Despite advancements in technology for physical activity measurement, no single existing technique can capture the full spectrum of this construct’s complexity and multidimensionality. Researchers have been encouraged to consider the qualities of physical activity that distinguish it from other health behaviors; it is a repeated-occurrence adoption (rather than cessation) behavior that can occur across multiple domains, and it requires time and effort beyond rest, the latter of which can impart feelings of displeasure that contribute to poor maintenance (Ekkekakis et al., 2011; Rhodes & Nigg, 2011). Understanding the complex nature of physical activity is necessary to inform approaches to measurement and enhance consistency across research studies (Pettee Gabriel et al., 2012). Ecological momentary assessment (EMA; repeated, real-time sampling of behavioral, affective, and contextual factors in the natural environment) has been promoted to better understand the synchronicity, sequentiality, and instability of physical activity and its hypothesized correlates (Dunton, 2017). However, because adults are subject to various reporting biases and recall errors, especially when physical activity is habitual or intermittent, the concurrent use of an accelerometer-based device has been recommended (Kanning et al., 2013). These devices measure changes in bodily accelerations that can then be translated into metrics that represent intensity and/or the volume of physical activity, such as total step count, minutes of moderate to vigorous physical activity (MVPA), and total activity counts (John & Freedson, 2012). Accelerometer-based devices are limited, in that they cannot provide behavioral, contextual, or perceptual information about captured movement (Troiano et al., 2014). Given their complementary strengths, data derived from both methods (EMA and accelerometer-based devices) could be integrated to better elucidate the structure (movement pattern and intensity) and meaning (purpose, enjoyment, and intention–behavior concordance) to gain a new perspective of physical activity as a singular research object, a process known as “triangulation” (Hackfort & Birkner, 2003).
Through a preliminary review of the literature, we found that, in most instances, researchers who have implemented both methods used EMA to assess variables specific to the research question (e.g., momentary affect, episodic anxiety, situational context) and used processed data from accelerometer-based devices to estimate MVPA. Various studies (Bruening et al., 2016; Dunton et al., 2005, 2012, 2015, 2016; Ehlers et al., 2016; Knell et al., 2017; Liao et al., 2015, 2020; Maher et al., 2018, 2021) have specifically implemented EMA with physical activity-specific survey items (i.e., type, context) and included concurrent accelerometer-based devices. However, the use of physical activity estimates based on accelerometer sensor processing has varied. In some cases, these estimates were used to determine covariates (Dunton et al., 2015) or dependent variables (Liao et al., 2015), or for assessing wear-time feasibility (Ehlers et al., 2016). A primary use in a number of studies (Bruening et al., 2016; Dunton et al., 2005, 2012, 2016; Knell et al., 2017; Maher et al., 2018, 2021) was to validate EMA reports of physical activity/exercise. That is, researchers isolated acceleration data pertaining to 5, 10, 15, or 30 min preceding each EMA prompt, which asked respondents to report activities that were being performed at the time of the prompt, and processed these data to estimate average counts per minute, MVPA, or step bouts across the target time frame. These estimates were examined to determine if greater values were observed with EMA reports of physical activity, exercise, and/or sport compared with EMA reports of nonphysical activities (e.g., reading a book, watching television; Bruening et al., 2016; Dunton et al., 2005, 2012, 2016). Another approach taken by researchers has been to examine EMA and accelerometer-based estimates of MVPA to utilize once daily EMAs to gather an estimate of self-reported MVPA across 7 days and correlate that with an accelerometer-based estimate of MVPA. This correlation was then compared with the correlation between the accelerometer-based estimate of MVPA and validated questionnaires (e.g., International Physical Activity Questionnaire, Behavioral Risk Factor Surveillance System; Knell et al., 2017).
Within this small body of literature, researchers have demonstrated innovative approaches to collecting and analyzing EMA data to move the field forward, but some important limitations should be noted. First, using estimates of physical activity based on accelerometer sensor processing to validate EMA reports of sedentary or active behavior assumes these methods are comparable. However, inferences of measurement equity should be interpreted with caution because one measure cannot serve as a criterion for the other, as they assess different constructs (behavior vs. bodily movement; Troiano et al., 2014). Second, analyzing a set portion of time from the accelerometer-based device that precedes an EMA prompt (e.g., 5–30 min) may not necessarily represent a bout of activity in its entirety, introducing error into estimates of intensity. Third, aggregating all sources of MVPA to construct a single dependent variable for predictive analyses assumes that activities share hypothesized correlates. While the Physical Activity Guidelines for Adults indicates that individuals can meet the recommendations by performing activities across all domains (Piercy et al., 2018), it is reasonable to speculate that disentangling sources of MVPA, particularly exercise from nonexercise, may be an important component of data processing prior to predictive modeling procedures or evaluating intervention outcomes. For example, walking and running are two frequently reported modes of leisure-time exercise in U.S. adults (Crespo et al., 1996; Watson et al., 2015). While deliberate bouts of exercise-related running may be relatively easy to disentangle from infrequently performed running activity in other domains (e.g., running to catch a bus), walking is a more common source of incidental and transportation-related physical activity. That is, an individual may deliberately walk for 10 min at a brisk pace as part of their daily exercise regimen, as well as engage in multiple, brisk, 10-min bouts of walking for transportation in the same day. It is commonly accepted that different modes of exercise (i.e., walking vs. running) will elicit different experiences, necessitating the ability to classify activity type. But it is also reasonable to consider that antecedents, in-task experiences, and cognitive evaluations may differ depending on the domain or context under which a target activity (i.e., walking for exercise vs. walking for transport) occurs, despite similarities in volume and/or intensity.
Physical activity researchers would benefit from leveraging the complementary features of EMA and accelerometer-based devices for the purposes of triangulation to describe and understand unsupervised physical activity with more depth. However, to our knowledge, methods for triangulating physical-activity-related data derived from EMA and accelerometer-based devices have not been reported in the literature. Additionally, while triangulation procedures necessitate that data from separate sources be in alignment (i.e., each EMA report of physical activity has corresponding data detected by a wearable activity monitor), the frequency of this occurrence has also not been examined. The current study represents an initial step in developing triangulation procedures to better assess the complexity of physical activity, with a specific focus on walking and running for exercise. In this paper, we present—with replicable detail—a preliminary set of novel procedures using previously collected data derived from EMA and accelerometer-based devices to isolate and describe episodes of exercise-related walking and running. Our aim was to explore the alignment between characteristics of exercise-related walking and running reported via EMA and accelerometer-based estimates of time spent walking and running.
