Abstract
Physical inactivity constitutes a pressing societal problem. To realize physical activity’s (PA) potential as a key health resource, mechanisms of PA engagement need to be understood. Laboratory and interventional studies documented that exercise relates to affective well-being (AWB) and suggested that AWB may shape PA behaviour. Digitalization enabled the investigation of how PA relates to AWB in everyday life, but findings from individual studies are ambiguous. Here we compiled 67 datasets (55.2% of eligible records) including 321,345 smartphone-based AWB ratings and nearly 1,000,000 h of accelerometer-measured PA (N = 8,223 participants) until December 2023 to clarify the nature and extent of PA–AWB associations. One- and two-stage individual participant data meta-analyses reveal that momentary AWB is associated with both prior (within, r = 0.05, 99.2% confidence intervals (CI) 0.03 to 0.06; between, r = 0.08, 99.2% CI 0.04 to 0.12) and subsequent (within, r = 0.04, 99.2% CI 0.03 to 0.05; between, r = 0.08, 99.2% CI 0.04 to 0.13) short-term PA in everyday life. Within persons, PA displays a positive association with energetic arousal, positive affective states and valence, yet a negative relation to calmness. The practical effect sizes are comparable to other daily life activities, with energetic arousal evincing the strongest relation to PA. Considerable heterogeneity in associations across individuals can be partially explained by sociodemographic moderators. Between participants, PA relates to positive affective states. The results document the critical relevance of PA–AWB relations in everyday life. They can contribute to the revision and development of health behaviour models and establish a starting point to approach behavioural, physiological and neuronal mechanisms underlying PA–AWB associations.
Similar content being viewed by others
Main
Worldwide, many humans are insufficiently physically active1,2,3. If people meet current physical activity (PA) recommendations by the World Health Organization4, global gross domestic product is estimated to increase by US$314–446 billion per year5, mostly attributable to PA counteracting physical6,7 and mental illness8,9,10. Recent health behaviour theories consider affective well-being (AWB) as one of the most potent psychological facilitators between PA and health outcomes and suggest it to be critically involved in motivating, maintaining and reinforcing PA11,12,13,14,15,16,17. In general, AWB is increasingly recognized as a key driver of human behaviour and experience; for example, AWB is shown to be involved not only in PA engagement but also in attention, learning, memory and decision-making processes18.
Since the 1980s, cross-sectional studies employing retrospective questionnaires and laboratory experiments provided insights into associations of PA with AWB19,20,21. These findings, however, are constrained by low ecological validity owing to artificial settings, recall biases and a focus on differences between individuals22,23,24. Critically, between-person differences cannot be employed as a proxy for intra-individual processes25, a well-known limitation referred to as the ecological fallacy26,27,28,29,30,31. Yet, to promote PA in daily life, it is of crucial importance to understand the within-person associations of PA and AWB in everyday life. In general, variations in AWB in everyday life plausibly affect how humans act and think and are currently suggested to drive key health behaviours such as PA and diet32. Insights into these daily life processes can personalize and tailor prevention and treatment by identifying affective facilitators and barriers therewith contributing to public health strategies32,33.
To this end, for the last two decades, researchers have collected intensive longitudinal data on the relationship of AWB and PA in daily life34. In practice, such research employs wearables including accelerometers for PA tracking and repeated ratings of AWB on smartphone-based electronic diaries as participants go about their daily routines33,35. This procedure bypasses aforementioned limitations by assessing momentary instead of retrospective experience complemented by device-based measures and by collecting extensive within-person data for each individual22,33,36,37. Of note, it is not only experimental control in laboratory and interventional studies that contrasts with ecological validity of real-life studies but also the type of PA being researched, that is, structured exercise sessions in laboratory and interventional studies versus incidental PA in daily life studies, which comprises exercise but also low-threshold non-exercise activity as PA facets.
Evidence from daily life studies regarding the existence, strength, directionality and circumstances of PA–AWB associations is ambiguous. Liao et al. were the first trying to resolve these ambiguities by conducting a short review in 201538. Differentiating three concepts of AWB (positive affect, negative affect and feeling states, that is, energy and fatigue), they found the strongest evidence for the bidirectional association of PA and positive affect, very weak evidence for PA and negative affect, and insufficient evidence for PA and feeling states. Similarly, in a systematic review on AWB–PA associations in children and adolescents from 2021, associations of PA and subsequent positive affect and energy emerged as most robust. There was some evidence for associations of valence and subsequent PA. Associations of PA with negative affect were not supported by literature39. Most recently, Timm et al.34 conducted a literature search, revealing that although the number of studies increased tremendously from 14 studies in 201538 to >66 in 2023, evidence remains ambiguous. In line with prior reviews as well as a range of prominent laboratory studies19,20,21, Timm et al. reported stronger support for bidirectional associations between PA and positive affect compared with negative affect. Reciprocal associations of PA with energetic arousal were most consistently supported by literature. For valence, less than half of the studies found significant positive associations with preceding or subsequent PA. Associations between calmness and PA appeared the most heterogeneous, exhibiting both positive and negative associations with PA. Finally, the review reported a high degree of heterogeneity of PA and AWB assessment and statistical methods34.
In sum, although intensive longitudinal data on AWB–PA associations are rapidly increasing, evidence remains ambiguous. The concepts and terminology of AWB strongly differ between studies (for example, mood, emotions, affect and core affect) and there is little consistency regarding the items employed. To extensively summarize all evidence, we chose to focus on AWB as a superordinate term of affective constructs18,21,40. Further, the above-discussed systematic reviews were based on vote counting, that is, the sorting of studies according to their positive, negative or lack of statistical significance. This provides essential summaries of existing evidence. However, vote counting cannot quantify the existence, strength, directionality and circumstances of PA–AWB associations. Further, given that all studies are weighted equally, conclusions based on the descriptive method of vote counting are limited. Outweighing a significant effect in a study with a large sample size by a non-significant effect in a study with a small sample size is generally inappropriate41,42. In the PA–AWB field, matters are further complicated by the variety of methods and statistical models, which results in heterogeneous effect measures and prevents the extraction of effect sizes for traditional meta-analytic approaches. The gold-standard solution is individual participant data (IPD) meta-analysis43,44, in which data from publications are collected and uniformly analysed so that original statistical modelling does not affect results. In addition, IPD compiles all data necessary to model within-person processes across time. Here, we employ this method to investigate how PA covaries with AWB in humans’ everyday life.
For the IPD meta-analysis on PA–AWB, we collected 67 datasets from 14 countries worldwide. These datasets included data on 8,223 participants and 321,345 e-diary AWB ratings associated to nearly 1 million hours of PA measurement via accelerometry. The preregistered primary aims of this IPD meta-analysis were to (1) investigate the existence of PA–AWB associations for specific AWB concepts (positive affective states, negative affective states, valence, energetic arousal and calmness), (2) quantitatively assess the strength of PA–AWB associations and (3) compare effect sizes of PA–AWB associations in both directions, that is, AWB and preceding PA, referred to as the antecedent model, as well as AWB and subsequent PA, referred to as the consequent model. Further, we explored the heterogeneity of PA–AWB associations and investigated potential moderators (for example, age and weekday versus weekend).
Results
Studies and samples descriptives
Overall, we combined 67 datasets including 8,223 participants (Fig. 1). Studies varied in observation duration (M = 17.72, median of 14, range 1–100 days), PA aggregation intervals (M = 159.44, median of 15, range 5–1,440 min) and AWB assessment frequency (M = 7.30, median of 5.50, range 1–25 prompts per day). See Supplementary Information Section 1 for details, including AWB and PA measurement as well as e-diary sampling procedures. See Table 1 for descriptives of participants.
Illustration of the search and retrieval process for the IPD meta-analysis following the PRISMA guidelines. Sed.-triggered, e-diary prompts triggered when participants were sedentary; krecords, numbers of records identified by database searching; kother, additional datasets that were not identified by database searching but by contacting the identified authors in the field; k, number of unique datasets when considering that several records relied on the same datasets, and several records combined datasets. See Supplementary Information Section 1 for details.
Effects, relevance and strength, and temporal directionality of PA–AWB associations in everyday life
In our first series of analyses, we tackled the most prominent questions on PA–AWB associations32,34,38,39 including (1) how PA and AWB are associated with each other in everyday life (effects), (2) how practical effect sizes compare with the multitude of other influences in everyday life (relevance and strength) and (3) the temporal directionality of PA–AWB associations (AWB and preceding PA versus AWB and subsequent PA, which we operationalized in antecedent and consequent models). Towards these aims and to gain comprehensive and multifaceted insights on the basis of statistically robust analyses, we applied a three-step procedure. First, we conducted a (multivariate) meta-analysis to estimate an overall effect across AWB concepts and to compare standardized effect sizes derived from multilevel within- and between-person correlations r of different AWB concepts (two-stage IPD; Supplementary Information Section 2, analysis step 1). Second, we built a series of multilevel models to investigate the role of individual participant and measurement characteristics including standardized PA values and distinct AWB concepts, controlled for age and sex/gender (one-stage IPD; Supplementary Information Section 3, analysis step 2). Third, to calculate practical effect sizes, we built another series of multilevel models with the same structure (step 2) but restricted the data to studies containing measures of raw PA data (movement acceleration intensity (MAI), in milli-g) and added PA aggregation interval as a control variable (MAI one-stage IPD; Supplementary Information Section 3, analysis step 3). A comprehensive range of sensitivity and robustness analyses including ordinal multilevel regressions and bootstrapped confidence intervals (CIs) are presented in Supplementary Information Section 4. We adjusted alpha to 0.05/6 = 0.008, as each AWB concept was tested in three analysis steps and two directions. See Fig. 2 for a graphical overview of the findings.
A summary of findings from all three analysis steps. Left: within-person correlations of AWB with prior (black, antecedent model) and subsequent (white, consequent) PA derived from two-stage IPD meta-analysis (Supplementary Information Section 2) are displayed. The dots represent the weighted mean effects across studies for each outcome. The whiskers represent the 99.2% CI. Middle: moderating influences on associations between AWB and PA derived from one-stage IPD (Supplementary Information Sections 3 and 12) models are illustrated. Right: practical effect sizes are depicted for significant associations of AWB and PA, which were derived from MAI one-stage IPD models (Supplementary Information Sections 3 and 6). Practical effect sizes are exemplified for a PA increase in the range of walking at 5.4 km h−1 versus sitting.
