EPA-600/5-77-010
July 1977 Socioeconomic Environmental Studies Series
AIR POLLUTION AND HEALTH IN
WASHINGTON, D.C.:
Some Acute Health Effects Of
Air Pollution In The
Washington Metropolitan Area
Environmental Research Laboratory
Office of Research and Development
U.S. Environmental Protection Agency
Corvallis, Oregon 97330
-------
RESEARCH REPORTING SERIES
Research reports of the Office of Research and Development. U.S. Environmental
Protection Agency, have been grouped into nine series These nine broad cate-
gories were established to facilitate further development and application of en-
vironmental technology. Elimination of traditional grouping was consciously
planned to foster technology transfer and a maximum interface in related fields
The nine series are:
1. Environmental Health Effects Research
2. Environmental Protection Technology
3. Ecological Research
4. Environmental Monitoring
5. Socioeconomic Environmental Studies
6 Scientific and Technical Assessment Reports (STAR)
7. Interagency Energy-Environment Research and Development
8 "Special" Reports
9. Miscellaneous Reports
This report has been assigned to the SOCIOECONOMIC ENVIRONMENTAL
STUDIES series. This series includes research on environmental management,
economic analysis, ecological impacts, comprehensive planning and fore-
casting, and analysis methodologies. Included are tools for determining varying
impacts of alternative policies; analyses of environmental planning techniques
at the regional, state, and local levels; and approaches to measuring environ-
mental quality perceptions, as well as analysis of ecological and economic im-
pacts of environmental protection measures Such topics as urban form, industrial
mix, growth policies, control, and organizational structure are discussed in terms
of optimal environmental performance. These interdisciplinary studies and sys-
tems analyses are presented in forms varying from quantitative relational analyses
to management and policy-oriented reports
This document is available to the public through the National Technical Informa-
tion Service, Springfield, Virginia 22161.
-------
EPA-600/5-77-010
July 1977
AIR POLLUTION AND HEALTH IN WASHINGTON, D.C.
Some Acute Health Effects
of Air Pollution in the
Washington Metropolitan Area
Eugene P. Seskin
National Bureau of Economic Research, Inc.
Washington, D.C. 20006
Contract No. 68-01-3144
Project Officer
John Jaksch
Criteria and Assessment Branch
Corvallis Environmental Research Laboratory
Corvallis, Oregon 97330
CORVALLIS ENVIRONMENTAL RESEARCH LABORATORY
OFFICE OF RESEARCH AND DEVELOPMENT
U.S. ENVIRONMENTAL PROTECTION AGENCY
CORVALLIS, OREGON 97330
EPA-RTF LIBRARY
-------
DISCLAIMER
This report has been reviewed by the Corvallis Environmental Research
Laboratory, U.S. Environmental Protection Agency, and approved for publica-
tion. Approval does not signify that its contents necessarily reflect the
views and policies of the U.S. Environmental Protection Agency nor does
mention of trade names or commercial products constitute endorsement or
recommendation for use.
The methods, results, and conclusions found in this study do not neces-
sarily reflect the opinions of the sponsoring agencies or the organizations
who provide data. Specifically, with regard to the data supplied by the
Grpup Health Association (GHA), it is recognized that: (1) The socio-demo-
graphic characteristics of the GHA membership may not be representative of
the socio-demographic characteristics of the entire metropolitan Washington
population; (2) utilization of medical services by GHA members may also be
different than utilization of medical services by the general Washington
population; and (3) only within-plan utilization is accounted for, i.e.,
some GHA members may, at times, seek outside medical care which is not
recorded in the data.
-------
FOREWORD
Effective regulatory and enforcement actions by the Environmental Protection
Agency would be virtually impossible without sound scientific data on pollu-
tants and their impact on environmental stability and human health. Responsi-
bility for building this data base has been assigned to EPA's Office of
Research and Development and its 15 major field installations, one of which
is the Corvallis Environmental Research Laboratory (CERL) in Oregon.
The primary mission of the Corvallis Laboratory is research on the effects of
environmental pollutants on terrestrial, freshwater, and marine ecosystems;
the behavior, effects and control of pollutants in lake systems; and the
development of predictive models on the movement of pollutants in the biosphere.
This report was initiated by the Washington Environmental Research Center,
Office of Research and Development, Washington, D.C. and completed at the
Corvallis Environmental Research Laboratory.
A.F. Bartsch
Director, CERL
111
-------
ABSTRACT
This study has attempted to assess some of the acute health effects of
air pollution. Specifically, the investigation has tested the hypothesis
that air pollution can aggravate the health status of a population and can
result in increased utilization of certain types of medical care services.
The study period was 1973-1974 and centered in the Washington, D.C.
Metropolitan Area. Statistical models were formulated, explaining health-care
utilization of a group practice medical care plan. Primary interest was
focused on the effects of mobile-source air pollutants including carbon
monoxide, nitrogen dioxide, non-methane hydrocarbons, and photochemical
oxidants. Meteorological conditions, as well as other variables thought to
influence the consumption of medical services, were included in the models
as explanatory variables.
The statistical results indicated that air pollution levels had a very
limited effect on the health-care utilization of the group practice.
This report is not an official National Bureau publication since the
findings reported herein have not undergone the full critical review accorded
the National Bureau's studies, including approval by the Board of Directors.
This report was prepared under contract number 68-01-3144 to U.S. Environ-
mental Protection Agency and the U.S. Department of Transportation. This
report covers the period from January 1, 1973, to December 31, 1974; and
work was completed as of March 1976.
IV
-------
CONTENTS
FOREWORD iii
ABSTRACT iv
ACKNOWLEDGEMENTS vll
SECTIONS
I INTRODUCTION 1
Background
Reseat'ch Objectives
An Overview of the Research
II SUMMARY OF FINDINGS 3
III INVESTIGATING THE AIR POLLUTION-HEALTH RELATIONSHIP
Introduction
Laboratory and ToxicoLogical Experimentation
Epidemiological Approaches
Multivariate Analyses
Results of a Similar Study
IV A DESCRIPTION OF THE STUDY AREA AND THE GROUP HEALTH
ASSOCIATION
The Study Area
The Group Health Association
COMPILING THE MEDICAL DATA 19
Inpatient vs. Outpatient Data
The Selected Outpatient Data
The Selection of Departments
Preparing the GHA Data for Use
VI COMPILING THE AIR POLLUTION AND METEOROLOGICAL DATA 27
The Air Pollution Data
Collecting and Preparing the Air Pollution Data for Use
The Meteorological Data
-------
VII STATISTICAL METHOD 35
Introduction
The Choice of Techniques
Theory and Method
Application and Hypothesis Testing
VIII EMPIRICAL RESULTS 42
Introduction
Discriminant Analyses
Regression Analyses
Lag and Episodic Effects
The Effects of Other Air Pollutants
IX DISCUSSION 74
Policy Implications
Economic Consequences
Future Research
REFERENCES 78
XI APPENDICES 82
Appendix A
Appendix B
Appendix C
Appendix D
Appendix E
Appendix F
-------
ACKNOWLEDGEMENTS
1 wish to acknowledge several people at the National Bureau who signifi-
cantly contributed to this research undertaking. Special appreciation must
go to Henry Peskin whose valuable suggestions provided guidance throughout the
research effort and whose authorship of Section VII provided a theoretical
background for the statistical analyses. 1 also wish to thank James Etnbersit
for the innumerable hours he spent at the computer terminal as well as his
other wide-ranging efforts in assisting the research. Research assistance
was also provided by Emily Eisner during the initial phase of the study and
additional computer programming skills were provided by Robert Chen. Great
patience was displayed by Lorraine Patterson in her typing and retyping of the
draft manuscript. John Nankin then did an excellent job in preparing the
final manuscript.
Several people associated with the Group Health Association must also be
acknowledged for their valuable contributions to the study. In particular, Bruce
Steinhardt provided many hours of guidance during our efforts to analyze the
Group Health data. Our initial discussions with Goldie Krantz led to the re-
sulting cooperation between GI1A and NBER. Peter Birk then continued this
cooperation between our two organizations.
Finally, those associated with this project, at both the Environmental
Protection Agency and the Department of Transportation, should be thanked
for their valuable comments and assistance throughout the endeavor.
VI1
-------
SECTION I
INTRODUCTION
BACKGROUND
A number of investigators have examined the health effects of air pollu-
tion episodes in cities such as New York!/ and Boston.I/ They have looked at
the effects of air pollution on indices of health (including mortality rates,
pediatric and adult clinic visits, emergency hospital admissions, and so on).
Although estimates of excess mortality, clinic visits, and hospital admissions
(over expected values of these indices) were computed, the studies did not
assess the economic impact of such air pollution episodes. Moreover, the air
pollutants considered in these studies were primarily those from stationary
sources (e.g., sulfur dioxide and smokeshade).^/
The absence of economic imputation in these studies represents a serious
deficiency. The ultimate objective of policies designed to deal with air pol-
lution problems is to attain ambient levels of pollution at which the corres-
ponding marginal social costs and benefits are balanced. Consequently, there
is a critical need for the translation of physical effects (e.g., increased
air pollution-associated illness) into economic terms that can be compared to
the societal costs associated with more stringent air quality criteria and
abatement activities.
RESEARCH OBJECTIVES
This study will attempt to assess the health effects and related impacts
of air pollution in the Washington Metropolitan Area. A major output of this
research will be the quantification of the economic costs of air pollution due
to its debilitating effects (if they exist) on the health status of the exposed
population. Special attention will be focused on mobile-source air pollution
(e.g., carbon monoxide, nitrogen oxides, and photochemical oxidants).
Specifically, the objectives to be accomplished by this research are as
follows;
(1) A test of the null hypothesis that there is no significant asso-
ciation between air pollution in Metropolitan Washington and
health effects.
(2) If the null hypothesis is rejected, an investigation of the
relationship between specific pollutants and specific health
effects.
(3) An assessment of the economic impacts of the air pollution-health
association both in terms of direct costs (e.?,. , medical treatment
expenditures) and indirect costs (e.g., work-days-lost).
!_/ See, for example, Greenburg, ej: al. f 19"|
2/ See Heimann [23],
_3/ For a survey of studies that have dealt with the health effects of air pollu-
tion, see Lave and Seskin [ 28"].
1
-------
AN OVERVIEW OF THE RESEARCH
The study area of this research is the Washington, D.C. Metropolitan Area
during the period 1973 and 1974. The nation's capital is of particular interest
because mobile-source air pollution is a dominant factor in Washington, whereas
stationary sources (e.g., power plants) are the chief contributors of air pol-
lution in other major eastern seaboard cities. In view of the controversies
surrounding environmental legislation of regulations on mobile-source emis-
sions, the results for the Washington area may be extremely significant for
policy purposes.
The data for this study came from several sources. The Group Health As-
sociation, a prepaid group practice medical care plan with approximately 100,000
members, provided the health data. The air pollution data used in the research
came from the Government of the District of Columbia, Department of Environ-
mental Services. Finally, the meteorological data were provided by the National
Weather Service.
Section II presents a summary of our findings. In Section III we shall
explore the methods that have been previously used to investigate the associa-
tion between air pollution and human health, and shall provide some illustra-
tions from the literature. Section IV includes a description of the study
area (Metropolitan Washington, D.C.) as well as a description of the primary
source of health data (The Group Health Association). In Section V, the
medical data and our method of compiling them are described in detail. Section
VI discusses the compilation of the air pollution and meteorological data. In
Section VII, the statistical methods used in the analysis are examined. Section
VIII presents the empirical results. Finally, Section IX presents some policy
implications, economic consequences, and future research needs.
-------
SECTION II
SUMMARY OF FINDINGS
We have analyzed health care utilization, air pollution, and weather
data in an effort to test the null hypothesis that there is no significant
association between air pollution levels in Metropolitan Washington, D.C. and
health effects.
According to an analysis of these data for 1973, there did not appear
to be a significant relationship between photochemical oxidant readings (as
measured by the maximum 1-hour average) on any given day and the number of
unscheduled visits on that day to three of the four departments at the
main clinic (Pennsylvania Avenue) of a local group practice medical plan.
These included the urgent visit clinic and the internal medicine and pedia-
tric departments. The only exception was for unscheduled visits to the
ophthalmology department. Here a suggestive relationship between oxidant
levels and unscheduled visits was noted for further investigation.
We then examined the 1974 data in an effort to replicate the 1973
findings. In general, the results were comparable to those for 1973.
Again, a positive and statistically significant relationship was exhibited
between the oxidant readings and unscheduled visits to the ophthalmology
department. In addition, a suggestive relationship between oxidant levels
and utilization of the urgent visit clinic was seen. This relationship was
further supported by the results of an analysis involving oxidant pollution
data monitored at another station. Finally, for 1974 oxidant data from one
of the monitoring stations, a positive and statistically significant associ-
ation was seen with unscheduled visits to the pediatric department. This
relationship, however, did not hold when oxidant data from the other
monitoring station were substituted.
Looking at both the 1973 and 1974 results together, the magnitude of
the association between unscheduled visits to the ophthalmology department
and photochemical oxidant levels indicated that a 10 percent decrease in
oxidant levels was related to between a 1.1 and 4.3 percent decrease in
unscheduled utiliEation of the ophthalmology department. The statistically
significant 1974 result pertaining to the urgent visit clinic indicated that
a 10 percent decrease in oxidant levels was related to a 0.5 percent decrease
in urgent visits. The single positive and statistically significant 1974 result
pertaining to pediatric visits indicated that a 10 percent decrease in oxidant
levels was related to a 0.9 percent decrease in unscheduled pediatric visits.
These somewhat mixed findings warranted further investigation.
Thus, we performed statistical analyses in an effort to uncover lag or
episodic effects of the air pollution levels on the health care utilization.
No such effects were found. In this connection it should be noted that the
oxidant readings in these data were sufficiently high to trigger six air
pollution alerts in 1973 and one alert in 1974 by the Metropolitan Washington
Council of Governments (COG) and were among the highest ever recorded in the
Metropolitan Washington Area.
-------
Next, we investigated the effects of three other air pollutants
related primarily to mobile sources (non-methane hydrocarbons, nitrogen
dioxide, and carbon monoxide) and one air pollutant primarily related to
stationary sources (sulfur dioxide).
We did not uncover any consistently signficant relationship
between unscheduled utilization in any department and the levels of
non-methane hydrocarbons or the levels of nitrogen dioxide during 1973.
Unfortunately, 1974 data on these two air pollutants were not satisfactory
to permit a replication of these analyses.
Our empirical results involving carbon monoxide did suggest an
association between unscheduled utilization of the ophthalmology depart-
ment and, to a lesser extent, unscheduled utilization of the pediatric
department during 1973. However, each result held for only one of the
two monitoring stations providing air pollution data. Furthermore, the
significant and positive associations between carbon monoxide levels and
department utilization were not found using 1974 data.
The last air pollutant we examined was sulfur dioxide. Two
monitoring stations provided sufficient sulfur dioxide data in 1973 to
permit separate analyses. The findings from these analyses indicated
two positive and statistically significant associations: one between
sulfur dioxide levels and unscheduled visits to the internal medicine
department, and the other between sulfur dioxide levels and unscheduled
visits to the ophthalmology department. Neither association was signifi-
cant when air pollution data from the other monitoring station were
substituted. Furthermore, the associations failed to hold when 1974 data
were analyzed.
Table 2.1 presents a summary of the positive and significant
associations between unscheduled department visits and levels of individual
air pollutants. Specifically, it displays the estimated percentage
reductions in unscheduled visits associated with 10 percent reductions in
the relevant air pollution measure.
Since there have been a number of studies suggesting the importance
of interactions between air pollutants and their combined effects on
health, we examined the possibility of synergistic effects. In
particular, we concentrated on synergistic effects of oxidants in combi-
nation with nitrogen dioxide, sulfur dioxide, and carbon monoxide. No
evidence of interactive effects was found.
We then examined the effects of two additional weather variables, a
measure of precipitation and a measure of temperature change. Predominant
interest lay in the effects that including these variables had on the magnitude
and significance of the coefficients associated with the air pollution
variables. Secondary interest was focused on the influence of the specific
weather variables on department utilization. In most cases the inclusion
and substitution of particular weather variables did not greatly affect the
magnitude and significance of the coefficicnts of the air pollution variables.
-------
TABLE 2.1. POSITIVE AND SIGNIFICANT ASSOCIATIONS BETWEEN
UNSCHEDULED VISITS AND AIR POLLUTION LEVELS a/
Air follutant^/
Monitoring
Station (year)
Photochemical
Oxidants
CAMP 1973
CAMP 1974
Civ. Park 1974
Carbon Monoxide
CAMP 1973
D.C. Hosp. 1973
Sulfur Dioxide
D.C. Hosp. 1973
ACS 1973
Unscheduled Visits to Departments j
Pediatrics
0.97.*
0.67o*
Internal
Medicine
1 . 57.*
Ophthalmology
1 . 17.*
4 . 37.****
1.5%****
0 . 47o*
Urgent Visit
Clinic
0.57.*
a/
— Numbers represent percentage reductions in unscheduled visits associated
with a 10 percent reduction in the relevant air pollution measure.
— All air pollution measures are based on maximum one-hour averages.
*Significant at the 10 percent level.
**Signifleant at the 5 percent level.
***Signifleant at the 2 percent level.
****Signifleant at the 1 percent level.
-------
The only notable exception occurred when the variable representing
temperature change was substituted for the average temperature variable.
In this case, the air pollution coefficient displayed mixed effects. This
was primarily attributed to the relatively high correlation of average
temperature with certain air pollution variables.
In general, the weather variables did not exhibit important
relationships with unscheduled department utilization. However, in
several cases, temperature change (as measured by the difference between
the maximum daily temperature and the minimum daily temperature) was
positively and significantly related to the utilization data. This was
especially true for the pediatric visits.
The next analysis we undertook was a comparison of the results just
discussed with those based on an analysis of Metro Transit Employees
(Metro). Because of small sample sizes, we examined total department
visits rather than only unscheduled department visits. In general, the
results based on the Metro sample did not display consistently significant
associations between visits to the clinic departments and air pollution
levels. The only exception was the association exhibited between visits
by the Metro employees to the ophthalmology department and photochemical
oxidant levels during 1974. We concluded that the findings for the Metro
sample were not at great variance with the previous results based on
unscheduled visitation by the total sample. However, given the limited
data and the difficulties associated with analyzing small samples, we
cautioned against overinterpretation of this conclusion.
Finally, we examined the findings pertaining to the data from a
smaller GHA facility located in Takoma Park. In general, the associations
involving the Takoma Park data were weaker than the associations based on
the data from the main facility. Several possible explanations for this
were presented. Since Takoma Park had no ophthalmology department and
since the sample sizes at Takoma Park were considerably smaller than those
pertaining to the Pennsylvania Avenue clinic, we do not feel that these
results negated our previous findings.
-------
SECTION III
INVESTIGATING THE AIR POLLUTION-HEALTH RELATIONSHIP
INTRODUCTION
For more than half a century, scientists have been accumulating evidence
that associates ill health with air pollution. One hypothesized relationship
between air pollution and human health involves long-term exposure to low
levels of air pollution. It is hypothesized that this chronic long-term ex-
posure exacerbates existing disease or increases the susceptibility to disease.
A second hypothesized relationship is more subtle and involves an acute re-
sponse in which high concentrations of air pollutants have an immediate effect
on health.
In a classical framework, testing these hypothesized relationships would
be based on the assumption that the functional specifications were given and
that the relevant variables were known. Statistical theory would then offer
procedures for testing competing hypotheses. However, in examining the asso-
ciations between air pollution and health, the functional forms of the rela-
tions are not known and there are only conjectures concerning the factors that
should be included. This is especially true with regard to the health effects
of mobile-source pollution where even the qualitative relationship is quite
uncertain. Finally, it should be stressed that for policy purposes it is not
sufficient to know only qualitatively whether air pollution is associated
with ill health; it is essential to quantify the air pollution-health rela-
tionship.
A basic difficulty in investigating the air pollution-health relationship
is isolating the effects of air pollution from the effects of numerous other
factors that influence health status. These include physical, socioeconomic,
and personal characteristics such as age, sex, race, income, smoking habits,
exercise habits, genetic history, nutritional history, and medical care as well
as other environmental factors such as climate. In order to estimate the
effect of any one of these factors on health the others must be held constant
experimentally or controlled statistically. !_/
An ideal investigation of the association between air pollution and
health would concrol for all of the above factors. Unfortunately, many of
these factors are difficult to measure conceptually (e.g., genetic history),
while others are poorly measured in existing statistics (e.g., medical care).
Since we do not even know all the relevant factors, the practical difficulty
is to control for as many factors as possible, either experimentally or sta-
tistically, explicitly or implicitly.
An additional problem surrounds the fact that there is a lack of data on
air pollution doses and dose rates. Available air pollution data are usually
in the form of air pollution level readings at a specific point in a geographi-
_!/ Only in the remote case in which a particular factor was uncorrelated with
all the others could one uncover the "true" effect of that factor using a
univariate approach.