Materials and Methods
Parent Study
To address the aims of the current study, a secondary analysis was conducted using EMA-derived self-report data and accelerometer-based data that were collected as part of a previous study that broadly examined various exercise behaviors (Strohacker, 2018). The parent study was approved by the The University of Tennessee, Knoxville Review Board, and all participants provided written informed consent prior to study enrollment. The aim of the parent study was to assess hypothesized antecedents (e.g., affective state, intentions to exercise, perceptions of physical discomfort) of exercise behavior. A convenience sample of participants (faculty, staff, and students at a large flagship university in the Southeast region of the United States) was recruited between June and November 2016. Regardless of current exercise behavior, individuals were eligible to participate if they (a) were at least 18 years old, (b) were not varsity athletes, and (c) owned a smartphone with text messaging and Internet capabilities. The participants who completed all study procedures consisted primarily of young adults (N = 29, 24 ± 6 years, 55% women, 76% non-Hispanic White) who were relatively active, on average (steps per day = 8,233 ± 4,708, MVPA per day = 78.9 ± 43.1 min). Internet survey links (Qualtrics) were sent via text message to the participants’ smartphones four times per day (9:30 a.m., 1:30 p.m., 5:30 p.m., and 9:30 p.m.) across 14 consecutive days. Of the 1,624 EMA prompts sent, 1,348 (83%) surveys were completed within the 60-min limit. The participants, on average, achieved 12 ± 2 days where the accelerometer-based device was worn for at least 10 hr. Further details regarding survey initiation, timing, and completion rates, as well as accelerometer wear-time statistics, can be found in a previous publication (Sheridan et al., 2019).
In line with the accepted definition of “exercise” (Caspersen et al., 1985), the participants were explicitly asked to only report activities as exercise if they were planned, structured, and performed with the purpose of maintaining and improving one or more component of physical fitness. It is important to note that, while researchers provided the participants with the definition of exercise, the researchers did not impose a minimum duration (e.g., 10 min) or intensity in order to count a bout of activity as exercise. Additionally, the participants were explicitly asked to avoid reporting nonexercise activity (e.g., transportation, chores, work). In each survey, the participants who reported that they had exercised in the previous 4-hr time block were prompted to indicate the exercise mode(s) (biking outdoors, jogging or running, brisk walking, aerobic group fitness, muscle-strengthening group fitness, swimming, hiking, and weight lifting). An “other” option allowed participants to write in exercise modes not listed. For each mode indicated, the participants reported the respective duration in minutes. The participants were also asked to wear an ActiGraph GT3X+ activity monitor on the right hip during all waking hours, except during water activities, for the duration of the study. These devices were initialized to collect data at 30 Hz.
EMA Survey Data Processing
Raw data from the Qualtrics Web Survey platform were downloaded in a single output file (comma-separated values) in a long-format organization. That is, up to 56 rows were dedicated to each participant, with each row representing one of the four daily surveys for each of the 14 study days. Columns contained data regarding each observation (e.g., participant ID, date, time of survey receipt), survey metadata (e.g., completion status, time of completion), and primary variables of interest (e.g., exercise mode, exercise duration). Survey metadata were then assessed to (a) remove incomplete reports, (b) remove reports submitted after 60 min of receipt, and (c) organize data to view all reports from each participant in chronological order. The cleaned and organized EMA data served as the source of self-reported exercise mode (walking or running) and self-reported exercise duration that researchers used to inspect the processed accelerometer sensor data.
ActiGraph Data Processing
Raw acceleration data (30 Hz) were downloaded and converted to counts per 10 s using the ActiLife software (ActiGraph Corp., version 6.13.1), with the low-frequency extension disabled (i.e., normal default filter was used). The refined Crouter two-regression model (Crouter 2RM; Crouter et al., 2010) was applied to the 10-s ActiGraph count data to establish the coefficient of variation (CV) of the count data and estimate the intensity of activity in terms of metabolic equivalent of task (METs). The development of the Crouter 2RM is described elsewhere (Crouter et al., 2006, 2010). Briefly, the Crouter 2RM first calculates the CV between each 10-s epoch and the surrounding five 10-s epochs (determining the CV across a minute’s time) to determine if the count data indicate continuous movement or not. For example, if the CV of the 10-s epoch and surrounding five epochs are ≤10% and >0%, this indicates consistent movement, such as walking or running. Next, for sedentary activities (e.g., sitting, lying) where the CV is 0% or the counts are <8 per 10-s epoch, then a MET value of 1.0 is assigned. Lastly, for each epoch, if the CV is between ≤10% and >0%, then a walking and running regression equation is applied, and for epochs where the CV >10%, then an intermittent lifestyle regression is applied. The Crouter 2RM produced comma-separated value files for each participant, showing the counts, CV, and predicted METs per 10 s across each week of participation (Crouter et al., 2010).
Participant-specific output files (comma-separated values) created after applying the Crouter 2RM were organized by participant identification number. Within each file, the date, time, count data, and Crouter 2RM output (CV and estimated MET level) were available in 10-s epochs for all 14 days of participation.
Visual Inspection Procedures
As this was an initial attempt at developing triangulation procedures, visual validation procedures (guided by human decision making) were utilized because it allowed for the discovery of unexpected patterns and idiosyncrasies in the data that would need to be accounted for in the future, automated iterations of these procedures. Therefore, two independent reviewers were provided with both aforementioned spreadsheets, which contained processed data derived from the accelerometer or EMA data. The reviewers then visually inspected CV in activity counts, based on EMA reports of walking and running exercise. During this process, each reviewer developed a third spreadsheet of their own (see Table 1 for an excerpt) to organize the data resulting from the processing steps below:
- 1.Review all EMA data from one participant to:
- a.Denote the date and time of all EMAs that indicated a bout of exercise-related walking or running during the 4 hr preceding the prompt. If a participant reported no exercise bouts via EMA across the active study period (allowed in the parent study to recruit individuals across a range of exercise levels), no further action was taken.
- b.Denote the mode (e.g., walking, running, both)
- c.Denote the duration in minutes of each mode listed in the participant’s EMA report (DurationEMA)
- 2.For each EMA containing a report of exercise-related walking or running, locate the corresponding block of activity count CV to inspect (i.e., if the 9:30 a.m. EMA indicated the target exercise, the reviewer would locate the CV data recorded between 5:30 a.m. and 9:29 a.m.).
- 3.Visually inspect the corresponding 4-hr block for a continuous pattern where CV values are ≤10% and >0% (indicative of walking or running).
- a.Researchers recorded the start and end time of each bout if at least 1 min of CV ≤10% and >0% was found. These are denoted as single episodes of activity.
