Full text of DOI:10.1080/10473289.2000.10464188

This article was downloaded by: [Mercer University]
On: 21 October 2014, At: 20:06
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer
House, 37-41 Mortimer Street, London W1T 3JH, UK
Journal of the Air & Waste Management Association
Publication details, including instructions for authors and subscription information:
http://www.tandfonline.com/loi/uawm20
Aerosol Research and Inhalation Epidemiological
Study (ARIES): Air Quality and Daily Mortality
Statistical Modeling— Interim Results
Rebecca J. Klemm a & Robert M. Mason Jr. a
a Klemm Analysis Group , Washington , DC , USA
Published online: 27 Dec 2011.
To cite this article: Rebecca J. Klemm & Robert M. Mason Jr. (2000) Aerosol Research and Inhalation Epidemiological
Study (ARIES): Air Quality and Daily Mortality Statistical Modeling— Interim Results, Journal of the Air & Waste
Management Association, 50:8, 1433-1439, DOI: 10.1080/10473289.2000.10464188
To link to this article: http://dx.doi.org/10.1080/10473289.2000.10464188
PLEASE SCROLL DOWN FOR ARTICLE
Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained
in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no
representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose
of the Content. Any opinions and views expressed in this publication are the opinions and views of the
authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should
not be relied upon and should be independently verified with primary sources of information. Taylor
and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses,
damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection
with, in relation to or arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any substantial or systematic
reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any
form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http://
www.tandfonline.com/page/terms-and-conditions

TECHNICAL PAPER
Klemm and Mason
ISSN 1047-3289 J. Air & Waste Manage. Assoc. 50:1433-1439
Copyright 2000 Air & Waste Management Association
Aerosol Research and Inhalation Epidemiological Study
(ARIES): Air Quality and Daily Mortality Statistical Modeling—
Interim Results
Rebecca J. Klemm andRobert M. Mason, Jr.
Klemm Analysis Group, Washington, DC
ABSTRACT
air quality relative to today’s number of deaths was posi-
tive for lag 0, lag 1, and 2-day average and positive only
for decedents at least 65 years of age using first differ-
ence. The t values ranged from 0.81 to 1.15 for lag 0, 1.04
to 1.53 for lag 1, 1.10 to 1.66 for 2-day average, and –0.32
to 0.33 for first difference with 346 or 347 days of data.
No statistically significant estimate of the linear coeffi-
cient was found for the other 14 air quality variables in
our interim analysis for the four mortality subgroups. We
discuss diagnostics to support these models.
These interim analyses did not include an evaluation
of sensitivity to a larger set of lag structures, nonlinear
model specifications, multipollutant analyses, alternative
weather model and smoothing model specifications, air
pollution imputation schemes, or cause-specific mortal-
ity indicators, nor did they include a full reporting of
model selection or goodness-of-fit indicators. No conclu-
sion can be drawn at this time about whether the find-
ings from subsequent studies have sufficiently greater
power to detect effects comparable to those found in other
U.S. cities including at least 2 or 3 years of data.
The Aerosol Research and Inhalation Epidemiological
Study (ARIES) is an EPRI-sponsored project to collect air
quality and meteorological data at a single site in north-
western Atlanta, GA. Seventy high-resolution air quality
indicators (AQIs) are used to examine statistical relation-
ships between air quality and health outcome end points.
Contemporaneous mortality data are collected for Fulton
and DeKalb counties in Georgia. Currently, 12 months of
air quality and weather data are available for analysis, from
August 1998 through July 1999.
The interim mortality analysis used Poisson regres-
sion in generalized additive models (GAMs). The estimated
log-linear association of mortality with various AQIs was
adjusted for smoothed functions of time and meteoro-
logical data. The analysis considered daily deaths due to
all nonaccidental causes, deaths to persons 65 years or
older, and deaths in each of the two constituent counties.
