIABLE AND SIMULTANEOUS VALUE FROM
COMPOSITE DATA SET AS INDEPENDENT VARIABLE
Code
B
01
01
01
01
01
01
D3
93
03
01
04
02
a
In-transit
subcategory
Motorcycle
Bus
Walking
Truck
Car
All
Linear regression
n
22
76
619
405
3631
4762
Intercept
4.50
3.17
0.06
3.27
6.01
5.15
Slope
1.14
1.02
0.47
0.54
-0.21
-0.08
R2
0.28
0.13
0.03
0.02
0.00
0.00
P
0.011
0.002
0.000
0.013
0.002
0.186
aIncludes D3 codes 01, 02, 03, 04, 92, and 93.
pProbability that slope = 0.
2
regression analysis yields high R values with negative slope coefficients in
such cases. All 13 of the microenvironments listed above have negative slopes
in Table 3-10.
The rationale for adjusting indoor PEM values is that indoor PEM values
should nearly equal simultaneous outdoor CO levels in the absence of indoor
sources. An adjusted value thus provides a measure of CO from indoor sources
if the fixed-site value subtracted from the PEM value equals the CO level
immediately outside the building. The fixed-site monitors in Denver tend to
be traffic-oriented and thus generally overestimate typical outdoor CO levels
near most indoor locations in Denver. For this reason, adjustment of PEM
values does not appear to be a promising approach to characterizing indoor
sources of CO.
3.5 DAILY MAXIMUM EXPOSURES VERSUS VALUES REPORTED BY DENVER FIXED-SITE
MONITORS
The analyses discussed in Sections 3.1 through 3.4 suggest that individual
PEM readings are not highly correlated with simultaneous fixed-site readings.
In a supplemental analysis, PEI investigated whether daily maximum 1-hour and
30
-------
8-hour exposures reported by PEM's in the Denver study were correlated with
daily maximum 1-hour and 8-hour CO concentration reported by fixed sites.
PEI developed a file which lists the daily maximum 1-h and 8-h CO exposures
for each person-day of the Denver study. Paired with these values are the
daily maximum 1-h and 8-h composite fixed-site values which occurred during
the midnight-to-midnight period containing the daily maximum 1-h exposure or
the first hour of the daily maximum 8-h exposure. The composite fixed-site
data were used because of the difficulty in pairing daily maximum exposures
which spanned two or more census tracts with a single fixed-site monitor.
Weighted linear regression analysis using the daily maximum 1-h exposure
as the dependent variable and the daily maximum 1-h composite fixed-site value
as the independent variable yields the regression equation
cexp,lh = 8'32 + (C
2
with R = 0.0067. Weighted linear regression analysis using the daily maximum
8-h exposure as the dependent variable and the daily maximum 8-h composite
fixed-site value as the independent variable yields the regression equation
2
with R = 0.057. Repeating the linear regression analyses without weighting
yields
Sxp.lh = 8'47
with R2 = 0.0075 and
2 2
with R = 0.058. The small R values suggest that composite fixed-site daily
maximum values are poor predictors of daily maximum exposures.
In a related task assignment for EMSL, PEI was directed to investigate
the magnitude of exposures among the Denver study participants on days when
31
-------
violations of the National Ambient Air Quality Standard (NAAQS) for carbon
monoxide (CO) occurred. For purposes of this analysis, PEI defined a violation
as the occurrence of a daily maximum 8-hour CO value at any one of the 15
fixed-site monitors operating in Denver during the study. Daily maximum
1-hour data were not considered because only one site (Map Code A) reported
daily maximum 1-hour values exceeding 35 ppm during the study period; whereas,
11 of the 15 fixed-site monitors reported daily maximum 8-hour values exceeding
9 ppm (Table 3-12). These results suggest that fixed-site monitors operating
in the Denver area are much more likely to report violations of the current
8-hour National Ambient Air Quality Standard (NAAQS) than the current 1-hour
NAAQS.
Table 3-13 lists the days during the study period for which one or more
fixed-site monitors reported daily maximum 8-hour values exceeding 9 ppm. If
the daily maximum 8-hour exposure calculated for a person-day of PEM data
started on one of these days, the person-day was included in Group H; other-
wise, it was included in Group L. BMDP program P2D was used to analyze the
daily maximum 8-hour exposure values in Group H, in Group L, and in the com-
bined group (i.e., all values). Only values for person-days with valid overall
data quality codes were analyzed. Table 3-14 lists summary statistics taken
from the BMDP runs. The mean of Group H is 6.69 ppm; the mean of Group L is
4.13 ppm. Figure 3-1 presents histograms for the two groups. Because both
distributions are skewed and have large kurtosis values (Table 3-14), PEI in-
vestigated taking the natural logarithms of the exposure values as a means of
obtaining more normal distributions. Table 3-15 lists summary statistics for
the transformed data; Figure 3-2 provides histograms. The values for skewness
and kurtosis are much smaller in Table 3-15 than in Table 3-14, but are still
significant. For this reason, PEI performed both parametric and nonparametric
tests on the grouped data. Table 3-16 lists the results of these tests.
The Levene test is a test for homogeneity of variance. The small p value
(0.0001) suggests the variances of the logarithms of Groups L and H are not
equal. Consequently, the t (separate) test is more appropriate than the t
(pooled) test for determining if the means of the two groups are equal under
the assumption of normality. Since p < 0.0001 for the t (separate) test, one
can conclude the means of the logarithms are not equal.
32
-------
TABLE 3-12. PERCENTAGE OF DAILY MAXIMUM 8-HOUR VALUES REPORTED BY
DENVER MONITORING SITES EXCEEDING SELECTED CARBON MONOXIDE
CONCENTRATIONS BETWEEN NOVEMBER 1, 1982, AND FEBRUARY 28, 1983
Map code
A
B
C
D
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
060580013F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824F05
062080817F05
Percentage of daily maximum 8-hour values
> 7 ppm
49.2
35.2
33.6
22.1
13.6
18.6
1.8
0.9
20.5
22.2
3.4
20.3
0
20.7
2.6
> 9 ppm
28.3
15.7
18.7
13.3
4.2
8.5
0
0
10.7
8.3
0.8
4.2
0
11.2
0
> 12 ppm
13.3
7.4
3.7
1.8
1.7
2.5
0
0
1.8
4 ."6
0
0.8
0
1.7
0
> 15 ppm
7.5
1.9
0
0.9
0
0.8
0
0
0
0.9
0
0
0
0
0
33
-------
TABLE 3-13. DAYS FOR WHICH ONE OR MORE DENVER FIXED-SITE MONITORS
REPORTED DAILY MAXIMUM 8-HOUR VALUES EXCEEDING 9 PPM
Date
11-05
11-06
11-10
11-12
11-13
11-15
11-16
11-17
11-18
11-19
11-24
11-25
11-27
11-29
11-30
12-03
12-04
12-06
12-09
12-12
Number of sites
reporting values
> 9 ppm
6
6
6
1
4
1
2
2
2
1
9
1
6
1
1
1
1
1
9
3
Composite
site value
7.4
7.1
7.2
5.3
5.0
5.8
6.1
5.3
5.8
4.7
9.4
3.1
8.2
4.5
4.5
5.6
6.2
5.3
9.2
6.3
Date
12-13
12-16
12-17
12-20
12-21
12-26
12-29
12-31
1-01
1-03
1-04
1-06
1-07
1-11
1-12
1-15
1-17
1-19
1-20
1-27
Number of sites
reporting values
> 9 ppm
7
8
5
2
3
1
5
2
1
3
1
2
4
1
4 '
10
2
4
2
8
Composite
site value
8.6
9.1
7.1
6.1
4.7
5.6
8.4
4.6
5.2
6.1
5.3
4.5
5.6
3.9
7.0
10.3
5.5
6.9
6.0
8.1
Daily maximum 8-hour concentration (ppm).
34
-------
TABLE 3-14.
SUMMARY STATISTICS FOR DAILY MAXIMUM 8-HOUR
CARBON MONOXIDE EXPOSURES
Statistic
Number of cases
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Skewness/std. error
Kurtosis/std. error
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
98th percentile, ppm
99th percentile, ppm
Group H
227
0.0
44.0
6.69
6.7
5.68
22.06
50.15
2.1
3.5
5.6
8.2
11.3
15.0
25.4
36.8
Group L
493
0.0
34.8
4.13
1.4
3.82
24.28
54.91
0.7
1.7
3.2
5.2
8.3
10.4
16.1
17.6
All
770
0.0
44.0
5.05
a
4.73
36.48
95.76
1.0
2.2
3.9
6.6
9.7
12.8
17.6
25.0
Not unique.
35
-------
o
-------
TABLE 3-15. SUMMARY STATISTICS FOR LOGARITHMS OF DAILY
MAXIMUM 8-HOUR CARBON MONOXIDE EXPOSURES
Statistic
Number of cases
Minimum, In(ppm)
Maximum, In(ppm)
Mean, In(ppm)
Mode, In(ppm)
Standard deviation, In(ppm)
Skewness/std. error
Kurtosis/std. error
10th percentile, In(ppm)
25th percentile, In(ppm)
50th percentile, In(ppm)
75th percentile, In(ppm)
90th percentile, In(ppm)
95th percentile, In(ppm)
98th percentile, In(ppm)
99th percentile, In(ppm)
Group H
277
-3.69
3.78
1.62
1.90
0.81
-9.04
22.41
0.74
1.25
1.72
2.10
2.42
2.71
3.23
3.61
Group L
493
-3.69
3.55
1.00
0.34
1.08
-12.49
16.34
-0.36
0.53
1.16
1.65
2.12
2.34
2.78
2.87
All
770
-3.69
3.78
1.22
a
1.04
-15.99
23.84
0.00
0.79
1.36
1.89
2.27
2.55
2.87
3.22
*Not unique.
37
-------
20
15
10
GROUP L
GROUP H
, n n
-4
-3
-2 -1 0.1 2
LOGARITHM OF DAILY MAXIMUM 8-HOUR EXPOSURE, LN (PPM)
Figure 3-2. Histograms of logarithms of daily maximum 8-hour
exposures to carbon monoxide.
TABLE 3-16. RESULTS OF STATISTICAL TESTS COMPARING THE LOGARITHMS OF
GROUP L AND GROUP H
Test
t (separate)
t (pooled)
Levene
Mann-Whitney3
Kruskal-Wallis3
Assumed distributions
Normal, unequal variances
Normal, equal variances
Normal
None
None
Test
statistic
-9.09
-8.41
14.80
41554.50
81.42
D.F.
705
768
1, 768
1
P
0.0000
0.0000
0.0001
0.0000
0.0000
Results are independent of log transformation.
38
-------
The two nonparametric tests (Mann-Whitney and Kruskal-Wallis) also yielded
p values less than 0.0001. These results suggest that the null hypothesis
that the two groups have the same distributions be rejected. Usually when the
null hypothesis is rejected, the assumption is made that one group has a
higher median.
The general conclusion from these analyses is that the median concentra-
tions of Group L and Group H differ significantly at the p = 0.0001 level.
The median for Group H is 5.6 ppm--an increase of 2.4 ppm (75%) over the Group
L median of 3.2 ppm. Because both distributions are nonnormal, it is difficult
to determine if the means of the two distributions are significantly different.
3.6 REFERENCES
1. Johnson, T. and R. Paul. The NAAQS Model (NEM) Applied to Carbon
Monoxide. U.S. Environmental Protection Agency, Research Triangle
Park, North Carolina. EPA-450/5-83-003, December 1983.
2. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983. •
3. Hartwell, T. D., C. A. Clayton, R. M. Michie, R. W. Whitmore, H. S.
Zelon, S. M. Jones, and D. A. Whitehurst. Study of Carbon Monoxide
Exposure of Residents of Washington, D. C. and Denver, Colorado.
Prepared for the U.S. Environmental Protection Agency. Research
Triangle Institute, Research Triangle Park, North Carolina. 1984.
4. Settergren, S. K., T. D. Hartwell, and C. A. Clayton. Study of
Carbon Monoxide Exposure of Residents of Washington, D. C.--Addi-
tional Analyses. Prepared for the U.S. Environmental Protection
Agency. Research Triangle Institute, Research Triangle Park, North
Carolina. 1984.
5. Clayton, C. A., S. B. White, and S. K. Settergren. Carbon Monoxide
Exposure of Residents of Washington, D. C.: Comparative Analysis.
Prepared for the U.S. Environmental Protection Agency. Research
Triangle Institute, Research Triangle Park, North Carolina. 1984.
39
-------
SECTION 4
RELATIONSHIPS BETWEEN EXPOSURES AND SELECTED EXPLANATORY VARIABLES
A major goal 1n analyzing data from the Denver study is the Identifica-
tion of factors associated with high CO exposure among the study subjects.
This section begins with an exploratory analysis which considered the rela-
tionships between high daily maximum exposures and a variety of candidate
factors, including the frequency that a subject occupies a microenvironment,
the duration of exposure in the microenvironment, the subject's occupation,
traffic density, mode of transportation, and fixed-site reading. Also included
in this section are analyses of factors which affect in-transit and indoor
exposures.
4.1 HIGH DAILY MAXIMUM 8-HOUR EXPOSURES
EMSL directed PEI to identify factors associated with person-days of
data for which the daily maximum 8-hour exposure exceeds 9 ppm. To facilitate
this investigation, PEI divided the person-days into two exposure groups.
Group H (high exposures) contains person-days with daily maximum 8-hour values
exceeding 9 ppm; Group L (low exposures) contains the remaining person-days.
Note that these definitions differ from those used in Section 3.5.
PEI began an exploratory analysis by comparing the microenvironment
listings of Group H person-days with those of Group L. The basic unit of
analysis was the occupancy period. As described in Section 6.7 of Reference
1, an occupancy period begins when a subject enters a microenvironment and
ends when the subject leaves the microenvironment. Table 4-1 lists by expo-
sure group the number of occupancy periods reported for each microenvironment
(n), the mean duration of the occupancy periods, and the mean CO concentra-
tion measured during the occupancy periods. It should be noted that mean
occupancy periods for the indoor residential microenvironment are likely to
be inaccurate because subjects were usually occupying residences before the
first diary entry and after the last diary entry. The microenvironments are
40
-------
TABLE 4-1. OCCUPANCY PERIOD STATISTICS BY MICROENVIRONMENT AND EXPOSURE GROUP
Code
B
52
01
54
01
72
01
71
62
01
55
58
51
07
01
bb
05
74
03
73
56
04
80
59
61
53
02
77
60
57
76
01
78
79
03
a
93
bb
03
a
02
bb
bb
04
bb
bb
bb
c
01
bb
bb
bb
c
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
92
bb
bb
Mlcroenvlronment
Category
Indoors
In transit
Indoors
In transit
Outdoors
In transit
Outdoors
Indoors
In transit
Indoors
Indoors
Indoors
Outdoors
In transit
Not specified
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
In transit
Outdoors
Outdoors
Subcategory
Public garage
Motorcycle
Service station or motor
vehicle repair facility
Bus
Public garage
Car
Residential garage or
carport
Other location
Truck
Other repair shop
Shopping mall
Residential garage
Within 10 yards of road
Walking
Not specified
Restaurant
Service station or motor
vehicle repair facility
Office
Parking lot
Auditorium, sports arena,
concert hall, etc.
Store
Other location
Health care facility
Other public building
Manufacturing facility
Residence
School grounds
School
Church
Residential grounds
Bicycle
Sports arena, amphitheater.
etc.
Park or golf course
Total
Occupancy periods
Group H
Sampl e
size
(nH)
29
5
38
9
13
304
6
16
53
3
6
5
74
49
41
42
3
73
9
9
47
16
5
3
0
293
6
10
8
6
4
0
1
TT86
Mean
duration,
m1n
26.9
19.2
59.4
13.5
5.7
27.7
2.2
144.5
29.1
138.5
71.0
3.5
28.0
14.9
49.5
149.2
11.0
163.2
33.2
93.6
69.5
21.8
182.9
43.4
322.5
16.3
138.9
80.4
29.6
14.0
89.0
Mean
cone.,
ppm
14.3
6.4
11.9
15.4
9.8
13.1
9.8
36.2
14.5
18.4
8.1
5.3
5.9
7.5
4.9
3.6
5.7
7.5
5.9
10.4
6.7
7.6
3.5
7.7
6.4
1.2
7.7
3.4
1.6
1.6
0.0
Group L
Sampl e
size
(nL)
45
10
14
49
11
1998
13
148
196
8
18
42
244
400
218
194
7
457
42
23
307
37
78
49
6
2055
8
82
47
39
5
6
10
6866
Mean
duration,
min
26.2
28.4
114.7
30.3
18.1
25.2
12.2
83.5
32.6
203.4
80.4
25.2
33.3
21.7
40.1
65.4
6.5
213.2
16.2
126.1
45.2
93.3
161.0
78.7
373.9
342'. 0
13.9
162.4
119.7
32.8
13.4
192.2
57.0
Mean
cone. ,
ppm
6.6
15.5
5.4
7.5
7.1
7.1
1.7
2.8
5.2
3.2
4.4
2.4
2.5
2.4
3.1
3.5
3.6
2.8
3.2
2.2
2.3
1.0
2.1
2.0
1.9
1.2
3.8
1.3
0.9
0.6
1.8
0.7
0.6
nrp
ucn
In /m 1
1VV
2.66e
2.26
4.96e
1.050
3.68e
0.90
2.14
'0.66Q
1.45e
1.85
1.70
0.72
1-58*
0.74e
1.07
1.21
2.04
0.94
1.20
1.91
0.90
2.05e
0.41
0.39
0.000
0.85!
2.91e
0.74
0.99
0.91
3.02
0.00
0.62
t statistic
Duration
means
0.03
-1.76
-1.60,
-4.06r
-1.04
0.99
-1.72f
2.16T
-0.63
-0.71
-0.23,
-3.44T
-0.82.
-2.20T
0.99,
2.69T
f
6.61,
-2.73
1.40
-1.23
1.39.
-2.24r
0.31
-2.88T
-0.90
0.73
-0.56.
-2.98T
-0.53
0.16
Cone.
means
1.34,
-2.88T
1.29
2.02
1.23f
6.56T
1.12,
4.271
7.60
4.46.
1.14
0.95.
3.30L
3.06T
1.80
0.13
1.23f
5.59T
1.66
*
2.99,
4.73J
2.74T
0.33
1.53
f
13. or
-1.36,
2.76T
1.13
0.67
-0.09
"includes D3 = bb, 01, and 02. bBlank. clncludes 03 = bb and 01. ' dlncludes 03 = bb, 01, 02, and 03.
eRatio Is significantly different from 1.0 at p = 0.05 level. Significant at p = 0.05 level.
-------
listed in the same order as they appear in Table 6-20 of Reference 1 (i.e.,
in descending order of weighted mean CO concentration as discussed in Section
6.7).
If there were no differences between Group H person-days and Group L
person-days with respect to the number of times a particular microenvironment
was occupied, then n,, would be expected to equal m,,, where
mH = (nH + n|.XNH)/(NH + NL). (4-1)
In this expression, NH equals 1189, the number of occupancy periods associated
with Group H person-days, and N. equals 6866, the number of occupancy periods
associated with Group L person-days. The ratio of n,, to m., is thus the ratio
of the number of observed events to the number of expected events. In the
discussion that follows, such ratios are referred to as observed-to-expected
ratios (OER's) and are calculated using the general expression
OER = x/m (4-2)
where x is the number of events observed and m is the number expected. Bailer
2
and Ederer provide tables for determining if an OER differs significantly
from unity for a given x.
Values for OER = nuMj are presented in Table 4-1 under the column head-
ing "OER (nH/m,,).M Nine of the OER's are flagged as significant at the
p = 0.05 level. The four largest flagged values correspond to the microen-
vironments labeled indoors-service station (4.96), outdoors-public garage
(3.68), outdoors-school grounds (2.91), and indoors-public garage (2.66).
Other microenvironments with flagged OER's greater than 1.0 are outdoors-
other location (2.05), outdoors-within 10 yards of road (1.58), and in
transit-truck (1.45). All of these microenvironments are associated with out-
door locations and/or motor vehicles.
Table 4-1 also lists t statistics by microenvironment for tests that
Group H and Group L duration means are equal and that Group H and Group L
concentration means are equal. The statistics were calculated assuming un-
equal variances; flagged t values indicate that the probability that the
42
-------
associated means are equal is 0.05 or less. A positive t value indicates
that the Group H mean exceeds the Group L mean. With respect to duration,
the only flagged positive t values are associated with outdoors-service
station (6.61), indoors-restaurant (2.69), and indoors-other location (2.16).
With respect to,concentration, there are 11 flagged positive t values in
Table 4-1. The seven largest flagged t values are associated with indoors-
residence (13.01), in transit-truck (7.60), in transit-car (6.56), indoors-
office (5.59), indoors-store (4.73), indoors-other location (4.27), and
outdoors-within 10 yards of road (3.30). Few of the t values are negative,
indicating that mean CO concentrations for most microenvironments were
higher for Group H person-days than for Group L person-days.
PEI also attempted to determine whether geographic location affects expo-
sure within a particular microenvironment. A file was compiled listing each
nontransit PEM value that has an activity diary code indicating it was
recorded in a census tract containing one of the 15 fixed-site monitors
operating during the study. Figure 3 in Appendix A shows the locations of
the 15 sites. To simplify the analysis, PEI divided the fixed sites into
three groups according to the median (50th percentile) 8-hour daily maximum
value reported during the study period at each site (Table 4-2). Note that
the same grouping would occur if the 90th percentile were used. Under
the headings NM and N. , Table 4-3 lists the number of PEM values recorded
on Group H and Group L person-days which fall into each combination of
microenvironment and site group. A total of 189 values were reported
for Group H person-days; 1798 PEM values were reported for Group L
person-days.
OER values were calculated for the various microenvironment-site
group combinations listed in Table 4-3 using Equation 4-1 with NH and N.
equal to the total PEM values in a particular site-group associated with
Group H and Group L, respectively. For example, the OER for indoors-
office/Site Group I is calculated as
OER = nH/mH
= (25)/[(25 + 81)(63)/(63 +. 761)]
to yield OER = 3.08. OER values that differ significantly from 1.0
at the p = 0.05 are flagged in the table. The largest flagged OER
43
-------
(6.56) is associated with Site Group III and the microenvironment labeled
"outdoors - within 10 yards of road." This is consistent with our
expectations that high exposures would occur near roadways in areas with
high ambient CO levels (i.e., the Group III areas) since it is likely
the high levels,are the result of heavy traffic.
TABLE 4-2. ASSIGNMENT OF DENVER FIXED-SITE MONITORS TO
SITE GROUPS I, II, AND III
Site group
I
II
III
SAROAD code
060080002F01
062080825 F05
062080823F05
060140002F01
062080817F05
062080822F05
062080819F05
062080818F05
062080824F05
060120002F01
062080821F05
062080820F05
060580013F01
060580014F01
060580002F01
Map code
M
H
G
K
0
E
J
I
N
L
D
F
C
B
A
Three-digit
code
HIG
825
823
AUR
817
822
819
818
824
ARV
821
820
013
014
002
Daily maximum
8-hour CO value, ppm
50th
percentile
1.3
2.1
2.6
2.7
. 2.7
3.4
4.2
4.2
4.4
4.5
4.5
4.9
5.7
5.9
6.8
90th
percentile
3.3
4.1
5.1
5.0
5.2
7.5
8.9
9.3
9.5
8.1
9.5
9.0
10.3
10.9
13.7
-------
TABLE 4-3. NUMBER OF PEM VALUES REPORTED FOR INDICATED COMBINATIONS OF MICROENVIRONMENT, EXPOSURE
GROUP, AND SITE GROUP (INCLUDES ONLY PEM VALUES RECORDED IN CENSUS TRACTS CONTAINING
A FIXED-SITE MONITOR)
B
52
54
72
71
62
55
58
51
07
bb
05
74
03
73
56
04
80
59
61
53
02
77
60
57
76
78
79
D3
a
bb
a
bb
bb
bb
bb
bb
c
bb
bb
bb
c
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
Microenvironment
Category
Indoors
Indoors
Outdoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Not specified
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Outdoors
Outdoors
Subcategory
Public garage
Service station or motor
vehicle repair facility
Public garage
Residential garage or
carport
Other location
Other repair shop
Shopping mall
Residential garage
Within 10 yards of road
Not specified
Restaurant
Service station or motor
vehicle repair facility
Office
Parking lot
Auditorium, sports arena,
concert hall , etc.
Store
Other location
Health care facility
Other public building
Manufacturing facility
Residence .'
School grounds
School
Church
Residential grounds
Sports arena, amphitheater,
etc.
Park or golf course
TOTAL
Site Group I
nH
0
0
0
0
0
0
0
0
0
0
2
0
25
1
0
10
0
0
0
0
25
0
0
0
0
0
0
63
nL
2
0
0
6
12
0
0
0
14
0
9
0
81
0
0
14
0
0
0
0
612
0
2
9
0
0
0
761
OER
0.00
-
-
0.00
0.00
-
-
-
0.00
-
2.38
n
3.086
-
_
5.45e
_
-
-
f\
0.516
0.00
0.00
-
-
-
Site Grou
nH
0
0
0
0
2
0
0
0
0
0
0
0
25
0
0
1
1
0
0
0
40
4
6
0.
0
0
0
79
nL
0
18
0
0
17
23
5
0
16
0
19
0
93
7
1
13
5
0
12
31
375
2
20
3
0
0
0
660
P H
OER
_
0.00
-
-
0.98
0.00
0.00
-
0.00
-
0.00
Łt
1.98e
0.00
0.00
0.67
1.56
-
0.00
0.00
0.90
6.24e
2.16
0.00
-
-
-
Site Grou
nH
0
0
0
0
0
0
0
0
8
0
5
0
14
1
0
16
0
0
0
0
3
0
0
0
0
0
0
47
"L
12
1
0
0
16
0
0
0
3
0
16
0
61
2
0
13
27
77
12
.24
78
0
34
0
1
0
0
377
p III
OER
0.00
0.00
-
-
0.00
-
-
A
6.56e
-
2.15
-
1.68
3.01
_
4.98e
0.00
0.00
0.00
0.00
0.33e
-
0.00
.-
0.00
-
-
-p.
en
'includes D3 = bb, 01, and 02. DBlank.
'Includes D3 = bb and 01.
•Observed-to-expected ratio (OER) is significantly different from 1.0 at p = 0.05 level.
Includes D3 = bb, 01, 02, and 03.
-------
There are many zeros in Table 4-3 due to the small number of PEM
values which were recorded in census tracts containing fixed-site monitors.
To increase the sample size, PEI prepared a file which matched each PEM
value in the SAMPLE-DATA file with the nearest fixed-site monitor and then
divided the PEM values into three groups based on the site group to which
the nearest fixed-site monitor was assigned. Table 4-4 lists the results
in a format similar to Table 4-3. OER's are calculated in a similar
manner. Combinations of microenvironments and site groups with OER's which
are positive and significant (p < 0.05 level) are listed in Table 4-5.
Significantly, the indoor-public garage and indoor-service station
microenvironments have very large ratios for Site Group I, the group
associated with low ambient CO levels. Apparently CO sources within these
microenvironments are associated with high exposures even in the absence of
high fixed-site monitor readings.
Associated with each in-transit PEM value are two census tract
listings, one associated with the start address and the other with the end
address. Realizing that neither may be a good indicator of the CO
conditions encountered during the trip, PEI prepared two tables based on
these data. Table 4-6 pairs each in-transit PEM value with the fixed-site
nearest to the start census tract. Table 4-7 pairs each in-transit PEM
value with the fixed-site nearest to the end census tract. The format of
each of these tables is similar to that of Table 4-4 and OER values are
calculated in a similar manner. The in-transit microenvironments are
listed in the same order as Table 6-26 in Reference 1 with the addition of
the bicycle microenvironment.
OER values significant at the p = 0.05 level are flagged in both
tables. Surprisingly, only the truck microenvironment is flagged for Site
Group III in the two tables. The largest car OER is only 1.01 and four of
the six car ratios in the two tables are less than 1.00. The combinations
of microenvironment and site group with the three largest significant OER
values in Table 4-6 are motorcycle - Site Group I (4.66), truck - Site
Group III (2.27, and truck - Site Group II (1.45). Only two combinations
have significant OER values in Table 4-7: motorcyle - Site Group I (4.52)
and truck - Site Group III (2.46).
46
-------
TABLE 4-4. NUMBER OF PEM VALUES REPORTED FOR INDICATED COMBINATIONS OF MICROENVIRONMENT, EXPOSURE
GROUP, AND SITE GROUP (PEM VALUES ARE MATCHED TO NEAREST FIXED-SITE MONITOR)
B
52
54
72
71
62
55
58
51
07
bb
05
74
03
73
56
04
80
59
61
53
02
77
60
57
76
78
79
D3
a
bb
a
bb
bb
bb
bb
bb
c
bb
bb
bb
c
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
Microenvironment
Category
Indoors
Indoors
Outdoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Not specified
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Outdoors
Outdoors
Subcategory
Public garage
Service station or motor
vehicle repair facility
Public garage
Residential garage or
carport
Other location
Other repair shop
Shopping mall
Residential garage
Within 10 yards of road
Not specified
Restaurant
Service station or motor
vehicle repair facility
Office
Parking lot
Auditorium, sports arena,
concert hall , etc.
Store
Other location
Health care facility
Other public building
Manufacturing facility
Residence
School grounds
School
Church
Residential grounds
Sports arena, amphitheater,
etc.
Park or golf course
TOTAL
Site Group I
nH
43
28
2
0
24
0
2
5
68
0
30
2
57
9
19
52
9
13
0
0
1154
0
9
6
2
0
0
1534
nL
19
10
0
13
142
0
25
35
159
0
199
5
726
36
33
297
42
61
29
0
10064
3
147
64
34
4
10
12157
OER
6.19e
6°58P
8.93e
0.00
1.29
0.66
1.12
2.67e
-
1.17
2.55
0.65e
1.79
o
3.26e
1.33
1.58
1.57
0.00
A
0.92e
o.oo
0.516
0.77
0.50
0.00
0.00
Site Grou
nH
6
63
9
6
29
14
9
0
32
0
76
1
143
3
5
26
29
1
1
0
1104
7
22
9
5
0
0
1600
\
12
25
13
2
159
.35
13
14
188
2
137
5
709
15
28
206
39
70
43
48
7667
7
179
61
31
12
4
972T
P n
OER
2.36
5'07p
2.90e
A
5.31e
1.09
2«02e
2.90e
0.00
1.03
0.00
2.53e
1.18
1.19
1.18
1.07
0.790
3.02*
o.io*
0.16e
0.00
0.89f
3.54e
0.77
0.91
0.98
0.00
0.00
Site group III
nH
2
0
3
'l
22
0
5
0
28
0
19
1
98
1
0
46
2
9
5
0
386
0
0
5
1
0
0
634
nL
43
2
4
4
103
6
4
16
98
0
128
0
828
15
22
178
35
223
35
25
3377
0
76
43
7
0
4
5276
OER
0.41
0.00
4.00
1.86
1.64e
0.00
5.18e
0.00
2.07e
-
1.20
9.32
0.99
0.58
0.00Q
1.916
0.500
0.36e
1.17
0.00
0.96
-
0.00
0.97
1.17
-
0.00
Includes D3 = bb, 01, and 02. DBlank. Includes D3 = bb and 01. Includes D3 = bb, 01, 02, and 03.
?0bserved-to-expected ratio (OER) is significantly different from 1.0 at p = 0.05 level.
-------
TABLE 4-5. COMBINATIONS OF MICROENVIRONMENTS AND SITE GROUPS WITH
SAMPLE SIZES EXCEEDING FIVE AND SIGNIFICANT
OBSERVED-TO-EXPECTED RATIOS IN TABLE 4-4 EXCEEDING 1.00
Code
52a
54bb
72a
71bb
62bb
58bb
07c
05bb
56bb
04bb
80bb
77bb
Microenvironment
Indoors-public garage '
Indoors-service station or motor
vehicle repair facility
Outdoors-public garage
Outdoors-residential garage or carport
Indoors-other location
Indoors-shopping mall
Outdoors-withln 10 yards of road
Indoors-restaurant
Indoors-auditorlum, sports arena,
concert hall , etc.
Indoors - store
Outdoors-other location
Outdoors - school grounds
Site group
I
I
II
II
II
III
II
III
I
III
II
I
III
II
II
Sample size
62
38'
88
22
8
125
22
9
. 227
126
213
52
224
68
14
OER
6.19
6.58
5.07
2.90
5.31
1.64
2.90
5.18
2.67
2.07
2.53
3.26
1.91
3.02
3.54
'includes D3 = bb, 01, and 02.
5Blank.
"Includes D3 = bb and 01.
48
-------
TABLE 4-6. NUMBER OF "START" IN-TRANSIT PEM VALUES REPORTED FOR INDICATED COMBINATIONS OF
MICROENVIRONMENT, EXPOSURE GROUP, AND SITE GROUP (PEM VALUES ARE MATCHED TO NEAREST
FIXED-SITE MONITOR)
Code
B
01
01
01
01
01
01
D3
93
03
01
04
02
92
In-transit
subcategory
Motorcycle
Bus
Walking
Truck
Car
Bicycle
TOTAL
Site grou
nH
11
1
10
35
207
0
264
\
8
22
107
191
1529
2
1859
p I
OER
4.66a
0.35
0.69
1.25a
0.96
0.00
Site group II
nH
0
6
39
36
206
4
291
nL
5
21
215
123
1199
7
1570
OER
0.00
1.42
0.98
1.45a
0.94
2.33
Site group III
nH
0
4
27
19
101
0
151
nL
2
38
276
50
725
3
1094
OER
0.00
0.79
0.73
2.27a
1.01
0.00
Observed-to-expected ratio (OER) is significantly different from 1.0 at p = 0.05 level.
