This study was conducted in Belgium. Healthy older adults (aged ≥65 y) were recruited between November 2019 and March 2020 using convenience sampling (ie, flyers on social media and contacting associations for older adults). Self-reported exclusion criteria were (1) impaired cognition (ie, diagnosed with dementia and Alzheimer or other cognitive diseases), (2) severe impairment of vision and hearing, (3) inability to walk 100 m and stand or sit independently, (4) impairment of fine motor skills, and (5) insufficient knowledge of the Dutch language.

The Checklist for Reporting EMA Studies reporting guidelines proposed by Liao et al [29] were used to describe this EMA study. Participants were visited twice at home. During the first visit, informed consent was signed; sociodemographics (ie, sex; age; BMI; educational level, ie, nontertiary education—none, primary education, or secondary education—and tertiary education—higher education or university education), main occupation before retirement, marital status, having children and grandchildren were collected through a paper-based questionnaire; and instructions for the measurement period were given. In addition, the EMA app was installed on the participants’ smartphones, followed by a brief training on how to use it (ie, opening the app and answering the EMA questionnaires) by providing printed screenshots of the app. The visit was followed by one monitoring period of 7 consecutive days, consisting of 5 weekdays and 2 weekend days, although not all participants started the measurement period on the same day of the week. During this measurement period, participants were asked to answer 6 EMA questionnaires per day using a smartphone app and to wear an Axivity AX3 accelerometer to monitor their PA. After these 7 days, a second home visit was performed, during which the measurement materials were reassembled.

Ethics approval was obtained from the Ghent University Hospital Ethics Committee before the start of the study (2019/0192).

The Smartphone EMA³ (SEMA³) app is a suite of software for intensive longitudinal survey research that can be used on iOS and Android smartphones [30]. The SEMA³ app triggered 6 time-based EMA questionnaires per day, between 9 AM and 10 PM, for 7 consecutive days (ie, the participants were required to answer 42 questionnaires in total). Each EMA questionnaire was randomly triggered within a predefined time frame of 1 hour (eg, the time frame from 9 AM to 10 AM, in which the EMA trigger randomly appeared at 9:38 AM). In total, 6 time frames were predefined per day. If the participants did not respond to the initial trigger, 2 reminders were given after approximately 5 minutes and 10 minutes; 20 minutes after the initial trigger, the questionnaire was unavailable until the next scheduled trigger. Participants were asked to use their own smartphone during the measurement period (the lowest acceptable operating systems were Android 5.0 and iOS 12.4), but participants who did not own a smartphone were provided with a Wiko Lenny 3 smartphone (Android 6.0).

The COM-B framework was used as a guiding framework to compile the first version of the questionnaire from items previously used in research [31-33]. Experts in psychology, health sciences, and EMA were involved at multiple stages in the development of the EMA questionnaire. Subsequently, cognitive interviews were conducted with 10 older adults, which led to further adjustments of the questionnaire, mainly concerning the comprehensibility of the items. The final version of the EMA questionnaire assessed the following components in this fixed order: emotions, physical complaints, and the constructs’ intention and self-efficacy. However, the order of the questions within the components’ emotions and physical complaints was randomized and therefore could change over the EMA triggers. In total, the EMA questionnaire consisted of 18 items [34]. However, based on the results of a previous analysis of the same EMA data [34], only those determinants with more within-subject variation (>50%) than between-subject variation were selected for this study. Consequently, only 7 items of the EMA questionnaire were analyzed in this study: relaxation, satisfaction, irritation, feeling down, fatigue, intention, and self-efficacy. The items for the emotions relaxation, satisfaction, irritation, and feeling down were originally developed by the research group of Philippe Delespaul at the University of Maastricht [31]. The item for physical complaint fatigue was selected from the validated Patient Health Questionnaire-15 [32]. Finally, the items assessing intention and self-efficacy toward PA were based on items that are frequently used in our research group [33] but were specifically adapted by the authors for this EMA study. All items were assessed on a 7-point Likert scale and were presented in Dutch (their English translation is available in Multimedia Appendix 1).