The fine particle effect associated with the four mortality
subgroups, using only today (lag 0), yesterday (lag 1),
2-day average (average of today and yesterday), and first
difference (today minus yesterday) measurements of the
INTRODUCTION
IMPLICATIONS
The Aerosol Research and Inhalation Epidemiological Study
(ARIES) is a multifaceted, multidisciplinary research pro-
gram. ARIES has developed a data set containing air qual-
ity indicators (AQIs) and health outcome end points. This
data set comprises 70 AQIs and meteorological measure-
ments, plus mortality counts and emergency department
visits. Van Loy, Bahadori, and Wyzga provide general in-
formation on the entire ARIES project.1 Previous studies
have identified associations between ambient (outdoor) PM
and adverse health outcomes, primarily a relationship be-
tween 2-day average fine particulates and daily mortality.2
To improve on these study designs, ARIES is designed to
record several daily AQIs simultaneously. The data are then
used to analyze the relationships between AQIs and health
outcomes. All AQIs in ARIES are ambient ecologic air qual-
ity measures, and no individual exposure is captured.
Numerous papers have purported to show a statistically
significant positive association of mortality with air pollu-
tion levels. However, no statistically significant estimate
of the linear coefficient was found in our analysis. Model
diagnostics provide little evidence for further investiga-
tion of the model behavior or inconsistencies of the esti-
mates. These results occurred with average PM2.5 levels
similar to those in prior studies. Additional modeling us-
ing other variables in the ARIES epidemiological analysis
data set must be completed to fully assess the
relationship(s) between daily mortality counts and AQIs.
Additional meteorological variables will also be examined
to model the effects of other weather variables. A better
understanding of the ambient environmental exposure in
targeted populations during the period immediately pre-
ceding death may also provide clues on our lack of sta-
tistically significant interim results.
Volume 50 August 2000
Journal of the Air & Waste Management Association 1433

Klemm and Mason
This paper covers exploration of the data, data min-
ing, and statistical modeling. Data exploration captures
properties of the data that may affect the selection of vari-
ables or analysis models. Missingness and outlier analy-
ses were performed to understand the robustness of the
data set. Principal component and factor analyses were
applied to understand the variance properties of the mul-
titude of AQIs, some of which are statistically related quan-
tities, such as 24-hr average and maximum daily 1-hr
average. Since statistical modeling relies on variance prop-
erties of the independent and dependent variables, data
mining provides useful information for the variable se-
lection process. This paper covers findings for the period
from August 1, 1998, to July 31, 1999.
Our interim model used the generalized additive
model (GAM) to estimate the relationship between AQIs
and daily mortality, and it had the Poisson regression form.
This model approach has been used in recent air quality
and health outcome studies. One attractive feature of the
model is the simplicity of the calculation of estimated
proportional changes in mortality from pollutants. It also
provides a foundation for comparison of ARIES results with
previous, similar studies.
averages; daily cumulative; and 5-min, 1-, 3-, 8-, and
24-hr maxima and minima. Each variable has specific units
associated with the measurements. The complete list of
measurements and the ARIES data dictionary are avail-
able from EPRI.3
By consensus, the ARIES epidemiology team of staff
from EPRI, Klemm Analysis Group, and Emory University
prioritized 15 of the original 58 pollutant measures. This
consensus was based on explicit or implicit linkage to pre-
vious mortality and morbidity studies and time limits to
complete the interim analysis. The priority list represents
many of the pollutants previously modeled. These vari-
ables included daily average PM2.5 (Federal Reference
Method [FRM]), PM10, coarse particulate mass, organic car-
bon, elemental carbon, oxygenated hydrocarbons, acid,
NO3, SO4, ultrafine particles of 10–100-nm diameter area
and count, and daily maximum 1-hr CO, O3,NO2, and SO2.
Our interim analysis used the data from the ARIES
epidemiological database without consideration of outli-
ers (except as noted for imputation of PM2.5 [FRM] described
in the Methods section). Future analysis will address other,
if not all, data in the ARIES epidemiological database.
