See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/23195345 Intra- and inter-individual variability in location data for two U.S. health-compromised elderly cohorts Article in Journal of Exposure Science and Environmental Epidemiology · August 2008 DOI: 10.1038/jes.2008.47 · Source: PubMed CITATIONS READS 6 21 5 authors, including: Emma Frazier Thomas Mccurdy 40 PUBLICATIONS 541 CITATIONS 65 PUBLICATIONS 960 CITATIONS Centers for Disease Control and Prevention SEE PROFILE United States Environmental Protection Age… SEE PROFILE Some of the authors of this publication are also working on these related projects: Relative variability in longitudinal time use and physical activity data View project All content following this page was uploaded by Thomas Mccurdy on 07 January 2017. The user has requested enhancement of the downloaded file. All in-text references underlined in blue are added to the original document and are linked to publications on ResearchGate, letting you access and read them immediately. Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 r 2008 Nature Publishing Group All rights reserved 1559-0631/08/$30.00 www.nature.com/jes Intra- and inter-individual variability in location data for two U.S. health-compromised elderly cohorts EMMA L. FRAZIERa, THOMAS MCCURDYb, RON WILLIAMSb, WILLIAM S. LINNc AND BARBARA JANE GEORGEb a Division of Science and Mathematics, Morehouse College, Atlanta, Georgia, USA Human Exposure and Atmospheric Sciences Division, National Exposure Research Laboratory, US Environmental Protection Agency, Research Triangle Park, North Carolina, USA c Environmental Health Service, Rancho Los Amigos Medical Center, Downey, California, USA b This study provides descriptive statistical data on daily time spent in three locations of exposure assessment interest for two panel studies of healthcompromised elderly individuals 465-year-old having multiple days of human activity data. The panel studies include individuals living in Los Angeles (CA) and Baltimore (MD) in various housing types. Three general locations are evaluated: outdoors, in vehicles, and total indoors. Of particular interest is providing information regarding the within- and between-individual variability in the time use data for the three locations. The data are analyzed using non-parametric statistics and alternative statistical models. Within and between variability are evaluated using intraclass correlation coefficients (ICCs); daily ‘‘lag-one’’ autocorrelation coefficients are also provided for the two samples. There were significant gender differences for selected seasonal and/or day-of-the-week metrics for: (1) outdoor time in Los Angeles, but not in Baltimore, and (2) in-vehicle time in both areas. Elderly women spent more time in these locations than similarly aged men. The ICC statistic indicates that most of the variability in the time spent in the three locations is due to intraindividual variability rather than to inter-individual variability. The results indicate that US Environmental Protection Agency should consider gender, day-of-the-week, and time-of-day data in its exposure modeling of daily activities undertaken by the health-compromised elderly population. Journal of Exposure Science and Environmental Epidemiology advance online publication, 27 August 2008; doi:10.1038/jes.2008.47 Keywords: elderly, gender, locations, time use. Introduction The National Exposure Research Laboratory (NERL) of the U.S. Environmental Protection Agency (EPA) is engaged in a research effort to better understand how changes in the aging population may differentially affect their exposures to environmental contaminants. One purpose of this research is to identify susceptible subgroups in the diverse population of older adults (Geller and Zenick, 2005). Susceptibility may be related to differences in activity patterns or physiological/ metabolic parameters in elderly individuals, and NERL is emphasizing those factors in their research. Alternative ‘‘lifestyles’’ are manifestations of activity patterns, and we feel that they are important in the identification of possible population subgroups. With respect to lifestyles, the elderly 1. Abbreviations: CHAD, Consolidated Human Activity Database; CHD, coronary heart disease; COPD, chronic obstructive pulmonary disease; EPA, US Environmental Protection Agency; ICC, intraclass correlation coefficient; KS, Kolmogorov–Smirnov; NHAPS, National Human Activity Pattern Survey 2. Address all correspondence to: Thomas McCurdy, US Environmental Protection Agency, MD 205-02, Research Triangle Park, NC 27712, USA. Tel.: þ 919 541 0782. Fax: þ 919 541 9444. E-mail: mccurdy.thomas@epa.gov Received 19 March 2008; accepted 30 June 2008 often are categorized into healthy, active, ‘‘the fit’’, working, and frail subgroups, but these subgroups must be explicitly defined for use in EPA’s exposure models that are used to evaluate the impacts of environmental exposure on population subgroups (McCurdy and Graham, 2003). Although lifestyle has some relationship with physical activity patterns, such as exercise and fitness levels, we focus here on the locational aspects of elderly persons activities as exposure is locationally dependent. If a person is not in a location containing a pollutant, then no exposure to that agent occurs (Zartarian et al., 2005). Thus, location is of prime importance, and our focus in this paper is not on ‘‘what’’ but on ‘‘where.’’ Even though there are hundreds of articles focused on elderly physical activities, ‘‘activities of daily living’’, leisure activities, and so forth, few of them contain explicit information about where these activities occur, except those studies that focus on a particular assisted-care facility or other institution; however, EPA does not routinely undertake exposure assessment in those locations. Of 15 leisure activities (both cognitive and physical in nature) analyzed in Verghese et al. (2003), location was not provided for any of them. Most could occur in many locations. The majority of the 41 activities surveyed in Stewart et al. (2001) could also occur in many locations, although many could Variability in elderly person’s location data Frazier et al. reasonably be assigned to a general location with some degree of uncertainly (playing golf does not usually occur inside, for instance; likewise gardening). Most articles on the elderly do not provide explicit information regarding where the elderly are in time and space (e.g., Moss and Lawton, 1982; Verbrugge et al., 1996; Federal Interagency Forum, 2006). The point here is that explicit locational information for elderly activities rarely is provided; the exceptions are mentioned in the Discussion section below. EPA’s exposure models usually focus on a year as the time period of analysis. Population ‘‘cohorts’’ of interest, represented by a set of simulated individuals with shared characteristics, are mathematically tracked over that time period on an event-by-event basis, where an event is a finite length of time that a person is in a specific location undertaking a single activity. If the location, type of activity, combustion status (cooking, fireplace use, etc.), or clock hour changes then a new event occurs. Event data are included in a database of almost 23,000 person-days of activity information gathered over the years by governmental and academic time use researchers; it is known as the Consolidated Human Activity Database (CHAD). It is publicly available at www.epa.gov/chadnet1/. There are 2149 person-days of data for people Z65 years of age currently in CHAD, and more will be available soon. Activity data on the elderly also are available from the American Time Use Study conducted by the Bureau of Labor Statistics; see: www.bls.gov. See, in particular: ‘‘How do older Americans spend their time?’’ (Krantz-Kent and Stewart, 2007), which is available on that website. CHAD contains 24 h sequential activity data from both ‘‘real-time’’ diary studies and ‘‘ex-post’’ recall surveys (usually of the form: ‘‘What did you do yesterday?’’ where the respondent recounts the previous day’s activities from midnight-to-midnight). The recall studies are random probability and national in scale, and constitute about 64% of the data in CHAD (14,802 person-days of data). There is another 6,334 days of information from random-probability surveys in California residents and in three metropolitan areas around the country. Thus, most of the data (96%) in CHAD is probability based. The remaining data are from panel studies. About 75% of the data in CHAD is cross-sectional in nature, with only 1 day of information being available for an individual. The remainder contains multiple days of time use data for individual respondents, varying from 2 to 24 days. One of the multiday studies in CHAD is analyzed in this paper, as described below. The disassociation between longitudinal exposure modeling and cross-sectional data requires that a set of decision rules be used to form appropriate cohorts for exposure modeling purposes. Recently, a statistical approach has been developed and applied to better combine cross-sectional activity data into reasonable long-term modeling cohorts 2 (Glen et al., 2008). It attempts to: (1) maintain intra- and inter-variability in locations frequented by people with similar anthropogenic characteristics, and (2) mimic day-today correlation in the time spent in those locations by individuals in each subgroup. The variability aspect of the approach is most succinctly described by an intraclass correlation coefficient (ICC) (Winer, 1971; Quackenboss et al., 1986; McGraw and Wong, 1996; Hruschka et al., 2005). The ICC is explicitly defined below, but basically it is the proportion of between-individual variance relative to total within- and between-individual variance explained by a model. A low ICC indicates a high level of intraindividual variability relative to between-individual variability. There is a need for more information on locational ICCs for all population subgroups, including the elderly, to determine how much emphasis should be placed on within-individual variability in developing exposure modeling cohorts. We know of no study to date that has characterized locational ICCs for Americans Z65-year-old. As mentioned above, there are few studies that even provide descriptive statistical information on where elderly people spend time. The intent of this paper is to analyze existing multiday human activity data from two US panel studies of people Z65 years and to provide descriptive information on where they spent time in locations important to exposure modeling. The ICC and 1-day autocorrelation statistics are two descriptive metrics that are estimated. Only two studies could be found that contain daily activity data suitable for this purpose (described below), and both include individuals with preexisting respiratory or cardiovascular impairment. Older persons with compromised health in general are negatively affected when exposed to airborne pollutants. EPA labels people with preexisting disease as being ‘‘susceptible,’’ meaning that they are more likely to have an adverse health effect associated with environmental exposures. The elderly groups evaluated here are a subgroup of prime concern to the EPA and are important from an exposure modeling viewpoint even if our findings have narrow applicability. Methods Los Angeles Study The first study that we evaluated was conducted in the Fall/ Winter time period of 1996–1997 in the Los Angeles area by scientists at the Rancho Los Amigos Medical Center in Downey California; it is described in Linn et al. (1999). They focused on 30 individuals aged 56–83 year with clinically diagnosed chronic obstructive pulmonary disease (COPD). All participants lived at home, mostly in detached singlefamily homes. Each subject filled out a contemporaneous activity dairy for two sets of 4 days starting on a Thursday. The smallest time unit used was 20 min. The diary was very detailed concerning respiratory symptoms encountered over Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data each day, but was much more general regarding locations visited during each hour. Ultimately, by combining two types of data and using Boolean logic, we were able to assign participants to one of three locations for each 20-min time segment for every day. The recorded general locations were ‘‘outdoors’’ and ‘‘motor vehicles.’’ Time spent indoors per hour was calculated as 60 min minus the time spent outdoors or in a motor vehicle. All three locational categories have exposure ramifications depending upon the environmental pollutant being evaluated, so while being very general, the locations are useful from a modeling perspective. The outdoor location, for instance, is very important for understanding exposures to ozone (O3) and smaller or fine levels of particulate matter (PM2.5), which are ubiquitous ambient pollutants that frequently exceed US national air standards. The motor vehicle location is important for those pollutants emitted by autos and trucks, including nitrogen oxides, carbon monoxide, and PM (fine and coarse size fractions). Even indoor locations may result in adverse health effects if they contain pollution emissions, such as unvented gas stoves, fireplaces, and other combustion products. All records were evaluated for missing and inconsistent data by the original researchers. If data were missing from the diaries, they called respondents and asked them to recall their activities and fill in missing information. Even so, the data we were given had a number of hours with 460 min of coded locations. Given the tendency of people to not record short ‘‘transition times,’’ particularly associated with travel and outdoor times (McCurdy et al., 2000), we allocated ‘‘extra’’ time thusly: per-hour time is equal to (1) time spent in a vehicle, (2) time spent outdoors, and (3) time spent indoors, in that order. If hourly time spent in the first two locations exceeded 60 min, we used vehicle time as reported; time spent outdoors was set equal to the time remaining in the hour (or up to the value recorded, whichever was highest). Indoors was assigned the residual time, if any. There were less than 1% of the records affected by this decision rule. A rudimentary analysis of the data to compare basic statistics before and after the ‘‘problem hours’’ were adjusted showed small, insignificant differences in the average and median locational values. If the ‘‘problem’’ hours exceeded 10% of any day, the day was deleted from the dataset. One person was removed from our analyses based on this criterion. The modified dataset is designated simply as ‘‘Los Angeles’’ in this paper. Baltimore-CHAD Studies We combined measurement data from two panel studies conducted by EPA staff in the Baltimore (MD) area (Williams et al., 2000a, b, c). The two seasons were January–February 1997 and July–August 1998. The winter 1997 study was small, providing only 13% of the total person days of data; the summer 1998 study provided the remainder. Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Frazier et al. Measurements during the winter months were recorded only on weekdays. Activity data were obtained from a total of 26 individuals aged 65–89 years (mean ¼ 81 years) for 15 days, on average. The number of days per person ranged from 4 to 24. A 15-min ‘‘block’’ diary format was filled out by a technician during a daily face-to-face recall interview session with each subject. All participants lived in separate ‘‘domiciles’’ (a 74 m2 bed-sitting room) or in larger apartments in two separate independent retirement facilities, and all subjects were ambulatory. The facilities were quite selfcontained and included restaurants, a convenience store, beauty/barber shop, and so on. The vast majority of study participant’s total time (96%) was spent indoors during both monitoring periods (Williams et al., 2000a, b). Only the three general locational categories noted above are analyzed for this paper even though more locations were originally coded. We performed internal consistency checks on this database, including verifying that total accounted time in every hour was 60 min. When an activity code was unknown for an event, time spent in the corresponding location was coded as ‘‘missing.’’ Therefore, some of the respondents had o1440 min/day. About one-half of the sample had some missing time. In total, 13 persons had 1–5 entire days of missing data. Additionally, one person in the original dataset had only 1 day of diary data and was deleted from our analyses. The two-season data are part of the CHAD database, and the modified dataset is designated here simply as ‘‘Baltimore-CHAD’’ study. Statistical Analyses Characteristics of both samples were examined using simple descriptive statistics. For each study, the mean personminutes per day was calculated for the three analyzed locations as the total number of minutes for each person in each activity divided by the total number of days of activity. These values were averaged across the number of respondents for each location. Even though the number of men in the Baltimore CHAD study was small (n ¼ 6), we retained gender comparisons that are based on the number of person days. There were 303 female and 87 male person-days of data in the Baltimore study. With respect to the male respondents, one had 4 days of data; three had between 10 and 15 days of locational data; and two had information for 420 days. We examined the mean person-minutes per day, standard deviation, range, and coefficient of variation for several characteristics stratified by gender. The characteristics examined included day-of-the-week (Sunday, Saturday, and weekday); time-of-day by 6 h ‘‘block’’ (midnight to 6 am, 6:01 am to noon, 12:01 to 6 pm, 6:01 pm to midnight); season-of-the-year (fall, winter, and summer); and the interaction of season by day-of-the-week. We assessed normality of the data by examining the standardized skewness and kurtosis moments of the distributions 3 Variability in elderly person’s location data Frazier et al. for the three locations. As most of the data were not normally distributed, but approximated a log-normal distribution, we used the non-parametric, distribution-free Kolmogorov–Smirnov (KS) test to determine if the genderspecific location data came from statistically different distributions using the Dn metric, an approximate w2-statistic, and a ¼ 0.050. We investigated a series of models for a repeated measures design using the SAS PROC MIXED procedure (SAS Institute, 2006). It was used to test the fit and significance of the repeated measures models for each locational category, and allowed us to include both fixed effects and random effects in the model. The PROC MIXED procedure also allowed us to handle unequally spaced values, as well as missing repeated measures for individual subjects. This procedure has several techniques to estimate variance– covariance parameters of a model (Wolfinger and Chang, 1995). Common structures that can be analyzed are compound symmetry, an autoregressive model, and an unstructured model. The simplest correlation model is compound symmetry that assumes a constant covariance among all pairs of values. To decide on the best variance– covariance structure to use, we generated models using all three structures and employed a ‘‘penalized’’ likelihood based criteria described by Wolfinger and Chang (1995) to compare them. Specifically, we compared the information criteria of Akaike (AIC) and Schwarz (BIC) used in SAS and selected the model with the largest value. Based on these criteria, the compound symmetry structure was the best for both studies. We used the restricted maximum likelihood method, as these estimators tend to be less biased than maximum likelihood estimators (Wolfinger and Chang, 1995; Kao et al., 2003). We tested several models, starting from the simplest where only gender differences in mean time spent in the three locations were evaluated. Our initial models only included factors that had at least one statistically significant difference between men and women in the descriptive analysis. Health factors, including the respiratory symptoms were not included in the models. Second, we generated models for all characteristics (independent variables) that showed significant univariate statistical differences: time-of-day, day-of-the-week, and season-of-the-year. These models were used to test for differences among these characteristics after adjusting for gender. When notable significant differences were found, contrasts were generated to determine which groups differed and to identify patterns of differences for time spent at each of the locations. The Intraclass Correlation Coefficient We were interested in determining if the time spent per day in each location (indoors, outdoors, and in a motor vehicle) varied across time within individuals, as well as between individuals. Repeated measurements for each subject in both samples permit us to compare the amount of variation that 4 exists between individuals relative to the amount of variation reported within each individual (Hruschka et al., 2005). To assess variability in activity patterns, we examined our data using the ICC obtained from the R correlation matrix in the SAS PROC MIXED procedure, which gives the common covariance under compound symmetry (s2cs) between parameters in a model and its residual error (s2e). The ICC is computed thusly: ICC ¼ (s2cs)/(s2cs þ s2e). A low ICC indicates that within-individual variability in the respective studies was high relative to between-individual variability. Conversely, a high ICC suggests that there is high between-individual variation in locations relative to withinindividual variability. All P-values o0.05 were considered to be statistically significant. Finally, we analyzed day-to-day time spent in the three analyzed locations to understand the correlation structure of the locational categories. ‘‘Lag-one’’ (adjacent days) Spearman’s rank-order correlations (rs) were computed for specific day-pairs in the week; these correlations are useful for identifying repeated patterns in locational choices made by subjects in the Baltimore-CHAD and Los Angeles studies. Results Summary Information Basic comparisons were