This study was designed to evaluate longitudinal variations in time-activity patterns. Data collected in this study allow us to evaluate inter- and intra-individual as well as longitudinal variations in human time-activity patterns.
The time-location/activity data we collected are substantially in agreement with two large time-activity studies conducted in the U.S. The California Study of Children's Activity Patterns survey (April 1989 - February 1990) collected 24-hour recall from 1200 children under 12-years-of age, half of whom were age 6 or above [4
]. This group reported that children under 12-years-of age spent 1,078 minutes/day (18 hours/day) at home and 109 minutes/day in school/childcare on average. Younger children (1 to 6-year-old) spent more time at home and less time in school than older children (6 to 11-year-olds), which echoes our results in SUPERB where children spent 1134 minutes/day (19 hours/day) at home and 105 minutes/day in school/childcare on average. The time children spent in transit was similar in both studies too, that is, 69 minutes/day in the California Children's Study vs. 61 minutes/day for children under 12-years-of age in our study on average.
A second key study is the National Human Activity Pattern Survey (NHAPS). NHAPS collected 24-hour recall from 9,368 respondents across all age groups in the U.S between October 1992 and September 1994 [5
]. We compared our data with a subset of NHAPS data, those collected from respondents in comparable age groups who lived in California. Note that adults in NHAPS did not necessarily have young children. The time spent in selected microenvironments/activities reported in the two studies were generally similar, with discrepancy mainly observed for the age group of parents of young children (Figure ). Parents of young children in SUPERB spent more time at home and less time in transit. In particular, fathers spent much less time in transit on weekends, and mothers reported shorter time spent working on weekdays but longer time on weekends compared to NHAPS adults in the same age group. Craig and Mullan (2010) also reported that parents of young children spent on average five more hours per day on child care and housework than childless men and women, and in particular, mothers spent 2.8 fewer hours working than childless women in the same age group [20
]. Apart from this difference, the time allocation observed in SUPERB is similar to the range reported in NHAPS.
Figure 2 Comparison of time-location/activity data between SUPERB and NHAPS by day-type, gender and age group. The solid bars represent the means and the error bars represent 95% confidence intervals of the means. For NHAPS data, the sample size was 84 for children (more ...)
Our findings concerning day-type and seasonal variations are consistent with Echols et al. (1999) [12
]. They examined variations in 80 individuals (primarily adults) by day-of-week (i.e., Monday through Sunday), cycle (six sampling cycles over a year), and individual. Significant day-of-week variations were observed for all seven time-location categories they examined, i.e., more time spent in transit and at work/school (both in and outdoors) on weekdays compared to more time spent at home (both in and outdoors) and other locations (both in and outdoors) on weekend days. Since they did not include season as a covariate in the model, the variations by cycle they reported were effectively the same as seasonal variability, e.g., less time spent at work/school indoors but more time spent at home outdoors and in other outdoor locations in the warm months (June to September), which are consistent with the seasonal variations of time spent at home, in transit, and in the school and office we reported for SUPERB participants.
Previous studies examining variations in time-location/activity patterns were based on very specific populations or were more restricted with regard to locations included. Our study provides additional evidence that confirms previous findings. A previous study conducted by McCurdy and Graham analyzed data on 57-60 year old males and found that day-type and season greatly influenced time spent indoors and outdoors but not time spent in motor vehicles [14
]. In contrast, Frazier et al. (2009) reported little day-type and seasonal variability for time spent indoors, outdoors or in motor vehicles in a cohort of elderly aged 56-83 with chronic obstructive pulmonary disease in Los Angeles; day-type variation was stronger in a cohort of elderly (aged 65-89) in Baltimore [7
]. In our study, 17% of the older participants had heart disease and 23% had asthma, while 63% of the older participants considered their overall health condition healthy. The time that older adults spent in total indoors, outdoors, or in vehicles did not significantly varied by day-type, but day-type variation was observed for time spent in specific microenvironments, i.e., home and places for work, shopping, eating or running errands. We also observed that older adults spent more time outdoors in the warm season than in the cool season (see Additional File 2
Graham and McCurdy (2004) considered age and gender as the primary factors to define a cohort in a time-activity study [21
]. We did observe statistically significant impacts of sex, age and employment status on time spent in some locations and on some activities. However, our younger adult population comprised solely of parents of young children, which may have influenced their time-activity patterns. Therefore, the variation by sex and age we observed may not be generalizable to populations of different characteristics (employment, family size).
Previous studies used ICC values, the ratio of between-subject variance to total variance, as a measure of for inter-individual variability. Xue et al. (2004) and Frazier et al. (2009), based on diaries collected daily from elementary school children and a cohort of elderly, respectively, obtained ICCs ranging from 0.10 to 0.35 for time spent indoors, outdoors and in a vehicle [2
]. We obtained ICCs in the same range if we allocate time in a similar way (see Additional File 2
). The ICCs for time spent on exposure-related activities that were asked in the web-survey questionnaire are generally higher than the ICCs for time spent in different locations that were asked in the 24-hour recall diary, indicating greater variability in activity patterns than space transition between individuals.