This paper focuses on representative examples of the
dozens of AQI variables and model combinations that the
ARIES epidemiology team of the EPRI, Klemm Analysis
Group, and Emory University established a priori.
Mortality Data
In many mortality studies, annual mortality data are ob-
tained from the National Center for Health Statistics, U.S.
Department of Health and Human Services. These data
are available 18 months–2 years after the time of death.
Klemm arranged to acquire copies of redacted (identities
removed) death certificates for the ARIES area (Fulton and
DeKalb counties, GA) from the County Offices of Vital
Records. These mortality records were entered into elec-
tronic files, reviewed, unduplicated, and coded for Inter-
national Classification of Disease (ICD)-9 cause of death.
This contemporaneous mortality database provided the
basis for our mortality analysis.
METHODS
All ARIES AQI and meteorology data are collected at a
single site in northwestern Atlanta, GA, which houses all
the measurement instruments. The site has been designed
to account for the local geography and avoid interference
between instruments, and it offers on-site diagnostic, re-
pair, and data-processing capabilities.
Meteorological Data
Mortality data were compiled from redacted copies
of the official Georgia death certificates. The Georgia cer-
tificate of death contains 32 fields of information, and
many of these fields, such as cause of death, contain mul-
tiple subfields. The redacted certificate contains counties
of death and residence; place, time and date of death;
age; sex; and causes of death. The records were prescreened
by the county departments of vital records before being
sent, in an effort to eliminate accidental causes of death
(ICD-9 code 800 or higher); those accidental causes not
identified by the county staff were eliminated at a later
step in the process. The data from the redacted copies of
each death certificate were entered into an electronic file.
Next, we sorted the mortality file on all data fields to
identify possible duplicate entries. Unique entries were
set aside and the potential duplicates reviewed manually.
Mortality entries with identical information were removed
The ARIES data set includes 10 meteorological variables
and daily 24-hr maximum and minimum temperatures.
Relative humidity is reported as a 24-hr average, daily
maximum 1-hr average, and daily minimum 1-hr aver-
age. Dew-point temperature, barometric pressure, and
solar flux are analyzed only as 24-hr averages. Precipita-
tion appears as a daily total. Complete definitions of the
meteorological data are available through EPRI.3
Air Quality Measurements
Sixty air quality measurement variables are stored in the
ARIES database, including fine, coarse, and PM10; ultrafine
particle counts, numbers, and size distributions; and car-
bons, hydrocarbons, gaseous pollutants, mold, pollen, and
other epidemiologically interesting species. These quan-
tities include multiple PM2.5 measurement methods; daily
1434 Journal of the Air & Waste Management Association
Volume 50 August 2000

Klemm and Mason
as duplicate when the counties of residence and death
were Fulton and DeKalb. This process also eliminated other
duplicated entries.
from the other measures of PM2.5. Since each model in-
volved only one AQI, missing FRM data were not associ-
ated with other covariates.
The file of mortality records was then processed to
calculate daily mortality counts. Each record was allocated
to the count for the day recorded as the date of death on
the certificate. We accumulated counts for the total file
by age (>64) and by county. This procedure produced four
mortality subgroups for the interim analysis.
Modeling
Our interim models employed Poisson regression in a
GAM framework, as determined by the ARIES team. This
model is analogous to the model from an October 1996
paper by Schwartz, Dockery, and Neas,2 and is consistent
with models used in other mortality–air quality research.
The interim-analysis model used day of study, aver-
age daily temperature, average dew-point temperature, day
of the week, and one air quality variable. The time and
temperature variables entered the model in a smoothed
form. Day of the week was a factor variable with seven
levels and the AQI variable entered in a linear form.