made to evaluate characteristics of the two elderly samples (Table 1). There were 29 subjects in the Los Angeles Study and 25 subjects in the BaltimoreCHAD studies after the modifications noted above were made. The age range was similar in both groups, but we could not compare their mean ages as the Los Angeles sample did not provide individual-specific ages. For characteristics that we could analyze, the two subpopulations were similar; one major difference was the subject’s gender breakdown: Baltimore was 76% women, whereas Los Angeles was 55% women. The most notable difference between the two cohorts was in living accommodations (Table 1). Climate is another important difference between the two areas. Winter in Baltimore certainly is more severe than in Los Angeles and should, among other impacts, modify the time spent outdoors in the two cities. A summary of the differences in time spent in the three locations for both cohorts is provided in Table 2. The Los Angeles subjects spent significantly more time outdoors than the Baltimore-CHAD population, whereas time spent in a vehicle was similar for the two samples. Box plots of the time spent outdoors by gender are depicted in Figures 1–4 for both cities. Univariate Statistics We investigated variability within an individual’s time spent outdoors (min/day) in each study sample. Most of the Baltimore participants had low day-to-day variation, with more variability Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data Frazier et al. Table 1. Selected characteristics of the Los Angeles and Baltimore-CHAD samples. Characteristics a Participants Age of subjects Subject-days of data Gender/racial distribution Diary instrument Living accommodations Calendar time period Health status Ambulatory status Smoking status Los Angeles sample Baltimore-CHAD sample 29 subjects 56–83b 8 days per subject: 232 person-days 16 females (55%) 13 males (45%) 100% Caucasians 20 min blocks for 24 h periods 100% lived at home, mostly in detached single-family homes. October–November 1996; December–February 1997 100% clinically diagnosed with severe COPD; some used O2 100% ambulatory 6.9% smokers; 93.1% non-smokers 25 subjects 72–89 (mean age±SD ¼ 81.2±5.4) 4–24 days per subject: 390 person-days 19 females (76%) 6 males (24%) 100% Caucasians 15 min blocks for 24 h periods 100% lived in retirement facilities, in separate ‘‘domiciles’’ or in apartments. January–February 1997; July–August 1998 Some individuals with COPD and/or cardiovascular disease; some were healthy. 100% ambulatory 100% non-smokers Abbreviations: CHAD, Consolidated Human Activity Database; COPD, chronic obstructive pulmonary disease. a One subject was excluded from our analyses for each dataset. b Age range was provided by Linn et al. (1999), but age for each person was not available. Table 2. Summary of time spent in selected locations in the Los Angeles and Baltimore-CHAD samples. Study Los Angeles Baltimore CHAD Time spent outdoors (min/day) Time spent in a motor vehicle (min/day) Time spent indoors (min/day) 62.7±62.3a (0–360)b 21.7±51.8 (0–450) 38.1±48.4 (0–240) 30.4±45.9 (0–225) 1,339.2±94.3 (1,060–1,440) 1,384.7±78.5 (885–1,440) Abbreviation: CHAD, Consolidated Human Activity Database. Mean±SD. Range of values for the location. a b within women. In the Los Angeles study, there was considerable variability in time spent outdoors for both genders. 300 Daily Time Time Spent Outdoors We examined time spent outdoors (Table 3), which indicates that women in Los Angeles spent, on average, more minutes per day outdoors than men; the differences are significant for the fall season, and during the fall on Sundays (Po0.05). For Baltimore participants, men generally spent more time outdoors compared to women, but the differences were not statistically significant. 400 200 100 0 26 24 23 20 19 17 16 15 14 12 11 9 Time Spent in Vehicles In the Los Angeles study, women reported more time spent in vehicles than men in most categories, except for the winter months and between the hours of 6:01 am and noon (Table 4). However, in the fall months of the year, as well as Saturdays during the fall, women were significantly more likely to spend time in motor vehicles than men in Los Angeles (Po0.05). In Baltimore, women were significantly more likely to spend time in vehicles than men, particularly in the summer, weekdays, weekdays in the summer, and between 12:01 and 6:00 pm. Men were more likely to be in a vehicle than women Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 8 7 6 5 4 3 2 1 Female Id Number Figure 1. Time spent (min/day) outdoors by women in the BaltimoreCHAD Study. during Sundays and between 6:01 pm and midnight in Baltimore. Time spent indoors Men were more likely to spend time indoors than women in both studies as shown in Table 5. In Los Angeles, men were significantly more likely to spent time indoors than women 5 Variability in elderly person’s location data Frazier et al. 500 Daily Time 400 300 200 100 0 25 22 21 18 Male Id Number 13 10 Figure 2. Time spent (min/day) outdoors by men in the BaltimoreCHAD Study. 400 Daily Time 300 200 100 0 1 4 5 6 8 14 15 16 17 18 19 21 28 30 Female Id Number Figure 3. Time spent (min/day) outdoors by women in the Los Angeles Study. 200 Daily Time 150 100 50 0 2 3 9 10 11 12 13 20 22 23 24 25 26 27 29 Male Id Number Figure 4. Time spent (min/day) outdoors by men in the Los Angeles Study. during the fall and Saturdays in the fall season. Men were significantly more likely to be indoors than women in the summer, weekdays during the summer, and during 12:01 to 6:00 pm in Baltimore. One-day lagged rs correlations for individuals were computed for both datasets. Due to the sampling frame 6 used, there were six paired-day values for subjects in Los Angeles: Thursday-Friday, Friday-Saturday, and Saturday-Sunday (each pair twice). Thus, the n is small for every individual, requiring a large rs value to be significant. As there were some people who did not go outside or use a vehicle at all, or only on 1 day of the six, there were a number of day-pair rs’s that could not be computed. For a Po0.05, no person in the Los Angeles study had a statistically significant 1-day rs for the in-vehicle location, and only 1 person had one for the outdoor location. Interestingly, however, there were five people with statistically significant 1-day rs’s for same-day time spent in a vehicle and outdoors. Time spent in those two locations were correlated on a daily basis for those individuals. These people constitute about 21% of the sample with valid 1-day lag pairs. If the a-critical value was relaxed to 0.10, 14% of the sample would have statistically significant 1-day rs’s for outdoor time and 23% for in-vehicle time. Same-day vehicle/outdoor Spearman’s correlations increase to 36% of the sample for the relaxed significance criterion. The Baltimore situation is more complex from the correlation perspective since the number of days of data per person varies between 4 and 24; seven had as many as 22 consecutive days. Again, because of missing days and/or zero time spent in the locations of interest, 1-day lag rs’s could not be computed for 4 of the 25 people in the Baltimore sample. Using Po0.05, no one had a significant 1-day lag rs for either in-vehicle or outdoor time. Between 15% and 19% of the elderly did have a significant rs for same-day in-vehicle and outdoor time, similar to Los Angeles, albeit a slightly smaller