To capture participants’ PA, all participants wore an Axivity AX3 accelerometer on their nondominant wrist during the same 7 consecutive days as the time-based EMA. The Axivity AX3 accelerometer is a reliable and valid device for measuring PA [35]. Participants were instructed to wear the accelerometer during the whole day and night and to remove the accelerometer only during water-based activities. Data were extracted using the OMGUI software (Open Movement) [36] and then processed in R (version 4.0.1; R Foundation for Statistical Computing) [37]. The raw data downloaded from Axivity AX3 were first reduced by averaging the Euclidean norm minus one values (in milligravitational units, ie, mg) over 1-second epochs and then 1-minute epochs, after which cut points for older adults by Sanders et al [38] were applied to categorize individual minutes as sedentary (≤57 mg), LPA (57-104 mg), or MVPA (≥104 mg). The total number of LPA and MVPA minutes was calculated for the time frames of 15, 30, 60, and 120 minutes after the trigger. In addition, the total number of TPA minutes was obtained for these time frames by summing LPA and MVPA minutes.

Multilevel regression analyses were performed using the lme4 package [39] in R [37]. Residuals were plotted and visually inspected to check for linearity and normality assumptions. Because the PA data in the 4 time frames after the EMA trigger (ie, 15, 30, 60, and 120 min) included an excessive number of “true” null values (ie, not all participants were physically active during all time frames), Hurdle models [40,41] were fitted. The Hurdle models consist of 2 parts: a logistic regression model, estimating the odds of engaging in PA after the EMA trigger, and a linear model, estimating associations with the amount of PA among those who performed at least some PA at 15, 30, 60, and 120 minutes after the EMA trigger. First, the logistic regression model was applied, for which all PA variables were recoded into dichotomous variables (ie, 0 vs at least 1 min of LPA, MVPA, and TPA, respectively). For each determinant, a separate model was created with each PA variable (ie, LPA, MVPA, and TPA in the 4 different time frames) as an outcome variable, which led to 84 models in total (ie, 7 determinants × 3 PA outcome variables × 4 time frames). Further model assumptions were visually checked (ie, outliers and influential observations). Second, to construct the linear model, 3 different models were applied to the original PA values for each determinant separately to check which model best fitted the data (ie, Gaussian, Poisson, and negative binomial). The models were fitted separately for each determinant and each PA outcome. Of these 3 models, the model with the lowest Akaike information criterion value indicating a better model fit was chosen. The Akaike information criterion was, in all cases, the lowest for the negative binomial model; therefore, the negative binomial was applied in all cases. The models were fitted for each determinant as between-subject (ie, mean of the variable at the subject level) as well as within-subject (ie, individuals’ score minus their mean score), but only the within-subject associations were considered in this study. In addition, for the linear models, further model assumptions were visually checked (ie, homoscedasticity, outliers, and influential observations). The level of significance was set at α<.05. To limit the number of results for this study, all results for TPA can be found in Multimedia Appendices 2-4.

In total, 67 older adults participated in this study. All participants completed the measurement period of 7 days. To be included in the analysis, participants had to respond to at least one-third of all triggers (ie, 14 out of 42). As a result, data from 3 participants were excluded, and 64 participants were included in the analysis. Their descriptive statistics are presented in Table 1. In total, 11 participants used a Wiko Lenny 3 smartphone because they did not have a smartphone of their own.

A total of 2690 triggers were sent during the study, of which 30 were not delivered as intended because of technical issues (ie, more than 6 triggers a day were sent or triggers were sent outside the predefined time frames, eg, during the night) and were excluded from the analysis. Of the remaining 2660 triggers, 2057 EMA questionnaires were completed (response rate of 2057/2660, 77.33% and a mean of 32.1 completed questionnaires per participant). The median response latency was 0.00 (quarter 1=0.00, quarter 3=4.00) minutes. The median time needed to complete the EMA questionnaire was 1.79 (quarter 1=1.43, quarter 3=2.31) minutes. Descriptive statistics for LPA and MVPA are presented in Table 2, and those for the EMA items are presented in Table 3.

This study examined the within-person associations of multiple determinants of the capability and motivation component of the COM-B framework with subsequent LPA, MVPA, and TPA. Multiple associations with LPA and MVPA in the 15, 30, 60, and 120 minutes after the trigger were found: irritation, feeling down, intention, and self-efficacy were positively associated with subsequent PA, whereas relaxation, satisfaction, and fatigue were negatively associated with subsequent PA. All results for TPA can be found in Multimedia Appendices 2-4.

In line with our hypothesis, intention and self-efficacy were positively associated with subsequent PA, and fatigue was negatively associated with subsequent PA.