7
-------
cal area for a given time period. However, local topography, location and
height of buildings, weather conditions, and location of emission sources,
often lead to considerable differences in the actual ambient air quality
across an area. Consequently, the air pollution measures are, at best, ap-
proximations to the air pollution doses individuals receive. Thus, the data
on the variable of primary interest are crude and any estimated air pollution-
health relationship is subject to additional uncertainty.
LABORATORY AND TOXICOLOGICAL EXPERIMENTATION
Laboratory experimentation on humans can sometimes provide useful infor-
mation. For example, short-term fumigation experiments may be helpful in
uncovering the physiological mechanisms by which air pollutants affect humans.
However, in many instances it is not practical to perform such experiments.
If extremely high levels of air pollution are hypothesized to harm people, it
would not be socially acceptable to expose subjects to these levels in order
to measure the physical effects. More importantly, since air pollution con-
centrations of the magnitude required to demonstrate these effects in the
laboratory are seldom experienced in urban air, it is questionable whether
the results of such experiments are valid in assessing the effects of air pol-
lution exposure on the general population. Finally, if the hypothesized
effect is small, it may not be feasible to test for it in the laboratory (e.g.,
a slight increase in the mortality rate of subjects with life expectancies of
70 years could require hundreds of thousands of subject-years to ascertain).
Since experimentation on humans may not be practical, laboratory experi-
ments with animals may be useful in obtaining important knowledge of pollution
effects. However, differences in physiology, life span, and dose rate make it
difficult to extrapolate results from animal studies to effects on humans.
Furthermore, policymakers are more interested in determining the extent to
which a pollutant increases the frequency of a disease in the general popula-
tion than whether high concentrations of the pollutant induce the disease in
white mice under controlled laboratory conditions. Thus, epidemiological
studies which examine humans in their natural setting are more relevant than
are laboratory experiments. 2/
EPIDEMIOLOGICAL APPROACHES
In reviewing these epidemiological studies, one is usually faced with two
types of investigations. The first type represents classical epidemiology and
attempts to compare groups that differ only in their exposure to air pollution.
The ultimate example would be a study that looked at the incidence of disease
among sets of identical twins, one of each set living in an area of low pollu-
tion and the other living in an area of high pollution. ^/
The second type of investigation examines the incidence of disease in
large, geographically-defined groups. ^/ Such large groups permit controls
2/ Studies of highly susceptible populations may also be useful in identifying
the potential increment in pollution-related illness and death.
3/ See, for example, Cederlo'f [9].
4/ See Hammond and Horn [21].
-------
for other important factors, but preclude the detailed comparisons that are
possible with more classical methods. Both types of investigations aid in
exploring the air pollution-health association.
The extensive literature that has evolved from investigations of the air
pollution-health relationship covers the gamut of epidemiological research.
One of the earliest and most common methods of analysis involved cross tabu-
lations or simple correlations. For example, the prevalence of a disease
(or even death) in specified population groups was correlated with some index
of air pollution. j>/ The difficulty in using the results of such studies is
that a host of other factors were allowed to vary across populations and it
is impossible to identify the "pure" pollution effects.
Recognizing the need to control for other important factors affecting
health, some investigations cross tabulate along several dimensions (e.g.,
age, race, income), while others compute partial correlations. b_f However,
even these more sophisticated statistical techniques will not produce reliable
estimates of the air pollution-health relationship if the groups being com-
pared are not well matched. For instance, if a study compares urban with
rural groups, the multitude of factors that differ between urban and rural
environments (in addition to pollution levels) cannot be controlled completely
by statistical methods. One is inevitably left with a number of important
uncontrolled factors known to vary systematically with urbanization.
An improvement in methodology is represented by community studies. _7/
Utilizing air pollution measurements across areas of a city, such investiga-
tions contrast measures of health status within those areas, hypothesizing
that many relevant factors are constant or vary randomly within communities.
However, such studies have problems associated with small sample sizes and
the consequent large sampling variation, _8/ In addition, systematic relation-
ships between such factors as low income and exposure to high air pollution
levels often confound the observed associations.
Thus, two major shortcomings with attempts to investigate the association
between air pollution and health have been (1) failure to control for the
numerous factors that influence health status and (2) sample sizes too small
to lend confidence to the results.
MULTIVARIATE ANALYSES _9/
To a large extent multivariate analyses of large populations can overcome
many of the difficulties in estimating the air pollution-health relationship.
5/ See Stocks [37].
1>/ See, for example, Daly [ 131.
]_/ See, for example, the Nashville studies undertaken by Zeidberg and his co-
workers [46-50] or the Buffalo studies done by Winkelstein and his
colleagues [40-45] listed in the bibliography.
8/ See the discussion on this topic in Lave and Seskin [28].
9/ Specific multivariate techniques are discussed in Section VII.
-------
Since these techniques can be used to control statistically for a number of
confounding factors simultaneously, they can reduce the likelihood that such
factors will obscure or bias the dose-effect estimates. In addition, the use
of these methods lessens the chances that an estimate will reflect a spurious
association.
For example, assume that a specific air pollutant that was not damaging
to plants could only be formed in the presence of certain meteorological con-
ditions. Further, assume that these weather conditions were damaging to
plants independent of pollution levels. Then if an analysis was undertaken
relating only levels of this air pollutant to plant damage across areas, it
would probably uncover a significant positive correlation. That is, the
analyst might erroneously conclude that this air pollutant caused plant
damage. The difficulty arises because the "true" cause of both the air pollu-
tion and its hypothesized effect (plant damage), i.e., the meteorological con-
ditions, were not controlled in the analysis. Thus, in general, to avoid or
minimize this type of spurious association, confounding factors must be ex-
plicitly taken into account. 10/
Two basic types of multivariate analyses can be used to explore the air
pollution-health association. The first, cross-section analysis, investigates
a measure of health status (e.g., mortality rates) across areas having differ-
ent air pollution levels, while controlling for other area differences (e.g.,
socioeconomic characteristics), ll/ The second type, time-series analysis,
examines a health measure within a single location experiencing changing air
quality over a period of time. 12/ Thus, one advantage of the time-series ap-
proach is that many of the factors (excluding air pollution) which lead the
mortality rate to be higher in one area than in another area, should be rela-
tively constant over time within an area.
In comparing the two approaches, it should be noted that any estimated
relationship between air pollution and health is likely to be different in a
time-series study than in a cross-section study. For example, analyzing daily
deaths in conjunction with daily air pollution levels should primarily un-
cover very short-term effects of air pollution (if they exist) as opposed to
long-term effects one would expect to observe in a comparison of annual death
rates across cities.
RESULTS OF A SIMILAR STUDY
Jaksch and Stoevener [25 ] attempted to quantify in monetary terms the
effect of air pollution on the utilization of outpatient medical services in
the Portland, Oregon area. The study focused on the impact of suspended par-
ticulate pollution on the consumption of outpatient medical services per dis-
ease episode. To isolate this effect, meteorological conditions in the area
as well as socioeconomic-demographic characteristics of the patients were con-
trolled. Health data were taken from a five percent ongoing random sample of
the membership of the Kaiser Foundation Health Plan, a prepaid group medical
plan.
10/ For an excellent discussion of spurious correlation, see Simon [35].
ll/ See, for example, Lave and Seskin [29],
12/ See Glasser and Greenburg ("16").
10
-------
The method of analysis involved the application of multivariate regres-
sion analysis to the data using models with both unlagged and lagged air
pollution and weather variables. The statistical results indicated signifi-
cant effects of air pollution (as measured by suspended particulates) on
medical services used to treat respiratory diseases, but not on medical
services used to treat circulatory-respiratory diseases.
In addition, each medical procedure, treatment, and clinic visit was
assigned a dollar value according to the California Relative Value System.
This scaling system was used by Kaiser to quantify the medical services per-
formed. As an economic cost, the researchers concluded that air pollution
had a minimal effect on increasing the quantity of outpatient medical ser-
vices consumed per contact with the medical system.
11
-------
SECTION IV
A DESCRIPTION OF THE STUDY AREA AND THE GROUP HEALTH ASSOCIATION
THE STUDY AREA I/
The area of study can be roughly described as the section of the Washing-
ton Standard Metropolitan Statistical Area encompassed by U.S. Interstate 495
(the Beltway). Shown in Figure 4A, this 164,753 square acre area includes
the District of Columbia, Arlington County, parts of Fairfax County, and the
city of Alexandria, Virginia as well as portions of Montgomery and Prince
Georges Counties in Maryland. With a population of 1,707,668, the study area
accounts for approximately 62 percent of the entire Washington SMSA population
of 2.9 million.
Metropolitan Washington is situated on the western edge of the Atlantic
coastal plain, 35 miles west of the Chesapeake Bay. The Blue Ridge Mountains
rise to 3,000 feet about 50 miles to the west with an orographic effect of
warming and drying westerly winds. The coastal plain to the east of the study
area is essentially flat. Although no topographical barriers exist between
the study area, the Chesapeake Bay, and the Atlantic Ocean, the Washington
Metropolitan Area is too far inland to be affected by summer sea breezes.
The District of Columbia, the largest city in the study area, (population
767,000) is located at the head of navigation of the Potomac River near its
confluence with the Anacostia River. The terrain of the city itself ranges
from sea level to slightly over 400 feet. Bluffs along the Potomac River and
Rock Creek, and to the southeast and east of the Anacostia River suggest some
channeling of the airflow, but, generally, the terrain does not impede the
free movement of air about the city.
The Washington area climate has the seasonal and daily variations charac-
teristic of the eastern seaboard, with moderate winters and frequent intervals
of high humidity and oppressive heat in the summer. During the summer, high
temperatures average in the upper eighties. The average annual precipitation
is about 40 inches, with no pronounced wet or dry season. Annual snowfall
averages about 20 inches.
The frequent movement of cold arctic air masses into the Middle Atlantic
States from Canada results in a prevailing northwest flow of unstable air in
Washington from late November through April. The unstableness and relatively
high wind speeds attending these air masses result in good atmospheric dilution
conditions. In general, the winter and spring months constitute the period of
most frequent unstable weather in Washington (i.e., storminess and high winds)
and thus result in good atmospheric dilution and good ventilation in the lower
atmosphere.
The summer and fall months are characterized by a reduction of wind speed
and the prevalance of southerly winds. These months also have a higher fre-
JY This information was taken from [39].
12
-------
FIGURE 4A. THE WASHINGTON METROPOLITAN AREA
METROPOLITAN WASHINGTON
Source; \10]
-------
quency of cloudiness and light wind conditions during nocturnal hours per-
mitting a higher frequency of radiational surface-based inversions to form.
Consequently, the summer and fall seasons constitute the period of lowest
ventilation and highest potential for atmospheric stagnation in the Washing-
ton area.
THE GROUP HEALTH ASSOCIATION
The Group Health Association (GHA) is the largest and oldest prepaid
group practice medical care plan for persons living in the Washington, D.C.
Metropolitan Area. 2/ As of December 31, 1973, GHA had approximately 90,000
members. Enrollment by residential location was as follows; District of
Columbia - 50%, Maryland - 35%, and Virginia - 15%. Until the end of 1973,
GHA operated from four health center locations: The Downtown Center (hence-
forth referred to as the Pennsylvania Avenue Center), the Labor-Management
Health Center (henceforth referred to as the Takoma Park Center), and two
smaller centers in Annandale, Virginia and in Northwest Washington. _3/ A
fifth center was opened in Rockville, Maryland at the beginning of 1974.
The membership of GHA is comprised of basically three types of enrollees:
Federal, metro, and general. The largest group (about 757,) consists of Federal
government employees. The next largest group (about 157«) is made up of employ-
ees of the Washington metropolitan transit system (Metro). The general member-
ship (about 107») includes college students, medicaid participants, and other
miscellaneous enrollees. Figures 4b and 4c and 4d and 4e indicate the GHA en-
rollment distribution by sex and major age groups, respectively, for 1973 and
1974. Figure 4f indicates the GHA enrollment by 5-year age groups for 1973. 4/
Members of GHA pay a monthly premium which in turn provides a broad range
of medical care. The services that are outpatient in nature are provided in
GHA health care centers. Inpatient services are provided in neighboring hos-
pitals. The services provided by GHA include physician visits in the health
centers or in the hospital, hospitalization, surgery, laboratory tests and
X-rays, emergency service, mental health, and routine physical examinations,
among other medical services. In 1973, the utilization of GHA facilities had
the following characteristics: 0.97 injections per enrollee, 1.32 X-ray films
per enrollee, 9.92 laboratory tests per enrollee, and 3.25 physician visits per
enrollee. 57 In 1974, the figures were: 0.89, 1.38, 9.60, and 3.29, respective-
ly.
A number of factors can affect the utilization of medical services at the
facilities, both from the supply side and the demand side. For example, supply,
or capacity to deliver, influences utilization rates. Physicians usually join
the GHA staff in July. Hence, in some instances backlogs may develop in areas
of elective treatment and this, in turn, can bias utilization upward during the
2/ As with other group practices, physicians are organized into a medical group
or partnership, from which they receive their salaries.
3/ The Northwest Washington center closed August 15, 1975.
4_/ These data were not available for 1974.
57 This does not include visits with referral for physicians or retainer physi-
cians; nor does it include (1) hospital calls, of which there were 35,853 for
1973, estimated on a minimal basis of one call for each hospital day, or (2)
home calls of which there were 619 in 1973.
14
-------
TOTAL
ENROLLMENT
4S6 _jj.
FEDERAL
GROUP
48.8
r-ill
ilpi:y:i:&i$"
GENERAL
GROUP
55.1
-••••: •:-;•;•
mm
^.:x--,-. -H
--"" '•','.'.'
METRO
GROUP
45.9
Figure 4b. GHA enrollment by sex (December 31, 1973)
Source: [141-
TOTAL
ENROLLMENT
31.4
•'.1.6
FEDERAL
GROUP
IS. 6
.......
GENERAL
GROUP
34.4
45.6
METRO
GROUP
S3 6
46.4
Figure 4c. GHA enrollment by sex (December 31, 1974),
Males Source fl5l.
Females
Note: Scales used in these figures differ.
15
-------
O- 19 years
20-64 years
65 and over
TOTAL
ENROLLMENT
3V 3
4.J
FEDERAL
GROUP
573
GENERAL
GROUP
597
7.2
METRO
GROUP
357
8.1
CD
Figure 4d. CHA enrollment by major age groups (December 31, 1973).
Source: [14 ].
TOTAL
ENROLLMENT
0-17 years
18-64 years
'' ^5 and over
4.3
FEDERAL
CROUP
64.1
GENERAL
GROUP
METRO
GUOUP
Figure 4e. GHA enrollment by major age groups (December 31, 1974)
Source: [15 ].
Note: Scales used in these figures differ.
16
-------
65 + •
60-64-
b5-59
50-54
40-44
35-39
30-34
2S-23
?0-24
15-
/:
XI
5- 9 —
0- 4 —
Figure Af. GHA enrollment by five-year ape groups (December 31, 1973).
Source: [15].
summer months when the staff has increased. _6/ In addition, there is some
movement of patients between facilities. For instance, if staffing is low in
a particular department at a specific center, patients can be shifted from one
clinic to another. A more extreme situation took place when the new GHA fa-
cility opened in Rockville, Maryland. Patients previously utilizing the
center in Takoma Park now had the option of using a new facility that was in
some cases more convenient.
On the demand side, it should be noted that the Federal "open season,"
i.e., the period in which government employees can join GHA, begins in November.
Consequently, since the majority of GHA members are civil service workers, this
means that utilization can increase substantially during the winter months as
new enrollees join the plan. Both the supply and demand factors will be taken
into account in the subsequent analyses.
Access to the GHA data for this study was indeed fortunate. Analysis in
the air pollution-health area has been hampered by the lack of basic data. For
example, very few measures of a population's health status are available. In
fact, the only information that is available on a comprehensive basis consists
of mortality rates. Since many of the health effects of air pollution (par-
ticularly mobile-sou'ce pollution) are likely to manifest themselves in less
severe conditions, the scarcity of other health measures has presented a serious
problem to the analyst and, in turn, to the policymaker. The extensive morbidity
data from GHA permit significant questions to be answered on the complex issue
of the health effects of mobile-source air pollution.
6/ This phenomenon is not likely to be an important factor in our analysis
since we will be focusing primarily on "urgent" or unscheduled non-elective
cases.
17
-------
At the same time, two additional caveats should be mentioned regarding
the use of these data. The first is that the socio-demographic character-
istics, such as the racial composition and income distribution, represented
by the GHA membership may not be typical of the socio-demographic character-
istics of the entire metropolitan Washington population. The second, and
related point, is that the utilization of medical services by GHA members
may also be different than the utilization of medical services by the general
Washington population.
18
-------
SECTION V
COMPILING THE MEDICAL DATA
IMPATIENT VS. OUTPATIENT DATA
As noted previously, there are basically two types of medical services
offered by GHA: outpatient care at health centers and inpatient care in
local hospitals. In this study only outpatient data will be examined. The
chief reason for this lies in the fact that the subtle health effects of
exposure to air pollution (especially mobile-source air pollution) are not
likely to manifest themselves in a time-series analysis of inpatient data
such as hospital admissions. Exceptions to this might take place if either
the sample were restricted to extremely vulnerable individuals (e.g., heart
patients) or if the sample were limited to emergency room admissions. Neither
of these limitations was feasible for this analysis.
One caveat that should be mentioned with regard to the use of outpatient
data is that only within-plan utilization is accounted for. If a patient re-
ceived medical services outside the GHA system, such services would not be
captured in the data. By their very nature, inpatient data are much less sub-
ject to this problem. Nevertheless, despite this disadvantage, outpatient
data would seem to be far more relevant than inpatient data in examining the
health effects of air pollution on a daily basis.
THE SELECTED OUTPATIENT DATA
Virtually all registration information recording visits to nine depart-
ments at the GHA Pennsylvania Avenue Health Center and at the Takoma Park
Health Center during 1973 and 1974 was collected for analysis. _!/
The departments were;
1. Urgent Visit Clinic (both centers)
2. Internal Medicine (both centers)
3. Pediatrics (both centers)
4. Optometry (both centers)
5. Ophthalmology (Pa. Ave. only)
The reasons for selecting these departments are discussed in the next section.
For each visit, both patient information and service information were
available. Briefly, the patient information consisted of six items; the
policy group to which the patient belonged, the type of membership plan, the
membership status of the patient, restrictions on policy coverage of the
patient, the patient's sex, and the patient's age.
_!/ The study was restricted to these two centers since they account for approxi-
mately 84 percent of all visits made to GHA facilities in the Washington
area.
19
-------
In addition to the department and the date in which the service was
rendered, the other service information included the doctor number and the
type of visit. The type of visit will be of particular importance to this
analysis. ^/ The six possible types of visits are: urgent visit, regular or
scheduled visit, general physical examination, well baby care, unscheduled
appointment, and OB/GYN (obstetrics/gynecology).
For our purposes the last category, OB/GYN, is not relevant since we did
not collect utilization data pertaining to obstetrical or gynecological visits.
The urgent visit classification applies only to visits at the Urgent Visit
Clinics. Regular or scheduled appointments can take place at any of the re-
maining four departments. General physical examinations are given in both
the internal medicine and pediatric departments. Well baby care only applied
to the pediatric department and involves routine checkups of infants following
their birth until the age of three or four. As with regular or scheduled ap-
pointments, unscheduled appointments can take place in all departments except
the Urgent Visit Clinic where such visits are termed "urgent visits."
THE SELECTION OF DEPARTMENTS
The departments listed above were chosen for several reasons. Since
July 1, 1973, the Urgent Visit Clinic (UVC) at the Pennsylvania Avenue Center
operated 24 hours a day seven days a week. Prior to that time, it operated
six days a week from 9:00 am to 10:00 pm. The UVC at Takoma Park operates
9:00 am to 4:30 pm Mondays through Fridays and 9:00 am to 11:00 am on Satur-
day. These two clinics operate on a basis whereby GHA members can walk in off
the street and seek immediate treatment for ailments without prior appointments.
At the discretion of the UVC staff, members are either treated immediately or
referred to more specialized departments. As a general rule visits are offi-
cially recorded in the department in which a member receives treatment. Thus,
UVC "walk-ins" referred to other departments would be recorded as unscheduled
visits in those departments. In 1973 the two Urgent Visit Clinics saw ap-
proximately 45,000 and 15,000 patients, respectively (see Tables 5.1 and 5.2).3_/
Since one might hypothesize that the type of health effects from air pollution
would cause patients to utilize such facilities, and since other investigators
have found that similar clinic visits have been correlated to air pollution
episodes, 4/ data from these clinics were thought to be of particular interest.
The internal medicine departments at both centers operate between the
hours of 9:00 am and 5:00 pm five days a week. ^/ In 1973 they saw about
45,000 and 10,500 patients, respectively (see Tables 5.1 and 5.2). The three
2/ We did not feel that the attending physician was a relevant factor for
analyzing the occurrence of unscheduled visits.
