Data were collected from children participating in the longitudinal observational Mothers’ and Their Children’s Health (MATCH) study. Baseline data were used for this analysis. The goal of the MATCH study was to examine the effects of maternal stress on obesity risk in children living in Southern California. Participant recruitment occurred via flyers and in-person research staff visits at public elementary schools and community events. The inclusion criteria for mother–child dyads were (1) the child is in the 3^rd-6^th grade (aged 8-12 years), (2) more than half of the child’s custody belongs to the mother, and (3) both mother and child are able to read English or Spanish. Dyads were excluded from the study if the mother or the child (1) was taking medications for thyroid function or psychological conditions, (2) had a health condition that limited PA, (3) was enrolled in a special education program, (4) was currently using oral or inhalant corticosteroids for asthma, (5) was pregnant, (6) the child was classified as underweight by a BMI percentile <5% adjusted for sex and age, or (7) the mother worked more than 2 weekday evenings (between 5-9 pm) per week or more than 8 h on any weekend day. The MATCH study protocol is described in further detail elsewhere [22].

Mothers provided in-person parental consent, and children provided written assent. Mothers completed paper and pencil questionnaires on their child’s age, sex, and race or ethnicity during a 90-min data collection session. Additionally, anthropometric measures of the child participants were taken at this time. Specifically, height (centimeters) and weight (kilograms) were collected in duplicate and averaged. Age- and sex-adjusted BMI z-scores were then calculated using the Centers for Disease Control EpiInfo 2005, Version 3.2 resource [22].

The children downloaded a custom-made EMA app for Android mobile phones (Google Inc., Mountain View, CA) on their personal mobile phones. If they did not have their own mobile phone, they were provided with a Moto G mobile phone (Motorola Mobility, Chicago, IL) to use for the duration of the study. After doing so, each child received random EMA prompts after 5:00 pm on the day of the data collection session (day 1) across the next 6 complete days (days 2-7) and up until 5:00 pm on the last day when the phone was returned to the researchers (day 8). On weekends, EMA surveys were prompted up to 7 times per day (between 7:00 am and 8:00 pm). On weekdays, EMA surveys were prompted up to 3 times per day (between 3:00 and 8:00 pm). Participants were instructed to proceed with their normal daily routines during the assessment period. The participant’s mobile phone would chime and vibrate to prompt the child to stop his or her current activity and answer EMA survey, which took approximately 2 min to complete. Assessments did not occur during school holidays or summer. At each prompt, children were asked: “In the past 2 HOURS, which of the following have you done? (choose all that apply).” Response options included “sports or exercise” and “TV, videos, and/or video games.”

Children were also provided Actigraph accelerometers (Model GT3X, Actigraph Corp., Pensacola, FL) and instructed to wear the devices on their right hip for the same 8 consecutive days as EMA data collection. MVPA was defined based on age-specific Freedson cut points [23], whereas ST was defined as <100 activity counts per minute [24]. Nonwear was defined as 60 min of consecutive zero count epochs [25], and only valid accelerometer wear time was used for this analysis. All accelerometer measurements were time-stamped so that they could be linked to the same time windows as EMA prompts. The University of Southern California Institutional Review Board approved all the procedures.

Frequencies and mean values were calculated for participant demographic characteristics and for EMA and accelerometer variables stratified across participant demographic characteristics and weekends versus weekdays. The mean and standard deviation of the accelerometer-measured activity stratified by yes or no EMA reports of PA (sports or exercise) and sedentary behavior (TV, videos, or video games) were also calculated. EMA prompt compliance was calculated as the proportion of prompts answered out of the total prompts. Additionally, multilevel logistic regression models were utilized to investigate whether age, sex, ethnicity, BMI-z, or weekends versus weekdays were associated with EMA prompt compliance (scored as yes or no). Linear mixed models were used to assess whether age, sex, ethnicity, BMI-z, or weekends versus weekdays were associated with nonvalid accelerometer time.