- b.In some instances, complete minutes CV ≤10% and >0% were found, but were interspersed with brief changes in CV (i.e., CV >10% or = 0%). If the change in CV was less than 1 min and was flanked (on both ends) by at least 1 min with CV ≤10% and >0%, this was considered a within bout break, and the bout was denoted as a single episode of activity. The number of within-bout breaks were recorded per bout. This allowance was put in place to account for natural breaks in outdoor running or walking (e.g., ascending/descending stairs, waiting at a crosswalk; Troiano et al., 2008).
- c.When complete minutes of CV ≤10% and >0% were found but were interspersed with changes in CV lasting longer than 1 min, a new episode of activity was considered. These were denoted as multiple episodes of activity.
- 4.Denote the total duration of each instance of the target repeating pattern (CV ≤10% and >0%, within bout breaks included; DurationCV).
- a.The reviewers specifically searched for a single episode of repeating instances of CV ≤10% and >0% lasting for a duration that was reasonably similar to the duration of walking/running reported in the corresponding EMA (minutes ± 20%). For example, if the EMA indicated a walking bout lasting 20 min, an episode of repeating low CVs was considered reasonably similar if it lasted 16–24 min.
- i.If a single episode meeting the above criteria was observed, additional instances of low CVs were not recorded.
- ii.If a single episode meeting the above criteria was not observed, reviewers were instructed to denote the total number of episodes of repeating low CVs lasting at least 1 min and their respective durations, as well as the time between the occurrence of multiple episodes.
- a.The reviewers specifically searched for a single episode of repeating instances of CV ≤10% and >0% lasting for a duration that was reasonably similar to the duration of walking/running reported in the corresponding EMA (minutes ± 20%). For example, if the EMA indicated a walking bout lasting 20 min, an episode of repeating low CVs was considered reasonably similar if it lasted 16–24 min.
- 5.For each EMA report of walking/running for exercise, classify the alignment between DurationEMA and DurationCV into one of the following categories:
- a.ActiGraph not worn (Category 1): Continuous zero counts observed for the entire 4-hr block preceding an EMA report of exercise.
- b.No evidence of exercise-related walking or running (Category 2): No observed instances where CV ≤10% and >0% repeated for at least 1 min.
- c.Nonaligned evidence of exercise-related walking or running (Category 3): Reviewers observed at least one instance of repeating CV ≤10% and >0% lasting 1 or more minutes, but the data could not be classified as a single episode where DurationCV = DurationEMA ± 20%.
- d.Aligned evidence of exercise-related walking or running (Category 4): Reviewers identified a single episode of repeating CV ≤10% and >0%, such that DurationCV = DurationEMA ± 20%.
Reviewer Data Sheet Excerpt
| Data from EMA | Data from Crouter 2-RM processing | Alignment between durationEMA and durationCV | Additional information | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Survey date | Survey time | Exercise mode | Exercise duration (min) | Target time block to inspect CV in activity counts | Accelerometer data available for processing | Observed pattern in CV | Start time of CV ≤10% and >0% | End Time of CV ≤10% and >0% | Total Duration where ≤10% and >0% (h:min:s) | Target duration range (EMA duration ± 20%) (h:min:s) | Duration of ≤10% and >0% within target range | Alignment category | Mean intensity (METs) |
| September 9, 2016 | 9:30 a.m. | Running | 12 | 5:30 a.m.–9:29 a.m. | Yes | Single episode | 7:38:20 a.m. | 7:39:20 a.m. | 0:01 :00 | 0:09:36–0:14:54 | No | 3 | 4.63 |
| September 9, 2016 | 1:30 p.m. | Running | 15 | 9:30 a.m.–1:29 p.m. | Yes | Single episode | 12:29:40 a.m. | 12:46:00 a.m. | 0:16:20 | 0:12:00–0:18:00 | Yes | 4 | 7.46 |
| September 9, 2016 | 5:30 p.m. | Running | 15 | 1:30 p.m.–5:29 p.m. | Yes | Multiple episodes | 4:51:20 p.m. 4:59:20 p.m. | 4:54:20 p.m. 5:04:40 p.m. | 0:08:20 | 0:12:00–0:18:00 | No | 3 | 7.85 |
| September 10, 2016 | 9:30 a.m. | Running | 18 | 5:30 a.m.–9:29 a.m. | Yes | Single episode | 9:11:50 a.m. | 9:28:50 a.m. | 0:17:00 | 0:14:24–0:21:36 | Yes | 4 | 7.88 |
| September 15, 2016 | 5:30 p.m. | Walking | 120 | 1:30 p.m.–5:29 p.m. | Yes | Multiple episodes | 1:44:40 p.m. 2:25:00 p.m. | 1:49:30 p.m. 2:27:20 p.m. | 00:07:10 | 1:36:00–2:24:00 | No | 3 | 5.84 |
| September 21, 2016 | 9:30 a.m. | Running | 35 | 5:30 a.m.–9:29 a.m. | Yes | Single episode | 6:41:30 a.m. | 7:17:40 a.m. | 0:36:10 | 0:28:00–0:42:00 | Yes | 4 | 7.09 |
Note. EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model; CV = coefficient of variation; MET = metabolic equivalent of task; durationEMA = time spent in exercise-related walking or running indicated in the EMA survey; durationCV = time spent walking or running based on CV ≤10% and >0%; Alignment Category 3 = nonaligned evidence of exercise-related walking or running; Alignment Category 4 = aligned evidence of exercise-related walking or running.
Analyses
To determine the reliability of the visual inspection procedures, interobserver agreement between reviewers was calculated for classifying CV data as providing “aligned” versus other classifications. The reviewers compared their independent results and calculated their initial interobserver agreement. Additional discussions were held between the reviewers to provide context to their independent decisions that led to discordant classifications, allowing reviewers the opportunity to amend their original decision in order for the team to reach agreement. In the single instance where the reviewers could not reach agreement, the first author served as the tiebreaker. Additionally, a two-way mixed intraclass correlation coefficient (ICC; 95% confidence intervals included) was used to determine the level of agreement (LOA) between the reviewers’ DurationCV values within the “aligned” and “nonaligned” classifications. The ICC, which ranges from .0 to 1.0, was interpreted based on accepted values (Cicchetti, 1994), indicating the degree of consistency: <.40 (poor), .40–.59 (fair), .60–.74 (good), ≥.75 (excellent). Frequency analyses were conducted to determine the proportion of EMA reports assigned to each classification (one through four) based on patterns of CV of the activity counts. Basic descriptive statistics were computed to summarize “aligned” and “nonaligned” accelerometer-based estimates for evidence of duration, intensity, and breaks. Bland–Altman plots including 95% LOA were constructed to visually assess the LOA between the DurationCV and DurationEMA (in minutes) for each walking and running bout classified as aligned and nonaligned. Basic descriptive statistics (M, SD, and range) were computed to summarize the MET estimates of walking and running (computed from 10-s epochs) bouts across classifications. Cases classified as “nonaligned evidence” were further explored to describe alternative patterns emerging from these data.