proportion of habitués for the two locations. Statistical Models Models were generated for each of the major factors: season of the year, time-of-day, and days of the week, after adjusting for the effects of gender. For our analyses, timeof-day was collapsed into two categories because of the large number of zero activities between midnight and 0600. The categories used for the model were 12:01 am to noon and 12:01 to midnight. Results of our mixed models analyses are depicted in Table 6. For time spent outdoors, gender alone did not explain differences in outdoor time (Model I), nor did season given (symbolized as: | ) gender or day-of-week|gender (Model II). However, there are significant differences in time spent outdoors in both studies when time-of-the-day is considered after adjusting for the effects of gender (Model II). The ICCs for gender alone are 0.35 in Los Angeles and a low 0.14 in Baltimore (indicating very high intraindividual variability). The ICCs for the season|gender Model II are 0.38 for Los Angeles and 0.13 for Baltimore. They are 0.37 and 0.15 for day-of-the-week|gender, and 0.26 and 0.09 for the time-of-day|gender for the two cities, respectively. Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data Frazier et al. Table 3. Descriptive statistics for time spent outdoors by gender and selected characteristics (min/day) for the Los Angeles and Baltimore-CHAD studies. Female Male KS P Mean±SD Range Mean±SD Range Dn Los Angeles Study Daily total 74.4±72.9 0–360 53.1±50.4 0–200 0.14 0.23 Time-of-day r0600 0601–1200 1201–1800 1801–2400 0.6±3.6 19.4±31.5 47.3±51.1 7.1±15.0 0–20 0–140 0–280 0–80 0.0±0.0 18.1±25.6 32.7±36.8 2.3±9.6 0–0 0–160 0–120 0–60 0.03 0.05 0.15 0.17 1.00 1.00 0.18 0.07 Season Fall Winter 85.0±67.5 62.1±77.7 0–320 0–360 47.3±48.0 78.3±54.0 0–180 0–200 0.31* 0.25 Day-of-the-week Weekday Saturday Sunday 73.1±68.9 76.9±77.5 74.6±78.7 0–320 0–360 0–260 56.6±53.8 53.1±46.7 46.3±47.8 0–200 0–140 0–180 0.14 0.21 0.27 0.62 0.57 0.24 Season  day type Fall weekday Fall Saturday Fall Sunday Winter weekday Winter Saturday Winter Sunday 82.9±75.2 84.3±44.5 90.0±74.3 61.7±60.4 68.3±105.6 56.7±83.0 0–320 0–140 0–260 0–220 0–360 0–220 52.3±52.9 46.2±46.9 38.5±37.9 75.0±56.0 83.3±34.4 80.0±72.7 0–180 0–140 0–140 20–200 20–120 0–180 0.24 0.43 0.48* 0.17 0.58 0.42 0.23 0.07 0.03 0.98 0.13 0.49 Baltimore-CHAD study Daily total 20.0±47.2 0–375 27.8±65.3 0–450 0.08 0.78 0.0±0.0 6.8±24.1 9.0±27.1 4.3±20.6 0–0 0–240 0–195 0–210 0.0±0.0 8.4±17.1 14.3±45.7 5.0±26.8 0–0 0–60 0–360 0–180 0.00 0.11 0.05 0.03 1.00 0.36 0.99 1.00 14.5±48.6 20.6±47.1 0–210 0–375 53.2±78.0 19.2±58.7 0–270 0–450 0.35 0.06 0.08 0.99 18.1±45.2 31.7±58.6 21.6±47.1 0–375 0–210 0–240 26.2±50.6 3.8±10.6 65.6±156.1 0–270 0–30 0–450 0.10 0.33 0.16 0.60 0.47 0.99 14.5±48.6 18.6±44.8 0–210 0–375 53.2±78.0 14.1±24.4 0–270 0–105 0.35 0.07 0.08 0.99 Time-of-day r0600 0601–1200 1201–1800 1801–2400 Season Winter Summer Day-of-the-week Weekday Saturday Sunday Season  day typea Winter weekday Summer weekday o0.01 0.27 Abbreviations: CHAD, Consolidated Human Activity Database; KS; Kolmogorov–Smirnov. Dn ¼ Dn metric computed from the KS test. *Indicates that the distributions are significantly different at Po0.05. a Data were collected only on Saturdays and Sundays during the summer. With respect to time spent outdoors by the elderly participants in the two cities, the Baltimore sample shows 2–3 times more intraindividual variability than the Los Angeles sample. Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 The opposite situation is found in the time spent in a motor vehicle and total time spent indoors. The ICCs for these two locations are lower in Los Angeles than in Baltimore, indicating more intraindividual variability in the 7 Variability in elderly person’s location data Frazier et al. Table 4. Descriptive statistics for time spent in a vehicle by gender and selected characteristics (min/day) for the Los Angeles and Baltimore-CHAD studies. Female Male KS Mean±SD Range Mean± SD Range Dn Los Angeles study Daily total 41.9±49.4 0–200 35.0±47.5 0–240 0.13 0.32 Time-of-day r0600 0601–1200 1201–1800 1801–2400 0.6±5.9 6.5±16.3 26.9±32.6 7.9±22.2 0–60 0–100 0–140 0–160 0.0±0.0 12.7±30.9 20.9±30.9 1.4±7.6 0–0 0–220 0–120 0–60 0.00 0.09 0.13 0.13 1.00 0.68 0.30 0.25 Season Fall Winter 53.2±51.8 28.8±43.4 0–200 0–180 32.3±49.0 46.7±39.0 0–240 0–120 0.24* 0.33 0.03 0.06 Day-of-the-week Weekday Saturday Sunday 39.6±51.7 50.0±39.7 38.5±54.2 0–200 0–160 0–200 39.4±45.9 33.1±52.9 28.1±45.4 0–180 0–240 0–220 0.05 0.35 0.18 1.00 0.06 0.77 Season  day type Fall weekday Fall/Saturday Fall/Sunday Winter/weekday Winter/Saturday Winter/Sunday 43.6±52.8 64.3±36.9 61.4±61.5 35.0±51.1 33.3±37.5 11.7±27.3 0–200 0–160 0–200 0–180 0–100 0–80 38.1±48.2 26.2±51.2 26.9±49.0 45.0±35.3 63.3±54.3 33.3±27.3 0–180 0–240 0–220 0–100 0–120 0–60 0.11 0.65* 0.33 0.38 0.33 0.50 0.98 o0.01 0.28 0.21 0.77 0.27 Baltimore-CHAD study Daily total 34.2±46.4 0–225 17.4±42.5 0–210 0.29* o0.01 Time-of-day r0600 0601–1200 1201–1800 1801–2400 0.0±0.0 11.1±19.8 19.6±30.3 3.6±15.6 0–0 0–120 0–180 0–120 0.0±0.0 8.8±23.7 3.6±11.2 5.0±20.7 0–0 0–120 0–60 0–150 0.00 0.16 0.32* 0.02 1.00 0.05 o0.01 1.00 Season Winter Summer 48.5±74.5 32.7±42.1 0–225 0–195 30.7±62.9 12.9±32.5 0–210 0–165 0.14 0.31* 0.97 o0.01 Day-of-the-week Weekday Saturday Sunday 34.4±48.2 29.6±42.8 38.4±35.5 0–225 0–150 0–105 12.3±38.7 3.8±10.6 76.9±51.0 0–210 0–30 30–165 0.36* 0.30 0.38 o0.01 0.59 0.33 Season  day typea Winter weekday Summer weekday 48.5±74.5 32.3±43.0 0–225 0–195 30.7±62.9 4.0±15.2 0–210 0–90 0.14 0.43* 0.97 o0.01 P Abbreviations: CHAD, Consolidated Human Activity Database; KS; Kolmogorov–Smirnov. Dn ¼ Dn metric computed from the KS test. *Indicates that the distributions are significantly different at Po0.05. a Data were collected only on Saturdays and Sundays during the summer. California city; however, the ICCs are more comparable for the in-vehicle and indoor locations in the two cities. With respect to explaining total variance in the two locations for both cities, time-of-day|gender provides significant results (a 8 Model II P-value o0.02 in both locations). In addition, dayof-week|gender is a significant explanatory variable for time spent in both in-vehicle and indoor locations in the Baltimore-CHAD data (Model II P-values o0.01). Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data Frazier et al. Table 5. Descriptive statistics for time spent indoors by gender and selected characteristics (min/day) for the Los Angeles and Baltimore-CHAD studies. Female Male KS Mean±SD Range Mean±SD Range Dn 1323.7±101.4 1060–1420 1351.9±86.6 1080–1440 0.14 0.22 358.8±6.7 334.0±40.3 285.8±70.3 345.0±35.0 300–360 200–360 80–360 160–360 360.0±0.0 329.2±47.3 306.4±61.7 356.3±16.8 360–360 100–360 120–360 240–360 0.04 0.07 0.15 0.18 1.00 0.97 0.11 0.05 Season Fall Winter 1301.8±97.9 1349.2±100.4 1080–1440 1100–1440 1360.4±86.3 1315.0±79.0 1080–1440 1200–1440 0.29* 0.27 Day-of-the-week Weekday Saturday Sunday 1327.3±100.1 1313.1±98.1 1326.9±110.1 1080–1440 1060–1440 1100–1440 1344.1±92.5 1353.8±85.4 1365.6±75.6 1080–1440 1140–1440 1180–1440 0.11 0.23 0.21 0.87 0.43 0.54 Season  day-of-week Fall weekday Fall Saturday Fall Sunday Winter weekday Winter Saturday Winter Sunday 1313.6±103.0 1291.4±71.3 1288.6±113.3 1343.3±96.3 1338.3±120.7 1371.7±91.2 1080–1440 1140–1440 1100–1440 1080–1440 1060–1440 1220–1440 1349.6±96.2 1367.7±81.2 1374.6±68.9 1320.0±72.9 1293.3±82.6 1326.7±96.9 1080–1440 1140–1440 1180–1440 1200–1420 1220–1420 1200–1440 0.21 0.51* 0.37 0.25 0.33 0.41 0.42 0.02 0.17 0.70 0.77 0.49 Baltimore-CHAD study Daily total 1383.8±72.5 1035–1440 1387.9±96.9 885–1440 0.12 0.24 359.9±1.7 341.7±32.9 330.5±43.9 351.7±32.3 330–360 120–360 150–360 60–360 360.0±0.0 341.0±30.3 339.1±53.3 347.8±44.2 360–360 225–360 0–360 30–360 0.00 0.06 0.21* 0.03 1.00 0.97 o0.01 1.00 Season Winter Summer 1368.0±92.9 1385.5±69.9 1140–1440 1035–1440 1339.8±126.3 1404.2±79.5 1020–1440 885–1440 0.14 0.19* 0.97 0.04 Day-of-the-week Weekday Saturday Sunday 1384.9±71.2 1379.6±93.1 1380.0±56.7 1035–1440 1125–1440 1200–1440 1393.1±85.3 1432.5±13.8 1297.5±174.7 1020–1440 1410–1440 885–1410 0.15 0.34 0.28 0.18 0.43 0.69 Season  day typea Winter weekday Summer weekday 1368.0±92.9 1387.4±67.5 1140–1440 1035–1440 1339.8±126.3 1417.0±41.8 1020–1440 1215–1440 0.14 0.25* 0.97 0.01 Los Angeles study Daily total Time-of-day r0600 0601–1200 1201–1800 1801–2400 Time-of-day r0600 0601–1200 1201–1800 1801–2400 P o0.01 0.19 Abbreviations: CHAD, Consolidated Human Activity Database; KS; Kolmogorov–Smirnov. Dn ¼ Dn metric computed from the KS test. * Indicates that the distributions are significantly different at Po0.05. a Data were collected only on Saturdays and Sundays during the summer. In sum, although gender alone does not explain time spent in any of the three location categories analyzed, it does when combined with time-of-day information, and, in the Baltimore-CHAD studies, when combined with day-of-week Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 information. The ICCs for the three locations are all lowto-modest, indicating considerable intraindividual variability in the data. The pattern of the ICCs in the two cities however is not consistent. 9 Variability in elderly person’s location data Frazier et al. Table 6. Results of the mixed models analyses for each location and their intraclass correlation coefficients for the Los Angeles and Baltimore-CHAD studies. Los Angeles Variables in Model F Time spent outdoors Model a Gender 1.98 P Baltimore CHAD ICC F P ICC 0.17 0.35 0.91 0.35 0.14 0.05 0.82 70.54*,c o0.01 0.18 0.84 0.38 0.26 0.37 1.49 0.24 9.94* o0.01 3.24*,c 0.05 0.13 0.09 0.15 0.78 0.17 1.31 0.26 0.30 Model II b Season|gender 0.22 0.64 Time-of-day|gender 43.21*,c o0.01 Day-of-week|gender 0.57 0.57 0.17 0.11 0.18 1.68 0.21 6.63*,c 0.02 12.06*,c o0.01 0.30 0.19 0.32 Time spent indoors Model I a Gender 0.14 0.26 Model II b Season|gender 0.00 0.97 Time-of-day|gender 76.21*,c o0.01 Day-of-week|gender 0.45 0.64 0.27 0.18 0.29 Model II b Season|gender Time-of-day|gender Day-of-week|gender Time spent in a vehicle Model I a Gender 0.08 2.33 0.00 0.95 0.30 3.24 0.09 18.04* o0.01 10.31*,c o0.01 0.29 0.20 0.33 Abbreviations: CHAD, Consolidated Human Activity Database; ICC, intraclass correlation coefficient. *Significant at Po0.050. Note: The | symbol is read as ‘‘given.’’ a Model I has one dependent variable (location) and one independent variable (gender). b Model II has one dependent variable (location), two independent variables (gender and the variables listed), and their interaction. c There was a significant interaction between the variable listed and gender for this model. Discussion The present study analyzes multiday human activity data for two elderly cohorts living in Los Angeles (CA) and Baltimore (MD). We evaluated various descriptive statistics to examine relationships between locational data and specific characteristics for these unique groups. Few studies exist that have provided statistics indicating how locational measurements are correlated among the elderly. Assessments using the ICC provide a better understanding of population variability in the time spent in the selected locations. In both the Los Angeles and Baltimore elderly groups, most of the daily time was spent indoors. Men spent slightly more time indoors than women in both study groups. The Los Angeles sample spent significantly more time outdoors than the Baltimore sample (about double). Although we do 10 not have daily meteorological information for the study periods, it is likely that the days were warmer and drier in California than in Baltimore, conducive to going outside. Correlations between adjacent day pairs were strong (r40.50) for both elderly groups for time spent indoors. This is not surprising as this group of elderly individuals spend over 90% of their day indoors. The lag-one correlation coefficients for the other two locations of interest are mixed, and do not indicate strong day-to-day patterns in either city. There are stronger rs correlations between outdoor time and in-vehicle time on a same-day basis than for either location on a between-day basis. This type of information is important to EPA in understanding relationships among locations in the time spent by the elderly over various days of the week. In reviewing the published data suitable for putting the work described here into perspective, there is location and activity information on persons 464-year-old contained in EPA reports that describe or use the Agency’s National Human Activity Pattern Survey (NHAPS). See for instance: Tsang and Klepeis (1996) and the National Center for Environmental Assessment (1997). NHAPS data are included in CHAD, as are other time-activity studies containing elderly data (known as: ‘‘California,’’ ‘‘Cincinnati,’’ ‘‘Denver,’’ ‘‘Washington DC,’’ and ‘‘Valdez’’). See McCurdy et al. (2000) for a more detailed discussion of studies in CHAD. Graham and McCurdy (2004) present selected activity data on persons 464 years contained in CHAD. Although all studies in CHAD were evaluated, the authors focused only on those persons-days of data having 429 events, the minimum number that they felt represented a ‘‘compliant’’ respondent. The number of days of compliant elderly data in CHAD is 1758, about 82% of the total available for persons 464 years. Most of the ‘‘dropped’’ days came from the recall surveys in CHAD. The authors differentiate between the total population and habitués – those people who actually spend time in a particular location (if a person does not spend any time in a location, then he or she cannot get an exposure in it). Almost 57% of people 464 years in CHAD spend 1 or more min/dayoutdoors. The median number of minutes spent outdoors in compliant 464 years habitués, with or without health issues, is 65 min/day, with a mean of 118±130 and a range of 1–840. The mean for all persons 464 years is 67.3 min/day, including those who do not go outdoors at all. Gender differences for the compliant elderly are: women: median ¼ 60 min/day (mean ¼ 88±98; range ¼ 1–630) and men: median ¼ 118 min/day (mean ¼ 164±156; range ¼ 1–840). Thus, the Graham and McCurdy (2004) analysis indicates that older men spend almost twice as much time outdoors per day as do older women. The time spent in motor vehicles was also analyzed for people 464 years, although gender differences were not presented by Graham and McCurdy (2004) as there was no significant difference in this location by sex (for any age group, with some minor exceptions). The median vehicular Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data Frazier et al. time for compliant 464 years habitués is 60 min/day (mean ¼ 86±85; range ¼ 4–789 min), with 72% of 464 years people in CHAD spending 41 min/day in a motor vehicle. The mean vehicular time for all compliant persons 464 