The relationship between the minutes of the accelerometer-measured ST within the last 2 h and EMA reports of sedentary screen behaviors within the same 2-h time window was investigated via linear mixed models using PROC MIXED in SAS v9.4 (SAS Institute, Cary, NC). Mixed models were used to adjust for the clustering of EMA responses that were nested within each child [26]. The dichotomous EMA reports of sedentary behaviors (ie, TV, videos, or video games) in the past 2 h were the independent variable, and the total minutes of the accelerometer-measured ST in the past 2 h were the dependent variable. All ST models were adjusted for age, sex, ethnicity, BMI-z, and weekends versus weekdays. These covariates were also tested as moderators by multiplying the main effect terms together to create 2-way interaction terms between EMA reports of ST with age, sex, ethnicity, BMI-z, and weekends versus weekdays. They were then entered into the models separately to test the significance of the interaction. Post hoc analyses were conducted where the models were stratified by any significant interaction variables identified. All models were also controlled for between-subject (BS) effects; this was done by creating BS and within-subject (WS) versions of the predictors to indicate an individual’s mean variation from the grand mean (using grand-mean centering) and one’s variation from his or her own mean (using person-mean centering) at any given prompt [27].

Traditional linear mixed models that assume a normal distribution are not appropriate for MVPA data because they are typically skewed with an inflated number of zero values [28]. Therefore, we used a 2-part model, which utilizes a mixture of logistic regression for zero MVPA values and gamma regression for positive MVPA values [28-30]. The logistic regression portion (zero portion) of the model predicts whether the participant was not active (odds of no activity), whereas the gamma regression portion (positive portion) predicts the expected amount of MVPA on occasions when the participant was active [28]. Thus, there are two interpretations of the results when utilizing this modeling method—the likelihood of no activity (zero portion) and the expected amount of activity when the participant was active (positive portion).

The 2-part models assessed the association between EMA reports of sports or exercise within the last 2 h and accelerometer-measured minutes of MVPA during that same 2-h time window using the “gsem” command in Stata 14.2. These models were adjusted for age, sex, ethnicity, BMI-z, and weekends versus weekdays. The aforementioned covariates were also tested as moderators by multiplying the main effects terms together to create 2-way interaction terms between EMA reports of PA with age, sex, ethnicity, BMI-z, and weekends versus weekdays. They were then entered into the models one at a time to test the significance of the interaction. Post hoc analyses were conducted, in which the models were stratified by any significant interaction variables identified. All models were also controlled for BS effects by creating BS and WS versions of the predictors using the same method, as previously mentioned [27].

Our sample consisted of 192 children with available EMA and accelerometer data of the 202 children, in total, enrolled in the MATCH study (Figure 1). As indicated by the flow diagram, exclusion may have occurred for a number of reasons, ranging from technical issues to unanswered EMA prompts. The mean (SD) age of the participants was 9.6 (0.9) years. Half (49.0%, 94/192) of the sample were boys, and 54.2% (104/192) were Hispanic. Of the participants, 38.0% (73/192) of the children were classified as overweight or obese based on their BMI-z. EMA prompt compliance was 75.7% (2158/2851), which is approximately the average level of compliance compared with EMA studies conducted on other samples of children [31]. A total of 157 participants (81.8%) completed 50% or more of possible EMA surveys, consistent with other studies on similar samples [32]. In total, 16 participants (8.3%) completed 100% of EMA surveys prompted during the assessment period of this study. Multilevel logistic regression analyses indicated that the likelihood of EMA prompt compliance was greater on weekends than on weekdays (odds ratio, OR 1.3, 95% CI 1.1 to 1.5). However, there were significantly more nonvalid accelerometer minutes on weekends than on weekdays (beta=16.9, P<.001). Additionally, as the child BMI-z score increased, there was a lower likelihood of EMA compliance (OR 0.8, 95% CI 0.7 to 0.9). No other demographic characteristics were significantly associated with EMA prompt compliance. No demographic characteristics were associated with nonvalid accelerometer time.

The mean (SD) of the accelerometer-measured MVPA in the 2 h before EMA prompt was 10.4 (14.6) min. The mean (SD) of the accelerometer-measured ST in the 2 h before EMA prompt was 65.8 (21.2) min. Children reported sports or exercise in 37.4% (807/2158) of EMA prompts, and TV, videos, video games were reported in 47.3% (1021/2158) of prompts. Table 1 presents additional descriptive statistics on the accelerometer and EMA variables across participant demographic characteristics and on weekends versus weekdays, whereas Table 2 presents the mean (SD) accelerometer-measured activity stratified by EMA-reported sports or exercise and TV, videos, or video games.