Higher levels of intention or self-efficacy were associated with higher odds of performing subsequent PA or with more minutes of subsequent PA. This finding is in line with previous EMA research in adults and older adults [20,22,24]. In adults, it was found that intentions and self-efficacy positively predicted subsequent accelerometer-assessed MVPA in the 2 hours after the EMA trigger [20]. In older adults, greater levels of self-efficacy predicted higher levels of subjectively measured subsequent MVPA in the 4 hours after the EMA trigger [24]. Intention and self-efficacy were also found to predict increases in the subsequent time spent upright in older adults [22]. Multiple theoretical frameworks [42-44] stress the importance of intention and self-efficacy as a gateway to or as a condition for behavior change, and therefore, positive associations between these constructs and subsequent PA seem natural. However, these theories often do not assess these concepts as dynamic, whereas in a previous study using the same data [34], we found that these are time dependent. Thus, the novelty observed in this study is that these generic associations among intention, self-efficacy, and PA also occur in shorter time frames.

Furthermore, in this study, a negative association was found between the physical complaint fatigue and subsequent LPA in the 15, 30, 60, and 120 minutes after the trigger and subsequent MVPA in the 60 and 120 minutes after the trigger. This means that a higher level of fatigue resulted in lower odds of being physically active and in fewer minutes of subsequent PA. In previous research, fatigue was often mentioned as a barrier to performing PA [25], which consequently might lead to less subsequent PA. However, Dunton et al [24] and Liao et al [19] observed that fatigue was unrelated to subsequent accelerometer-assessed LPA in the 15 and 30 minutes after the EMA trigger in adults [19] and unrelated to subjectively measured subsequent MVPA in the 4 hours after the EMA trigger in older adults [24]. In contrast to the study by Liao et al [19], in this study, significant negative associations of fatigue with LPA were found in all time frames. Although Liao et al [19] did not examine longer time frames, differences in the associations detected 15 and 30 minutes after the trigger might be explained by differences in participants (ie, low-active adults in the study by Liao et al [19]). Differences in results with the study by Dunton et al [24] might be explained by the large time frame and subjective assessment of subsequent PA.

Some of the associations found in this study were unexpected and in contrast with our hypothesis. Irritation and feeling down were seen as negative emotions and were expected to be negatively associated with subsequent PA. However, in this study, the opposite was observed, that is, higher levels of irritation and feeling down resulted in higher odds of being physically active or in more minutes of subsequent PA. Previous research found that higher negative affect (ie, measured as emotionally upset, annoyed, angry, sad, or depressed [45] and as anxious, stressed, depressed, and angry [19]) was associated with lower levels of MVPA in subsequent time frames in older adults [24] and in adults [19]. However, in some other studies, negative affect was positively associated with LPA in the 4 hours after the trigger in older adults [24] and with bodily movement in 15- and 30-minute time frames in adults aged 18 to 73 years [46]. Engaging in PA can improve mental health [1,47-50]; therefore, PA might be used as a coping strategy to counter or reduce the negative emotions of irritation and feeling down to improve affect [51]. In a previous study using the same data [34], we found that for both irritation and feeling down, participants reported in more than 90% of their answers that they were not feeling irritated or feeling down at all, which might have caused a floor effect and influenced the associations identified here. Finally, in this study, higher levels of relaxation and satisfaction resulted in lower subsequent PA. Although relaxation and satisfaction are positive emotions and therefore positive associations with PA were expected, the association with PA might be different than in the case of other positive emotions (eg, cheerfulness). It is possible that older adults who felt relaxed or satisfied did not wish to be active but rather further enjoyed their state of relaxation or satisfaction. However, this is merely an assumption and should be examined further.

In this study, associations were found in different or multiple time frames depending on the determinant that was examined. In light of JITAIs, insight into the importance of determinants in specific time frames is crucial to achieving the greatest behavioral change. For example, when associations are found in the 30 and 60 minutes after the trigger, it might be interesting to examine whether suggestions that encourage the older adult to be active in the next hour instead of “right now” might lead to an increase in their subsequent PA or to examine whether suggestions to be active can be sent after 30 to 60 minutes (after the assessment of the determinant) and in this way increase the older adults’ PA. Both can be examined by using multiple mini interventions. Intention and self-efficacy were positively associated with subsequent PA over all time frames, which shows the importance of these constructs in relation to PA in both short and long time frames. When intention and self-efficacy are already high, it might be sufficient for JITAIs to provide an activity suggestion to increase PA levels. In contrast, when intention and self-efficacy are low, it might be an opportunity for JITAIs to offer behavior change techniques aiming to increase intention and self-efficacy and consequently increase PA. For example, previous research showed that the behavior change technique “provide information on the consequences of behavior in general” is associated with positive changes in intention [52] and that “action planning” was associated with higher self-efficacy [53].