_3/ Tables 5.1 and 5.2 present a breakdown of the department visitation data for
1973 by type of visit for the Pennsylvania Avenue Center and the Takoma Park
center, respectively. Tables 5.3 and 5.4 present a similar breakdown for
the 1974 data.
4_/ See Greenburg e^t al. [18].
5_/ In addition, the internal medicine department at the Pennsylvania Avenue
Center sees patients between 9:00 am and 12:00 noon on Saturdays and the
internal medicine department at Takoma Park sees patients between 9:00 am
and 11:00 am on Saturdays.
20
-------
TABLE 5.1. 1973 GHA DATA (PENNSYLVANIA AVENUE)
Type of
Visit
Urgent
Visit
Regular
Visit
General
Physical
Well Baby
Care
Unscheduled
Appointment
Total
Department
Optometry
0
14,668
(97.2%)
0
0
427
(3.8%)
15,095
(10.5%)
Internal
Medicine
0
29,925
(66.6%)
10,352
(23.0°X)
0
4,682
(10.4%)
44,959
(31.2%)
Pediatrics
0
1,873
(5.9%)
6,924
(21.8%)
9,814
(31 .0%)
1 3,078
(41.3%)
31,689
(22.0%)
Ophthal-
mology
0
5,227
(70.2%)
0
0
2,221
(29.8%)
7,448
(5.2%)
Urgent Visit
Clinic
44,795
(100.0%)
0
0
0
0
44,795
(31.1%)
Total
44,795
(31.1%)
51 ,693
(35.9%)
17,276
(12.0%)
9,814
(6.8%)
20,408
(14.2%)
143,986
(100.0%)
Note; Except for the last row of numbers, the percentage figures in parentheses all sum by
column (department) to represent the composition of department visits by type of visit.
The last row sums across to represent the composition of total visits by department.
-------
TABLE 5.2. 1973 GHA DATA (TAKOMA PARK)
Type of
Visit
Urgent
Visit
Regular
Visit
General
Physical
Well Baby
Care
Unscheduled
Appointment
Total
Department
Op tome try
0
A, 503
(99.4%)
0
0
29
(0.62)
4,532
(9.7%)
Internal
Medicine
0
7,337
(70.3%)
2,871
(27.4%)
0
235
(2.3%)
10,443
(22.3%)
Pediatrics
0
704
(4.3%)
3,012
(18.3%)
3,412
(20.8%)
9,297
(56.6%)
16,425
(35.1%)
Ophthal-
mology
0
0
0
0
0
0
Urgent Visit
Clinic
15,398
(100%)
0
0
0
0
15,398
(32.9%)
Total
15,398
(100%)
12,544
(26.8%)
5,883
(12.6%)
3,412
(7.3%)
9,561
(20.4%)
46,798
(100.0%)
Note: Except for the last row of numbers, the percentage figures in parentheses all sum by
column (department) to represent the composition of department visits by type of visit.
The last row sums across to represent the composition of total visits by department.
-------
TABLE 5.3. 1974 GHA DATA (PENNSYLVANIA AVENUE)
Type of
Visit
Urgent
Visit
Regular
Visit
General
Physical
Well Baby
Care
Unscheduled
Appointment
Total
Department
Op tome try
0
14,436
(92.8%)
0
0
1,106
(7.2%)
15,542
(10.3%)
Internal
Medicine
0
33,628
(77.6%)
6,840
(15.8%)
0
2,849
(6.6%)
43,317
(28.7%)
Pediatrics
0
2,008
(6.470
6,81.6
(21.9%)
6,442
(20.7%)
15,883
(51.0%)
31,149
(20.6%)
Ophthal-
mology
0
5,544
(73.6%)
0
0
1,99.3
(26.4%)
7,537
(5.0?0
Urgent Visit
Clinic
53,439
(100.0%)
0
0
0
0
53,439
(35.4/0
Total
53,439
(35.4%)
55,616
(36.8%)
13,656
(9.0%)
6,442
(4.3%)
2] ,831
(14.5%)
150,984
(100.0%)
Note: Except for the last row of numbers, the percentage figures in parentheses all sum by
column (department) to represent the composition of department visits by type of visit.
The last row sums across to represent the composition of total visits by department.
-------
TABLE 5.4. 1974 GHA DATA (TAKOMA PARK)
Type of
Visit
Urgent
Visit
Regular
Visit
General
Physical
Well Baby
Care
Unscheduled
Appointment
Total
Department
Optometry
0
3,768
(99.0%)
0
0
37
(1.0%)
3,805
(9.0%)
Internal
Medicine
0
8,096
(73.2%)
2,464
(22. W)
0
503
(4.5%)
11,063
(26.1%)
Pediatrics
0
1,056
(7.7%)
2,183
(15.8%)
1,823
(13.2%)
8,714
(63.3%)
13,776
(32.4%)
Ophthal-
mology
0
0
0
0
0
0
Urgent Visit
Clinic
13,793
(100.0%)
0
0
0
0
13,793
(32.5%)
Total
13,793
(32.5%)
12,920
(30.4%)
4,647
(11.0%)
1,823
(4.3%)
9,254
(21.8%)
42,437
(100.0%)
Note: Except for the last row of numbers, the percentage figures in parentheses all sum by
column (department) to represent the composition of department visits by type of visit.
The last row sums across to represent the composition of total visits by department.
-------
types of visits to the internal medicine departments are: regular scheduled
appointments, general physical, and unscheduled appointments. Although a rela-
tively small percentage of the visits to the internal medical departments are
unscheduled appointments, there is a procedure whereby patients may call the
physicians and if deemed- necessary an appointment can be scheduled within
a day or two. Thus, even though the appointment will be recorded as being
previously scheduled, it may be air pollution-related. j>/ In addition, since
studies have linked respiratory illness and cardiac distress to air pollu-
tion, _?/ the internal medicine department which sees patients with these con-
ditions seemed quite relevant for our analysis.
The pediatric departments at both the Pennsylvania Avenue and Takoma Park
facilities follow approximately the same schedule as the internal medicine
departments. 8/ In 1973 they saw about 31,500 and 16,500 children, respective-
ly. As shown in the tables, there are basically four types of visits that
occur in the pediatric departments: regular scheduled, general physicals,
well baby care, and unscheduled appointments. Primary attention will be
focused on the unscheduled visits; however, in general, the visitation data
pertaining to small children are of interest because there is evidence in the
literature that the very young as well as the very old are especially suscep-
tible to health effects from air pollution. _9/
The final departments from which we collected data were the optometry
and ophthalmology departments. As noted above, Takoma Park has no ophthal-
mology department. In addition, their optometry department only operates
during the hours 9:00 am to 1:00 pm and 2:00 pm to 5:00 pm Monday through
Friday and 9:00 am to 1:00 pm on Saturday. At the main Pennsylvania Avenue
facility, both departments operate five days a week between 8:40 am and 5:00 pm
and on Saturday between 8:30 am and 12:30 pm. (See Tables 5.1 to 5.4 for a
breakdown of these departments by type of visit.) These departments see indi-
viduals with eye problems and since eye discomfort is one of the primary symp-
toms that has been related to mobile-source pollution 10/, it was felt that
examining the frequency of visits to these departments would be particularly
relevant for the study, ll/
PREPARING THE GHA DATA FOR USE
The visitation data were received from GHA on a magnetic tape. The tape
contained information representing 190,784 visits for 1973 and 195,903 visits
6/ These visits are called "write-ins." Unfortunately, our data do not permit
identification of them.
]_f See Cohen et al. [11].
jj/ The only difference is that the pediatric department at Pennsylvania Avenue
is open until 1:00 pm on Saturdays.
£/ See for example the study that was undertaken by Pearlman et al. [33 1. It
investigated the relationship between illness of the lower respiratory tract
and nitrogen dioxide exposure in school children and infants.
10/ See for example Hammer et al. [20].
ll/ After the fact, it was learned that the optometry departments almost exclu-
sively saw patients with problems relating to vision correction; hence,
these data were not utilized in the subsequent analyses.
25
-------
for 1974. Since interest lay in the day-to-day relationship between GHA uti-
lization and both daily air pollution and daily weather conditions, it was
necessary to aggregate these data on a daily basis.
Specifically, for each day of the year at each of the two health centers
in each of the nine departments, the following items were calculated and a
new tape containing the information was created;
1. The number of patient visits affiliated with each of the five
policy groups.
2. The number of patient visits in each of the nine membership
p1an s.
3. The number of patients having each of the five possible member-
ship statuses.
4. The number of department visits of patients with restricted
policy coverage.
5. The number of department visits by male patients and the number
of department visits by female patients.
6. The number of department visits of patients in each of five age
categories: under 1 year, 1 to 14 years, 15 to 44 years, 45 to
64 years, and 65 years and older. 12/
7. The number of clinic visits in each of the five types discussed
above. 13/
12/ These age classifications were chosen because they correspond to the age
breakdown used by Vital Statistics. In addition, there is reason to believe
that air pollution could exhibit differential health effects across these
age groups. See Lave and Seskin [28 ].
13/ This information was not all used in this analysis. The means and standard
deviations of the specific health utilization variables used in the analysis
appear in Appendices A and B.
26
-------
SECTION VI
COMPILING THE AIR POLLUTION AND METEOROLOGICAL DATA
THE AIR POLLUTION DATA
The air pollution data used in this research came from the Government of
the District of Columbia, Department of Environmental Services. The specific
ways in which the raw data were transformed for use in the analysis will be
discussed below.
Metropolitan Washington's primary air pollution problem involves photo-
chemical oxidants. This air pollutant, which is often referred to as photo-
chemical smog, is a secondary air pollutant. It is not emitted directly into
the atmosphere by any source, but is created in the atmosphere by complex
chemical reactions triggered by sunlight acting on precursors--hydrocarbons
and nitrogen oxides. Figures 6a and 6b show the maximum one-hour readings in
relation to the primary and secondary national standard (0.08 ppm) across
several monitoring stations in the region for 1973 and 1974, respectively.
As can be seen, each station exhibited readings in excess of the standard. I/
Note, too, that in most cases the levels reported at these stations did not
differ greatly.
In addition to measurements of photochemical oxidants, readings were
also available for its two precursors, hydrocarbons and nitrogen oxides. With
regard to hydrocarbons, principal interest is focused on the levels of non-
methane hydrocarbons. 2/ Levels of non-methane hydrocarbons are determined
by measuring separately the total hydrocarbons and methane at & monitoring
station. 3/ In 1973, the primary and secondary standard for non-methane hydro-
carbons was exceeded on 165 days of the 225 days for which data were valid.
(Bar graphs were not available.)
Probably the most frequently mentioned oxide of nitrogen is nitrogen
dioxide (NO,,). Nitrogen dioxide is formed when fuels are burned at high
temperatures with air containing nitrogen. Automobiles and power plants are
_!/ It should be noted that the actual air pollution data used in the analysis
were not identical to those data represented in the graphs. The data repre-
sented in the graphs were obtained from publications of the Metropolitan
Washington Council of Governments and only depict maximum readings taken at
a number of stations throughout the region. For our statistical analyses,
we required more comprehensive data on a daily basis. As noted earlier,
these were obtained from the District of Columbia, Department of Environ-
mental Services.
2/ In fact, the national standards were formulated in terms of non-methane
hydrocarbons because they were thought to be more photochemically reactive.
2/ The readings of non-methane hydrocarbons are probably subject to more un-
certainty than the other air pollution readings because of difficulties
associated with the measurement of methane and the fact that two machines
are involved.
27
-------
,
0.20 -
0.10 -
0.16 -
0.14-
"
0.12-
cL o.io-
0.08 -
0,06 -
0.0'. -
0.02 -
i a a r
J C 8
1 H f. C
9 M D
i « r -
ft t a
i c n i
i n o
i a n r
inn
i a u i
u o n
i a ii '.
c • n
1331.
a Q n
i a o i
tr n n
•» n , i f
n ti o
i a n i
n n t?
i n n t
u it a
i o n t
•I!!;!
\
•J
• - — (
•* B « 3
n a ^ i
HOC1
i E r n
• 11 B 1
i a a c
fl n n •
H C B tl
n u n i
;:: :
7^77]
1C L i
' r; E I.
" ti ;•.
i U D f
«• r r i
• :i n f
• • r r' '
i' D ii '
: » o t
[' U !l
• 'I n PK
-M (t .1
• "t :: :
', 'JLZI.
— ii L '
i T i: •_
r u u i
: J U ;.
,• n f .
> :) n r
•: n u
MI a f ,
• T n i
1 ! 3 I'
N 1 1 fl •'
: ;i u 41
i!i!i;
\
—^
s
0
jd p u
.1 « o;
rs:»fr
T^/J14
3<= t_:
30 3-
aocc1
q ar i
'J IJ V-O
jja;a
3j- i«*i
C;;^
y J*1
Hl^
flHt^'
•vit'-l
rJ.Tfl-1
tin an
13 J.'l
HFjrJf]
J3 95
;o 'j --
-J J-.1 1)
H
>•
.:::::
V
__
°^
\ "
»DO€
Ooa^
Sour
ao.T7
0303
ajua
joa y
3 33 J
j^oa
ajju
jn'if5
Rl''n?
m T a
iiTPtl
'n* n. 3
_.MA:)
3^03
rfljj «
rl
ff.inj
07";;
ajfao
BBiC
uapo
U;-,nn
?^
;rri'"
J6
\
—
s
s
»jjor>
O.IOE
0393
o nrrr
nil n1-
n/7 (7^
!J OO 1-
nUnf
17 0 ^
O ono
D0«f
aooa
(J0D3
3 d DC
vsfin
OO-O^J
rji33 a
jrtTiC?
aODQ
OJflO
f? 93fc
•'.'^j'jp
3 rjofl
crtjca
J«UO
703 t
ocr&
l^-O'J
5iTDi
C'0i£
i • :
: : : •' :
>
S
—
s
"H,
IT 0 '1
) I/ID
•jOilll
3 nun
7,,'ja
MU"
XrAi
U^cr3
POJJ
ncB--i
(I a on
anaq
3onn
P n&-^
oQ-fio
annf
1 3 b ^
— A f £
c ^-^ **
1 aau
a4»a
.)UU3
'•^aua
034"
i njft
^ fl Bn
•i:::
S
~~
•^
?
6 Olj j
) T ^ ll
n.j^i
a fl T) '&
.. ; j fl u
jjo.;
J L? •» -J
fj U'j I.
n.i'3.:
rj n j a
o^aa
vt-iJ'.T-T
a aa tr
-jano
J3 3*J
Qotia
Jitijj
JD ag
OOJU
naau
lipjK'j
g CIUL
: i:
S
National
Standard
V
\
> **
MARYLAND
D.C.
VIRGINIA
Figure 6a, Concentrations of photochemical ox id ants
(maximum hour - 1973).
Source:
28
-------
0.22 -
0 .20 .
0.1B -
0.16 •
0.14 -
0.12 -
a 0-10 •
c.
0.08.
0.06 -
0.04 -
0.02 -
0
"
xj
'.'•'.•
.','
:'•:'
S\L
y
•I;
_v
v?
:•:•
|
^
-TT
. *
—
+ •*•-•
I
•—
—
-n-
'"
—
:'•:•
•*•'
!;!;
•:•:
!
1
~
|
!•*•
•:•
|
::-;
j-'-t.
T T '
$
'•_:'•
|
Itli
-
F * ^
J.
x|
x|
:•:;
:j:|
tp.^
'x.
r—
!y
:':'
j:j:
tV/
—
TTn
•:•
*.'
|
*^*-"
N
National StanJai !
Figure 6b. Concentrations of photochemical oxidants fmaximum hour •
Source: [2].
29
-------
the principal emission sources for this air pollutant. In 1973, the levels
of nitrogen dioxide obtained from the sampling stations in the Metropolitan
Washington Area were somewhat below the national primary and secondary stan-
dard (0.05 ppm).
Carbon monoxide is another air pollutant emitted by mobile-sources--
automobiles, trucks, and planes. This colorless, odorless, tasteless gas,
is caused by combustion of fuels with insufficient oxygen. Levels of carbon
monoxide across the metropolitan region have also been at or above the pri-
mary and secondary national standard (9 ppm maximum 8-hour average). Figures
6c and 6d show the relation of monitored levels to the standard at a number
of stations in 1973 and 1974, respectively.
Other air pollutants that would be of secondary interest in a study such
as this one include suspended particulate matter and sulfur dioxide. 4/ Par-
ticulates are fine pieces of dirt and dust that are so small they tend to
float--suspended in air. Man-made particulates are primarily emitted by
stationary sources (e.g., power plants, businesses, schools, residences,
and industrial operations) when fuels such as oil and coal are burned. _5/
Similarly, sulfur dioxide is a gas formed when a fuel containing sulfur (coal
and oil in particular) is burned. Consequently, stationary sources are, again,
the primary emitters of this air pollutant.
COLLECTING AND PREPARING THE AIR POLLUTION DATA FOR USE
Specifically, the air pollution data were taken from f5 & 6]. These
publications by the District Government report air pollution readings moni-
tored at a number of stations within the District of Columbia (see Table 6.1).
A number of missing observations occurred in the air pollution data
series. Consequently, to handle the problem of missing air pollution readings,
we adopted the technique of simple linear interpolation. This method replaces
the missing data in a series by a linear interpolation of the numeric values
before and after the missing point(s). The method can be used to handle con-
secutive missing observations as well as missing observations that may occur
at the beginning or end of a series. 6/ The general formula for linear inter-
polation is;
value^ - value_ ..
n . t+n t-1
v.ilue^. = value n +
t t-1 n
where value is the numeric estimate used to replace the missing data point
and n is the number of consecutive observations that are missing (it can, of
course, equal one). Ir. the case where the missing observations occur at either
kj Unfortunately, comprehensive data on suspended particulates were not avail-
able for analysis.
5/ It should be noted that a large percentage of total suspended particulates
can be attributed to natural sources.
£/ It should be noted that in these cases, the estimates are based on less
information.
30
-------
— 1
20 -
,
18 -
16-
14 -
12 •
—
G
a.. ip
o. 1U
3 -
6 -
4-
2-
—
0
t
i
i
\
\
\
\
n n i,
-T r: c !
n n •:
n n o
n n n
c R n
n a r
•z c n
~ .''. O '
•!:':':
;:::;•
%
]
o
\
'0
TT73
::•:
:::;:
\
\
?
^
\
\
i
•&-
;JT]
: n n »
neC|
: nfi i
n n n
HHr.
n n r
7 r r- •
r. c ':
? r, r
.-^ n .-] r
n r; '
r. ii ?> *
, - r .'
D* 00
LLU.LL
s
"
I
?
1
!
j
i
i
1
S
^
,.otfl
3O up
ODD
ggao
JDO 0
a a aa
; p DQ
3 u UH
D aon
3 dot
3D or
DCHU
U00»
ycjy.
naa=
«_•:-•::
a BUS
a I.O-A
c occ
a irts
'.insi.
?^:fl
OBM
b
\
r~ ; - ' 1
[Bflaoj
ipULJQj
:^ a P o;
ioanLT.
nnl
if VJ41-1!
DDOJ
bfln*i
k&Btt
ibflfipj
In jj PI n
u bjqu
|DQ*?3
fi£S.1
if-n^J
it^iaa
|gn(<3
IIS ';e~>
I ««i u •#*!
;u?jj
n£23
|H2g3
DPOtf
I :::•': :
i:::.:
. i . . - i
-;;
j i : : . :
.'.'.'.'. i
;;;;:
i
% \
\
"'r- *V-
;; -'..
\
%
1
\
r
L*
~~ ":"
13 •• -•
. " -
-
•' ^ .,
- -l"!
'• inf
- r • ' \
\
X
Mj5
i^C
c r
£ L
(TTp
!' ''
•
i
- !
• ::
ill
i: ::S
Oil
^
\ \
• \ \
\
\
National
MARYLAND
D.C.
VIRGINIA
Figure 6c. Concentrations of carbon monoxide (maximum 8 hr. averaee - 1973).
Source: [11.
-------
22 -
20 -
18 -
16 -
14 -
12 -
n. 10 -
o.
G -
G -
4 -
2 .
0
rr-r-
~_
TT-
•V',
;X
• '•'
• •
• "•*
X*
*«"
1
•r '. -'•
•W
","."
• *l"
i
•s'v-i
r^n
A"i
«
National
Standard
VIRGINIA
Figure 6d. Concentrations of carbon monoxide
(maximum 8 hr. average - 1974).
Source: (21.
32
-------
TABLE 6.1. AIR POLLUTION PROFILE
Air Pollutant
Photochemical
Oxidants
Non-Methane
Hydrocarbons
Nitrogen
Dioxide
Carbon
Monoxide
Sulfur
Diox ide
Primary
Standard
0.08 ppma'b
^ -,/ a.e
0.24 ppm
0.05 ppmg
9 ppm'
35 ppm
0.14 ppm '
0.03 ppmg
Secondary
Standard
0.08 ppma>b
0. 24 ppm' '
0.05 ppmg] ;
Beckman 1R31 5A
Technical!