PA predicting subsequent AWB: antecedent model
First, we examined how PA is related to subsequent AWB. To test if there is an overarching association between PA and AWB, we computed a meta-analytic model. In this model, all AWB concepts were subsumed in one larger category, and distinct AWB signs were rectified. This analysis revealed significant within- and between-person associations (two-stage IPD; within, t(63) = 9.07, P < 0.001, r = 0.05, 99.2% CI 0.03 to 0.06; between, t(63) = 5.56, P < 0.001, r = 0.08, 99.2% CI 0.04 to 0.12; Fig. 3a,b). This evidences that overall AWB is indeed positively associated with preceding PA. That is, when a participant is more active than their own average, their AWB in the subsequent AWB assessment is higher compared with assessments with lower preceding PA (within-person association), and on average, more active participants exhibit higher AWB compared with less active participants (between-person association).
a,b, Within-person (a) (P = 0.0000000000005) and between-person associations (b) (P = 0.0000006) of PA with all AWB concepts subsumed in one category. A total of kcum = 287 dependent effects from k = 65 independent studies serve as experimental units and are included in the analysis. For the overall effect, effects for negative affective states were inverted, as negative effects reflect positive associations with well-being. Black dots represent the weighted mean effect across studies for each outcome. The whiskers represent the 99.2% CI. The cumulative number of effects from all AWB concepts is reflected by kcum. Significance was assessed on the basis of an alpha threshold of 0.8%. Each dot represents an individual effect measured for that particular outcome and its size reflects the precision and weight in the analysis. Larger dots indicate higher precision as estimated by 1/s.e.m. Significance for the overall moderator effect was assessed using an omnibus F test, and the significance of individual moderator levels was determined via a t-test. All statistical tests were conducted two-sided. c,d, Plots are organized according to a and b but reflect the effects separately for all assessed concepts of AWB within-person (c), for example, positive affective states: P = 0.00000005; valence, P = 0.0000002; energetic arousal, P = 0.000000008; and calmness, P = 0.0005 and between-person (d). The whiskers represent the 99.2% CI. Significance for the overall moderator effect was assessed using an omnibus F-test, and the significance of individual moderator levels was determined via a t-test. All statistical tests were conducted two-sided.
We next investigated associations of PA with distinct AWB concepts. Here, we differentiated between two unipolar AWB concepts (positive affective states and negative affective states) and three bipolar AWB concepts (valence, energetic arousal and calmness). We found evidence for differential associations (two-stage IPD; F(9, 55) = 19.38, P < 0.001). At the within-person level, PA showed the strongest positive associations with energetic arousal, that is, feeling energized and awake45, compared with all other AWB measures (two-stage IPD, t(55) = 6.79, P < 0.001, r = 0.14, 99.2% CI 0.09 to 0.20; Fig. 3c and Supplementary Information Section 2). One-stage IPD analysis robustly supports this finding (t(17.35) = 7.50, P < 0.001, β = 0.14, 99.2% CI 0.09 to 0.19, Fig. 4g and Supplementary Table 5.4). Effects translate to an increase in energetic arousal of 0.62 points on a 1 to 4 scale when a participant was walking (499 milli-g)46,47 versus sitting (20 milli-g (refs. 46,47); MAI one-stage IPD, t(13.83) = 7.32, P < 0.001; Supplementary Table 6.4 and Supplementary Fig. 6). Associations of PA with positive affective states, for example, feeling happy, cheerful or active48, and valence (that is, feeling well and content)45 were smaller but also significantly positive across all analysis steps (positive affective states: two-stage IPD, t(55) = 6.31, P < 0.001, r = 0.09, 99.2% CI 0.02 to 0.15, one-stage IPD, t(35.25) = 7.30, P < 0.001, β = 0.05, 99.2% CI 0.03 to 0.07; Fig. 4a and Supplementary Table 5.1; valence: two-stage IPD, t(55) = 5.90, P < 0.001, one-stage IPD, t(14.16) = 6.14, P < 0.001, β = 0.04, 99.2% CI 0.02 to 0.06; Fig. 4e and Supplementary Table 5.3). Translated to practice, participants’ positive affective states increased by 0.24 (MAI one-stage IPD, t(3.70) = 7.51, P = 0.002; Supplementary Table 6.1 and Supplementary Fig. 6) and their valence increased by 0.16 (MAI one-stage IPD, t(13.88) = 6.84, P < 0.001; Supplementary Table 6.3 and Supplementary Fig. 6) points on a 1–4 scale when they were walking versus sitting. Conversely, we found a significant negative association with PA and subsequent calmness, that is, feeling relaxed and calm45, across all analysis steps (two-stage IPD, t(55) = 3.70, P < 0.001, r = −0.06, 99.2% CI −0.10 to −0.01; one-stage IPD, t(13.26) = −3.95, P = 0.002, β = −0.05, 99.2% CI −0.09 to −0.02; Fig. 4i and Supplementary Table 5.5). The typical participant’s calmness decreased by −0.22 points on a 1–4 scale when they were walking versus sitting in their everyday life (MAI one-stage IPD, t(9.36) = −4.56, P = 0.001; Supplementary Table 6.5 and Supplementary Fig. 6). Negative affective states, that is, feeling sad, anxious or angry48, did not reveal a robust significant association with preceding PA on the within-person level (two-stage IPD, t(55) = 1.09, P = 0.280, r = −0.01, 99.2% CI −0.03 to 0.01; Fig. 3c; one-stage IPD, t(22.23) = −1.93, P = 0.067, β = −0.01, 99.2% CI −0.02 to 0.00; Fig. 4c and Supplementary Table 5.2; MAI one-stage IPD, t(2.85) = −2.04, P = 0.140; Supplementary Table 6.2).
a,c,e,g,i, The within-person associations of PA and AWB for positive affective states (a) (P = 0.00000002, i = 225,932), negative affective states (c) (i = 235,537), valence (e) (P = 0.0000246, i = 53,274), energetic arousal (g) (P = 0.0000008, i = 50,105) and calmness (i) (i = 48,399). The individual regression lines (thin lines) of participants are shown in red (negative slope) and blue (positive slope). The bold line and associated interval represent the conditional fixed effect of person-mean-centred PA with a 99.2% CI. b,d,f,h,j, The between-person associations of PA and AWB for positive affective states (b) (P = 0.000000003, n = 6,159), negative affective states (d) (n = 6,429), valence (f) (n = 1,947), energetic arousal (h) (P = 0.0001, n = 1,751) and calmness (j) (n = 1,641). Each dot represents one participant’s average PA (x axis) and average AWB (y axis). The bold line and associated interval represent the conditional fixed effect of person-mean PA with a 99.2% CI. AWB was regressed on person-mean-centred and person-mean PA. Person-mean-centred PA was nested in individuals, which were nested in studies. Multilevel models were controlled for gender/sex and age, and regression coefficients were standardized. Significance was derived using two-sided t-tests of regression coefficients, which were adjusted for multiple comparisons (P < 0.008). n indicates the number of participants and i the number of ratings.
For between-person associations, we found significant associations between PA and positive affective states (two-stage IPD, t(55) = 3.46, P = 0.001, r = 0.09, 99.2% CI 0.02 to 0.15; Fig. 3d; one-stage IPD, t(4,304.66) = 5.95, P < 0.001, β = 0.09, 99.2% CI 0.05 to 0.13; Fig. 4b and Supplementary Table 5.1; MAI one-stage IPD, t(496.20) = 3.78, P < 0.001; Supplementary Table 6.1) as well as valence (two-stage IPD, t(55) = 2.80, P = 0.007, r = 0.09, 99.2% CI 0.00 to 0.19; Fig. 3d; one-stage IPD, t(1,647.42) = 3.24, P = 0.001, β = 0.05, 99.2% CI 0.01 to 0.09; Fig. 4f and Supplementary Table 5.3; except for MAI one-stage IPD, t(1,372.52) = 2.01, P = 0.045; Supplementary Table 6.3). Hence, participants who engaged in more PA in general appeared to also experience higher positive affective states and valence than the average less active participant. Evidence for a between-person association of preceding PA and energetic arousal was mixed across analysis steps (two-stage IPD, t(55) = 2.22, P = 0.031, r = 0.10, 99.2% CI −0.02 to 0.22; Fig. 3d; one-stage IPD, t(350.13) = 3.87, P < 0.001, β = 0.06, 99.2% CI 0.02 to 0.09; Fig. 4h and Supplementary Table 5.4; MAI one-stage IPD, t(341.36) = 2.33, P = 0.020; Supplementary Table 6.4).
AWB and subsequent PA: consequent model
Next, we investigated associations of momentary AWB with subsequent PA in everyday life. Our two-stage IPD across all AWB concepts revealed significant within- and between-person associations (within, t(35) = 7.43, P < 0.001, r = 0.04, 99.2% CI 0.03 to 0.05; between, t(35) = 4.01, P < 0.001, r = 0.08, 99.2% CI 0.04 to 0.13; Fig. 5a,b). That is, when a participant exhibited higher AWB than their own average, their subsequent PA was higher compared with assessments with lower prior AWB (within-person association), and on average, participants with higher AWB showed higher overall levels of PA compared with participants with lower AWB (between-person association).
a,b, Within-person (a) (P = 0.0003) and (b) between-person associations (P = 0.00000001) of PA with all AWB concepts subsumed in one category. A total of kcum = 164 dependent effects from k = 37 independent studies serve as experimental units and are included in the analysis. For the overall effect, effects for negative affective states were inverted, as negative effects reflect positive associations with well-being. The black dots represent the weighted mean effect across studies for each outcome. The whiskers represent the 99.2% CI. The cumulative number of effects from all AWB concepts is reflected by kcum. Significance was assessed on the basis of an alpha threshold of 0.8%. Each dot represents an individual effect measured for that particular outcome and its size reflects the precision and weight in the analysis. Larger dots indicate higher precision as estimated by 1/s.e.m. Significance for the overall moderator effect was assessed using an omnibus F-test, and the significance of individual moderator levels was determined via a t-test. All statistical tests were conducted two-sided. c,d, Plots are organized according to a and b but reflect the effects separately for all assessed concepts of AWB within-person (c), for example, positive affective states, P = 0.0006; valence, P = 0.0005; energetic arousal, P = 0.000000006; and calmness, P = 0.00001; and between-person (d), for example, positive affective states: P = 0.00009. The whiskers represent the 99.2% CI. Significance for the overall moderator effect was assessed using an omnibus F test, and the significance of individual moderator levels was determined via a t-test. All statistical tests were two sided.
