In this study, adults accurately reported that they spent most of their time at work/school and doing office work, such as filing papers, desk work, and working at the computer. In contrast, adolescents most accurately reported their time out in the community and in leisure pursuits, such as watching television, talking with friends, or playing games. In general, participants were most accurate in classifying the location and purpose of the behaviors in which they spent the most time. Adults tended to report the location and purpose of their behavior more accurately than adolescents. This may be partially due to the more structured nature of adults activities (e.g., time at work) compared to leisure activities that were more commonly reported by adolescents. The accuracy in estimating location and purpose was similar for active and sedentary behaviors, with the exception of transportation where adult participants reported sedentary transport time more accurately than active transport time. Our study suggests that participants report the location of their activities with considerable accuracy and also report useful information about the purpose of behaviors, particularly those in which they spend the most time. To our knowledge, this is the first study to validate estimates of behavioral context for active and sedentary behaviors compared to a criterion measure of direct observation.
There are other methods available to gather location and purpose of physical activities and sedentary behaviors. Some questionnaires include contextual information to improve estimates of duration, intensity and frequency of physical activity or sedentary behavior [27, 29, 30]. Questionnaires typically ask about time domains of activity (e.g., transportation, household, leisure and occupation) and are often validated compared to an activity monitor. The domain specific constructs have not been routinely validated [31, 32], in part because of the absence of strong domain-specific criterion measures. As more research has focused on domain-specific associations, researchers may be interested in validating these specific constructs but it is challenging to do so. Some studies use a behavioral log as a criterion, where people record all activities to validate the questionnaire but this still relies on the participants record rather than an objective measure . The gold-standard is direct observation, which is costly, time-intensive and logistically challenging. Other objective methods (described below) may be able to estimate location and purpose of activity and be used as a criterion measure in future studies.
Activity monitors used in combination with GIS and SenseCam technology objectively assess the context of active and sedentary behaviors. GIS systems have been used to explicitly link location-specific physical activity outcomes with attributes of the built environment (e.g., walkability) [33–35]. While GIS is able to characterize where physical activity or sedentary behavior is taking place, this system cannot directly assess the purpose of the behavior. The SenseCam takes first-person pictures of the environment approximately every 20 seconds . The pictures are manually annotated and classify the activity context into categories based on the Compendium of Physical Activities [12, 22, 37]. This information is combined with accelerometer data to estimate the purpose of active and sedentary behaviors and identify sedentary behaviors that were misclassified by an accelerometer [12, 37]. However, when attempting to classify free-living behavior, only 81% of episodes could be classified . Both ethical (e.g., privacy)  and practical barriers (e.g., poor lighting) currently limit widespread application of this technology . Further, both GIS and SenseCam require expensive equipment and time-intensive data processing, thus these methods may be less feasible for use in large-scale epidemiological studies. Ecological momentary assessment uses a cell phone platform to gather information about the context and affect within a naturalistic setting [14, 39]. This method provides valuable insight into what the participant is doing at an exact moment in time but it does not gather estimates of total time in either active or sedentary behaviors in a given context, as participants are prompted about their current activity multiple times throughout the day [13, 39, 40].
Understanding where and why active and sedentary behaviors take place can inform individual-level and environmental-level interventions targeting these behaviors [10, 41]. All of the tools described above have both strengths and weakness and the choice of instrument should be driven by the research question of interest. For example, if a researcher is increased in long-term activity (i.e., over the past year) then a questionnaire could be preferable to the PDR . Studies examining trips to a particular green-space or trail may consider GPS. Conversely, if the researcher is interested in the location and purpose of active and sedentary behaviors that are currently performed by the individual, the PDR may be preferred. The short-term recall reduces challenges associated with recalling activity over a long prolonged period and gathers detailed information on behavioral context, posture and activity intensity [17, 19]. The current data demonstrate that the PDR is a valid tool for measuring location and purpose of physically active and sedentary behaviors compared to direct observation. Previous reports demonstrate its validity for estimating posture and intensity , with correlation coefficients that tended to be higher than those of questionnaires that rely upon long-term recall to estimate time spent in different behaviors . The development of self-administered short-term recalls furthers the potential utility of this approach in large-scale epidemiologic studies [16–18].
Despite efforts to match the direct observation time with specific PDR segments, there was still approximately 9% more time reported in the PDR than for direct observation. Compared to direct observation, adults reported 28.6 (6.9, 50.3) min more and adolescents reported 24.4 (10.5, 38.2) min more on the PDR. This difference may be due to errors recalling when a segment started and stopped. For example, if the observation started at 11:50 AM and the participant reported eating lunch at 12:00 PM, there will be 10 additional minutes in the observation period that were not reported in the PDR. Because the PDR used in this study was not time stamped, it was not possible to distinguish error due to incorrect recall of specific activity versus error due to difficulties remembering when the segment started and stopped. For the majority of the segments this was not a major problem, but there were two segments included in the dataset where participant reported the segment length was >1hour more than the observation time, increasing bias. This error recalling when a segment starts and stops may be an artifact of the current study design as it was necessary for us to perform direct observation during segments of the day due to the infeasibility of observing participants for 24 consecutive hours.
The use of direct observation as a criterion is an important strength of this study. The challenges of implementing criterion measures such as direct observation to assess location and purpose of activity may explain the paucity of data in this area. Furthermore, we matched the direct observation system to the PDR in order to ensure both our criterion measure and the PDR were measuring the same constructs. This study has other important strengths as well. The study participants completed a range of activities and the time distribution of activities was fairly consistent with how Americans typically spend their time outside of work/school activities [21, 42]. Inclusion of a range of locations and purpose supports the use of the PDR in future studies where participants may complete a variety of behaviors.
There are limitations of the current study that should be considered. Participants may recall their behavior better because they were being observed, though there is no way to empirically test this. To mitigate this effect, the participants were not told the PDR was going to be conducted the following day. The ICCs between direct observation and PDR active and sedentary time in this study (0.73 to 0.83) were similar to the correlations between activPAL and PDR estimates from previous work (0.75 to 0.81), with the exception of adolescent girls who had lower correlations (0.64 to 0.80) in the previous study . This suggests, at least for active and sedentary time, that the accuracy of the recall was not substantially impacted by the observation session. Another limitation to our study is that the sample size was small and may not generalize to other populations. In particular, the sample was relatively young (range 12-62y) with a mean of 42y for adults. Future studies should assess the validity of the PDR with an older adult population. The sample did include a range of BMI (15.7 to 41.0 kg/m2) and a roughly equal distribution of males and females (Table 1). There are challenges for both the observer and the PDR when they complete multiple activities at the same time, which could have introduced error. The observers were instructed to code the purpose for the primary activity. For example, if a participant was eating while watching TV, this would be coded as “self-care”, until the participant stopped eating and then it would be considered leisure time. Similarly, for the recall participants were instructed to code the primary activity. If the participant reported doing two activities with substantially different energy expenditure (e.g., reading and walking on a treadmill) the activity with the higher energy expenditure was recorded.
For some locations and purposes, there were a very small number of participants who completed the activity. The accuracy for these less common physical behaviors could not be addressed. It is possible this reflects the actual portion of time individuals spend in these activities (e.g., adults doing educational activities) . For other activities, such as time at work/school for adolescents, the low number of segments (n = 5) was largely due to safety and logistic concerns of the school district in having an outside observer in the classroom. Time at school makes up a large portion of an adolescent’s day and future studies should attempt to validate measurement tools within a school setting.