-------
TABLE 4-7. NUMBER OF "END" IN-TRANSIT PEM VALUES REPORTED FOR INDICATED COMBINATIONS OF
MICROENVIRONMENT, EXPOSURE GROUP, AND SITE GROUP (PEM VALUES ARE MATCHED TO NEAREST
FIXED-SITE MONITOR)
Code
B
01
01
01
01
01
01
D3
93
03
01
04
02
92
In-transit
subcategory
Motorcycle
Bus
Walking
Truck
Car
Bicycle
TOTAL
Site group I
"H
9
1
11
36
209
0
266
nL
7
21
107
190
1547
1
1873
OER
4.52a
0.37
0.75
1.28
0.96
0.00
Site grou
nH
2
8
35
35
207
4
291
"L
5
18
215
129
1173
9
1549
3 II
OER
1.81
1.95
0.89
1.35
0.95
1.95
Site group III
nH
0
2
30
19
101
0
152
nL
3
42
278
45
737
2
1107
OER
0.00
0.38
0.81
2.46a
1.00
0.00
in
O
Observed-to-expected ratio (OER) is significantly different from 1.0 at p = 0.05 level.
-------
Another exploratory analysis investigated whether high exposures
occurred more often on days when ambient CO levels were high. Table 4-8 lists
the calendar days in the Denver study monitoring period and indicates the
number of Group H and Group L person-days associated with each calendar day.
A person-day is-associated with a calendar day if the daily maximum 8-hour
exposure of the person-day begins with an hour that falls in the calendar day.
Also listed for each calendar day is an OER value (x is the number of Group
H person-days observed and m is the number of Group H persondays expected)
and the daily average and daily maximum 8-hour values reported by the
"composite" site. Two calendar days have significant (p < 0.05) OER values:
November 5 and December 9. Daily average values at the composite site
were quite high on these days (5.87 ppm and 6.32 ppm, respectively).
Each OER in Table 4-8 was paired with the corresponding daily average
value and daily maximum 8-hour value, and the Spearman rank correlation test
was performed. Both tests yielded a Spearman rho statistic which was positive
and significant at the p = 0.001 level (one-sided test). This result suggests
that the ratio is generally larger on days with high ambient CO concentrations.
This in turn implies that person-days in Group H are strongly associated with
days with high ambient CO concentration.
In an attempt to identify high exposure occupations, PEI determined
the number of person-days in Groups H and L which were reported by persons
with each three-digit occupation code used by the Bureau of Census. The
results are listed in Table 4-9. OER values were calculated for each
occupation and those found to be significant (p < 0.05) are listed in the
table. Only three occupations are associated with significant OER values:
Code 139 - education teachers (OER = 5.84), Code 243 - supervisors and
proprietors, sales occupations (OER = 6.67), and Code 877 - stock handlers
and baggers (OER = 5.84). In evaluating these results, it should be noted
that some of the person-days assigned occupation categories in Table 4-9
may be nonwork days and thus do not represent occupation-related exposures.
At this early stage in the analysis, no attempt was made to adjust for this
potential bias.
51
-------
TABLE 4-8. NUMBER OF DAILY MAXIMUM 8-HOUR VALUES OCCURRING ON
INDICATED DATE BY EXPOSURE GROUP
Date
11-1
11-2
11-3
11-4
11-5
11-6
11-9
11-10
11-11
11-12
11-13
11-14
11-15
11-16
11-17
11-18
11-19
11-20
11-21
11-22
11-23
11-29
11-30
12-1
12-2
12-3
12-4
12-5
12-6
12-7
12-8
12-9
12-10
12-11
12-12
12-13
12-14
12-15
12-16
12-17
12-18
Number of daily maximum
8-hour values beginning
on indicated date
Group^H
0
1
1
2
5
1
0
4
0
1
0
0
2
2
1
2
1
0
0
0
1
0
1
0
0
1
4
0
1
1
0
6
2
1
1
1
0
0
2
0
0
Group L
1
11
6
6
7
7
2
9
1
14
10
5
4
10
5
1
4
13
4
7
2
2
4
5
2
8
12
1
14
2
6
6
0
12
4
7
3
6
8
6
2
OER
0.00
0.65
1.11
1.95a
3.24a
0.97
0.00
2.40
0.00
1.56
0.00
0.00
2.60
1.30
1.30
5.19
1.56
0.00
0.00
0.00
2.60
0.00
1.56
0.00
0.00
0.87
1.95
0.00
0.52
2.60
0.00a
3.89a
7.79
0.60
1.56
0.97
0.00
0.00
1.56
0.00
0.00
Composite fixed-site
concentration for date
Daily
average
2.48
2.25
1.91
2.85
5.87
3.37
3.17
5.94
1.39
2.34
2.91
1.76
3.77
4.17
3.85
4.07
2.68
2.32
1.73
1.87
1.76
2.75
2.33
2.58
2.76
3.15
3.20
1.87
3.46
0.92
1.42
6.32
1.60
2.42
3.33
4.27
1.44
2.06
5.72
4.00
1.20
Daily
maximum 8-h
3.6
2.9
2.5
4.9
7.4
7.1
6.4
7.2
1.6
5.3
5.0
3.9
5.8
6.1
5.3
5.8
4.7
5.0
2.9
3.2
3.7
4.5
4.5
3.6
3.6
5.6
6.2
3.1
5.3
1.3
2.0
9.2
2.1
5.3
6.3
8.6
2.2
3.6
9.1
7.1
1.7
(continued)
52
-------
TABLE 4-8 (.continued)
Date
12-19
12-20
12-21
12-22
12-23
1-5
1-6
1-7
1-8
1-9
1-10
1-11
1-12
1-13
1-14
1-15
1-16
1-17
1-18
1-19
1-20
1-21
1-22
1-24
1-25
1-26
1-27
1-28
1-29
1-30
1-31
2-1
2-2
2-3
2-4
2-5
2-6
2-7
2-8
2-9
2-10
NumBer of datly maximum
8-hour values beginning
on indicated date
Group H
0
0
0
0
0
0
1
4
0
0
1
0
1
0
2
4
1
4
1
3
1
1
2
0
1
2
4
0
4
0
0
0
0
2
2
2
1
1
0
0
1
Group L
3
9
4
3
2
3
8
11
9
2
6
10
14
7
6
10
3
5
6
8
3
7
9
7
7
13
8
8
8
3
6
9
13
8
14
4
6
1
11
6
7
OER
0.00
0.00
0.00
0.00
0.00
0.00
0.87
2.08
0.00
0.00
1.11
0.00
0.52
0.00
1.95
2.22
1.95
3.46
1.11
2.12
1.95
0.97
1.42
0.00
0.97
1.04
2.60
0.00
2.60
0.00
0.00
0.00
0.00
1.56
0.97
2.60
1.11
3.89
0.00
0.00
0.97
Composite fixed-site
concentration for date
Daily
average
1.44
4.14
3.09
3.02
1.55
2.58
2.83
4.25
2.33
0.60
0.89
2.39
4.77
2.61
2.60
6.09
2.23
4.20
3.46
5.00
3.79
2.45
2.90
2.57
• 3.04
3.60
5.40
2.09
1.67
2.01
1.34
1.13
1.74
1.62
2.70
1.82
1.15
1.81
3.17
2.06
1.43
Daily
maximum 8-h
1.9
6.1
4.7
4.1
3.0
3.9
4.5
5.6
4.0
0.9
1.2
3.9
7.0
3.5
3.4
10.3
4.2
5.5
5.2
6.9
6.0
2.9
3.7
4.6
4.7
6.3
8.1
4.0
3.6
2.9
1.7
1.5
3.2
2.5
3.7
2.5
1.5
3.0
4.7
3.0
2.4
(continued)
53
-------
TABLE 4-8 (continued)
Date
2-11
2-12
2-13
2-14
2-15
2-16
2-17
2-18
2-19
2-20
2-21
2-22
2-23
2-24
2-25
2-26
2-27
2-28
Number of dally maximum
8-hour values beginning
on Indicated date
Group H
0
1
2
1
0
0
2
2
0
0
2
0
0
0
0
1
1
0
Group L
14
5
13
4
10
11
12
13
3
10
9
10
12
9
10
16
3
9
OER
0.00
1.30
1.04
1.56
0.00
0.00
1.11
1.04
0.00
0.00
1.42
0.00
0.00
0.00
0.00
0.46
1.95
0.00
Composite fixed-site
concentration for date
Dally
average
2.10
1.87
2.13
1.12
1.95
2.85
1.86
1.91
0.83
0.85
1.90
2.07
2.41
1.95
2.24
1.58
1.22
2.06
Dally
maximum 8-h
3.4
3.1
3.3
1.5
2.9
4.6
2.5
3.0
1.2
1.8
3.5
2.7
3.7
2.4
3.3
2.6
1.5
3.4
TOTAL
103
699
Observed-to-expected ratio (OER) is significantly different from
1.0 at p = 0.05 level.
54
-------
TABLE 4-9. NUMBER OF PERSON-DAYS IN OCCUPATIONAL CATEGORIES USED BY
U.S. BUREAU OF CENSUS BY EXPOSURE GROUP
Code
005
007
008
009
013
014
015
016
019
023
025
026
027
029
037
055
059
064
065
067
075
077
078
084
095
096
097
124
137
138
139
143
148
155
156
157
158
Occupational category
Administrators and officials, public admini-
stration
Financial managers
Personnel and labor relations managers
Purchasing managers
Managers, marketing, advertising, and public re-
lations
Administrators, education
Managers, medicine and health
Managers, properties and real estate
Managers and administrators, not elsewhere coded
(n.e.c.)
Accountants and auditors
Other financial officers
Management analysts
Personnel, training, and labor relations
specialists
Buyers, wholesale and retail, except farm
products
Management-related occupations, n.e.c.
Electrical and electronic engineers
Engineers, n.e.c.
Computer systems analysts and scientists
Operations and systems researchers and analysts
Statistician
Geologists and geodesists
Agricultural and food scientists
Biological and life scientists
Physicians
Registered nurses
Pharmacists
Dietitians
Political science teachers
Art, drama, and music teachers
Physical education teachers
Education teachers
English teachers
Trade and industrial teachers
Teachers, prekindergarten and kindergarten
Teachers, elementary school
Teachers, secondary school
Teachers, special education
Number of
person-days
Group H
0
0
0
1
'l
0
0
0
3
0
1
0
0
0
1
0
0
1
0
1
0
0
0
1
3
0
0
0
0
2C
3C
0
0
0
2
1
0
Group L
4
3
2
0
6
4
9
5
17
9
3
3
2
2
3
4
1
2
2
1
6
2
2
11
15
5
2
2
2
4
1
1
2
4
27
3
2
(continued)
55
-------
TABLE 4-9 (continued).
Code
159
163
164
173
174
176
178
185
186
189
195
198
203
207
208
213
216
228
229
235
243
253
254
255
257
259
263
268
274
275
276
277
278
303
309
Occupational category
Teachers, n.e.c.
Counselors, educational and vocational
Librarians
Urban planners
Social workers
Cl ergy
Lawyers
Designers
Musicians and composers
Photographers
Editors and reporters
Announcers
Clinical lab technologists and technicians
Licensed practical nurses
Health technologists and technicians, n.e.c.
Electrical and electronic technicians
Engineering technicians, n.e.c.
Broadcast equipment operators
Computer programmers
Technicians, n.e.c.
Supervisors and proprietors, sales occupations
Insurance sales occupations
Real estate sales occupations
Securities and financial services sales
occupations
Sales occupations, other business services
Sales representatives, mining, manufacturing and
wholesale
Sales workers, motor vehicles and boats
Sales workers, hardward and building supplies
Sales workers, other commodities
Sales counter clerks
Cashiers
Street and door-to-door sales workers
News vendors
Supervisors, general office
Peripheral equipment operators
Number of
person-days
Group H
0
1
0
0
0
'0
3
0
0
0
0
0
0
1
0
0
0
0
1
0
6d
0
0
0
0
0
0
2
0
0
0
0
0
0
0
Group L
12
5
4
1
5
2
11
4
2
1
4
1
3
4
1
5
4
2
4
2
1
4
3
2
2
5
2
0
12
2
1
2
2
2
4
(continued)
56
-------
TABLE 4-9 (continued}
Code
313
315
319
323
336
337
354
357
375
376
377
378
379
385
406
418
426
427
434
435
436
444
446
447
449
453
467
468
495
503
506
507
508
509
514
(contii
Occupational category
Secretaries
Typists
Receptionists
Information clerks, n.e.c.
Records clerks
Bookkeepers, accounting, and auditing clerks
Postal clerks, except mail carriers
Messengers
Insurance adjusters, examiners, and investigators
Investigators and adjusters, except insurance
Eligibility clerks, social Welfare
Bill and account collectors
General office clerks
Data-entry keyers
Child care workers
Police detectives, public service9
Guards and police, except public service
Protective services occupations, n.e.c.
Bartenders3
Waiters and waitresses
Cooks, except short order
Miscellaneous food preparation occupations
Health aides, except nursing
Nursing aides, orderlies, and attendants
Maids and housement
Janitors and cleaners
Homemakers
Child care workers
Forestry workers, except logging
Supervisors, mechanics, and repairers3
Auto mechanic apprentices
Bus, truck, and stationary engine mechanics
Aircraft engine mechanics
Small engine repairers3
Auto body and related repairers3
iued)
Number of
person-days
Group H
1
3
0
0
0
1
0
•0
0
2
0
0
5
0
0
0
1
0
0
0
3
2
0
0
0
1
17
0
0
1
0
2
0
0
1
Group L
23
8
2
1
5
4
2
2
1
2
2
2
21
2
2
1
3
4
2
2
3
0
1
4
7
4
150
4
2
0
1
0
1
2
0
57
-------
TABLE 4-9 (.continued)
Code
516
518
529
534
535
536
539
555
558
567
577
588
599
617
633
637
644
657
677
678
696
734
736
743
744
763
765
774
777
779
787
796
804
805
856
Occupational category
Heavy -equipment mechanics3
Industrial machinery repairers
Telephone Installers and repairers
Heating, air conditioning, and refrigeration
mechanics
Camera, watch, and musical Instrument repairers
Locksmiths and safe repairers
Mechanical controls and valve repairers
Supervisors, electricians and power transmission
Installers
Supervisors, construction occupations, n.e.c.
Carpenters
Electrical power installers and repairers
Concrete and terrazzo finishers
Construction trades, n.e.c.
Mining occupations, n.e.c.
Supervisors, production occupations
Machinists
Precision grinders, fitters, and tool sharpeners
Cabinet makers and bench carpenters
Optical goods workers
Dental lab and medical appliance technicians
Stati onary engi neers
Printing machine operators
Typesetters and compositors
Textile cutting machine operators
Textile sewing machine operators
Roasting and baking machine operators, food
Folding machine operators
Photographic process machine operators
Miscellaneous machine operators, n.e.c.
Machine operators, not specified
Hand molding, casting, and forming occupations
Production inspectors, checkers, and examiners
Truck drivers, heavy3
Truck drivers, light
Industrial truck and tractor equipment
operators
Number of
person-days
Group H
0
0
0
0
0
1
2
t
1
1
1
1
0
0
0
0
0
0
0
0
0
0
1
2
0
0
0
0
0
1
0
0
1
0
2
0
Group L
2
2
3
1
0
0
0
0
3
7
1
1
2
2
4
2
2
2
2
2
2
1
0
0
2
2
2
2
0
2
2
1
3
4
2
(continued)
58
-------
TABLE 4-9 (continued)
Code
869
877
885
888
889
998b
999°
Blank
Occupational category
Construction laborers
Stock handlers and baggers
Garage and service station related occupations
Hand packers and' packagers
Laborers, except construction
Not employed
Retired
Total person-days
Number of
person-days
Group H
2.
3C
2
0
0
5
1
1
103
Group L
4
1
1
2
0
19
37
8
699
Potential for high occupation-related exposure to CO because of proximity
to motor vehicles, gas appliances, or cigarette smokers.
'Unofficial code used by PEL
:OER =5.84 (significant at p = 0.05 level).
OER = 6.67 (significant at p = 0.05 level).
Table 4-10 lists the number of person-days by exposure group for
selected aggregate occupation categories which might be expected to have
higher CO exposures because of proximity to motor vehicles or gas
appliances. Only the aggregate category "work which may involve proximity
to running motor vehicles or internal combustion engines in enclosed space"
has a significant (p < 0.05) OER value.
59
-------
TABLE 4-10. NUMBER OF PERSON-DAYS IN SELECTED AGGREGATE
OCCUPATION CATEGORIES BY EXPOSURE GROUP
Aggregate occupational category
1. Work may require travel in
vehicles in addition to
commuting
2. Work may involve proximity to
running motor vehicles or
internal combustion engines
in enclosed space
3. Work may involve proximity to
street
4. Work may involve proximity to
gas appliances (excluding
homemakers)
Codes
253,
357,
804,
503,
508,
516,
277,
434,
763
263,
418,
805,
506,
509,
885
357
435,
277,
529,
856
507,
514,
444,
Person-days
Group H
2
6
0
2
Group L
23
7
•
4
6
OER
0.62
•*
3.59a
0.00
1.95
OER value is significantly different from 1.0 at p * 0.05 level.
The following tentative conclusions are suggested by the exploratory
data analyses discussed above.
1. Person-days in Group H exhibit higher CO levels in most micro-
environments.
2. The mlcroenvironments which were visited more often during
Group H person-days than would be expected are all associated
with either outdoor locations and/or motor vehicles. Indoors-
service station and outdoors - public garage have particularly
large OER values.
3. The average durations of visits to the outdoors-service
station and indoor-restaurant mlcroenvironments are larger
for Group H person-days than for Group L person-days.
4. The microenvironment "outdoors - within 10 yards of road" is
associated with Group H person-days when it is located in a area
with high ambient CO levels.
5. The microenvironments "indoor - public garage" and "indoor - service
station" are associated with Group H person-days in areas
with relatively low ambient CO levels.
6. Of the in-transit microenvironments, only truck is associated with
Group H person-days when the start or end location of the trip is
in a high ambient CO location. "Motorcycle" is associated with
Group H for low ambient CO locations.
60
-------
7. Group tt person-days are strongly associated with periods of high
ambient CO concentration, particularly November 5 and December 9,
1982.
8. The aggregate occupation category "work which may involve
proximity to running motor vehicles or internal combustion
engines in enclosed space" is strongly associated with Group H
person-days.
Although the determination that a strong association exists is
supported by a statistical significance level for each of the above
conclusions, the conclusions should be considered as tentative where
sample sizes are small or where confounding factors are likely to exist.
More detailed analyses of the factors associated with high in-transit and
high indoor exposures are described in Sections 4.2 and 4.3.
4.2 IN-TRANSIT EXPOSURES
Under the direction of EMSL, PEI investigated the effects of transit
mode, smoking, time of day, and duration on in-transit exposures among the
subjects of the Denver study. As in previous analyses, PEI assumed that
personal GO exposures were accurately represented by the PEM values obtained
for each subject. Mode of transit, smoking status, time of day, and
duration were determined from the activity diary entries associated with
each PEM value. Since PEI had files containing these data for Washington
as well as Denver, a less extensive analysis of the Washington data was
also performed.
4.2.1 General Approach and Notation
The general approach to analyzing the data was to perform analyses of
variance (ANOVA) with the PEM values as the response variable and the remaining
variables as factors which may explain variations in the PEM variable. Mode
of transit, smoking status, and time of day were considered categorical varia-
bles. Duration was considered a continuous variable. Table 4-11 lists the
variables considered in the analyses and indicates how the activity diary
61
-------
TABLE 4-11. VARIABLES CONSIDERED IN ANALYSES
OF VARIANCE AND COVARIANCE
Variable
CO concentration recorded
by PEM, ppm
Mode of transit, Denver
Mode of transit, Washington
Smoking status
Time of day
Duration, minutes
Abbreviation
PEM
MDTR
MDTR
SM
TM
DR
Category
a
walk
car
other
walk
car
" other
yes
no
6-9
9-4
4-7
7-mid
b
c
Responses in category
01: walking
02: car
03: bus
04: truck
93: motorcycle
100: walking
661: jogging
200: car
300: bus
400: truck
500: train/ subway
664: van
Smokers present
Uncertain
Smokers not .present
6:00 < time < 9:00
-9:00 < time < 16:00
16:00 < time < 19:00
19:00 < time < 24:00
24:00 < time < 6:00
Continuous variable.
}No category.
%
'Continuous variable: 0 < DR < 60.
responses were grouped into categories. The number of categories developed
for each variable is smaller than the number of distinct activity diary responses
to avoid the occurrence of empty cells in the ANOVA layout.
The mode-of-transit (MDTR) categories were selected so that walking/
jogging and car would be distinct categories. The remaining category contains
all other modes involving motor vehicles or trains. In developing the two
smoking status (SM) categories, the "uncertain" response was combined with the
62
-------
"smokers present" response after noting that the distribution of "uncertain"
PEM values was statistically more similar to the "smokers present" distribu-
tion than the "smokers not present" distribution. The time-of-day (TM)
categories were selected so that the rush-hour periods of 6 a.m. to 9 a.m.
and 4 p.m. to 7 p.m. would fall into distinct categories. The other categories
correspond to late morning through early afternoon and post rush hour through
midnight. The period from midnight through 6 a.m. was not assigned a category
because of the small number of intransit PEM values which were recorded during
these hours.
The ANOVA's were performed using BMDP program P2V. The Denver data set
contained 4094 valid in-transit PEM values; the Washington data set contained
3176 valid in-transit PEM values. Cell sizes were unequal. The program
provided statistical tests for null hypotheses such as "the means of PEM
values corresponding to different transit modes are equal."
All analyses were performed on PEM values transformed using the Box-
Cox transformation
y - (xX - 1)A, (4-3)
where x was the PEM value and X was set equal to 0.25. A preliminary
analysis suggested that setting X equal to 0.25 would significantly
reduce the skewness of the empirical distributions.
Two approaches were used to evaluate the effect of exposure duration.
First, duration (DR) was used as a regressor in an analysis of covariance
using MDTR, SM, and TM as factors. Second, DR was used as a cell weight in
a weighted three- factor ANOVA. Note that duration values pertain to indi-
vidual PEM readings and do not necessarily equal total "trip" time.
4.2.2 Statistical Procedures
Three-Way ANOVA and Interations--
A three-factor, fixed-effect linear model was assumed for the initial
analysis of variance. The response variable Y was exposure; the factors were
MDTR, TM, and SM. The method of incorporating the fourth variable, DR, in
the model, is discussed below.
63
-------
The general equation for the linear model is
Yijkm * yijk + eijkm
where
y... » cell mean
ijk ,
2
eijkm * ^d N(0»a ' random variable
i = 1,... ,a
k = 1 ..... c
m = 1,. . . ,n.
In general, the number of observations n is different for each unique
set of values of the indices i, j, and k. The cell mean y. .. is expanded as
follows:
*ijk = °1 + ej + 6k (4'5)
+ <<*>ij + (W)jk + (6a)ki
+ (a66)ijk •
Here y... is the overall mean; a., B . , and 6. are the main effects; (08)..^.
(86)^k and (6a). . are the two-factor interactions, and (a86).--k is the three-
factor interaction. So-called Z restrictions are imposed by the BMDP routine
to remove rank deficiency.
In a two-factor study, the interaction of the i-th level of factor A with
the j-th level of factor B is defined as the difference between the cell mean
y. .. and the value y. . + a. + 8. which would be expected if the two factors
were additive; in other words,
(08)^ ^yy-j - (y.. + a. + 3^. (4-6)
64
-------
In a three-factor study the three-factor interaction (aS<5).j -k is defined as
the difference between the cell mean y... and the value that would be ex-
pected if main effects plus two-factor interactions were sufficient to account
for all factor effects; thus
(°36)1jk ^y1jk - [y... + a. + 31 + \ + (ctf)^. + (06)Jk + (6o)k1l. (4-7)
In the special case where cell sizes are equal (e.g., in a "designed"
study), the total sum of squares
a b c n 9
Z Z Z 'Z (Y... - Y....r (4-8)
1=1 j=l k=l m=l 1Jkm
can be expanded as a sum of component sums-of-squares. The components include
main effects, two-factor interactions, and three-factor interactions. Orthogo-
nality exists and there is additivity of effects.
In the Denver and Washington data sets the cell sizes are generally un-
equal; thus, orthogonality is lost and the effects are no longer additive.
The main consequences of the loss of additivity is that it becomes more
difficult to test hypotheses concerning the presence of specified component
effects. It is still possible to test for equality of means across levels of
a given factor. The approach is to first test for significance of two-factor
interactions. If these interactions are not significant, the test for equality
of means across levels of each of the three factors can proceed either directly,
or by first reducing the model to a no-interaction model. On the other hand,
if two-factor interactions are significant, tests for equality of level means
of one factor can proceed only if one conditions on each level of the other
factor. This approach assumes that the three-factor interaction is not signifi-
cant. If the three-factor interaction is significant, then the interactions
between any two factors need to be studied separately for each level of the
third factor.
Analysis of Covariance--
In covariance analysis one uses the relationship between the response
variable and a regressor, or explanatory variable, to improve the model fit,
G5
-------
I.e., to reduce the experimental error. Although one is usually not concerned
with a test of the hypothesis that the regression coefficient equals zero, it
was of interest in this study to discern a relationship between exposure and
duration of exposure.
Weighted ANOVA—
As an alternative to the use of duration as a concomitant variable in the
analysis of covariance, a weighted ANOVA was implemented with weight equal to
duration.
4.2.3 Results
As indicated in the introduction, data were available for both Denver,
Colorado and Washington, D.C. The results for Denver are more extensive
and are given in Tables 4-12 through 4-17 and in Tables 4-21 through 4-23.
The results for Washington, D.C. are given in Tables 4-18 through 4-20.
All results are in the form of standard ANOVA tables. Tables 4-12 and 4-13
are unweighted ANOVA1s with and without the variable DR as covariable, or
regressor. Results for the weighted ANOVA, with weight * DR, are presented
in Table 4-14. Note that all results apply to transformed (X = 0.25) PEM
values.
In Tables 4-15, 4-16, and 4-17 are presented the results of two-factor
ANOVA's conditioned on the levels of the third factor. Conditioning factors
are TM, MDTR, and SM, respectively.
In Tables 4-18, 4-19, and 4-20 are presented the results for Washington,
D.C. Tables 4-18 and 4-19 are unweighted ANOVA1s with and without the varia-
ble DR as covariable. Table 4-20 is the weighted ANOVA with weight equal to
DR.
Tables 4-21, 4-22, and 4-23 present the Denver cell means, the cell
standard deviations, and the cell counts for each of the possible cross-
tabulations, conditioned on the alternate factor.
The goal of this analysis is to test for equality of means of the main
effects. Such a test can be done provided interactions between the factor of
interest and any other factor in the study are not significant. In the
presence of interaction effects the test for equality of means of the main
effects can still be done provided one conditions on the interacting factor(s).
66
-------
TABLE 4-12. ANALYSIS OF VARIANCE TABLE FOR DENVER PERSONAL EXPOSURE
DATA (CELL WEIGHT = 1, COVARIABLE = NONE)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Error
Sums of
squares
138.06
91.36
4.44
20.30
0.52
52.47
90.88
20817.20
Degrees
of freedom
2
3
1
6
2
3
6
4070
Mean
square
69.03
30.45
4.44
3.38
0.26
17.49
15.15
5.1148
F
13.50
5.95
0.87
0.66
0.05'
3.42 •
2.96
Tail
prob.
0.0000
0.0005
0.3517
0.6808
0.9503
0.0167
0.0069
TABLE 4-13.
ANALYSIS OF VARIANCE TABLE FOR DENVER PERSONAL EXPOSURE DATA
(CELL WEIGHT = 1, COVARIABLE = DR)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Covariable3
Error
Sums of
squares
150.08
87.59
3.42
21.88
0.28
54.35
93.50
35.80
20781.40
Degrees
of freedom
2
3
1
6
2
3
6
1
4069
Mean
square
75.04
29.20
3.42
3.65
0.14
18.12
15.58
35.80
5.11
F
14.69
5.72
0.67
0.71
0.03
3.55
3.05
7.01
Tail
prob.
0.0000
0.0007
0.4129
0.6383
0.9727
0.0139
0.0056
0.0081
Regression coefficient estimate = -0.0063.
67
-------
TABLE 4-14. ANALYSIS OF VARIANCE TABLE FOR DENVER PERSONAL EXPOSURE DATA
(CELL WEIGHT = DR, COVARIABLE = NONE)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Error
Sums of
squares
2064.76
1139.51
121.37
688.57
577.74
43.04
1202.91
340554.16
Degrees
of freedom
2
3
1
6
2
3
6
3993
Mean
square
1032.38
379.84
121.37
114.76
288.87
14.35
200.49
85.29
F
12.10
4.45
1.42
1.35
3.39
0.17
2.35
Tail
prob.
0.0000
0.0040
0.2330
0.2331
0.0339
0.9179
0.0287
TABLE 4-15. ANALYSIS OF VARIANCE TABLE FOR DENVER
PERSONAL EXPOSURE DATA (CONDITIONED ON TM)
Conditioning
factor,
TM
6-9
9-16
16-19
19-24
Source
MDTR
SM
MDTR*SM
ERROR
MDTR
SM
MDTR*SM
ERROR
MDTR
SM
MDTR*SM
ERROR
MDTR
SM
MDTR*SM
ERROR
Sum of
squares
18.51
2.87
6.76
20817.20
109.97
12.32
71.75
20817.20
92.43
31.91
45.91
20817.20
34.56
7.87
7.69
20817.20
Degrees
of freedom
2
1
2
4070
2
1
2
4070
2
1
2
4070
2
1
2
4070
Mean
square
9.25"
2.87
3.38
5.11
54.98
12.32
35.87
5.11
46.21
31.91
22.96
5.11
17.28
7.87
3.84
5.11
F
1.81
0.56
0.66
10.77
2.41
7.02
9.05
6.25
4.49
3.38
1.54
0.75
Tail
prob.
0.1634
0.4532
0.5159
0.0000
0.1205
0.0009
0.0001
0.0125
0.0112
0.0340
0.2145
0.4713
68
-------
TABLE 4-16. ANALYSIS OF VARIANCE TABLE FOR DENVER
PERSONAL EXPOSURE DATA (CONDITIONED ON MDTR)
Conditioning
factor,
MDTR
Walk
Car
Other
Source
TM
SM
TM*SM
ERROR
TM
SM
TM*SM
ERROR
TM
SM
TM*SM
ERROR
Sum of
squares
50.27
0.64
44.04
20817.20
103.43
2.74
5.68
20817.20
16.90
1.46
51.98
20817.20
Degrees
of freedom
3
1
3
4070
3
1
3
4070
3
1
3
4070
Mean
square
16.76
0.64
14.68
5.11
34.48
2.74
1.89
5.. 11
5.63
1.46
17.33
5.11
F
3.28
0.12
2.87
6.75
0..54
0.37
1.10
0.29
3.39
Tail
prob.
0.0200
0.7244
0.0349
0.0002
0.4638
0.7743
0.3465
0.5931
0.0172
TABLE 4-17. ANALYSIS OF VARIANCE TABLE FOR DENVER
PERSONAL EXPOSURE DATA (CONDITIONED ON SM)
Conditioning
factor,
SM
Yes
No
Source
TM
MDTR
TM*MDTR
ERROR
TM
MDTR
TM*MDTR
ERROR
Sum of
squares
39.75
44.67
55.32
20817.20
a
a
a
a
Degrees
of freedom
3
2
6
4070
Mean
square
13.25
22.23
9.22
5.11
F
2.59
4.35
1.81
Tail
prob.
0.0509
0.0129
0.0940
Analysis not performed.
69
-------
TABLE 4-18. ANALYSIS OF VARIANCE TABLE FOR WASHINGTON, D.C. PERSONAL EXPOSURE
DATA (CELL WEIGHT = 1, COVARIABLE =» NONE)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Error
Sums of
squares
, 92.09
112.30
36.73
21.45
18.48
3.58
10.01
8696.98
Degrees
of freedom
2
3
1
6
2
3
6
3152
Mean
square
46.04
37.43
36.73
3.57
9.24
1.19
1.67
2.76
F
16.69
13.57
13.31
1.30
3.35
0.43
0.60
Tall
prob.
0.0000
0.0000
0.0003
0.2556
0.0353
0.7298
0.7268
TABLE 4-19. ANALYSIS OF VARIANCE TABLE FOR WASHINGTON, D.C. PERSONAL EXPOSURE
DATA (CELL WEIGHT = 1, COVARIABLE = DR)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Covariable3
Error
Sums of
squares
91.00
112.39
36.16
21.26
18.36
3.54
10.02
0.13
8696.85
Degrees
of freedom
2
3
1
6
2
3
6
1
3151
Mean
square
45.50
37.46
36.16
3.54
9.18
1.18
1.67
0.13
2.76
F
"16.48
13.57
13.10
1.28
3.33
0.43
0.60
0.05
Tail
prob.
0.0000
0.0000
0.0003
0.2611
0.0360
0.7333
0.7266
0.8310
Regression coefficient not significant.
70
-------
TABLE 4-20. ANALYSIS OF VARIANCE TABLE FOR WASHINGTON, D.C. PERSONAL EXPOSURE
DATA (CELL WEIGHT = DR, COVARIABLE = NONE)
Source
MDTR
TM
SM
MDTR*TM
MDTR*SM
TM*SM
MDTR*TM*SM
Error
Sums of
squares
•1303.48
2743.88
919.10
751.60
1141.14
100.06
472.87
168622.53
Degrees
of freedom
2
3
1
6
2
3
6
3146
Mean
square
651.74
914.63
919.10
125.27
570.57
33.35
78.81
53.60
F
12.16
17.06
17.15
2.34
10.65,
0.62
1.47
Tail
prob.