The AWB concepts showed distinct associations with subsequent PA (two-stage IPD, F(9, 27) = 29.71, P < 0.001). At the within-person level, energetic arousal showed stronger positive associations with subsequent PA compared with all other measures of AWB (two-stage IPD, t(27) = 8.35, P < 0.001, r = 0.12, 99.2% CI 0.08 to 0.16; Fig. 5c and Supplementary Information Section 2). Strong associations of energetic arousal and subsequent PA were fully supported by one-stage IPD analyses (t(7.24) = 8.47, P < 0.001, β = 0.12, 99.2% CI 0.08 to 0.16; Fig. 6g and Supplementary Table 7.4; MAI one-stage IPD, t(614.55) = 21.10, P < 0.001; Supplementary Table 8.4). Moreover, positive affective states (two-stage IPD, t(27) = 3.89, P < 0.001, r = 0.05, 99.2% CI 0.01 to 0.09; Fig. 5c; one-stage IPD, t(18.26) = 4.87, P < 0.001, β = 0.04, 99.2% CI 0.02 to 0.06; Fig. 6a and Supplementary Table 7.1; except for MAI one-stage IPD, t(2.87) = 5.63, P = 0.013; Supplementary Table 8.1) and valence (two-stage IPD, t(27) = 3.98, P < 0.001, r = 0.03, 99.2% CI 0.01 to 0.05; Fig. 5c; one-stage IPD, t(8.74) = 4.20, P = 0.002, β = 0.03, 99.2% CI 0.01 to 0.05; Fig. 6e and Supplementary Table 7.3; except for MAI one-stage IPD, t(6.03) = 3.38, P = 0.015; Supplementary Table 8.3) demonstrated significant positive and calmness (two-stage IPD, t(27) = 5.30, P < 0.001, r = −0.06, 99.2% CI −0.10 to −0.03; Fig. 5c; one-stage IPD, t(5.96) = −5.71, P = 0.001, β = −0.06, 99.2% CI −0.09 to −0.03; Fig. 6i and Supplementary Table 7.5; MAI one-stage IPD, t(616.34) = −10.48, P < 0.001; Supplementary Table 8.5) significant negative associations with subsequent PA. Negative affective states showed no robust significant association with subsequent PA in any analysis step (two-stage IPD, t(27) = 1.94, P = 0.063, r = −0.02, 99.2% CI −0.04 to 0.01; Fig. 5c; one-stage IPD, t(12.42) = −2.13, P = 0.053, β = −0.01, 99.2% CI −0.02 to 0.00; Fig. 6c and Supplementary Table 7.2; MAI one-stage-IPD: t(257.20) = −2.48, P = 0.014; Supplementary Table 8.2).
a,c,e,g,i, The within-person associations of AWB and PA for positive affective states (a) (P = 0.0001, i = 140,170), negative affective states (c) (i = 150,118), valence (e) (i = 44,251), energetic arousal (g) (P = 0.00005, i = 43,270) and calmness (i) (i = 41,561). The individual regression lines (thin lines) of participants are shown in red (negative slope) and blue (positive slope). The bold line and associated interval represent the conditional fixed effect of person-mean-centred AWB with a 99.2% CI. b,d,f,h,j, The between-person associations of AWB and PA for positive affective states (b) (P = 0.0000000000007, n = 3,864), negative affective states (d) (P = 0.00006, n = 4,179), valence (f) (n = 1,354), energetic arousal (h) (n = 1,271) and calmness (j) (n = 1,161). Each dot represents one participant’s average AWB (x axis) and predicted average PA (y axis). The bold line and associated interval represent the conditional fixed effect of person-mean AWB with a 99.2% CI. Of note, in k = 5 studies included, the intercept of the outcome is systematically larger owing to the limited variability or the higher average of the outcome, leading to larger values on the standardized PA score. This can be seen in all panels where several lines/dots are clustered above the fitted lines. Removing these data points (see Supplementary Information Section 4.4 for sensitivity analysis) did not alter the conclusions drawn, suggesting that these between-study differences in mean outcome levels did not bias our results. PA was square-root transformed and regressed on person-mean-centred and person-mean AWB. Person-mean-centred AWB was nested in individuals, which were nested in studies. Multilevel models were controlled for gender/sex and age, and regression coefficients were standardized. Significance was derived using two-sided t-tests of regression coefficients, which were adjusted for multiple comparisons (P < 0.008). n indicates the number of participants and i the number of ratings. For visualization, predictions were back-transformed to the original response scale (instead of square-root transformed scale). Standard errors remain on the square-root-transformed scale.
At the between-person level, only positive affective states consistently showed a positive association with subsequent PA across all analysis steps (two-stage IPD, t(27) = 4.60, P < 0.001, r = 0.13, 99.2% CI 0.05 to 0.21; Fig. 5d; one-stage IPD, t(3,697.86) = 7.20, P < 0.001, β = 0.03, 99.2% CI 0.02 to 0.05; Fig. 6b and Supplementary Table 7.1; MAI one-stage IPD, t(619.35) = 3.27, P = 0.001; Supplementary Table 8.1). Whether participants with on average lower negative affective states (two-stage IPD, t(27) = 2.68, P = 0.012, r = −0.09, 99.2% CI −0.18 to 0.01; Fig. 5d; one-stage IPD, t(3,861.76) = −4.04, P < 0.001, β = −0.02, 99.2% CI −0.03 to −0.01; Fig. 6d and Supplementary Table 7.2; one-stage MAI IPD: t(604.85) = −1.17, P = 0.242; Supplementary Table 8.2) or higher valence (two-stage IPD, t(27) = 1.02, P = 0.319, r = 0.05, 99.2% CI −0.09 to 0.18; Fig. 5d; one-stage IPD, t(1,188.32) = 3.21, P = 0.001, β = 0.03, 99.2% CI 0.01 to 0.06; Fig. 6f and Supplementary Table 7.3; one-stage MAI IPD: t(794.75) = 2.12, P = 0.035; Supplementary Table 8.3) also exhibited higher PA was mixed across analysis steps and robustness analyses; therefore, we assume no association.
Drawing on a conceptual framework from affective psychology49, we conducted sensitivity analyses for AWB operationalizations. For bipolar conceptualizations of AWB, we jointly analysed studies (kant = 17, kcon = 8) assessing valence, energetic arousal and calmness with the psychometrically evaluated short-version of the Multidimensional Mood Questionnaire45. The short-version of the Multidimensional Mood Questionnaire has been referenced as a measure of core affect50,51,52,53,54. For unipolar conceptualizations of AWB, there is no similarly established and psychometrically evaluated measure in daily life research and the collected data. We concentrated our analysis on all studies that followed the Positive and Negative Affect Schedule (PANAS)48 and summarized high-activation items in positive and negative affect (kant = 10, kcon = 4). It needs to be kept in mind that this operationalization is not undisputed (see, for example, refs. 55,56 for a critical discussion of the PANAS). The sensitivity analyses support our main findings. See Supplementary Information Sections 4.5, 4.6 and 9 for a more detailed description of the theoretical considerations and the resulting analyses.
Effect directions
In summary, this IPD meta-analysis on PA–AWB associations in everyday life consistently shows PA to be associated with subsequent valence, calmness, energetic arousal and positive affective states but not with subsequent negative affective states on a within-person level. For the reverse time-order, we consistently found within-person associations of positive affective states, valence, energetic arousal and calmness with subsequent PA but not for negative affective states. Between participants, PA was robustly associated with subsequent positive affective states and valence but with none of the other AWB concepts (calmness, energetic arousal and negative affective states). For between-associations of AWB and subsequent PA, we found only positive affective states to be consistently associated with subsequent PA.
Relevance and strength
When translating these effect sizes into practice to estimate their relevance in humans’ everyday life, their magnitude is comparable to other daily life activities. Changing one’s activity in the range of, for example, from sitting to walking is associated with an increase in AWB from 0.16 points (scale 1–4) for valence (the weakest yet significant effect) to 0.62 points (scale 1–4) for energetic arousal (the strongest effect). Comparing PA–AWB effects to an existing and prominent large-scale daily life study57 makes their relevance and strength tangible: overall, our data across five AWB concepts are in line with the Killingsworth and Gilbert study on happiness during self-reported activities (17,775 ratings)57, in which taking a walk was related to an increase in happiness of approximately 0.21 points compared with a person’s average happiness and exercising to an increase of approximately 0.36 points when transferring effects to a scale from 1–4 (Supplementary Fig. 6). In the Killingsworth and Gilbert study, other daily life activities such as listening to music, playing and talking exhibited effects of similar size; watching television, reading, shopping and relaxing were related to a smaller increase in happiness of approximately 0.06 points.
Temporal directionality
We compiled evidence on the temporal directionality of PA–AWB versus AWB–PA associations. Our analyses revealed significant within-person associations of both preceding and subsequent PA with positive affective states, valence, energetic arousal and calmness but no within-person associations with negative affective states. To statistically test for temporal directionality, we conducted a two-stage IPD meta-analysis considering both temporal directions in one model (Supplementary Information Section 10). Post-hoc comparisons revealed no significant differences in effects between both time-orders for all PA–AWB associations. Hence, there was no evidence that the PA–AWB association is larger in one or the other temporal direction.
Heterogeneity of PA–AWB associations
Variation in the direction and strength of PA–AWB associations between individuals has not been representatively studied to date but promises to come with high relevance for precision medicine and therefore remains under intensive discussion58,59,60,61,62. Employing the full strength of the IPD meta-analysis, we here found considerable differences in the direction and strength of PA–AWB associations between individuals (Figs. 4a,c,e,g,i and 6a,c,e,g,i). This offers a promising treatment target for individualized and thus expedient interventions. In particular, the association between PA and subsequent negative affective states appears the most heterogeneous (40.22% positive and 59.78% negative) across participants, whereas we found the most homogeneous effects for PA and energetic arousal. Nearly all participants (>95%) felt more energetically aroused when PA prior or subsequent to the e-diary prompt was higher.
Moderators of PA–AWB associations
The evidence on the heterogeneity of PA–AWB associations motivated us to inspect potential correlates of person and measurement characteristics with the within-person associations of AWB and PA (for moderator analyses for study characteristics see Supplementary Information Section 11). Although the range of possible moderators appears immense, we focused our analyses on four epidemiologically potent moderators63,64: age (person-level), gender/sex (person-level), body mass index (BMI; person-level) and weekday (relative to weekend; measurement-level). Moderating associations emerged, which differed depending on the AWB concept (only significant moderation analyses are presented; see Supplementary Information Section 12 for a comprehensive report of moderator analyses).
Age
Younger individuals engaged in more PA before (t(720.63) = −3.74, P < 0.001, β = −0.01, 99.2% CI −0.02 to −0.00; Fig. 7a and Supplementary Table 12.1a) and after (t(179.58) = −3.01, P = 0.003, β = −0.02, 99.2% CI −0.04 to −0.00; Fig. 7b and Supplementary Table 12.2a) reporting high positive affective states compared with older individuals, for whom the bidirectional association of positive affective states and PA was less pronounced.
a,b, The moderating influence of age on the association of positive affective states with PA. Younger individuals engaged in more PA before (a) and after (b) reporting high positive affective states compared with older individuals, for whom the bidirectional association of positive affective states and PA was less pronounced. c,d, The moderating influence of sex/gender on the association of energetic arousal and calmness with PA; women felt more energetically aroused after being physically active than men (c); men were more physically active than women when feeling less calm than usual; however, they were less physically active than women when reporting more calmness than usual (d). e, The moderating influence of BMI on the association of valence with preceding PA. Individuals with a lower BMI reported higher valence after being more physically active than usual compared with individuals with a higher BMI. Given that height and weight-based BMI values are, in general, inappropriate for underaged individuals, this moderator analysis was conducted for adults only. f,g, The moderating influence of weekday versus weekend on the association of energetic arousal and PA. On weekends, the bidirectional association of PA and energetic arousal was stronger than on weekdays.
Gender/sex
Women felt more energetically aroused after being physically active than men (t(808.61) = 3.18, P = 0.002, β = 0.04, 99.2% CI 0.01 to 0.07; Fig. 7c and Supplementary Table 12.3d). Men were more physically active than women when feeling less calm than usual. However, they were less physically active than women when reporting more calmness than usual (t(630.44) = 3.52, P < 0.001, β = 0.10, 99.2% CI 0.02 to 0.18; Fig. 7d and Supplementary Table 12.4e).