Results from the linear mixed models investigating the relationship between EMA-reported sedentary behaviors and accelerometer-measured minutes of ST are shown in Table 3. Child EMA reports of engagement in the past 2-h TV, videos, or video games were associated with the greater minutes of the accelerometer-measured ST during the same time window (beta=7.3, 95% CI 5.5 to 9.0, P<.001). Additionally, the interaction between EMA-reported ST and sex was significant (beta=4.9, 95% CI 1.4 to 8.5, P=.01, Table 3); the strength of the association between EMA and accelerometer measures of ST was significantly different between boys and girls. Stratified post hoc analyses indicated that this association was twice as strong in boys (beta=9.9, 95% CI 7.2 to 12.6, P<.001) than that in girls (beta=4.9, 95% CI 2.6 to 7.2, P≤.001). No other significant moderators were found.

The results of the models investigating the relationship between EMA-reported sports or exercise and accelerometer-measured minutes of MVPA are shown in Table 4. The positive portion (the expected amount of activity measured by the accelerometer on occasions when the participant reported sports or exercise via EMA) and zero portion (likelihood of no activity measured by the accelerometer on occasions when the participant reported being active via EMA) of the 2-part model indicate a significant relationship between EMA and the accelerometer-measured activity in the total sample (zero portion estimate=−0.8, 95% CI −1.1 to −0.6, P<.001; positive portion estimate=0.6, 95% CI 0.5 to 0.7, P<.001). When the participant was active (according to the accelerometer) and reported sports or exercise via EMA in the last 2 h, this was associated with an 80.4% increase in the minutes of MVPA within that same time frame. Furthermore, when sports or exercise was reported via EMA, the odds of no MVPA measured by the accelerometer decreased by a factor of 56.8%.

The interaction term between EMA reports of sports or exercise and weekends versus weekdays was statistically significant, indicating a moderation of the association between EMA-reported sports or exercise and accelerometer-measured MVPA (zero portion estimate=−0.3, 95% CI −0.8 to 0.2, P=.24; positive portion estimate=0.2, 95% CI 0.05 to 0.4, P=.01). The relationship between the accelerometer-measured MVPA and EMA reports was stronger on weekends than on weekdays (Table 4). According to the positive portion of the model, if the participant had any accelerometer-measured MVPA, reporting sports or exercise via EMA was associated with a 103.4% increase in minutes of MVPA on weekends compared with only a 53.7% increase in minutes of MVPA on weekdays. Furthermore, when sports or exercise was reported via EMA, the odds of no accelerometer-measured MVPA decreased by 65.7% on weekends, whereas this was 30.2% on weekdays, according to the zero portion of the model.

In addition to weekends versus weekdays, age was found to be a significant moderator of the relationship between EMA-reported sports or exercise and accelerometer-measured MVPA (zero portion estimate=−0.1, 95% CI −0.4 to 0.2, P=.43; positive portion estimate=0.1, 95% CI 0.02 to 0.2, P=.02). This relationship was stronger in participants above the mean age of 9.6 years than in those below the mean age (Table 4). In children with an age above the mean age, if he or she had accelerometer-measured MVPA and reported sports or exercise, then there was a 101.4% expected increase in minutes of MVPA, whereas this expected increase was just 60% in children below the mean age, according to the positive portion of the model in each subsample. Furthermore, the odds of no accelerometer-measured MVPA decreased by a factor of 60.7% when exercise or sports were reported via EMA in children above the mean age compared with the reduced odds by a factor of 49.8% in children below the mean age, as indicated by the zero portions of each model.

Finally, ethnicity was also determined to be a significant moderator of the relationship between the accelerometer-measured MVPA and EMA-reported sports or exercise (zero portion estimate=−0.1, 95% CI −0.6 to 0.5, P=.86; positive portion estimate −0.2, 95% CI −0.4 to −0.1, P=.01). This relationship was stronger in non-Hispanic versus Hispanic children (Table 4). According to the positive portion of the model, if the participant had accelerometer-measured MVPA and reported sports or exercise, there was a 107.5% expected increase in minutes of MVPA in non-Hispanic children, whereas this increase in daily minutes of MVPA was only expected to be 61.6% in Hispanic children. The estimates from the zero portion of the model indicated that when the participant reported activity via EMA, the odds of no accelerometer-measured MVPA reduced by a factor of 56.0% in the non-Hispanic children. Similarly, these odds were reduced by a factor of 57.3% in the Hispanic participants. No other significant moderators were found between the accelerometer-measured MVPA and EMA reports of sports or exercise.