We used natural splines rather than loess to smooth
terms in our models. Natural splines offer several advan-
tages over loess, including piecewise polynomial proper-
ties, more direct control over the tradeoff between local
and global features of the smoothing operation (through
specifying the number and placement of knots), and ex-
act theory for statistical inference and diagnostics. In our
initial spline models, we set time-based knots at monthly
midpoints. For temperature and dew point, we chose three
knots each, based on the quartiles of each variable.
To provide diagnostics on the model fit, we used the
results of the partial fit of the model, given the smoothed
and factor variables. The partial fits were plotted to inves-
tigate the deviation of the linear model from smoothed
estimates of the air quality variable.
A certified nosologist coded the mortality records fol-
lowing the ICD-9 cause of death scheme. This process lagged
the mortality data collection by several months. The ICD-
9 coding will be used to prepare daily mortality counts by
specific illness-related subgroups for future analysis.
ANALYTIC APPROACH
Our analysis methods included the exploration of the
time-based behavior of the air quality variables, the iden-
tification of missingness, the appropriateness and ap-
proach to imputation, the assessment of variance patterns
with principal component and cluster analysis, and the
modeling. The Study of Particulates and Health in Atlanta
(SOPHIA) team from Emory University has provided stan-
dard parametric analysis of the data.4 The SOPHIA project
examines the relationship between the ARIES data and
morbidity health end points.
Missingness and Imputation
Missing observations reduce the degrees of freedom in a
statistical model, and they may also bias results. The
missingness among the ARIES meteorological and AQI
ranged from 5 to 102 out of 365 among the variables con-
sidered in this paper. The primary reason for missing data
was instrument problems. Each missing value corresponds
to a day with that observation missing. Partial measure-
ments result in missing observations since all air quality
and meteorological data are averages, cumulative sums,
or maxima or minima from the 24-hr daily period. Mor-
tality counts are available for every day in the interim
study period.
RESULTS
Table 1 presents summary statistics for the 4 mortality
counts, 2 meteorological variables, and 15 AQIs in our
interim database. The average annual PM values are espe-
cially interesting. PM2.5 had a mean value of 19.9 µg/m3,
which is similar to the averages for Knoxville, TN (20.8)
and St. Louis, MO (18.7) in the model of Schwartz,
Dockery, and Neas.2
ARIES provides several measures of PM2.5, including
FRM, tapered element oscillating microbalance (TEOM),
and particle composition monitor (PCM), with 24-hr av-
erages and daily minimum and maximum 1-hr averages.
It was agreed by the ARIES team that initial models would
focus on the FRM. However, the FRM variable had 46
missing values. Of those 46 missing days, 42 had
nonmissing values among at least one of the other PM2.5
variables, and the various measures were highly correlated.
The classification and regression tree methodology pro-
vided a mechanism for exploiting available information
even on incomplete days. Thus, a regression tree was used
to impute the missing values of FRM by predicting them
The 1998 U.S. Census population estimate for Fulton
and DeKalb counties is 1,333,217. The mean daily total
mortality for the 1-year period (nonaccidental causes of
death) was 22.9. For the elderly, the mean daily mortality
was 14.9. Fulton County contributes almost twice as many
daily deaths as DeKalb County, but both counties have
similar population age distributions. Among 8349 interim-
analysis deaths, 293 (3.5%) were under 25 years of age,
774 (9.3%) were between 25 and 45, 1858 (22.2%) were
between 45 and 65, and 5424 (65.0%) were 65 years or
older. The age distribution for the ARIES mortality had
slightly less elderly by percentage than previous studies,
in which the 65+ age group often represents 70–75% of
Volume 50 August 2000
Journal of the Air & Waste Management Association 1435

Klemm and Mason
Table 1. Summary statistics for measures in interim ARIES analysis.