BMI
Adults with a lower BMI reported higher valence after being more physically active than usual compared with adults with a higher BMI (t(620.19) = −2.87, P = 0.004, β = −0.02, 99.2% CI −0.03 to −0.00; Fig. 7e and Supplementary Table 12.5c).
Weekday versus weekend
On weekends, the bidirectional association of PA and energetic arousal was stronger than on weekdays (antecedent model, t(17,918.03) = −3.36, P < 0.001, β = −0.02, 99.2% CI −0.03 to −0.00; Fig. 7f and Supplementary Table 12.7d; consequent model, t(35,430.89) = −4.04, P < 0.001, β = −0.03, 99.2% CI −0.05 to −0.01; Fig. 7g and Supplementary Table 12.8d).
Finally, given the urgent need to further research heterogeneity in PA–AWB associations between participants towards precision medicine approaches, we explored potential low-base effects. We plotted participants’ individual slopes and intercepts against each other and explored the random effects correlations within the one-stage IPD multilevel models (Supplementary Information Section 13). The data suggest that individuals exhibiting higher average negative affective states show stronger reductions in their momentary negative affective states after engaging in more PA than usual compared with individuals with low negative affective states (r = −0.13, 99.2% CI −0.23 to −0.04). For positive affective states, the data suggest that individuals exhibiting lower average positive affective states show stronger increases in their momentary positive affective states after engaging in more PA than usual compared with individuals with high positive affective states (r = −0.09, 99.2% CI −0.17 to −0.01).
Discussion
This preregistered IPD meta-analysis provides evidence on directions and strength, relevance, temporal directionality, heterogeneity and moderators of PA–AWB relations in humans’ everyday life. Associations vary depending on AWB concepts and level of analysis: Within-persons, PA is consistently and positively associated with positive affective states, valence and energetic arousal. Yet, PA is negatively associated with calmness and, overall, not significantly linked to negative affective states. Between-persons, PA only robustly associates with positive affective states. The practical effect sizes are comparable to other daily life activities, with energetic arousal evincing the strongest within-person association with PA. Associations substantially vary in direction and strength between individuals, but age, gender/sex, BMI and weekday/weekend partially explain PA–AWB association differences between persons.
Overall, the data show that PA is robustly associated with AWB as humans go about their daily routines. The evidence on pronounced within-person PA–AWB relations substantiates the general sentiment of cutting-edge health behaviour theory development. Traditional health behaviour theories emphasizing the rational and cognitive determination of health behaviour65,66,67,68,69 show limited explanatory power11,14,17,70. To tackle the vital issue of health behaviour decision-making and public health, novel models of health behaviour build on hedonic theories of behaviour11,12,13,14,71,72,73,74,75,76 and take into account that affective, impulsive processes shape health behaviours, including PA engagement. These affective, impulsive processes contrast against rational, reflective processes and regained attention subsequent to Kahnemann’s ‘thinking slow, thinking fast’ proposal77. Put simply, novel health behaviour models emphasize that it is not only the knowledge on positive health effects that drives PA engagement but actual experiences and affective feelings related to PA.
This large-scale IPD provides empirical evidence to understand the micro-temporal processes of PA–AWB relations. This is crucial for future research on precision medicine approaches such as personalized micro-interventions in everyday life based on real-time data (so-called ecological momentary interventions or just-in-time adaptive interventions)78. Ultimately, the micro-temporal processes promise to allow for individually and context-tailored real-time support to prevent and intervene on a broad range of non-communicable diseases and thereby contribute to public health. As variations in AWB are not only thought to be linked to PA engagement but also to human experiences and behaviours in general, it is tempting to speculate that PA may indirectly also influence other aspects of daily life, such as attention, memory, decision-making or learning, via fluctuations in AWB. General models of decision-making, for example, postulate that incidental factors, such as lingering moods and emotions, can impact decisions79,80. Hence, PA-related changes in AWB unrelated to the object of decisions may nevertheless subtly affect how individuals make decisions in their daily lives79. However, these implications need to be considered against the background of the observational nature of the data meta-analysed and the resulting potential for residual confounding. Although this meta-analysis on intensive longitudinal data of daily life studies on PA and AWB comes with highest ecological validity, the designs of PA–AWB studies included are observational. That is, they do not include experimental manipulation to control the impact of confounding variables (that is, variables impacting the PA–AWB associations). As a result, unmeasured (time-varying) confounding variables (for example, weather conditions) could have affected the findings. Ideally, future Ecological Momentary Assessment (EMA) studies consider the impact of these variables more thoroughly. For studies focusing on the PA–AWB association, a quasi-experimental design could be adopted81,82. For studies examining the AWB–PA association, potential confounding variables should be systematically identified and accounted for in the EMA design and the analyses83,84.
The data reveal an especially pronounced covariation of PA with feelings of energy across time, with PA–energetic arousal associations yielding considerably higher effect sizes than PA–calmness and PA–valence associations. The PA–positive affective states association strength positions itself in between. Accordingly, PA–energetic arousal and PA–positive affective states relations present themselves as particularly salient targets to shape theories of health behaviour change and develop future precision medicine approaches. For example, PA–energy associations can be especially valuable in the prevention and intervention of fatigue symptomatology85,86,87 and in the treatment of affective disorders8,36,88. Overall, the practical effect sizes of PA–AWB relations derived from our IPD meta-analysis are comparable to other daily life activities (for example, reading, listening to music and eating) that were investigated in a large-scale study by Killingsworth and Gilbert57. For feelings of energy, they appear considerable. It has to be taken into account, however, that the Killingsworth and Gilbert57 study investigated happiness, whereas we present different concepts of AWB.
Our IPD meta-analytic findings can guide future research towards individualized, tailored and expedient interventions utilizing PA–AWB relations. First, given its considerable effect size, interactions of PA with energetic arousal appear as important future research targets. Considering the overall differential AWB associations documented, it can critically contribute to existing health behaviour theories and even bring up novel health behaviour models. Of note, our meta-analysis revealed a pronounced variability between studies in the instruments applied for the assessment of AWB, as well as a limited coverage of certain AWB dimensions. This appears as a challenge to daily life studies. We conducted sensitivity analyses revealing that our main analyses are robust against assignments of AWB constructs to positive affective states, negative affective states, valence, energetic arousal and calmness (Supplementary Information Sections 4.5, 4.6 and 9). We recommend future studies to provide clear conceptual rationales and rely on psychometrically evaluated measures when conducting a study on PA–AWB associations in daily life, for example, guided by a framework from affective psychology49. The development of formalized and testable psychological theories89,90,91 as well as item pools (for example, the Experience Sampling Method Item Repository92) for daily life research can further contribute to greater consensus. Second, evidence is in favour of temporally bidirectional PA–AWB associations. However, aggregated data are not suited to evidence causal conclusions. Dynamic modelling approaches such as continuous-time analyses93,94 enriched by experimental manipulation in everyday life are essential to approach a future PA–AWB Frequency, Intensity, Time and Type framework95.
Third, our data clearly show PA–AWB relations to vary in direction and strength between individuals and accumulate evidence for sex/gender-, BMI- and weekday-dependent relationships. Although we received no clear evidence for uniform moderation patterns across AWB concepts, distinct interactions emerged. Younger and low-BMI individuals exhibit stronger associations of PA with subsequent valence or positive affective states. In light of motivational and particularly hedonic theories, which postulate that individuals strive to maximize their well-being, the less positive affective response to PA in high-BMI and older individuals plausibly fits epidemiological observations of reduced PA in these groups2,96,97,98. High-BMI and older adults may have weaker associations between PA and subsequent positive affective states and valence because PA may trigger more feelings of discomfort, pain, heat, stain or other unpleasant physiological states in these groups. To enhance PA engagement, it thus appears especially important to identify putative catalysts of the PA–AWB associations in vulnerable groups99: Sociodemographic, biological and contextual factors may facilitate effects of PA on pleasure to reinforce future PA engagement and lower the threshold for AWB to evoke PA. This may comprise aspects such as walkability100,101, green spaces102,103,104 or the social context105,106,107. Given neurobiological108 and laboratory109,110,111 findings of blunted PA–AWB responses in high-BMI individuals, it is also tempting to speculate that the intensity of PA differentially affects neurobiological processes and, thus, constitutes a candidate for precision intervention111,112. Next, for men, calmness is more strongly associated with subsequent PA than for women, yet women feel more energetically aroused after being physically active than men. Although the assessment of gender/sex differed between studies and only few studies included a non-binary gender option, this differential pattern of PA–AWB associations may relate to distinct autonomic nervous system properties seen in people with female versus male biology, such as differences in hormones113 or heart rate variability114,115. If substantiated in future proposals, this suggests that biologically informed prevention and treatment approaches may be particularly salient for PA engagement and AWB. Finally, on weekends, the bidirectional association of PA and energetic arousal is stronger than on weekdays, which partially matches findings on differences in work-related versus leisure-time PA impacts on both mental116,117 and physical118 health. Particularly, studies have found that although leisure-time PA is consistently linked with reduced cardiovascular risk and depression, occupational PA appears to have negative health effects. Hence, accounting for the day-of-week context appears of particular importance for future precision medicine approaches. Intriguingly, the data also reveal that high-negative affective states individuals can exhibit substantial benefit from PA. Thus, this work sets the basis for differential investigations within specific individuals, subgroups and contexts for leveraging diverse personal or situational characteristics of the ideographic AWB–PA relationship in follow-up randomized clinical trials. Following recent criticism on the exclusive focus on social and contextual determinants to the detriment of biological and genetic determinants of PA engagement119,120, future research can gain extensive knowledge on the idiographic associations between PA and AWB by focusing not only on a wide range of person–situation–environment characteristics but also through studying biological and genetic factors as well.
Several limitations need to be taken into account. First, although we searched the most prominent literature databases and asked all authors for additional unpublished data, we cannot rule out that representative datasets were overlooked by our literature search. However, we received and compiled a total of 67 datasets across 14 countries comprising a highly representative sample. Second, although a major strength of this IPD meta-analysis is the sole inclusion of device-based assessed PA, studies vary in measurement methods. To rule out residual confounding influences, we adjusted to these differences by means of (1) statistical standardization, (2) testing for potential moderation and (3) securing findings by introducing a third analysis step only applying streamlined raw PA data. Findings remained robust across all analysis steps and regardless of PA assessment procedures. Third, studies varied in the AWB concepts investigated and, within concepts, the specific items employed. We streamlined our analyses across unipolar (positive affective states and negative affective states) and bipolar (valence, calmness and energetic arousal) AWB concepts. Despite the heterogeneity of AWB assessments, our meta-analysis reveals that PA can explain meaningful variance across AWB concepts.
In conclusion, this work shows the critical relevance of PA–AWB associations within humans and extends the state of knowledge on how PA relates to AWB in everyday life. It supports the general sentiment of ‘affectivism’18 and of motivational theories that highlight the relevance of affective processes. The findings argue for a dominant role of momentary AWB fluctuations within humans to shape behaviour and experiences. The heterogeneity of PA–AWB associations underlines the importance of considering contextual factors, especially in vulnerable groups. Considering aspects of the social and physical environment of individuals, such as walkability or urban green spaces, appears of utmost importance for human decision-making and public health strategies. The collected data provide an evidence base for precision medicine interventions that may ultimately alleviate the inactivity pandemic and enhance human health around the globe.