This is the first study comparing ST and MVPA measured concurrently by accelerometry and EMA-reported sedentary behaviors (ie, TV, videos, or video games) and PA (ie, sports or exercise) in children while also testing the moderators of the aforementioned relationships. Results indicated that EMA-reported sedentary behaviors were strongly positively associated with the accelerometer-measured minutes of ST during the same 2-h time frame. These findings indicate that EMA may be a promising method for capturing the specific forms of sedentary behavior through self-report with a very short-term recall window. Furthermore, EMA can provide contextual information such as where and with whom the behavior was performed [13]. The social and physical environments are important in understanding complex health behaviors, including sedentary behaviors [33]. For example, the built environment [34], peer relationships [35], and the day of the week [36] have all been shown to influence the levels of sedentary behavior in children. Other retrospective self-report tools for assessing sedentary behavior, such as the outdoor playtime recall questionnaire, are unable to provide such details surrounding sedentary behaviors and have demonstrated weak correlations with the accelerometer-measured ST [37,38]. Therefore, EMA may be more effective at capturing factors relevant to ST than other self-report measures previously utilized by investigators.

Although this evidence suggests that EMA is a helpful tool for gaining a better understanding of sedentary behavior in children, the results suggest that it may perform better in boys. The relationship between the accelerometer-measured ST and EMA reports of TV, videos, or video games was stronger in boys than in girls. These differences may emerge from the differences in leisure time sedentary behavior preferences in boys versus girls [39]. Studies indicate that boys spend more time playing computer games [40], whereas girls may prefer sedentary activities such as painting or drawing and playing musical instruments [39]. Therefore, EMA item capturing screen-based behaviors such as video games may have been a better indicator of the boys’ ST in our sample.

Results also showed a strong association between the accelerometer-measured MVPA and EMA reports of sports or exercise. When sports or exercise was reported in the past 2 h via EMA, significantly more minutes of MVPA were recorded by accelerometers during this time frame, and this was consistent with previous findings in adults [41]. A recent study comparing retrospective self-reports of PA and accelerometer-measured PA in youth found no relationship between the two [37]. Moreover, participants have been shown to overestimate the amounts of PA that they engaged in by an average of 596 minutes per week when utilizing retrospective questionnaires [17], highlighting the need for more effective self-report methods, particularly in children. The results of this study suggest that EMA self-reported PA highly relates to the accelerometer-measured PA, and children did not necessarily overestimate physical activity to the same degree when reporting via EMA. Secondary analyses of our data further support this notion; when there were zero minutes of the accelerometer-measured MVPA in the previous 2 h, participants in our sample only self-reported sports or exercise via EMA on 18% of occasions. Therefore, EMA has the potential to address the aforementioned weaknesses of other self-report PA measures by reducing the prevalence of over reports of PA.

However, when utilizing EMA as a PA data capture tool in children, investigators should be cognizant of time-variant and time-invariant variables that may influence the construct validity of EMA prompts. Specifically, the accelerometer-measured MVPA was more strongly related to EMA reports of sports or exercise on weekends than on weekdays. Multiple studies have demonstrated that the amount and types of PA that children participate in can differ between weekdays and weekends [36,42]. Children may be more likely to engage in nonrecreational types of PA during weekdays, such as active school transport (ie, walking), which may not be effectively captured by EMA item assessing leisure time PA in this study. Furthermore, a recent study conducted in more than 6200 children aged 9-11 years indicated that engaging in active school transport during the week was related to a greater accelerometer-measured MVPA during those days [43]. Therefore, activities such as active school transport during the week may contribute to the greater discrepancy observed between the accelerometer-measured MVPA and leisure time PA reported via EMA during weekdays.

Additionally, we found that the strength of the association between EMA self-reports of sports or exercise and accelerometer-measured MVPA differed between age groups. In those above the mean age of 9.6 years, the observed association between the two measures was stronger. In a study of more than 1000 active children aged 5-15 years, the investigators found that younger participants engaged in a more intermittent type of active play, whereas older children accumulated PA through walking and organized sport [44]. These differences in PA patterns may account for the age variations in the measurement associations that were observed within the current sample. Therefore, EMA item measuring sports or exercise, specifically, may be more successful at capturing the types of PA accumulated by older children, whereas the objective measures may be considered as the preferred method for capturing younger children’s physical activity behaviors, consistent with previous findings [45].