'ieckman 906a
Reckman 906
icckman 906a
ii M
lierkman 906
Toclin ican
; Method
Chem i luminescence
ii ti it
Fl a me i on izat ion
Chem i luminescence
Nond i .spcrs i vc
Infrared
Parnrosanaline
Coul nmet i c
" "
1. I.
11 II
II '•
Piircitosanaline
Type of 1 Number oZ
Number of Times
Average johservat ions Standard Exceeded
max . 1 -hr .
" "
3-lir.
&:00am - 9: 00am
24-hr.
max . 1 -hr .
„
24-hr.
331 (354)d
(344)
225
4SC (20)d
(37)
|
165
,
348 ; na
337 2
310 (311) i 0 (0)
O23)
24-hr.
(0)
310 1
" " : 322
" " I (303)
max. 1-hr. ! 309
; 322
(322)
f)
0
0
(0)
•''Monitor i ng S t a t ion l.ocat ions
CA"P - 427 Xow .lersi-y Ave. . X. W
Cleveland Park l.ihrarv -
Conn. Avo. 6 M.-icomh SLs.
D.C. Get L-rol Hospital -
9th 6 Mass. Ave. , S.!'..
American Chemical Society
1155 16th St. . N.v:.
a - Not to be ex.ceded more than
once a vear.
b - Based on max mum one hour
average.
c - Data for 197 .
d - Data /or 197 .
e - Based on 3-hour average
(6:00 am - 9:00 am)
£ - Readings were obtained by mea-
suring separately total hydro-
carbons and methane.
R - Annual arithmetic mean.
li - [lased on 24-hour averages.
i - Based on 8-hour avera«es.
j - Based on 3-hour averages.
Source: [5 & 6j
-------
the beginning or the end of the series, they are replaced by the increment
computed from the two nearest numeric values. ]_/
THE METEOROLOGICAL DATA
Since weather conditions are the other major environmental factors hypoth-
esized to affect illness on a day-to-day basis, they must be controlled
for in the analysis. Consequently, we obtained meteorological data from the
National Weather Service. These data, in the form of a magnetic tape, con-
tained daily weather information for 1973 and 1974 from the station located
at Washington National Airport. National Airport was selected as being the
most representative weather station for the Washington Metropolitan Area. £/
The specific climatological variables that were collected on a daily basis
and that will enter this analysis were measures of temperature, wind, and
precipitation. 9/
Since the study was using daily health data, and since the air pollution
data were also being utilized as daily observations, the daily form of the
weather data posed no difficulties of compatibility. Consequently, daily
vectors of each climatological variable were created for use in conjunction
with the health and air pollution data.
]_l The means and standard deviations of the specific air pollution variables
used in the analysis appear in Appendix C.
8/ The alternatives were the weather stations located at either Dulles Inter-
national Airport or Baltimore-Washington International Airport. Dulles is
24 miles from downtown Washington and Baltimore International is 34 miles
from the downtown area.
9/ The means and standard deviations of the specific weather variables appear
in Appendix D.
34
-------
SECTION VII
STATISTICAL METHODS
INTRODUCTION
As discussed in Section III, techniques of multivariate analysis may be
the most promising approach to investigating the air pollution-health relation-
ship. Consequently, several multivariate methods were used to analyze the
health, air pollution, and weather data. Before presenting the results of
applying these procedures, two specific multivariate techniques, discriminant
analysis and multivariate regression analysis, will be discussed.
THE CHOICE OF TECHNIQUES
Discriminant analysis and regression analysis should not be viewed as two
alternative ways of solving the same problem. On the contrary, these tech-
niques address different questions under different statistical assumptions and
make different demands on the talents of the investigator. Neither technique
is superior on every count; it is for this reason that both are being employed
in this study.
A discriminant analysis permits the investigator to draw the following
type of conclusion: "A 20 percent increase in the oxidant level will lead to
a 0.005 increase in the probability of a person experiencing a sudden illness
on any given day." A regression analysis, on the other hand, permits the
investigator to draw the following sort of conclusion: "A 20 percent increase
in the oxidant level will lead to a 5 percent increase in the number of persons
experiencing a sudden illness on any given day."
If both of these conclusions could be made with 100 percent certainty,
the latter, more quantitative conclusion would be preferable to the former,
more probabilistic conclusion. Quantitative statements from a regression
analysis are more readily translated into information that a policymaker can
use to develop specific programs. For this reason, everything else being
equal, a regression analysis produces more powerful results. However, it is
seldom the case that everything else is equal. Hence, at times, the weaker
probability statement generated by the discriminant analysis may provide not
only sufficient, but also better guidance for the policymaker.
One reason why everything else is not equal is that the two techniques
require different statistical assumptions. The discriminant analysis assumes
that the variables explaining the probabilistic outcome are being randomly
drawn from two normally distributed multivariate populations having different
means but identical variances. I/ On the other hand, the regression analysis
assumes that the outcome is functionally or algebraically determined by certain
explanatory variables and a single, usually additive, random term (having zero
I/ For certain discriminant analyses more than two populations may be assumed.
35
-------
mean). 2/ It is important that the functional relationship between the out-
come and the explanatory variables is stable for all values of the explanatory
variables and that this function is correctly specified. Equally important
is the assumption that the values of the random error term are distributed
independently of the values of the explanatory variables. Although it is not
necessary to assume that this random error term is drawn from a normal popu-
lation, this assumption is usually made to enable certain types of hypothesis
testing.
If the two important assumptions for regression analysis hold -- the
independence of the error term and the specification of a correct and stable
functional relationship -- then it is known that "on average" statements
relating changes in an explanatory variable (e.g., oxidant levels) to changes
in the outcome (e.g., the number of sudden illnesses) will be correct. More-
over, the likelihood that the statements will be correct improves as the
amount of data (number of observations) used in the analysis increases. How-
ever, if these two assumptions fail to hold, such statements are known to be
biased (often in uncertain direction and magnitude) and the situation will
not improve even if the amount of data becomes quite large.
Thus, much of the analyst's research effort in applying the techniques
of regression analysis involves an attempt to specify the "correct" functional
relationship. This task is especially difficult if a theory relating the out-
come to the explanatory variables is lacking. Unfortunately, as pointed out
earlier, this lack of theory accurately describes the air pollution-health
relationship. This explains the attractiveness of the discriminant approach,
since it makes relatively weaker theoretical demands (albeit stronger statis-
tical demands) than the regression approach.
Fortunately, the two techniques can be complementary. For example, if it
is found from a discriminant analysis that a particular variable does not affect
the probability of an outcome, then this variable is unlikely to make a signifi-
cant quantitative contribution to explaining an outcome in a regression analy-
sis. Consequently, since the discriminant approach is relatively easy to under-
take (see below), it can serve as a useful "screening" device to limit the num-
ber of variables that must be considered in specifying a relationship for a
regression analysis. ^/
THEORY AND METHOD
The basic theories underlying multivariate regression and discriminant
analysis are thoroughly discussed in several well-known texts. 4/ There is no
need to repeat the discussion of regression analysis, since it is clearly pre-
21/ This remark applies to the type of single-equation analysis used in this
study. The assumptions underlying multiple-equation regression analysis
are somewhat more involved.
_3/ There are several theoretical reasons for wanting to minimize the number of
explanatory variables in a regression analysis. These include saving degrees
of freedom and reducing the likelihood of multicollinearity.
4/ See, for example, Johnston [26] or Goldberger [17] on regression analysis;
and Kendall [27] or Anderson [4] on discriminant analysis.
36
-------
sented in these texts. However, our use of discriminant analysis is somewhat
unconventional. Even though we do not alter the basic theoretical principles
or the mathematical formulation, it may be difficult for the reader to see the
connection between the conventional textbook presentation and our unconven-
tional application. Thus, we feel the following discussion is warranted.
Classical discriminant analysis is a procedure for optimally classifying
data into two or more groups. Each piece of data represents a set of measure-
ments and the classification procedure attempts to p,i"oup data with similar
measurements together. Our use of discriminant analysis starts with the as-
sumption that the data have already been classified. U'e then use the dis-
criminant analysis to draw certain inferences about that classification.
Formally, we let our datum be represented by X. We assume that X is a
m variable that could be drawn from either of tw
with probabilities q and q , respectively. That is,
random variable that could be drawn from either of two populations, TT or n.
Prob (Xen ) = q
1 1
and
Prob
Let the ith observed value of X be X.. Suppose we know that X will have
the value X. with a known probability provided we also know from which group
X was drawn. In particular, suppose we know that
Prob(X = X. Xen ) = P (X.)
and
Prob(X = X. I Xen ) = P2(X.)
We then pose the following problem: Given a value of X = XJL, what is
the probability chat X, in fact, came from population TI ? That is, what is
xi>?
From Baye's Theorem
Prob (Xerr ^ X,. ) =
1) Prob(X. Xen
Prob(X = X.-JXeT^) + Prob(XerO Prob(X = X^
37
-------
This can be rewritten
+ 1
In other words, two pieces of information are required to solve our prob-
lem. First, we must know (or estimate) the relative chances that X came from
population ff ; that is, we must determine q.-Xq.,. Since we have assumed that
our data have already been classified into two groups of, say, size N. and N ,
the ratio N./N? could be used to estimate q,/q,,. However, often it may be
desirable to neglect the sample count and instead set q-i/q/, by other, a priori
considerations. (For example, qi/q? is often set at unity indicating equal
chances.)
The other piece of required information is the relative probability that
X will have the value Xj[ given that X was drawn from population TJ, i.e.,
Pi(Xi)/?2(Xi). To estimate this ratio we assume that X is a function of p
variables represented by a vector Z. Moreover, if X is from population TT^
we assume that Z has a multivariate normal distribution with a vector mean
of M-I. If X belongs to population T^, we assume that Z has a multivariate
normal distribution with a vector mean of u^. Both normal distributions are
assumed to have the same variance, £.
Under this assumption it can be shown that
where Z± is the ith value of Z.
Each element of P-^ and M"~ can be estimated by the sample means. That is,
the kth mean of the first group can be estimated;
N
where 7^ represents the NI x p matrix associated with group 1.
The variance £ can also be estimated from the data by computing the
matrix of_ sums of squared deviations about the means and cross products.
Letting Z represent the vector of mean values of the p variables taken over
all N observations, we estimate 2 as.
38
-------
While the above is a perfectly valid computational procedure for calcu-
lating Prob(Xerrj| Xj^) , we have employed an alternative procedure that yields
certain additional useful information.
With a little manipulation, the expression
+ 1
can be written as
Prob(X6TT X.) = —
U
where
u
- - P2(xt) - "
It can be shown that the parameter u in the above expression can be
estimated by
TUT
u M KZjlft - 1^ + Z2)' Kb + Jin ^
where
N + N - 2
K =
N1N2 N1K2 £ ^vfi
V^ " Vr^(^ " 2
As before N,/N^ is an estimate of q,/q?.
39
-------
A
The parameter b is a vector of least-squares regression coefficients
estimated from the regression equation:
Yi = Zib
where Y is an observation vector of binary variables defined such that
and
Yt = 0 if X^
The advantage of this procedure is that the parameter b can be used to
calculate the marginal contribution of the variables Z to the Prob(Xerr I X ),
In particular, the vector of slopes is:
3prob(X6TT I x.)
and the vector of elasticities is
aprob(Xerr I X )
X.)
where the symbol @ indicates that direct matrix product.
It should be noted that both the slopes and elasticities change with
changes in Prob (Xe^r | X.) and changes in Z. Therefore, as a matter of
convention we will present these results only for the mean values of
Prob(XeTT, I X ) and the mean values of the p variables in Z.
APPLICATION AND HYPOTHESIS TESTING
As previously noted, the application of the above procedure starts with
the assumption that the data have already been properly classified into one
population group or the other (TT, or n2) • In practice, this classification
activity often relies on subjective judgment or guesswork. Fortunately, how-
ever, the validity of the classification can be assessed statistically.
Suppose, for example, our data consist of a set of daily observations
on the number of visits (X^) to a clinic treating sudden illness. Suppose,
further, that we hypothesize these visits depend on a set of variables (Z^)
including the average daily temperature, the day of the week, and the average
daily air pollution level. We might then wish to group the data into two
populations, TT^ and T^, where TT^ contains the daily observations in which the
number of visits exceeds the average value of daily visits and T^ contains the
daily observations less than or equal to this average. Our hypothesis is that
Zi will "explain" why any given Xi is greater or less than the average. In
40
-------
particular, given the concerns of this report, we hypothesize that the greater
the air pollution level the greater the probability that Xi will fall in popu-
lation T^.
Yet our classification scheme may be erroneous in the sense that Z may
not "explain" the classification. We can test for this by computing the fol-
lowing quadratic form:
U = (Zj_ - Z2)
2
It can be shown that U has a t distribution with Nj_ -f N2 - 1 degrees of free-
dom. The appropriate null hypothesis is that the means of the explanatory
variables _after being grouped into the two populations are the same, i.e.,
that Z^ = Z2« (If the means were equal one would not expect the variables to
"explain" why a particular X^ fell in one population as opposed to the other.)
Should we fail to reject the null hypothesis, we have the choice of either
altering the set of explanatory variables (Z) or of changing the classification
rule. The correct choice cannot be based on abstract statistical criteria
but rather it must be based on theoretical considerations, common sense, and,
in some cases, pure luckl
-------
SECTION VIII
EMPIRICAL RESULTS
INTRODUCTION
In the first phase of our empirical analysis we decided to perform simple
time-series plots of subsets of the data. The purpose of employing such
graphical techniques was to elicit any temporal patterns that might exist in
the data (e.g., day-of-the-week or seasonal effects).
From examinations of resulting graphs, we concluded that there were no
clear-cut day-of-the-week effects (except for weekend effects when clinic
hours were shorter or nonexistent) nor unambiguous seasonal effects. Conse-
quently, we proceeded with the analysis by applying our first multivariate
statistical technique, discriminant analysis. In doing so, we explored further
the possibility of temporal effects.
DISCRIMINANT ANALYSES
As we have discussed previously, the correct mathematical form of a model
to explain utilization of health care facilities is not known. Furthermore,
there are a large number of potential variables that could ostensibly fit such
a model. For both of these reasons, we began our multivariate investigation
by applying discriminant analysis. We felt that by adopting this approach
initially, we would be able to look at a large number of potential relation-
ships without the quantitative restrictions inherent in a multivariate regres-
sion analysis. Significant associations that were found could then be refined
by further examination.
We began by applying discriminant analyses to the 1973 data from the
Pennsylvania Avenue clinic. Specifically, we focused on the daily number of
unscheduled visits arriving at each of the departments from which we had col-
lected data. We then classified the data according to whether the number of
unscheduled visits was greater than (or equal to) or less than the average
number of unscheduled visits coming to that department on that particular day
of the week. \] Once classified, these data were then discriminated against
several possible explanatory variables.
In Section III we discussed that a number of factors are hypothesized to
affect the health status of a population. However, in analyzing day-to-day
variations in an index of health such as health clinic utilization, many of
these factors can be assumed to remain essentially constant. For example, the
population at risk, its habits, housing, and occupation mix are not likely to
exhibit substantial changes on a day-to-day basis if the time period under con-
sideration is not too long. On the other hand, air pollution levels, weather
I/ We did this procedure for each day of the week in an effort to control
implicitly for a day-of-the-week effects. Saturdays and Sundays were ex-
cluded, since as explained in Section V, most departments had shorter
hours on Saturdays and were closed on Sundays.
42
-------
conditions, and specific day-of-the-week factors (e.g., weekends, holidays,
working patterns) would exhibit daily fluctuations. Consequently, in our
initial analyses we controlled for these factors explicitly by including vari-
ables representing average daily temperature, average daily wind speed, 2/ and
daily pollution levels (as measured by the maximum 1-hour average photochemical
oxidant reading taken at the CAMP station located in downtown Washington). 3/
As discussed in footnote 1, day-of-the-week factors were controlled impliciTly
by our classification scheme.
The results of one such discriminant analysis are presented in Table 8.1.
The classification vector was based on the number of unscheduled visits to
the internal medicine department on a given day. Days with a greater than
average number of unscheduled visits (for that day of the week) were assigned
a one, while days with less than average numbers of unscheduled visits were
assigned a zero. This classification vector was then discriminated against
the three environmental measures described above.
TABLE 8.1. DISCRIMINANT ANALYSIS ON INTERNAL MEDICINE
DEPARTMENT (PENNSYLVANIA AVENUE - 1973)
Oxidant (ppm)
Temp. ( F)
Wind (mph)
Elasticities
-.105
.492
-.021
107, A Mean
.005
5.974
.851
A Prob.*
-.004
.019
-.001
U-Statistic = 0.037 (distributed as t )
* The mean probability of being classified as an unscheduled visit was
0.395.
As can be seen by the U-statistic below the table, the three environmental
variables did poorly in discriminating between above average and below average
unscheduled utilization of the internal medicine department. The signs of the
elasticities displayed in the table indicate that air pollution and wind were
negatively related to utilization, while temperature was positively related.
The magnitude of the elasticities permit "unit-free" comparisons of the rela-
tive associations. For example, a 10 percent increase in the oxidant level
was associated with a 1.05 percent decrease in the probability, i.e., an in-
crease of 0.005 ppm in the oxidant level was associated with a decrease of
0.004 in the probability of being classified as a day having greater than
2/ These climatic variables were selected for initial use on the basis of find-
~ ings presented in Lave and Seskin [28]. The analysis of other weather vari-
ables will be discussed below.
3/ Later, we will explore the results of using photochemical oxidant data from
~~ another monitoring station as well as data on other air pollutants.
43
-------
average unscheduled visits to the internal medicine department, kj
Similar results were exhibited by discriminant analyses involving two of
the three remaining departments (urgent visit clinic and pediatrics). No
consistent pattern was uncovered between daily air pollution levels (as mea-
sured by photochemical oxidants) and the daily number of unscheduled visits
arriving at these departments. However, the data from the ophthalmology de-
partment exhibited slightly different characteristics. The results of the
discriminant analysis analogous to the one presented above for internal
medicine, suggested a possible association between air pollution and un-
scheduled visits for eye problems. The specifics of the discriminant analysis
for the ophthalmology department are shown in Table 8.2.
TABLE 8.2. DISCRIMINANT ANALYSIS ON OPHTHALMOLOGY
DEPARTMENT (PENNSYLVANIA AVENUE - 1973)
Oxidant (ppm)
Temp. (°F)
Wind (mph)
Elasticities
.379
-.376
.218
107. A Mean
.005
5.974
.851
A Prob.*
.016
-.016
.009
U-Statistic = 0.125 (distributed as t )
* The mean probability of being classified as an unscheduled visit was
0.422.
Except for the classification vector, the variables were identical to
those described above. Although the U-statistic increased, it was still not
statistically significant. The three environmental variables did not dis-
criminate well between above average and below average unscheduled utilization
of the ophthalmology department. However, in this case, air pollution (as
measured by photochemical oxidants) and wind were positively related to utili-
zation, while temperature was negatively related. As can be seen from the
elasticities, air pollution and temperature had effects of roughly the same
magnitude, the effect of wind was less pronounced. Although not reported,
the t-statistics corresponding to the variables in the 1-0 regression (see
previous footnote) indicated that the air pollution variable was statistically
significant. This warranted further exploration.
4/ As would be expected from these results, the t-statistics (not reported) for
each of the variables derived in the 1-0 regression used in performing the
discriminant analysis (see Section VII) were all statistically insignificant.
In the context of this analysis, a factor or variable was considered "sta-
tistically significant" if its regression coefficient was significantly
different from zero at (at least) the 20 percent level. That is, we had at
least 80 percent confidence in the individual statements regarding the asso-
ciation.
44
-------
The next step we took was to rerun the discriminant analyses using only
the data for the summer months of 1973. This time period was of special
interest because of the relatively high air pollution levels that occurred.
(Four air pollution alerts were called by the Metropolitan Washington Council
of Governments.) The results of these further analyses were similar to those
for the entire year in that no consistent pattern was found between daily
photochemical oxidant levels and the daily number of unscheduled visits ar-
riving at these departments.
After examining the summer months, we decided to use the 1974 data in an
effort to replicate the 1973 results. Tables 8.3 and 8.4 present discriminant
analyses using 1974 data that are analogous to those shown in Tables 8.1 and
8.2, respectively.
With regard to unscheduled visits to the internal medicine department,
it can be seen from Table 8.3 that the U-statistic still indicates that the
environmental variables did poorly in discriminating between above average
and below average utilization. However, the signs of the elasticities now
indicate that oxidant levels and temperature were positively related to utili-
zation, while wind was negatively related. The magnitudes of the elasticities
show little change for the air pollution variable between the two years, but
substantial differences for the two climatic variables. 5/
8.3. DISCRIMINANT ANALYSIS ON INTERNAL MEDICINE
DEPARTMENT (PENNSYLVANIA AVENUE - 1974)
Oxidant (ppm)
Temp. (°F)
Wind (mph)
Elasticities
.150
.061
-.353
10% A Mean
.004
5.873
.869
6 Prob.*
.007
.003
-.017
U-Statistic = 0.131 (distributed as t )
* The mean probability of being classified as an unscheduled visit was
0.471.