Methods
This IPD meta-analysis followed established procedures (PRISMA-IPD; Supplementary Information Section 14) and was preregistered on 21 March 2022 (CRD42022303509; https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022303509). The present IPD meta-analysis complies with all relevant ethical regulations. An ethics exemption was obtained from the ethics committee at the Faculty of Sport Science, Ruhr University Bochum, Germany, based on the ethics approvals of individual studies included in the meta-analysis (reference: EKS S 03/2022).
All preprocessing, harmonization, checking, merging and analysing of data was performed using R (version 4.4.0 and 4.5.2121), the interface RStudio122 and various R packages123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140.
Data sources and study selection
Records were considered for inclusion if (1) PA was quantified using technical devices (for example, accelerometers) to reduce distortions (for example, cognitive heuristics35,141,142); (2) AWB was self-reported (for example, electronic diaries on smartphones) to comply with state-of-the-art procedures for reliable and ecologically valid assessments of psychological states142; (3) assessments were conducted for at least 1 day to minimize (diurnal) confounds143; and (4) aggregated time frames did not exceed 24 h to reduce well-known recall-bias142. Records studying people of all ages with and without mental or physical diseases were included to provide a summary of PA–AWB associations across populations.
We excluded records if (1) PA measurements were conducted under controlled or artificial conditions (for example, laboratory or research settings, interventions and in-patient treatments); (2) records employed retrospective questionnaires on PA and AWB; and (3) data collection occurred one time only.
We searched the databases Web of Science, PubMed, Scopus, SPORTDiscus and PsycINFO until December 2023 using the search terms ‘ecological momentary assessment’, ‘physical activity’ and ‘sedentary behavior’ as well as their synonyms (see Supplementary Information Section 15 for the adapted Boolean operators) and search terms for the umbrella term AWB; these comprise ‘mood’ or ‘emotion’ or ‘affect’. The search scope was confined to articles published in English, without restrictions on publication year. Records were first selected on the basis of a screening of titles. Abstracts and full-texts were independently reviewed for eligibility (M.G., I.T.; in case of non-correspondence: M.R. as third reviewer; Fig. 1).
Data synthesis
Corresponding authors of selected studies were contacted to retrieve IPD and also asked for additional unpublished IPD meeting the inclusion criteria. The deadline for all data transfers was set to 31 October 2024. During data collection, two retrieved datasets were excluded. In one dataset, PA and AWB were only available for exercise bouts, not for time slots preceding or following the e-diary prompt144. The dataset of another study only included one aggregate measure of PA per participant so that the within-associations of PA and AWB could not be investigated145. Variables to be harmonized were chosen upon intensive discussion in the IPD-coordinating team (J.R., I.T., G.B., M.G., J.P. and M.R.).
Outcomes
In the antecedent models, AWB served as the primary outcome. There is an ongoing debate on how to conceptualize, define and operationalize AWB18,55,76,146,147,148,149,150,151,152. A variety of terms such as core affect, feelings, emotions and moods, exists, which are, to date, not consistently used (see18 and Supplementary Information Section 9 for an extensive discussion). Prominent models of AWB include Russell’s153 circumplex model, Watson et al.’s two-dimensional approach operationalized within the PANAS48 and Thayer’s concept of activation comprising tension and wakefulness146,154,155. Recently, three-dimensional theories of AWB, which include the dimensions of valence, calmness and energetic arousal45,156, have gained increasing interest. To extensively summarize the heterogeneous literature on PA and AWB, we focused on five constructs from two- and three-dimensional models of AWB that were frequently assessed in real-life studies on AWB–PA associations34: positive affective states, negative affective states, valence, energetic arousal and calmness. We differentiated positive and negative affective states (as unipolar AWB concepts) from bipolar AWB concepts (that is, valence, energetic arousal and calmness). If studies included none of the predefined AWB dimensions, the assessed AWB measures were assigned to positive affective states, negative affective states, valence, energetic arousal and calmness by five researchers independently (M.R., M.G., G.B., I.T. and J.R.); a unipolar item of ‘fatigue’ was, for example, assigned to negative affective states, a unipolar item of ‘pleasure’ (that is, ‘not at all pleasurable’ to ‘totally pleasurable’) to positive affective states. ‘Mood’ was depending on whether it was assessed unipolarly (for example, DABS Mood157) or bipolarly (that is, ‘very unpleasant’ to ‘very pleasant’, ‘sad’ to ‘happy’ and ‘very negative’ to ‘very positive’) assigned to positive affective states and negative affective states (unipolar) or to valence (bipolar).
In the consequent models, PA was the primary outcome. If multiple PA measures were available, we followed a predefined procedure. To support the investigation of momentary PA–AWB associations, we preferred PA aggregated over the smallest available time interval. If multiple PA operationalizations were included in a dataset, we preferred (1) measures comprising the full PA intensity spectrum (MAI, counts, steps, MET > moderate-to-vigorous PA (MVPA), light PA (LIPA) and time stepping), (2) streamlined measures involving less preprocessing (MAI > counts, steps) and (3) measures capturing more frequent activities (LIPA > time stepping > MVPA). The full hierarchy is: MAI > counts > steps > MET > LIPA > time stepping > MVPA.
All studies assessed AWB at multiple discrete time points. PA was measured continuously in all studies. To analyse the relation of AWB and PA, researchers aggregated PA in certain time intervals, for example, 60 min preceding or following the AWB rating. Thus, antecedent models incorporated PA aggregated before the AWB rating. Consequent models incorporated PA aggregated following the AWB rating. See Extended Data Fig. 1 for a rough visualization of the study design across studies.
To increase robustness of our findings, we included three main effect measures: First, we chose a two-stage IPD approach, in which multilevel correlations served as effect measures. In the second and third analysis step, we chose a one-stage IPD approach and set up separate multilevel models for each AWB dimension. Here, beta coefficients served as effect measures (see Figs. 4 and 6 for standardized and Supplementary Information Sections 5–8 for unstandardized effects).
Analysis
We tackled the most prominent questions on PA–AWB associations32,33,34,38,39 on (1) how PA and AWB are associated to each other in everyday life (effects), (2) how practical effect sizes compare with the multitude of other influences in human everyday life (relevance and strength) and (3) the temporal directionality of PA–AWB associations (PA and subsequent AWB versus AWB and subsequent PA). We calculated a two-stage and two one-stage IPD analyses for both antecedent and consequent models: First, we conducted a (multivariate) meta-analysis to estimate an overall effect across AWB dimensions and compared standardized effect sizes derived from multilevel within- and between-person correlations r and their respective 99.2% CIs of different AWB dimensions (two-stage IPD). Second, we built a series of three-level models to investigate the role of individual participant and measurement characteristics including standardized PA values and distinct AWB dimensions, controlling for age and sex/gender (one-stage IPD). Third, to calculate practical effect sizes, we built an additional series of multilevel models with the same structure (step 2) but restricted the data to studies containing measures of raw PA data (MAI, in milli-g) and added PA aggregation interval as a control variable (MAI one-stage IPD). The decision to employ MAI for the deduction of practical effect sizes (instead of, for example, counts or MVPA) was data-driven: we compared PA data distributions and statistics across studies and decided on the most homogeneous PA measure. Finally, we explored variations in effects based on study-level, participant-level and measurement-level characteristics and conducted moderator analyses. A detailed description of all analysis steps can be found in Supplementary Information Sections 2, 3, 11 and 12.
Risk of bias
To assess and account for risk of bias between studies, we calculated small study bias in the two-stage IPD analysis and visually inspected funnel plots. We used the precision-effect test and precision-effect estimate with standard errors to apply a small study bias correction by using either the standard error and the effect size variance as regressors158. Influential studies were quantified using Cook’s D. A threshold of D > 0.5 was used to qualify a study as influential159. Heterogeneity was assessed using Cochran’s Q, which indicates if the extracted effect sizes estimate a common population effect size.
To assess risk of bias in individual studies, we applied the ROBINS-E tool160. Ratings were conducted and discussed by three researchers (I.T., J.R. and M.G.) and are presented in Supplementary Information Section 16. Further, we also employed the modified quality assessment (QA) tool developed by Timm et al.34 as a QA tool specifically developed for ambulatory assessment studies on PA and AWB. The modified QA tool includes categories such as accelerometer technology employed, e-diary sampling schema applied and compliance rates received. It consists of 16 questions (that is, the total score equals 16). For the current meta-analysis, all included records’ quality was assessed independently by two researchers (I.T. and M.G.) and divergent ratings were rated by a third author (M.R.). Ratings can be found in Supplementary Information Section 17.
Additional analyses
In the two-stage IPD analyses, sensitivity analyses were carried out with values of ρ = 0, ρ = 0.25, ρ = 0.75 and ρ = 1.0 (Supplementary Information Section 18).
To ensure robustness of findings against violated assumptions in one-stage IPD analyses, we followed established procedures161,162 and calculated residual/wild bootstrapped CIs using the package lmeresampler131. In addition, as there is an increasing call for ordinal instead of metric models to analyse Likert-scaled data163, for the antecedent hypothesis, we also calculated ordinal multilevel regression models using the package ordinal130. In most cases, findings aligned (Supplementary Information Section 4).
Moreover, to ensure robustness of AWB operationalizations, we conducted sensitivity analyses in the one-stage IPD analysis. These analyses followed the same procedure as our main analysis but were restricted to a subset of studies that followed a narrowly defined assessment of AWB (Supplementary Information Sections 4.5, 4.6 and 9).
Finally, we set up a logistic regression model and predicted data provision by study characteristics to explore systematic differences between studies that provided data and studies that did not provide data for our meta-analysis (Supplementary Information Section 19). In the studies included, PA was more often investigated before AWB than in studies that did not provide data (P = 0.01).
Deviations from preregistration
In contrast to preregistration, the synthesized data did not allow to investigate the preregistered research questions RQ6 (does the context of physical behaviour influence the effect size, for example, gym versus home?) and RQ8 (how long does the effect on well-being last?) as the information was missing in individual datasets. Further, RQ4 (does the type of physical behaviour influence the effect size?) and RQ5 (does the duration of physical behaviour influence the effect size?) could only be approximated by two-stage IPD moderator analyses (that is, moderation of PA type and aggregation interval; Supplementary Information Section 11). In addition, our meta-analysis is based on peer-reviewed records identified by Timm et al.34 and also includes records identified by an updated literature search and unpublished/unrequested data. Data harmonization and merging was not conducted by two researchers separately but by J.R. using a standardized R script. The R script underwent independent code review by J.P. As effect measures in two-stage IPD analyses, we employed multilevel correlations instead of standardized beta coefficients as correlations allow for easier interpretations and are comparable to beta coefficients in case of no included covariates. Standardized betas were used in one-stage IPD analyses. We forwent imputations of missing data due to the lack of consistent predictors of missing data across individual datasets. Finally, influential studies in two-stage IPD analysis were identified using Cook’s D instead of bootstrapping to follow established procedures.
