Study design

Adults in the United States and Germany were recruited to participate in research-related activities for 10 consecutive weeks. While enrolled, participants used their personally-owned smartphones to enact event-contingent ecological momentary assessments (pre- and post-activity) in response to purposeful (performed with intention, awareness) physical activity sessions lasting ten or more minutes. The EMA data were used to computed person-specific and variable-centered associations. All procedures described were approved for each site (The University of Tennessee, Knoxville Institutional Review Board; Ethics Committee of the Faculty of Economics and Social Sciences at University Tübingen) and all participants provided written, informed consent to engage in research-related activities by signing a standardized paper consent form, which was witnessed and also signed by the overseeing research associate.

Inclusivity in global research

Additional information regarding the ethical, cultural, and scientific considerations specific to inclusivity in global research is included in the S1 Checklist.

Procedures

Instrumentation

Data processing

Pre-, post-, and retrospective physical activity databases were merged and organized, such that all single LTPA session reports from a given participant were chronologically listed in subsequent rows to support person-specific analyses of association. Columns included survey meta-data (e.g., survey duration, date and time of completion) as well as target study variables. The database was further organized to only include exercise-related and non-exercise-related LTPA (i.e., excluding any sessions designated as domestic or occupational). One report from one participant was excluded, as it pertained to an extreme outlier (11-hour activity session; mean duration of remaining 33 sessions = 47.78±14.53 min). Four participants from analyses due to reporting less than 5 LTPA sessions (minimum engagement considered to be study active). A separate database contained within-person averages for all target variables to conduct variable-centered analyses of association.

Statistical and visual analyses

All analyses described were conducted using the Statistical Package for Social Sciences (SPSS). Basic descriptive and frequency analyses were conducted to describe the sample and summarize within-person readiness states, behavior, and experiences. Coefficients of variation were computed (person-specific standard deviation divided by person-specific mean) to understand degrees of within-person variability regarding LTPA duration, intensity, and affective valence. Individuals demonstrating the lowest, median, and highest coefficient of variation for each outcome were visualized using standard line graphs.

Each individual’s most important motives and goals for LTPA were computed using their baseline BMZI data to further describe the participants based on existing classification approach of typical motive and goal profiles for LTPA [47]. Furthermore, radar plots were constructed for each individual to visually demonstrate within-person variability across all reported LTPA sessions regarding the 5 domains of readiness and 11 types of situational motivation.

To account for measuring most variables using ordinal scales, Spearman rank correlation procedures were used to compute size and direction of association (rho; ρ), as well as 95% confidence intervals to explore how pre-activity ratings of situational motivation and readiness related to subsequent activity duration, perceived exertion, and affective responses at the person-specific level. Of the 1,056 idiographic analyses (22 participants x 16 domains for situational motivation and readiness x 3 LTPA outcomes) there were 48 instances (across 6 participants) where analyses could not be conducted due to the individual reporting no variance in one correlate. When averaging session-to-session ratings within each person, data were kept in their ordinal form (zero decimal places) and we repeated the Spearman rank correlation processes to explore variable-centered associations. Forest plots were constructed to visualize interindividual variance in person-specific associations, whereas scatterplots of ranked values were constructed to visualize variable-centered associations.

Heat maps were constructed to broadly visualize the proportion of individuals (red = 0 individuals; dark green = highest number of individuals) assigned across seven categories: negligible association; negative associations (small, medium, large); positive associations (small, medium, large) based on idiographic analyses. Each heat map was organized first based on hypothesized determinant (motivation type or readiness domain), then listed in order from the strongest positive to the strongest negative nomothetic association. For the purpose of highlighting potential differences in interpretation, associations are referred to as small (.10 to .29), medium (.30-.49), or large (>.50) based on standard demarcations [48].

Summary of findings

The current study was designed to explore interindividual variability in person-specific associations regarding hypothesized predictors of volitional LTPA effort and experience, as well as highlight the potential for differing interpretations when data are assessed across individuals. By leveraging event-contingent EMA procedures to allow participants to report on purposeful (to them) instances of LTPA, this work demonstrates that–within a relatively small sample (N = 22) over a multi-week period of time (10-weeks)–situational motivations, readiness states, LTPA behavior (session-to-session mode, duration and intensity), and affective experience can vary generally across and dynamically within individuals. Further, this exploratory data provides proof-of-concept that how researchers and practitioners interpret relationships between physical activity-related variables is likely to be impacted by the level of analysis (idiographic vs. nomothetic). This work provides the requisite foundation to stimulate future research based on both our preliminary findings and study limitations.