Another observation that could be drawn from the results of this study is that, for some determinants, stronger associations were found with LPA, whereas other determinants were associated with MVPA. For example, emotional satisfaction and feeling down are mainly important for LPA, whereas fatigue is also associated with MVPA. However, these findings are not surprising because they concern different behaviors (ie, examples of LPA are walking and climbing the stairs, while running and cycling are classified as MVPA); therefore, different determinants might be important. Nevertheless, these are important findings for the development of tailored interventions, as different individual determinants may be important for different types of PA that older adults wish to increase. Specifically, suggestions can be tailored according to the behavior of interest, for example, in the case of LPA, providing suggestions to be active when the older adult feels down, and in the case of MVPA, avoiding giving suggestions to be active when the older adult feels fatigued. Intention and self-efficacy showed positive associations with both LPA and MVPA, which provides interesting information for future PA interventions, because both LPA and MVPA can be enhanced in older adults by targeting both determinants.

A previous study using the same data [34] found that multiple individual-level determinants (eg, satisfaction, intention, self-efficacy, and fatigue) are time dependent and therefore can vary within subjects within days. These time-dependent fluctuations are important to keep in mind for the personalization of JITAIs, so that the moment to provide suggestions to be active can be adjusted to the emotional, cognitive, and physical state of the individual. By regularly assessing the time-dependent determinants (eg, multiple times per day) in JITAIs, individually tailored suggestions to improve PA can be provided. Furthermore, it was found that the variation between days was limited in older adults [34]. Therefore, it might be possible to monitor older adults for just a few days before introducing a JITAI to capture the most receptive moments without overburdening them with an extensive monitoring period (eg, 2 weeks). This information, combined with the identified within-person associations found in this study, provides useful information for the development of future JITAIs.

A first strength of the study was the repeated assessment throughout the day using EMA. The use of EMA reduces recall biases by capturing present experiences rather than beliefs or ratings based on memory. Furthermore, the assessment occurs in the individuals’ natural environments and social contexts, which increases ecological validity. Finally, fine-grained information was provided using multiple repeated assessments over time. A second strength is that this study fills an important gap in the literature, as previous research on the associations of determinants with subsequent PA is limited, especially in older adults. Third, participants’ PA was accelerometer-assessed using Axivity AX3 accelerometers rather than using self-reported measures.

This study had some limitations. The first limitation is the limited generalizability of the study results, as nonprobability sampling was used to recruit older adults. Therefore, the findings of this study might be specific to the study sample used in this study. It is recommended that future studies use a more random sampling approach to obtain a more heterogeneous study sample and generalize the study results to a wider population of older adults. Second, in this study, PA was considered an outcome variable, with determinants as predictors. However, examining accelerometer-assessed PA as a predictor and the determinants as outcome variables in an EMA study might contribute to a better understanding of the influence of PA on subsequent emotions, physical complaints, and intention and self-efficacy. Third, it is possible that in some cases there was less than 2 hours between 2 triggers, which might have caused some overlap and influenced the associations found in the 120 minutes time frames. This aspect should be considered in future studies. Fourth, although most items of the EMA questionnaire (ie, 4 out of 7) were specifically developed for EMA research, psychometric information was unavailable. It is recommended that future EMA studies use items that are validated for EMA or conduct a validation study before the start of the study. Fifth, we did not consider possible confounders, such as preceding PA, emotions, or other unmeasured variables (eg, sleep). Possible confounders, especially preceding PA, should be considered in future EMA studies. To do so, we first need to explore which time frame of the preceding PA is best to use, since many different time frames can be considered (eg, 15 min, 30 min, and 60 min). However, this exploration was beyond the scope of this study. Sixth, although wrist-worn devices are a common wearing position for the assessment of PA, it might still be possible that there is an overestimation of PA. However, the possibility of an overestimation of PA is limited because the validated cutoff points [38] were used in this study. Although a different accelerometer and different sampling rates were used in the study of Sanders et al [38], the equivalence of the key PA outcomes for GENEActiv and Axivity accelerometers has been proven in previous research [54].

Overall, this study showed that the emotions irritation and feeling down and the constructs intention and self-efficacy were positively associated with subsequent PA in older adults, whereas the emotions relaxation and satisfaction and physical complaint fatigue were negatively associated with subsequent PA. Using EMA, this study yielded new knowledge about these associations and the time dependency of the determinants, which is valuable for future interventions. By monitoring older adults for a few days, the most receptive moments for triggering them to be more active can be captured, and this information can be used in JITAIs to provide individual tailoring and promote PA more effectively.