Reporting summary
Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Data availability
For licence and ethical reasons, IPD cannot be made publicly available. Data from the individual studies need to be requested from the respective data contributor. To facilitate data requests, a list of included datasets is publicly available (https://osf.io/2tn8u). Furthermore, we are happy to assist in establishing contact with the data contributors.
Code availability
The analysis code of our analyses is publicly available (https://osf.io/2tn8u).
References
Brauer, M. et al. Global burden and strength of evidence for 88 risk factors in 204 countries and 811 subnational locations, 1990–2021: a systematic analysis for the Global Burden of Disease Study 2021. Lancet 403, 2162–2203 (2024).
Strain, T. et al. National, regional, and global trends in insufficient physical activity among adults from 2000 to 2022: a pooled analysis of 507 population-based surveys with 5.7 million participants. Lancet Glob. Health 12, e1232–e1243 (2024).
Guthold, R., Stevens, G. A., Riley, L. M. & Bull, F. C. Global trends in insufficient physical activity among adolescents: a pooled analysis of 298 population-based surveys with 1.6 million participants. Lancet Child Adolesc. Health 4, 23–35 (2020).
Global Action Plan On Physical Activity 2018–2030: More Active People For A Healthier World (WHO, 2018).
Hafner, M. et al. Estimating the global economic benefits of physically active populations over 30 years (2020-2050). Br. J. Sports Med. 54, 1482–1487 (2020).
Chaabene, H. et al. Home-based exercise programmes improve physical fitness of healthy older adults: A PRISMA-compliant systematic review and meta-analysis with relevance for COVID-19. Ageing Res. Rev. 67, 101265 (2021).
Kivimäki, M. et al. Physical inactivity, cardiometabolic disease, and risk of dementia: an individual-participant meta-analysis. Br. Med. J. 365, l1495 (2019).
Noetel, M. et al. Effect of exercise for depression: systematic review and network meta-analysis of randomised controlled trials. Br. Med. J. 384, q1024 (2024).
Gordon, B. R., McDowell, C. P., Lyons, M. & Herring, M. P. The effects of resistance exercise training on anxiety: a meta-analysis and meta-regression analysis of randomized controlled trials. Sports Med. 47, 2521–2532 (2017).
Dauwan, M., Begemann, M. J. H., Heringa, S. M. & Sommer, I. E. Exercise improves clinical symptoms, quality of life, global functioning, and depression in schizophrenia: a systematic review and meta-analysis. Schizophr. Bull. 42, 588–599 (2016).
Williams, D. M., Rhodes, R. E. & Conner, M. T. Conceptualizing and intervening on affective determinants of health behaviour. Psychol. Health 34, 1267–1281 (2019).
Ekkekakis, P. & Brand, R. Affective responses to and automatic affective valuations of physical activity: fifty years of progress on the seminal question in exercise psychology. Psychol. Sport Exerc. 42, 130–137 (2019).
van Cappellen, P., Rice, E. L., Catalino, L. I. & Fredrickson, B. L. Positive affective processes underlie positive health behavior change. Psychol. Health 33, 77–97 (2018).
Stevens, C. J. et al. Affective determinants of physical activity: a conceptual framework and narrative review. Front. Psychol. 11, 568331 (2020).
Ekkekakis, P. Pleasure and displeasure from the body: perspectives from exercise. Cogn. Emot. 17, 213–239 (2003).
Dohnalová, L. et al. A microbiome-dependent gut–brain pathway regulates motivation for exercise. Nature 612, 739–747 (2022).
Rhodes, R. E., McEwan, D. & Rebar, A. L. Theories of physical activity behaviour change: a history and synthesis of approaches. Psychol. Sport Exerc. 42, 100–109 (2019).
Dukes, D. et al. The rise of affectivism. Nat. Hum. Behav. 5, 816–820 (2021).
Reed, J. & Buck, S. The effect of regular aerobic exercise on positive-activated affect: a meta-analysis. Psychol. Sport Exerc. 10, 581–594 (2009).
Reed, J. & Ones, D. S. The effect of acute aerobic exercise on positive activated affect: a meta-analysis. Psychol. Sport Exerc. 7, 477–514 (2006).
Buecker, S., Simacek, T., Ingwersen, B., Terwiel, S. & Simonsmeier, B. A. Physical activity and subjective well-being in healthy individuals: a meta-analytic review. Health Psychol. Rev. 15, 574–592 (2021).
Trull, T. J. & Ebner-Priemer, U. Ambulatory assessment. Annu. Rev. Clin. Psychol. 9, 151–176 (2013).
Mehl, M., Conner, T. & Csikszentmihalyi, M. Handbook of research methods for studying daily life. Choice Rev. Online https://doi.org/10.5860/choice.50-1159 (2014).
Kim, J., Kikuchi, H. & Yamamoto, Y. Systematic comparison between ecological momentary assessment and day reconstruction method for fatigue and mood states in healthy adults. Br. J. Health Psychol. 18, 155–167 (2013).
Hamaker, E. L. in Handbook of Research Methods for Studying Daily Life (eds Mehl, M. R. & Conner, T. S.) 43–61 (Guilford Press, 2012).
Robinson, W. S. Ecological correlations and the behavior of individuals. Am. Sociol. Rev. 15, 351–357 (1950).
Zawadzki, M. J., Smyth, J. M., Sliwinski, M. J., Ruiz, J. M. & Gerin, W. Revisiting the lack of association between affect and physiology: contrasting between-person and within-person analyses. Health Psychol. 36, 811–818 (2017).
Kamarck, T. W., Schwartz, J. E., Janicki, D. L., Shiffman, S. & Raynor, D. A. Correspondence between laboratory and ambulatory measures of cardiovascular reactivity: a multilevel modeling approach. Psychophysiology 40, 675–683 (2003).
Reichert, M. et al. The association of stress and physical activity: mind the ecological fallacy. Ger. J. Exerc. Sport Res. 52, 282–289 (2022).
Molenaar, P. C. M. A manifesto on psychology as idiographic science: bringing the person back into scientific psychology, this time forever. Meas. Interdiscip. Res. Perspect. 2, 201–218 (2004).
Meyer-Lindenberg, A. The non-ergodic nature of mental health and psychiatric disorders: implications for biomarker and diagnostic research. World Psychiatry 22, 272–274 (2023).
Dunton, G. F. Sustaining health-protective behaviors such as physical activity and healthy eating. J. Am. Med. Assoc. 320, 639–640 (2018).
Dunton, G. F. Ecological momentary assessment in physical activity research. Exerc. Sport Sci. Rev. 45, 48–54 (2017).
Timm, I., Giurgiu, M., Ebner-Priemer, U. W. & Reichert, M. The within-subject association of physical behavior and affective well-being in everyday life: a systematic literature review. Sports Med. 54, 1667–1705 (2024).
Reichert, M. et al. Ambulatory assessment for physical activity research: State of the science, best practices and future directions. Psychol. Sport Exerc. 50, 101742 (2020).
Reichert, M. et al. A neural mechanism for affective well-being: subgenual cingulate cortex mediates real-life effects of nonexercise activity on energy. Sci. Adv. 6, eaaz8934 (2020).
Benedyk, A. et al. Real-life behavioral and neural circuit markers of physical activity as a compensatory mechanism for social isolation. Nat. Mental Health 2, 337–342 (2024).
Liao, Y., Shonkoff, E. T. & Dunton, G. F. The acute relationships between affect, physical feeling states, and physical activity in daily life: a review of current evidence. Front. Psychol. 6, 1975 (2015).
Bourke, M., Hilland, T. A. & Craike, M. A systematic review of the within-person association between physical activity and affect in children’s and adolescents’ daily lives. Psychol. Sport Exerc. 52, 101825 (2021).
Williams, D. M., Rhodes, R. E. & Conner, M. T. (eds) Affective Determinants of Health Behavior (Oxford Univ. Press, 2018).
Bushman, B. J. & Wang, M. C. in The Handbook of Research Synthesis and Meta-Analysis (eds Cooper, H., Hedges, L. V. & Valentine, J. C.) 207–220 (Russell Sage Foundation, 2009).
Friedman, L. Why vote-count reviews don’t count (multiple letters). Biol. Psychiatry 49, 161–162 (2001).
Simmonds, M. C. & Stewart, L. A. in Systematic Reviews in Health Research (eds Egger, M., Higgins, J. P. T. & Davey Smith, G.) 220–237 (Wiley, 2022).
Stewart, L. A. & Parmar, M. K. Meta-analysis of the literature or of individual patient data: is there a difference? Lancet 341, 418–422 (1993).
Wilhelm, P. & Schoebi, D. Assessing mood in daily life: structural validity, sensitivity to change, and reliability of a short-scale to measure three basic dimensions of mood. Eur. J. Psychol. Assess. 23, 258–267 (2007).
Olfermann, R., Ebner-Priemer, U., Reichert, M. & Giurgiu, M. Comparability of accelerometry outcomes across popular metrics and widespread sensor positions. PLoS One 20, e0337897 (2025).
Giurgiu, M. et al. Validating accelerometers for the assessment of body position and sedentary behavior. J. Meas. Phys. Behav. 3, 253–263 (2020).
Watson, D., Clark, L. A. & Tellegen, A. Development and validation of brief measures of positive and negative affect: the PANAS scales. J. Pers. Soc. Psychol. 54, 1063–1070 (1988).
Ekkekakis, P. & Zenko, Z. in Emotion Measurement, (ed. Meiselman, H. L.) 299–321 (Elsevier, 2016).
Bourke, M., Hilland, T. A. & Craike, M. Contextual influences on the within-person association between physical activity and affect in adolescents: an ecological momentary assessment study. J. Behav. Med. 44, 296–309 (2021).
Bourke, M., Hilland, T. A. & Craike, M. Daily physical activity and satisfaction with life in adolescents: an ecological momentary assessment study exploring direct associations and the mediating role of core affect. J. Happiness Stud. 23, 949–968 (2022).
Bourke, M., Hilland, T. A. & Craike, M. Domain specific association between physical activity and affect in adolescents’ daily lives: an ecological momentary assessment study. Psychol. Health 38, 369–388 (2023).
Seiferth, C. et al. How to e-mental health: a guideline for researchers and practitioners using digital technology in the context of mental health. Nat. Mental Health 1, 542–554 (2023).
Timm, I., Reichert, M., Ebner-Priemer, U. W. & Giurgiu, M. Momentary within-subject associations of affective states and physical behavior are moderated by weather conditions in real life: an ambulatory assessment study. Int. J. Behav. Nutr. Phys. Act. 20, 117 (2023).
Ekkekakis, P. The Measurement of Affect, Mood, and Emotion: A Guide for Health-Behavioral Research (Cambridge Univ. Press, 2013).
Larsen, R. J. & Diener, E. in Emotion: Review of Personality and Social Psychology Vol. 13 (ed. Clark, M. S.) 25–59 (Sage, 1992).
Killingsworth, M. A. & Gilbert, D. T. A wandering mind is an unhappy mind. Science 330, 932 (2010).
Stavrakakis, N. et al. Temporal dynamics of physical activity and affect in depressed and nondepressed individuals. Health Psychol. 34S, 1268–1277 (2015).