0.0000
0.0000
0.0000
0.0296
0.0000
0.6006
0.1842
TABLE 4-21.
CROSS TABULATED EXPOSURE MEANS AND STANDARD DEVIATIONS FOR DENVER
MDTR VERSUS SM CONDITIONED ON TM
TM
6-9
9-16
16-19
19-24
MDTR
walk
car
other
walk
car
other
walk
car
other
walk
car
other
SM
Yes
Mean
1.360
1.908
2.919
0.934
1.037
1.214
0.657
1.747
0.193
-0.842
1.249
0.826
Sd
2.14
2.65
1.29
1.70
2.47
1.60
2.27
2.23
2.32
2.93
2.52
3.09
n
17
37
4
68
161
25
33
69
17
5
43
9
- No
Mean
1.420
2.072
1.658
-0.124
1.367
1.105
0.685
1.769
2.174
0.556
1.302
1.288
Sd
2.27
2.41
2.54
2.26
2.15
2.53
2.55
2.31
1.48
2.46
2.27
1.91
n
59
453
55
248
1393
179
86
648
74
37
336
38
71
-------
TABLE 4-22. CROSS TABULATED EXPOSURE MEANS AND
STANDARD DEVIATION FOR DENVER
TM VERSUS SM CONDITIONED ON MDTR
MDTR
Walk
Car
Other
TM
6-9
9-16
16-19
19-24
6-9
9-16
16-19
19-24
6-9
9-16
16-19
19-24
SM
Yes
Mean
1.360
0.934
0.657
-0.842
1.908
1.037
1.747
1.249
2.919
1.214
0.193
0.826
sd
2.14
1.70
2.27
2.93
2.65
2.47
2.23
2.52
1.29
1.60
2.32
3.09
n
17
68
33
5
37
161
69
43
4
25
17
9
No
Mean
1.420
-0.124
0.685
0.556
2.072
1.367
1.769
1.302
1.658
1.105
2.174
1.288
sd
2.27
2.26
2.55
2.46
2.41
2.15-
2.31
2.27
2.54
2,53
1.48
1.91
n
59
248
86
37
453
1393
648
336
55
179
74
38
TABLE 4-23. CROSS TABULATED EXPOSURE MEANS AND
STANDARD DEVIATION FOR DENVER
TM VERSUS MDTR CONDITIONED ON SM
SM
Yes
No
TM
6-9
9-16
16-19
19-24
6-9
9-16
16-19
19-24
MDTR
Walk
Mean
1.360
0.934
0.657
-0.842
1.420
-0.124
0.685
0.556
sd
2.14
1.70
2.27
2.93
2.27
2.26
2.55
2.46
n
17
68
33
5
59
248
86
37
Car
Mean
1.908
1.037
1.747
1.249
2.072
1.367
1.769
1.302
sd
2.65
2.47
2.23
2.52
2.41
2.15
2.31
2.27
n
37
161
69
43
453
1393
648
336
Other
Mean
2.919
1.214
0.193
0.826
1.658
1.105
2.174
1.288
sd
1.29
1.60
2.32
3.09
2.54
2.53
1.48
1.91
n
4
25
17
9
55
179
74
38
72
-------
Denver--
The three-factor interaction was found to be significant; consequently,
a series of two-factor ANOVA's conditioned on the level of the third factor
were run. Displays are presented in Tables 4-15, 4-16, and 4-17. Two-factor
interactions were found to be significant in some, though not all, combina-
tions; consequently, some additional one-factor ANOVA's, conditioned on the
levels of the interacting factor, are warranted. The .indicated one-factor
ANOVA's have not been carried out.
Some interim conclusions can be formulated, however. From Table 4-15 it
can be inferred that interactions between MDTR and SM are not significant
for the conditioning intervals 6-9 and 19-24. In these intervals the
ANOVA's also yield tests for the equality of means in the main effects MDTR
and SM. The null hypothesis is rejected for MDTR in the 19-24 interval,
but cannot be rejected for SM in either interval.
From Table 4-16 it can be inferred that interaction is not significant
for the conditioning factor MDTR = car. The main effect means for TM are
not all equal, whereas such a claim cannot be made for the main effect
means for SM.
Table 4-17 is incomplete at present. However, it can be inferred that
the interaction is not significant for the conditioning factor SM = yes.
The main effect MDTR is significant, while the main effect TM barely misses
the critical region at the 0.05 level.
The results presented in Tables 4-12 (no covariable) and 4-13
(covariable = DR) are qualitatively identical. Entering the covariable DR
into the model does not change the fit. It is of interest that the
regression coefficient is significant; consequently, an association between
personal exposure and duration of exposure does exist.
Table 4-14 presents the ANOVA results for a weighted analysis, with weight
equal to DR. A comparison of the results of unweighted and weighted ANOVA's
(Tables 4-12 and 4-14) reveals that the three-factor interaction is significant
in both tables, suggesting the need for additional two-factor ANOVA's (un-
weighted and weighted) conditioned on the levels of the third factor. The un-
weighted analysis has been implemented while the weighted analysis remains to
be carried out. In the unweighted analysis (Table 4-12), the two-factor in-
teraction MDTR*SM is not significant, whereas TM*SM is significant. In the
73
-------
weighted analysis, the reverse is true: MDTR*SM is significant and TM*SM is
not significant (Table 4-14). Further exploration via two-way and one-way
conditioned ANOVA's is indicated.
Washington—
The results presented in Tables 4-18, 4-19, and 4-20 indicate that the
three-factor interaction in each table is not significant. Table 4-19
suggests that no association between personal exposure and duration exists
for Washington. Tables 4-18 and 4-19 suggest that two-factor interactions
involving TM are not significant. The null hypotheses for equality of
means of the main effect TM is rejected. It is concluded that the personal
exposure means are different according to the time of day and independent
of MDTR and of SM. Two one-factor ANOVA's, conditioned on the levels of
the remaining factor, are needed to clarify the roles of MDTR and SM. They
remain to be carried out.
Results from the weighted ANOVA are given in Table 4-20. The two-factor
interactions between MDTR and TM, and between MDTR and SM, are both significant,
whereas the interaction between TM and SM is not significant. Future analyses
should consider (a) a set of one-way ANOVA's with MDTR as the main effect and
conditioned on each of the levels of TM and SM; and (b) a "set of two-way
ANOVA's with TM and SM as main effects and MDTR as the conditioning variable.
4.2.4 Summary of Conclusions
The ANOVA's, analyses of covariance, and cross-tabulations presented
above support the following general conclusions.
Denver—
1. The two motor vehicle categories (car and other) are associated
with higher exposures than walking. Exposures in these two
categories are particularly high during the time period 6-9 and
16-19. These two periods bracket the morning and afternoon rush
hours.
2. The presence of smokers does not increase exposure.
3. Exposure decreases as duration increases.
Washington—
1. Exposure varies with mode-of-travel, time-of-day, and presence of
smokers.
74
-------
2. No associations exist between exposure and duration.
As indicated in Section 4.2.3, the Washington data received an
incomplete analysis. In future analyses, cross-tabulation tables similar
to Tables 4-21, 4-22, and 4-23 should be constructed so that the effect
of the various variables can be quantified. Recommended ANOVA's for
future Washington analyses are discussed in Section 4.2.3.
4.3 INDOOR EXPOSURES
EMSL directed PEI to identify specific factors which significantly affect
indoor exposures in Denver. This section describes a number of analyses
performed by PEl in this area.and summarizes the major findings. Note that
the principal data bases used in these analyses consisted of 1) PEM values,
2) the activity diary entries of the Denver study participants, and 3)
participant responses to the study questionnaire.
4.3.1 Candidate Exposure Factors
Each PEM value in the Denver SAMPLE-DATA file (described in Section
4.10 of Reference 1) has a two-digit location code. Sixteen of these codes
correspond to indoor microenvironments. Table 4-24 lists these microenviron-
ments and provides selected summary statistics based on data with acceptable
overall (i.e., PEM plus activity diary) quality codes. The minimum value
reported for each microenvironment is zero. Note the large values of
skewness and kurtosis listed for most microenvironments. To facilitate the
use of statistical analyses requiring normal distributions, PEI investigated
the use of the Box-Cox transformation as a means of reducing skewness and
kurtosis. The general form of the transformation is given in Equation 4-3
with y the transformed value, x the PEM value, and X a constant selected
by the user. A preliminary analysis suggested that the choice of A = 0.35
produced low skewness and kurtosis values for most microenvironments.
Table 4-25 lists summary statistics of the transformed data. Note the
dramatic reduction in most of the skewness and kurtosis values after trans-
formation..
75
-------
TABLE 4-24. SUMMARY STATISTICS FOR CARBON MONOXIDE CONCENTRATION VALUES RECORDED
BY PERSONAL EXPOSURE MONITORS IN INDOOR MICROENVIRONMENTS
Code
02
03
04
05
51
52
53
54
55
56
57
58
59
60
61
62
Indoor
microenvironment
Residence
Office
Store
Restaurant
Residential garage
Public garage
Manufacturing
facility
Service station or
auto repair
facility
Other repair shop
Auditorium
Church
Shopping mall
Health care facility
School
Other public building
Other indoor location
n
21518
2287
734
524
66
115
42
125
55
100
179
58
351
426
115
425
Carbon monoxide concentration, ppm
Maximum
76.4
59.1
56.3
35.0
28.4
81.2
8.0
73.1
33.1
31.2
21.7
33.9
31.3
21.6
21.8
66.4
Mean
2.212
3.248
3.385
4.313
3.364
11.968
1.750
9.409
7.620
4.523
1.824
5.271
2.334
2.056
2.937
4.923
s.d.a
4.030
4.970
4.754
4.674
5.059
11.984
2.366
9.704
8.575
5.649
2.998
6.493
3.632
3.090
3.760
7.958
s.e.b
0.027
0.104
0.175
0.204
0.623
1.118
0.365
0.868
1.156
0.565
0.224
0.853
0.194
0.150
0.351
0.386
Median
0.90
1.90
2.00
3.15
1.40
8.40
0.00
5.80
3.00
3.50
1.00
3.10
1.20
0.80
1.50
2.90
s.e.b
0.029
0.058
0.115
0.231
0.318
0.635
0.577
1.212
1.617
0.577
0.087
0.462
0.173
0.144
.0.491
0.202
Std.
skewness
e .
110.40
47.05
21.76
9.02
10.85
2.75
13.13
2.98
9.72
19.50
7.35
29.28
24.52
9.77
34.45
Std. d
kurtosis
e
483.29
167.26
44.53
14.49
19.71
-0.30
31.32
-0.26
13.97
42.54
9.58
79.80
47.92
15.02
85.67
Standard deviation.
JStd. kurtosis = g
Standard error.
cStd. skewness = gj//6/n.
"Not computed because of large sample size.
-------
TABLE 4-25. SUMMARY STATISTICS FOR CARBON MONOXIDE CONCENTRATION VALUES RECORDED BY
PERSONAL EXPOSURE MONITORS IN INDOOR MICROENVIRONMENTS AFTER TRANSFORMATION
Code
02
03
04
05
51
52
53
54
55
56
57
58
59
60
61
62
Indoor
microenvironment
Residence
Office
Store
Restaurant
Residential garage
Public garage
Manufacturing
facility
Service station or
auto repair
facility
Other repair shop
Auditorium
Church
Shopping mall
Health care facility
School
Other public building
Other indoor location
n
21518
2287
734
524
66
115
42
125
55
100
179
58
351
426
115
425
Carbon monoxide concentration, ppm
Maximum
10.175
9.055
8.854
7.059
6.360
10.456
3.059
9.975
6.867
6.668
5.532
6.949
6.679
5.518
5.545
9.551
Mean
-0.311
0.469
0.592
1.095
0.431
3.336
-0.814
2.744
1.731
0.967
-0.284
1.511
-0.033
-0.255
0.258
0.983
s.d.a
2.170
2.154
2.111
2.108
2.244
2.085
2.287
2.111
2.833
2.364
1.888
2.017
2.044
2.045
2.205
2.467
s.e.b
0.015
0.045
0.080
0.092
0.276
0.194
0.353
0.189
0.382
0.236
0.141
0.265
0.109
0.099
0.206
0.120
Median
-0.103
0.720
0.784
1.412
0.357
3.161
-2.857
2.430
1.340
1.570
0.000
1.388
0.188
-0.215
0.436
1.290
s.e.b
0.032
0.038
0.074
0.111
0.245
0.157
1.051
0.348
0.637
0.289
0.084
0.265
0.156
0.163
0.357
0.103
Std.
skewness
e
0.56
-0.25
-3.21
0.68
2.28
0.91
0.37
-0.27
-0.60
1.12
0.65
1.19
1.65
-0.14
1.55
Std. .
kurtosis
e
-0.52
-0.62
-1.54
-0.67
0.79
-2.24
. 1.53
-1.71
-1.15
-0.55
0.62
-1.65
-3.46
-2.26
1.38
Standard deviation.
Std. kurtosis = g2//24/n.
Standard error.
'Std. skewness =
'Not computed because of large sample size,
gjX/6/n.
-------
Section 5.3 summarizes the results of an analysis which used multiple
comparison tests as a means of identifying and grouping similar mlcroenvlron-
ments. The analysis found that the 16 microenvlronments could be grouped
into four aggregate microenvironments such that the microenvironments 1n each
group were statistically similar with respect to exposure and the groups were
statistically different. Thus microenvlronment In general 1s a factor which
affects exposure, but the effects of two specific microenvironments may not
be statistically different.
Other data items appearing in the activity diary which might affect
indoor exposure include the activity (e.g., laundry, cooking), address,
attached garage, gas stove, and smokers present entries. A variety of
analyses evaluating the relationships between exposures and activity diary
items are described in Section 6 of Reference 1 and previous sections of
this report. Table 4-26 lists the format of a special computer file
created by PEI which provides the indoor PEM values and activity diary
entries for each person-day of data. For each PEM value associated with a
particular person, the file lists the background questionnaire entries for
that person. In analyzing this file, it is understood that not all
questionnaire responses listed for a given PEM value are pertinent to that
exposure. For example, the presence of a heat pump in the home should not
affect exposures at work.
4.3.2 Determination of Subject Location
Some of the analyses that follow Involve relating exposures to question-
naire responses dealing with the home and workplace environments. Determin-
ing when a subject is "home" is not straightforward. Although the SAMPLE-
DATA file indicates when a subject is in a residence, that residence is not
necessarily his or her home. The address of the residence does not appear in
the SAMPLE-DATA file, only the census tract. The subject could be visiting
a neighbor or relative. Fortunately, one can identify many "indoors-residence"
entries in the activity diary as not being the home residence. Usually the
subject will be home when the first diary entry is made, when "sleeping" is
recorded, and when "final entry" is recorded. If the same census tract is
given in all three cases, there is a very high probability that the reported
census tract is the "home census tract." We can assume that all "indoors-
78
-------
TABLE 4-26. FORMAT OF COMPUTER FILE LISTING INDOOR PEM VALUES, ACTIVITY
DIARY ENTRIES, AND SELECTED BACKGROUND QUESTIONNAIRE RESPONSES
Columns
1-6
7-13
14-19
20-21
22-25
26-29
30-31
32-33
34-39
40-44
45-49
50-51
52-53
54-55
56-57
58
59
60
61
62
63
64-65
66-71
72-76
77-85
86-94
95-103
104-108
109-110
111-115
116-117
118-126
127-132
133-136
137-138
139-140
141-145
146-147
148-149
150-155
156-157
158-159
160-161
162-166
167-169
170-171
172-173
174-175
176-181
Format
A6
A7
A6
12
A4
F4.1
12
12
F6.2
F5.2
F5.2
12
12
12
12
11
11
11
11
12
11
12
A6
F5.1
F9.3
F9.3
F9.3
F5.1
12
15
12
9tH)
5(I1),A1
3(11), Al
12
12
5(11}
12
12
16
12
12
12
5(11)
13
12
12
12
F6.2
Description
Monitor ID
PID #
Date sampled (MMDDYY)
Sequence number
Time (0000-2359)
PEM CO concentration, ppm
Activity Diary Item A
Activity Diary Item B
Activity Diary Item C
Activity Diary Item Dl
Activity Diary Item 02
Activity Diary Item 03
Activity Diary Item El
Activity Diary Item E2
Activity Diary Item F
Class code for slope
Class code for intercept
Highest class code
PEM quality code (1 = flag)
Diary quality code
Overall quality code
Duration, minutes
Nearest fixed-site code
Nearest fixed-site CO concentration, ppm
Weight (PEM)
Weight (Diary)
Weight (Overall)
Composite site CO concentration
Housing type
Living area (sq. ft)
Packs smoked per week
CO sources in residence
Energy-saving devices in residence
Fans in residence
Main heating system in residence
Air conditioning in residence
Pollution sources near residence
Year of automobile
Type of enclosed work area (EWA)
Square feet in EWA
Air conditioning in EWA
Fan in EWA
Main heating system in EWA
Pollution -sources near work
Occupation type
Sex
Age
Construction of residence
Census tract of residence
79
-------
residence" entries with census tracts other than the "home census tract" were
not recorded in the subject's home.
PEI wrote a computer program which performed the indicated logic tests
and identified the home census tract. Where some of the necessary diary
entries were missing, PEI reviewed the actual diaries and determined the
correct home census tracts. As indicated in Table 4-26, home census tracts
appear in columns 176 through 181. For the analyses described here, PEI
assumed a subject was home when the activity diary LOCATION code indicated
"indoors-residence" and. the census tract listed under ADDRESS was
identical to the "home" census tract appearing in columns 176 through 181.
A subject was considered to be indoors at work when the ACTIVITY code
indicated "work or study." PEI required that the work location be a place
other than residence for the analyses described in this memorandum. Since
only indoor LOCATION codes were included in the special computer file,
outdoor work exposure situations were automatically omitted from the
analysis.
4.3.3 Results of One-Way Analyses of Variance
One-way analyses of variance were performed by BMDP program P2V on indoor
exposures at home and indoor exposures at work using responses to selected
questions on the background questionnaire as the grouping variables. Tables
4-27 through 4-33 present the results of these analyses for untransformed and
transformed (X » 0.35) PEM values. The p values listed in these tables indi-
cate the probability that the means associated with the various responses to
a question are identical.
Table 4-27 presents the results of an ANOVA of the effect of area on home
exposure. Area responses were divided into four groups using cutpoints of
1000, 1425, and 2056 ft2--the quartiles listed in Table 6-9 of Reference 1.
As might be expected, the largest mean (2.49 ppm) is associated with small
i)
homes (area _< 1000 ft ), which have less air volume to dilute CO from indoor
sources. However, the smallest mean is not associated with large homes but
2 2
with homes having areas between 1000 ft and 1425 ft . This result suggests
a possible confounding effect. Perhaps gas stoves are more common in certain
size homes.
80
-------
TABLE 4-27. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS
Approximate area of
living quarters, square feet
Area Ł 1000
1000 < area _< 1425
1425 < area < 2056
2056 < area
n
4828
4401
4685
4620
Untransformed data
Mean
2.49
1.80
2.46
1.94
Std. dev.
3.42
3.30
4.98
4.24
Transformed data
Mean
0.01
-0.62
-0.22
-0.50
Std. dev.
2.14
2.11
2.23
2.09
F test indicates p < 0.0001 for untransformed and transformed data,
TABLE 4-28. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME VERSUS
NUMBER OF CIGARETTE PACKS SMOKED PER WEEK BY OTHER HOUSEHOLD MEMBERS
Code
0
01
02
03
04
Cigarette packs smoked
per week
No smokers in residence
Less than 1 pack
1 to 4 packs
5 to 7 packs
8 or more packs
n
14865
2159
1951
620
705
Untransformed data
Mean
2.15
2.00
2.41
2.22
3.80
Std. dev.
3.80
3.57
4.59
3.55
7.22
Transformed data
Mean
-0.34
-0.51
-0.18
-0.21
0.59
Std. dev.
2.15
2.17
2.17
2.13
2.25
F test indicates p < 0.0001 for untransformed and transformed data.
81
-------
TABLE 4-29.
RESULTS OF ANALYSES OF VARIANCE OF INDOOR EXPOSURES VERSUS RESPONSE TO SELECTED
QUESTIONS CONCERNING COMBUSTION SOURCES IN LIVING QUARTERS
Combustion source
In living quarters
Fireplace3
Woodstove
Gas furnace0
Gas cooking stove0
Gas hot water heater
Gas clothes dryerc
Gas or kerosene space
heater
Code
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
1
2
3
Response
Used and vented
Used and not vented
Not used
Used and vented
Used and not vented
Not used
Used and vented
Used and not vented
Not used
Used and vented
Used and not vented
Not used
Used and vented
Used and not vented
Not used
Used and vented
Used and not vented
Not used
i
Used and vented
Used and not vented
Not used
n
7953
292
11911
1963
96
18097
16188
786
3182
1837
2837
15482
15271
1638
3218
3729
476
15951
291
1286
18579
Untransformed data
Mean
2.12
1.94
2.29
2.20
2.31
2.22
2.18
3.10
2.19
3.66
3.24
1.86
2.21
2.35
2.22
2.68
2.33
2.11
1.82
3.07
2.17
Std. dev.
4.43
3.11
3.76
4.96
3.15
3.92
3.99
6.61
3.31
5.09
4.34
3.76
4.04
4.94
3.46
4.62
2.95
3.90
4.85
4.90
3.95
Transformed data
Mean
-0.38
-0.60
-0.24
-0.50
-0.30
-0.28
-0.33
0.01
-0.23
0.48
0.53
-0.54
-0.30
-0.30
-0.26
-0.03
-0.09
-0.37
-0.59
0.18
-0.33
Std. dev.
2.13
2.21
2.19
2.26
2.29
2.16
2.16
2.37
2.12
2.38
2.11
2.09
2.16
2.21
2.16
2.24
2.14
2.15
2.03
2.29
2.16
00
ro
F test indicates p = 0.006 for untransformed data and p < 0.0001 for transformed data.
}F test indicates p = 0.958 for untransformed data and p = 0.0001 for transformed data,
'F test indicates p < 0.0001 for untransformed and transformed data.
F test indicates p = 0.415 for untransformed data and p = 0.654 for transformed data.
-------
TABLE 4-30., RESULTS OF ANALYSES OF VARIANCE OF INDOOR EXPOSURE VERSUS
RESPONSE TO SELECTED QUESTIONS CONCERNING ENERGY-SAVING DEVICES
IN LIVING QUARTERS
Energy-saving
devices in
living
quarters
Storm windows3
Storm door(s)a
Extra insula-
tion5
Special dampers
in stove or
fireplace
Code
1
2
1
2
1
2
1
2
Response
Yes
No
Yes
No
Yes
No
Yes
No
n
12750
7259
13319
6481
9859
7027
3602
12915
Untrans formed data
Mean
2.35
1.97
2.43
1.78
2.18
2.17
2.53
2.06
Std. dev.
4.47
3.08
4.32
3.36
3.91
4.34
5.08
3.59
-
Transformed data
Mean
-0.23
-0.43
-0.17
-0.58
-0.29
-0.36
-0.18
-0.38
Std. dev.
2.19
2.12
2.20
2.07
2.14
2.16
2.25
2.12
F test indicates p < 0.0001 for untransformed and transformed data.
5F test indicates p = 0.885 for untransformed data and p = 0.0546 for
transformed data.
83
-------
TABLE 4-31. RESULTS OF ANALYSES OF VARIANCE OF INDOOR EXPOSURES VERSUS MAIN HEATING SYSTEM
IN LIVING QUARTERS
Code
01
02
03
04
05
06
07
08
11
Main heating system in living quarters3
Steam or hot water system
Central warm air furnace with ducts to individual
rooms, or central heat pump (forced air)
Built-in electric units, permanently installed in
wall, ceiling, or baseboard
Floor, wall, or unvented furnace
Circulating, radiant, or room heaters, WITH flue or
vent, burning gas, oil, or kerosene
Circulating radiant, or room heaters (not portable)
WITHOUT flue or vent, burning gas, oil, or kerosene
Portable room heaters of any kind
Fireplace(s) or stove(s) burning coal, wood, or coke
Gravity gas
n
2942
15552
195
229
139
119
60
727
136
Untransformed data
Mean
2.37
2.13
0.96
3.08
2.45
1.96
11.84
2.34
5.03
Std. dev.
3.77
4.05
2.06
3.53
3.69
3.22
4.34
3.94
6.44
Transformed data
Mean
-0.10
-0.37
-1.26
0.51
-0.44
-0.23
3.81
-0.21
1.32
Std. dev.
2.13
2.16
1.76
2.06
2.43
1.96
0.96
2.17
2.16
CO
F test indicates p < 0.0001 for untransformed and transformed data,
-------
TABLE 4-32. RESULTS OF ANALYSES OF VARIANCE OF INDOOR EXPOSURES VERSUS MAIN HEATING SYSTEM
IN WORKPLACE
Code
01
02
03
05
06
07
Main heating system in workplace
Steam or hot water system
Central warm air furnace with ducts to individual
rooms, or central heat pump (forced air)
Built-in electric units, permanently installed in
wall, ceiling, or baseboard
Circulating, radiant, or room heaters, WITH flue or
vent, burning gas, oil, or kerosene
Circulating radiant, or room heaters (not portable)
WITHOUT flue or vent, burning gas, oil, or kerosene
Portable room heaters of any kind
n
742
1610
284
71
25
23
Untransformed data
Mean
3.36
3.47
3.09
8.48
6.04
2.00
Std. dev.
4.27
5.93
3.17
13.73
9.47
1.61
Transformed data
Mean
0.40
0.44
0.55
2.08
1.22
0.15
Std. dev.
2.32
2.27
2.05
2.47
2.67
1.75
CO
en
F test indicates p < 0.0001 for untransformed and transformed data,
-------
TABLE 4-33.
RESULTS OF PAIRWISE COMPARISONS OF INDOOR EXPOSURES ASSOCIATED
WITH COMBUSTION SOURCES IN LIVING QUARTERS
Combustion source
1n living quarters
Fireplace
Woodstove
Gas furnace
Gas cooking stove
Gas hot water
heater
Gas clothes dryer
Gas or kerosene
space heater
Means of untransformed data, ppm
Response A
uva
UV
UNV
UV
UV
UNV
UV
UV
UNV
UV
UV
UNV
UV
UV
UNV
UV
UV
UNV
UV
UV
UNV
Value
2.12
2.12
1.94-
2.20
2.20
2.31
2.18
2.18
3.10
3.66
3.66
3.24
2.21
2.21
2.35
2.68
2.68
2.33
1.82
1.82
3.07
Response B
UNV^
NUC
-'NU
UNV
NU
NU
UNV
NU
NU
UNV
NU
NU
UNV
NU
NU
UNV
NU
NU
UNV
NU
NU
Value
1.94
2.29
2.29
2.31
2.22
2.22
3.10
2.19
2.19
3.24
1.86
1.86
2.35
2.22
2.22
2.33
2.11
2.11
3.07
2.17
2.17
p value
Untrans-
formed
data
0.3317.
0.0040°
0.0543
0.7519
0.8722
0.7827
0.0001d
0.8483.
0.0002°
0.0038d
0.0000°.
0.0000°
0.2729
0.9449
0.3268
0.0263.
0.0000°
0.1050
0.0001d
0.2317.
0.0000°
Transformed
data
0.0922H
0.0000°
0.0063
0.3953H
0.0000°
0.9255
0.0001d
0.0226.
0.0090°
0.4972,
0.0000°
0.0000°
0.8911
0.3551
0.6428
0.5150H
0.0000°,
0.0056°
0.0000°
0.0297,
0.0000°
UV = used and vented.
3UNV = used and not vented.
:NU = not used.
Jp < 0.0167.
86
-------
Table 4-28 shows a large increase in CO exposure for homes where eight or
more packs of cigarettes are smoked per week. The mean exposure associated
with this response (3.80 ppm) is almost twice the mean exposure for homes
where less than one pack is smoked per week (2.00 ppm).
Table 4-29 presents results of seven separate ANOVA's concerning combus-
tion sources in the home. The possible responses to each question are 1) the
source is used and vented, 2) the source is used and not vented, and 3) the
source is not used. If a source produces significant quantities of CO, one
would expect the means associated with response 2 to be significantly larger
than those associated with response 3. If the mean for response 2 is signifi-
cantly larger than the mean for response 1, one can assume that venting re-
duces CO exposure in the home. To identify these situations, PEI performed
pairwise comparisons on transformed data using BMDP program P7D. Table 4-
33 summarizes the results of these comparisons. If a pairings yields
p < 0.05/N where N = 3 (the number of possible pairings of 3 responses),
the test is considered significant at the 0.05 level according to the
Bonferroni test, and the associated means are assumed to be different.
Vented source which yield significantly higher exposures (p < 0.0167) than
those that occur in the absence of the source include gas.cooking stoves
and gas clothes driers. Unvented sources which yield significantly higher
exposures than those that occur in the absence of the source include gas
furnaces, gas cooking stoves, gas clothes dryers, and space heaters (gas or
kerosene). Sources for which venting appears to significantly decrease
exposure include gas furnaces and space heaters (gas or kerosene). Some
results are counter-intuitive; for example, exposures in homes with a
vented woodstove are slightly lower than exposures in homes without a
woodstove. Perhaps homes with woodstoves are less likely to have gas
appliances.
Table 4-30 presents the results of four F tests comparing exposures with
and without certain energy-saving devices in the home. The results suggest
that storm windows, storm doors, and special dampers significantly (p < 0.05)
increase exposure, whereas extra insulation does not appear to have a sig-
nificant effect.
87
-------
Table 4-31 presents the results of an ANOVA evaluating the effect of the
main heating system on home exposures. The highest mean (11.84 ppm) is
associated with portable room heaters. Gravity gas has the second highest
mean (5.03 ppm). The lowest mean exposure is associated with built-in
electric units (0.96 ppm).
Table 4-32 presents results of a similar analysis of the main workplace
heating system. Contrary to the results in Table 4-31, portable room heaters
yield the lowest mean exposure listed in the table. However, the sample
size is quite small (n = 23) and probably represents the exposures of only a
few subjects. The largest means (8.48 ppm and 6.04 ppm) are associated with
nonportable room heaters burning gas, oil, or kerosene. Again, sample sizes
are small.
4.3.4 Results of Two-Hay Analyses of Variance
To separate the effects of area from those of indoor CO sources in the
home, PEI used BMDP program P2V to perform a series of two-way ANOVA1s with
area as one grouping variable and indoor source type or source status as the
other grouping variable. Tables 4-34 through 4-39 present the results of
these ANOVA1s. Note that the Box-Cox transformation was not used 1n these
analyses.
Table 4-34 presents the results for an ANOVA where cigarette packs smoked
per week is the indoor source variable. To prevent the occurrence of empty
cells, the two classifications "area <. 1000" and "1000 < area <_ 1425" listed
in Table 4-27 were combined into the classification "area <_ 1425." When area
is held constant, one finds the expected general increase in CO with increas-
ing number of packs smoked for only one area category, "area _< 1425," and only
if one ignores the mean for "8 or more packs." When smoking is held constant,
one does not find the expected general decrease in CO with increasing area.
Table 4-35 presents the results for an ANOVA where gas furnace status is
the indoor source variable. Status categories are used and vented (UV),
used and not vented (UNV), and not used (NU). No general patterns are
evident in the cell means other than the UV mean exceeds the NU mean for
all but the smallest area classification.
In Table 4-36, gas cooking stove status is the indoor source variable.
The main pattern evident in this table is that NU means tend to be less than
UV and UNV means when area is held constant.
88
-------
TABLE 4-34. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND CIGARETTE PACKS SMOKED PER WEEK
BY OTHER HOUSEHOLD MEMBERS
Approximate
area of living
quarters, ft2
Area < 1425
1425 < area <
2056
2056 < area
Cigarette packs
smoked per week
No smokers
Less than 1 pack
1 to 4 packs
5 to 7 packs
8 or more packs
No smokers
Less than 1 pack
1 to 4 packs
5 to 7 packs -
8 or more packs
No smokers
Less than 1 pack
1 to 4 packs
5 to 7 packs
8 or more packs
n
8574
1121
1006
200
94
3158
613
570
135
209
3133
425
375
285
402
CO concentration, ppm
Mean
2.19
2.31
2.36
2.90
2.00
2.23
1.20
3.24
1.87
7.82
1.95
2.30
1.28
1.92
2.13
Std. dev.
3.31
3.97
3.84
3.68
1.71
4.07
2.38
6.28
2.60
11.88
4.64
3.70
2.85
3.79
2.37
TABLE 4-35. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR -EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND STATUS OF GAS FURNACE
Approximate
area of living
quarters, ft2
Area < 1000
1000 < area < 1425
1425 < area < 2056
2056 < area
Status of
gas furnace
uvb
UNVD
NUC
UV
UNV
NU
UV
UNV
NU
UV
UNV
NU
n
4530
224
1696
3820
159
422
3868
214
603
3970
189
461
CO concentration, ppm
Mean
2.46
2.25
2.77
1.81
2.99
1.23
2.58
1.22
2.10
1.84
6.31
1.08
Std. dev.
3.43
2.30
3.81
3.43
2.58
1.97
5.30
2.75
3.03
3.54
12.12
1.82
Used and vented.
Used and not vented.
"Not used.