Results
Interrater Reliability of Visual Inspection Outcomes
Initial interobserver agreement was 91% regarding the classification of CV patterns as “aligned” versus other categories. ICCs for reviewers’ estimates of DurationCV across data classified as “aligned” (ICC = .992; 95% confidence interval [.988, .995]) and “nonaligned” (ICC = .960; 95% confidence interval [.940, .974]) were considered excellent.
Visual Inspection Outcomes
Figure 1 provides a summary of outcomes from the visual inspection procedures. In total, 302 bouts of exercise were reported across participants. Of these reports, 59.7% indicated walking exercise, 34.5% indicated running exercise, and 5.8% indicated both running and walking exercise. For 75% of the EMA reports indicating exercise-related walking and running, the reviewers were unable to identify a single episode of repeating CV ≤10% and >0% using corresponding processed accelerometer data, such that DurationCV = DurationEMA ± 20%.
—Classification of alignment flow chart. Note. EMA = ecological momentary assessment; CV = coefficient of variation in Crouter 2RM activity counts across all participants; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Classification of alignment flow chart. Note. EMA = ecological momentary assessment; CV = coefficient of variation in Crouter 2RM activity counts across all participants; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Classification of alignment flow chart. Note. EMA = ecological momentary assessment; CV = coefficient of variation in Crouter 2RM activity counts across all participants; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
Figure 2 provides representative examples for the patterns of CV in activity counts over time. The gray bands denote instances of CV ≤10% and >0%, which indicate either walking or running. The self-reported exercise mode(s) and duration(s) from the corresponding EMA are included for each representative example, but these classifications should not be extrapolated to the CV data, as the classification of running versus walking cannot be distinguished without the addition of MET values. Figure 2a is representative of patterns observed in the 34 cases classified as “aligned evidence.” Figure 2b and 2c show patterns of data classified as “nonaligned evidence.” Figure 2b shows one of the six instances where the EMA reports indicated both walking and running, but the pattern in CVs were described by the reviewers as resembling interval-type exercise. Figure 2c represents the most commonly observed pattern (56 cases), wherein multiple episodes of walking or running occurred over the 4-hr time block. While in 18 cases, single episodes of nonaligned walking/running were observed (with no additional instances of CV ≤10% and >0% lasting one or more minutes), this pattern is not shown, as it resembles that demonstrated in Figure 2a. Figure 2d demonstrates one of the 15 instances where the reviewers did not observe CV ≤10% and >0% for at least 1 min.
—Common patterns in Crouter 2RM data. Note. Gray bands demonstrate the threshold of ≤10% and >0%. (a) Single episodes classified as “aligned,” (b) walking/running resembling interval training, (c) multiple episodes of walking/running across the target time frame, and (d) no evidence of CV ≤10% and >0% for at least 1 min. CV = coefficient of variation; EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Common patterns in Crouter 2RM data. Note. Gray bands demonstrate the threshold of ≤10% and >0%. (a) Single episodes classified as “aligned,” (b) walking/running resembling interval training, (c) multiple episodes of walking/running across the target time frame, and (d) no evidence of CV ≤10% and >0% for at least 1 min. CV = coefficient of variation; EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Common patterns in Crouter 2RM data. Note. Gray bands demonstrate the threshold of ≤10% and >0%. (a) Single episodes classified as “aligned,” (b) walking/running resembling interval training, (c) multiple episodes of walking/running across the target time frame, and (d) no evidence of CV ≤10% and >0% for at least 1 min. CV = coefficient of variation; EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
Bland–Altman plots showing the individual variability between DurationEMA and DurationCV for each bout of walking and running exercise reported are shown in Figure 3. The aligned bouts had the lowest individual variability (95% LOA: −4.5 to 3.1 min; Figure 3a) compared with the nonaligned single bouts (95% LOA: −108.2 to 58.2 min; Figure 3b) and nonaligned multiple bouts (95% LOA: −57.7 to 26.6 min; Figure 3c). DurationCV was underestimated when compared with DurationEMA for the aligned bouts (−0.7 min) and nonaligned cases occurring in single bouts (−25.0 min) or in multiple bouts (−15.5 min).
—Bland–Altman plots illustrating data variability. Note. (a) Aligned bouts, (b) nonaligned bouts when a single exercise episode was found in the Crouter 2RM data that corresponded with EMA reporting time frame, and (c) nonaligned bouts with multiple exercise episodes. EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Bland–Altman plots illustrating data variability. Note. (a) Aligned bouts, (b) nonaligned bouts when a single exercise episode was found in the Crouter 2RM data that corresponded with EMA reporting time frame, and (c) nonaligned bouts with multiple exercise episodes. EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Bland–Altman plots illustrating data variability. Note. (a) Aligned bouts, (b) nonaligned bouts when a single exercise episode was found in the Crouter 2RM data that corresponded with EMA reporting time frame, and (c) nonaligned bouts with multiple exercise episodes. EMA = ecological momentary assessment; Crouter 2RM = Crouter two-regression model.
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
Table 2 provides descriptive statistics regarding duration and intensity across reports categorized as aligned and nonaligned, with the latter category split based on observed patterns indicating single or multiple episodes of activity. Reports including both walking and running found to resemble interval-type training were not accounted for in these outcomes. Overall, the majority of cases found to be in alignment pertained to running exercise (59%), whereas the majority of nonaligned cases pertained to walking exercise (72%). Multiple episodes corresponding to self-reported walking exercise generally occurred across a total time interval of 70.5 ± 59.8 min and were separated by 19.0 ± 28.4 min. Episodes corresponding to running occurred across 100.1 ± 69.1 min, separated by 19.1 ± 29.5 min.