Difrancesco, S. et al. Within-day bidirectional associations between physical activity and affect: a real-time ambulatory study in persons with and without depressive and anxiety disorders. Depress. Anxiety 39, 922–931 (2022).
Wright, A. G. C. & Woods, W. C. Personalized models of psychopathology. Annu. Rev. Clin. Psychol. 16, 49–74 (2020).
Rosmalen, J. G. M., Wenting, A. M. G., Roest, A. M., Jonge, P. D. & Bos, E. H. Revealing causal heterogeneity using time series analysis of ambulatory assessments: application to the association between depression and physical activity after myocardial infarction. Psychosom. Med. 74, 377–386 (2012).
Ekkekakis, P., Hall, E. E. & Petruzzello, S. J. Variation and homogeneity in affective responses to physical activity of varying intensities: an alternative perspective on dose-response based on evolutionary considerations. J. Sports Sci. 23, 477–500 (2005).
Alcántara, C. et al. Social determinants as moderators of the effectiveness of health behavior change interventions: scientific gaps and opportunities. Health Psychol. Rev. 14, 132–144 (2020).
Mauvais-Jarvis, F. et al. Sex and gender: modifiers of health, disease, and medicine. Lancet 396, 565–582 (2020).
Ajzen, I. The theory of planned behavior. Organ. Behav. Hum. Decis. Process. 50, 179–211 (1991).
Rosenstock, I. M. Historical origins of the health belief model. Health Educ. Monogr. 2, 328–335 (1974).
Bandura, A. & Adams, N. E. Analysis of self-efficacy theory of behavioral change. Cogn. Ther. Res. 1, 287–310 (1977).
Deci, E. L. & Ryan, R. M. in Intrinsic Motivation and Self-Determination in Human Behavior (eds Deci, E. L. & Ryan, R. M.) 11–40 (Springer, 1985).
Prochaska, J. O. & DiClemente, C. C. Stages and processes of self-change of smoking: toward an integrative model of change. J. Consult. Clin. Psychol. 51, 390–395 (1983).
Feil, K., Fritsch, J. & Rhodes, R. E. The intention–behaviour gap in physical activity: a systematic review and meta-analysis of the action control framework. Br. J. Sports Med. 57, 1265–1271 (2023).
Stults-Kolehmainen, M. A., Dunton, G., Boullosa, D., Ash, G. I. & Filgueiras, A. Editorial: motivation states and hedonic motivation for physical activity, exercise, and sport vs. sedentary behaviors. Front. Sports Act. Living 5, 1282118 (2023).
Stults-Kolehmainen, M. A. et al. Motivation states for physical activity and sedentary behavior: desire, urge, wanting, and craving. Front. Psychol. 11, 568390 (2020).
Williams, D. M. in Affective Determinants Of Health Behavior (eds Lliams, D. M., Rhodes, R. E. & Conner, M. T.) 204–234 (Oxford Academic, 2018).
Ekkekakis, P., Parfitt, G. & Petruzzello, S. J. 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 Med. 41, 641–671 (2011).
Ekkekakis, P. The dual-mode theory of affective responses to exercise in metatheoretical context: I. Initial impetus, basic postulates, and philosophical framework. Int. Rev. Sport Exerc. Psychol. 2, 73–94 (2009).
Williams, D. M. & Evans, D. R. Current emotion research in health behavior science. Emot. Rev. 6, 277–287 (2014).
Kahnemann, D. Thinking, Fast and Slow (Farrar, Straus and Giroux, 2011).
Heron, K. E. & Smyth, J. M. Ecological momentary interventions: incorporating mobile technology into psychosocial and health behaviour treatments. Br. J. Health Psychol. 15, 1–39 (2010).
Loewenstein, G. & Lerner, J. S. in Handbook of Affective Sciences (eds Davidson, R. J., Scherer, K. R. & Goldsmith, H. H.) 619–642 (Oxford Academic, 2002).
Mendl, M. & Paul, E. S. Animal affect and decision-making. Neurosci. Biobehav. Rev. 112, 144–163 (2020).
Giurgiu, M., Timm, I., Ebner-Priemer, U. W., Schmiedek, F. & Neubauer, A. B. Causal effects of sedentary breaks on affective and cognitive parameters in daily life: a within-person encouragement design. npj Ment. Health Res. 3, 64 (2024).
Schmiedek, F. & Neubauer, A. B. Experiments in the wild: introducing the within-person encouragement design. Multivariate Behav. Res. 55, 256–276 (2020).
Ruissen, G. R. Establishing causal inferences from experimental and observational data: a critical review and primer for sport and exercise psychology. Sport Exerc. Perform. Psychol. 14, 7–23 (2025).
Poppe, L. et al. Experience sampling method studies in physical activity research: the relevance of causal reasoning. Int. J. Behav. Nutr. Phys. Act. 22, 28 (2025).
Weis, J. Cancer-related fatigue: prevalence, assessment and treatment strategies. Expert Rev. Pharmacoecon. Outcomes Res. 11, 441–446 (2011).
Puetz, T. W. Physical activity and feelings of energy and fatigue: epidemiological evidence. Sports Med. 36, 767–780 (2006).
Huizinga, F. et al. Home-based physical activity to alleviate fatigue in cancer survivors: a systematic review and meta-analysis. Med. Sci. Sports Exerc. 53, 2661–2674 (2021).
Merikangas, K. R. et al. Real-time mobile monitoring of the dynamic associations among motor activity, energy, mood, and sleep in adults with bipolar disorder. JAMA Psychiatry 76, 190–198 (2019).
Robinaugh, D. J., Haslbeck, J. M. B., Ryan, O., Fried, E. I. & Waldorp, L. J. Invisible hands and fine calipers: a call to use formal theory as a toolkit for theory construction. Perspect. Psychol. Sci. 16, 725–743 (2021).
Jekauc, D., Voelkle, M. C., Giurgiu, M. & Nigg, C. R. Unveiling the multidimensional nature of the intention–behavior gap: dynamics of intentions and behaviors. Eur. J. Health Psychol. 32, 34–50 (2025).
Bennett, D., Davidson, G. & Niv, Y. A model of mood as integrated advantage. Psychol. Rev. 129, 513–541 (2022).
Vermaelen, N. et al. The experience sampling method (ESM) item repository. OSF https://doi.org/10.17605/OSF.IO/KG376 (2025).
Ruissen, G. R. et al. Continuous-time modeling of the bidirectional relationship between incidental affect and physical activity. Ann. Behav. Med. 56, 1284–1299 (2022).
Ruissen, G. R., Zumbo, B. D., Rhodes, R. E., Puterman, E. & Beauchamp, M. R. Analysis of dynamic psychological processes to understand and promote physical activity behaviour using intensive longitudinal methods: a primer. Health Psychol. Rev. 16, 492–525 (2022).
Barisic, A., Leatherdale, S. T. & Kreiger, N. Importance of frequency, intensity, time and type (FITT) in physical activity assessment for epidemiological research. Can. J. Public Health 102, 174 (2011).
O’Donoghue, G. et al. A systematic review of correlates of sedentary behaviour in adults aged 18-65 years: a socio-ecological approach. BMC Public Health 16, 163 (2016).
Davis, M. G. et al. Objectively measured physical activity in a diverse sample of older urban UK adults. Med. Sci. Sports Exerc. 43, 647–654 (2011).
Sun, F., Norman, I. J. & While, A. E. Physical activity in older people: a systematic review. BMC Public Health 13, 449 (2013).
Hofmann, W. et al. Rethinking behaviour change interventions in policymaking. Nat. Hum. Behav. 9, 1765–1767 (2025).
Althoff, T. et al. Large-scale physical activity data reveal worldwide activity inequality. Nature 547, 336–339 (2017).
Althoff, T. et al. Countrywide natural experiment links built environment to physical activity. Nature 645, 407–413 (2025).
Tost, H. et al. Neural correlates of individual differences in affective benefit of real-life urban green space exposure. Nat. Neurosci. 22, 1389–1393 (2019).
Nigg, C. et al. Time to go green? Sports Psychiatry 1, 116–118 (2022).
Lu, Y. et al. Wearable data link urban green space to physical activity. Nat. Health 1, 67–77 (2026).
Boyle, H. K. et al. Affective response as a mediator of the association between the physical and social environment and physical activity behavior. J. Behav. Med. 43, 773–782 (2020).
Dunton, G. F., Liao, Y., Intille, S., Wolch, J. & Pentz, M. A. Physical and social contextual influences on children’s leisure-time physical activity: an ecological momentary assessment study. J. Phys. Act. Health 8, 103–108 (2011).
Kanning, M. et al. Doing exercise or sport together with one’s child is positively associated with mothers’ momentary affect in daily life, but not with higher levels of overall physical activity. BMC Public Health 20, 715 (2020).
Jackson, T., Gao, X. & Chen, H. Differences in neural activation to depictions of physical exercise and sedentary activity: an fMRI study of overweight and lean Chinese women. Int. J. Obes. 38, 1180–1185 (2014).
Elsangedy, H. M. et al. Poorer positive affect in response to self-paced exercise among the obese. Physiol. Behav. 189, 32–39 (2018).
Ekkekakis, P., Lind, E. & Vazou, S. Affective responses to increasing levels of exercise intensity in normal-weight, overweight, and obese middle-aged women. Obesity 18, 79–85 (2010).
Ekkekakis, P. & Lind, E. Exercise does not feel the same when you are overweight: the impact of self-selected and imposed intensity on affect and exertion. Int. J. Obes. 30, 652–660 (2006).
Williams, D. M. et al. Recommending self-paced exercise among overweight and obese adults: a randomized pilot study. Ann. Behav. Med. 49, 280–285 (2015).
Dart, A. M., Du Jun, X. & Kingwell, B. A. Gender, sex hormones and autonomic nervous control of the cardiovascular system. Cardiovasc. Res. 53, 678–687 (2002).
Koenig, J. & Thayer, J. F. Sex differences in healthy human heart rate variability: a meta-analysis. Neurosci. Biobehav. Rev. 64, 288–310 (2016).
Tobaldini, E. et al. Depression and cardiovascular autonomic control: a matter of vagus and sex paradox. Neurosci. Biobehav. Rev. 116, 154–161 (2020).
White, R. L. et al. Domain-specific physical activity and mental health: a meta-analysis. Am. J. Prev. Med. 52, 653–666 (2017).
Cho, D., Armeli, S., Weinstock, J. & Tennen, H. Daily- and person-level associations between physical activity and alcohol use among college students. Emerg. Adulthood 8, 428–434 (2018).
Holtermann, A., Schnohr, P., Nordestgaard, B. G. & Marott, J. L. The physical activity paradox in cardiovascular disease and all-cause mortality: the contemporary Copenhagen General Population Study with 104 046 adults. Eur. Heart J. 42, 1499–1511 (2021).
Geus, E. J. C. D. Genetic pathways underlying individual differences in regular physical activity. Exerc. Sport Sci. Rev. 51, 2–18 (2023).
Lightfoot, J. T. et al. Biological/genetic regulation of physical activity level: consensus from GenBioPAC. Med. Sci. Sports Exerc. 50, 863–873 (2018).
R Core Team. R: A Language and Environment for Statistical Computing (R Foundation for Statistical Computing, April, 2024 and November, 2025).