89
-------
TABLE 4-36. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND STATUS OF GAS COOKING STOVE
Approximate
area of living
quarters, ft2
Area Ł 1000
1000 < area _< 1425
1425 < area < 2056
2056 < area
Status of gas
cooking stove
UVK
UNV°
NU
UV
UNV
NU .
UV
UNV
NU
UV
UNV
NU
n
941
1427
4082
244
599
3558
313
389
3983
339
422
3859
CO concentration, ppm
Mean
4.33
3.28
1.86
4.21
2.84
1.45
3.14
4.71
2.18
1.84
2.32
1.91
Std. dev.
4.55
3.71
2.90
7.27
3.14
2.74
5.05
7.51
4.59
4.08
3.31
4.34
Used and vented.
}Used and not vented,
:Not used.
TABLE 4-37. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR-EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND STATUS OF GAS CLOTHES DRYER
Approximate
area of living
quarters, ft2
Area _< 1000
1000 < area <: 1425
1425 < area <. 2056
2056 < area
Status of gas
clothes dryer
uvh
UNV°
NUC
UV
UNV
NU
UV
UNV
NU
UV
UNV
NU
n
1036
286
5128
915
31
3455
541
135
4009
1237
24
3359
CO concentration, ppm
Mean
2.98
2.66
2.44
2.43
0.00
1.64
2.38
1.11
2.51
2.73
8.40
1.60
Std. dev.
4.00
2.99
3.41
3.66
0.00
3.19
4.11
1.34
5.15
5.80
1.42
3.41
Used and vented.
5Used and not vented,
:Not used.
90
-------
TABLE 4-38. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND STATUS OF GAS OR KEROSENE
SPACE HEATER
Approximate
area of living
quarters, ft2
Area Ł 1000
1000 < area _< 1425 *
1425 < area < 2056
2056 < area
Status of
space heater
Used
Not used
Used
Not used
Used
Not used
Used
Not used
n
381
6213
286
4115
482
4203
572
4048
CO concentration, ppm
Mean
3.60
2.45
2.93
1.72
1.81
2.53
2.90
1.81
Std. dev.
5.16
3.35
5.10
3.12
2.43
5.19
5.66
3.98
TABLE 4-39. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES AT HOME
VERSUS AREA OF LIVING QUARTERS AND IDENTITY OF MAIN HEATING SYSTEM
Approximate
area of living
quarters, ft2
Area _< 1000
1000 < area
Main heating system in
living quarters
Steam or hot water system
Central furnace or heat
pump
Built-in electric units
Floor, wall, or unvented
furnace
Heater with flue burning
gas, oil , or kerosene
Fireplace or stove
Gravity gas
Steam or hot water system
Central furnace or heat
pump
Built-in electric units
Floor, wall, or unvented
furnace
Heater with flue burning
gas, oil , or kerosene
Fireplace or stove
Gravity gas
n
1406
4483
102
99
51
124
62
1536
11039
93
130
88
603
74
CO concentration, ppm
Mean
2,51
2.32
1.30
5.44
5.01
4.25
2.48
2.24
2.06
0.58
1.28
0.97
1.95
7.16
Std. dev.
3.49
3.29
2.67
3.94
4.55
3.39
1.94
4.00
4.31
0.91
1.61
1.93
3.93
7.96
91
-------
In Table 4-37, gas clothes dryer status is the indoor source variable.
No general patterns are evident in the cell means.
Status of space heater is the indoor source variable in Table 4-38. The
UV and UNV responses have been combined into a "used" category to avoid empty
cells. When area is held constant, the "used" means exceed the "not used"
means for all area categories except "1425 < area _< 2056." Means for the
"used" category decrease with increasing area except for the "2056 < area"
category.
Ł
Table 4-39 presents results for an ANOVA where identity of the main heat-
ing source in the living quarters is the indoor source variable. To avoid
empty cells, only two area categories are defined and two heating systems
(06--heater without flue burning gas, oil, or kerosene, and 07--portable room
heaters of any kind) are omitted. If heating system is held constant, cell
means decrease with increasing area for all heating systems except gravity
gas. Built-in electric units have the lowest cell mean in both area categories.
The ANOVA's summarized in Tables 4-34 through 4-39 all yield p values
less than 0.0001 for the area variable, for the indoor source variable, and
for the interaction of the two variables. However, clear patterns in the cell
means occur only in Tables 4-36 (gas cooking stove), 4-38-(space heater), and
4-39 (main heating source). To determine if the effects of gas cooking stoves
and space heaters are additive, PEI performed the ANOVA summarized in Table
4-40. As would be expected, the smallest cell mean (1.78 ppm) is associated,
with the absence of both indoor sources. However, the two largest cell means
are associated with the presence of a gas cooking stove (either UV or UNV)
and the absence of a space heater. An additive relationship between gas stoves
and space heaters is not apparent, although the occurrence of several cells
with small sample sizes makes it difficult to completely dismiss such a rela-
tionship. The ANOVA summary table lists p = 0.9641 (nonsignificant) for the
space heater effect, p < 0.0001 for the gas stove effect, and p < 0.0001 for
their interaction.
To determine if space heaters affect exposure in homes without gas stoves
PEI performed a t test comparing two cell means: space heater used - gas stove
92
-------
TABLE 4-40. RESULTS OF ANALYSIS OF VARIANCE OF INDOOR EXPOSURES VERSUS
STATUS OF GAS OR KEROSENE SPACE HEATER AND STATUS OF
GAS COOKING STOVE
Status of
space heater
Used
Not used
Status of gas
cooking stove
uva
UNVb
NUC
UV
UNV
NU
n
165
171
1241
1672
2666
14241
CO concentration, ppm
Mean
3.02
2.96
2.80
3.72 •
3.26 .
1.78
Std. dev.
3.01
3.33
5.29
5.25
4.39
3.58
Used and vented.
'Used and not vented,
'Not used.
not used (mean = 2.80 ppm) and space heater not used - gas stove not used
(mean = 1.78 ppm). The results were highly significant (p < 0.0001), indicat-
ing that space heaters do increase exposure in homes without gas stoves.
4.3.5 Summary of Results
The ANOVA's discussed in Section 4.3 support the following conclusions:
2
1. CO exposures are higher in homes having living areas of 1000 ft
or less.
2. CO exposures are higher in homes where gas cooking stoves or gas
clothes dryers are used (vented or not vented).
3. CO exposures are higher in homes where unvented gas furnaces or
space heaters are used.
4. Venting of gas furnaces and space heaters decreases CO exposure
in the home.
5. CO exposures are higher in homes which have storm windows, storm
doors, or special dampers.
6. CO exposures are higher in homes where the main heating source is
either a portable room heater or gravity gas system.
93
-------
7. CO exposures are lower in homes where the main heating system
consists of built-in electric units.
8. CO exposures are higher in work places where the main heating
system consists of nonportable heaters burning gas, oil, or kerosene.
9. In homes without a gas cooking stove, the presence of a space
heater significantly increases exposure.
10. In homes with gas cooking stoves, the presence of a space heater
does not significantly increase exposure.
All of these conclusions are based on the assumption that all significant
confounding factors have been identified and properly considered in the
analyses. This assumption is probably unwarranted.
The dilution of CO from indoor sources is a function of the enclosed air
volume and the air exchange rate. The basis for using area as an explanatory
variable 1n some of the ANOVA's is that it is the best indicator of the volume
of the enclosed living space available. Possible indicators of air exchange
rates considered in this section are the heating system type and the presence
of energy-saving devices. Other possible indicators of home air exchange
rates listed in Table 4-26 are housing type, fans in residence, air-condition-
ing in residence (probably not pertinent for a winter study), and construction
of residence. These factors should be considered in future analyses.
The analysis of factors affecting work exposures presented here is less
detailed than the analysis of factors affecting home exposures. One reason
is the small number of identifiable indoor sources. Another is the ambiguity
concerning what constitutes a subject's enclosed work area. This ambiguity
may affect the validity of the responses to many of the questions concerning
work place. For this reason, the analyses discussed here focus more on the
home environment which has more definite boundaries.
4.4 REFERENCES
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. Report by PEDCo Environmental, Inc., to the U.S.
Environmental Protection Agency, Research Triangle Park, North Carolina.
June 1984.
2. Bailar, J.C., and F. Ederer. 1964. Significance Factors for the
Ratio of a Poisson Variable to its Expectation. Biometrics, 639-643.
3. Wesolowsky, G. 0. Multiple Regression and Analysis of Variance. John
Wiley and Sons. New York. Page 251, 1976.
94
-------
SECTION 5
MODELS FOR PREDICTING EXPOSURE IN DENVER
This section describes the evolutionary development of a model which, in
its final form, explains approximately 34 percent of the observed variation
in PEM values reported during the Denver CO study. As indicated below, a
series of 14 general models were proposed and evaluated in a sequential man-
ner such that the results of each evaluation were considered in constructing
the next general model. The parameters considered in the general models in-
cluded data obtained from the activity diaries, the background questionnaire
completed by each participant, the fixed-site monitors, and a meteorological
file containing data on temperature and daily average wind speed. Model
evaluation was accomplished by performing step-wise linear regression on each
general model and noting 1) which terms were retained in the "best-fit"
2
model and 2) the R value associated with the best-fit model.
Sections 5.1 and 5.2 present PEI analyses in the orde'r in which they
were performed. Section 5.1 discusses General Models 1 through 4; Section
5.2 discusses General Models 5 through 14. Section 5.3 discusses a method
involving pairwise comparisons which was used to aggregate indoor microen-
vironments. These aggregate microenvironments were used in General Models 5
through 14 but not in General Models 1 through 4. Section 5.3 also discusses
how the Box-Cox transformation can be used to reduce the skewness and kurtosis
of PEM values grouped by microenvironment.
5.1 GENERAL MODELS 1 THROUGH 4
Table 5-1 lists the candidate exposure factors which were combined in
various combinations to form the independent variables of General Models 1
through 4. The dependent value in each of these models is either PEM value
or the logarithm of PEM value. Step-wise linear regression (weighted) with
forward and backward stepping was performed on each general model to deter-
mine a "best-fit" model containing only those terms which make significant
95
-------
TABLE 5-1. GROUP CODES USED IN STEPWISE LINEAR REGRESSION ANALYSIS
Data source
item
Factor
code
Description
Responses coded
as 1 1n file
Diary item A
(Activity)
Al
A2
A3
A4
A5
A6
Diary Item B + D3
CMicroenviron-
ment]
Bl
Travel-related activities
Indoor activities near
CO sources
Other indoor activities
Outdoor activities
Work-related activities
Other activities
In-transit microenviron-
ments (ME's) involving
motor vehicles
01:all travel
03:cooking
04: laundry
08:eating
12:cafe or pub
05:other indoor
chores and child
care '
ll:social, political,
or religious
activities
09:sleeping
10:other personal
needs
17: interview
18:final entry
20:begin breath
sample
21:end breath sample
06:yard work and
other outdoor
activities
13:walking, bicycl-
ing, or jogging
02:work and study
07:errands and shop-
ping
14:other leisure
activities
15:uncertain of code
16:no entry in diary
0102:car
Ol03:bus
Ol04:truck
0193-.motorcycle
Ccontinued)
96
-------
TABLE 5-1 (continued)
Data source
item
r . . v
Factor
code
B2
B3
B4
B5
B6
B7
Description
Outdoor ME's near CO source
Other outdoor ME's
—
Public garage, carport,
or service station
Indoor residential garage
Indoor ME's with possible
CO source
Other indoor ME's
Responses coded
as 1 in file
0101 :wa1 king
07c:outdoors within
10 yards of road
73d: parking lot
74bb:service station
or motor vehicle
repair facility
75bb:construction
site
76bb:residential
grounds
77t>b:school grounds
78bb:sports arena,
etc.
79bb:park or golf
course
80bb:other location
52a: indoor public
garage
54bb:indoor service
station or motor
vehicle repair
facility
55bb:other indoor
repair shop
71bb:outdoor carport
72a:outdoor public
garage
51bb:indoor residen-
tial garage
05bb:restaurant
58bb:shopping mall
62bb:other location
02bb:residence
03c:office
04bb:store
53bb:manufacturing
facility
56bb:auditorium, etc.
57bb:church
59bb:health care
facility
[continued)
97
-------
TABLE 5-1 (continued)
Data source
item
Factor
code
Description
Responses coded
as 1 in file
Diary item El
Diary item E2
Diary item F
Diary time entry
Fixed-site data
file
GARAGE
GAS
SMOKE
Tl
T2
Garage attached to build-
ing?
Gas stove in use?
Smokers present?
Time between 6:00 a.m. and
10:00 a.m.
Time between 10:00 a.m.
and 5:00 p.m.
Simultaneous concentration
at the nearest fixed-site
for nontransit ME's and at
composite site for in-
transit ME's
60bb:school
61bb:other public
building
01:yes
01:yes
01:yes
6:00 <_ time < 10:00
10:00 <. time < 17:00
Continuous variable
Includes D3 = bb, 01, and 02.
'Blank.
'Includes 03 =
Includes D3 =
bb and 01.
bb, 01, 02,
and 03.
contributions to explaining the observed variation in the dependent variable.
In these regression analyses, most of the exposure factors are treated as
binary variables. A binary variable has a value of 1 when one or more of the
response codes listed in Table 5-1 are present and 0 when none of these codes
is present. Note that one of the factors (C) is treated as a continuous
variable.
98
-------
General Model 1 is
CpEM = a + C^KAl) + (32)(A2) + ••• + (3g)CA6) +
(37)(B1) + (38)(B2) + ... + (313)(B7) +
(314)(GARAGE) + (315)(GAS) + (6lg)(SMOKE). (5-1)
The coefficients a, $,, 3o» "'* 3i6 were estimated during the regression
analysis. Table 5-2 lists the sequence of steps followed by BMDP program P2R
in adding and deleting terms to the model. The suggested "best fit" model is
CpEM = 3.705 - (0.528)(A3) + (4.165)(B1)
+ (5.145)(B4) - (1.694)(B7)
+ (0.546)(GARAGE) + (1.923)(6AS)
+ (1.221)(SMOKE). (5-2)
2
The R value for this model is 0.1285. The variables in Equation 5-2 entered
the model in the following order: Bl (in-transit ME's), B7 (other indoor
ME's), SMOKE, B4 (public garage, carport, or service station), GAS, GARAGE,
and A3 (other indoor activities). It is interesting to note that inclusion
of the single term Bl (in-transit ME) yields an R value of 0.0917. Of the
various explanatory variables considered in General Model 1, transit status
appears to be the best single predictor of CO exposure. The activity codes
(Al, A2 A6) are generally poor predictors of CO exposure. Only A3 was
included in the stepwise model, and it was the last term added. Note that
both A3 and B7 have negative coefficients in Equation 5-2. The occurrence of
"1" for either of these explanatory variables indicates the subject is indoors
but not near a CO source. Multiplying 1 by a negative coefficient yields a
reduction in estimated PEM value. This result is consistent with our expec-
tations.
Stepwise linear regression (weighted) was likewise performed using
General Model 2:
99
-------
TABLE 5-2. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 1
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
Operation
Add Bla
Add B7a
Add SMOKE3
Add B4a
Add GASa
Add GARAGEa
ADD A3a
ADD A4
ADD B3
ADD A2
ADD A5
Remove A5
Remove A2
Remove B3
Remove A4
2
Resulting R
0.0917
0.1041
0.1150
0.1205
0.1257
0.1271
0.1285
0.1290
0.1293
0.1295
0.1295
0.1295
0.1293
0.1290
0.1285
Change in R
0.0917
0.0124
0.0109
0.0055
0.0052
0.0014
0.0014
0,0005
0.0003
0.0002
0.0001
-0.0001
-0.0002
-0.0003
-0.0005
Retained in best-fit model.
TEM
(B7)(B7) + (S8)(C) + (69) (GARAGE)
+ (610)(6AS) + (Bn)(SMOKE) +
(e12)(Bi)(c) + (e13)(B2)(c) + •••
(B18)(B7)(C)
(5-3)
where C is the simultaneous concentration at the nearest fixed site for non-
transit microenvironments and at the composite site for in-transit micro-
environments. Note that this General Model 2 contains a series of
interaction terms between microenvironment group (Bl, B2, ..., B7) and C.
Table 5-3 lists the results of the stepwise linear regression. The
suggested "best-fitting" model is
1QO
-------
TABLE 5-3. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 2
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Operation
Add Bla
Add Ca
Add (B7)(C)
Add SMOKE3
Add GASa
Add B4
Add GARAGE
Add B7a
Add (B6)(C)
Add B6
Add (B2)(C)
Add B3
Add (B5)(.C)
Add B5
Add (B4)(C)
Add B2
Add (B3)(C)a
Add (Bl)(C)a
Remove B5
Remove (B6)(C)
Remove B6
Remove (B5)(C)
Remove (B7)(C)
Remove (B3)(C)
Remove B3
Remove B4
Remove B2
Remove (B2)(C)
Remove GARAGE
2
Resulting R
0.0924
0.1441
0.1589
0.1690
0.1734
0.1780
0.1788
0.1795
0.1799
0. 1804
0.1806
0.1807
0.1808
0.1809
0. 1809
0.1810
0.1810
0.1811
0.1811
0.1811
0.1811
0.1810
0.1809
0.1805
0.1802
0.1796
0.1787
0.1776
0.1764
2
Change in R
0.0924
0.0517
0.0148
0.0101
0.0044
0.0045
0.0008
0.0007
0.0004
0.0005
0.0002
0.0001
0.0001
0.0001
0.0001
0.0000
0.0000
0.0001
-0.0000
-0.0000
-0.0000
-0.0000
-0.0001
-0.0004
-0.0003
-0.0006
-0.0009
-0.0011
-0.0011
Retained in best-fit model.
= 2.72 + (2.809)(B1) - (1.925)(B7)
+ (0.440)(C) + (1.799)(GAS) +
(1.343)(SMOKE) + (0.440)(B1)(C)
+ (0.897)(B4)(C).
(5-4)
The R value for this model is 0.1764; thus, Equation 5-4 is superior to Equa-
tion 5-2 in explaining the observed variation in PEM values. The addition of
C to the general model appears to be the main cause of this improvement.
101
-------
It is interesting to note that Equations 5-2 and 5-4 both contain microen-
vironment group codes Bl, B4, and B7 and no other microenvironment group
codes. The appearance of interaction terms (B1)(C) and (B4)(C) in Equation
5-4 suggests that exposures in these microenvironment groups reflect simul-
taneous fixed-site data to some degree. However, Table 5-3 demonstrates that
2
these terms add little to the overall R value.
Figures 6-8, 6-9, and 6-10 in the Denver CO report suggest 1) that the
composite fixed-site value exceeds average exposure between 6:00 and 10:00
and 2) that average personal exposure exceeds the composite fixed-site value
between 10:00 and 17:00. Consequently, two new explanatory variables (Tl and
T2) were defined as follows:
Tl = 1 if 6:00 <_ time < 10:00
Tl = 0 otherwise;
T2 • 1 1f 10:00 <. time < 17:00
T2 - 0 otherwise.
General Model 3 is
(B2)(B4)' + (B3)(B7)
(64) (C) + (65) (GARAGE) + (B6)(GAS)
(S7)(SMOKE) + (B8)(B1)(C) + (Bg)(B4)(C)
(B10)(B7)(C) + (Bn)(Tl) + (B12)(T2)
+ (B14)(T2)(B1) +
(B16)(T2)(C). (5-5)
Note that only microenvironment group codes Bl, B4, and B7 appear in this
general model. The results of the stepwise regression analysis are listed
in Table 5-4. The suggested "best-fit" model is
= 0.845 + (4.576)(B1) + (4.977)(B4)
+ (1.010KO + (1.983)(GAS) + (1.322)(SMOKE)
- (0.521)(B7)(C) - (0.252)(T1)(C). (5-6)
102
-------
TABLE 5-4. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 3
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Operation
Add Bla
Add Ca
Add (B7)(C)a
Add SMOKE3
Add GASa
Add (Tl)(C)a
Add B4a
Add GARAGE
Add (T1)(B1)
Add (T2)(B1)
Add B7
Add Tl
Add (B4)(C)
Add (B1)(C)
Add (T2)(C)
Remove (T2)(C)
Remove (B1)(C)
Remove (B4)(C)
Remove Tl
Remove B7
Remove GARAGE
Remove (T2)(B1)
' Remove (T1)(B1)
Resulting R
0.0924
0.1441
0.1589
0.1690
0.1734
0.1782
0.1829
0.1835
0.1841
0.1851
0.1856
0.1860
0.1861
0.1864
0.1865
0.1864
0.1861
0.1860
0.1856
0.1851
0.1846
0.1835
0.1829
2
Change in R
0.0924
0.0517
0.0148
0.0101
0.0044
0.0048
0.0046
0.0007
0.0006
0.0010
0.0005
0.0003
0.0002
0.0003
0.0001
-0.0001
-0.0003
-0.0002
-0.0003
-0.0005
-0.0006
-0.0010
-0.0006
Retained in best-fit model.
2
The R value obtained with this model is 0.1829—a slight improvement over
Equation 5-4. Note that T2 does not appear in Equation 5-6.
/\
General Model 4 is the same as General Model 3 except that ln(cpEM) is
substituted for cpEM in Equation 5-5. The results of the stepwise regression
analysis are listed in Table 5-5. The suggested "best-fit" model is
ln(CREM) = - 1.033 + (1.703)(B1) + (1.361)(B4)
- (0.613)(B7) + (0.337)(C) - (0.196)(GARAGE)
+ (0.897)(GAS) + (0.788)(SMOKE) - (0.133)(B1)(C)
- (0.600)(T1) - (0.073)(T1)(C) + (0.825)(T1)(B1). (5-7)
2
The R obtained with this model is 0.2694--a significant improvement over
S*
Equation 5-6. Note that the log transformation of cDrM is totally responsible
for this improvement in R .
PEM
103
-------
TABLE 5-5. RESULTS OF STEPWISE LINEAR REGRESSION
USING GENERAL MODEL 4
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Operation
Add Ca
Add Bla a
Add SMOKE3
Add Tla
Add B7a
Add GASa
Add B4a
Add (Tl)(Caa
Add GARAGE3 a
Add (Tl)(Bl]a
Add (Bl)(C)a
Add (T2)(C)
Add T2
Add (B4)(C)
Add (T2)(B1)
Add (B7)(C)
Remove (B7)(C)
Remove (T2)(B1)
Remove (B4)(C)
Remove T2
Remove (T2)(C)
2
Resulting R
0.1190
0.2066
0.2315
0.2449
0.2543
0.2620
0.2640
0.2653
0.2666
0.2677
0.2694
0.2703
0.2710
0.2715
0.2718
0.2718
0.2718
0.2715
0.2710
0.2703
0.2694
2
Change in R
0.1190
0.0876
0.0248
0.0135
0.0094
0.0077
0.0020
0.0013
0.0013
0.0011
0.0018
0.0009
0.0006
0.0005
0.0003
0.0000
-0.0000
-0.0003
-0.0005
-0.0006
-0.0009
Retained in best-fit model.
General Models 5 through 14 are discussed in Section 5.2. Section 5.3
describes a method for aggregating indoor microenvironments which was
implemented prior to the analyses discussed in Section 5.2.
5.2 GENERAL MODELS 5 THROUGH 14
2
The analyses described in Section 5.1 yielded a model with an R value
of 0.27 (i.e., a model which explains 27 percent of the variation in PEM
values). The model contains terms corresponding to various activity diary
entries, fixed-site data, and time of day. Subsequent to these analyses, PEI
performed the analyses described in Sections 2.2, 4.2, and 4.3. Results of
the analyses suggest that wind speed; maximum daily temperature; exposure
duration; and use of gas furnaces, gas cooking stoves, and gas or kerosene
space heaters in a participant's home make significant contributions toward
explaining variations in personal exposure. The analysis described in Sec-
tion 5.3 provided alternatives to the aggregate indoor microenvironments
104
-------
considered in General Models 1 through 4. After considering these results,
PEI developed a new set of candidate exposure factors, which are listed in
Table 5-6.
These factors were combined in various ways to form the independent
variables of General Models 5 through 14. PEM value or a function of PEM
value is the dependent variable. Stepwise linear regression (weighted) with
forward and backward stepping was performed on each general model to deter-
mine a "best-fit" model containing only those terms which make significant
contributions to explaining the observed variation in the dependent variable.
As in Section 5.1, most of the exposure factors are treated as binary varia-
bles. Note that five of the new factors (DUR, Cl, C2, WIND, and'TMAX) are
treated as continuous variables.
General Model 5 is
CpEM = a + (B^tAl) + (B2)(A2) + ... +
(36)(A6) + (B7)(M1) + (B8)(M2) + ... +.
(.614)(M8) + (B15.) (GARAGE) + (Blg)(GAS) + (B17)_(SMOKE) +
(B18)(DUR) + (B19)(PD + (B20)(P2) + (B21)(P3) + (B22)(P4) +
(B23)(GFN) + (824)(GCS) + (B25)(SHN) + (B26)(IH) + (B27)(C) +
(B28)(WIND)-1 + (B29(TMAX). (5-8)
Note that this model contains all factors listed in Table 5-6 except Cl and
C2 and that there are no interactions among the factors. Table 5-7 lists the
sequence of steps followed by BMDP program P2R in adding and removing terms
from the model. The suggested "best- fit" model is
= 1.86 + (6.38)(M1) - (1.27)M3)
- (1.62)(M4) + (4.07)(M5) + (1.64)(6AS)
+ (1.21)(SMOKE) - (0.0141)(DUR)
- (0.82)(P2) + (0.75)(6CS)
+ (0.48)(C) - (6.62)(WIND)"1. (5-9)
105
-------
TABLE 5-6. CANDIDATE EXPOSURE FACTORS USED IN STEPWISE LINEAR
REGRESSION ANALYSES INVOLVING GENERAL MODELS 5 THROUGH 14
Data source
Factor
code
Description
Responses coded
as 1 1n file
Diary item A
(Activity)
Diary item B
+ D3
(Microen-
vironment)
Al
A2
A3
A4
A5
A6
Ml
M2
Travel-related activities
Indoor activities near
CO sources
Other indoor activities
Outdoor activities
Work-related activities
Other activities
Very high indoor exposures
High indoor exposures
01:all travel
03:cooking
04:laundry
08: eating
12:cafe or pub
05:other indoor chores and
child care
ll:social, political, or
religious activities
09-.sleeping
10:other personal needs
17:interview
18:final entry
20:begin breath sample
21:end breath sample
06:yard work and other
outdoor activities
13:walking, bicycling, or
jogging
02:work and study
07:errands and shopping
14:other leisure
activities
15uncertain of code
16:no entry in diary
52a:public garage
54bb:service station or
motor vehicle repair
facility
05bb:restaurant
55bb:other repair shop
56bb:auditorium, sports
arena, concert hall,
etc.
58bb:shopping mall
62bb:other location
(continued)
106
-------
TABLE 5-6 (continued)
Data source
Factor
code
Description
Responses coded
as 1 in file
M3
M4
M5
M6
M7
M8
Diary item El
Diary item E2
Diary item F
(continued)
GARAGE
GAS
SMOKE
Medium indoor exposures
Low indoor exposures
High in-transit exposures
High outdoor exposures
Medium in-transit and
outdoor exposures
Low in-transit and out-
door exposures
Garage attached to
building?
Gas stove in use?
Smokers present?
03c:office
04bb:store
51bbresidential garage
61bb:other public building
02bb:residence
53bb:manufacturing
facility
57bb:church
59bb:health care facility
60bb:school '
0102:car
0103:bus
0104:truck
0193:motorcycle
71bb:residential garage or
carport
72a:public garage
0101:walking
07c:wdth 10 yards of road
73d:parking lot
74bb:service station or
motor vehicle repair
facility
80bb:other location
0192:bicycle
75bb:construction site
76bb:residential grounds
77bb:school grounds
78bb:sports arena, amphi-
theater, etc.
79bb:park of golf course
01:yes
01:yes
01:yes
.107
-------
TABLE 5-6 (continued)
Data source
Factor
code
Description
Responses coded
as 1 in file
Diary time
entry
Questionnaire
item 4c
Questionnaire
item 4d
Questionnaire
item 4g
Special file
Fixed-site
data file
Meteorologi-
cal data
file
DUR
PI
P2
P3
P4
GFN
GCS
SHN
IH
Cl
C2
WIND
TMAX
Duration of exposure in
minutes
Period from 24:00 to 6:00
Period from 6:00 to 10:00
Period from 10:00 to 16:00
Period from 16:00 to 24:00
Gas furnace used and not
vented
Gas stove used (vented or
not vented)
Gas or kerosene space
heater used and not
vented
Indoors at home
Simultaneous concentration
at the nearest fixed site
for non-trans it ME's and
at composite site for in-
transit ME's
C value one hour earlier
C value two hours earlier
Daily mean wind speed, mph
Daily maximum temperature,
Continuous variable
24:00 < time < 6:00
6:00 Ł time < 10:00
10:00 <_ time < 16:00
16:00 <_ time. < 24:00
2:used and not vented
l:used and vented
2:used and not vented
2:used and not vented
B = indoors residence when
census- tract is "home"
census tract
Continuous variable
Continuous variable
Continuous variable
Continuous variable
Continuous variable
Includes D3 = bb, 01, and 02.
'Blank.
Includes D3
Includes D3
"See Section
= bb and 01.
= bb, 01, 02, and 03.
4.3.2.
108
-------
TABLE 5-7. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 5
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
Operation
Add M5a
Add Ca
Add M4a
Add Mla
Add GASa
Add P2a
Add SMOKE3
Add M3a
Add GCSa
Add DURa
Add (WIND)'1(a)
Add GARAGE
Add TMAX
Add A4
Add SHN
Add M6
Add A6
Add M8
Add IH
Add A5
Add M2
Add PI
Remove PI
Remove M2
Remove A5
Remove IH
Remove M8
Remove A6
Remove M6
Remove SHN
Remove A4
2
Resulting R
0.0924
0.1441
0.1565
0.1627
0.1676
0.1706
0.1732
0.1763
0.1785
0.1808
0.1828
0.1838
0.1847
0.1854
0.1858
0.1862
0.1864
0.1866
0.1868
0.1869
0.1870
0.1870
0.1870
0.1869
0.1868
0.1866
0.1864
0.1862
0.1858
0.1854
0.1847
2
Change in R
0.0924
0.0517
0.0123
0.0062
0.0049
0.0030
0.0026
0.0031
0.0022
0.0023
0.0020
0.0011
0.0009
0.0007
0.0004 '
0.0003
0.0002
0.0002
0.0002
0.0002
0.0001
0.0000
-0.0000
-0.0001
-0.0002
-0.0002
-0.0002
-0.0002
-0.0003
-0.0004
-0.0007
(continued)
109
-------
TABLE 5-7 (continued)
Step
32
33
Operation
Remove TMAX
Remove GARAGE
2
Resulting R
0.1838
0.1828
2
Change 1n R
-0.0009
-0.0011
aRetained in best-fit model .
2
The R value for this model is 0.1828. As indicated in Table 5-7, the varia-
bles 1n Equation 5-9 entered the model in the following order: M5 (high in-
transit exposures), C (simultaneous fixed-site concentration), M4 (low indoor
exposures), Ml (very high indoor exposures), GAS, P2 (6:00 Ł time < 10:00),
SMOKE, M3 (medium indoor exposures), GCS (gas cooking stove in residence), DUR,
and (WIND) . None of the activity codes (Al, A2 ..... A6) were retained in
the model .
General Model 6 consists of the 32 possible interactions between aggregate
microenvironments (Ml, M2, ..., M8) and time periods (PI, P2, P3, and P4); that
is,
pEM = a + (01)(P1)(M1) + (e2)(P1)(M2) + ... +
+ (B9)(P2)(M1) + (B1Q)(P2)(M3) + ... + (B16)(P2)(M8)
+ ... + (B25)(P4)(M1) + (B26)(P4)(M2) + ... + (B32)(P4)(M8)
The best-fit model (R2 = 0.1330) is
(5-10)
'PEM
3.69 + (14.67)(P1)(M2) - (2.43)(P1)(M4) +
(8.16)(P2)(M1) + (3.32)(P2)(M2) - (2.15)(P2)(M4)
+ (5.87)(P2)(M5) + (5.69)(P3)(M1) - (1.95)(P3)(M4)
(3.99)(P3)(M5) + (11.21)(P4)(M1) - (1.00)(P4)(M4)
(3.92)(P4)(M5) .
(5-11)
Table 5-8 lists the sequence of add/remove steps which produced the best-fit
model. Note that terms containing M3, M6, M7, and M8 do not appear in Equation
110
-------
TABLE 5-8. ABRIDGED RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 6
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
•
26
27
28
34
35
36
37
38
39
40
Operation
Add (P3)(M5)a
Add (P2)(M5)a
Add (P4)(M5)a
Add (Pl)(M4)a
Add (Pl)(M2)a
Add (P4)(Ml)a
Add (P2)(M4)a
Add (P3)(M4)a
Add (P4)(M4)a
Add (P2)(Ml)a
Add (P3)(Ml)a
Add (P2)(M2)a
Add (P2)(M3)
e
•
Add (P4)(M8)
Remove (P4)(M8)
Remove (P4)(M6)
e
Remove (P4)(M2)
Remove (P4)(M7)
Remove (P2)(M7)
Remove (P4)(M3)
Remove (P2)(M6)
Remove (P3)(M8)
Remove (P2)(M3)
.Resulting R
0.0324
0.0653
0.0936
0.1014
0.1072
0.1128
0.1178
0.1226
0.1274
0.1298
0.1316
0.1330
0.1341
•
0.1373
0.1372
0.1372
0.1360
0.1357
0.1353
0.1350
0.1346
0.1341
0.1330
2
Change in R
0.0324
0.0329
0.0283
0.0078
0.0058
0.0056
0.0050
0.0049
0.0048
0.0024
0.0019
0.0014
0.0011
0.0000
-0.0000
-0.0001
-0.0004
-0.0003
-0.0004
-0.0003
-0.0004
-0.0005
-0.0011
Retained in best-fit model.