Variability in reported LTPA characteristics

The observed variability across and within participants regarding LTPA in this naturalistic study reflects prior findings from single time-point or shorter term studies designed to quantify the complexity of physical activity behavior. Prior data collected for cross-sectional [49], as well as 7-day [50] and 14-day [51, 52] EMA designs demonstrate that sample populations report engaging various primary modes relating to aerobic, resistance, combined aerobic and resistance (e.g., Tae Bo, CrossFit), and group sports in their leisure time, in line with data provided in Table 2. Low to moderate consistency in session-by-session intensity (objectively measured and perceived) and volume (intensity multiplied by duration) has also been documented for volitional LTPA [50].

Research has also demonstrated that experiential responses vary across individuals [38, 40], as well as within individuals over several imposed exercise sessions [53] or over 7 to 28 days [50, 54–56]. Such variability reflects the expectation that physical activity-related experiences are multifaceted (physical, psychological, social effects), which may differ across mode and context [57]. Resultant experiences from physical stimuli are likely to be influenced, in part, by pre-activity perceptions of readiness and multi-thematic situational motivations, which have been shown to fluctuate over time [50, 58]. As demonstrated in this study (Fig 2), the degree of fluctuation may also be relatively person-specific, warranting further research to understand how low-to-high variance explains or predicts individuals’ physical activity behavior and related experiences.

Level of analysis and differing data interpretation

Our findings present the possibility of Simpson’s paradox regarding relationships between situational motivation, readiness, and target physical activity outcomes. Simpson’s paradox refers to instances where an association at the population level may be reversed within the subgroups comprising that population [59]. Using simulation models, Molenaar [60] suggests that–mathematically–data derived from interindividual variation cannot be generalized to results based on intraindividual variation. In acknowledging this possibility, Kievit [13] cautioned that differences in interpretation, especially when the direction of association are in opposition, can have significant implications for delivery of care in medical and social contexts. We offer two scenarios to demonstrate how differing interpretations may unfavorably impact practice related to physical activity promotion.

From our data, we consider the observed nomothetic pattern that individuals with stronger general motives to replenish energy tended to perform longer LTPA, on average. If this relationship is uncritically assumed to exist within individuals, an exercise professional may urge clients with relatively low energy levels to engage in longer-than-normal bouts of activity, reflecting the traditionally held belief that ‘exercise makes people feel better’ [40]. However, this interpretation opposes our preliminary observation that the larger proportion of participants demonstrated an inverse association between variables, meaning that stronger motives for replenishing energy were associated with lower reported duration over time.

The current analyses also revealed that the nomothetic interpretation that physical readiness had almost no association with recalled affective valence. This lack of association opposes a growing body of qualitative findings, which identify factors underlying perceptions of physical readiness (e.g., energy, fitness, body integrity) as influencing physical activity-related cognitions and affective experiences [61–63]. Relying solely on the nomothetic interpretation, a client’s poor perception of physical readiness may not dissuade a practitioner from imposing a scheduled, more strenuous exercise session–potentially eliciting a rather unpleasant experience. Conversely, based on observations that individuals reported LTPA sessions that were preceded by more unfavorable readiness states, it would also be shortsighted to conclude that physical activity should be generally be avoided completely if physical, mental, and emotional states are suboptimal.

Future directions in data analysis and intervention development

The potential for emergent subgroups, based on direction and magnitude of idiographic associations, has implications for future statistical analyses. The process of first examining person-specific data is recommended prior to estimating within-person and between-person effects using repeated-measures correlation or subsequent multi-level modeling [32], as the observation of small effects could be due to consistently small effects across subjects (i.e., good model fit) or an artifact of heterogeneous slopes (i.e., poor model fit). Sufficient time-series data can also be leveraged to conduct network analyses [21]. For example, group iterative multiple model estimation (GIMME) procedures can be applied to create personalized maps comprised of nodes (variables operationalizing target constructs) and edges (indicating relation between nodes that are unique to each individual or common across the sample) [64, 65]. Given that GIMME can adequately map between 3 and 20 variables, this approach may be particularly suitable to understand which types of situational motivation and domains of readiness are most powerful–for whom and in general–for predicting physical activity-related effort and experience. In line with an advancing focus on precision health [66], the need for robust idiographic exploration (and later replication and confirmation of findings) extends beyond our selected variables and pertains broadly to hypothesized determinants of physical activity adoption and adherence [10, 11] to effectively inform person-adaptive treatment approaches.