The results of the 1974 analysis of the ophthalmology department (Table
8.4) were comparable to the 1973 results. The U-statistic increased; however,
it remained statistically insignificant. The signs of the elasticities were
unchanged and their magnitudes were slightly larger in the 1974 replication. 6>/
5/ The t-statistics corresponding to the variables in the 1-0 regression (not
reported) were all statistically insignificant. (See footnote 4 in this
section.)
6/ Again, the t-statistic for the oxidant variable was statistically signifi-
cant.
45
-------
TABLE 8.4. DISCRIMINANT ANALYSIS ON OPHTHALMOLOGY
DEPARTMENT (PENNSYLVANIA AVENUE - 1974)
Oxidant (ppm)
Temp. (°F)
Wind (mph)
Elasticities
.469
-.673
.334
107, A Mean
.004
5.873
.869
A Prob.*
.022
-.031
.016
U-Statistic = 0.284 (distributed as t )
* The mean probability of being classified as an unscheduled visit was
0.467.
REGRESSION ANALYSES
Given the findings of the discriminant analyses which in general did not
evidence an association between photochemical oxidant pollution and clinic
utilization (except possibly a suggestive relationship between oxidant pollu-
tion and unscheduled visits to the ophthalmology department), we decided to
proceed by analyzing the same data using multiple regression. As discussed
earlier, this approach facilitates the investigation of more specific quanti-
tative hypotheses. Initially, we began by running some simple regressions.
One such model is shown below and corresponds to regression 1 in Table 8.5:
UIM = 17.859 - 12.317 Ox + 0.003 Av T + 0.024 Av Wind
(-1.00) (0.15) (0.24)
-9.709 Sat - 17.622 Sun
(-11.59) (-22.44)
(R = 0.638)
UIM represents the number of unscheduled visits to the internal medicine de-
partment on a given day, Ox is the maximum 1-hour average oxidant reading on a
given day, Av T is the average temperature for the day, and Av Wind is the
average wind speed for that day. Sat and Sun are dummy variables representing
the weekends. ]_/ The equation above represents the number of unscheduled visits
to the internal medicine department during 1973. &/ As can be seen, 63.8 per-
7/ Dummy variables are specially constructed variables that may be used to
represent various factors such as temporal effects, spatial effects, quali-
tative variables, and broad f»roupin»s of quantitative variables. In our
analysis we employed two dummy variables for the two days of the weekend.
Specifically, if an observation represented data pertaining to a Saturday,
the Saturday dummy variable (Sat) was assigned a value of 1 and for all
other days it was assigned a value of 0. Similarly, if an observation
represented data pertaining to a Sunday, the Sunday dummy variable (Sun)
was assigned a value of 1 and for all other days it was assigned a value of
0.
8_/ Seven days in April and twenty-three days in May were omitted because no
air pollution data were available.
46
-------
TABLE 8.5. REGRESSION ANALYSIS OF OXIDANT EFFECTS
ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
1
UOPHTH
2
UPED
3
R
Constant
0.638 a/
17.859
0.460
8.178
0.646
51.952
Air Pollution:
Ox
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
-12.317 b/
(-1.00) c/
0.003
(0.15)
0.024
(0.24)
-9.709
(-11.59)
-17.622
(-22.44)
13.867
(1.73)
-0.013
(-0.90)
-0.005
(-0.07)
-4.444
(-8.10)
-8.055
(-15.67)
-22.764
(-0.83)
-0.051
(-1.02)
-0.509
(-2.33)
-4.387
(-2.34)
-42.499
(-24.16)
a/ The coefficient of determination; a value of 0.638 indicates that 63.8
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
47
-------
cent of the variance in daily unscheduled visits was explained by the right-
hand-side variables (R = 0.638). The t-statistics in parentheses below the
coefficients indicate that none of the environmental variables was statisti-
cally significant; 9/ the signs of the coefficients indicate that oxidant pol-
lution was negatively related to clinic visits, while both average temperature
and average wind speed were positively related. It is apparent that the ex-
planatory power of the regression was derived primarily from the influence of
the dummy variables for Saturdays and Sundays. As expected, utilization was
negatively associated with weekends.
Similar regression results were found for the pediatric department
(Table 8.5, regression 3). That is, there was no evidence of a daily effect
of photochemical oxidant pollution on daily visitation. The ophthalmology
department, however, displayed suggestive results of a possible association.10/
This can be seen in the equation below which corresponds to regression 2 in
Table 8.5:
UOPHTH = 8.178 + 13.867 Ox - 0.013 Av T - 0.005 Av Wind
(1.73) (-0.90) (-0.07)
-4.444 Sat - 8.005 Sun (R2 = 0.460)
(-8.10) (-15.67)
UOPHTH represents the number of unscheduled visits to the ophthalmology depart-
ment on a given day, and the remaining variables are as defined above. As can
be seen, 46.0 percent of the variance in the daily unscheduled visits was ex-
plained by the right-hand-side variables (R = 0.460). In this case the coef-
ficient and t-statistic of the air pollution variable indicate a positive and
statistically significant (at approximately the 10 percent significance level)
association with the unscheduled ophthalmology visits. One interpretation of
the magnitude of this result is that an increase of 0.01 parts per million
(ppm) in the maximum 1-hour average oxidant level (raising the mean from 0.048
to 0.058 ppra) was related to an increase of 0.14 (0.01 x 13.867) daily un-
scheduled visits to the ophthalmology department (raising the mean number of
daily unscheduled visits from 6.03 to 6.17). An alternative interpretation
of this result is that a 10 percent decrease in the mean of the air pollution
variable (decreasing the mean by 0.0048 ppm) would be associated with a 1.1
((0.0048 x 13.867)/6.03) percent decrease in the number of unscheduled visits
to the ophthalmology department. The signs of the coefficients of the other
environmental variables indicated a negative relationship with both average
temperature and average wind speed; both were statistically insignificant. The
weekend dummy variables exhibited similar results to those reported above. 11_/
97 A value of 1.65 indicates significance at the 10 percent level using a two-
tailed test; a value of 1.28 indicates significance at the 20 percent level
using a two-tailed test.
10/ Regressions were not run for the 1973 urgent visit clinic data because of
the schedule change that occurred during the year (see Section V, p. 20).
ll/ We experimented with specifications omitting the weekend dummy variables
along with all data pertaining to the weekends. As expected, the coef-
ficient of variation (R^) decreased dramatically; however, the magnitude
48
-------
Similar results to those just reported were found when regressions were
run on only the summer months of 1973 (not shown). That is, a significant as-
sociation was seen for visits to the ophthalmology department, while no ap-
parent associations were seen for visits to the other departments.
The results of replicating the 1973 regressions using 1974 data are shown
in Table 8.6. 12/ Replications for the internal medicine department and the
ophthalmology department using air pollution data from the GAMP station are
shown in the equations below (they correspond to regressions 2 and 4, respec-
tively) :
UIM = 11.195 -f 8.873 Ox + 0.004 Av T - 0.156 Av Wind
(0.48) (0.13) (-1.32)
-6.092 Sat - 10.135 Sun (R2 = 0.266)
(-5.88) (-10.32)
UOPHTH = 5.996 + 60.424 Ox - 0.040 Av T + 0.163 Av Wind
(4.54) (-1.72) (1.89)
-5.017 Sat - 7.528 Sun (R2 = 0.300)
(-6.67) (-10.54)
One notes that the variance explained of the 1974 unscheduled
visits to the internal medicine department dropped substantially (R2
decreased from 0.638 to 0.266). Again, all three environmental
variables were statistically insignificant; only the dummy variables
representing the weekends were statistically important.
The results for the ophthalmology department also evidenced a decrease in
the coefficient of variation (R ). However, the t-statistics now indicated that
all three environmental variables were statistically significant. The sugges-
tive association between photochemical oxidant levels and unscheduled visits to
the ophthalmology department still held. The magnitude of the association ex-
hibited by the 1974 data indicated that an increase of 0.01 ppra in the maximum
1-hour average oxidant level (raising the mean from 0.038 to 0.048 ppm) was
related to an increase of 0.60 (0.01 x 60.424) daily unscheduled visits to the
ophthalmology department (raising the mean number of daily unscheduled visits
from 5.38 to 5.98). An alternative interpretation of this result is that a
10 percent decrease in the mean of the air pollution variable (decreasing the
mean by 0.0038 ppm) would be associated with a 4.3 ((0.0038 x 60.424)/5.38)
percent decrease in the number of unscheduled visits to the ophthalmology de-
partment.
11/ (continued) and statistical significance of the other explanatory variables
were not significantly affected. Evidently, there are many unaccounted
for variables influencing unscheduled utilization of the various depart-
ments, but the statistically significant associations we observed were not
simply artifacts of a particular specification.
12/ Since homogeneous data for the urgent visit clinic were available, regres-
sion results for that department are shown. In addition, photochemical oxi-
dant data were obtained from another monitoring station in Cleveland Park
(C.P.); hence, two regressions for each department were analyzed.
49
-------
TABLE 8.6. REGRESSION ANALYSIS OF OXIDANT EFFECTS
ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1974)
UIM
UOPHTH
UPED
UVC
0.270 a/ 0.266 0.255 0.300 0.476 0.485 0.302 0.307
Constant
8.732 11.195 5.956 5.996 62.613 69.824 154.942 152.459
Air Pollution;
Ox (CAMP)
Ox (C.P.)
Weather;
Av T
Av Wind
Dummy:
Sat
Sun
-28.270
(-1.82)
8.
(0.
b/
£/
873
48)
-7.432
(-0.61)
60.424
(4.54)
-15
(-0
.353
.31)
108.
(1.
634
87)
153
(1
.071
.80)
138.844
(1.35)
0.066
(1.98)
-0.137
(-1.13)
0.004
(0.13)
-0.156
(-1.32)
0.023 -0.040
(0.88) (-1.72)
-0.332 -0.498
(-3.10) (-4.86)
0.008
(0.04)
0.067
(0.71)
0.163
(1.89)
1.195
(3.10)
1.044
(2.78)
0.114
(0.63)
-0.023 -0.256
(-0.09) (-0.39)
-6.136 -6.092 -5.188 -5.017 -7.526 -8.242 -31.560 -32.867
(-5.86) (-5.88) (-6.28) (-6.67) (-2.25) (-2.21) (-5.50) (-5.67)
-9.829 -10.135 -7.362 -7.528 -50.766 -53.058 -61.257 -64.155
(-9.97) (-10.32) (-9.48) (-10.54) (-16.14) (-17.01) (-11.34) (-11.66)
a/ The coefficient of determination; a value of 0.270 indicates that 27.0 percent of the
variation in unscheduled visits was "explained" by the independent variables.
b/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent level, using a
two-tailed test.
50
-------
The 1974 results, unlike the 1973 findings, also revealed a statistically
significant association between unscheduled visits to the pediatric department
and levels of photochemical oxidants. The relevant equation is shown below
and corresponds to regression 6 in Table 8.6:
UPED = 69.824 + 108.634 Ox - 0.498 Av T + 1.044 Av Wind
(1.87) (-4.86) (2.78)
-8.242 Sat - 53.058 Sun (R2 = 0.485)
(-2.21) (-17.01)
The variables are as above except for the dependent variable UPED which repre-
sents the number of unscheduled visits to the pediatric department on a given
day. As can be seen, 48.5 percent of the variance in daily visits to the pe-
diatric department was explained by the right-hand-side variables (R = 0,485)
The coefficent and t-statistic of the air pollution variable indicate a posi-
tive and statistically significant (at the 10 percent significance level) as-
sociation with the unscheduled visits. One interpretation of the magnitude of
this result is that an increase of 0.01 ppra in the maximum 1-hour average
oxidant level (raising the mean from 0.038 to 0.048 ppm) was related to an
increase of 1.08 (0.01 x 108.634) daily visits to the pediatric department
(raising the mean number of daily unscheduled visits from 43.70 to 44.78). An
alternative interpretation of this result is that a 10 percent decrease in the
mean of the air pollution variable (decreasing the mean by 0.0038 ppm) would
be associated with a 0.9 ((0.0038 x 108.634)743.70) percent decrease in the
number of unscheduled visits. The coefficients and t-statistics of the other
environmental variables indicated statistically significant effects of tempera-
ture and wind on pediatric visits; temperature was negatively related and wind
was positively related to such visits. The weekend dummy variables again
reflected decreased visitation on weekends.
In addition to the photochemical oxidant data monitored at the
CAMP station, comprehensive data for this air pollutant were available from
another station located at Cleveland Park (see Table 6.1). When those readings
were substituted for the CAMP station readings, only the urgent visit clinic
exhibited a positive and statistically significant association between air
pollution levels and unscheduled utilization. The relevant equation is shown
below and corresponds to regression 7 in Table 8.6:
UVC = 154.942 + 153.071 Ox + 0.008 Av T - 0.023 Av Wind
(1.80) (0.04) (-0.09)
-31.560 Sat - 61.257 Sun (R2 = 0.302)
(-5.50) (-11.34)
Again, the variables are as above except for the dependent variable UVC
which represents the number of urgent visits to the urgent visit clinic on a
given day. As can be seen, 30.2 percent of the variance in daily visits to
the urgent visit clinic was explained by the right-hand-side variables
(R = 0.302). The coefficient and t-statistic of the air pollution variable
indicate a positive and statistically significant (at the 10 percent signifi-
cance level) association with the urgent visits. One interpretation of the
magnitude of this result is that an increase of 0.01 ppm in the maximum 1-hour
51
-------
oxidant level (raising the mean from 0.044 to 0.054 ppm) was related to an
increase of 1.53 (0.01 x 153.071) daily visits to the urgent visit clinic
(raising the mean number of daily urgent visits from 147.50 to approximately
149.03). An alternative interpretation of this result is that a 10 percent
decrease in the mean of the air pollution variable (decreasing the mean by
0.0044 ppm) would be associated with a 0.46 ((0,0044 x 153.071)7147.50) per-
cent decrease in the number of urgent visits. The coefficients and t-sta-
tistics of the other environmental variables indicated statistically
insignificant effects of temperature and wind on urgent visits. The week-
end dummy variables again reflected decreased utilization on weekends.
Taken together, the results for 1973 and 1974 relating photochemical
oxidant pollution to unscheduled utilization suggest a number of possible
associations worthy of further exploration. Both 1973 and 1974 data
exhibited a relationship between unscheduled ophthalmology visits and
oxidant levels, although 1974 air pollution data from a second monitoring
station failed to demonstrate a similar association. In addition, 1974 data
indicated mixed associations between unscheduled pediatric visits and oxi-
dant levels; air pollution data from one monitoring station exhibited a
'positive and statistically significant relationship with unscheduled visits,
while data from a second station displayed a negative and statistically
insignificant relationship. Finally, 1974 data suggested an association
between urgent clinic visits and oxidant levels. Air pollution readings
from one monitoring station were related positively and significantly with
such visits and readings from a second station were also related positively
although not quite significantly. Given these findings, further analysis
was warranted.
LAG AND EPISODIC EFFECTS
In addition to the fact that a time-series multiple regresssion allows one
to investigate contemporaneous effects while assuming that many of the factors
affecting health status remain essentially constant on a day-to-day basis, it
also permits one to investigate other possible effects. In particular we
examined lag effects and episodic effects of photochemical oxidants.
Since we might expect that our measure of health status, namely clinic
utilization, would be affected by levels of air pollution on immediately pre-
ceding days, we included lagged air pollution variables representing the air
pollution readings on the three preceding days. 13/ The results (summarized
in Table 8.7 and Table 8.8) of including these simple lagged variables indi-
cated that there were no consistent, significant simple lag effects relating
department utilization and oxidant levels for both years of data. Specifically,
the combined net effect of the contemporaneous and lagged pollution variables
was positive for only one of the three departments in 1973 (regression 1,
Table 8.7) and it failed to hold for that department in 1974 (regression 2,
Table 8.8). Furthermore, the combined net effect for the ophthalmology de-
partment in 1974 (regression 4, Table 8.8) was not significantly greater than
the simple contemporaneous association estimated previously (regression 4,
Table 8.6).
13/ Several studies have used lags of up to three days. See, for example
Greenburg et^a1« [19 ]•
52
-------
TABLE 8.7. LAG EFFECTS OF OXIDANTS ON UN-
SCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
R2
Constant
Air Pollution:
Oxt
<*t-l
°*t.2
°*t-3
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
UIM
1
0.636 a/
16.823
26.034 b/
(1.78) c/
-7.190
(-0.46)
-5.108
(-0.34)
18.309
(1.39)
-0.034
(-1.49)
0.058
(0.54)
-9.340
-(10.94)
-16.833
(-20.88)
UOPHTH
2
0.524
9.412
1.968
(0.21)
12.717
(1.24)
-10.803
(-1.08)
-6.345
(-0.73)
-0.014
(-0.92)
0.016
(0.22)
-5.347
(-9.54)
-8.517
(-16.10)
UPED
3
0.702
55.257
10.246
(0.34)
37.819
(1.16)
-69.348
(-2.19)
-6.880
(-0.25)
-0.113
(-2.38)
-0.335
(-1.48)
-4.417
(-2.48)
-42.739
(-25.46)
a/ The coefficient of determination; a value of 0.636 indicates that 63.6
percent of the variation in unscheduled visits was "explained" by the
Independent variables,
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
53
-------
TABLE 8.8. LAG EFFECTS OF OXIDANTS ON UN-
SCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1974)*
R2
Constant
UIM
1 2
0.276 a/ 0.278
8.474 12.268
UOPHTH
3 4
0.292 0.318
4.136 7.472
UPED UVC
5 6 7
0.515 0.526 0.318
60.687 75.134 159.077
8
0.339
165.601
Air Pollution:
Oxc
°Vi
°V2
°V3
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
* The
the
a/ The
the
b/ The
£/ The
-18.765 b/ 10.178
(-1.02) c/ (0.51)
-13.116 -0.145
(-0.68) (-0.01)
-12.769 -6.439
(-0.67) (-0.31)
0 007 -7.580
(0.01) (-0.39)
0.079 0.006
(1.84) (0.16)
-0.108 -0.205
(-0.86) (-1.70)
-6.059 -6.310
(-5.66) (-5.85)
-10.031 -10.423
(-9.90) (-10.28)
8.314 51.902
(0.58) (3.07)
-19.418 14.866
(-1.29) (0.99)
-26.272 0.149
(-1.77) (0.01)
-14.526 27.646
(-1.09) (1.99)
0.081 -0.082
(2.39) (-2.90)
0.114 0.139
(1.16) (1.59)
-5.386 -4.863
(-6.42) (-6.23)
-7.722 -7.548
(-9.74) (-10.28)
31.908 114.358 88.693
(0.56) (1.87) (0.91)
-8.579 82.336 102.977
(-0.14) (1.30) (1.01)
-95.987 -111.195 -226.244
(-1.64) (-1.77) (-2.23)
-92.119 -32.386 122.890
(-1.75) (-0.55) (1.35)
-0.184 -0.508 0.021
(-1.37) (-4.26) (0.09)
1.188 0.893 -0.184
(3.06) (2.43) (-0.27)
-7.065 -8.718 -30.759
(-2.14) (-2.66) (-5.38)
110.178
(1.03)
77.995
(0.70)
-94.740
(-0.86)
15.747
(0.15)
0.001
(0.01)
-0.590
(-0.92)
-35.811
(-6.24)
-51.510 -54.769 -61.566 -65.407
(-16.46) (-17.75) (-11.38) (-12.12)
odd-numbered regressions aie based oxidant data collected at Cleveland Park and
even-numbered regressions are based on oxidant data collected at the CAMP station.
coefficient of determination; a value of 0.276 indicates that 27.6 percent of
variation in unscheduled visits was "explained" by the independent variables.
regression coefficient.
t-statistic; a value of 1.65 indicates significance at the 10 percent level,
using a two-tailed test.
54
-------
In addition to testing for simple lags, we employed the Almon distributed
lag technique in order to examine the data for more complicated lag effects.
This procedure imposes structure on the coefficients of the lagged air pollu-
tion variables by constraining them to fit a polynomial curve of a specified
degree. The method often results in the reduction of large standard errors
in the distributed lag coefficients that may arise from multicollinearity in
the lagged values of the independent variables. It also allows for consider-
able flexibility. We began by fitting second- and third-degree polynomials,
using lags of five days. The results (not reported) did not uncover any con-
sistent, significant lag effects.