Ill
-------
5-11. Terms containing M5 (high in-transit exposures) contribute 0.0936 to
2
value of the best
General Model 7 is
2
the R value of the best-fit model.
CpEM - a + (B^CPIKC) + (B2)(P2)(C) + (33)(P3)(C) + (B4)(P4)(C)
+ (35)(M1)(C) + (36)(M2)(C) + ... + (312)(M8)(C) . (5-12)
This model is composed of interactions between time period and simultaneous
fixed-site concentration and between aggregate microenvironment and fixed-site
2
concentration. The best-fit model (R = 0.1443) is
CpEM = 1.70 + (0.77)(P1)(C) + (0.64)(P2)(C) + (1.06)(P3)(C) +
(0.91)(P4)(C) + (0.94)(M1)(C) - (0.35)(M3)(C) -
(0.55)(M4)(C) + (0.91(M5)(C) . (5-13)
Table 5-9 lists the add/remove steps of the stepwise regression.
General Model 8 is
CpEM = a + (gj)(GARAGE)(Ml) + (32)(GARAGE)(M2) +
(33)(GARAGE)(M3) + (34)(GARAGE)(M4) + (B5)(GAS)(M1) +
(36)(GAS)(M2) + (B7)(GAS)(M3) + (3g)(GAS)(M4) +
(39)(SMOKE)(M1) + (31Q)(SMOKE)(M2) + ... + (B16)(SMOKE)(M8). (5-14)
It contains interaction terms pairing aggregate indoor microenvironments with
GARAGE and GAS and all eight aggregate microenvironments with SMOKE. The
best-fit model (R2 = 0.0426) is
CpEM = 3.10 + (8.05)(GARAGE)(M1) + (1.62)(GARAGE)(M3) -
(0.85)(GARAGE)(M4) + (2.30)(GAS)(M3) + (1.53)(GAS)(M4) +
(1.73)(SMOKE)(M2) + (7.10)(SMOKE)(M5) . (5-15)
112
-------
TABLE 5-9. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 7
Step
1
2
3
4
5
6
•
7
8
9
10
11
12
13
14
15
16
Operation
Add (C)(M5)a
Add (C)(P4)a
Add (C)(P3)a
Add (C)(Ml)a
Add (C)(M7)
Add (C)(M2)
Add (C)(M3)a
Add (C)(M6)
Add (C)(P2)a
Add (C)(Pl)a
Add (C)(M4)a
Add (C)(M8)
Remove (C)(M2)
Remove (C)(M6)
Remove (C)(M7)
Remove (C)(M8)
2
Resulting R
0.0918
0.1092
0.1219
0.1311
0.1353
0.1390
0.1420
0.1428
0.1434
0.1440
0.1446
0.1448
0.1448
0.1447
0.1446
0.1443
2
Change in R
0.0918
0.0174
0.0127
0.0092
0.0042
0.0037
0.0029
0.0009
0.0006
0.0006
0.0007
0.0002
-0.0001
-0.0001
-0.0001
-0.0003
Retained in best-fit model.
Table 5-10 lists the add/remove steps of the stepwise regression. Only the
2
(SMOKE) (M5) term contributes more than 0.02 to the R value of the best-fit
model .
General Model 9 is
= a + (B1)(IH)(GFM)(P1) + (B2)(IH)(GFN)(P2) + (B3)(IH)(GFN)(P3) +
(B4)(IH)(6FN)(P4) + (B5)(IH)GCS)(P1) + (Bg)(IH)(GCS)(P2) +
(B7)(IH)(GCS)(P3) + (B8)(IH)(GCS)(P4) + (Bg)(IH)(SHN)(Pl) +
(B1Q)(IH)(SHN)(P2) + (Bn)(IH)(SHN)(P3) + (B12)(IH)(SHN)(P4). (5-16)
113
-------
TABLE 5-10. RESULTS OF STEPUISE LINEAR REGRESSION USING GENERAL MODEL 8
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
Operation
Add (SMOKE) (M5)a
Add (GARAGE) (Ml )a
Add (GARAGE )(M4)a
Add (GAS)(M3)a
Add (GARAGE) (M3)a
Add (GAS)(M4)a
Add (SMOKE )(M2)a
Add (SMOKE )(M4)
Add (SMOKE )(M3)
Add ( SMOKE )(M7)
Add (GAS)(M2)
Add (SMOKE) (Ml)
Add (GARAGE )(M2)
Add ( SMOKE )(M8)
Remove ( SMOKE )(M8)
Remove (GARAGE) (M2)
Remove (SMOKE) (Ml)
Remove (GAS)(M2)
Remove ( SMOKE )(M7)
Remove ( SMOKE )(M3)
Remove ( SMOKE )(M4)
2
Resulting R
0.0218
0.0295
0.0347
0.0374
0.0394
0.0411
0.0426
0.0432
0.0435
0.0438
0.0440
0.0441
0.0442
0.0442
0.0442
0.0441
0.0440
0.0438
0.0435
0.0432
0.0426
2
Change in R
0.0218
0.0077
0.0052
0.0028
0.0021
0.0016
0.0014
0.0006
0.0003
0.0003
0.0002
0.0001
0.0001
0.0000
-0.0000
-0.0001
-0.0001
-0.0002
-0.0003
-0.0003
-0.0006
Retained in best-fit model.
114
-------
Each term contains the IH variable which is equal to 1 only when a subject is
indoors at home. The GFN, GCS, and SHN variables indicate, respectively, the
existence of a gas furnace, a gas cooking stove, and a space heater in the
home (Table 5-6). Time periods are also included in the interaction terms.
The best-fit model is
= 3.12 + (2.18)(IH)(6CS)(P4);
(5-17)
the R value is only 0.0062. The add/remove steps are listed in Table 5-11.
Only the term (IH)(GCS)(P4) is retained in the best-fit model.
TABLE 5-11. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 9
Step
1
2
3
4
5
6
7
8
9
10
11
Operation
Add (IH)(GCS)(P4)a
Add (IH)(GFN)(P1)
Add (IH)(GCS)(P2)
Add (IH)(SHN)(P4)
Add (IH)(6CS)(P1)
Add (IH)(GFN)(P4)
. Remove (IH)(GFN)(P4)
Remove (IH)(GCS)(P1)
Remove (IH)(SHN)(P4)
Remove (IH)(GCS)(P2)
Remove (IH)(GFN)(P1)
2
Resulting R
0.0062
0.0064
0.0066
0.0067
0.0068
0.0068
0.0068
0.0067
0.0066
0.0064
0.0062
2
Change in R
0.0062
0.0002
0.0001
0.0001
0.0001
0.0001
-0.0001
-0.0001
-0.0001
-0.0002
-0.0002
Retained in best-fit model.
General Model 10,
CpEM = a +
'1
+ (62)(WIND)"1(M2) + ... +
(68)(WIND)'(M8) + (Bg)[ln(WIND)](Ml)
(e16)[ln(WIND)](M8)
[In(WIND)] (M2)
(5-18)
115
-------
contains interactions between aggregate microenvironment and WIND" and
between aggregate microenvironment and In (WIND). The best-fit model
(R2 = 0.1227) is
CREM = 3.79 + (42.6)(WIND)"1(M1) +
(9.38)(WIND"1)(M2) + (28.7)(WIND"1)(M5) -
(0.913)[ln(WIND)](M4)
Table 5-12 lists the stepwise regression results. These results are consistent
with the findings in Section 2.2, where WIND" was found to be a generally
better predictor of ambient CO levels (and presumably CO exposures) than
In(WIND). The aggregate microenvironments jnost a-ffected by wind speed are M5
(high in-transit exposures), M4 (low indoor exposures), and Ml (very high
indoor exposures). Other general models containing WIND" are evaluated
later in this section.
General Model 11 contains interactions between aggregate microenviron-
ments and C (simultaneous fixed-site values), Cl (fixed-site values one hour
earlier), and C2 (fixed-site values two hours earlier). The general model is
+ (B2)(C)(M2) + ... + (38)(C)(M8) +
(B1(J)(C1)(M2) + ... + (B16)(C1)(M8) +
(617)(C2)(M1) + (B18)(C2)(M2) + ... + (B24)(C2)(M8). (5-20)
The best-fit model (R2 = 0.1470) is
= 1.66 + (1.77)(C)(M1) + (0.48)(C)(M3) + (1.99)(C)(M5) +
(1.01)(C)(M6) + (0.93)(C)(M7) + (0.93)(C1)(M2) +
(0.34)(C1)(M4) - (1.18)(C1)(M5) + (1.11)(C2)(M5). (5-21)
Table 5-13 lists the stepwise regression sequence. Note that (C1)(M2) and
(C1)(M4) appear in the best-fit model and that (C)(M2) and (C)(M4) do not.
Apparently PEM values for M2 (high indoor exposures) and M4 (low indoor
exposures) are better predicted by fixed-site values one hour earlier than
the PEM reading than by simultaneous fixed-site values. However, (C1)(M2) and
116
-------
TABLE 5-12. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 10
Step
1
2
3
4
5
6
7
8
9 -
10
11
12
13
14
15
16
17
18
19
20
Operation
Add (WINDr^MS)3
Add [ln(WIND)](M4)a
Add (WIND)"1(Ml)a
Add [ln(WIND)](M3)
Add [ln(WIND)](M8)
Add [ln(WIND)](M7)
Add (WINOr^MS)
Add (WIND)'1(M2)a
"Add [ln(WIND)](M5)
Add (WIND)"1(M4)
Add (WIND)"1(M7)
Add (WIND)"1(M3)
Remove [In(WIND)] (M3)
Remove (WIND)"1(M3)
Remove [In(WIND)] (M7)
Remove (WIND)"1(M7)
Remove [In(WIND)] (M8)
Remove (WINDj'^MG)
Remove (WIND)"1(M4)
Remove [In(WIND)] (M5)
Resulting R
0.0908
0.1132
0.1212
0.1228
0.1238
0.1246
0.1249
0.1251
0.1254
0.1259
0.1262
0.1262
0.1262
0.1262
0.1260
0.1258
0.1253
0.1247
0.1237
0.1227
2
Change in R
0.0908
0.0224
0.0080
0.0016
0.0009
0.0008
0.0003
0.0002
0.0003
0.0005
0.0003
0.0001
-0.0001
-0.0000
-0.0002
-0.0002
-0.0005
-0.0007
-0.0010
-0.0010
Retained in best-fit model.
117
-------
TABLE 5-13. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 11
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
Operation
Add (C)(M5)a
Add (C)(Ml)a
Add (C)(M2)a
Add (Cl)(M4)a
Add (C)(M3)a
Add (C)(M7)a
Add (C2)(M5)a
Add (Cl)(M5)a
Add (C)(M6)a
Add (C2)(M3)
Add (C2)(M4)
Add (C2)(M2)
Add (C1)(M1)
Add (C)(M4)
Remove (C)(M4)
Remove (C1)(M1)
Remove (C2)(M2)
Remove (C2)(M4)
Remove (C2)(M3)
2
Resulting R
0.0944
0.1046
0.1128
0.1206
0.1316
0.1406
0.1436
0.1456
0.1470
0.1482
0.1486
0.1488
0.1489
0.1489
0.1489
0.1488
0.1486
0.1482
0.1470
2
Change in R
0.0944
0.0102
0.0082
0.0078
0.0111
0.0090
0.0031
0.0020
0.0014
0.0012
0.0004
0.0001
0.0001
0.0001
-0.0001
-0.0001
-0.0001
-0.0004
-0.0012
Retained in best-fit model.
11B
-------
(C1)(M4) together add only 0.0160 to the R2 value of the best-fit model. The
2
sole term containing C2—(C2)(M5)— contributes only 0.0031 to the R value.
Three general models were constructed using selected terms retained in
the best-fit models corresponding to General Models 5 through 11. Each term
2
contributed at least 0.0042 to the R value of the corresponding best-fit
model. These terms are listed in Table 5-14. Because of the large number of
terms, the complete general models will not be listed here. However, they
all have the form
F(CpEM) = a + (BjMMS) + (32)(C) + ... + (C)(M3). ' (5-22)
^ A A
The dependent variable tF(CpEM)] is CpEM for General Model 12, ln(CpEM) for
General Model 13, and (CREMX - 1)/X for General Model 14. The best-fit
model (R2 = 0.2013) for General Model 12 is
CpEM = 2.22 + (3.50)(M5) + (1.39)(GAS)
+ (15.9)(P1)(M2) - (1.54)(P4)(M5)
+ (0.86)(C)(M5) + (0.16)(C)(P4)
+ (1.63)(C)(M1) + (0.73)(C)(M7)
+ (2.92)(SMOKE)(M5) + (2.45)(IH)(GCS)(P4)
- (0.75)[ln(WIND)](M4) + (0.64)(C1)(M2)
+ (0.34)(C1)(M4) + (0.28)(C)(M3). (5-23)
2
Table 5-15 lists the stepwise regression sequence. The best-fit model (R =
0.3046) for General Model 13 is
ln(CpEM) = -0.672 + (0.12)(C) + (1.93)(M1) + (0.59)(6AS) -
(0.38)(P1)(M4) - (0.76)(P2)(M4) + (1.62)(P2)(M5) +
(1.56)(P3)(M5) + (1.57)(P4)(M5) + (0.18)(C)(M7) +
(0.65)(SMOKE)(M5) - (0.25)(GARAGE)(M4) +
(0.90)(IH)(6CS)(P4) - (0.49)[ln(WIND)](M4) +
(0.15)(C1)(M2) + (0.22)(C1)(M4) + (0.11)(C)(M3) . (5-24)
Table 5-16 lists the stepwise regression sequence.
119
-------
TABLE 5-14. TERMS SELECTED FROM GENERAL MODELS 5 THROUGH 11 FOR
INCLUSION IN GENERAL MODELS 12, 13, AND 14
General model
5
6
7
8
9
10
11
Term
M5
C
M4
Ml
GAS
(P3)(M5)
(P2)(M5)
(P4)(M5)
(P1)(M4)
(P1)(M2)
(P4MM1)
(P2)(M4)
(C)(M5)
(C)(P4)
(C)(P3)
(C)(M1)
(C)(M7)
(SMOKE) (M5)
(GARAGE) (Ml)
(GARAGE )(M4)
(IH)(GCS)(P4)
(WIND)-1(M5)
[ln(WIND)](M4)
(WIND)"1 (Ml)
(C)(M5)a
(C)(Ml)a
CC1)(M2)
(C1)(M4)
(C)(M3).
(C)(M7)a
2
Contribution to R
0.0924
0.0517
0.0123
0.0062
0.0049
0.0324
0.0329
0.0283
0.0078
0.0058
0.0056
0.0050
0.0918
0.0174
0.0127
0.0092
0.0042
0.0218
0.0077
0.0052
0.0062
0.0852
0.0224
0.0080
0.0944
0.0102
0.0082
0.0078
0.0111
0.0090
Term also listed under General Model 7.
120
-------
TABLE 5-15. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 12
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
Operation
Add M5a
Add C
Add [1n(WIND)](M4)a
Add (IH)(GCS)(P4)a
Add (Pl)(M2)a
Add (C)(Ml)a
Add (C)(M5)a
Add (SMOKE) (M5)a
Add (Cl)(M4)a
Add (Cl)(M2)a
Add (C)(M7)a
Add GASa
Add (P2)(M4)
Add (WINDr^Ml)
Add (C)(M3)a
Add (P4)(M5)a
Add (C)(P4)a
Add ( GARAGE )(M4)
Add (WINDr^MS)
Add (C)(P3)
Add (P4(M1)
Add Ml
Add M4
Remove M4
Remove Ml
Remove (P2)(M4)
Remove C
Remove (P4)(M1)
Remove (WIND)"1^)
Remove (C)(P3)
Remove ( GARAGE )(M4)
Remove (WIND)'1^!)
Resulting R
0.0953
0.1475
0.1584
0.1693
0.1754
0.1811
0.1849
0.1885
0.1905
0.1929
0.1955
0.1977
0.1989
• 0.1999
0.2009
0.2018
0.2028
0.2036
0.2042
0.2045
0.2047
0.2049
0.2049
0.2049
0.2047
0.2045
0.2044
0.2042
0.2038
0.2032
0.2024
0.2013
2
Change in R
0.0953
0.0522
0.0109
0.0109
0.0061
0.0057
0.0038
0.0037
0.0020
0.0024
0.0026
0.0022
0.0012
0.0010
0.0009
0.0009
0.0010
0.0009
0.0005
0.0003
0.0002
0.0002
0.0001
-0.0001
-0.0002
-0.0002
-0.0001
-0.0002
-0.0005
-0.0005
-0.0009
-0.0011
Retained in best-fit model.
121
-------
TABLE 5-16. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 13
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
l_ Operation
Add [ln(WIND)](M4)a
Add Ca
Add M5
Add CCl)(M4)a
Add (IH)(GCS)(P4)a
Add (P2)(M4)a
Add GASa
Add (Pl)(M4)a
Add ( GARAGE )(M4)a
Add Mla
Add ( SMOKE )(M5)a
Add (C)(M7)a
Add (C)(M3)a
Add (Cl)(M2)a
Add CC)-(M5)
Add M4
Add (P1)(M2)
Add (P3)(M5)a
Add (C)(P4)
Add (C)(P3)
Add (W1ND)"1(M1)
Add (P2)(M5)a
Add CP4)(M5)a
Add (WINDj'^Ml)
Remove M5
Remove (C)(P3)
Remove (C)(P4)
Remove (P1)(M2)
Remove M4
Remove (C)(M5)
Resulting R
0.1276
0.2221
0.2470
0.2661
0.2802
0.2883
0.2918
0.2947
0.2968
0.2988
0.3001
0.3009
0.3023
0.3041
0.3047
0.3053
0.3054
0.3055
0.3056
0.3057
0.3057
0.3057
0.3061
0.3060
0.3060
0.3059
0.3058
0.3057
0.3051
0.3046
Change in R
0.1276
0.0945
0.0249
0.0191
0.0142
0.0081
0.0035
0.0029
0.0022
0.0020
0.0013
0.0008
0.0014
0.0018
0.0006
0.0006
0.0001
0.0001
0.0001
0.0001
0.0000
0.0000
0.0003
-0.0000
-0.0001
-0.0001
-0.0001
-0.0001
-0.0006
-0.0006
Retained in best-fit model.
122
-------
Four values of X (0.30, 0.35, 0.40, and 0.45) were used in exploratory
stepwise linear regression analyses to determine the optimal X for the "Box-
2
Cox" function in General Model 14. The largest R value (0.3366) resulted
from X = 0.40. The corresponding best-fit model is
°°40
(CpEM° - 1J/0.40 = -0.068 + (0.11)(C) + (0.68)(6AS) -
(0.32)(P1)(M4) - (0.65)(P2)(M4) + (1.56)(P2)(M5) +
(1.72)(P3)(M5) + (1.43)(P4)(M5) + (0.19)(C)(M5) +
(0..27)(C)(M7) + (1.05)(SMOKE)(M5) +
(1.15)(IH)(6CS)(P4) - (0.51)[ln(WIND)](M4) +
(15.2)(WIND)"1(M1) + (0.23)(C1)(M2) +
(0.22)(C1)(M4) +
(0.13)(C)(M3). (5-25)
Table 5-17 lists the stepwise regression sequence. Note that this best-fit
2
model yields the highest R value yet obtained. The three most important
2
terms with respect to increasing R are related to wind sp'eed given to
low exposure indoor microenvironment [In(WIND)] (M4), to simultaneous
fixed-site readings (C), and to high exposure in-transit microenvironments
(MS).
Because wind speed appears to be particularly significant in
explaining the variation in PEM values, future models may benefit from
higher resolution wind data. For example, five-minute average windspeeds
collected every three hours are listed in monthly summaries published by
the National Weather Service. A reasonable assumption is that eight
windspeed readings per day would provide better resolution of wind-
related effects than WIND which is a 24-hour average.
5.3 AGGREGATION OF INDOOR MICROENVIRONMENTS
PEI was directed by EMSL to compare concentrations observed in different
indoor locations (e.g., residence, school, office) and to test for stati-
stically significant differences between the locations. The results of these
123
-------
TABLE 5-17. RESULTS OF STEPWISE LINEAR REGRESSION USING GENERAL MODEL 14
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
Operation
Add [ln(WIND)](M4)a
Add Ca
Add M5
Add (IH)(GCS)(P4)a
Add (Cl)(M4)a
Add (P2)(M4)a
Add GASa
Add (WIND)"1^!)3
Add ( SMOKE )(M5)a
Add (Pl)(M4)a
Add (C)(M7)a
Add (Cl)(M2)a
Add (C)(M5)a
Add (C)(M3)a
Add (C)(M1)
Add (P1)(M2)
Add (P3)(M5)a
Add (C)(P4)
Add (C)(P3)
Add (P2)(M5)a
Add (P4)(M5)a
Add ( GARAGE )(M4)
Add M4
Add Ml
Add (P4)(M1)
Remove (P4)(M1)
Remove Ml
Remove M5
Remove M4
2
Resulting R
0.1430
0.2384
0.2806
0.3005
0.3097
0.3159
0.3199
0.3236
0.3264
0.3280
0.3296
0.3312
0.3328
0.3357
0.3364
0.3368
0.3372
0.3375
0.3378
0.3382
0.3385
0.3388
0.3389
0.3390
0.3390
0.3390
0.3389
0.3389
0.3387
2
Change in R
0.1430
0.0954
0.0422
0.0200
0.0091
0.0062
0.0040
0.0037
0.0028
0.0016
0.0016
0.0016
0.0015
0.0030
0.0006
0.0005
0.0004
0.0003
0.0003
0.0003
0.0003
0.0003
0.0002
0.0000
0.0000
-0.0000
-0.0000
-0.0001
-0.0002
(continued)
124
-------
TABLE 5-17 (continued)
Step
30
31
32
33
34
Operation
Remove ( GARAGE )(M4)
Remove C(C)(P3)
Remove (C)(P4)
Remove tPl)(M2)
Remove CO(Ml)
Resulting R
0.3385
0.3381
0,3376
0.3372
0.3366
2
Change in R
-0.0003
-0.0004
-0.0004
-0.0005
-0.0006
aRetained in best-fit model.
tests were to be used to identify locations with similar pollutant concentra-
tion distributions. This section describes a methodology developed by PEI
which uses the statistical technique known as pairwise comparisons to
aggregate similar microenvironments into groups which differ significantly
one from another.
5.3.1 The Indoor Microenvironments
Associated with each PEM value is a two-digit location code. Sixteen
of these codes correspond to indoor microenvironments. Table 5-18 lists these
microenvironments and provides selected summary statistics based on data with
acceptable overall (i.e., PEM plus activity diary) quality codes. Data
quality codes are explained in Section 4.11 of Reference 1. The minimum
value reported for each microenvironment was zero. Note the large values of
skewness and kurtosis listed for most microenvironments. To facilitate the
use of statistical analyses requiring normal distributions, PEI investigated
the use of the Box-Cox transformation as a means of reducing skewness and
kurtosis. The general form of the transformation is
y = (XA -
(5-29)
125
-------
TABLE 5-18. SUMMARY STATISTICS FOR CARBON MONOXIDE CONCENTRATION VALUES RECORDED
BY PERSONAL EXPOSURE MONITORS IN INDOOR MICROENVIRONMENTS
Code
02
03
04
05
51
52
53
54
55
56
57
58
59
60
61
62
Indoor
mi croenv i ronment
Residence
Office
Store
Restaurant
Residential garage
Public garage
Manufacturing
facility
Service station or
auto repair
facility
Other repair shop
Auditorium
Church
Shopping mall
Health care facility
School
Other public building
Other indoor location
n
21518
2287
734
524
66
115
42
125
55
100
179
58
351
426
115
425
Carbon monoxide concentration, ppm
Maximum
76.4
59.1
56.3
35.0
28.4
81.2
8.0
73.1
33.1
31.2
21.7
33.9
31.3
21.6
21.8
66.4
Mean
2.212
3.248
3.385
4.313
3.364
11.968
1.750
9.409
7.620
4.523
1.824
5.271
2.334
2.056
2.937
4.923
s.d.a
4.030
4.970
4.754
4.674
5.059
11.984
2.366
9.704
8.575
5.649
2.998
6.493
3.632
3.090
3.760
7.958
s.e.
0.027
0.104
0.175
0.204
0.623
1.118
0.365
0.868
1.156
0.565
0.224
0.853
0.194
0.150
0.351
0.386
Median
0.90
1.90
2.00
3.15
1.40
8.40
0.00
5.80
3.00
3.50
1.00
3.10
1.20
0.80
1.50
2.90'
s.e.b
0.029
0.058
0.115
0.231
0.318
0.635
0.577
1.212
1.617
0.577
0.087
0.462
0.173
0.144
0.491
0.202
Skewness
s.e.
c
110.40
47.05
21.76
9.02
10.85
2.75
13.13
2.98
9.72
19.50
7.35
29.28
24.52
9.77
34.45
Kurtosis
s.e.
c
483.29
167.26
44.53
14.49
19.71
-0.30
31.32
-0.26
13.97
42.54
9.58
79.80
47.92
15.02
85.67
ro
Standard deviation.
'standard error.
'Not computed because of large sample size.
-------
where y is the transformed value, x is the PEM value, and X is a constant
selected by the user. Evaluation of x values between 0 and 1 suggested
that X = 0.35 was a nearly optimal choice in that it produced low skewness
and kurtosis values for most microenvironments. Table 5-19 lists summary
statistics of the transformed data. Note the dramatic reduction in most
of the skewness and kurtosis values after transformation.
Table 5-20 ranks the microenvironments by mean value before and after
transformation and by median value (transformation has no effect on the rank-
ing of median values). In all three cases, the microenvironment with the
largest value is public garage (Code 52). Other microenvironments which are
ranked high on all lists (though not in the same order) are service station
or auto repair facility (Code 54), other repair shop (Code 55), auditorium
(Code 56), shopping mall (Code 58), and restaurant (Code 05). Manufacturing
facility (Code 53) ranked last in all four lists. This result is somewhat
surprising but may be partially explained by the small sample site (n = 42).
5.3.2 Iterative Aggregation Based on Pairwise Comparisons
Pairwise comparisons were performed on the transformed (x = 0.35) data
using BMDP program P7D. Figure 5-1 is the output of the program. Since there
are 16 microenvironments, there are 120 possible pairings." For each pairing
the program output provides the mean for each microenvironment, the difference
in means, the results of a t test assuming unequal variances ("separate
variance t test"), and the results of a t test assuming equal variances
("pooled variance t test"). Asterisks indicate the signficance level of the
Bonferroni test as explained at the top of the printout.
PEI considered all pairings which were not significant at the 0.05 level
as indicated by the Bonferroni test (i.e., all pairings with a separate
variance t test p value greater than 0.05/number of pairings) as candidates
for aggregation. The pairing in Figure 5-1 with the largest p value (0.9505)
consists of Code 56 (auditorium, etc.) and Code 62 (other indoor location).
Based on the assumption that these two microenvironments were not significantly
different, they were combined into an aggregate microenvironment with Code
5662. The pairwise comparisons analysis was then repeated on the resulting
15 microenvironments. This time the largest p value (0.8930) was associated
with Code 03 (office) and Code 51 (residential garage). These were aggregated
127
-------
TABLE 5-19. SUMMARY STATISTICS FOR CARBON MONOXIDE CONCENTRATION VALUES RECORDED BY
PERSONAL EXPOSURE MONITORS IN INDOOR MICROENVIRONMENTS AFTER TRANSFORMATION
Code
02
03
04
05
51
52
53
54
55
56
57
58
59
60
61
62
Indoor
ml croenvi ronment
Residence
Office
Store
Restaurant
Residential garage
Public garage
Manufacturing
facility
Service station or
auto repair
facility
Other repair shop
Auditorium
Church
Shopping mall
Health care facility
School
Other public building
Other indoor location
n
21518
2287
734
524
66
115
42
125
55
100
179
58
351
426
115
425
Carbon monoxide concentration, ppm
Maximum
10.175
9.055
8.854
7.059
6.360
10.456
3.059
9.975
6.867
6.668
5.532
6.949
6.679
5.518
5.545
9.551
Mean
-0.311
0.469
0.592
1.095
0.431
3.336
-0.814
2.744
1.731
0.967
-0.284
1.511
-0.033
-0.255
0.258
0.983
S.d.a
2.170
2.154
2.111
2.108
2.244
2.085
2.287
2.111
2.833
2.364
1.888
2.017
2.044
2.045
2.205
2.467
S.e.b
0.015
0.045
0.080
0.092
0.276
0.194
0.353
0.189
0.382
0.236
0.141
0.265
0.109
0.099
0.206
0.120
Median
-0.103
0.720
0.784
1.412
0.357
3.161
-2.857
2.430
1.340
1.570
0.000
1.388
0.188
-0.215
0.436
1.290
s.e.b
0.032
0.038
0.074
0.111
0.245
0.157
1.051
0.348
0.637
0.289
0.084
0.265
0.156
0.163
0.357
0.103
Skewness
s.e.
c
0.56
-0.25
-3.21
0.68
2.28
0.91
0.37
-0.27
-0.60
1.12
0.65
1.19
1.65
-0.14
1.55
Kurtosis
s.e.
c
-0.52
-0.62
-1.54
-0.67
0.79
-2.24
1.53
-1.71
-1.15
-0.55
0.62
-1.65
-3.46
-2.26
1.38
rv>
oo
Standard deviation.
Standard error.
°Not computed because of large sample size.
-------
TABLE 5-20. MICROENVIRONMENTS LISTED IN DESCENDING ORDER OF MEAN AND MEDIAN
VALUES BASED ON UNTRANSFORMED AND TRANSFORMED CARBON
MONOXIDE VALUES
Rank
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
Microenvironment
Mean
Data untransformed
52
54
55
58
62
56
05
04
51
03
61
59
02
60
57
53
: Public garage
: Service station
: Other repair
: Mall
: Other indoor
: Auditorium
: Restaurant
: Store
: Res. garage
: Office
: Other public
: Health care
: Residence
: School
: Church
: Manufacturing
Data transformed
52: Public garage
54: Service station
55: Other repair
58: Mall
05: Restaurant
62: Other indoor
56: Auditorium
04: Store
03: Office
51: Res. garage
61: Other public
59: Health care
60: School
57: Church
02: Residence
53: Manufacturing
.Median
52:
54:
56:
05:
58:
55:
62:
04:
03:
61:
51:
59:
57:
02:
60:
53:
Public garage
Service station
Auditorium
Restaurant
Mall
Other repair
Other indoor
Store
Office
Other public
Res. garage
Health care
Church
Residence
School
Manufacturing
129
-------
PftE
NAME
CO
o
PAIPWlf>F COMP»RI SONS \"ONG •")•'Ł fPTY C U. I ( GRr UT I MEANS.
ASTERISKS UENME T»,E LEVELS (F SICMtlCAKCE OF TKE
TESTS.
TH VALUE GIVCf F"R THE bCf f trFr*C I TEST If T*E 51
SIe*lFlCAflT P \i»LUL OF COP0*"! ?0\S CF ALL PA17S
THAT IS, AFTL" *bJim"LNT fff THt PLITIFLL COMFAPISON OF ALL
PAI?S OF "EANSt TO fL SIGNIFICANT AT THE .15 LEVEL THE
V»ll'E FUST (L LESC THAf .r».rii7
MEAN
.17
.59
1 .1"
.13
3.31
- .et
2.71
1 .77
.97
- .2f
1.51
-.03
-.26
.26
.98
.59
1 .1"
.13
3.31
-.81
2.7*
1 .73
.97
- .2f
1.51
-.31
-. 31
-.31
-. 31
-.31
-.31
-. 31
-.31
-. 31
-.31
-. 31
-. 31
-.31
-. 31
-.31
. 17
.47
. 17
.01
.17
. 1
2 7 1 f. ; 1
27101
27101
271P1
P-VALUt
. tnoo
.0000
.COOO
.0055
.0000
.1333
.0000
.COOO
.0000
.8653
.000"
.0169
.5960
.0051
. OCOO
.1818
.area
.3889
.CPOO
.0002
.0000
.cnoi
.0218
.0001
.0001
.0001
.CPOP
.3377
. CC01
.0001
.5613
.O^OO
.oroi
.0"00
.0*502
.1019
. or n i
.0020
.cnoi
.0000
.1713
.fC31
.0191
.COCO
*»•
**•
• •*
* •»
«••
**
• *
**
• •»
»•
• •*
•
• *•
**
»•
*
«
• •
• *
•
*•*
»
• * *
*
4 •
• *
• *
• •«
Figure 5-1. Output of pairwise comparison test program (step one), (continued)
-------
51
* 1
51
51
51
51
51
51
51
5?
5'?
3?
5?
5?
5P
5'
53
52
5"
54
51
51
c t
5f
5'-
i .1 '•
1.1-3
1.1-
1. I"1
1.1 •'
1. 1 '
1.1-1
1. 1C
i .1'-
.13
.43
.13
.13
.13
.13
. 13
.13
.13
.13
3. 31
3.31
3. 31
3.34
3.34
3. 31
3.31
3. 31
3.31
3. 31
-.81
-. 81
-.Ł1
-. eil
-.61
-. 81
-.61
-. LI
-.61
.74
. 71
.74
. 71
.71
2. 74
2. 74
2.7-1
1. 73
1 .72
1. 7.1
1 .73
1. 73
i .73
1. 71
. 97
.97
'4
5 5-
56
f. 7
JF
59
t'1
61
t?