One of the unique contributions of the SUPERB web survey was that time-activity data was collected over an extended 18-month period, allowing us to evaluate longitudinal variation over a longer period than previous studies. We found that time-activity patterns were basically consistent over the study period, with some exceptions possibly related to human physiological changes and socio-economic factors. Results suggest that day-type is a primary source of variation for time-activity patterns, with season a second and usually predictable source; beyond these, human time-activity patterns were basically consistent over time. Therefore, for exposure modeling purposes, researchers should use cross-sectional time-activity surveys to collect baseline human activity pattern on different types of days (weekday and weekend), and then account for seasonal variations. For seasonal variations, one could either collect data in different seasons or estimate seasonal variation by incorporating known seasonal variation into the model. In addition, long-term time-activity patterns due to social and economic changes, which have not been paid attention to before, should be included. For young children and older people, physiological changes also need to be considered. We recommend use of supplemental questionnaires to collect the frequency of exposure-related activities that happen less often and therefore may not be captured on a single sampling day, e.g., vigorous outdoor activity, pumping gas, barbeque, etc.
Another contribution of our study is the investigation of intra- and inter-individual variation in many more locations and for more activities than studied before. We extended the limited number of micro-environments (home, work/school, transit and other locations) or standard activity categories (indoors, outdoors and in-vehicle). We collected longitudinal data on ~30 location/activity categories and examined the intra- and inter-individual variation in half of them, including restaurants, several types of stores, parks, health clubs, etc. Our findings thus greatly expand the current knowledge about the variation in human time-activity patterns for the three age groups.
Furthermore, we introduced mixed-distribution mixed-effects modeling to analyze time-activity data in which a large percentage of participants are assigned a zero value. Compared to the traditional non-parametric method, e.g., relying on the Kolmogorov-Smirnov test, this method provides a better solution for time-activity data that are not normally distributed, because we were able to simultaneously assess both the likelihood and the duration of the time spent in microenvironments/activities.
A limitation of this study is that participants gradually withdrew or dropped out over time, hindering the evaluation of longitudinal variation. As reported in Wu et al., out of the 206 households, 56% of parents of young children and 24% of older adults did not complete the study. A large number (84%) of parents of young children could not complete all required surveys due to limited time and family responsibility [17
]. To retain maximum information, we included all valid diaries into the statistical summary, but only the participants who completed two or more surveys into the longitudinal analysis. In addition, web surveys allow participants to select any day for recall, and they may select a convenient day. Specifically, participants may select an unrepresentative day with fewer activities than typical in order to minimize reporting effort, for example, a nurse could select a weekday that he/she was not on shift for recall. We did not include the diaries with very few location/activity changes in this analysis, but this trend helps explain why participants reported longer sleep time but shorter work hours in our study, and thus may influence the estimation of intra- and inter-individual variations.
Secondly, our definitions of moderate and vigorous activities were not precise, and tend to over-estimate the activity level. According to the activity metabolic equivalent (MET) intensity levels published by Ainsworth et al. (2000), activities with METs ranging 3-6 were considered moderate and those with METs above 6 were considered vigorous. For example, walking can be light to vigorous depending on the speed (METs ranging 2.0 to 12.0), and bicycling varies from moderate to vigorous with METs of 4.0 to 16.0 [22
]. Food preparation is more commonly considered to be a light activity with METs between 2.0 and 3.0. Comparisons with available data on strictly-defined moderate and vigorous physical activities (MPA & VPA) also suggest overestimation of time spent on these activities. According to the State Indicator Report on Physical Activity, 67% and 45% of adults in California are physically active and highly active, respectively [23
]. Ainsworth et al. (2000) reported that adults in their 40's spent 16 min/day in MPA and 18 min/day in VPA [22
]. Sallis et al. (1985) investigated moderate and vigorous physical activities among 2126 adults between 20 and 74 years old [24
]. They found that 84% of respondents engaged in moderate activities and spent 50-83 minutes/day on average depending on respondents' age; 15% of respondents reported vigorous activities and spent 17-68 minutes/day on vigorous activities. Compared to their study, we over-estimated the percentages of doers of moderate and vigorous activities by around 7% and 12%, respectively, and over-estimated the time spent on moderate and vigorous activities by approximately 3 times.
Lastly, in our method evaluation step, we compared the web survey data with the time-activity data collected by telephone interview from the same respondents and obtained similar distributions of the time-allocation measured by these two methods [17
]. However, since these two types of the surveys were not conducted for the same day for a respondent, such comparison does not fully establish the validity of the web survey method. More objective methods, such as Global Position System (GPS) recordings, may provide better reference values to determine the validity of the self-reported location data [25
]. Gold standards for activity data are more difficult to obtain.