Time-series multiple regression also allows one to investigate whether
a consecutive period of several days of high air pollution are more "detri-
mental" to health than isolated days of high air pollution. That is, it per-
mits one to examine the implications of air pollution episodes. This issue
is of particular interest since several studies have uncovered such episodic
effects. 14 / One way to explore the possibility of episodic effects is to
define a new air pollution variable as the product of the air pollution read-
ings on several consecutive days and substituting this variable in place of a
simple contemporaneous air pollution measure. Specifically, we defined a new
air pollution variable as the product of the photochemical oxidant levels for
the current day and the two preceding days. 15/ The results (Table 8.9 and
Table 8.10) of using this variable in place of the simple contemporaneous air
pollution measure indicated that the data did not evidence episodic effects
of this nature. (In fact, when the coefficients for the episodic variables were
statistically significant, they were negative.) We then examined "episodes" of
different lengths and found similar results. In other words, we did not find
that department utilization significantly increased during periods in which air
pollution levels were elevated for a successive number of days.
THE EFFECTS OF OTHER AIR POLLUTANTS
In addition to looking at the association between photochemical oxidant
levels and clinic utilization, we were able to investigate the effects of
three other air pollutants primarily related to mobile sources (non-methane
hydrocarbons, nitrogen dioxide, and carbon monoxide) and one air pollutant
primarily related to stationary sources (sulfur dioxide). This section will
present the regression results pertinent to these air pollutants.
Non-methane Hydrocarbons
We reported in Section VI that the levels of non-methane hydrocarbons
exceeded the national standards on most days for which data were available
(see Table 6.1). Nevertheless, the results of substituting readings on non-
methane hydrocarbons for photochemical oxidants in statistical analyses similar
to those reported above did not uncover any consistently significant relation-
ship between unscheduled utilization in any department and the level of non-
methane hydrocarbons (as measured by the 3-hour average taken at the CAMP
station) during 1973 (Table 8.11). Unfortunately, adequate 1974 non-methane
hydrocarbon data were not available to permit a replication of these findings.
14/ For a detailed review of several air pollution episodes, see Ashe [7] and
McCarroll [30 |. 2
15/ In notational terms, Ox- = TT Qx
— t k=0 t-k
55
-------
TABLE 8.9. EPISODIC EFFECTS OF OXIDANTS ON
UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
1
UOPHTH
2
UPED
3
R
Constant
0.639 a/
18.120
0.448
7.930
0.634
52.054
Air Pollution:
k=0 °*t-k
Weather :
Av T
Av Wind
Dummy:
Sat
Sun
-0.535 b/
(-0.72) c/
-0.005
(-0.22)
0.022
(0.22)
-10.189
(-11.74)
-17.855
(-21.39)
0.152
(0.32)
-0.001
(-0.07)
-0.015
(-0.24)
-4.352
(-7.87)
-7.787
(-14.65)
-1.214
(-0.72)
-0.067
(-1.47)
-0.527
(-2.35)
-3.618
(-1.86)
-42.176
(-22.56)
a/ The coefficient of determination; a value of 0.639 indicates that 63.9
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
56
-------
TABLE 8.10. EPISODIC EFFECTS OF OXIDANTS ON-
UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 197'
UIM
UOPHTH
UPED
UVC
0.282 a/ 0.271 0.275 0.257 0.512 0.500 0.308 0.326
Constant
9.A87 11.646 5.436 4.794 64.184 71.825 156.890 161.652
Air Pollution:
2
iA)c
(CP)
-2.344 b/
(-2.47) c/
-1.997
(-2.65)
-8.067
(-2.73)
1.189
(0.23)
k=0
(CAMP)
-0.418
(-0.16)
1.354
(0.70)
4.459
(0.56)
10.907
(0.79)
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
0.042
(1.50)
-0.131
(-1.05)
0.010
(0.34)
-0.186
(-1.55)
0.034
(1.52)
0.074
(0.77)
0.018
(0.83)
0.147
(1.63)
-0.321
(-3.67)
1.107
(2.93)
-0.445
(-4.87)
0.938
(2.50)
0.135
(0.89)
0.096
(0.61)
-0.289 -0.539
(-0.44) (-0.84)
-6.208 -6.144 -5.531 -4.838 -8.231 -8.084 -31.258 -35.131
(-5.87) (-5.77) (-6.58) (-6.03) (-2.50) (-2.43) (-5.47) (-6.13)
-10.219 -10.230
(-10.33) (-10.32)
-7.595 -7.159 -51.337 -52.911 -60.700 -63.847
(-9.66) (-9.60) (-16.65) (-17.10) (-11.36) (-11.98)
a/ The coefficient of determination; a value of 0.282 indicates that 28.2 percent of the
variation in unscheduled visits was "explained" by the independent variables.
b/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent level, using a
two tailed test.
57
-------
TABLE 8.11. REGRESSION ANALYSIS OF NON-METHANE.HYDROCARBON
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
1
UOPHTH
2
UPED
3
Constant
0.616 a/
15.475
0.542
10.673
0.715
53.761
Air Pollution:
Nm H
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
0.181 b/
(0.27) c/
-0.005
(-0.24)
0.064
(0.49)
-8.307
(-8.81)
-15.738
(-17.29)
-0.021
(-0.05)
-0.022
(-1.66)
-0.070
(-0.81)
-5.207
(-8.36)
-8.777
(-14.61)
0.826
(0.59)
-0.123
(-2.91)
-0.208
(-0.76)
-6.350
(-3.20)
-43.082
(-22.53)
a/ The coefficient of determination; a value of 0.616 indicates that 61.6
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
58
-------
Nitrogen Dioxide
Unlike the situation with non-methane hydrocarbons, we reported in Sec-
tion VI that the levels of nitrogen dioxide in the Washington Metropolitan Area
were below the national primary and secondary standard. When we substituted
a nitrogen dioxide variable for the photochemical oxidant variable (in speci-
fications similar to those reported previously) we did not find evidence of
consistently significant associations between unscheduled utilization in any
department and the level of nitrogen dioxide (as measured by the 24-hour
average taken at the CAMP station) during 1973 (Table 8.12). Again, we did
not have sufficient 1974 nitrogen dioxide data to test these findings by
replication.
Carbon Monoxide
The final mobile-source air pollutant that we examined was carbon monoxide.
The discussion in Section VI indicated that levels of this air pollutant were
at or above the national standards at times during 1973 and 1974. Our empiri-
cal results from substituting a carbon monoxide variable for the photochemical
oxidant variable suggested an association between unscheduled utilization of
the ophthalmology department and levels of carbon monoxide (as measured by the
maximum 1-hour average taken at the GAMP station) during 1973. The relevant
equation is shown below and corresponds to regression 3 in Table 8.13:
UOPHTH = 6.301 +0.119 CO + 0.009 Av T + 0.035 Av Wind
(2.82) (0.74) (0.52)
-4.405 Sat - 7.653 Sun (R2 = 0.470)
(-7.86) (-14.36)
The variables are similar to those in the specification above, although CO
(the maximum 1-hour average carbon monoxide reading on a «iven day) is substi-
tuted for the photochemical oxidant variable, Ox. Rather than discussing this
equation in detail, we will interpret the results relative to those obtained
for photochemical oxidants. One interpretation of the equation is that an
increase of 1.0 ppm in the maximum 1-hour average carbon monoxide level
(raising the mean from 6.94 to 7.94 ppm) was related to an increase of 0.12
daily unscheduled visits to the ophthalmology department (raising the mean
number of unscheduled visits from 6.07 to 6.19). An alternative interpreta-
tion of this result is that a 10 percent decrease in the mean of the air pol-
lution variable would be associated with a 1.5 percent decrease in the number
of unscheduled visits to the department.
in addition to the carbon monoxide data monitored at the CAMP station,
comprehensive data for this air pollutant were available from another station
located at i). C. General Hospital (see Table 6.1). When those readings were
substituted for the CAMP station readings, the results indicated that the
association for the ophthalmology department failed to hold. However, a
positive and statistically significant association was seen between
unscheduled pediatric visits and carbon monoxide readings from the D. C.
59
-------
TABLE 8.12. REGRESSION ANALYSIS OF NITROGEN DIOXIDE
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
1
UOPHTH
2
uvc
3
R
Constant
0.615 a/
18.715
0.456
7.979
0.645
51.865
Air Pollution:
N02
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
-17.414 b/
(-0.78) c/
-0.010
(-0.53)
-0.007
(-0.07)
-9.729
(-11.48)
-17.365
(-21.57)
3.275
(0.23)
0.002
(0.20)
-0.022
(-0.32)
-4.497
(-8.21)
-8.068
(-15.49)
-38.977
(-0.80)
-0.474
(-1.18)
-0.465
(-2.01)
-4.634
(-2.52)
-42.599
(-24.41)
a/ The coefficient of determination; a value of 0.615 indicates that 61.5
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
60
-------
TABLE 8.13. REGRESSION ANALYSIS OF CARBON MONOXIDE
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
UOPHTH
UPED
0.599 a/ 0.634
Constant 18.377 16.339
0.470 0.446
6.301 7.939
0.650 0.631
48.570 44.760
Air Pollution:
CO (CAMP) -0.077 W
(-1.15) c/
CO (B.C.)
Weather.-
Av T
Av Wind
Dummy:
Sat
Sun
0.102
(1.50)
-0.010 -0.011
(-0.52) (-0.60)
-0.010 0.132
(-0.09) (1.27)
-9.645 -9.597
(-10.76) (-11.12)
-17.329 -17.375
(-20.32) (-21.26)
0.119
(2.82)
0.009
(0.74)
0.035
(0.52)
-4.405
(-7.86)
-0.037
(-0.83)
0.007
(0.56)
-0.026
(-0.38)
-4.241
(-7.51)
0.136
(0.96)
-0.057
(-1.32)
-0.312
(-1.36)
0.306
(1.97)
-0.032
(-0.75)
-0.168
(-0.71)
-7.653 -7.880
(-14.36) (-14.74)
-3.394 -4.400
(-1.78) (-2.22)
-42.192 -42.125
(-23.34) (-22.41)
§_/ The coefficient of determination; a value of 0.599 indicates that 59.9 percent o£
the variation in unscheduled visits was "explained" by the independent variables.
b/ The regression coefficient.
c_/ The t-statistic; a value of 1.65 indicates signficance at the 10 percent level,
using a two-tailed test.
61
-------
General Hospital station. 16/ The equation corresponding to regression 6
in Table 8.13 for the pediatric department is presented below:
UPED = 44.760 + 0.306 CO - 0.032 Av T - 0.168 Av Wind
(1.97) (-0.75) (-0.71)
-4.400 Sat - 42.125 Sun (R2 = 0.631)
(-2.22) (-22.41)
The variables are as before except for the fact that the air pollution variable
was based on readings monitored at D.C. General Hospital. One interpretation
of this equation is that an increase of 1.0 ppm in the maximum 1-hour average
carbon monoxide level (raising the mean from 6.86 ppm to 7.86 ppm) was related
to an increase of 0.31 daily unscheduled visits to the pediatric department
(raising the mean number of unscheduled visits from 36.03 to 36.34). This
can also be stated in percentage terms: a 10 percent decrease in the mean of
the air pollution variable would be associated with a 0.58 percent decrease in
the number of unscheduled visits to the department.
Carbon monoxide data were also available for 1974 from the monitoring
station located at D.C. General Hospital. The results of statistical analyses
using these data in conjunction with 1974 visitation and climatic data failed
to uncover any statistically significant associations between carbon monoxide
levels and unscheduled department visits (Table 8.14).
Following these results, we examined both the 1973 and 1974 carbon monoxide
data in conjunction with the health data in an attempt to uncover lag or epi-
sodic effects of air pollution (not reported). As was the case for photo-
chemical oxidants, no consistent lag or episodic effects were exhibited.
Sulfur Dioxide
The last air pollutant for which we had data was sulfur dioxide. As noted
in Section VI, this air pollutant is primarily attributed to stationary sources.
Notwithstanding, since the air pollution-health literature is dominated by
studies involving stationary-source pollutants such as sulfur dioxide, 17/ we
felt it would be useful to examine our data for possible associations. We
recognized at the outset that the sulfur dioxide levels experienced in the
Washington Metropolitan Area are substantially below those in most other metro-
politan areas. Hence, we were somewhat surprised that our findings indicated
possible associations between levels of sulfur dioxide and unscheduled visits
to the internal medicine and ophthalmology departments.
Specifically, two monitoring stations (D.C. General Hospital and American
Chemical Society) provided sufficient sulfur dioxide data to permit statistical
analyses (see Table 6.1). In addition, each station reported both the 24-hour
16/ Note that the results using the CAMP station carbon monoxide data
indicated a positive, statistically insignificant association with
unscheduled pediatric visits (see regression 5, Table 8.13).
17/ See, for example, Hodgson [24] and Glasser and Greenburg [16].
62
-------
TABLE 8.14. REGRESSION ANALYSIS OF CARBON MONOXIDE
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1974)
UIM
1
UOPHTH
2
UPED
3
UVC
4
Constant
0.256 a/
10.914
0.235
4.991
0.476
66.326
0.306
145.649
Air Pollution:
CO (D.C.) -0.065 b/
(-0.70) c/
Weather:
Av T 0.024
(0.96)
Av Wind -0.174
(-1.35)
Dummy :
Sat -6.037
(-5.56)
Sun -10.050
(-9.79)
0.0002
(0.01)
0.019
(0.99)
0.126
(1.27)
-4.852
(-5.82)
-7.023
(-8.92)
-0.025
(-0.09)
-0.364
(-4.74)
0.892
(2.25)
-7.599
(-2.27)
-51.597
(-16.35)
0.037
(0.07)
0.268
(1.90)
-0.048
(-0.07)
-31.233
(-5.17)
-64.436
(-11.36)
a/ The coefficient of determination; a value of 0.256 indicates that 25.6
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
63
-------
average reading and the maximum 1-hour reading. Our empirical results
indicated that the latter measure was more important in terms of signifi-
cant associations with the health data. 18/ Hence, Table 8.15 presents
the results of separate regressions using 1973 sulfur dioxide levels
(as measured by maximum 1-hour readings) from each station for each
department.
As can be seen from the table, none of the associations between
unscheduled department visits and sulfur dioxide levels was significant
for readings taken from both stations. However, positive and statis-
tically significant associations were exhibited for unscheduled visits
to the internal medicine department (regression 1) and for unscheduled
visits to the ophthalmology department (regression 4). The relevant
equations are reproduced below:
UIM = 14.222 + 19.502 SO- + 0.012 Av T + 0.147 Av Wind
(1.78) (0.56) (1-35)
-9.353 Sat - 16.811 Sun „
(-10.50) (-19.19) (R = 0.583)
UOPHTH = 6.166 + 10.680 SO, + 0.017 Av T + 0.012 Av Wind
(1.65) (1.12) (0.17)
-4.508 Sat - 7.875 Sun .
(-8.08) (-14.45) (R - 0.452)
The variables are similar to those in the previous specifications,
although S02 (the maximum 1-hour average sulfur dioxide reading on a
given day) -is the air pollution variable. One interpretation of the first
equation is that an increase of 0.01 ppm in the maximum 1-hour average
sulfur dioxide level (raising the mean from 0.041 to 0.051 ppm) was
related to an increase of 0.20 daily unscheduled visits to the internal
medicine department (raising the mean number of unscheduled visits from
13.02 to 13.22). An alternative interpretation of this result is that a
10 percent decrease in the mean of the air pollution variable would be
associated with a 1.5 percent decrease in the number of unscheduled
visits to the department.
With respect to the ophthalmology department, one interpretation of
the second equation is that an increase of 0.01 ppm in the sulfur dioxide
level was related to an increase of 0.11 daily unscheduled visits to the
ophthalmology department (raising the mean number of unscheduled visits
from 6.06 to 6,17). In percentage terms this result signifies that a 10
percent decrease in the mean of the air pollution variable would be
associated with a 0.4 percent decrease in the number of unscheduled
visits to the department.
Maximum 1-hour readings for sulfur dioxide were available for only one
station in 1974 (see Table 6.1). Analyses of these data in conjunction
with the 1974 utilization data failed to replicate the positive associations
18/ This is a somewhat interesting result since the national primary
standard for sulfur dioxide is based on a 24-hour average rather than
a maximum 1-hour average.
64
-------
TABLE 8.15. REGRESSION ANALYSIS OF SULFUR DIOXIDK
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATUV
(PENNSYLVANIA AVENUE - 1973)
R2
Constant
Air Pollution:
S02 (D.C.)
S02 (ACS)
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
1
0.583
14.222
19.502
(1.78)
0.012
(0.56)
0.147
(1.35)
-9.353
(-10.50)
-16.811
(-19.19)
U1M
2
a/ 0.585
15.153
b/
£/
7.593
(0.72)
0.007
(0.29)
0.116
(1.03)
-9.682
(-10.62)
-16.937
(-19.03)
UOPHTH
3
0.466
8.230
0.781
(0.12)
0.004
(0.30)
-0.032
(-0.48)
-4.763
(-8.71)
-8.216
(-15.29)
4
0.452
6.166
10.680
(1.65)
0.017
(1.12)
0.012
(0.17)
-4.508
(-8.08)
-7.873
(-14.45)
UPED
5
0.626
48.399
18.986
(0.81)
-0.049
(-1.07)
-0.331
(-1.42)
-4.150
(-2.18)
-42.383
(-22.68)
6
0.643
46.208
23.058
(1.07)
-0.024
(-0.47)
-0.317
(-1.38)
-5.152
(-2.76)
-42.157
(-23.14)
a/ The coefficient of determination; a value of 0.583 indicates that 58.3 percent of
the variation in unscheduled visits was "explained" by the independent variables.
b/ The regression coefficient.
c/ The t-statistic; a value of 1.65 indicates significance at the 10 percent level,
using a two-tailed test.
65
-------
exhibited in Table 8.15 for the 1973 data (Table 8.16). We also examined
both the 1973 and 1974 data for evidence of lag or episodic effects
involving sulfur dioxide data, but none were discovered (not reported).
SYNERGISTIC EFFECTS
There have been a number of studies suggesting the importance of
interactions between air pollutants and their combined effects on health.
Particular reference has been made to possible synergistic effects of
ozone in combination with nitrogen oxides 19/, and oxidants in combination
with sulfur dioxide. 20/ Consequently, in an effort to investigate our
data for possible synergistic effects, we regressed the variables repre-
senting unscheduled utilization of the various departments on variables
representing the possible interactions between photochemical oxidants
and nitrogen dioxide, and between photochemical oxidants and sulfur
dioxide. In addition, we examined the interaction between photochemical
oxidants and carbon monoxide. 21/
The empirical analyses involving interaction terms for photochemical
oxidants and nitrogen dioxide failed to uncover any evidence of synergism
(Table 8.17). When the interaction term was substituted in the previous
specifications, its coefficient failed to attain statistical significance.
The results for the interaction between photochemical oxidants and sulfur
dioxide were generally similar (Table 8.18). However, in this case, the
interaction term was statistically significant in explaining the variation
in 1974 unscheduled visits to the ophthalmology department (regression 4).22/
Nevertheless, upon comparing the estimated effect represented by the
interaction term with the estimated effects corresponding to each of these
air pollution variables in previous regressions, there was no indication
of a synergistic relationship. That is, the magnitude of the association
between the interaction variable (Ox x S02) and unscheduled ophthalmologic
visits was less than the sum of the associations between unscheduled
ophthalmologic visits and both photochemical oxidant levels and sulfur
dioxide levels (regression 4, Table 8.6 and regression 2, Table 8.16,
respectively). Finally, the results for the interaction between photo-
chemical oxidants and carbon monoxide are presented in Table 8.19. Again,
despite positive and statistically significant coefficients for the air
pollution interaction term in regressions 4 and 7, the magnitude of the
estimated effects did not indicate any evidence of synergism.
19/ See Thorp [38].
2_0/ See Amdur [3].
21/ In notational terms the interactions were equal to the product of the
photochemical oxidant reading and the nitrogen dioxide reading, Ox x N0~;
the product of the photochemical oxidant reading and the sulfur dioxide
reading, Ox x S02J and the product of the photochemical oxidant reading
and the carbon monoxide reading, Ox x CO.
22 / This might be expected since the previous results indicated significant
associations between unscheduled visits to the ophthalmology department
and both photochemical oxidant levels and sulfur dioxide levels
(measured at the American Chemical Society).
66
-------
TABLE 8.16. REGRESSION ANALYSIS OF SULFUR DIOXIDE
EFFECTS ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1974)
UIM
1
UOPHTH
2
UPED
3
uvc
4
Constant
0.276 a/
15.038
0.234
3.844
0.472
75.379
0.292
149.633
Air Pollution:
S02 (CAMP)
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
-29.222 b/
(-2.17) £/
-0.024
(-0.87)
-0.217
(-1.62)
-6.284
(-5.72)
-10.329
(-9.88)
1.511
(0.14)
0.033
(1.56)
0.015
(1.43)
-5.009
(-5.88)
-6.974
(-8.54)
-54.658
(-1.25)
-0.467
(-5.33)
0.960
(2.21)
-9.397
(-2.66)
-51.772
(-15.32)
-19.604
(-0.27)
0.246
(1.66)
-0.159
(-0.22)
-34.847
(-5.83)
-60.162
(-10.50)
a/ The coefficient of determination; a value of 0.276 indicates that 27.6
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b_/ The regression coefficient.