Finally, EMA reports of sports or exercise were more strongly related to the accelerometer-measured MVPA in non-Hispanic versus Hispanic children. In a nationally representative survey of children aged 9-13 years, it was determined that non-Hispanic children were significantly more likely to be involved in organized sports than their Hispanic counterparts [46]. To further support this notion, a large study of Hispanic children determined that first-generation Hispanic participants were less likely to report engaging in sports compared with their second- and third-generation peers [47]. Therefore, it may be that Hispanic children are accumulating PA through activities other than sports as a result of cultural preferences [47]. Thus, the current EMA items regarding leisure time sports or exercise may not be optimal for capturing PA behaviors in Hispanic participants.

This study highlights the strengths and weaknesses of EMA as a self-report tool for assessing leisure time PA and sedentary behavior data in children. Overall, EMA reports relate highly to the accelerometer-measured MVPA and ST. However, the moderators of this relationship reveal the limitations of EMA. EMA prompts asking about TV, videos, or video games might be a better indicator of ST in boys than in girls. Additionally, EMA prompts measuring sports or exercise appear to be a better indicator of PA on weekends than on weekdays. Finally, EMA self-reports of sports or exercise may be more effective for assessing PA in older (above 9.6 years old in our sample) and non-Hispanic children. To address the limitations of EMA, investigators may tailor EMA items to capture the types of PA and sedentary behaviors typically performed by individual participants in their samples. If tailoring EMA items is not feasible, the moderators of the relationship between EMA reports and objectively measured activity levels should be considered when analyzing and interpreting EMA data. As a general recommendation, it is also suggested that future investigators utilize both accelerometers and EMA simultaneously, depending on the scope of the investigation. It may be more useful to utilize EMA in studies assessing different types of behaviors (eg, reading vs computer use) that are being performed at any given moment, whereas accelerometers may be more useful in circumstances when investigators are interested in the overall frequency or duration of activity behaviors. Informed instrument selection will ultimately increase our understanding of PA and sedentary behaviors and how they relate to preventable health issues.

Although our study has several strengths, there are limitations to note. Depending on their placement, accelerometers cannot detect all bodily movements such as upper body activities, and their detection of activity is sensitive to chosen cut points. Furthermore, EMA responses may be subject to recall bias, though to a lesser extent, compared with the traditional retrospective self-report strategies [13]. Therefore, a study comparing two tools that have inherent limitations may ultimately be considered as a weakness. Additionally, we were not able to assess whether the amount of time elapsed between EMA prompt and the provision of an answer moderated EMA-accelerometer associations in this study. Furthermore, children were not EMA-prompted before 3 pm on weekdays; therefore, the results from this study may not generalize the behaviors and activities performed during school hours. In addition to this, the first prompt after 3 pm on weekdays asked participants about their behaviors “since you woke up this morning,” as opposed to asking about the previous 2-h behaviors. Therefore, the behaviors reported during this EMA prompt may not reflect the past 2-h activity levels measured via accelerometer. However, post hoc exploratory analyses removing the first prompts of weekdays (the prompts asking about behaviors since the participant woke up that morning) minimally altered our parameter estimates and results. This pattern suggests that these EMA items did not influence our findings.

Another limitation of this study is that contextual (eg, environmental or social) data were not assessed, and this may influence EMA-accelerometer associations; future studies should attempt to address this limitation. EMA prompt compliance was greater on weekends and in children with lower BMI-z. Thus, our findings may not be as generalizable to data captured on weekdays as well as data collected in heavier participants. Additionally, there was more nonvalid accelerometer time on weekends, which presents as an additional limitation for generalizability. Finally, our sample from Southern California metropolitan community, which contains more than 50% Hispanic participants, may differ from the general population of youth living in the United States and therefore may limit the generalizability of our results. Future studies should attempt to address these generalizability issues.

Findings indicate that EMA reports of TV, videos, or video games were strongly related to the accelerometer-measured ST during the same 2-h time frame. However, this relationship was stronger in boys than in girls. Although EMA reports of sports or exercise were associated with the accelerometer-measured MVPA, time-variant (weekends vs weekdays) and -invariant (age and ethnicity) variables were found to be the moderators of this relationship. EMA reports of sports or exercise and accelerometer-measured MVPA were more strongly associated on weekends, in older children, and in non-Hispanic participants. These moderators can be addressed by tailoring EMA items designed to capture PA and sedentary behaviors based on specific participant demographics and the day of the week. Taken together, this study supports EMA as a useful self-report tool for capturing PA and sedentary behavior in children because it demonstrates a high correlation with objectively measured activity levels.