D2
53
51
S C
5*
57
5fc
5 9
t<~'
61
12
53
54
55
56
57
58
5-9
61
ol
6?
54
55
56
57
56
59
6f
61
62
c f
56
57
58
59
60
61
t2
56
57
56
cc
61
tl
63
57
5E
-.fcl
2 .74
1. 7'
.97
-.2?
1 .5-1
-. d?
- ,2t
. 2t
,9P
3.31
- .61
2.74
1.73
.97
-.28
1 .51
-.07
- .26
.26
.98
-.81
2 .71
1.7?
.97
-.28
1.51
- .C3
-.26
.26
.98
2 .74
1.73
.97
-.28
1 .51
-.03
- .26
.26
.9?
1 .72
.97
- .26
1.51
- .0?
-.26
. 26
.96
.97
- .26
1.51
- .t-2
-.26
.'it
.98
-.28
1 .51
1.
-1 .
- .
.
1.
- «
1 .
1 .
.
.
-2.
1 .
-2.
- 1 .
- .
.
-1 .
.
.
.
- .
4.
.
1.
2.
3.
1.
3 .
3.
3.
2.
-3.
-2.
-1 .
~ *
-2.
- .
- .
-1.
-1 .
1 .
1 .
3.
1.
2.
3.
2.
1 .
.
2.
.
1 .
1.
1 .
.
1.
- .
91
t 5
ol
13
3c
12
13
35
Ł1
11
91
1-5
21
30
51
72
L'6
16
69
17
55
15
59
61
i7
62
83
27
59
08
35
56
55
78
53
23
78
16
07
Ł0
01
73
u3
23
76 i
00
19
76
76
. 2
22
76
99
47
75
f. 5
51
5.23
-7.15
-1.62
.51
8.19
-1 .If
7 . 90
9.96
3. 72
.7*
-8.60
2 .76
-6.91
-2.76
-1.17
2.31
-2.82
1.56
2.24
.50
-1 .62
10.30
2.19
3.74
7.74
15.07
5.56
15.1 1
16.46
1C. tl
10.31
-8.61
-4.89
-4 .IS
-1.40
-5.27
-2.12
-1 .52
-2.62
-4 .82
2.36
5.87
12.64
3.79
12 .72
14.06
8.91
7.Łf
1.70
4.95
.47
4.44
5. S3
3.40
1.87.
4.54
-1 .52
46
167
60
13U
"14 '3
71
766
S19
162
Ł36
127
86
125
.75
.It
.44
.74
.91
C -I
.16
. .:•- 1
.9>l
.65
.54
.21
.57
1 07.05
144
100
121
86
82
133
91
67
236
82
199
225
1 17
190
177
227
208
66
94
79
54
81
49
47
70
50
61
200
247
1 15
212
1 97
234
232
95
69
97
63
61
66
65
170
134
.29
.83
.92
.47
.60
.58
.2i
.44
.79
.97
.12
.56
.90
.Of
.67
.30
.66
.04
.68
.42
.B3
.57
.16
.6<
.62
.91
.49
.45
,3S
.91
.21
.34
.22
.66
.64
.34
.12
.10
.43
.27
.04
.06
.77
. J 0 0 -"! • • •
« w 'J O fi * * *
.11J7
.6134
.OjuO «••
.1 42f
. OJul • «»
. o 9 u f- •«*
.0003 *
.45t3
.OOiO «*»
.OJo7
.u3uC- *•»
.0069
.1131
.0232
.0356
.12iO
.U217
.6150
.07JC
.0000 *»•
.0296
.3003 *
.0000 »»*
.O-JOO «J.
.03 JO *•*
.OOUO • »•
.0000 «**
.0000 »••
.0000 «»»
.0000 ••»
.0003 •••
.uJGl «»
.1686
.0000 • »»
.C394
.1339
.0106
.oOJf' «»
.0199
.00 JO ...
.OOUO ••*
.0302 »
.uOuf *•*
.0000 •»•
.000° «**
. joijo •»«
. J920
.coon »•»
.6365
.J3u0 ••
.0000 ••«
. U 0 1 n
.0661 '
.0000 • »
.1276
5
-7
— ..
7
-1
7
9
3
-8
t-
-7
-3
-1
2
~t
1
2
-1
10
2
4
E
13
5
14
15
If
10
-9
-5
-4
-1
-5
-2
-1
. -2
-5
2
6
11
3
12
13
6
7
2
• 6
F
6
4
2
4
-1
.49
.64
.07
.54
.35
.39
.54
.55
.75
.79
.68
.SI
.Cl
.28
.56
.29
.77
.59
.39
.52
.92
.62
.12
.52
.C'O
.97
.23
.46
.77
.77
.33
.2:0
.73
.47
.43
.29
.21
.59
.74
.13
.69
.11
.96
.58
.3C
.60
.88
.98
.10
.(•3
.54
.fcl
.40
.15
.41
.62
.52
27104
27104
27104
27104
27104
27104
27104
27104
27101
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
271P4
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27101
27104
27104
27104
27104
27104
27101
27104
27104
27104
27104
27101
27104
27104
.0000
.rooo
.0385
.5870
.0000
.1658
.0000
.0000
.0002
.4285
.0000
.0037
.0000
.0011
. 1195
.0221
.0057
.1109
.0167
.6041
.C544
.0003
.0345
.0001
.CCOO.
.0000
.0000
.COOO
.0000
.rooo
.0000
.nooo
.0000
.0001
.1537
.C001
.0273
.1110
.0062
<>oroi
.0040
.0003
.0000
.0004
.onoo
.0000
.0000
.nnoo
.0357
.0000
.5891
.0000
.0000
.0001
.0161
.0001
.1?83
«»
* •
«o
• »
*»
*
*«
*•
• *»
•
*
»«
*
«»
• #
«*
»«
•
»
*
*•
* *
**
• «•
*
**«
***
*»
»*
*«
**
«*
*
**
Figure 5-1. Output of pairwise comparison test program (step one), (continued)
-------
57
57
57
;7
bl
f F
5P
5 =
59
59
tr
6"1
61
. 97
.97
. 97
.97
-. 26
-.28
-. 20
-.28
-.28
1.51
1. 51
l.t.1
1. 51
-.(13
-. JJ
-. 0?
-.26
-.26
.26
•39
t 1
t?
?9
6"
tl
6?
tr
61
6?
<1
6?
t?
-
-
1
-
-
-
-
-
. J?
• 2Ł
.26
• 9'
.51
.C"
.26
.21
. If
.r.J
.26
.2f
.96
.it
.26
.9P
.26
.9fl
.?"
1
1
-
-1
-
-
-
-1
1
1
1
-
-1
-
-1
-
.J"l
.i?
.71
• Ii2
.79
.25
.03
.54
.77
.b4
.77
.15
.53
.22
.29
.02
.51
.24
.73
3.64
4 .77
2.26
- «i^i
-5.9?
-1 .41
-.17
-2.17
•6.85
5.3S
6.?5
3.74
1.82
1 .51
-1.25
-6.27
-2.21
-7.97
-3.G5-
143.7"
135. St
204.01
153. Ł7
91.58
284. 5 f
360.23
;i5.9f
432.3ft
77. t2
73. 88
123.6V
82.19
747. J3
182.52
773.99
170.6?
820.10
198.24
.1)302 *
.boon »•*
.0247
.95u5
• UO JO * * *
.1596
.8682
• J3J9
.00 JO •••
•oOOO ••*
.0000 •«•
.0003 •
.0730
.1312
.2139
.tnon ***
.0259
.0000 ***
.U026
4
r
2
-
-5
-1
-
-2
-6
C
5
1
1
1
-1
-6
-2
-8
_ i
.07
.17
.39
.07
.48
.26
.15
.'.'9
.56
.02
.82
.59
.74
.42
.25
. 5J
.25
.33
.18
27104
27104
27104
271C4
271C4
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
27104
.0001
. COC1
.0169
.9455
.0000
.2H7C
.3821
.C367
.0000
.C001
.0000
.0004
.0821
.1543
.2128
.0(100
.0244
.0000
.0015
•
• •
• *
• *
• »
• *
*
*•
*•
O*
Figure 5-1. Output of pairwise comparison test program (step one).
-------
into a new microenvironment (Code 0351), and the pairwise comparison analysis
was repeated with 14 microenvironments. This iterative procedure was continued
until the largest p value was found to be less than the critical Bonferroni
value. Table 5-21 summarizes the results of these runs. The procedure termi-
nated at Step 13 which yielded a largest p value less than the indicated
critical value.
TABLE 5-21. RESULTS OF PAIRWISE COMPARISON TESTS
Step
1
2
3
4
5
6
7
8
9
10
11
12
13
Na
120
105
91
78
66
55
45
36
28
21
15
10
6
Largest t-test
p value"
0.9505
0.8930
0.8682
0.6355
0.5632
0.4141
0.3193
0.1612
0.1343
0.0298
0.0169
0.0118
0.0000
Bonferroni
critical p value
0.00042
'0.00048
0.00055
0.00064
0.00076
0.00091
0.00111
0.00139
0.00179
0.00238
0.00333
0.00500
0.00833
, Resulting
group code
5562
0351
5760
5558
025760
055662
035161
02535760
03045161
5254
0555565862
0253575960
None
Number of pairwise comparisons,
'Separate variance t test.
Critical value = 0.05/N.
Four microenvironment groups were obtained through the aggregation
process (Table 5-22). The results are generally consistent with our expecta-
tions. Public garages and service stations make up the group with the
largest mean. The group with the smallest mean contains health care facili-
ties, schools, churches, residences, and manufacturing facilities.
The procedure described above appears to be a .reasonable basis for
partitioning a list of user-defined microenvironments into groups which are
133
-------
TABLE 5-22. MICROENVIRONMENT GROUPS SUGGESTED BY PAIRWISE
COMPARISONS ANALYSIS
Group code
5254
0555565862
03045161
0253575960
n
240
1162
3202
22516
Untransformed
data
Mean
10.635
4.759
3.271
2.207
Std.
dev.
10.909
6.457
4.883
3.998
Transformed
data
Mean
3.028
1.094
0.489
-0.307
Std.
dev.
2.115
2.304
2.148
2.164
Constituent
mi croenvi ronments
Public garage
Service station or
auto repair
Other repair shop
Shopping. mall
Restaurant
Other Indoor location
Auditorium
Store
Office
Residential garage
Other public
building
Health care facility
School
Church
Residence
Manufacturing
facility
statistically similar. In future analyses of the Denver data, this procedure
could be applied easily to alternative sets of indoor microenvironments. For
example, the residential microenvironment could be subdivided according to
gas stove use or type of heating system.
5.4 REFERENCE
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983.
134
-------
SECTION 6
COMPARISON OF CONSECUTIVE DAILY MAXIMUM EXPOSURES
The subjects in the Denver study were requested to participate for two
consecutive 24-hour sampling periods. An analysis of the resulting PEM data
revealed that a pair of valid daily maximum 8-hour exposure values (i.e., one
for each sampling period) could be calculated from the data obtained from each
of 335 subjects. The first of the two values in each pair is hereafter
referred to as the A value; the second of the two values is the B value. PEI
performed a series of statistical analyses to determine 1) if the distribu-
tion of A values differs significantly from the distribution of B values,
2) if the mean difference between paired values differs significantly from
zero, and 3) if there is a high correlation between A and B values. This
section presents the results of these analyses.
6.1 DISTRIBUTIONS OF A AND B VALUES
Table 6-1 lists summary statistics for the A and B values. The mean of
the A values is 4.70 ppm; the mean of the B values is 5.24 ppm. Figure 6-1
presents histograms for the two groups. Because both distributions are
skewed and have large kurtosis values (Table 6-1), PEI investigated taking
the natural logarithms of the exposure values as a means of obtaining more
normal distributions. Table 6-2 lists summary statistics for the transformed
data; Figure 6-2 provides histograms. The values for skewness and kurtosis
are much smaller in Table 6-2 than in Table 6-1, but are still significant.
For this reason, both parametric and nonparametric tests were performed on
the grouped data. Table 6-3 lists the results of these tests.
The Levene test is a test for homogeneity of variance. The large p
value (0.4354) suggests the variances of the logarithms of the two groups
are equal. Consequently, the t (pooled) test is more appropriate than the
t (separate) test for determining if the means of the two groups are equal
135
-------
TABLE 6-1. SUMMARY STATISTICS FOR DAILY MAXIMUM 8-HOUR
CARBON MONOXIDE EXPOSURES
Statistic
Number of cases
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Skewness/std. error
Kurtosis/std. error
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
98th percentile, ppm
99th percentile, ppm
A values
335
0.0
44.0
4.70
a
4.86
28.65
83.04
0.9
1.9
3.5
6.0
9.6
12.0
18.1
24.4
B values
335
0.0
38.7
5.24
3.1
4.65
22.34
55.47
1.1
2.4
4.2
6.9
9.6
12.6
16.9
25.4
All
670
0.0
44.0
5.05
a
4.73
36.48
95.76
1.0
2.2
3.9
6.6
9.7
12.8
17.6
25.0
Not unique.
136
-------
CD
<
20
15
10
A SAMPLING PERIODS
B SAMPLING PERIODS
.-. ,-f
10 15 20 25
DAILY MAXIMUM 8-HOUR EXPOSURE. PPM
30
35
40
45
Figure 6-1. Histograms of daily maximum 8-hour exposures to carbon monoxide.
-------
TABLE 6-2. SUMMARY STATISTICS FOR LOGARITHMS OF DAILY
MAXIMUM 8-HOUR CARBON MONOXIDE EXPOSURES
Statistic
Number of cases
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Skewness/std. error
Kurtosis/std. error
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
98th percentile, ppm
99th percentile, ppm
A values
335
-3.69
3.78
1.11
a
1.09
-11.10
17.17
-0.11
0.64
1.25
1.79
2.26
2.48
2.90
3.19
B values
335
-3.69
3.66 '
1.28
1.13
1.01
-10.14
13.50
0.10
0.88
1.44
1.93
2.26
2.53
2.83
3.23
All
670
-3.69
3.78
1.22
a
1.04
-15.99
23.84
0.00
0.79
1.36
1.89
2.27
2.55
2.87
3.22
Not unique.
138
-------
cc.
UJ
Q-
20
15
10
A SAMPLING PERIODS
B SAMPLING PERIODS
I
n H
-4
-3
-2-1012
LOGARITHM OF DAILY MAXIMUM 8-HOUR EXPOSURE, LN(PPM)
Figure 6-2. Histograms of logarithms of daily maximum 8-hour exposures to carbon monoxide.
-------
TABLE 6-3. RESULTS OF STATISTICAL TESTS COMPARING THE LOGARITHMS OF
A VALUES AND B VALUES
Test
t (separate)
t (pooled)
Levene
Mann-Whitney3
Kruskal-Wallis3
Assumed distributions
Normal, unequal variances
Normal, equal variances
Normal
None
None
Test
statistic
-2.08
-2.08
0.61
49967.00
6.02
D.F.
664
668
1, 668
1
P
0.0381
0.0381
0.4354
0.0141
0.0141
Results are independent of log transformation.
under the assumption of normality. Since p < 0.05 for the t (pooled) test,
it can be concluded that the means of the logarithms are not equal.
The two nonparametric tests (Mann-Whitney and Kruskal-Wallis) also yielded
p values less than 0.05. These results suggest that the null hypothesis that
the two groups have the same distributions should be rejected. Usually when
the null hypothesis is rejected, the assumption is made that one group has
a higher median.
The general conclusion suggested by these analyses is" that the median
concentrations of the two groups differ significantly. The median for the
B values is 4.2 ppm--an increase of 0.7 ppm (202.) over the A value median
of 3.5 ppm. Because both distributions are nonnormal, it Is not possible
to determine if the means of the two groups are significantly different.
6.2 DISTRIBUTION OF DIFFERENCES BETWEEN A AND B VALUES
For each pair of A and B values, there is the difference C = B - A. It
is also possible to determine a difference between the natural logarithms of
the values, i.e., D = ln(B) - ln(A). Table 6-4 lists summary statistics for
the C and D values. The mean of the C values is 0.55 ppm; the mean of the D
values is 0.17. The D values have less skewness and kurtosis than the C
values. If normality is assumed, the one-sample (a.k.a. matched pairs) t
test can be used to determine if the means of the C and D values are signifi-
cantly different than zero. The t statistic for the C values is 1.92 and the
140
-------
TABLE 6-4. SUMMARY STATISTICS FOR DIFFERENCE VALUES
Statistic3
Number of cases
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Skewness/std. error
Kurtosis/std. error
5th percent! le, ppm
10th percenti le, ppm
25th percenti le, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
C = B - A
335
-41.4
32.1
0.55
b
5.21
-9.14
65.66
-6.9
-4.9
-1.6
0.9
2.8
5.3
6.5
D = ln(B) - ln(A)
335
-5.15
5.86
0.17
0.69
1.28
' 0.27
12.14
-2.12
-1.29
-0.43
0.29
0.84
1.44
2.17
aReplace ppm with In(ppm) for D statistics.
Not unique.
p value is 0.0563. This result suggests that the mean of the C values is not
different than zero at the p = 0.05 significance level. In the case of the
D values, the t statistic is 2.42 and the p value is 0.0160. This result
suggests that the mean of the D values is different than zero at the p = 0.05
significance level. Note that D > 0 implies B/A > 1 since D = ln(B/A) and
ln(l) = 0.
Two nonparametric tests were also performed on the paired values. The
sign test yielded a p value of 0.0000 to four decimal places; the Wilcoxon
signed rank test yielded a p value of 0.0004. The sign test results suggest
the A and B values have different medians; the Wilcoxon results suggest that
the difference between A and B values is not zero.
141
-------
6.3 CORRELATION BETWEEN A AND B VALUES
Linear regression analysis with B as the dependent variable yielded the
regression equation
B = 3.44 ppm + (0.383)(A); (6-1)
2 2
R = 0.16. The small R value suggests that the A and B values are not highly
correlated. Repeating the analysis using 1n(B) as the dependent variable and
ln(A) as the independent variable yielded the regression equation
ln(B) = 1.01 + (0.244) [ln(A)]. (6-2)
o
In this case, the correlation was even less (R = 0.07).
The Kendall and Spearman rank correlations for the paired A and B values
were also computed. The Kendall rank correlation (t.) is 0.2431; the Spearman
rank correlation (r ) is 0.3477. Both of the statistics suggest that the
ranks of the paired A and B values are not highly correlated.
The analyses discussed above did not yield an explanation for the finding
that B values tend to be slightly larger (~ 0.5 ppm) than A values. The
analyses discussed below were subsequently performed to determine if the
occurrence of unequal sample sizes with respect to day of the week could
account for the observed difference between A and B values.
6.4 DISTRIBUTIONS OF A AND B VALUES BY DAY OF WEEK
The data base under investigation contains 670 valid daily maximum 8-hour
exposures (i.e., 335 A values and 335 B values). Each exposure is associated
with a sample period which started around 7 p.m. one night and ended around
7 p.m. the next night. In the discussion which follows, each sampling period
is referred to in terms of the day the period ends. For example, a sampling
period which starts Friday night and ends Saturday night is referred to as a
Saturday sampling period. This labeling method was selected because the
maximum 8-hour exposure during a sampling period usually occurs during the
latter half of the sampling period. Using this labeling method, the daily
maximum 8-hour exposure values are distributed as follows:
142
-------
Monday
Tuesday
Wednesday
Thursday
Friday
Saturday
Sunday
If the values were evenly distributed among the days of the week, each day
would have 95.7 values. The extremes are Monday (19% low) and Wednesday (10%
high). If the values were proportionally distributed between weekdays and
weekend days, the weekdays would have 478.6 values and the weekend days would
have 191.4 values. The actual breakdown is 478 values for weekdays and 192
values for weekend days.
Table 6-5 lists summary statistics for the maximum 8-hour exposure values
of the week. Monday and Friday vie for the high exposure day. Monday has
the largest median exposure (4.6 ppm), the largest 75th percentile exposure
(7.9 ppm), and the largest 99th percentile exposure (44.0 ppm). Friday has
the largest mean exposure (6.4 ppm), the largest 90th percentile exposure
(18.1 ppm), and the largest 98th percentile exposure (35.8 ppm). In a
similar manner, Tuesday and Sunday vie for low exposure day.
The weighted average of the weekday means is 5.1 ppm; the weighted
average of the weekend means is 4.6 ppm. The weighted average of the weekday
medians is 4.1 ppm; the weighted average of the weekend medians is 3.4 ppm.
These results suggest that weekdays have higher daily maximum 8-hour exposure
values than weekend days (~ 0.6 ppm higher).
6.5 DISTRIBUTION OF DIFFERENCES BETWEEN A AND B VALUES BY DAY OF THE WEEK
For each pair of A and B values, the differences C = B - A and
D = ln(.B) - ln(.A) were determined. Tables 6-6 and 6-7 list summary statistics
for the C and D values according to the day of the week of the A value.
PEI tested the null hypothesis that the median C or D value for each day
is not greater than zero, i.e.,
H : median < 0.
o —
143
-------
TABLE 6-5. SUMMARY STATISTICS FOR DAILY MAXIMUM 8-HOUR CARBON MONOXIDE EXPOSURES
Statistic
Number of cases
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Std. skewness
Std. kurtosisc
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
98th percentile, ppm
99th percentile, ppm
Monday
78
0.1
44.0
5.7
a
5.6
15.2
46.2
1.4
2.7
4.6
7.9
9.6
13.5
16.1
44.0
Tuesday
90
0.0
16.1
3.8
2.6
2.8
6.6
8.1
0.8
2.0
3.2
5.0
6.7
8.5
13.4
16.1
Wednesday
105
0.0
15.0
4.6
a
3.0
4.7
3.1
1.2
2.6
4.0
5.9
8.7
10.3
13.7
14.2
Thursday
104
0.1
22.1
5.0
2.3
3.7
5.8
7.2
1.0
2.3
4.3
7.2
9.7
11.2
12.0
16.9
Friday
101
0.2
38.7
6.4
a
6.9
11.9
20.0
1.4
2.7
4.5
7.5
12.3
18.1
35.8
36.8
Saturday
100
0.0
34.8
4.9
1.4
5.3
11.1
20.7
0.5
1.4
3.5
6.3
8.9
15.3
19.3
24.4
Sunday
92
0.0
25.5
4.3
a
4.3
10.8
19.3
0.7
1.6
3.2
5.2
9.4
11.1
25.0
25.5
All
670
0.0
44.0
5.1
a
4.7
36.5
95.8
1.0
2.2
3.9
6.6
9.7
12.8
17.6
25.0
Not unique.
'std. skewness = g,//6/n.
g2//24Ai.
'Std. kurtosis
-------
TABLE 6-6. SUMMARY STATISTICS FOR C DIFFERENCE VALUES0
Statistic
Number of differences
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation,, ppm
Std. skewnessc
Std. kurtosis
5th percent! le, ppm
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
Percentage exceeding zero
Day of A value
Monday
36
5.4
-41.4
-1.5
b
7.5
-10.0
23.4
-5.7
-5.6
-3.3
0.1
1.5
3.7
5.4
52.8
Tuesday
53
-5.2
6.0
1.4
1.7
2.4
-0.7
-0.2
-2.3
-1.7
0.1
1.7
2.8
4.6
5.8
77.4
1
Wednesday
52
-7.0
11.8
1.1
b
4.0
1.4
-0.2
-4.9
-3.4
-1.5
0.6
3.0
6.1
8.8
57.7
Thursday
52
-8.1
32.1
2.3
b
5.9
8.4
16.1
-2.6
-2.0
-0.3
1.2
2.7
6.4
16.6
73.1
Friday
49
-20.3
6.4
-1.0
b
5.2
-3,4
2.6
-9.6
-7.2
-4.2
0.4
3.1
4.3
5.0
53.1
Saturday
50
-23.7
16.7
-1.0
b
5.7
-2.6
6.5
-9.6
-8.0
-2.8
-0.6
2.6
4.2
4.7
48.0
Sunday
42
-7.3
10.4
1.9
2.0
3.7
0.1
0.3
-4.1
-2.1
-0.1
1.4
3.7
7.0
8.6
73.8
aC = B - A.
Not unique.
°Std. skewness =
Std. kurtosis = g2//24/n.
-------
TABLE 6-7. SUMMARY STATISTICS FOR D DIFFERENCE VALUES*
Statistic
Number of differences
Minimum, ppm
Maximum, ppm
Mean, ppm
Mode, ppm
Standard deviation, ppm
Std. skewnessc
Std. kurtosisd
5th percentile, ppm
10th percentile, ppm
25th percentile, ppm
50th percentile, ppm
75th percentile, ppm
90th percentile, ppm
95th percentile, ppm
Percentage exceeding zero
Day of A value
Monday
36
-2.8
2.9
0.0
b
1.1
0.4
0.9
-1.7
-1.0
-0.8
0.0
0.5
1.2
2.1
52.8
Tuesday
53
-2.3
4.2
0.6
b
1.0
1.9
3.3
-1.1
-0.4
0.0
0.4
1.0
1.8
2.3
77.4
•
Wednesday
52
-3.6
5.9
0.1
b
1.5
2.0
5.1
-2.4
-1.8
-0.4
0.1
0.8
1.5
2.2
57.7
Thursday
52
-1.4
2.5
0.4
1.1
0.7
1.4
1.3
-0.9
-0.4
-0.1
0.4
0.6
1.2
2.0
73.1
Friday
49
-4.2
2.3
-0.3
b
1.4
-1.7
-0.4
-2.7
-2.4
-1.2
0.2
0.8
1.2
1.6
53.1
Saturday
50
-5.1
5.0
-0.1
b
1.6
0.1
2.6
-2.8
-1.9
-0.9
-0.2
0.7
1.6
2.3
48.0
Sunday
42
-2.4
5.3
0.5
0.4
1.1
4.0
8.1
-0.6
-0.4
0.0
0.4
0.9
1.5
2.2
73.8
aD = ln(B) -
b
Not unique.
cStd. skewness
d
g,//67n.
Std. kurtosis = g2//247n.
-------
The test statistic is
T = (2N - n)//n~ (6-3)
where N is the number of positive values and n is the total number of values.
For n > 25, T follows the unit normal distribution. Table 6-8 lists the
results of this test. The results apply to both the C and D values. Since
Tuesday, Thursday, and Sunday have p values less than 0.05, one can conclude
that the median C and D values for these days are significantly greater than
zero and thus that B values tend to be larger than A values. The medians for
the other days are not significantly larger than zero.
As Table 6-6 indicates, the median C values for Tuesday, Thursday, and
Sunday are 1.7 ppm, 1.2 ppm, and 1.4 ppm, respectively. Since one would
expect weekday exposures to be somewhat uniform, it is surprising that for 50
percent of the subjects Wednesday exposures exceed Tuesday exposures by at
least 1.7 ppm and Friday exposures exceed Thursday exposures by at least 1.2
ppm. The Sunday result is not particularly surprising, however, as one would
expect the Sunday exposure of a typical subject to be less than his or her
Monday exposure.
One possible explanation for the unexpected results for Tuesday and
Thursday is that the ambient CO concentrations during the Denver CO study may
have been higher on Wednesdays and Fridays than on other days. An investiga-
tion of this hypothesis had not been carried out at the time of this report.
6.6 REFERENCE
1. Pollard, J. H. A Handbook of Numerical and Statistical Techniques.
Cambridge University Press, London. 1977.
147
-------
TABLE 6-8. RESULTS OF NONPARAMETRIC TEST OF NULL HYPOTHESIS THAT MEDIAN C OR D
VALUE IS NOT GREATER THAN ZERO
Statistic
n
N
T
P
Day of A value
Monday
36
19
0.33
0.37
Tuesday
53
41
3.98
<0.01
Wednesday
52
30
1.11
0.13
Thursday
52
38
3.33
<0.01
Friday
49
26
0.43
0.33
Saturday
50
24
-0.28
0.61
Sunday
42
31
3.09
<0.01
00
-------
SECTION 7
TIME SPENT IN SELECTED MICROENVIRONMENTS
One of the principal reasons for collecting activity diary data during
the Denver study was to provide a means for relating exposure to a subject's
microenvironment, i.e., the subject's immediate physical surroundings. As
illustrated in Section 3.6 of Reference 1, the codes assigned to activity
diary entries can be combined in a variety of ways to designate microenviron-
ments of interest. In the initial analyses, four-digit microenvironment codes
were created by combining the two-digit B (location,) code with the two-digit
D3 (transit mode) code. Examples of microenvironment codes occurring in the
SAMPLE-DATA file include 0193 (in transit - motorcycle), 52bb (indoors -
public garage), and 09bb (uncertain).
Table IV in Appendix A lists the weighted mean (in minutes) of the
occupancy periods for each microenvironment except indoors - residence (Code
02bb). An occupancy period begins when a subject enters a" new microenviron-
ment and ends when the subject leaves the microenvironment. Mean occupancy
periods for the indoor residential microenvironment could not be determined
accurately from activity diary data because subjects were usually occupying
residences before the first diary entry and after the last diary entry.
Mean occupancy periods range from 431.9 minutes (indoors - manufacturing
facility) to 7.4 minutes (outdoors - residential garage or carport). Mean
occupancy periods for in-transit microenvironments associated with motor
vehicles and high CO levels are 30.8 minutes for trucks, 28.0 minutes for
buses, 25.9 minutes for cars, and 23.0 minutes for motorcycles. The value
for indoors - public garage (29.4 minutes) is higher than expected and may be
the result of errors in recording activity diary information.
Statistics have also been compiled on the total time spent per day in
selected microenvironments. Table 7-1 lists 10 microenvironments of
particular interest to EMSL and percentiles (weighted) for the total time
spent per day in each (here labeled total person-day exposure duration). Also
149
-------
TABLE 7-1. WEIGHTED SUMMARY STATISTICS FOR CO EXPOSURES BY MICROENVIRONMENT
CONSIDERING ONLY PERSON-DAYS WITH NONZERO EXPOSURE DURATIONS*
Microenvironment
Indoors - parking garage
In transit - car
In transit *• other
Outdoors - near roadway
In transit - walking
Indoors - restaurant
Indoors - office
Indoors - store/mall
Indoors - residence
Indoors - total
Number of
person-days
with nonzero
durations
31
643
107
188
171
205
283
243
776
776
CO exposure, ppm
Mean
18.8
8.0
7.9
3.8
4.2
4.2
3.0
3.0
1.7
2.1
Std. dev.
27.6
8.1
6.3
4.9
5.9
4.1
3.4
3.5
2.7
2.5
Std. error
4.96
0.32
0.61
0.36
0.45
0.29
0.20
0.22
0.10
0.09
Percentiles for
total person-day exposure
duration, minutes
10%
3
30
19
5
6
22
70
11
668
1162
25%
6
48
44
8
11
34
229
21
820
1263
50%
14
71
66
33
28
58
478
50
975
1343
75%
60
114
125
77
52
85
541
88
1225
1381
90%
120
166
196
102
124
232
628
170
1359
1408
Standard error statistics are approximations,
-------
listed are the mean CO exposures reported for each microenvironment. Note
that person-days with zero time spent in a particular microenvironment were
not considered in calculating these statistics.
The Strategies and Air Standards Division of EPA has developed the NAAQS
Exposure Model (NEM) as a means of estimating human exposure to criteria pollu-
tants such as CO and ozone. Six microenvironments have been defined for NEM
analyses of CO exposure. Table 7-2 shows how the microenvironments defined
for the Denver study can be aggregated into the six NEM microenvironments.
The E2 diary entry (gas^tove) was used to determine whether or not a gas stove
was in operation.
Omitting all data flagged as invalid, PEI calculated various unweighted
summary statistics on time spent per day in each microenvironment (Table 7-3).
Median times are 76 minutes for motor vehicles, 0 minutes for indoors-residence
(gas stove on), 980 minutes for indoors-residence (no gas stove or gas stove
off), 207 minutes for indoors-other locations, 0 minutes for outdoors-near
road, and 0 minutes for outdoors-other locations. Statistics on time spent
in activities for which the microenvironment was not recorded are provided in
the column labeled "uncategorized time." Statistics on the duration of each
sampling period (nominal duration = 24 hours or 1440 minutes) are provided
in the column labeled "all microenvironments."
Table 7-4 is similar to Table 7-3 except that person-days with zero time
spent in a microenvironment are not considered in calculating the summary
statistics for that microenvironment. As expected, median times are higher:
83 minutes for motor vehicles, 80 minutes for indoors-residence (gas stove
on), 985 minutes for indoors-residence (no gas stove or gas stove off), 299
minutes for indoors-other locations, 35 minutes for outdoors-near road, and
26 minutes for outdoors-other locations.
Although the study covered a period of cold weather (November-February),
it is nevertheless surprising that the Denver subjects spent so little
time outdoors. Subjects were instructed to record all activities expected
to last 5 minutes or more. Yet 59.0 percent of the person-days contained
no entries categorized as outdoors-near road, and 85.5 percent of the
subject-days contained no entries categorized as outdoors-other.
Another somewhat surprising result is the large quantity of time spent
in motor vehicles. More than 90 percent of the person-days contained entries
151
-------
TABLE 7-2. AGGREGATION OF MICROENVIRONMENTS DEFINED FOR DENVER CARBON MONOXIDE
STUDY INTO MICROENVIRONMENTS DEFINED FOR NEM ANALYSES OF CARBON
MONOXIDE EXPOSURE
NEM microenvironment
Denver CO study mlcroenvironment
Code
Description
Motor vehicle
Indoors-residence
(gas stove on)
Indoors-residence
(other)
Indoors-other locations
Outdoors-near road
0102
0103
0104
0193
5202
7202
7302
7303
02bb!