Measure
Units
Mean
Standard
Deviation
Minimum
First
Median
Third
Maximum
Number
Missing
Quartile
Quartile
All deaths
#
#
22.9
14.9
8.3
5.3
4.2
8.0
5.0
19.0
12.0
6.0
23.0
15.0
8.0
27.0
18.0
10.0
17.0
24.2
20.2
25.7
38.4
12.5
62.5
63.4
1806.6
25.1
65.5
18,357.8
0.0
41.0
30.0
0
0
Deaths 65+ years old
Deaths in DeKalb County
Deaths in Fulton County
Avg. temperature
Avg. dew point
PM2.5 FRM (imputed)
PM10
#
2.9
2.0
16.0
0
#
14.5
18.1
13.1
19.9
30.4
10.0
45.5
50.7
1467.9
18.7
62.5
15,172.3
0.0
4.2
4.0
12.0
12.6
6.9
14.0
19.4
14.7
18.3
29.3
9.4
27.0
0
°C
°C
7.2
1.0
29.4
16
14
0
8.2
–9.6
1.0
26.1
µg/m3
µg/m3
µg/m3
ppbv
ppbv
ppbv
ppbv
µm2/cm3
#/cm3
µg/m3
µg/m3
µg/m3
ppbC
µg/m3
µg/m3
9.3
13.2
21.3
6.3
54.8
12.2
5.3
6.1
69.0
72
65
21
78
24
26
102
102
113
16
16
67
16
16
Coarse mass
0.2
39.5
O3 max 8 hr
NO2 max 1 hr
CO max 1 hr
25.5
19.7
1304.5
17.7
115.9
26,584.5
0.0
3.5
25.0
35.6
603.6
6.7
40.6
49.4
979.2
12.7
43.4
10,917.0
0.0
130.7
107.9
9422.3
106.5
1400.8
299,239.2
20.2
7.6
271.1
1.0
SO2 max 1 hr
PM 10–100-nm surface area
PM 10–100-nm count
Acid
1.6
27.2
4529.6
0.0
550.3
0.0
Elemental carbon
Organic carbon
Oxygenated hydrocarbons
SO4
2.3
1.7
0.1
1.3
1.9
2.6
15.8
4.6
2.2
0.3
3.1
4.3
5.6
16.1
38.0
5.2
28.9
3.4
6.3
21.5
2.6
30.8
4.4
45.0
6.9
299.5
20.8
0.6
NO3
1.1
0.9
0.1
0.5
0.9
1.4
6.7
Aug 1998
Oct 1998
Dec 1998
Feb 1999 April 1999
Date
June 1999
Aug 1999
Figure 1. Total mortality in Fulton and DeKalb counties, August 1998–July 1999.
1436 Journal of the Air & Waste Management Association
Volume 50 August 2000

Klemm and Mason
Table 2. Increase in mortality associated with 10 µg/m3 PM2.5 FRM (imputed),
August 1998–July 1999.
the daily mortality counts.2 Figure 1 shows a time-series
plot of the daily mortality count for all causes of deaths.
The number peaked in February and troughed in July, but
the pattern of variation appears rather weak.
Mortality
% Increase
95% CI
t Value
Table 1 also lists number of missing observations for
each of the measurements of interest. Most missing ob-
servations occurred for short periods of time. However,
instruments used in ARIES periodically need lengthy re-
medial work, leading to long periods without measure-
ments. The missingness for our interim variables was not
a major problem, especially since data collection is ongo-
ing. Figure 2 displays the regression tree-based imputa-
tion of the PM2.5 (FRM) measurements. As described in
the methods section, this imputation used the other 14
measurements of PM2.5 in the database. The imputed val-
ues appear to be reasonable, but as with most regression-
based imputations, they tend to fall in the middle range
of the raw data values.