Duration and Intensity (Mean ± SD [Range]) of Aligned and Nonaligned Walking and Running Episodes
| Aligneda Single episodes | Nonaligned Single episodes | Nonaligned Multiple episodes | ||||
|---|---|---|---|---|---|---|
| Walkingb (n = 14 cases) | Runningb (n = 20 cases) | Walkingb (n = 13 cases) | Runningb (n = 5 cases) | Walkingb (n = 40 cases) | Runningb (n = 16 cases) | |
| DurationEMAc (min) | 18.6 ± 12.8 (10.0–50.0) | 32.9 ± 16.1 (15.0–69.0) | 37.3 ± 44.6 (10.0–180.0) | 30.8 ± 15.4 (12.0–50.0) | 31.3 ± 26.6 (10.0–120.0) | 35.8 ± 20.0 (14.0–90.0) |
| Episodes of low CV | 1 | 1 | 1 | 1 | 4 ± 3 (2–16) | 5 ± 3 (2–11) |
| DurationCVd (min) | 17.3 ± 14.4 (8.0–52.3) | 32.6 ± 16.6 (15.3–67.2) | 4.6 ± 3.6 (1.2–12.5) | 16.3 ± 20.5 (1.2–56.7) | 15.8 ± 14.5 (2.0–72.8) | 26.3 ± 13.5 (7.7–54.5) |
| Total per episode | N/A | N/A | N/A | N/A | 4.1±3.6 (0.8–15.3) | 5.4±7.2 (0.8–39.7) |
| METse per episode | 4.5 ± 0.6 (3.6–5.7) | 6.8 ± 1.4 (4.8–10.4) | 4.2 ± 0.4 (3.3–4.8) | 5.3 ± 0.5 (4.8–6.0) | 3.4 ± 0.4 (3.1–4.9) | 5.9 ± 1.2 (3.9–7.9) |
Note. CV = coefficient of variation; EMA = ecological momentary assessments; MET = metabolic equivalent of task; N/A = not applicable.
aDurationCV = DurationEMA ± 20%. bModes designated based on self-report from EMA. cDuration of exercise self-reported in EMA survey. dBased on continuous minutes of CV ≤10% and >0% in GT3X+ vertical axis counts. eMET estimated using the Crouter two-regression model (Crouter et al., 2010).
Figure 4 provides an overview of each participant’s data regarding the frequency of self-reported exercise bouts (walking and/or running) and the proportion of these bouts that were found to be in alignment with reviewers’ visual inspection of accelerometer-derived data (Category 4). Out of the 29 participants, three individuals reported no exercise on EMA reports during the 2-week assessment period. Of the 26 participants reporting exercise, 24 had at least one case where an EMA report was categorized as having no evidence (Category 2, n = 15 total cases) or nonaligned evidence (Category 3, n = 80 total cases) of exercise-related walking or running.
—Histogram illustrating proportions of aligned and nonaligned bouts within each participant (N = 29).
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Histogram illustrating proportions of aligned and nonaligned bouts within each participant (N = 29).
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
—Histogram illustrating proportions of aligned and nonaligned bouts within each participant (N = 29).
Citation: Journal for the Measurement of Physical Behaviour 5, 3; 10.1123/jmpb.2022-0016
Discussion
The current study demonstrates a set of procedures for visually assessing CV in ActiGraph activity counts to isolate exercise-related walking and running that corresponds to bouts reported through EMA. Interrater reliability of this method was found to be high, but alignment between exercise-related walking and running estimated from the Crouter 2RM and self-reported in EMA was observed in less than 25% of cases. The methods outlined in this study represent a novel approach regarding the complementary use of EMA and accelerometer-based devices in preparation for data triangulation. However, the inability to align EMA reports and accelerometer-derived data for the majority of reported cases of exercise-related walking and running raises several concerns that should be considered in future research designs.
The results from this study suggest a potential limitation of the standard EMA item design when assessing the alignment of exercise-related walking and running with accelerometer-derived data. To date, EMA items have reflected distinct modes of exercise (e.g., walking, jogging, weight lifting) and the total duration spent in each, an approach that was replicated in the parent study. An implied assumption of this wording is that episodes of exercise are performed in a single bout over a given period of time. However, 78% of nonaligned cases consisted of multiple, shorter bouts, or integration of walking and running intervals. The traditional item wording may limit researchers’ understanding of individuals’ exercise behavior performed in a free-living environment, which would preclude the application of triangulation procedures. Given that interval training is considered a top fitness trend in the United States (Kercher, 2018; Thompson, 2022) and accumulating exercise in shorter bouts is a common strategy assessed in comparison to continuous exercise in training studies (Murphy et al., 2009), adapting EMA items to capture these common patterns should be considered. For example, display logic could be strategically implemented, such that selecting “walking” as the exercise mode would trigger an initial set of follow-up questions to gauge continuity (continuous walking in a single bout vs. discontinuous walking in multiple bouts). Based on the pattern of walking, further probes could be prompted to appear: the selection of “continuous walking” initiates a question about pace (single speed vs. intervals of higher and lower intensity), whereas the selection of “discontinuous walking” could trigger items gauging number/duration/average intensity of bouts. Including items to indicate the approximate start time of exercise would also simplify the alignment procedures by narrowing the search window.
Differences in EMA study design may also impact alignment quality. In the parent study, EMAs were signal contingent, occurring at regular intervals to achieve a full coverage view of the participants’ waking hours. However, it is possible that asking participants to retrospectively report exercise details across 4-hr time frames introduced excess error in recall. To improve alignment, researchers may consider interval designs with shorter recall windows to minimize bias associated with memory reconstruction. Alternatively, the implementation of event-contingent EMA (i.e., respondents initiate surveys when in specific contexts; Shiffman et al., 2008)—in this case, immediately pre- or postexercise—may improve data alignment by removing reliance on autobiographical memory. However, as event-contingent EMA still relies on individuals remembering to initiate context-specific surveys, researchers could also consider leveraging monitor detection technology as well. For example, Dunton et al. (2014, 2016) constructed a smartphone application that utilizes built-in motion sensors to trigger EMA prompts relative to “activity,” “no activity,” or “no data” based on transitions in detected movement. Each of these design alternatives may prove more useful when individuals choose to walk for exercise (especially in shorter durations or in multiple bouts over time) but also regularly walk for transport or work in the same time frame when exercise occurred.
In addition to refined EMA prompt wording, participants should be sufficiently educated and reminded of the differences between general physical activity and exercise. In the parent study, the orientation session included instructions to only report planned, structured exercise instead of physical activity due to transportation, chores, or work, and no further reminders were provided. In a previous assessment of feasibility of these procedures, the “other” category was the fourth highest reported mode of exercise, but 31 of these 54 reports contained write-in details that indicated nonexercise physical activity in the domains of transportation (“taking lots of stairs”), housework (“painting a room”), or occupation (“bagging groceries”; Sheridan et al., 2019). Thus, it is entirely possible that some of the reports of walking exercise were transportation related (e.g., walking between classes) rather than episodes of structured, planned exercise. However, in a 1985 publication, Caspersen surmised that one could choose to do household or other tasks in a labor-producing, rather than a labor-saving approach (choosing to walk to class rather than ride the bus), and that this could be considered exercise (Caspersen et al., 1985). Researchers’ operationalizations of constructs, as well as detailed strategies for communicating these to participants, should be included with primary methodological details to better interpret findings across studies.