years in CHAD, including those who do not enter a vehicle at all, is 61.9 min/day (Graham and McCurdy, 2004). Similar data are presented in Leech and Smith-Doiron (2006) for 70 elderly persons with diagnosed COPD that participated in a time-activity recall survey in Ottawa, Canada in May of 2002. The mean age of the patients was 69 years, but an age range was not provided (probably it was 60–75 years). The time spent outdoors by this subgroup was 93.6±14.4 min/day (converting their data from percentage of daily time: 6.5%±1.0%). Similar data for the time spent in motor vehicles is 53.3±8.6 min/day (3.7%±0.6%). The authors compared their COPD patients with 272 people aged 60–75 in a Canadian time survey conducted in waves in 1994–1995 (Leech et al., 1996). That survey, called CHAPS, was for the general population and included people with a history of chronic bronchitis. There was no statistically significant difference in the time spent outdoors or in motor vehicles between the COPD and similar CHAPS cohorts using an unpaired t-test and unequal variance (Leech and Smith-Doiron, 2006). Liu et al. (2003) provide time-activity data for a convenience sample of elderly individuals 465 years of age in the Seattle area during the 1999–2001 time period. They were all considered to be ‘‘susceptible’’ from a PM exposure assessment perspective, and were classified into three groups (the fourth group, asthmatic children, is not of concern here). The elderly groups were: (1) healthy, defined to be people without any signs or symptoms of cardiorespiratory disease (n ¼ 28), (2) individuals with diagnosed COPD (n ¼ 34), and (3) individuals with coronary heart disease (CHD; n ¼ 27). Age or gender distinctions for these groups were not provided, although all but one was 465 years, and 85% of them were between 71 and 90 (Liu et al., 2003; p. 909). Data on time spent in various locations were provided as a percentage of total time in that study. We converted it into mean min/day estimates, with the following results compared to the Los Angeles and Baltimore-CHAD elderly cohort data. Time spent outdoorsa Time spent in vehicles (min/day) Range (min/day) Range 41.7±24.5 27.4±18.7 25.9±21.6 0–210 0–156 0–229 57.6±36.0 51.8±33.1 46.1±27.3 7–134 1–132 0–105 Current study Los Angeles elderly 62.7±62.3 Baltimore-CHAD elderly 21.7±51.8 0–360 0–450 38.1±48.4 30.5±46.0 0–240 0–225 Liu et al. (2003) Healthy elderly CHD patients COPD patients a Two locations in Liu et al (2003) were combined for these estimates: ‘‘Yard’’ and ‘‘Outdoors.’’ Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 There was no reported testing to identify statistical significant differences among the three elderly groups for the two locations in Liu et al. (2003). The authors present results of using multiple regression analysis to determine which of their measured independent variables explained the most variance in the proportion of time spent outdoors; the independent variables used were age, gender, season (a categorical variable), and health status (also a categorical variable) for healthy, CHD, and COPD persons. The regression results were indeterminate (R2 ¼ 0.23, P ¼ 0.43), and only age (P ¼ 0.01) and COPD status (P ¼ 0.03) were statistically significant explanatory variables, both having a negative sign (Liu et al., 2003). Finally, Krantz-Kent and Stewart (2007) provide information on ‘‘travel’’ for age/gender cohorts based on data obtained by the American Time Use Study (ATUS) conducted by the US Bureau of Labor Statistics. Mean travel time was higher in older men than women, and decreased between the 65 and 69 and 70 þ age brackets in both genders. That gender-difference finding is just the opposite of what we found in both Baltimore-CHAD and Los Angeles studies. The mean ATUS travel time estimates for both genders are 50% or so higher than in our two study areas, but that probably is due to the difference in measuring ‘‘travel,’’ as an activity, versus ‘‘in vehicles’’ as a location. This ambiguity also occurs in earlier studies of elderly time use, such as Lawton et al. (1986), Moss and Lawton (1982), and Verbrugge et al. (1996). The ICCs for the two elderly cohorts in this study are marginally higher than those seen in previous analyses of middle-aged working adults (Isaacs et al., 2007) and children (Xue et al., 2004). This indicates that there is more relative intraindividual variability in the elderly than these other groups relative to the other cohorts. This superficially is a counter-intuitive finding. Perhaps it reflects the fact that because elderly people in the two studies analyzed here are retired, they do not have a daily ‘‘fixed’’ work schedule that effectively reduces inter-individual variability in the sampled population. The lack of a fixed schedule allows elders to visit a location 1 day but not the next, if so desired. Also, since the mean time spent in the outdoor and in-vehicle locations are generally less for elderly people than for the other two cohorts analyzed in Isaacs et al. (2007) and Xue et al. (2004), within-person variability becomes relatively more important than between-person variability, even if the variances of the two metrics do not change. Undoubtedly, more information on this topic is needed before any definitive conclusions can be drawn about these inferences. Shortcomings of this study are those associated with any panel study: small sample sizes and lack of representativeness, even in the elderly with preexisting respiratory disease. Not all housing types are covered, for instance, nor were all possible lifestyles addressed. In addition, we used secondary data for analysis, although two first-author investigators 11 Variability in elderly person’s location data Frazier et al. from the original studies were involved with this one. Given the intensive nature of gathering longitudinal human activity data, and the costs involved, elderly time-budget data will of necessity be from panel studies whose findings will have to be generalized to larger populations using meta-analytic approaches. Conclusions This paper presents descriptive statistical analyses of selected time use data on some general locations frequented by (mostly) health-compromised individuals aged 56–89 in two US cities, Baltimore, MD (n ¼ 26) and Los Angeles, CA (n ¼ 30). Between 4 and 24 days of activity data exist for each participant. Both studies were conducted in the 1996– 1998 time period. The locations specifically analyzed are outdoors, in a motor vehicle, and total indoors; the unit of measurement is minutes per day in each location. The mean time spent outdoors by subjects in Los Angeles (63 min/day) is almost three times that of Baltimore subjects (22 min/day). The difference in mean time spent in motor vehicles is much narrower, 38 versus 30 min/day for the two cities, respectively. Between-study testing of these data was not undertaken because of (1) the different data gathering protocols used, and (2) the quite different housing types found in our two samples (predominantly single-family residences in Los Angeles and apartment-like domiciles in Baltimore). Gender differences within the cities were tested, however, and indicate that women spend significantly more time in motor vehicles in Baltimore than men. Other significant gender differences exist, but they are time-of-day, season-ofthe-year, and/or day-of-week dependent and no overall pattern emerges (Tables 2–4). There