£/ The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
67
-------
TABLE 8.17. SYNERGISTIC EFFECTS OF OXIDANTS AND NITROGEN
DIOXIDE ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973)
UIM
1
UOPHTH
2
UPED
3
Constant
0.629 a/
18.314
0.445
8.198
0.632
49.565
Air Pollution:
Ox x NO-
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
-204.237 b/
(-1.12) £/
-0.003
(-0.14)
0.008
(0.07)
-9.823
(-11.36)
17.748
(-21.49)
168.168
(1.40)
-0.007
(-0.52)
-0.016
(-0.24)
-4.319
(-7.58)
-7.992
(-14.70)
-447.909
(-1.10)
-0.018
(-0.38)
-0.491
(-2.13)
-5.019
(-2.60)
-42.535
(-23.03)
a/ The coefficient of determination; a value of 0.629 indicates that 62.9
percent of the variation in unscheduled visits was "explained" by the
independent variables.
b/ The regression coefficient.
cj The t-statistic; a value of 1.65 indicates significance at the 10 percent
level, using a two-tailed test.
68
-------
TABLE 8.18. SYNERGISTIC EFFECTS OF OXIDANTS AND SULFUR
DIOXIDE ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973 and 1974)*
DIM
1 2
UOPHTH UPED UVC
34567
0.636 a/ 0.276
0.450
0.257
0.641
0.470
0.293
Constant
18.034 12.746 7.775 2.706 52.409 71.133 149.154
Air Pollution:
Ox x S02 (ACS)
Ox x S02 (CAMP)
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
146.976
(-1.21)
-0.001
(-0.06)
0.018
(0.18)
b/
£/
-480
(-1
0
(0
-0
(-1
.82
.76)
.004
.16)
.185
.41)
105.
(1.
0.
(0.
-0.
(-0.
900
33)
000
01)
008
13)
397
(1
0
(1
0
(2
.534
.99)
.029
.54)
.233
.43)
-69
(-0
-0
(-1
-0
(-2
.059
.25)
.077
.81)
.488
.13)
390.
(0.
-0.
(-5.
1.
(2.
013
44)
456
23)
129
64)
819.820
(0.55)
0.229
(1.60)
-0.111
(-0.15)
-9.777 -6.233 -4.444 -4.780 -3.499 -9.073 -35.214
(-10.82) (-5.67) (-7.49) (-5.95) (-1.73) (-2.54) (-5.85)
-17.8 -10.303 -7.892 -6.897 -42.570 -52.361 -60.703
(-21.39) (-9.67) (-14.44) (-8.85) (-22.77) (-15.10) (-10.39)
*Regressions 1, 3, and 5 are based on 1973 data and regressions 2, 4, 6 and 7 are based
on 1974 data.
a/ The coefficient of determination; a value of 0.636 indicates that 63.6 percent of the
variation in unscheduled visits was "explained" by the independent variables.
b_/ The regression coefficient.
c_/ The t-statistic; a value of 1.65 indicates significance at the 10 percent level, usinR a
two-tailed test.
69
-------
TABLE 8.19. SYNERGISTIC EFFECTS OF OXIDANTS AND CARBON
MONOXIDE ON UNSCHEDULED DEPARTMENT UTILIZATION
(PENNSYLVANIA AVENUE - 1973 and 1974)*
R2
Constant
Air Pollution:
Ox x CO
Weather:
Av T
Av Wind
Dummy:
Sat
Sun
0
17
-0
(-1
-0
(-0
0
(0
-10
(-11
-17
(-20
1
.634
.865
.838
•52)
.004
.18)
.056
.52)
.231
.09)
.926
.23)
UIM
a/ 0.
3.
b/ 1.
£/ (0.
-3.
(-2.
0.
(3.
-0.
(-0.
0.
(6.
UOPHTH
2
192
182
161
69)
223
83)
252
35)
019
63)
132
25)
3
0.447
7.665
0.285
(0.80)
0.003
(0.22)
0.001
(0.01)
-4.426
(-7.39)
-7.865
(-13.6?)
4
0.197
1.291
2.995
(2.46)
-3.272
(-3.99)
0.004
(0.07)
0.062
(2.92)
0.057
(3.76)
0.
51.
-0.
(-0.
-0.
(-1-
-0.
(-1.
-5.
(-2.
-42.
(-21.
UPED
5
632
140
124
10)
065
53)
440
86)
498
71)
346
73)
6
0.659
6.003
-1.359
(-0.34)
2.132
(0.78)
0.399
(2.21)
0.912
(13.08)
0.391
(7.73)
uvc
0.
104.
20.
(2.
-17.
(-2.
0.
(2.
0.
(4.
0.
(4.
7
292
409
359
23)
789
89)
818
01)
697
42)
567
96)
*Regressions 1, 3 and 5 are based on 1973 data and regressions 2, 4, 6, and 7 are based
on 1974 data.
a/ The coefficient of determination; a value of 0.634 indicates that 63.4 percent of the
variation in unscheduled visits was "explained" by the independent variables.
b_/ The regression coefficient.
c/ The t-statistic; a value of 1.65 indicates significance at the 10 percent level, using
a two-tailed test.
70
-------
THE EFFECTS OF OTHER METEOROLOGICAL VARIABLES
Thus far we have reported results using only two meteorological variables,
average daily temperature and average daily wind speed. Here, we investigate
the effects of two additional climatic variables, daily precipitation and
daily temperature change. These two variables were selected on the basis of
findings from other studies. For example, precipitation has been included in
a study of emergency hospital admissions to account for the fact that rain (or
snow) might render transportation by foot or automobile more hazardous and
consequently cause accidents. 23/ We feel this factor has only limited appli-
cability to our analysis; however, for completeness, we felt it appropriate to
examine the effects of including a precipitation variable. Similarly, mention
has been made in some studies that various illnesses such as asthma can be
aggravated by sudden temperature changes; _2*V hence, we defined a temperature
change variable as the difference between the maximum daily temperature and
minimum daily temperature and explored the effects of including this variable
in our statistical analyses.
In general, the inclusion and substitution of the various climatic vari-
ables did not greatly affect the magnitude and significance of the coefficients
of the air pollution variables. The only noteworthy exception indicated that
substitution of the temperature change variable for the average temperature
variable had a moderate effect on the importance of certain air pollution
variables. This was not unexpected, since the air pollution variables were,
in general, more highly correlated with average temperature than with tempera-
ture change (see Appendix E).
As with the previous results, the additional weather variables seldom
exhibited statistically significant associations with the unscheduled visita-
tion data from the various departments. However, some generalisations can be
stated. Without question, the climatic variable that most often was statisti-
cally sis'nificant was the temperature change variable. At those times when
the variable was significant in explaining unscheduled visits, the signs of
its coefficient indicated that the greater the difference between the minimum
and maximum daily temperature the more unscheduled visits occurred. This
result was particularly true for visits to the pediatric department and to a
lesser extent for visits to the internal medicine department. 25/
THE EFFECTS ON METRO TRANSIT EMPLOYEES
As discussed in Section IV, approximately 15 percent of the membership
of the Group Health Association is comprised of Washington metropolitan
transit system employees. We felt that an examination of this subset of GHA
23/ See Silverman [34].
24/ See [8].
25/ To put this result into perspective we note that a 10 percent increase in
the temperature difference variable (raising the mean from about 18 to
about 20 degrees) would be related to an increase of between 0.3 and 0.4
percent in the mean number of unscheduled pediatric visits. This can be
contrasted with the elasticities corresponding to the significant associ-
ations with the air pollution variables in Table 2.1.
71
-------
members would be of particular interest. This was based on the presumption
that many of the Metro employees would be likely to be exposed to significantly
higher doses (or more prolonged exposure) to mobile-source air pollution in
their occupations as bus drivers or repair workers than other members of GHA.
Consequently, we replicated most of the analyses already discussed for this
group. In doing so, however, we found it necessary to aggregate unscheduled
utilization and scheduled utilization. 26/
In general, the results based on the Metro sample did not display signifi-
cant associations between monitored levels of the various air pollutants and
visits to the clinic departments. 27/ In fact, the only consistently significant
associations were exhibited between visits by Metro employees to the ophthal-
mology department and photochemical oxidant levels during 1974.
The following two equations present these results:
METRO-OPHTH = 1.331 + 9.526 Ox + 0.006 Av T + 0.038 Av Wind
(2.00) (0.79) (1.34)
-1.771 Sat - 2.454 Sun (R2 = 0.299)
(-7.14) (-10.39)
METRO-OPHTH = 1.595 + 6.012 Ox + 0.005 Av T + 0.031 Av Wind
(1.59) (0.59) (1.06)
-1.778 Sat - 2.428 Sun (R2 = 0.293)
(-6.97) (-10.13)
The top equation is based on 1974 photochemical oxidant data monitored at the
CAMP station, while the bottom equation is based on 1974 photochemical oxidant
data monitored at the Cleveland Park Public Library (see Table 6.1). In other
respects, these equations are similar to previous specifications except that
the dependent variable is now total department visits by the Metro sample
rather than unscheduled department visits by the total sample.
One interpretation of these results is that a 10 percent decrease in the
mean of the relevant air pollution variables would be associated with between
a 2.0 and a 1.5 percent decrease in the number of visits by Metro employees to
the ophthalmology department. Our conclusion from this analysis is that, in
general, the findings for the Metro sample were not at significant variance
from the findings for the total sample. Given the limited data and the afore-
mentioned difficulties associated with statistical analyses of small samples,
we caution against overinterpretation of these results.
26/ This was necessary since the sample size was greatly reduced resulting
in relatively large sampling variation. Large sampling variation greatly
impedes attempts to uncover significant statistical associations in such
data. See Lave and Seskin [28].
211 We concentrated primarily on visits to the internal medicine and ophthal-
mology departments, since pediatric visits were not thought to be relevant
for this subsample.
72
-------
THE FINDINGS FOR ANOTHER HEALTH CLINIC
As discussed in Section V, in addition to the data for the main GHA
facility in downtown Washington, we had data from a smaller suburban
clinic located in Takoma Park (see Tables 5.2 and 5.4). In this section
we will report on the empirical analyses pertaining to these data.
Emphasis will be placed on the differences between the Takoma Park
results and the findings for the Pennsylvania Avenue clinic previously
discussed.
Our "strongest" results involve the association between unscheduled
visits to the ophthalmology department and lvi-.-uls of photochemical
oxidants. Unfortunately, as mentioned in Section V, the Takoma Park clinic
does not run an ophthalmology department; hence, we could not examine this
association at that facility. With regard to our other findings, the
general statement can be made that the associations between unscheduled
utilization at Takoma Park and levels of air pollution were weaker than
the associations seen in using the data from the main Pennsylvania Avenue
facility.
At least two possible explanations for this fact come to mind. The
first involves difficulties that may be associated with analyzing the
relatively small samples represented by department utilization at Takoma
Park (see Appendix B). As discussed in the preceding section, small
samples and the accompanying large sampling variation raise statistical
problems in analyses of this nature.
Another possible explanation for the "poor" results from using the
Takoma Park data concerns the fact that both air pollution exposures and
the relevant population group may exhibit different characteristics than
the data represented by the analysis involving the Pennsylvania Avenue
Clinic. This hypothesis is more difficult to assess.
73
-------
SECTION IX
DISCUSSION
POLICY IMPLICATIONS
The lack of information on the effects of mobile-source air pollutants,
especially with regard to health, has made it difficult to evaluate the
potential benefits ^including hc-alth benefits) from abatement of air pollu-
tion attributed to mobile sources. !_/ Given that the annual cost of control-
ling mobile-source emissions from automobiles has been estimated to be as
high as $11 billion by 1985, this is a serious shortcoming. _2/
The only association we found to be consistent and statistically
significant for data from both 1973 and 1974 was between daily unscheduled
visits to the ophthalmology department at GHA and levels of photochemical
oxidant pollution. Even here, 1974 air pollution data from a second moni-
toring station did not confirm the relationship. In addition, a relationship
between urgent clinic visits and photochemical oxidant levels during 1974
(both stations) was noted with interest. Isolated positive and significant
associations were also found between photochemical oxidant levels and
unscheduled pediatric visits, carbon monoxide levels and both unscheduled
ophthalmologic and unscheduled pediatric visits, and sulfur dioxide levels
and both unscheduled internal medicine and unscheduled ophthalmologic
visits. What can we say about these results?
The association between mobile-source air pollutants and eye irritation
has been well documented. 3/ What is of particular interest here is the
fact that the levels of photochemical oxidant pollution in our data were
considerably below most "threshold" levels noted in the literature. 4/
Unfortunately, the specific complaint associated with the urgent clinic
visits cannot be determined from our data. Hence, we have no way of examin-
ing the association between specific ailments and photochemical oxidant
levels in light of the literature. Nevertheless, we should note the associa-
tion, although statistically significant, was of smaller magnitude than the
association with eye problems (see Table 2.1). That is, even if the associa-
tion proved to be causal in nature, it is questionable as to how serious the
policy implications (or economic consequences) would be. Finally, despite
the fact that positive and significant associations were noted between levels
of other air pollutants (carbon monoxide and sulfur dioxide) and unscheduled
department visits (see above), these results were mixed. Hence, we Relieve
that it would be overinterpreting the data to draw policy conclusions from
the isolated findings pertaining to these air pollutants.
While investigators may later find a strong association between ill health
and automobile emissions (as represented by ambient levels of carbon monoxide,
nitrogen oxides, hydrocarbons, and photochemical oxidants), current evidence is
somewhat to the contrary. This is in contrast to the relatively large number
of studies that document significant associations between ill health and
ambient levels of such stationary-source air pollutants as suspended particu-
lates and sulfates. In general, the results from this study pertaining to the
Washington Metropolitan Area support this difference.
_!/ See National Academy of Sciences [31'.
2/ Ibid., p. 12.
3/ See Hammer et al. [201-
47 Ibid., p. 257. 74
-------
ECONOMIC CONSEQUENCES
Despite our very limited findings, we will make a crude calcula-
tion of the economic consequences pertaining to the relationship
between oxidant levels and ophthalmologic visits. First we will deter-
mine the percent reduction in oxidant levels that would be necessary
in order that the national standard of 0.08 ppm (maximum 1-hour
average) not be exceeded. From our data, we estimate the necessary
reductions to be 55.6 percent for 1973 oxidant levels and 42.9 percent for
1974 oxidant levels. Thus, if oxidant levels were reduced by about 50
percent in the Washington D.C. Metropolitan Area, the air quality in the
region would probably be in compliance with the national ambient air
quality standard for photochemical oxidants. Using the estimated reduc-
tions, we can then apply them to the elasticities computed pertaining to
the oxidant-ophthalmology association (see Table 2.1). The results
indicate that the above reductions in oxidant levels would correspond to
a 6.1 percent reduction in unscheduled ophthalmologic visits during 1973
and an 18.4 percent reduction during 1974. This, in turn, represents
approximately 136 unscheduled visits to GHA in 1973 and 367 unscheduled
visits to GHA in 1974.
The second step involves attaching a monetary value to these visits.
To estimate the direct medical costs, we will assign the value of
$20.00 per visit. We have no specific cost data but that value is
representative of the average medical costs associated with visits for
simple eye problems in Maryland [22 1 . This translates into direct medical
costs of approximately $2700 in 1973 and $7300 in 1974.
However, the direct medical costs calculated above represent only
one component of the benefits that would be obtained from abating oxidant
air pollution if a causal relationship exists between oxidant levels and
unscheduled ophthalmologic visits. Additional indirect costs would be
associated with lost worker productivity and restricted activity.
Unfortunately, it is difficult to estimate these indirect costs as they
might relate to ophthalmologic problems. Nevertheless, Cooper and Rice
[121 recently estimated the economic cost of illness in the United States.
They found that for diseases of the nervous system and sense organs indirect
costs (including losses due to illness of homemakers who cannot perform
their housekeeping duties) were approximately 80 percent as great as direct
medical costs. Hence, using this proportion, we can obtain crude estimates
of the indirect costs associated with reduced ophthalmologic problems.
Specifically, the indirect costs were calculated to be $2160 for 1973 and
$5840 for 1974.
Finally, there are additional costs associated with pain and suffering.
However, no one has successfully quantified this dimension of illness.
Rather than assign completely arbitrary numbers to this category, we prefer
to conclude that the sum of the direct and indirect costs estimated
above represent an underestimate of the "true" costs associated with
reduced ophthalmologic visits. Thus, $4860 in 1973 and $13,140 in
1974 represent underestimates of the "true" benefits of reducing
ophthalmologic visits that would be obtained from the assumed reductions
in photochemical oxidant levels (if the relationship between oxidant
levels and ophthalmologic visits were a causal one).
75
-------
To extrapolate these benefit estimates to the entire Metropolitan
Washington Area is presumably stretching the data beyond its limits.
We know, for example, that the population embodied in the Group Health
membership is not representative of the Metropolitan Washington Area
population. J5/ Furthermore, it is difficult to determine the direction
of bias this introduces. Nevertheless, with these caveats in mind, for
illustrative purposes we will assume that the Group Health members are
characteristic of the metropolitan population. This would imply that
the benefits relating to decreased ophthalmologic problems from the
assumed reductions in photochemical oxidant levels could be as great
as twenty (the ratio of the Metropolitan Washington Area population
to the GHA membership) times the monetary estimates derived above.
That is, the 1973 benefit estimate could be as high as §97,200 and
the 1974 benefit estimate could be as high as $262,800.£/
The benefit estimates should not be taken out of context. To
put these numbers in perspective one must remember that there are costs
associated with reducing photochemical oxidant levels. The primary
method of abating photochemical oxidants in the Washington Metropolitan
Area is by policies designed to control emissions from automobiles. A
conservative estimate of the annual costs of emission-control devices
on automobiles is $100 per auto [321. The automobile population for
our study area is approximately 700,000. This translates into costs of
about $70 million per year. Clearly, the Metropolitan Washington popu-
lation is incurring substantial costs for the control of mobile-source
emissions. Although benefits in addition to reduced ophthalmologic
problems may accrue from such control policies, our limited findings
cannot be used to justify these large expenditures.
Thus, our findings along with other existing evidence, suggest
that mobile-source air pollution exhibits to a limited degree acute health
effects. Evidence supporting chronic health effects from mobile-source
air pollutants is virtually non-existent. It is possible that, in certain
areas of the country, the relationship between the levels of mobile-source
air pollutants and acute effects such as eye irritation warrant stringent
regional emission control policies. However, the adoption of stringent
national emission control policies is questionable. Additional information
on the health effects of automobile emissions may, of course, alter these
conclusions. Furthermore, if other effects (e.g., aesthetic effects)
associated with mobile-source air pollution are found to be of substantial
magnitude and importance, policymakers may justify the control of mobile-
source emissions on that basis alone.
FUTURE RESEARCH
The nature of the data available for statistical analyses has suggested
a number of future research needs. Perhaps the most serious deficiency in
investigating the air pollution-health relationship is that of good air
quality data. As noted in Section III, air quality measurements taken at
j>/ See Section IV, especially p. 18.
6/ Note, these benefit estimates do not include possible improvements in
~~ other health areas or benefits that might be associated with improved
visibility, aesthetic effects, and so on.
76
-------
a single sampling station must often be assumed to be representative of a
large geographical area. Furthermore, often inadequate equipment takes
infrequent readings of relatively few air pollutants. 7/ If we are to
provide better answers as to how much specific air pollutants should be
abated, better air quality data are required.
While one category of future needs involves obtaining better measure-
ments of the quality of air at a number of sites, another involves measuring
the quality of the air actually breathed by individuals. Thus, a personal,
portable device that would enable one to measure the dose of each air pol-
lutant the individual actually experienced would be invaluable in examining
the air pollution-health association.
Thus far we have confined the discussion to the air pollution exposure
data. However, the particular measure of health status is itself an important
concern. As noted in Section III, morbidity data such as those used in this
study should be more sensitive indicators of air pollution effects than
mortality data. However, comprehensive morbidity data are seldom available.
A possible approach to this problem might involve constructing a panel, the
members of which are monitored closely for changes in their health status.
Variations in the morbidity rate over time for each individual (as a function
of changes in air pollution exposure) would provide an almost ideal measure
of the air pollution-health association. 8/ While such data are likely to
be expensive to gather, they seem to be a prerequisite for sorting out the
health effects of various air pollutants.
Needless to say, a great deal of work remains to be done in establishing
the relationship between air pollution and human health. For epidemiological
investigations of the sort reported in this study, one key to further know-
ledge is better data.
T_l Many references have been made throughout this study to the inadequacy
of many of the air pollution data series. In addition, for a given air
pollutant, there was some evidence that conflicting results may have
been related to differences in the specific type of monitoring equipment
in place.
87 See Speizer [36] .
77
-------
SECTION X
REFERENCES
1. "Air Quality in the National Capital Interstate Air Quality Region - 1972,"
Technical Report No. 2, Addendum, Metropolitan Washington Council of Gov-
ernments, September 1973.