Upon making the decision to include acceleration-based physical activity assessments with EMA research, one must be aware that many considerations must be made when selecting a monitor, monitor calibrations, data processing methods, and physical activity variables of interest. While consumer-grade monitors are appealing because of user familiarity and, in some instances, lower cost per monitor (Wright et al., 2017), the granularity of data (e.g., access to continuous minute-by-minute data) must be considered. Data in minute-by-minute or shorter epochs are rarely directly available from consumer-grade monitors, most offering summary data across the day or each hour of the day. For the purposes of this study, or in any case when identifying specific instances of activity is necessary, summary variables are insufficient. For that reason, research-grade monitors are optimal, as the user can access continuous shorter epoch-level data (e.g., minute-by-minute, second-by-second). Moreover, some research-grade monitors provide numerous options for interpreting data, as in selecting from available algorithms or the ability to download continuous acceleration or count data for further postprocessing.
Our selected method (use of an ActiGraph GT3X+ with the Crouter 2RM for postprocessing) revealed necessary insights into patterns of ambulatory activity accumulation, but additional research is needed to determine optimal approaches to alignment.
Theoretically, under ideal conditions, bouts of exercise-related walking and running could be easy to identify in accelerometer-derived data reported in 10-s epochs, but visual inspection revealed otherwise. Only 25% of ambulatory exercise bouts could be easily identified as a single, continuous session, while other reports were more challenging to distinguish. Through this process we identified two important considerations for those intending to engage in similar processes to assess the alignment of data from EMA reports and data derived from wearable accelerometer-based devices.
The first consideration pertains to handling within-bout breaks. An important difference between self-reported physical activity and accelerometer-assessed physical activity is the granularity of data. Self-reported activity is a mean estimate of time spent completing one continuous activity, which does not account for short breaks taken during that bout of activity. These short breaks or pauses could be due to environmental factors (e.g., waiting for a car to pass before crossing a street) or could be other inert factors, such as pausing to tie a shoe or stopping for a water break. Thus, these inert or subconscious breaks in activity go unreported. However, when activity is assessed with a wearable accelerometer, these pauses are evident when data are reported in 10-s epochs, where a pause of at least 10 s reveals a “break” during a continuous bout. Chastin et al. (2009) suggested that self-reported activity and accelerometer-derived activity are two different constructs of perceived and actual activity, respectively. The authors suggested “stitching together” bouts that are separated by breaks lasting <10 s. Interestingly, Troiano et al. (2008) also made allowances for short breaks, but used a longer cut point for “stitching together” bouts that were separated by <2 min. In the current study, we “stitched” bouts that were separated by ≤1-min break, with the requirement of one complete minute of activity occurring before and after the break. These ranges have reasonable face validity based on how long these common activity interrupters last. However, to consistently triangulate data across studies for a meaningful synthesis of results, it would be important for researchers to investigate pauses in activity—their purpose and duration—to craft guidelines for when a break in activity is naturally occurring within a bout or indicative that a new activity has begun.
The second consideration pertains to handling competing bouts. While not observed in the data set used in this study, there exists the possibility of reviewers identifying two or more bouts within the same observation window that are approximately the same duration as the reported exercise. A question then arises: How would a researcher distinguish the exercise bout from other activity (e.g., transportation)? To do so, this concern needs to be considered in the planning stages of study design, and specific items need to be added into the EMA surveys. For example, items could be included that distinguish context (e.g., walking for exercise vs. walking to get somewhere). In addition to providing behavioral context, individuals could be asked to report the time of day when a given activity (exercise or a bout of transportation-related activity) occurred (start time, end time, or both). Together, such items would allow reviewers to distinguish between competing bouts of activity observed in a single target time block of accelerometer-derived data.
This study is not without limitations. The first set of limitations pertains to our use of the Crouter 2RM in a manner outside of its primary design scope. The Crouter 2RM was developed using the ActiLife 7164, potentially adding errors in the estimates of activity intensity. However, the variability in the count values will not be impacted, resulting in a negligible effect on the CV whether or not the low-frequency extension was enabled. While the ability to identify walking and running was an integral part of the Crouter 2RM design, making it suitable to address the aim of the current study, its primary purpose is to predict MET values (averaged over 1-min epochs). Averaging MET values across 10-s epochs likely added some variability in our intensity estimates. A more accurate classification of specific activities across various dimensions (e.g., intensity, duration, timing) necessitates that movement detection and data processing methods be developed and refined with these purposes in mind.
A second limitation is that only one method (visual inspection of ActiGraph data processed using the Crouter 2RM) was employed and procedures were conservatively, yet arbitrarily, set (i.e., acceptable pattern, within-bout break length, over/under summed duration) in the absence of empirically defined parameters and guidelines. For example, we initially considered classifying instances of multiple episodes and walking/running intervals as “aligned” if the total duration met the stated criteria, which would have applied to 12 additional cases. However, given the limits of the EMA items, the more conservative criteria were ultimately retained. It is certainly possible that other methods may be applied and found to be optimal in terms of error and researcher time burden. Related to this limitation, we did not clearly define the required wear time per 4-hr block. For our analysis, we included data for analysis if there was any wear time at any point during each 4-hr block. This could result in misalignment of EMA-reported activity and accelerometer-derived activity in the event that the accelerometer was worn during the block but was removed during activity. Third, this approach is limited to ambulation and cannot be directly extrapolated to other common modes of exercise (e.g., swimming, cycling, weight lifting). However, with advances in the classification of activity type using accelerometer sensor data with machine-learning algorithms (e.g., artificial neural networks), this method could be applied to the alignment of other activity types (e.g., household chores, recreational sports) beyond just walking and running.
Perhaps most importantly, the use of previously collected data to conduct the current analyses is inherently limiting, and the results reported here should be understood to be a beta test for data alignment protocols that can guide future, hypothesis-driven work. For example, addressing the aims of the parent study required a lower resolution view of the exercise mode—items were not then designed to capture all possible exercise subtypes (running vs. long slow duration or sprint intervals, weight lifting vs. machines or body weight, and swimming vs. continuous freestyle or kickboard drills). Furthermore, our reliance on data that were previously collected from a convenience sample (i.e., minimal variance in participant age, race, activity level) limited the ability to conduct follow-up analyses to explain the large degree of misalignment observed in the current study. In a prior examination using this data set, we determined that 9:30 a.m. surveys were less frequently completed than surveys sent at later times (Sheridan et al., 2019). Thus, it is possible that certain EMA distributions and participant characteristics are associated with differences in data alignment (e.g., more misalignment noted in morning surveys, on weekend days, or in low-active participants). Taken together, it is important to replicate the described methods in a study designed to solely focus on aligning data specifically for walking and/or running activity and powered for subgroup analyses. Such work would provide more accurate estimates regarding the proportions of aligned versus nonaligned bouts and necessary insight on contributors to misalignment to further optimize such procedures.