Total
lag 0
lag 1
1.87
2.54
–0.48
3.14
3.15
2.52
0.61
4.04
2.19
2.86
–0.47
3.58
1.72
2.43
–0.52
2.94
(–1.29, 5.14)
(–0.70, 5.89)
(–3.36, 2.49)
(–0.70, 7.12)
(–0.80, 7.25)
(–1.48, 6.69)
(–3.00, 4.35)
(–0.73, 9.03)
(–2.99, 7.64)
(–2.48, 8.48)
(–5.16, 4.46)
(–2.72, 10.30)
(–2.24, 5.83)
(–1.62, 6.64)
(–4.14, 3.23)
(–1.83, 7.95)
1.15
1.53
–0.32
1.60
1.56
1.22
0.33
1.66
0.81
1.04
–0.19
1.10
0.84
1.17
–0.28
1.20
1st difference
2-day average
lag 0
Deaths 65+ years old
DeKalb County
lag 1
1st difference
2-day average
lag 0
lag 1
1st difference
2-day average
lag 0
Fulton County
lag 1
1st difference
2-day average
Table 2 presents the results of the GAM for PM2.5 across
the four mortality counts and across lag 0, lag 1, the first
difference, and the 2-day average. An increase of 10
µg/m3 was associated with a 1.7–4.0% increase in mortal-
ity across all outcomes (except the first difference). How-
ever, none of these estimates achieved statistical
significance at 95% confidence. For these interim esti-
mates, there is no statistical evidence of an association
between mortality and PM2.5.
Note: All models had complete data for 347 or 348 out of 365 days.
Table 3 displays results for total mortality across all of
the initial AQI variables. As with PM2.5, we found no statis-
tical significance among any of the associations. The model
ran for all four mortality outcomes by all 15 measures at
Aug 1998
Oct 1998
Dec 1998
Feb 1999 April 1999
Date
June 1999
Aug 1999
Figure 2. PM2.5 FRM imputation.
Volume 50 August 2000
Journal of the Air & Waste Management Association 1437

Klemm and Mason
Table 3. Association of total daily mortality with interim ARIES pollutant variables, August 1998–
lag 0, lag 1, 2-day averaging, and the first differ-
ence of the air quality variable (not shown). Not
one of these initial 240 models exhibited statisti-
cal evidence of a nonzero association. No multiple-
pollutant models were attempted, given the
apparent overall lack of association.
July 1999.
Measure
Unit
Log-Linear
Coefficient
Standard
Error
t Value
PM2.5 FRM (imputed)
PM10
Coarse mass
O3 max 8 hr
NO2 max 1 hr
CO max 1 hr
µg/m3
µg/m3
µg/m3
ppbv
0.00186
0.00167
0.00057
–0.00041
0.00069
0.00002
0.00034
–0.00014
0.00000
0.94144
0.00258
0.00096
–0.00009
0.00506
–0.01552
0.00161
0.00140
0.00273
0.00110
0.00090
0.00001
0.00074
0.00013
0.00000
0.94281
0.00790
0.00606
0.00048
0.00535
0.01698
1.15
1.19
The mean ambient and dew-point temperatures
were also not generally statistically significant in
the model. Both temperatures entered the model
as smoothed estimates. This lack of significance
indicates the smoothed term for study time is the
only statistically significant variable in our interim
model; some models also exhibit an effect for day
of the week.
Partial fit plots for four selected air quality vari-
ables are presented in Figure 3. These plots show
the relationship of the data, the linear coefficient
estimate of the model, and the approximation of
the smoothed estimator. None of the partial fits
demonstrated substantial differences between the
linear and smoothed estimators of association.
However, when all diagnostic plots are considered,
0.21
–0.37
0.77
ppbv
ppbv
1.76
SO2 max 1 hr
PM 10–100-nm surface area
PM 10–100-nm count
Acid
ppbv
0.46
µm2/cm3
#/cm3
µg/m3
µg/m3
µg/m3
ppbC
–1.06
–1.21
1.00
Elemental carbon
Organic carbon
Oxygenated hydrocarbons
SO4
0.33
0.16
–0.20
0.95
µg/m3
µg/m3
NO3
–0.91
PM2.5 FRM daily 24-hr average(imputed)
PM 3-10 nm daily 24-hr average
PM 3-10 nm surface area (µm2/cm3)
PM2.5 FRM (imputed)
CO daily maximum 1-hr average
O3 daily maximum 8-hr average
CO(ppbv)
O3 (ppbv)
Figure 3. Partial fit of mortality vs pollutants.