are low day-to-day correlations in time spent in the outdoor and in-vehicle locations for individuals in the two panel studies analyzed here. The ICC metric is used to evaluate relative intra- and interindividual variability in the time spent in various locations, using three different mixed effects models (Table 5). The ICCs generally are low-to-modest, indicating a high degree of within-person variability in time spent in the analyzed locations. This is found in both cities. City-to-city variability in the time that older subjects spend in three locations of interest, plus the high degree of withinperson variability in this time, indicates that exposure models focused on the elderly will have to be city specific, evaluate the genders separately, and capture intraindividual variability in daily patterns to adequately mimic reality. Acknowledgements This research was supported by a grant to Dr. Frazier through the NAFEO (National Association for Equal 12 Opportunity in Higher Education) Program funded by the US. Environmental Protection Agency’s Washington D.C. Office of Research and Development (ORD). Troy Rutkofske of ORD’s Integrated Services Staff and Rachel Cooke of NAFEO expedited the grant, and we thank them for their assistance. We also greatly appreciate the assistance of Dr. Graham Glen and Dr. Luther Smith of Alion Scientific in Durham NC for the Baltimore data from CHAD. The research reported by Linn et al. (1999) was supported by the Electrical Power Research Institute (grant no. WO-3215). This paper has been subjected to Agency review and approved for publication. Mention of trade names, commercial products, and organizations does not constitute endorsement or recommendation for use. We thank the Journal’s two peer-reviewers for their thorough reading of the paper; we made many changes in the paper based on their comments and suggestions. References Federal Interagency Form on Aging. Older Americans: Update 2006. Government Printing Office, Washington DC, 2006. Geller A.M., and Zelnick H. Aging and the environment: a research framework. Environ Health Perspect 2005: 13: 1257–1262. Glen G., Smith L., Isaacs K., McCurdy T., and Langstaff J. A new method of longitudinal diary assembly for human exposure modeling. J Expo Sci Environ Epidemiol 2008: 18: 299–311. Graham S.E., and McCurdy T. Developing meaningful cohorts for human exposure models. J Expo Sci Environ Epidemiol 2004: 14: 23–43. Hruschka D.J., Kohrt B.A., and Worthman C.M. Estimating between and withinindividual variation in cortisol levels using multilevel models. Psychoneuroendocrinology 2005: 30: 698–714. Isaacs K., Errickson A., Forbes S., Glen G., McCurdy L., McCurdy T., et al. Statistical properties of longitudinal time-activity data for use in human exposuremodels. Paper presented at the Annual Meeting of the International Society of Exposure Analysis; Durham NC 2007. Kao T.C., Sparling Y., and Rochon J. Mixed-effect models to assess consistency and reliability across multiple evaluations. J Biopharm Stat 2003: 3: 539–548. Krantz-Kent R., and Stewart J. How do older Americans spend their time? Monthly Labor Rev 2007: 130: 8–26. Lawton M.P., Moss M., and Fulcomer M. Objective and subjective uses of time by older people. Int J Aging Human Develop 1986: 24: 171–188. Leech J.A., and Smith-Doiron M. Exposure time and place: do COPD patients differ from the general population? J Expo Sci Environ Epidemiol 2006: 16: 238–241. Leech J.A., Wilby K., McMullen E., and Laporte K. The Canadian human activity pattern survey: report of methods and population surveyed. Chron Dis Canada 1996: 17: 118–123. Linn W.S., Gong Jr H., Clark K.W., and Anderson K.R. Day-to-day particulate exposures and health changes in Los Angeles area residents with severe lung disease. J Air Waste Manage Assoc 1999: 49: PM108–PM115. Liu L.-J.S., Box M., Kalman D., Kaufman J., Koenig J., Larson T., et al. Exposure assessment of particulate matter for susceptible populations in Seattle. Environ Health Perspect 2003: 111: 909–918. McCurdy T., Glen G., Smith L., and Lakkadi Y. The National exposure research laboratory’s consolidated human activity database. J Expo Anal Environ Epidemiol 2000: 10: 566–578. McCurdy T., and Graham S. Using human activity data in exposure models: analysis of discriminating factors. J Expo Anal Environ Epidemiol 2003: 13: 294–317. McGraw K.O., and Wong S.P. Forming inferences about some intraclass correlation coefficients. Psychol Methods 1996: 1: 30–46. Moss M.S., and Lawton M.P. Time budgets of older people: a window on four lifestyles. J Gerontol 1982: 17: 115–122. Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 Variability in elderly person’s location data National Center for Environmental Assessment. Exposure Factors Handbook. Vol. 3. U.S. Environmental Protection Agency, Washington DC, 1997 (EPA/600/ P-95/002Fc). Quackenboss J.J., Spengler J.D., Kanarek M.S., Letz R., and Duffy C.P. Personal exposure to nitrogen dioxide: relationship to indoor/outdoor air quality and activity patterns. Environ Sci Technol 1986: 20: 775–783. SAS Institute. SAS/STAT User’s Guide, Version 9.2 Edition SAS Institute Inc, Cary, NC, 2006. Stewart A.L., Mills K.M., King A.C., Haskell W.L., Gillis D., and Ritter P.L. CHAMPS physical activity questionnaire for older adults: outcomes for interventions. Med Sci Sports Exercise 2001: 33: 1126–1141. Tsang A.M., and Klepeis N.E. Descriptive Statistics Tables from a Detailed Analysis of the National Human Activity Pattern Survey (NHAPS) Data. U.S. Environmental Protection Agency (EPA/600/R-96/148), Las Vegas, 1996. Verbrugge L.M., Gruber-Baldini A.L., and Fozard J.L. Age differences and age changes in activities: Baltimore Longitudinal Study of Aging. J Gerontol B: Psychol Sci Soc Sci 1996: 51B: S30–S41. Verghese J., Lipton R.B., Katz M.J., Hall C.B., Derby C.A., et al. Leisure activities and the risk of dementia in the elderly. N Engl J Med 2003: 348: 2508–2516. Williams R., Creason J., Zweidinger R., Watts R., Sheldon L., and Shy C. Indoor, outdoor, and personal exposure monitoring of particulate air pollution: the Journal of Exposure Science and Environmental Epidemiology (2008), 1–13 View publication stats Frazier et al. Baltimore elderly epidemiology-exposure pilot study. Atmos Environ 2000a: 34: 4193–4204. Williams R., Suggs J., Creason J., Rodes C., Lawless P., Kwok R., et al. The 1988 Baltimore particulate matter epidemiology-exposure study: part 2-personal exposure associated with an elderly study population. J Expo Anal Environ Epidemiol 2000b: 10: 533–543. Williams R., Suggs J., Zweidinger R., Evans G., Creason J., Kwok R., et al. The 1988 Baltimore particulate matter epidemiology-exposure study: part 1comparison of ambient, residential outdoor, indoor and apartment particulate matter monitoring. J Expo Anal Environ Epidemiol 2000c: 10: 518–532. Winer BJ Statistical Principles in Experimental Design. 2nd edn. McGraw-Hill Book Co., New York, 1971. Wolfinger R., and Chang M. Comparing the SAS GLM and MIXED Procedures for Repeated Measures. Proceedings of the Twentieth Annual SAS Users Group Conference SAS Institute, Cary NC, 1995. Xue J., McCurdy T., Spengler J., and Özkaynak H. Understanding variability in time spent in selected locations for 7–12-year old children. J Expo Anal Environ Epidemiol 2004: 4: 222–233. Zartarian V., Bahadori T., and McKone T. Adoption of an official ISEA glossary. J Expo Anal Environ Epidemiol 2005: 15: 1–5. 13