To continue optimizing the process of data alignment, methods of direct observation should be implemented to validate the interpretations of behavior arising from concurrent EMA and accelerometer-based devices. It is important to consider that direct observation with a researcher present may increase the risk for social desirability bias, potentially negating the “ecological” aspect of EMA. In this regard, video-based direct observation via wearable digital cameras may be a useful method to incorporate (Kelly et al., 2011). By implementing appropriate validation procedures, improvements can be made to data-processing procedures regarding allowable within-bout breaks, CV threshold, and EMA item wording, as well as determining participants’ ability to correctly contextualize physical activity as exercise versus nonexercise. Additionally, it would be beneficial for researchers to develop programming codes to automate (a) data extraction from multiple sources and (b) data inspection for evidence of target behaviors and movement patterns. In the current study, visual inspection procedures were used, following guidelines set a priori. The benefit of this approach was the potential for human reviewers to denote emerging characteristics or patterns of free-living movement that were not expected. Being an exploratory study, this is an important aspect of developing triangulation procedures. However, this process was time consuming; reviewers reported that visual inspection required approximately 1 hr per participant, depending on the number of exercise bouts reported and the complexity of the activity count data. As such, the present approach would not be particularly feasible with a larger sample of participants or if real-time, ongoing analyses are needed to guide personalized interventions. In this regard, the methods presented in the current study could be automated with a coding language (e.g., R, MATLAB) to reduce researcher burden, which would improve the feasibility of triangulation procedures and allow comparisons of results based on minor shifts in decision making (e.g., LOA achieved when using 1-min vs. 2-min breaks in activity).
In conclusion, this research provides a preliminary approach for assessing the alignment between self-reported minutes of walking or running from EMA and estimates of time spent walking or running based on CV in ActiGraph GT3X+ counts. The assessment of alignment is a necessary precursor to triangulation procedures that would provide added insight to human behavior that involves physical movement. This approach departs from the current body of literature by considering both methods as complementary, rather than comparative or interchangeable. Given the call for researchers to collect, analyze, and use intensive longitudinal data to measure and intervene upon human behavior (Dunton, 2017; Kiely, 2011; Nahum-Shani et al., 2018), approaches used in the current study may be refined and leveraged to better understand the topography (i.e., shape and features) of exercise and physical activity behavior within individuals, over time.
References
Bruening, M., van Woerden, I., Todd, M., Brennhofer, S., Laska, M.N., & Dunton, G. (2016). A mobile ecological momentary assessment tool (devilSPARC) for nutrition and physical activity behaviors in college students: A validation study. Journal of Medical Internet Research, 18(7), e209. https://doi.org/10.2196/jmir.5969
Caspersen, C.J., Powell, K.E., & Christenson, G.M. (1985). Physical activity, exercise, and physical fitness: Definitions and distinctions for health-related research. Public Health Reports, 100(2), 126–131. https://www.ncbi.nlm.nih.gov/pubmed/3920711
Chastin, S.F., Dall, P.M., Tigbe, W.W., Grant, M.P., Ryan, C.G., Rafferty, D., & Granat, M.H. (2009). Compliance with physical activity guidelines in a group of UK-based postal workers using an objective monitoring technique. European Journal of Applied Physiology, 106(6), 893–899. https://doi.org/10.1007/s00421-009-1090-x
Cicchetti, D. (1994). Guidelines, criteria, and rules of thumb for evaluating normed and standardized assessment instruments in psychology. Psychological Assessment, 6(4), 284–290.
Crespo, C.J., Keteyian, S.J., Heath, G.W., & Sempos, C.T. (1996). Leisure-time physical activity among US adults. Results from the Third National Health and Nutrition Examination Survey. Archives Internal Medicine, 156(1), 93–98. https://www.ncbi.nlm.nih.gov/pubmed/8526703
Crouter, S.E., Clowers, K.G., & Bassett, D.R., Jr. (2006). A novel method for using accelerometer data to predict energy expenditure. Journal of Applied Physiology, 100(4), 1324–1331. https://doi.org/10.1152/japplphysiol.00818.2005
Crouter, S.E., Kuffel, E., Haas, J.D., Frongillo, E.A., & Bassett, D.R., Jr. (2010). Refined two-regression model for the actigraph accelerometer. Medicine & Science in Sports & Exercise, 42(5), 1029–1037. https://doi.org/10.1249/MSS.0b013e3181c37458
Dunton, G.F. (2017). Ecological momentary assessment in physical activity research. Exercise and Sport Sciences Reviews, 45(1), 48–54. https://doi.org/10.1249/JES.0000000000000092
Dunton, G.F., Dzubur, E., & Intille, S. (2016). Feasibility and performance test of a real-time sensor-informed context-sensitive ecological momentary assessment to capture physical activity. Journal of Medical Internet Research, 18(6), e106. https://doi.org/10.2196/jmir.5398
Dunton, G.F., Dzubur, E., Kawabata, K., Yanez, B., Bo, B., & Intille, S. (2014). Development of a smartphone application to measure physical activity using sensor-assisted self-report. Frontiers in Public Health, 2, 12. https://doi.org/10.3389/fpubh.2014.00012
Dunton, G.F., Liao, Y., Intille, S., Huh, J., & Leventhal, A. (2015). Momentary assessment of contextual influences on affective response during physical activity. Health Psychology, 34(12), 1145–1153. https://doi.org/10.1037/hea0000223
Dunton, G.F., Liao, Y., Kawabata, K., & Intille, S. (2012). Momentary assessment of adults’ physical activity and sedentary behavior: Feasibility and validity. Frontiers in Psychology, 3, 260. https://doi.org/10.3389/fpsyg.2012.00260
Dunton, G.F., Whalen, C.K., Jamner, L.D., Henker, B., & Floro, J.N. (2005). Using ecologic momentary assessment to measure physical activity during adolescence. American Journal of Preventive Medicine, 29(4), 281–287. https://doi.org/10.1016/j.amepre.2005.07.020
Ehlers, D.K., Huberty, J., Buman, M., Hooker, S., Todd, M., & de Vreede, G.J. (2016). A novel inexpensive use of smartphone technology for ecological momentary assessment in middle-aged women. Journal of Physical Activity & Health, 13(3), 262–268. https://doi.org/10.1123/jpah.2015-0059
Ekkekakis, P., Parfitt, G., & Petruzzello, S.J. (2011). The pleasure and displeasure people feel when they exercise at different intensities: Decennial update and progress towards a tripartite rationale for exercise intensity prescription. Sports Medicine, 41(8), 641–671. https://doi.org/10.2165/11590680-000000000-00000
Hackfort, D., & Birkner, H. (2003). Triangulation as a basis for diagnostic judgements. International Journal of Sport and Exercise Psychology, 1(1), 82–94. https://doi.org/10.1080/1612197X.2003.9671705
John, D., & Freedson, P. (2012). ActiGraph and Actical physical activity monitors: A peek under the hood. Medicine & Science in Sports & Exercise, 44(1 Suppl. 1), S86–S89. https://doi.org/10.1249/MSS.0b013e3182399f5e
Kanning, M.K., Ebner-Priemer, U.W., & Schlicht, W.M. (2013). How to investigate within-subject associations between physical activity and momentary affective states in everyday life: A position statement based on a literature overview. Frontiers in Psychology, 4, 187. https://doi.org/10.3389/fpsyg.2013.00187
Kelly, P., Doherty, A., Berry, E., Hodges, S., Batterham, A.M., & Foster, C. (2011). Can we use digital life-log images to investigate active and sedentary travel behaviour? Results from a pilot study. International Journal of Behavioral Nutrition and Physical Activity, 8(1), 44. https://doi.org/10.1186/1479-5868-8-44
Kercher, V. (2018). International comparisons: ACSM’s Worldwide survey of fitness trends. ACSM’s Health & Fitness Journal, 22(6), 24–29. https://doi.org/10.1249/FIT.0000000000000431
Kiely, J. (2011). Periodization, planning, and prediction: A new perspective? Palestrica Mileniului III—Civilizatie si Sport, 12(2), 164–169.