In anticipation of applying network approaches to broadly understand and quantify determinants of physical activity performance and experience (including our selected variables of situational motivation and readiness), we propose several key considerations. Given the potential for numerous different types, contexts and experiences of physical activity, the first consideration is to narrow the focus of subsequent studies. For instance, specifically assessing the impact of situational motivation and readiness on perceived pleasure during cycling exercise in cardiac rehabilitation may yield more meaningful, directly translatable information for that context more so than analyzing various forms for physical activity and assuming inferences are broadly transferable to all physical activity-related situations. The second consideration is that longer (or repeated) observation periods may be required. While a generalized target of 60 data points per person [67] has been recommended for GIMME, simulation studies should also be conducted to tailor sample size needs to specific research questions [21]. For example, Dai et al. [68] produced person-specific models for physical activity (in addition to other variables, such as mental sharpness and social engagement) with 36 participants who had 13–25 data points. As repeated performance of physical activity elicits physiological and psychological adaptations, it is reasonable to consider that initial person-specific maps of determinants may evolve (or become obsolete) over time. Thus, our third consideration is for future research to assess the utility of creating person-specific maps based on EMA of forecasted responses to hypothetical physical activity sessions, in order to collect the requisite amount of data in a more timely manner, without repeated physical stimuli.

It is critical to understand which determinants of LTPA effort and experience are specific to an individual or generalizable across groups to develop, refine, and tailor behavioral interventions. For example, developing hypothesized pathways by which a treatment can support physical activity from both nomothetic and idiographic perspectives aligns with calls for applying flexible, iterative optimization processes to candidate treatments prior to conducting randomized controlled trials [69]. Accounting for intraindividual variation and integrating person-specific determinants stands to improve recipients’ acceptability of a physical activity intervention, thusly enhancing engagement, compliance to the program, and long-term behavioral adherence [70].

Optimal intervention development and research also necessitates sufficient and replicable descriptions of active components. Given that enactment of FNLP–the exercise programming model that informed study variables–is hypothesized to foster more self-determined forms of motivation [28], intervention development efforts should incorporate terms and definitions according to the taxonomy of motivation and behavior-change techniques proposed by Teixeira and colleagues (2020) for health promotion contexts [71]. For example, FNLP ‘provides choice’ and encourages self-experimentation [28] as autonomy-support techniques to ‘assist in setting optimal challenge’ based on situational motivation and readiness states as a competence-support technique. As FNLP may be implemented by a professional (e.g., certified exercise physiologist, behavioral interventionist) who educates patients or clients, a complementary taxonomy proposed by Ahmadi and colleagues (2023) will also benefit intervention development and assessment efforts [72]. This taxonomy importantly specifies behaviors of the educator that can support or thwart autonomous motivation, in recognition outcomes are differently impacted via separate pathways (e.g., psychological need satisfaction and need frustration) [73].

Limitations

While we prefaced this work as exploratory and caution against premature inferences based solely our findings, we also acknowledge several limitations. First, these data were collected using a convenience sample of individuals who self-selected into the study, who can be described as rather active with high competences in support of a physically active lifestyle. For example, the overall perceived physical fitness and physical activity-related competency scores in the current sample were higher compared to prior reports [43, 74, 75]. Further research must be conducted to understand within-person associations in varied sample populations. A second limitation is that results based on event-contingent reporting of volitional LTPA may be skewed more positive, as individuals could be less likely to initiate reports in response to strongly negative experiences. Additionally, despite the prompts specifically asking individuals to recall feelings during activity, capturing data in the post-activity stay may still yield higher-than-actual ratings of pleasure. While this approach had been used previously [61] to overcome limitations in collecting in-task data, it will be necessary to replicate these methods in intervention or treatment settings, paired with wearable devices or direct observation to more accurately monitor physical stimuli and implement approaches to capture actual in-task ratings of affect.

This work is also limited by potential influences of translating survey items; the BMZI and PAHCO were translated from German to English and the ARMS, BREQ-3, and PPFS were translated from English to German. While structured forward-backward translation procedures (including bi-lingual native speakers) were applied, further validation is needed for surveys targeting hypothesized determinants and outcome variables. Specifically, the predictive validity of single-item measures is imperative [76], as their use has practical importance for minimizing survey fatigue [77]. A final limitation relates to the high degree of variability in the number of sessions reported; less data can compromise the quality of correlation analyses for certain individuals. Conversely, while participants in the current sample were explicitly asked to maintain normal patterns of behavior and were not given a minimum number of sessions to perform or incentivized to initiate EMA reports, we cannot with certainty rule out the possibility for reactivity. It is important to acknowledge that perceived enforcement of behavior (e.g., performing additional or undesired LTPA due to being observed by researchers) would compromise the aim to understand volitional behavior and experiential aspects thereof. In future studies, it may be beneficial to conduct debriefing assessments (quantitative surveys or qualitative interviews) to understand the potential for reactivity and determine strategies to minimize such effects. Thus, while the 10-week study described in this paper is substantially longer that most existing single-wave EMA studies pertaining to physical activity, which generally aim to capture data across several days or several weeks [78–80], we re-iterate the necessity of subsequent research to appropriately estimate person-specific or person-oriented effects with minimal reporting bias.