1438 Journal of the Air & Waste Management Association
Volume 50 August 2000

Klemm and Mason
it is clear that some variables may have highly influential
extreme values. These values must be evaluated to deter-
mine whether they are anomalous outliers, or perhaps
whether some measures should be transformed.
tistical programming for the paper, and David Cowan for
his editing assistance. Finally, the authors must thank Ann
Parr and her staff at the Fulton County Office of Vital
Records and Wanda Cason and her staff at the DeKalb
County Office of Vital Records. Without their understand-
ing of the importance of this project and their coopera-
tion with the redaction, copying, and transmission of the
mortality records, our analysis would never have been
possible during the available time.
DISCUSSION
The ARIES team constructed an a priori statistical model
for interim consideration, and prioritized 15 of the origi-
nal 58 pollutant measures. The priority list represents
many of the pollutants modeled previously. However, by
focusing the range of initial considerations, we have
guarded against wholesale data dredging. Further analy-
sis will account for additional AQIs, alternative approaches
for adjusting for meteorological variables, and subsets of
mortality by cause of death. Nonetheless, among the ini-
tial 240 models, there is no statistical evidence whatso-
ever of an association between daily mortality count and
the 15 pollutants.
REFERENCES
1. Van Loy, M.; Bahadori, T.; Wyzga, R. Aerosol Research and Inhala-
tion Epidemiology Study (ARIES): PM2.5 Mass and Aerosol Compo-
nent Concentrations and Sampler Intercomparisons; J. Air & Waste
Manage. Assoc., in press.
2. Schwartz, J.; Dockery, D.W.; Neas, L.M. Is Daily Mortality Associated
Specifically with Fine Particles? J. Air & Waste Manage. Assoc. 1996,
46, 927-939.
3. EPRI Headquarters: 3412 Hillview Ave., P.O. Box 10412, Palo Alto,
CA 94303; Attention: Michael Van Loy.
4. Tolbert, P.E.; Klein, M.; Metzger, K.B.; Flanders, W.D.; Todd, K.;
Mulholland, J.A.; Ryan, P.B.; Frumkin, H. Interim Results of the Study
of Particulates and Health in Atlanta (SOPHIA); J. Expos. Anal. Environ.
Epidemiol., in press.
The use of 1 year of data may be cause for concern. We
do not believe that the duration of the data necessarily
influences the lack of findings. However, if the effect sizes
and intrinsic variability remain nearly the same with longer
periods of time, then most of the t statistics (other than for
first difference) could become statistically significant with
2–3 years of data. Since the study is ongoing, however, this
issue will be reconsidered later with additional data.
ACKNOWLEDGMENTS
The authors thank EPRI, the program director Dr. Ronald
E. Wyzga, and the ARIES project manager Dr. Tina
Bahadori for their advice and direction. This work was
supported by EPRI Contract WO3253-05. We also want
to thank the SOPHIA research team for their cooperation
and collaboration during this project. The authors also
wish to thank Charles M. Heilig, Carol Link, Michael Jones,
Rhonda Jackson, Jeff Carter, and Clinton J. Alverson for
their assistance with the research, computation, and sta-
About the Authors
Rebecca J. Klemm, Ph.D., and Robert M. Mason, Jr., are
senior statisticians at Klemm Analysis Group, Inc. Please
address all correspondence to Rebecca J. Klemm, Ph.D.,
Klemm Analysis Group, Inc., 1785 Massachusetts Ave. NW,
Fifth Fl., Washington, DC 20036.
Volume 50 August 2000
Journal of the Air & Waste Management Association 1439