Knell, G., Gabriel, K.P., Businelle, M.S., Shuval, K., Wetter, D.W., & Kendzor, D.E. (2017). Ecological momentary assessment of physical activity: Validation study. Journal of Medical Internet Research, 19(7), e253. https://doi.org/10.2196/jmir.7602
Liao, Y., Intille, S.S., & Dunton, G.F. (2015). Using ecological momentary assessment to understand where and with whom adults’ physical and sedentary activity occur. International Journal of Behavioral Medicine, 22(1), 51–61. https://doi.org/10.1007/s12529-014-9400-z
Liao, Y., Song, J., Robertson, M.C., Cox-Martin, E., & Basen-Engquist, K. (2020). An ecological momentary assessment study investigating self-efficacy and outcome expectancy as mediators of affective and physiological responses and exercise among endometrial cancer survivors. Annals of Behavioral Medicine, 54(5), 320–334. https://doi.org/10.1093/abm/kaz050
Maher, J.P., Rebar, A.L., & Dunton, G.F. (2018). Ecological momentary assessment is a feasible and valid methodological tool to measure older adults’ physical activity and sedentary behavior. Frontiers in Psychology, 9, 1485. https://doi.org/10.3389/fpsyg.2018.01485
Maher, J.P., Sappenfield, K., Scheer, H., Zecca, C., Hevel, D.J., & Kennedy-Malone, L. (2021). Feasibility and validity of assessing low-income, African American older adults’ physical activity and sedentary behavior through ecological momentary assessment. Journal for the Measurement of Physical Behaviour, 4(4), 343–352. https://doi.org/10.1123/jmpb.2021-0024
Murphy, M.H., Blair, S.N., & Murtagh, E.M. (2009). Accumulated versus continuous exercise for health benefit: A review of empirical studies. Sports Medicine, 39(1), 29–43. https://doi.org/10.2165/00007256-200939010-00003
Nahum-Shani, I., Smith, S.N., Spring, B.J., Collins, L.M., Witkeiwitz, K., Tawari, A., & Murphy, S.A. (2018). Just-in-time adaptive interventions (JITAIs) in mobile health: Key components and design principles for ongoing health behavior support. Annals of Behavioral Medicine, 52(6), 446–462. https://doi.org/10.1007/s12160-016-9830-8
Pettee Gabriel, K.K., Morrow, J.R., Jr., & Woolsey, A.L. (2012). Framework for physical activity as a complex and multidimensional behavior. Journal of Physical Activity & Health, 9(Suppl. 1), S11–S18. https://doi.org/10.1123/jpah.9.s1.s11
Piercy, K.L., Troiano, R.P., Ballard, R.M., Carlson, S.A., Fulton, J.E., Galuska, D.A., . . . Olson, R.D. (2018). The physical activity guidelines for Americans. JAMA, 320(19), 2020–2028. https://doi.org/10.1001/jama.2018.14854
Rhodes, R.E., & Nigg, C.R. (2011). Advancing physical activity theory: A review and future directions. Exercise and Sport Sciences Reviews, 39(3), 113–119. https://doi.org/10.1097/JES.0b013e31821b94c8
Sheridan, L.F., Toth, L., & Strohacker, K. (2019). Feasibility of using participant owned smartphone features to conduct ecological momentary assessment of planned exercise behavior in college-aged adults. Pursuit: The Journal of Undergraduate Research at the University of Tennessee, 9(1), Article 1. https://trace.tennessee.edu/pursuit/vol9/iss1/1
Shiffman, S., Stone, A.A., & Hufford, M.R. (2008). Ecological momentary assessment. Annual Review of Clinical Psychology, 4, 1–32. https://doi.org/10.1146/annurev.clinpsy.3.022806.091415
Strohacker, K., O’Neil, M., Springer, C.M., & Sheridan, L. (2018). Using ecological momentary assessment to explore proposed indices of exercise readiness and subsequent exercise behavior. Medicine and Science in Sports and Exercise, 50(5S), 311
Thompson, W.R. (2022). Worldwide survey of fitness trends for 2022. ACSM’s Health & Fitness Journal, 26(1), 11–20. https://doi.org/10.1249/FIT.0000000000000732
Troiano, R.P., Berrigan, D., Dodd, K.W., Masse, L.C., Tilert, T., & McDowell, M. (2008). Physical activity in the United States measured by accelerometer. Medicine & Science in Sports & Exercise, 40(1), 181–188. https://doi.org/10.1249/mss.0b013e31815a51b3
Troiano, R.P., McClain, J.J., Brychta, R.J., & Chen, K.Y. (2014). Evolution of accelerometer methods for physical activity research. British Journal of Sports Medicine, 48(13), 1019–1023. https://doi.org/10.1136/bjsports-2014-093546
Watson, K.B., Frederick, G.M., Harris, C.D., Carlson, S.A., & Fulton, J.E. (2015). U.S. adults’ participation in specific activities: Behavioral risk factor surveillance system–2011. Journal of Physical Activity & Health, 12(Suppl. 1), S3–S10. https://doi.org/10.1123/jpah.2013-0521
Wright, S.P., Hall Brown, T.S., Collier, S.R., & Sandberg, K. (2017). How consumer physical activity monitors could transform human physiology research. American Journal of Physiology-Regulatory Integrative and Comparative Physiology, 312(3), R358–R367. https://doi.org/10.1152/ajpregu.00349.2016