51bba
02bbJ
51bbc
03bb
0301
04bb
05bb
52bb
5201
53bb
54bb
55bb
56bb
57bb
58bb
59bb
60bb
61bb
62bb
0101
0192
07bb
0701
72bb
7201
7202
73bb
7301
In transit - car
In transit - bus
In transit - truck
In transit - motorcycle
Indoors - public garage (in car)
Outdoors - public garage (in car)
Outdoors - parking lot (in car)
Outdoors - parking lot (in bus)
Indoors - residence
Indoors - residential garage
Indoors - residence
Indoors - residential garage
Indoors - office
Indoors - office
Indoors - store
Indoors - restaurant
Indoors - public garage
Indoors - public garage
Indoors - manufacturing facility
Indoors - service station or motor
vehicle repair facility
Indoors - other repair shop
Indoors - auditorium, sports arena,
concert hall, etc.
Indoors - church
Indoors - shopping mall
Indoors - health care facility
Indoors - school
Indoors - other public building
Indoors - other location
In transit - walking
In transit - bicycle
Outdoors - within 10 yards of road
Outdoors - within 10 yards of road
Outdoors - public garage
Outdoors - public garage
Outdoors - parking lot
Outdoors - parking lot
(continued)
152
-------
TABLE 7-2 (continued)
NEM microenvironment
Outdoors-other locations
Denver CO study microenvironment
Code
71bb
74bb
76bb
77bb
78bb
79bb
80bb
Description
Outdoors - residential garage or
carport
Outdoors - service station or motor
vehicle repair service
Outdoors - residential grounds
Outdoors - school grounds
Outdoors - sports arena, amphi-
theater, etc.
Outdoors - park or golf course
Outdoors - other location
'E2 = 01.
'Blank.
:E2 = 02.
153
-------
TABLE 7-3. UNWEIGHTED SUMMARY STATISTICS FOR TIME SPENT PER DAY BY DENVER
SUBJECTS IN MICROENVIRONMENTS USED IN NEM ANALYSES
Statistic
Cases
Minimum, minutes
Maximum, minutes
Mean, minutes
Mode, minutes
Standard deviation,
minutes
5th percentile,
minutes
10th percentile,
minutes
25th percentile,
minutes
50th percentile,
minutes
75th percentile,
minutes
90th percentile,
minutes
95th percentile,
minutes
Percentage of
zero values
Motor
vehicle
851
0.0
1313.0
99.7
0.0
105.1
0.0
8.0
41.0
76.0
125.0
198.0
282.0
9.5
Indoors-
residence
(gas on )
851
0.0
1373.0
26.4
0.0
118.4
0.0
0.0
0.0
0.0
0.0
49.0
144.0
85.3
Indoors-
residepce
(other0)
851
0.0
1594.0
1004.2
0.0
273.1
636.0
717.0
805.0
980.0
1231.0
1360.0
1412.0
0.7
Indoors-
other
locations
851
0.0
1329.0
259.7
0.0
222.7
0.0
0.0
49.0
207.0
475.0
557.0
608.0
17.3
Outdoors-
near
road
851
0.0
778.0
25.0
0.0
61.4
0.0
0.0
0.0
0.0
25.0
71.0
134.0
59.0
Outdoors-
other
locations
851
0.0
972.0
10.2
0.0
58.7
0.0
0.0
0.0
0.0
0.0
12.0
41.0
85.5
Uncategor-
ized time
851
0.0
445.0
17.8
0.0
50.8
0.0
0.0
0.0
0.0
2.0
54.0
126.0
73.8
All micro-
environments
851
1134.0
1671.0
1443.0
1442.0
44.7
1375.0
1407.0
1430.0
1442.0
1454.0
1481.0
1515.0
0.0
Gas stove on.
No gas stove or gas stove off.
'Includes uncharacterlzed time.
-------
TABLE 7-4. UNWEIGHTED SUMMARY STATISTICS FOR TIME SPENT PER DAY BY DENVER SUBJECTS IN
MICROENVIRONMENTS USED IN NEM ANALYSES. STATISTICS FOR EACH MICROENVIRONMENT
OMIT PERSON-DAYS WITH ZERO TIME SPENT IN THE MICROENVIRONMENT
Statistic
Cases
Minimum, minutes
Maximum, minutes
Mean, minutes
Mode, minutes
Standard deviation,
minutes
5th percentile,
minutes
10th percentile,
minutes
25th percentile,
minutes
50th percentile,
minutes
75th percentile,
minutes
90th percentile,
minutes
95th percentile,
minutes
Motor
vehicle
770
2.0
1313.0
110.2
d
105.1
20.0
29.0
51.0
83.0
136.0
204.0
299.0
Indoors-
residence
(gas ona)
125
1.0
1373.0
179.9
45.0
261.1
6.0
10.0
44.0
80.0
185.0
527.0
819.0
Indoors-
residence
(otherb)
845
3.0
1594.0
1011.3
d
260.6
655.0
728.0
810.0
985.0
1231.0
1360.0
1412.0
Indoors-
other
locations
704
2.0
1329.0
313.9
85.0
207.2
29.0
54.0
117.0
299.0
504.0
572.0
617.0
Outdoors-
near
road
349
1.0
778.0
61.1
d
83.7
4.0
6.0
13.0
35.0
70.0
142.0
214.0
Outdoors-
other
locations
123
1.0
972.0
70.8
2.0
140.5
2.0
4.0
9.0
26.0
62.0
-161.0
307.0
Uncategor-
ized time
222
1.0
445.0
68.2
d
80.3
2.0
4.0
12.0
37.0
102.0
187.0
245.0
All micro-
environments
851
1134.0
1671.0
1443.0
1442.0
44.7
1375.0
1407.0
1430.0
1442.0
1454.0
1481.0
1515.0
en
en
Gas stove on.
'Includes uncharacterized time.
No gas stove or gas stove off.
Not unique.
-------
categorized as "1n motor vehicle." The median time spent in motor vehicles
per day by persons using motor vehicles is 76 minutes (Table 7-4). Ten percent
of those using motor vehicles spent 204 minutes (3.4 hours) or more in motor
vehicles.
156
-------
SECTION 8
FACTORS ASSOCIATED WITH INSTRUMENT FAILURE
The personal exposure monitors (PEM's) used in the Denver study received
zero-span checks before and after each sampling period and were>frequently
checked for loose connections, clogged pumps, and other conditions likely
to cause failure under use. Despite these precautions, a small percentage
of the units did fail. PEI was directed by EMSL to determine if some of
these failures were related to temperature or other meteorological
variables. This section contains a summary of PEI's analyses in this area.
8.1 PERSONAL EXPOSURE MONITOR
The PEM is a modified General Electric (GE) Carbon Monoxide Detector,
Model 15EC53003, mated with a modified Magus DL-1 Data Logger and mounted in
a compact, tamperproof casing. The PEM records the time and a CO concentra-
tion value every time the "activity button" on the top of the instrument is
pushed and every hour on the hour. In both cases, the CO value is the absolute
value of the integrated average CO concentration since the last recorded value.
The PEM is capable of operating continuously for 24 hours and logging up to
113 data points.
Figure 8-1 shows the PEM controls, liquid crystal display (LCD), and
access points with cover in place. The activity button is located under a
flexible black cover on the left side of the instrument. The on/off push-
button switch controls power from the battery pack to the pump. Power is
constantly applied to the CO sensor cell to maintain its stability regardless
of the position of the on/off switch. The on/off switch is located under a
hard plastic cap and is not accessible to study participants.
An ambient air sample is drawn into the detector through a potassium
permanganate filter by an integral sample air pump. The sample is discharged
to the CO sensor cell, the CO is oxidized, and an electrical signal proportional
157
-------
iNtrr
PUMVMETt*
(SCALB 1:2 AWMOXIMATILY)
Figure 8-1. Personal exposure monitor.
158
-------
to the CO level is produced. The air sample is then exhausted through the
flow indicator inside the case. The filter provides selectivity of the
detector to CO by removing most interfering gases. Hydrogen, ethylene, and
acetylene are potential interferents that may not be completely removed by
the filter.
Under normal operating conditions, the LCD displays the time. When the
activity button is pressed, the LCD momentarily displays the word GULP to
indicate that the instrument has received the signal to log data.
Procedures for calibrating and programming the PEM are discussed in Sec-
4.5 of the report by Johnson. The general
ment are listed in Table 3-1 of the same report.
tion 4.5 of the report by Johnson. The general specifications of the instru-
8.2 DATA BASE
Two sets of PEM's were used during the course of the Denver CO Study.
The primary set contained 26 PEM's that were used throughout the Denver
study. A supplementary set of six PEM's were put into operation on
February 9, 1983, and were used during the remainder of the study. These
six PEM's had been used previously by RTI as-part of the parallel study of
CO exposure in Washington, D.C., which ended February 7, t983. Consequently,
each of the 32 PEM's used in the Denver study experienced several months of
constant use.
A log was maintained on each PEM, which listed all instances during the
Denver study in which the PEM failed a zero-span check, all other instrument
failures that occurred (whether or not data were affected), and any other
situations in which nonroutine servicing was performed. Table N-l in the
report by Johnson lists all malfunctions recorded in these logs. PEI
identified a subset of these malfunctions which might be temperature-related.
Table 8-1 lists these malfunctions and assigns each an individual failure
mode code (1, 2 7) and group failure mode (A, B, or C). Group A con-
tains failure modes related to zero-span problems. Group B contains failure
modes related to problems with the MAGUS unit. In each case the MAGUS unit
switched from the normal data-recording mode (LOG) into an alternative mode
(ALOG, SCLR, DONE, or unspecified). A third group, Group C, was formed by
combining Group B and Failure Mode 7 ("lock-up"). When a PEM locked-up,
159
-------
TABLE 8-1. PEM FAILURE MODES ANALYZED BY LOGISTIC REGRESSION MODEL
Group3
A
B
-
Failure mode
1
2
3
4
5
6
7
Failure description
PEM failed zero-span check
PEM exhibited excessive zero and/or span
MAGUS switched to unspecified mode
MAGUS switched to ALOG mode
MAGUS switched to SCLR mode
MAGUS switched to DONE mode
No change in time displayed and no data
("lock up")
drift
stored
Group C combines Failure Modes 3 through 7.
the liquid crystal display (LCD) display and MAGUS data subsystem stopped
working, the LCD time did not advance, and the MAGUS data subsystem would
not log CO data.
A computer file was constructed listing 1) the number of PEM's which mal
functioned during each 24-hour sampling period by individual failure mode,
2) the maximum temperature (TMAX) of the day on which the sampling period
ended, 3) the minimum temperature (TMIN), and 4) the mean temperature (TAVG).
This computer file comprised the entire data base used in the analysis
discussed below.
8.3 MODEL USED FOR STATISTICAL ANALYSIS
The proportion (PROPN) of PEM's associated with a given temperature and
a given individual failure mode (or failure mode group) was assumed to follow
the logistic regression model (LRM)
(8-1)
where E is expectation; a is a .constant; g., i = 1,2,3, are regression coeffic-
ients; Gdj) = TMAX; G(T2) = TMIN; and G(T3) = TAVG. Regression coefficients
160
-------
were estimated by the method of maximum likelihood, as implemented in program
PLR of the BMDP statistical software package. Models with more than one
explanatory variable (EV); e.g.,
2
a + Z B. * 6(1,),
1=1 1 1
etc.
were found to provide no improvement in fit over models with a single
explanatory variable and are not discussed here.
8.4 TEST STATISTICS
Two basic test statistics were provided in the output of each run of the
PLR program. They were 1) the approximate F statistic, and 2) the improve-
ment chi-square statistic.
8.4.1 Approximate F Statistic
In linear logistic regression, the expected value of the logit of the
observed proportions is a linear function of one (or more) explanatory
variable(s). Consider a design with one EV. For each discrete value of
the EV, the corresponding observed proportion is an unbiased estimate of a
Bernouilli parameter 9 which is the probability of failure in a single event
given the value of the EV; that is,
E(X) = n6 (8-3)
VAR(X) = ne(l-e) (8-4)
where X = number of failures and n = number of events (or trials). In the
case that 9 is known for each discrete value of the EV, the weights for
weighted linear regression are given by the expression
weight = /n9(l-9) , (8-5)
161
-------
and the response variables are given by the expression
response variable = logfi . (8-6)
Because the value of 6 is unknown, the weighted linear regression can be done
only approximately. In fact, one uses in place of 6, its estimate x/n. Con-
sequently, an "approximate" F statistic is computed in weighted linear
regression to test the significance of the addition, or removal, of regression
terms.
8.4.2 Improvement Chi -Square Statistic
The extremum of the logarithm of the likelihood function is" determined
with, and without^a designated regression term in the model. Twice the
natural logarithm of the ratio of the two extrema is distributed, asymptotically,
as a chi-square random variable with one degree of freedom. The results of
applying this test to the data are similar to those obtained with the
approximate F test discussed above.
8.5 DISCUSSION OF RESULTS
Results for individual failure modes are displayed in Table 8-2. For
Failure Modes 1, 2, and 7 the regression coefficients are significant (i.e.,
p < 0.05), and the LRM is judged to provide a good fit. For Failure Mode 1,
each of the temperature measures can be used to "explain" the data. For
Failure Mode 2, the variable TMIN is an appropriate EV. For Failure Mode 7,
both TMAX and TAVG are qualified EV's. With respect to Failure Modes 3 through
6, no statistical significance was found; consequently, no temperature
dependence is inferred.
The regression coefficients have a negative sign for Failure Mode 1 and
positive signs for Failure Modes 2 and 7. With increasing temperature, the
proportion of failures decreases for Failure Mode 1, increases for Failure
Mode 2, and increases for Failure Mode 7.
The analysis was repeated using grouped failure modes. As discussed
above, Group A contains Failure Modes 1 and 2, which involve zero/span
problems. Group B contains Failure Modes 3 through 6, which involve problems
with the MAGUS unit.
162
-------
TABLE 8-2. RESULTS OF FITTING LOGISTIC REGRESSION MODEL TO FAILURE DATA
FMa
1
2
3
4 .
5
6
7
Ae
Bf
C9
FCb
31
2
2
25
6
4
22
33
37
59
Temperature-related variable
TMAX
0
-0.076
d
d
d
d
d
0.044
-0.065
d
0.023
c
F
X2
F
X2
F
X2
F
2
V
F
2
V
F
X2
F
X2
F
X2
F
2
r
F
2
r
Value
15.20
15.79
d
d
d
d
d
d
d
d
d
d
4.31
4.28
11.97
12.34
d
d
3.06
3.05
P
0.000
0.000
d
d
d
d
d
d
d
d
d
d
0.038
0.039
0.001
0.000
d
d
0.081
0.081
TMIN
0
-0.098
0.381
d
d
. d
d
d
-0.068
d
d
c
F
X2
F
X2
F
X2
F
2
r
F
2
r
F
X2
F
X2
F
X2
F
2
r-
F
2
X^
Value
7.41
7.77
6.75
6.78
d
d
d
d
d
d
d
d
d
d
3.95
4.08
d
d
d
d
P
0.007
0.005
0.010
0.009
d
d
d
d
d
d
d
d
d
d
0.047
0.043
d
d
d
d
TAVG
3
-0.117
d
d
d
d
d
0.060
.
-0.093
d
0.036
c
F
X2
F
X2
F'
X2'
F
2
Y
F
2
r
F
X2
F
X2
F
X2
F
2
F
2
X^
Value
15.02
16.29
d
d
d
d
d
d
d
d
d
d
4.06
3.95
10.76
11.46
d
d
3.62
3.56
P
0.001
0.000
d
d
d
d
d
d
d
d
d
d
0.044
0.047
0.001
0.001
d
d
0.057
0.059
FM = failure mode.
FC = failure count.
Statistic (.F = approximate F, X = improvement chi-square).
Not significant at p = 0.10 level.
elncludes Failure Modes 1 and 2.
Includes Failure Modes 3 through 6.
Includes Failure Modes 3 through 7.
163
-------
Group C consists of Failure Modes 3 through 7; in other words Groups B and
Failure Mode 7 have been combined into one group.
The results for Group A are only slightly different than the results
reported above for Failure Mode 1. Group B showed no temperature dependence
based on 37 failure events. Statistical significance for Group C was not
quite reached at a critical level of 0.05 (p = 0.0574 for TAVG) based on a
total of 59 failure events. Grouping did not affect the sign of the
regression coefficients.
8.6 CONCLUSIONS
The results of applying the logistic regression model to the data base
described above suggest that PEM's used in the Denver study were more likely
to experience zero-span problems on cold days and were more likely to
experience lock-up on warm days. Problems with the MAGUS unit other than
lock-up do not appear to be associated with temperature.
8.7 REFERENCE
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection" Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983.
164
-------
SECTION 9
SUMMARY OF RESULTS
This section summarizes the major results of the statistical analyses
described in Sections 2 through 8. Results of earlier analyses are
summarized in Appendix A.
9.1 FIXED-SITE MONITORS
Ten of the 15 maximum values reported by Denver fixed-site monitors
occurred during either the morning or the evening high traffic period
(8:00, 17:00, or 18:00). Four of the 15 maximum values occurred on
January 27, 1983. The maximum value in the composite data set was 15.8 ppm
and occurred at 8:00 on December 17, 1982.
Two of the permanent Denver monitors (Map Codes A and B) and three of
the temporary monitors (Map Codes D, F, and J) reported daily maximum 8-hour
values exceeding 15 ppm. These five monitors were all located in the
central business district of Denver, an area of high traffic density.
Stepwise linear regression results suggest that hourly average CO
concentration at the Denver composite site increases with maximum daily
temperature, decreases with minimum daily temperature, and decreases with
2
windspeed. The modeled relationship is weak, however (R = 0.12).
Denver experienced much higher ambient CO levels during the study
period than did Washington. With respect to the composite daily maximum
1-hour CO concentrations, Denver had a mean of 6.6 ppm--more than twice
Washington's mean of 3.2 ppm.
9.2 RELATIONSHIPS BETWEEN EXPOSURES AND FIXED-SITE READINGS
Stepwise linear regression analyses of Denver data suggest that 1-hour
values reported by a particular fixed-site monitor or "optimized" group
of monitors do not provide a good means of predicting simultaneous PEM
165
-------
2
values. The site at 2105 Broadway produced the largest R value (0.049)
when sites were considered individually.
Linear regression analyses relating Washington PEM values grouped by
microenvironments to 1-hour readings reported by the nearest fixed-site or the
2
composite site yielded R values from 0.00 to 0.66 for nontransit micro-
-*
environments and from 0.01 to 0.61 for in-transit microenvironments. Micro-
2
environments with relatively large R values include indoors-hospital (0.66),
indoors-church (0.60), indoors-garage (0.19), outdoors-park (0.15), train/
subway (0.61), jogging (0.30), truck (0.17), and bicycle (0.16). Results
•
suggest that in-transit PEM values are better paired to fixed-site values
reported by the composite site or site nearest the end address than to those
reported by the site nearest the start address. A similar analysis of Denver
data is summarized in Appendix A.
Based on linear regression analysis, the adjustment of Denver PEM values
by subtracting the simultaneously recorded fixed-site value does not appear
to be a promising approach for characterizing indoor sources of CO.
Linear regression analyses suggest that composite fixed-site daily
maximum values are poor predictors of daily maximum exposures (1-hour
and 8-hour) in Denver.
The results of t tests and various nonparametric tests suggest that
daily maximum 8-hour exposures in Denver are higher on days when fixed-site
daily maximum 8-hour values exceed 9 ppm.
9.3 RELATIONSHIP BETWEEN EXPOSURES AND SELECTED EXPLORATORY VARIABLES
A series of exploratory analyses were conducted to identify factors
associated with "Group H" person-days, that is, person-days for which the
daily maximum 8-hour exposure exceeds 9 ppm. Group L contains the remaining
person-days. The following results apply to the Denver study only.
1. Person-days in Group H exhibited higher CO levels in most
microenvironments.
2. The microenvironments which were visited more often during
Group H person-days than would be expected are all associated with
either outdoor locations and/or motor vehicles. Indoors-service
station and outdoors-public garage have particular large OER values.
166
-------
3. The average durations of visits to the outdoors-service station
and indoor-restaurant microenvironments are larger for Group H
than for Group L person-days.
4. The microenvironment "outdoors - within 10 years of road" is
associated with Group H person-days when it is located in an area
with high ambient CO levels.
5. The microenvironments "indoor - public garage" and "indoor - service
station" are associated with Group H person-days in areas with
relatively low ambient CO levels.
6. Of the in-transit microenvironments, only "truck" is associated
with Group H person-days when the start or end location of the trip
is in a high ambient CO location. "Motorcycle" is associated with
Group H for low ambient CO locations.
7. Group H person-days are strongly associated with periods of high
ambient CO concentration, particularly November 5 and December 9,
1982.
8. The aggregate occupation category "work which may involve
proximity to running motor vehicles or internal combustion engines
in enclosed space" is strongly associated with Group H person-days.
Results of analyses of variance and covariance performed on Denver
in-transit exposure data support the following conclusions:
1. The two motor vehicle categories (car and other) are associated
with higher exposures than walking. Exposures In these two
categories are particularly high during the time periods 6-9
and 16-19. These two periods bracket the morning and afternoon
rush hours.
2. The presence of smokers does not increase exposure.
3. Exposure decreases as duration increases.
Similar analyses performed on the Washington in-transit exposure data
support the following conclusions:
1. Exposure varies with mode-of-travel, time-of-day, arid presence of
smokers.
2. No associations exists between exposure and duration.
Analyses of variance performed on indoor exposure data from the Denver
study support the following conclusions:
2
1. CO exposures are higher in homes having living areas of 1000 ft
or less.
167
-------
2. CO exposures are higher in homes where gas cooking stoves or gas
clothes dryers are used (vented or not vented).
3. CO exposures are higher in homes where unvented gas furnaces or
space heaters are used.
4. Venting of gas furnaces and space heaters decreases CO exposure
in the home.
5. CO exposures are higher in homes which have storm windows, storm
doors, or special dampers.
6. CO exposures are higher in homes where the main heating source is
either a portable room heater or gravity gas system.
7. CO exposures are lower in homes where the main heating system
consists of built-in electric units.
8. CO exposures are higher in work places where the main heating
system consists of nonportable heaters burning gas, oil, or
kerosene.
9. In homes without a gas cooking stove, the presence of a space
heater significantly increases exposure.
10. In homes with gas cooking stoves, the presence of a space heater
does not significantly increase exposure.
9.4 MODELS FOR PREDICTING EXPOSURE IN DENVER
A series of 14 general models for predicting exposure in Denver were
proposed and evaluated in a sequential manner such that the results of each
evaluation were considered in constructing the next general model. The
parameters considered in the general models included data obtained from the
activity diaries, the background questionnaire completed by each participant,
the fixed-site monitors, and a meteorological file containing data on
temperature and daily average wind speed. Model evaluation was accomplished
by performing step-wise linear regression on each general model and noting
2
1) which terms were retained in the "best-fit" model and 2) the R value
associated with the best-fit model.
2
The best-fit model yielding the largest R values (0.34) is
168
-------
(CpEM°'4° - D/0.40 = -0.068 + (0.11)(C) + (0.68)(6AS) -
(0.32)(P1)(M4) - (0.65)(P2J(M4) + (1.56)(P2)(M5) -
(1.72)(P3)(M5) + (1.43)(P4)(M5) + (0.19)(C)(M5) +
(0.27)(C)(M7) + (1.05)(SMOKE)(M5) +
(1.15)(IH)(6CSJ(P4) - (0.51)[ln(WIND)](M4) +
(15.2)(WIND)~1(M1) + (0.23)(C1)(M2) +
(0.22)(C1)(M4) + (0.13)(C)(M3),
where Cpr» is the estimated PEM value. The other terms are defined in
Table 5-6. The three most important terms in the model with respect to
2
increasing the R value are related to wind speed given a low exposure
indoor microenvironment [In(WIND)] (M4), to simultaneous fixed-site
readings (C), and to high-exposure in-transit microenvironments (M5).
Four terms (Ml, M2, M3, and M4) which appear in the general models
relate to aggregate indoor microenvironments. These were defined through
the use of a pairwise comparison procedure which aggregated similar
microenvironments into groups which differ significantly one from another
with respect to CO exposure. Table 5-22 lists the four aggregate indoor
microenvironments.
9.5 COMPARISON OF CONSECUTIVE DAILY MAXIMUM EXPOSURES
The subjects in the Denver study were requested to participate for two
consecutive 24-hour sampling periods. An analysis of the resulting PEM
data revealed that a pair of valid daily maximum 8-hour exposure values
(i.e., one for each sampling period) could be calculated from the data
obtained from each of 335 subjects. The first of the two values in each
pair is referred to as the A value; the second of the two values is the B
value. The results of a series of statistical analyses support the
following conclusions:
1. The distribution of A values differs significantly from the
distribution of B values (p < 0.05).
2. The mean difference between paired A and B values differs
significantly from zero (p < 0.05).
o
3. The A and B values are not highly correlated (R = 0.16).
4. Weekdays have higher daily maximum 8-hour exposures than
weekend days (~ 0.6 ppm higher).
169
-------
9.6 TIME SPENT IN SELECTED MICROENVIRONMENTS
The Strategies and Air Standards Division of EPA has developed the NAAQS
Exposure model (NEM) as a means of estimating human exposure to criteria
pollutants such as CO and ozone. Six microenvironments have been defined
for NEM analyses of CO exposure. Table 7-2 shows how the microenvironments
defined for the Denver study can be aggregated into the six NEM micro-
environments. The E2 diary entry (gas stove) was used to determine whether
or not a gas stove was in operation.
Omitting all data flagged as invalid, PEI calculated various unweighted
summary statistics on time spent per day in each microenvironment. Median
times are 76 minutes for motor vehicles, 0 minutes for indoors-residence
(gas stove on), 980 minutes for indoors-residence (no gas stove or gas
stove off), 207 minutes for indoors-other locations. Statistics on time
spent in activities for which the microenvironment was not recorded are
provided in the column labeled "uncategorized time." Statistics on the
duration of each sampling period (nominal duration = 24 hours or 1440 minutes)
are provided in the column labeled "all microenvironments."
Different results occur when person-days with zero time spent in a
microenvironment are excluded when calculating the summary statistics for
that microenvironment. As expected, median times are higher: 83 minutes
for motor vehicles, 80 minutes for indoors-residence (gas stove on),
985 minutes for indoors-residence (no gas stove or gas stove off), 299
minutes for indoors-other locations, 35 minutes for outdoors-near road,
and 26 minutes for outdoors-other locations.
More than 90 percent of the person-days contained entries categorized
as "in motor vehicle." The median time spent in motor vehicles per day
by persons using motor vehicles is 76 minutes. Ten percent of those using
motor vehicles spent 204 minutes (3.4 hours) or more in motor vehicles.
9.7 FACTORS ASSOCIATED WITH INSTRUMENT FAILURE
A logistic regression model was used to evaluate the relationship
between the fraction of PEM's failing each day and three indicators of
ambient temperature: daily maximum temperature (TMAX), daily minimum
170
-------
temperature (JMIN), and daily mean temperature (TAVE). PEM failures
were grouped into seven failure modes. The results of the analysis suggest
that PEM's used in the Denver study were more likely to experience zero-span
problems on cold days and were more likely to experience lock-up on warm
days. Problems with the MAGUS unit other than lock-up do not appear to
be associated with temperature.
171
-------
84-121.3
APPENDIX A
A Study of Personal Exposure to Carbon
Monoxide in Denver, Colorado
Ted R. Johnson
Ted R. Johnson, M.S., 1s an Environmental Engineer with
PEDCo Environmental, Inc., 505 S. Duke St., Suite 503
Durham, North Carolina 27701
For presentation at the 77th Annual Meeting of the
Air Pollution Control Association
San Francisco, California
June 24-29, 1984
172
-------
84.121.3
Introduction
The National Ambient Air Quality Standard (NAAQS) for carbon
monoxide (CO) states that 1-hour CO concentrations shall not exceed 35
ppm more than once per year and that 8-hour CO concentrations shall not
exceed 9 ppm more than once per year. Compliance with these standards
is usually determined by fixed-site monitoring data. However, fixed-
site monitoring data may not provide an accurate indication of personal
exposure within an urban population, which is a function of both geo-
graphic location (e.g., downtown versus suburbia) and immediate physical
surroundings (e.g., indoors versus outdoors). Better estimates of
personal exposure can be developed by equipping a large number of
subjects with portable monitors and activity diaries. If the subjects
are properly selected, their exposures can be extrapolated to the larger
urban population.
Such a study was conducted in Denver, Colorado, by PEDCo Environ-
mental, Inc., for the Environmental Monitoring Systems Laboratory (EMSL)
of the U.S. Environmental Protection Agency (EPA). Each of 454 subjects
was asked to carry a personal exposure monitor (PEM) and an activity
diary for two consecutive 24-hour sampling periods and to provide a
breath sample at the end of each sampling period. Each participant.also
completed a detailed background questionnaire. The questionnaire results
and approximately 900 subject-days of PEM and activity diary data col-
lected between November 1, 1982, and February 28, 1983, were analyzed to
determine if factors such as microenvironment and the presence of indoor
CO sources significantly affect personal CO.exposure. In addition, the
exposure of the entire Denver population was .extrapolated from exposures
recorded by the study participants. PEDCo also compared CO levels record-
ed by fixed-site monitors to levels recorded simultaneously by PEM's.
Sample Selection
The target population of the study included all noninstitutional-
ized, nonsmoking residents of the urbanized portion of the Denver,
Colorado, metropolitan area who were between 18 and 70 years of age at
the time of the study. Research Triangle Institute (RTI) and PEDCo
developed a two-phase scheme for sampling this population, which is
estimated to be 245,000. In the first phase, a two-stage sample of
housing units was selected. Data on the individuals residing within
these housing units were collected using a brief screening questionnaire
administered by telephone or in the field. Individuals who exhibited
rare characteristics with respect to CO exposure were identified and
oversampled in the second-phase of sample selection.
Individuals entered the sample by three paths. The majority of
study participants (402) were identified by means of a telephone screen-
ing questionnaire administered to members of housing units appearing on
a list prepared by Donnelley Marketing Information Services. The remain-
ing 52 study participants were identified by field screening of housing
units which 1) appeared on the Donnelley list but for which no telephone
number was available or 2) were identified through a special survey of
housing units which did not appear on the Donnelley list. The original
sample selection protocol was designed to yield 500 study participants.
The reduced sample size (454) resulted from a higher than expected
refusal rate and unexpected equipment problems early in the study.
173
-------
84-121.3
Further information on sample selection is provided by Johnson and by
Hartwell, et a!.2
Data Collection Instruments and Procedures
The data collection instruments used in the Denver CO study included
three questionnaires [screening questionnaire, computer model input
questionnaire (CMIQ), and participant questionnaire] providing background
data on subjects and their families, a network of 15 fixed-site monitors,
the PEM's and activity diaries carried by each subject, and breath sample
bags. These instruments and the procedures employed in using them are
described in detail by Johnson.1
The screening questionnaire was administered on a household basis as
a means of identifying persons eligible for the study. It requested the
name of each household member, relationship to head of household, sex,
age, smoking status, occupation, and typical commute time. The completed
screening questionnaires yielded a list of 2232 eligible individuals from
which were selected a stratified sample of 1139 potential subjects. An
attempt was made to administer the CMIQ to each potential subject. Part
A of the CMIQ requested detailed data about the commuting habits of the
respondent's household and determined if any member of the household was
employed in one of nine occupational categories associated with high CO
exposure. These data were collected for use in SHAPE, a population
exposure model developed by Wayne Ott,3 and NEM, a population exposure
model developed by the Strategies and Air Standards Division of EPA.4
Part B of the CMIQ verified the respondent's address and attempted to set
up-an*appointment for the first visit by an interviewer. The participant
questionnaire was administered to each of the 454 persojis who actually
participated in the study. It included detailed questions about the
subject's home environment, work environment, commuting habits, occupa-
tion, leisure-time activities, and shopping habits. The participant
questionnaire also requested age, sex, and education data.
A PEM and an activity diary were provided to each subject for each
of two 24-hour periods. The PEM was a modified General Electric (GE)
Carbon Monoxide Detector, Model 15EC53003, mated with a modified Magus
DL-1 Data Logger and mounted in a compact, tamperproof casing (Figure 1).
The PEM recorded the time and a CO concentration value every time the
"activity button" on the top of the instrument was pushed and every hour
on the hour. In both cases, the CO value was the integrated average CO
concentration since the last recorded value. Each PEM was capable of
operating continuously for 24 hours and logging up to 113 data points.
Quality assurance activities associated with the PEM's included daily
zero-span checks, frequent multipoint calibrations, special studies
evaluating precision, and two independent audits.
The activity diary contained instructions for completing the diary,
examples of properly completed diary pages, and 64 blank pages for re-
cording activities. The subject was instructed to fill out a diary page
whenever the subject changed location or activity. Data entered on each
diary page included time, activity (e.g., cooking dinner), location
(e.g., indoors residence), address, mode of transit if applicable, and
whether smokers were present (Figure 2). For indoor locations, subjects
174
-------
84-121.3
CAP
INLST
PLOWM6TEH
(SCAUH 1:2 APPROXIMATELY)
Figure 1. Personal exposure monitor.
175
-------
TIME FROM MONITOR
A. ACTIVITY
0. ONLY IF IN TRANSIT
(1) Start address
B. LOCATION
In transit 1
Indoors, residence 2
Indoors, office 3
Indoors, store 4
Indoors, restaurant 5
t
Other Indoor location 6
•_ Specify: •
Outdoors, within 10 yards of road
or street 7
Other outdoor location
Specify:
. 8
Uncertain 9
C. ADDRESS (if not in transit)
(2) End address
(3) Mode of travel:
Walking 1
Car 2
Bus . . . . 3
Truck 4
Train/subway 5
Other ........... 6
Specify:
ONLY IF INDOORS
(1) Garage attached to building?