2. "Air Quality in the National Capital Interstate Air Quality Region - 1974,"
Technical Report No. 9, Metropolitan Washington Council of Governments,
August 1975.
3. Amdur, M. "Toxicological Guidelines for Research on Sulfur Oxides and
Particulates," Proceedings of the Fourth Symposium on Statistics and
the Environment, forthcoming.
4. Anderson, T.W. An Introduction to Multivariate Statistical Analysis (New
York: John Wiley & Sons, 1958).
5. Annual jleport on the Quality of the Air in Washington, D.C. 1973, Govern-
ment of the District of Columbia, Department of Environmental Services,
Environmental Health Administration, Bureau of Air & Water Quality, Air &
Water Monitoring Division.
6. Annual Report on the Quality of the Air in Washington, D.C. 1974, Govern-
ment of the District of Columbia, Department of Environmental Services,
Environmental Health Administration, Bureau of Air & Water Quality, Air &
Water Monitoring Division.
7. Ashe, W.F. "Exposure to High Concentrations of Air Pollution (1) - Health
Effects of Acute Episodes," Proceedings of the National Conference on Air
Pollution, Washington, D.C., Nov. 18-20, 1958, PHS Publ. No. 654, 188 (1959)
8. Berke, J., and V. Wilson. Watch Out for the Weather (New York: Viking,
1951); and Climate and Man. 1941 Yearbook of Agriculture, U.S. Department
of Agriculture Publication (1941).
9. CederTdf, R. "Urban Factor and Prevalence of Respiratory Symptoms and
Angina Pectoris," Archives of Environmental Health 13, 743 (1966).
10. The Chesapeake & Potomac Telephone Company, Yellow Pages, 1976.
11. Cohen, C.A. et al. "Carbon Monoxide and Survival from Myocardial Infarc-
tion," Archives of Environmental Health 19, 510 (1969).
12. Cooper, B.S., and D.P. Rice. "The Economic Cost of Illness Revisited."
Washington, D.C., U.S. Government Printing Office, Social Security
Bulletin, Social Security Administration, DHEW Publication No. (SSA)
76-11703, February 1976.
13. Daly, C. "Air Pollution and Causes of Death," British Journal of Preventive
and Social Medicine 13, 14 (1959).
78
-------
14. GHA News. Annual Report Issue, May 2, 1974, Vol. XXXVII, No. 3, Group
Health Association, Inc.
15. GHA News, Annual Report Issue, April-May, 1975, Vol. XXXVII, No. 2,
Group Health Association, Inc.
16. Glasser, M., and L. Greenburg. "Air Pollution, Mortality, and Weather,"
Archives of Environmental Health 22, 334 (1971).
17. Goldberger, A. Econometric Theory (New York; John Wiley & Sons, 1964).
18. Greenburg, L. et al. "Air Pollution and Morbidity in New York City,"
Journal of the American Medical Association 182, 161 (1962).
19. Greenburg, L. et al. "Report of an Air Pollution Incident in New York
City, November 1953," Conference Reports, Public Health Reports 77, 7 (1962),
20. Hammer, D.I. gg.jal. "Los Angeles Student Nurse Study," Archives of En-
vironmental Health 28, 255 (1974).
21. Hammond, E.C., and D. Horn. "Smoking and Death Rates—Report on Forty-four
Months of Follow-up of 187,783 Men. II. Death Rates by Cause," Journal of
the American Medical Association 166, 1294 (1958).
22. Hatem, T.J, "Medical Costs in the State of Maryland." A Report
of the Insurance Commissioner.
23. Heimann, H. "Episodic Air Pollution in Metropolitan Boston -- A Trial
Epidemiological Study," Archives of Environmental Health 20, 230 (1970).
24. Hodgson, T.A. "Short-term Effects of Air Pollution on Mortality in New
York City," Environmental Science and Technology 4, 589 (1970).
25. Jaksch, J.A., and H..H. Stoevener. "Outpatient Medical Costs Related to
Air Pollution in the Portland, Oregon Area." Washington, D.C., U.S.
Government Printing Office, Environmental Protection Agency Publication
No. 600/5-74-017, 1974.
26. Johnston, J. Econometric Methods 2nd Edition (New York; McGraw Hill, 1972).
27. Kendall, M.G. A Course in Multivariate Analysis (New York: Hafner, 1968).
28. Lave, L.B., and E.P. Seskin. Air Pollution and Human Health, forthcoming
for Resources for the Future, Inc., Johns Hopkins University Press.
29. Lave, L.B., and E.P. Seskin. "An Analysis of the Association Between U.S.
Mortality and Air Pollution," Journal of the American Statistical Associa-
tion 68, 284 (1973).
30. McCarroll, J. "Measurements of Morbidity and Mortality Related to Air Pol-
lution," Journal of the Air Pollution Control Association 17, 203 (1967).
79
-------
31. National Academy of Sciences, Air Quality and Automobile Emission Con-
trol, Volume 4. The Costs and Benefits of Automobile Emission Control,
A report by the Coordinating Committee on Air Quality Studies, National
Academy of Sciences, National Academy of Engineering, Serial No. 93-24,
Washington, B.C.: U.S. Government Printing Office, 1974.
32. National Academy of Sciences, Report by the Committee on Motor Vehicle
Emissions. Washington, D.C.: National Academy of Sciences,
February 12, 1973.
33. Pearlman, M.E. et al. "Nitrogen Dioxide and Lower Respiratory Illness,"
Pediatrics 47, 391 (1971).
34. Silverman, L.P. "The Determinants of Daily Emergency Admissions to Hos-
pitals," Public Research Institute, Center for Naval Analyses, 1973.
35. Simon, H. "Spurious Correlation: A Causal Interpretation," Journal of
the American Statistical Association 49, 467 (1954).
36. Speizer, F. "An Epidemiological Approach to Health Effects of Air Pollu-
tants," Proceedings of the Fourth Symposium on Statistics and the Environ-
ment, forthcoming.
37. Stocks, P. British Medical Journal. 1, 74 (1959).
38. Thorp, C.E. "influence of Nitrogen Oxides on the Toxicity of Ozone,"
Journal of the American Chemical Society (News Edition) 19, 686 (1941).
39. U.S.D.C. National Oceanic and Atmospheric Administration Environmental
Service, Local Climatological Data Annual Summary of Comparative Data,
Washington, D.C., National Airport.
40. Winkelstein, W., Jr., and M.L. Gay. "Suspended Particulate Air Pollu-
tion," Ar^]ii2ej_oJLJLnyJjro^^ 22, 174 (1971).
41. Winkelstein, W. , Jr., and S. Kantor. "Prostatic Cancer: Relationship
to Suspended Particulate Air Pollution," American Journal of Public
Health 59, 1134 (1969).
42. Winkelstein, W., Jr., and S. Kantor. "Respiratory Symptoms and Air
Pollution in an Urban Population of Northeastern United States,"
Archives of Environmental Health 18, 760 (1969).
43. Winkelstein, W. , Jr., and S. Kantor. "Stomach Cancer: Positive Asso-
ciation with Suspended Particulate Air Pollution," Archives of Environ-
mental Health 18, 544 (1969).
44. Winkelstein, W. , Jr. et al. "The Relationship of Air Pollution and
Economic Status to Total Mortality and Selected Respiratory System
Mortality in Men, I: Suspended Particulates," Archives of Environmen-
tal Health 14, 162 (1967).
45. Winkelstein, W., Jr. et al. "The Relationship of Air Pollution and
Economic Status to Total Mortality and Selected Respiratory System
Mortality in Men, II: Suspended Particulates." Archives of Environmen-
tal Health 16, 401 (1968).
80
-------
46. Zeidberg, L.D., R.J.M. Horton, and E. Landau. "The Nashville Air Pollu-
tion Study, V: Mortality From Diseases of the Respiratory System in
Relation to Air Pollution," Archives of Environmental Health 15, 214
(1967).
47. Zeidberg, L.D., R.J.M. Horton, and E. Landau. "The Nashville Air
Pollution Study, VI: Cardiovascular Disease Mortality in Relation to
Air Pollution, Archives of Environmental Health 15, 225 (1967).
48. Zeidberg, L.D., and R.A. Prindle.. "The Nashville Air Pollution Study,
II: Pulmonary Anthracosis as an Index of Air Pollution," American
Journal of Public Health 53, 185 (1963).
49. Zeidberg, L.D., R.A. Prindle, and E. Landau. "The Nashville Air Pollu-
tion Study, I: Sulfur Dioxide and Bronchial Asthma," American Review
of Respiratory Disease 84, 489 (1961).
50. Zeidberg, L.D., R.A. Prindle, and E. Landau. "The Nashville Air Pollu-
tion Study, III: Morbidity in Relation to Air Pollution," American
Journal of Public Health 54, 85 (1964).
81
-------
Appendix A
DAILY DEPARTMENT UTILIZATION VARIABLES USED IN THE ANALYSIS
(Pennsylvania Avenue)
Department
Internal Medicine
Unscheduled
Metro
Ophthalmology
Unscheduled
Metro
Pediatrics
Unscheduled
Metro
Urgent Visit Clinic
Urgent Visit
Metro
Minimum
oa (0)b
0 (0)
0 (0)
0 (0)
0 (0)
NA
13 (2)
0 (0)
i
Max uiium
34a (46)b
25 (26)
20 (37)
10 (S)
79 (1.31)
NA
260 (324)
24 (42)
;:ean
133 (8)b
8 (8)
6 (5)
2 (2)
36 (44)
NA
123 (146)
9 (13)
Standard
Deviation
8.5* (7.7)b
5.2 (5.8)
4. fa (6.0)
1.7 (1.9)
19. 4 (29.2)
NA
42.1 (44.4)
4.4 (5.5)
f
Figures for 1973.
Figures for 1974.
82
-------
Appendix B
DAILY DEPARTMENT UTILIZATION VARIABLES USED IN THE ANALYSIS
(Takoma Park)
Depar tment
Internal Medicine
Unscheduled
Pediatrics
Unscheduled
Urgent Visit Clinic
Urgent Visit
Minimum
oa (0)b
0 (0)
0 (0)
Maximum
83 (7)b
64 (89)
95 (144)
Mean
I3 (1)'
25 (24)
42 (38)
Standard
Deviation
.35* (5.2)b
14.6 (17.7)
22.5 (24.3)
Figures for 1973.
Figures for 1974.
83
-------
Appendix C
DAILY AIR POLLUTION VARIABLES USED IN THE ANALYSIS*
Air Pollution Measure
Photochemical. Oxidants
CAMP (Max. 1-hr, av.)
Civ. Park
(Max. 1-hr, av.)
Non-methane Hydrocarbons
CAMP (3-hr, av.)
Nitrogen Dioxide
CAMP (24-hr, av.)
Carbon Monoxide
CAMP (Max. 1-hr, av.)
D.C. Hosp.
(Max. 1-hr, av.)
Sulfur Dioxide
CAMP (24-hr, av.)
ACS (24-hr, av.)
D.C. Hosp. (24-hr, av.)
CAMP (Max. 1-hr, av.)
ACS (Max. 1-hr, av.)
D.C. Hosp.
(Max. 1-hr, av.)
Minimum
.ooia(.ooi)b
(.001)
.000
.005
1.00
1.00 (1.0)
(.002)
.002
.002 (.002)
(.002)
.008
.006
Maximum Mean
.180a(.140)b
(.220)
4.10
.101
43.0
30.0 (28.0)
(.082)
.134
.070 (.062)
(.076)
.222
.320
.048a(.038)b
(.044)
.564
.045
6.94
6.86 (6.76)
(.017)
.035
.020 (.014)
(.037)
.059
.041
Standard
Deviation
.032a(.026)b
(.033)
.542
.015
5.07
4.64 (4.34)
(.014)
.022
.012 (.009)
(.031)
.038
.032
*A11 figures in parts per million (ppm)
for 1973.
Figures for 1974.
-------
Appendix D
DAILY WEATHER VARIABLES USED IN THE ANALYSIS
Climatic Measure
Average Temperature (°F.)
Temperature
Difference (°F.)
(Max. - Min.)
Average Wind
Speed (m.p.h.)
Total Precipation (in.)
Minimum
19a (26)b
3 (3)
1.6 (2.7)
0 (0)
Maximum
86a (85)b
38 (38)
22.9 (20.4)
2.88 (1.90)
Mean
59.4a(58.8)b
18.1 (18.8)
8.4 (8.6)
.096 (.099)
Standard
Deviation
16.70a(15.1)b
6.58 (6.41)
3.08 (3.09)
.256 (.272)
Figures for 1973.
Figures for 1974.
85
-------
Appendix E
CORRELATION MATRIX OF 1973 EXPLANATORY VARIABLES AND OF 1973-1974 EXPLANATORY VARIABLES
00
1 D.C. Hosp-S07-24-Hr
ACS-SO -Max 1-Hr
! D.C. Hosp-S07-Max 1-Hr
> CAMP-CO-Max 1-Hr
~ D.C. Hosp-CO-Max 1-Hr
2 CAMP-NO -24-Hr
'• CAMP-NmH-3-Hr
'• CAMP-Ox-Max 1-Hr
i Ave Temp
i Temp Diff
Ave Wind
Precip
SAT
SUN
CA>lP-S07-24-Hr
! D.C. Hosp-SO -24-Hr
i CAMP-SO -Max 1-Hr
I D.C. Hosp-CO-Max 1-Hr
L; CAMP-Ox-Max 1-Hr
Civ Pk-Ox-Max 1-Hr
Ave Terap
Temp Diff
Ave Wind
Precip
T
ACS
SO,
24-Hr
Ave
.48
.89
.1*3
.25
.10
-.13
.08
-.-8
-.57
-.06
-.003
-.02
-.06
-.03
.25
.12
.18
-.03
-.49
-.36
-.55
-.10
.10
.007
D.C.
Heap-
so2
24-Hr
Ave
.52
.80
.23
.17
-.01
.11
-.24
-.44
.02
-.08
-.06
.21
-.08
.16
.05
. 12
-.03
-.35
-.28
-.43
-.06
.02
-.06
ACS
so2
Max
1-Hr
.48
.31
.15
-.06
.10
-.28
-.57
.04
-.06
-.05
-.06
-.03
.24
. 14
.17
-.05
-.49
-.41
-.57
-.08
.15
-.02
D.C.
Hosp
so2
Max
1-Hr
.35
.24
.05
.23
-.11
-.36
.16
-.15
-.08
-.04
-.06
.09
.06
.07
.02
-.30
-.25
.36
.04
.05
-.04
CAMP
CO
Max
1-Hr
.51
.36
.56
.28
-.10
.24
-.29
-.02
-.13
-.18
-.03
.12
-.02
.003
-.18
-.14
-.16
-.06
-.01
-.04
D.C.
Hosp
CO
Max
1-Hr
.22
.53
.08
-.09
.25
-.25
-.03
-.08
-.06
.09
.10
.09
.04
-.15
-.13
-.13
.07
.02
-.12
- 1973
CAM15
N'°2
24-Hr
Ave
.39
.29
.23
.28
-.42
-.05
-.15
-.11
-.15
-.03
-.09
.12
.20
.21
.22
.18
-.08
-.08
CAMP
NtnH
3-Hr
Ave
.45
.09
.24
-.34
.04
-.05
-.12
-.15
-.01
-.12
.06
.03
.02
.01
-.01
-.06
.03
CAMP
Ox
Max
1-Hr
.66
.22
-.30
-.05
-.01
.02
-.39
- . 22
-.35
-.05
.48
.53
.59
.03
-.15
. 12
Ave
Temp
. 12
- . ^ i
.01
-.01
-.02
-.34
-.34
-.28
- . 02
.67
.65
.80
.11
-.14
.06
Temp
Diff
-. 15
-.24
-.03
-.003
-.07
-.07
-.09
.01
.06
.03
.04
.06
.05
-.03
Ave
Wind
.04
.01
-.08
.11
. 12
.08
-.03
-.12
-.16
-.19
-.Of,
.06
-.02
s
Precip
-.06
-.02
.08
. 02
.12
.02
-.09
.07
-.007
-.04
.002
.003
-------
Appendix F
CORRELATION MATRIX OF 1974 EXPLANATORY VARIABLES
co
D.C. Hosp-S09-24-Hr
CAMP-SO -Max 1-Hr
D.C. Hosp-CO-Max 1-Hr
CAMP-Ox-Max 1-Hr
Civ Pk-Ox-Max 1-Hr
Ave Temp
Temp DifC
Ave Wind
Precip
SAT
SON
^
CAMP
SO
2
24-Hr
.74
.11
-.25
-.31
-.39
.11
-.16
-.05
.02
.06
D.C.
Hosp
SO
2
24-Hr
.09
.15
-.13
-.19
-.26
.J9
-.12
-.15
-.03
-.12
CAMP
SO..
2
Max
1-Kr
.22
-.21
-.21
-.30
.13
-.23
-.07
.03
.003
D.C.
i.'osp
CO
' !a x
J-Hr
-.001
.04
.05
.35
-.35
-.07
-.07
-.09
CAMP
Ox
Max
1-Hr
.67
.71
.31
-.16
-.08
.007
.1)
1 y t H
Civ.
Ox
Max
1-Hr
.71
.21
-.21
-.08
-.04
.05
Ave
Temp
.19
.22
.05
.02
.01
Temp
Diff
1'rerip
Wind
. 14
.19
.10
.03
. 17
-.05
.01
-.01
.06
-------
TECHNICAL REPORT DATA
(Please read Instructions on llic reverse before completing]
1. REPORT NO.
EPA-600/5-77-010
4. TITLE ANDSUBTITLE
Air Pollution and Health in Washington, D.C.
Some Acute Health Effects of Air Pollution in the
Washington Metropolitan Area
3. RECIPIENT'S ACCESSION NO.
5. REPORT DATE
July 1977
6. PERFORMING ORGANIZATION CODE
7. AUTHOR(S)
8. PERFORMING ORGANIZATION REPORT NO.
Eugene P. Seskin
9. PERFORMING ORGANIZATION NAME AND ADDRESS
10. PROGRAM ELEMENT NO.
National Bureau of Economic Research, Inc
1750 New York Ave, NW
Washington, D.C. 20006
PF 1 HA094
1 1. CONTRACT/
ACT/GRANT NO.
EPA Contract 168-01-3144
12. SPONSORING AGENCY NAME AND ADDRESS
Environmental Research Laboratory-Corvallis
Office of Research and Development
U.S.Environmental Protection Agency
Corvallis, Oregon 97330
13. TYPE OF REPORT AND PERIOD COVERED
extra-mural,final ,1973-74
14. SPONSORING AGENCY CODE
EPA/600-02
15. SUPPLEMENTARY NOTES
16. ABSTRACT
This study has attempted to assess some of the acute health effects of air pollution.
Specifically, the investigation has tested the hypothesis that air pollution can
aggravate the health status of a population and can result in increased utilization
of certain types of medical care services.
The study period was 1973-1974 and centered in the Washington, D.C. Metropolitan
Area. Statistical models were formulated, explaining health-care utilization of a
group practice medical care plan. Primary interest was focused on the effects of
mobile-source air pollutants including carbon monoxide, nitrogen dioxide, non-methane
hydrocarbons, and photochemical oxidants. Meteorological conditions as well as
other variables thought to influence the consumption of medical services were includec
in the models as explanatory variables.
The statistical results indicated
on the health-care utilization of
that air pollution levels
the group practice.
had a very limited effect
KEY WORDS AND DOCUMENT ANALYSIS
DESCRIPTORS
Air Pollution
Economic Analysis
Benefit/Cost Analysis, Air Pollution
Economic Effects, Air Pollution
Economic Effects, Health
b.IDENTIFIERS/OPEN ENDED TERMS
Air Pollution Economics
Economic Impact
Air Pollution Effects
(Health)
COSATI 1 iolil/Ciroup
06/F,S
18. DISTRIBUTION STATEMENT
Release to Public
19. SECURITY CLASS (This Report/
Unclassified
21. NO. OF PAGES
94
20 SECURITY CLASS (Tills page I
Unclassified
22. PRICE
EPA Form 2220-1 (Rev. 4-77)
88
O U.S. GOVERNMENT POINTING OCClCE 1977-79/1-504/197 REGION 10
-------
U.S. ENVIRONMENTAL PROTECTION AGENCY
Office of Research and Development
Technical Information Staff
Cincinnati, Ohio 45268
POSTAGE AND FEES PAID
U S ENVIRONMENTAL PROTECTION AGENCY
EPA-335
OFFICIAL BUSINESS
PENALTY FOR PRIVATE USE. S3OO
AN EQUAL OPPORTUNITY EMPLOYER
Special Fourth-Class Rate
Book
$
z
a
If your address is incorrect, please change on the above Inhcl,
tear off; and return to the above address.
If you do not desire to continue receiving this technical report
series, CHECK HERE d; tear off label, and return it to the
above address.
------- |