Yes . ." 1
No 2
Uncertain 3
*
(2) Gas'stove 1n use?
Yes 1
No 2
Uncertain 3
ALL LOCATIONS
Smokers present?
Yes i
No 2
Uncertain 3
Figure 2. Page from activity diary.
176
-------
Indicated whether a garage was attached to the building and whether a gas
stove was in use.
Thirteen interviewers were employed during the course of this study
to deliver PEM's, activity diaries, and participant questionnaires to the
subjects according to prescheduled appointments. Because different PEM's
and activity diaries were used for the two sampling periods, an inter-
viewer visited each subject on three consecutive days. In most cases,
the first PEM and activity diary were delivered between 7 p.m. and
9 p.m. on Day,A and picked up 24 hours later on Day B. During pickup,
problems encountered during the first sampling period were addressed and
a second PEM and a second activity diary were delivered. These were
subsequently picked up 24 hours later on Day C. Breath samples were
taken during pickups on Days B and C. The participant questionnaire was
delivered on Day A and picked up on Day C.
A field data sheet was used to record the PEM values and correspond-
ing coded activity diary data for each subject-day. These sheets were
validated using a special computer program which checked for 83 different
types of-'data anomalies, Including missing entries, illegal entries, and
logical inconsistencies.
Breath samples were taken by having each subject blow through a
disposable mouth piece into a 600 ml plastic carboxyhe'moglobin bag. To
measure the CO concentration of the breath sample, a prefilter contain-
ing potassium permanganate and activated carbon was inserted between the
mouthpiece and a General.Electric CO-3 portable CO monitor.
Fifteen fixed-site monitors operated in Denver during the period
of the study (Figure 3). Nine of these monitors were temporary and were
discontinued at the conclusion of the study. All of the monitors re-
ported hourly-average CO data and operated continuously.
Study Results
Response Rates and Instrument Performance
A total of 1094 subject-days of participation were scheduled. The
454 individuals who actually participated in the study yielded 900
subject-days; 446 subjects participated in two sampling periods, while
8 subjects participated in only one sampling period. Of the remaining
194 subject-days scheduled, 120 were lost because subjects requested
rescheduling, 33 were lost because of last-minute refusals to partici-
pate, and 41 were lost for other reasons (e.g., subject missed appoint-
ment, interviewer experienced car problems).
Of the 899 person-days of data obtained from the participants, 808
data sets (90%) were coded as acceptable for statistical analysis of PEM
values. Of the remaining 91 data sets, 50 were coded as unacceptable
because the difference between pre and post zero-span values was judged
excessive. Other frequently occurring instrument problems included
clogged pumps, low battery voltage, instances when the PEM logic system
switched out of the data recording mode, and fragile parts.
177
-------
--J
CO
CO
r\j
Figure 3. Locations of fixed-site monitors,
-------
Multipoint calibrations performed early in the study revealed a
potential nonlinearity problem in the low concentration portion of the
PEM's operating range. The adverse affects of this nonlinearity on the.
overall data quality were minimized by insuring that the PEM GE sensor
outputs were properly balanced to the output of the Magus data subsystem
outputs.
The accuracy of PEM measurements was determined daily based on a
pre- and post-sampling check of zero and span. Using the change in slope
as a measure of accuracy, 95 percent of the measurements were estimated
to be within +10 percent of the true concentration value. PEM's operated
in pairs showed a mean percent difference in paired values of 5.0 percent
with a standard deviation of 14.2 percent. PEM's attached to manifolds
supplying sample ambient air to'fixed-site monitors yielded paired values
with a mean difference of 8.3 percent (fixed-site being higher) and a
standard deviation of 22 percent.
A total of 859 data sets (96%) were coded as acceptable for statis-
tical analysis of diary entries. In addition, 778 data sets (87%) were
coded as acceptable for statistical analyses involving both PEM and
diary data.
A total of 859 breath samples were obtained and successfully analy-
zed for CO content. Thirty samples were lost because of leaks in the
sample bag. One subject refused to provide a breath sample, and another
was unable to provide a sample because of illness. Nine samples were not
obtained -for other reasons (e.g., subject could not fill breath bag). An
analysis relating breath sample CO concentrations to CO exposures has
been performed by L.A. Wallace, et al.5
Fixed-Site Monitoring Data
The highest 1-hour CO concentration reported by any of the 15 fixed-
site monitors during the study period was 44.1 ppm. Only one fixed-site
monitor (060580002F01) reported any daily maximum 1-hour values exceeding
35 ppm, the current 1-hour NAAQS. The highest 8-hour CO concentration
reported by any of the 15 fixed-site monitors was 20.7 ppm. Eleven of
the 15 fixed-site monitors reported daily maximum 8-hour values exceeding
9 ppm, the current 8-hour NAAQS (Table I). Five fixed-site monitors
reported daily maximum 8-hour values exceeding 15 ppm.
Frequency Distribution of Daily Maximum
One-Hour and Eight-Hour Exposures
The daily maximum 1-hour and 8-hour exposures calculated for the
study sample were extrapolated to the Denver target population using
weighting factors which accounted for the probability of selecting a
particular subject into the sample and for nonresponse caused by re-
fusals, instrument problems, and unacceptable activity diary data.
Table II summarizes these results. The weighted means for daily maximum
1-hour and 8-hour exposures during the study period are 10.0 ppm and 4.9
ppm, respectively. Approximately 3 percent of the daily maximum 1-hour
exposures exceeded 35 ppm; approximately 11 percent of the daily maximum
8-hour exposures exceeded 9 ppm.
179
-------
Table I. Summary statistics for dally maximum 8-hour carbon monoxide values reported
by Denver monitoring sites between November 1, 1982, and February 28, 1983.
Hap
code
A
B
C
0
E
F
G
H
I
J
K
L
M
N
0
SAROAD code
060580002F01
060580014F01
06058001 3F01
062080821F05
062080822F05
062080820F05
062080823F05
062080825F05
062080818F05
062080819F05
060140002F01
060120002F01
060080002F01
062080824 F05
0620808 17F05
Composite
Number
of
dally max.
values
120
108
107
113
118
118
114
113
112
108
118
118
116
116
114
120
»
Dally maximum 8-hour carbon monoxide concentration, ppm
Minimum
1.2
0.4
1.4
0.8
0.4
0.5
0.7
0.6
0.7
0.5
1.0
0.7
0.2*
1.2
0.1
0.6
Maximum
20.7
18.5
13.1
15.2
14.1
15.1
7.8
7.2
13.6
15.2
9.3
13.2
5.8 ,
13.5
8.6
10.3
Mean
7.66
6.11
6.31
5.13
4.14
5.16
3.05
2.48
4.97
5.00'
2.98
4.86
1.57
5.01
3.01
4.16
Std. dev.
3.97
3.60
2.86
2.85
2.45
2.84
. 1.51
1.40
2.93
2.96
1.51
2.40
1.08
2.79
.1,58
2.01
Percentlles
10
3.4
1.9
2.5
1.9
1.5
1.9
1.6
1.1
1.8
2.1
1.5
2.0
0.5
2.2
1.3
1.7
25
4.9
3.3
3.9
3.1
2.4
3.0
2.0
1.4
2.8
3.1
1.9
2.8
0.9
2.7
1.9
2.9
50
6.8
5.9
5.7
4.5
3.4
4.9
2.6
2.1
4.2
4.2
2.7
4.5
1.3
4.4
2.7
3.7
75
9.6
7.9
8.5
6.2
5.6
6.6
4.0
3.2
6.4
6.2
3.7
6.5
1.8
6.4
3.8
5.3
90
13.7
10.9
10.3
9.5
7.5
9.0
5.1
4.1
9.3
8.9
5.0
8.1
3.3
9.5
5.2
7.0
CO
o
-------
Table II.
84-121.3
Summary statistics for daily maximum 1-hour and 8-hour
carbon monoxide exposures (weighted).
Statistic
Minimum
Maximum
Mean
10th percentile
25th percentile
50th percentile
75th percentile
90th percentile
95th percentile
99th percentile
Dail^ maximum exposure, _pp_m
1-hour
0.0
91.2
10.0
2.2
4.1
8.0
12.7
18.5
26.3
47.0
8-hour
0.0
44.0
4.9
1.1
2.0
3.5
6.9
9.4
12.1
25.6
Linear regression analyses were performed to relate daily maximum
1-hour exposures (c, ) to daily maximum 8-hour exposures (CQ „, ).
i jUiax o, max
Omitting one outlier and using the daily ma-ximum 8-hour value as the
independent variable, weighted linear regression of 801 person-days of
data yields the-relationship
1, max
(1)
with R = 0.69. Omitting the same outlier and using the daily maximum
1-hour value as the independent variable yields the relationship
8, max ' °'36 +<0.45)(Cl
(2)
again R = 0.69. Equation 1 predicts that a member of the Denver target
population who receives a daily maximum 8-hour exposure of 9 ppm would
receive a daily maximum 1-hour exposure of 16.3 ppm. Similarly, Equation
2 predicts a person receiving a daily maximum 1-hour exposure of 35 ppm
would receive a daily maximum 8-hour exposure of 16.1 ppm.
Variation in Exposure with Microenvironment
One of the principal reasons for collecting activity diary data is
to provide a means for relating exposure to a subject's microenvironment,
i.e., the subject's immediate physical surroundings. The activity diary
codes used in the Denver study can be combined in a variety of ways to
designate microenvironments of interest. The initial analyses discussed
here considered the four-digit code created by combining the two-digit
location code (diary item B) with the two-digit transit mode code (diary
item D3).
181
-------
84-121.3
Using valid Individual PEM values with durations of 60 minutes or
less, the weighted means and standard deviations of PEM values grouped
by microenvlronment code were calculated. Listing the mlcroenvlronments
1n descending order by mean CO concentration (Table III) suggests that
mlcroenvlronments associated with motor vehicles had the highest CO
levels 1n Denver during the study period.
Occupancy period was defined as the time a subject spends in a
microenvlronment during a single visit. Table IV lists the weighted
mean of the occupancy periods for each microenvlronment except Indoors -
residence (Code 02bb). Mean occupancy periods for the Indoor residential
microenvlronment could not be determined accurately from activity diary
data because subjects were usually occupying residences before the first
diary entry and after the last diary entry.
Mean occupancy periods range from 431.9 minutes (indoors - manufac-
turing facility) to 7.4 minutes (outdoors - residential garage or car-
port). Mean occupancy periods for in-transit microenvlrotiments associ-
ated with motor vehicles and high CO levels are 30.8 minutes for trucks,
28.0 minutes for buses, 25.9 minutes for cars, and 23.0 minutes for
motorcycles. The value for Indoors - public garage (29.4 minutes)- 1s
higher than expected and may be the result of errors In recording
activity diary Information.
An analysis was conducted of residential Indoor exposures to deter-
mine the contribution of three potential CO sources. Mean exposure was
Increased 2.59 ppm (134 percent) by gas stove operation, 1.59 ppm (84
percent) by smokers other than study participants, and 0.41 ppm (22 per-
cent) by attached garages. As noted previously, only- nonsmokers were
invited to participate in the study.
Relationships Between Fixed and Personal Monitor Values
Some models used for estimating population exposure assume that a
strong, linear relationship exists between CO levels in certain micro-
environments and CO levels measured simultaneously at fixed-site moni-
tors. This assumption was Investigated by performing linear regression
analyses that used PEM values grouped by mlcroenvironment as the
dependent variable and fixed-site values as the independent variable.
For In-transit mlcroenvlronments, the independent variable was the mean
of the simultaneously-recorded values at all 15 sites. For nontransit
mlcroenvlronments, the Independent variable was the simultaneously-
recorded value at the nearest fixed-site monitor. Coefficients of
determination (R2) range from 0 to 0.58 (Tables V and VI). Most are
less than 0.50. Mlcroenvlronments with R2 values exceeding 0.30 Include
parks and golf courses, motorcycles, and buses. The residential garage
microenvlronment has an R2 value of zero.
An untried method of assigning fixed-site monitors to census tracts
1s to determine the traffic density of each census tract and then select
a fixed-site monitor located in a census tract with similar traffic
density. Such an approach may yield higher correlations between the
nontransit PEM values and the fixed site values than those listed in
Table V.
182
-------
84-121.3
Table III. Microenvironments listed in descending order of
weighted mean CO concentration.
Code
B
52
01
54
01
72
01
71
62
01
55
53
51
07
01
bb
05
74
03
71
56
04
80
59
61
53
02
77
60
57
76
01 -
78
79
03
a
93
bb
03
a
02
bb
bb
04
bb
bb
bb
c
01
bb
bb
bb
"c
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
92
bb
bb
Microenvironment
Category
Indoors
In transit
Indoors
In transit
Outdoors
In transit
Outdoors
Indoors
In transit
Indoors
Indoors
Indoors
Outdoors
In transit
Not specified
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Outdoors
In transit
Outdoors
Outdoors
Sub-category
Public garage
Motorcycle
Service station or motor
vehicle repair facility
Bus
Public garage
Car
Residential garage or
carport
Other location
Truck
Other repair shop
Shopping mall
Residential garage
Within 10 yards of road
Walking
Not specified
Restaurant
Service station or motor
vehicle repair facility
Office
Parking lot
Auditorium, sports arena,
concert hall, etc.
Store
Other location
Health care facility
Other public building
Manufacturing facility
Residence
School grounds
School
Church
Residential grounds
•Bicycle
Sports arena, amphitheater,
etc.
Park or golf course
n
116
22
125
76
29
3632
22
427
405
55
58
66
496
619
586
524
12
2287
61
100
734
• 126
351
115
42
21543
16
426
179
74
9
29
21
CO concentration,
ppm
'Mean
13.46
9.79
9.17
8.52
8.20
8.10
'7.53
7.40
7.03
5". 64
4.90
4.35
4.05
3.88
3.79
3.71
3.68
3.59
3.45
3.37
3.23
3.17
2.22
2.15
2.04
2.04
1.99
1.64
1.56
1.36
1.34
0.97
0.69
Std. dev.
18.14
8.15
9.33
7o08
5.33
9.88
8.93
17.97
9.89
7.67
6.50
- 7.06
5.44
6.61
6.57
4.35
.
3.84
4.18
4.23
4.76
5.56
5.47
4.25
3.26
2.55
4.06
3.39
2.76
3.35
2.24
3.61
2.80
1.01
Includes D3 = bb, 01, and 02.
JB1ank.
"Includes 03 = bb and 01.
Includes 03 = bb, 01, 02, and 03,
183
-------
Table IV. Microenvironments listed in descending order of weighted
mean occupancy period.
Code
B
53
03
55
78
60
59
57
56
80
62
05
61
58
54
79
04
bb
07
01
76
52
01
01
01
51
01
73
01
77
72
74
71
03
bb
c
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
c
04
bb
d
03
02
93
bb
01
e
92
bb
d
bb
bb
Microenvironment3
Category
Indoors
Indoors
Indoors'
Outdoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Indoors
Not specified
Outdoors
In transit
Outdoors
Indoors
In transit
In transit
In transit
Indoors
In transit
Outdoors
In transit
Outdoors
Outdoors
Outdoors
Outdoors
Subc'ategory
Manufacturing facility
Office
Other repair shop
Sports arena, amphitheater,
etc.
School
Wealth care facility
Church
Auditorium, sports arena,
concert hall, etc.
Other location
Other location
Restaurant
Other public building
Shopping mall
Service station or motor
vehicle repair facility
Park or golf course
Store
Not specified
Within 10 yards of road
Truck
Residential grounds
Public garage
Bus
Car
Motorcycle
Residential garage
Walking
Parking lot
Bicycle
School grounds
Public garage
Service station or motor
vehicle repair service
Residential garage or carport
n
8
610
11
6
100
91
57
38
60
188
267
52
24
53
11
400
320
379
-300
49
77
68
2593
17
49
511
67
13
15
26
12
20
Occupancy period,
minutes
Mean
431.9
206.9
192.4
191.6
165.0
162.4
115.7
112.0
93.1
91.1
79.1
77.8
77.3
72.3
60.4
48.4
37.3
31.2
30.8
30.8
29.4
28.0
25.9
23.0
23.0
20.0
16.4
16.1
12.8
10.7
7.5
7.4
Std. dev.
199.8
167.3
121.3
• 111.0
159.8
142.5
57.3
77.0
182.8
103.9
109.2
' 60.8
65.0
87.1
84.3
73.2
57.1
56.6
43.4
36.2
77.1
24.3
46.1
9.9
39.5
46.9
37.8
10.2
6.8
26.0
2.4
17.5
aOmits indoor residential microenvironment (Code 02bb).
bBlank.
Includes 03
Includes 03
Includes 03
bb and 01.
bb, 01, and 02.
bb, 01, 02, and 03.
184
-------
Table V. Results of weighted linear regression analyses with
nontransit PEM value as dependent variable and simultaneous
value at nearest fixed site as independent variable.
Code
B
80
79
77
54
05
74
07
57
73
55
78
61
58
04
59
02
60
03
71
bb
76
e
56
53
51
62
03
bb
bb
bb
bb
bb
bb
c
bb.
d
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
bb
Microenvironment3
Category
Outdoors
Outdoors
Outdoors
Indoors
Indoors
Outdoors
Outdoors
Indoors
Outdoors
Indoors
Outdoors
Indoors
Indoors
-Indoors
Indoors
Indoors
Indoors
Indoors
Outdoors
Not
specified
Outdoors
-
Indoors
Indoors
Indoors
Indoors
Subcategory
Other location
Park or golf course
School grounds
Service station or
motor vehicle
repair facility
Restaurant
Service station or
motor vehicle
repair facility
Within 10 yards of
road
Church
Parking lot
Other repair shop
Sports arena, amphi-
theater, etc.
Other public building
Shopping mall •
Store
Health care facility
Residence
School
Office
Residential garage
or carport
Not specified
Residential grounds
Public garage
Auditorium, sports
arena, concert
hall, etc.
Manufacturing
facility
Residential garage
Other location
Linear regression
n
115
18
15
112
486
-
11
468
178
51
46
16
111
55
675
336
20969
342
2090
22
583
70
139
94
41
66
381
Intercept
0.35
-0.09
-0.37
4.18
1.69
1.61
1.58
0.09
2.26
3.69
3.05
0.74
1.24
1.67
0;97
1.00-
0.97
2.53
5.67
0
2.07
0.84
8.44
2.25
1.41
3.98
7.94
Slope
1.11
0.39
1.15
1.68
0.76
1:21
0.89
0.70
' 0.60
0.88
-1.76
0.42
1.43
0.56
0.45
0.43
0.32
0.34
0.61
0.63
0.30
0.72
0.38
0.18
0.14
0.07
R*
0.46
0.44
0.27
0.27
0.25
0.23
0.21
0.21
0.21
0.18
0.15
0.14
0.14
0.09
0.09
0.07
0.07
0.05
0.05
0.05
0.04
0.04
0.04
0.03
0.00
0.00
P
0.000
0.003
O.Q4S
0.000
0.000
0.134
0.000
0.000
0.001
0.003
0.128
0.000
0.005
0.000
0.000
0.000
0.000
p. ooo
0.304
0.000
0.099
0.019
0.060
0.246
0.662
0.791
a 2
Listed in order of R value.
bBlank.
Includes 03 = bb and 01.
Includes 03 = bb, 01, 02, and 03.
Includes codes 52bb, 5201, 5202, 72bb, 7201, and 7202.
pProbability that slope = 0.
185
-------
84-121.3
Table VI. Results of weighted linear regression analyses with
in-transit PEM value as dependent variable and simultaneous
value from composite data set as independent variable.
Code
B
01
01
01
01
01
01
03
93
03
01
04
02
a
In-transit
subcategory
Motorcycle
Bus
Walking
truck
Car
All
Linear regression
n
22
76
619
405
3632
4763
Intercept
4.50
3.17
0.06
3.27
6.01
5.15
Slope
2.14
2.02
1.47
1.54
0.78
0.92
R2
0.58
0.36
0.23
0.11
0.04
0.05
P
0.000
0.010
0.000
0.000
0.000
0.000
dIncludes D3 codes 01, 02, 03, 04, 92, and 93,
pProbab1lity that slope » 0.
Diurnal patterns for weekdays (Figure 4), Saturdays, and Sundays
were developed for hourly average exposures, and composite fixed-site
values. In general, diurnal patterns for exposure were similar in shape
ta those for fixed-site data, although the morning rush hour peaks were
much higher in the composite fixed-site patterns than in the exposure
patterns.
Conclusions
In developing and implementing the Denver study, the attempt was
made to investigate the appropriateness of a general approach to deter-
mining the exposure of a large urban population. The overall success of
the Denver study suggests that the approach is valid. The study has also
provided a rich data base that should prove invaluable in answering
questions concerning the factors which affect exposure, the ability of
fixed-site data to represent personal exposures, the performance of
newly-developed instruments, and similar issues. The analyses .discussed
in this report suggest that 1) CO exposures in microenvironments assoc-
iated with motor vehicles are higher than exposures in microenvironments
not associated with motor vehicles, 2) CO exposures in the microenviron-
ments defined for this study are not strongly correlated with CO concen-
trations simultaneously recorded at fixed-site monitors, and 3) indoor
residential exposures are increased by gas stoves, smokers, and attached
garages.
References
1. T. Johnson, A Study of Personal Exposure to Carbon Monoxide in
Denver. Colorado, report by PEDCo Environmental, Inc., to the U.S.
Environmental Protection Agency, Research Triangle Park, N.C.,
December 1983.
186
-------
84-121.3
2. T. Hartwell, et al., ^tud_y_ of Carbon Monoxide Exposure of Residents
of Washington. DC and Denver, Colorado, report by Research Triangle
Institute to U.S. Environmental Protection Agency, Research
Triangle Park, N.C., January 1984.
3. W.R. Ott, "Exposure estimates based on computer generated activity
patterns," paper no. 81-57.6, 74th Annual Meeting of the Air Pollu-
tion Control Association, Philadelphia, June 21-26, 1981.
4. T. Johnson and R. Paul, The NAAQS Exposure Model (NEM). Applied to.
Carbon Monoxide, draft report by PEDCo Environmental, Inc., to the
U.S. Environmental Protection Agency, Research Triangle Park, N.C.,
September 1983.
5. L.A. Wallace, et al., "Alveolar CO Measurements of 10,00 Residents
of Denver and Washington, D.C.--A Comparison with Preceding Personal
Exposure," paper no. 121.5, to be presented at the 77th Annual
Meeting of the Air Pollution Control Association, San Francisco,
June 24-29, 1984.
6.0
- S.o
4.0
u
§ 3.0
§ 2.0
CO
1.0
•n
I
I
I
I
I
I
I
I
I
PERSONAL EXPOSURE
COMPOSITE FIXED SITE
i—
J I L
•4
10 12 14
HOUR
16 18 20
22 24
Figure 4. Weekday diurnal patterns.
187
-------
APPENDIX B
SITE DESCRIPTIONS AND SUMMARY STATISTICS FOR WASHINGTON
FIXED-SITE MONITORS
EMSL provided PEI with a file containing hourly average carbon monoxide
(CO) data for 11 fixed sites in the Washington, D.C., area. Tables 1 and 2
provide site characteristics for these 11 sites. As discussed in Section
5.1 of Reference 1, TDEN is an indicator of traffic density, and RDEN is an
indicator of population density. Figure 1 shows the approximate location of
each site.
A twelfth data set was created by taking the hour-by-hour mean of the
hourly values reported by the 11 fixed sites. This data set is referred
to as the "composite site" in the discussion that follows.
Table 3 summarizes the results of analyzing the hourly average values for
November 8, 1982, through February 25, 1983, using BMDP program P2D. None of
the sites reported hourly average values exceeding 35 ppm. Table 4 lists the
date and time of the maximum value reported at each site. Eight of the 11
maximum values occurred during either the morning or the evening high traffic
periods. Three days (11-8-82, 2-15-83, and 2-22-83) 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 2-15-83.
PEI created a supplementary file which contains daily maximum 1-hour and
8-hour values. Tables 5 and 6 summarize the results of analyzing these data
using BMDP program P2D. As indicated in Table 6, 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.
Reference
1. Johnson, T. A Study of Personal Exposure to Carbon Monoxide in
Denver, Colorado. U.S. Environmental Protection Agency, Research
Triangle Park, North Carolina. EPA-600/54-84-014, March 1983.
188
-------
TABLE 1. SITE CHARACTERISTICS OF WASHINGTON CARBON MONOXIDE MONITORS OPERATING DURING STUDY
Map
code
A
B
C
D
E
F
G
H
I
Location
District of
Columbia
Bladensburg,
MD
•Suitland-
Silver
Hill, MD
Alexandria,
VA
Arlington,
VA
Fairfax, VA
Mt. Vernon,
VA
SAROAD code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H01
480200020G01
481040005G01
481060018G01
Address
24th and L. Sts.,
NW (West End
Library)
L St. between 20th
and 21st Sts.,
NW
First and C Sts. ,
SW
Educational Media
Bldg.
Suitland Parkway
(near Bramley
Ave.)
517 N. St. Asaph
St. (near
Pendleton)
S. 18th and
S. Hayes Sts.
10600 Page Avenue
2675 Sherwood Hall
Cn.
Scale9
N
M
?
?
N
i
N
1 N
N
N
Probe
ht,
ft
33
11
11
13
14
36
16
12
12
Distance
to road,
ft
83
70
345
18
210
250
50
80
180
150
40
40
200
200
180
<100
190
250
180
Vehicles
per dayb
12,000
4,500
27,000
20,900
13,100
12,800
?
?
37,000
19,700
3,900
3,700
<500
<200
6,000
<200
17,900
8,250
Immediate area
land use
Commercial
Street corridor
Office build-
ings
Residential and
light commercial
Field near com-
mercial street
Light commercial
and residential
Commercial and
residential
Office build-
ings
Light commer-
cial and resi-
dential
(continued)
-------
TABLE 1 (continued)
Map
code
J
K
Location
McLean, VA
Seven Cor-
ners, VA
SAROAD code
481850001G01
482870004G01
Address
1437 Balls Hill Rd.
6100 Arlington
Blvd. (roof of
Montgomery Ward)
Scale3
N
N
Probe
ht,
ft
12
30
Distance
to road,
ft
260
216
430
328
800
800
Vehicles
per dayb
31,750
3,189
10,832
50,000
11,658
1,265
Immediate area
land use
Light commer-
cial and resi-
dential
Strip develop-
ment near resi-
dential
N = neighborhood, M = micro.
Estimate, accuracy uncertain.
-------
TABLE 2. LAND USE CHARACTERISTICS OF CENSUS TRACTS
CONTAINING WASHINGTON CO MONITORS
Map
code
A
B
C
D
E
F
G
H
I
J
K
Location
District of Columbia
Bladensburg, MD
Suitland-Silver Hill,
MD
Alexandria, VA
Arlington, VA
Fairfax, VA
Mt. Vernon, VA
McLean, VA
Seven Corners, VA
SAROAD code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H01
480200020G01
481040005G01
481060018G01
481850001G01
482870004G01
1980
census
tract
55.02
54.01
62.02
8043.00
8024.01
2018.02
1035.00
4405.00
4159.00
4706.00
5003.00
1970
census
tract
55.00
54.10
62.00
8040.00
8024.01
18.00
1035.00
4031.00
4008.00
4083.00
5003.00
TDEN
• 62.94
243.97
19.39
7.96
7.34
12.45
18.63
1.64
0.76
8.11
5.93
RDEN
26.19
21.51
0.04
7.44
6.03
6.82
19.88
3.08
3.03
5.29
6.58
191
-------
IO
ro
Figure 1. Locations of fixed-site monitors (base
map(pRand McNally and Company, used by
permission).
-------
TABLE 3. SUMMARY STATISTICS FOR HOURLY AVERAGE CARBON MONOXIDE VALUES REPORTED
BY WASHINGTON MONITORING SITES BETWEEN NOVEMBER 8, 1982 AND FEBRUARY 25, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
SAROAD code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H01a
480200020G01a
481040005G01a
481060018G01a
481850001G013
482870004G01a
Composite
Number
of
hourly
values
2626
2611
2377
2317
2015
2563
2619
2527
2608
2572
2564
2640
Hourly average carbon monoxide concentration, ppm
Minimum
0.1
0.1
0.1
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.2
Maximum
11.2
16.0
10.0
22.4
9.1
14.0
8.5
8.0
17.0
9.5
13.0
8.6
Mean
1.48
2.72
1.84
1.65
1.39
1.38
1.29
1.26
1.78
1.05
i
0.98
1.54
Std. dev.
1.22
2.22
1.14
2.22
0.97
1.24
1.05
0.94
1.82
1.06
0.91
1.09
Percent! les
10
0.5
0.5
0.8
0.2
0.4
0.5
0.5
0.5
0.5
0.0
6.0
0.6
25
0.7
1.1
1.1
0.4
0.7
0.5
0.5
1.0
1.0
0.5
0.5
0.8
50
1.2
2.1
1.6
0.9
1.2
1.0
1.0
1.0
1.0
0.5
1.0
1.2
75
1.8
3.7
2.2
1.9
1.8
1.5
1.5
1.5
2.0
1.5
1.0
1.9
90
2.8
5.7
3.1
4.0
2.6
3.0
2.5
2.0
4.0
2.5
2.0
2.8
98
5.4
9.0
5.4
8.9
4.1
5.0
5.0
4.5
7.5
4.5
3.5
5.0
00
Data reported in units of 0.5 ppm.
-------
TABLE 4. 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
2-15-83
11-08-82
2-22-83
2-22-83
2-15-83
2-22-83
2-22-83
12-08-82
2-15-83
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
194
-------
TABLE 5. SUMMARY STATISTICS FOR DAILY MAXIMUM 1-HOUR CARBON MONOXIDE VALUES REPORTED
BY WASHINGTON MONITORING SITES BETWEEN NOVEMBER 8, 1982 AND FEBRUARY 25, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
SAROAD code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H013
480200020G01a
481040005G01a
481060018G013
481850001G01a
482870004G01a
Composite
Number
of
daily
values
110
108
96
93
80
108
110
105
108
106
104
110
Daily maximum 1-hour carbon monoxide concentration, ppm
Minimum
0.7
1.5
0.9
'0.2
0.4
0.5
0.5
0.5
0.5
0.5
0.5
0.8
Maximum
11.2
16.0
10.0
22.4
9.1
14.0
8.5
8.0
17.0
9.5
13.0
8.6
Mean
3.56
6.74
3.61
5.12
2.84
3.59
2.97
2.86
4.86
2.85
*
2.32
3.24
Std. dev.
2.28
3.30
1.97
4.31
1.66
2.34
1.86
1.78
3.20
2.00
1.79
1.78
Percent iles
10
1.4
2.4
1.8
1.4
0.9
1.5
1.0
1.0
1.5
1.0
1.0
1.4
25
1.8
4.0
2.4
2.4
1.8
1.5
1.5
1.5
2.0
1.5
1.0
1.9
50
2.7
6.6
3.0
3.9
2.3
3.0
2.5
2.5
4.5
2.5
2.0
2.8
75
4.5
9.3
4.3
6.5
3.6
5.0
4.5
3.5
6.5
4.5
3.0
4.1
90
6.9
11.0
6.9
9.2
5.2
7.0
5.5
5.5
9.5
5.5
4.5
5.9
98
9.3
14.1
9.5
19.8
7.6
9.0
7.5
7.0
12.0
8.0
7.5
7.9
Data reported in units of 0.5 ppm.
-------
TABLE 6. SUMMARY STATISTICS FOR DAILY MAXIMUM 8-HOUR CARBON MONOXIDE VALUES REPORTED
BY WASHINGTON MONITORING SITES BETWEEN NOVEMBER 8, 1982 AND FEBRUARY 25, 1983
Map
code
A
B
C
D
E
F
G
H
I
J
K
SAROAD code
090020017101
090020023102
090020031102
210220001F01
211560001F01
480080009H013
480200020G013
481040005G013
481060018G013
481850001G01a
482870004G01a
Composite
Number
of
daily
values
110
109
96
89
79
105
110
105
107
105
103
110
Daily maximum 8-hour carbon monoxide concentration, ppm
Minimum
0.6
0.8
0.7
0.2
0.3
0.5
0.5
0.5
0.6
0.4
0.3
0.7
Maximum
7.7
10.0
5.9
12.5
5.6
7.1
6.3
6.1
8.5
6.5
6.5
6.4
Mean
2.35
4.33
2.55
3.12
1.97
2.22
2.04
1.86
3.11
1.79
i
1.46
2.29
Std. dev.
1.34
1.98
1.14
2.76
1.05
1.42
1.27
1.05
1.98
1.21
0.86
1.20
Percent lies
10
1.1
1.9
1.4
0.8
0.7
0.8
0.9
0.9
1.0
0.6
0.5
1.1
25
1.3
2.8
1.8
1.4
1.2
1.3
1.0
1.3
1.6
0.8
0.9
1.4
50
2.0
4.4
2.2
2.1
1.9
1.8
1.6
1.6
2.3
1.5
1.3
1.9
75
2.9
5.6
3.1
4.1
2.6
2.8
2.8
2.3
4.2
2.4
1.8
2.7
90
4.2
7.1
4.2
7.2
3.4
4.2
3.6
2.9
6.3
3.6
2.3
4.1
98
5.6
8.2
5.5
12.2
4.3
6.4
5.2
5.0
8.2
4.9
3.6
5.4
Data reported in units of 0.5 ppm.
------- |