RStudio: integrated development environment for R. Posit Software https://posit.co/ (2024).
Wickham, H. et al. Welcome to the Tidyverse. J. Open Source Softw. 4, 1686 (2019).
Viechtbauer, W. & Constantin, M. esmpack: R package to facilitate the preparation and management of ESM/EMA data. GitHub https://github.com/wviechtb/esmpack (2023).
Bates, D., Mächler, M., Bolker, B. M. & Walker, S. C. Fitting linear mixed-effects models using lme4. J. Stat. Softw. 67, 1–48 (2015).
Kuznetsova, A., Brockhoff, P. B. & Christensen, R. H. B. lmerTest package: tests in linear mixed effects models. J. Stat. Softw. 82, 1–26 (2017).
Ben-Shachar, M. S., Lüdecke, D. & Makowski, D. effectsize: estimation of effect size indices and standardized parameters. J. Open Source Softw. 5, 2815 (2020).
Rosseel, Y. lavaan: an R package for structural equation modeling. J. Stat. Softw. 48, 1–36 (2012).
Viechtbauer, W. Conducting meta-analyses in R with the metafor package. J. Stat. Softw. 36, 1–48. (2010).
Christensen, R. H. B. Regression models for ordinal data (R package ordinal version 2023.12-4.1). Comprehensive R Archive Network (CRAN) https://CRAN.R-project.org/package=ordinal (2024).
Loy, A., Steele, S. & Korobova, J. lmeresampler: bootstrap methods for nested linear mixed-effects models_. R package version 0.2.4. Comprehensive R Archive Network (CRAN) https://cran.r-project.org/web/packages/lmeresampler/index.html (2023).
Lüdecke, D. ggeffects: tidy data frames of marginal effects from regression models. J. Open Source Softw. 3, 772 (2018).
Lüdecke, D. sjPlot: data visualization for statistics in social science. Comprehensive R Archive Network (CRAN) https://doi.org/10.32614/CRAN.package.sjPlot (2023).
Lüdecke, D., Ben-Shachar, M., Patil, I., Waggoner, P. & Makowski, D. performance: an R package for assessment, comparison and testing of statistical models. J. Open Source Softw. 6, 3139 (2021).
Kassambara, A. ggpubr: ‘ggplot2’ based publication ready plots. Comprehensive R Archive Network (CRAN) https://doi.org/10.32614/CRAN.package.ggpubr (2016).
Firke, S. janitor: simple tools for examining and cleaning dirty data. Comprehensive R Archive Network (CRAN) https://doi.org/10.32614/CRAN.package.janitor (2016).
Becker, J., Chan, C., Schoch, D. & Leeper, T. J. rio: a swiss-army knife for data I/O. Comprehensive R Archive Network (CRAN) https://doi.org/10.32614/CRAN.package.rio (2013).
Auguie, B. gridExtra: miscellaneous functions for ‘grid’ graphics. Comprehensive R Archive Network (CRAN) https://doi.org/10.32614/CRAN.package.gridExtra (2010).
Maier, M., Bartoš, F. & Wagenmakers, E. J. Robust Bayesian meta-analysis: addressing publication bias with model-averaging. Psychol. Methods 28, 107–122 (2023).
Nakagawa, S. et al. orchaRd 2.0: an R package for visualising meta-analyses with orchard plots. Methods Ecol. Evol. 14, 2003–2010 (2023).
Burchartz, A. et al. Assessing physical behavior through accelerometry—state of the science, best practices and future directions. Psychol. Sport Exerc. 49, 101703 (2020).
Ebner-Priemer, U. W. & Trull, T. J. Ambulatory assessment: an innovative and promising approach for clinical psychology. Eur. Psychol. 14, 109–119 (2009).
Egloff, B., Tausch, A., Kohlmann, C. W. & Krohne, H. W. Relationships between time of day, day of the week, and positive mood: exploring the role of the mood measure. Motiv. Emot. 19, 99–110 (1995).
Smith, K. E. et al. Momentary affective response to bouts of moderate-to-vigorous physical activity predicts changes in physical activity and sedentary behavior during behavioral weight loss. Psychol. Sport Exerc. 57, 102056 (2021).
Hancock, K. E., Downward, P. & Sherar, L. B. Exploring feelings of pleasure and purpose associated with older people’s activities using ecological momentary analysis: an observational study. J. Aging Phys. Act. 29, 670–677 (2021).
Russel, J. A. Core affect and the psychological construction of emotion. Psychol. Rev. 110, 145–172 (2003).
Barrett, L. F. Solving the emotion paradox: categorization and the experience of emotion. Pers. Soc. Psychol. Rev. 10, 20–46 (2006).
Scherer, K. R. What are emotions? And how can they be measured? Soc. Sci. Inf. 44, 695–729 (2005).
Cunningham, W. A., Dunfield, K. A. & Stillman, P. E. Emotional states from affective dynamics. Emot. Rev. 5, 344–355 (2013).
Barrett, L. F. The theory of constructed emotion: an active inference account of interoception and categorization. Soc. Cogn. Affect. Neurosci. 12, 1833 (2017).
Kuppens, P. & Verduyn, P. Emotion dynamics. Curr. Opin. Psychol. 17, 22–26 (2017).
Scarantino, A. Emotion Theory: The Routledge Comprehensive Guide (Routledge, 2024).
Russell, J. A. A circumplex model of affect. J. Pers. Soc. Psychol. 39, 1161–1178 (1980).
Thayer, R. E. The Biopsychology of Mood and Arousal (Oxford Univ. Press, 1991).
Posner, J., Russell, J. A. & Peterson, B. S. The circumplex model of affect: an integrative approach to affective neuroscience, cognitive development, and psychopathology. Dev. Psychopathol. 17, 715–734 (2005).
Steyer, R., Schenkmezger, P., Notz, P. & Eid, M. The Multidimensional Mood State Questionnaire (MDBF) (Hogrefe, 1997).
Kamarck, T. W. et al. in Technology and Methods in Behavioral Medicine (eds Krantz, D. S. & Baum, A. S.) 163–193 (Psychology Press, 1998).
Rodgers, M. A. & Pustejovsky, J. E. Evaluating meta-analytic methods to detect selective reporting in the presence of dependent effect sizes. Psychol. Methods 26, 141–160 (2021).
Cook, R. D. in International Encyclopedia of Statistical Science (ed. Lovric, M.) 301–302 (Springer, 2011).
Higgins, J. P. T. et al. A tool to assess risk of bias in non-randomized follow-up studies of exposure effects (ROBINS-E). Environ. Int. 186, 108602 (2024).
Modugno, L. & Giannerini, S. The wild bootstrap for multilevel models. Commun. Stat. Theory Methods 44, 4812–4825 (2015).
Hox, J. & van de Schoot, R. in The SAGE Handbook of Multilevel Modeling (eds Scott, M. A., Simonoff, J. S. & Marx, D. B.) 387–402 (Sage, 2013).
Liddell, T. M. & Kruschke, J. K. Analyzing ordinal data with metric models: what could possibly go wrong?. J. Exp. Soc. Psychol. 79, 328–348 (2018).
Curtiss, J. E., Pinaire, M., Fulford, D., McNally, R. J. & Hofmann, S. G. Temporal and contemporaneous network structures of affect and physical activity in emotional disorders. J. Affect. Disord. 315, 139–147 (2022).
Haucke, M., Heinz, A., Liu, S. & Heinzel, S. The impact of COVID-19 lockdown on daily activities, cognitions, and stress in a lonely and distressed population: temporal dynamic network analysis. J. Med. Internet Res. 24, e32598 (2022).
Hachenberger, J. et al. Investigating associations between physical activity, stress experience, and affective wellbeing during an examination period using experience sampling and accelerometry. Sci. Rep. 13, 8808 (2023).
Acknowledgements
We thank M. Beisemann (TU Dortmund University) for her statistical advice and helpful discussions on the project. Further, we thank C. Hausmann (Karlsruhe Institute of Technology), S. Brüßler (Ruhr University Bochum), R. Hohmann (Ruhr University Bochum), C. Limpinsel (Ruhr University Bochum) and E. Rulf (CIMH Mannheim) for their support regarding literature search and manuscript preparation. Funding: I.T. and J.R. received funding from the DGPs Peer Mentoring programme to collaborate on this project. Further, the project was supported by the German Research Foundation through the Collaborative Research Center TRR265 (grant no. 402170461), project C05 (to M.R.) and by the ERA-NET NEURON Project MASE (BMBF, grant no. 01EW2404A, to M.R.). Additional funding was received by the DIAPASON project (Italian Ministry of Health), the Hellman Foundation (to S.L.), the US Department of Agriculture (grant no. 3092–51000-056-04A; ClinicalTrials.gov: NCT02784509, to A.S.), the NORC Center (grant no. P30DK072476, principal investigator E. Ravussin) from the NIDDK entitled ‘Nutrition, Obesity, and Metabolic Health through the Lifespan’, the NIGMS funding the Louisiana Clinical and Translational Science Center (grant no. U54 GM104940; principal investigator J. Kirwan) and the Research Foundation—Flanders (FWO; to L.P.); the I-GENDO project was funded by the German Federal Ministry of Education and Research (BMBF; grant no. 01GL1719A/B), U.S. R01 AG026010 (to L.M.), the DYNAMIC center, funded by the LOEWE programme of the Hessian Ministry of Science and the Arts (grant no. LOEWE1/16/519/03/09.001(0009)/98), the Ministry of Science, Research and the Arts of the State of Baden-Wuerttemberg, Germany (MWK; grant no. 42-04HV.MED(16)/16/1 to A.M.L.), the Research of Excellence on Digital Technologies and Wellbeing (grant no. CZ.02.01.01/00/22_008/0004583; to S.E.), the Healthy Aging in Industrial Environment (grant no. HAIE CZ.02.1.01/0.0/0.0/16_019/0000798) and the Collaborative Research Center TRR265 (grant no. 402170461), project S02 (to U.E.P. and A.M.L.). The funders had no role in study design, data collection and analysis of this meta-analysis nor in the decision to publish or the preparation of the manuscript.
Funding
Open access funding provided by Paris Lodron University of Salzburg.
Author information
Authors and Affiliations
Consortia
Contributions
M.R., J.R., J.P., I.T., M.G. and G.B. conceptualized the study and selected the methods. M.R. and J.R. administered the project. J.R. and J.P. conducted the formal analyses, and J.R., J.P., I.R., A.B.N. and M.R. validated the analyses. J.R., J.P. and M.R. visualized the data. All authors contributed to data collection and curation. J.R., M.R. and J.P. wrote the original manuscript draft. All authors reviewed, edited and approved the manuscript. M.R., U.E.P. and O. Güntürkün supervised the project. The corresponding author attests that all listed authors meet authorship criteria and that no others meeting the criteria have been omitted.
Corresponding author
Ethics declarations
Competing interests
The authors declare no competing interests.
Peer review
Peer review information
Nature Human Behaviour thanks Matthew N. Ahmadi, Taren Sanders and the other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer reviewer reports are available.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Extended data
Extended Data Fig. 1 Study design of ambulatory assessment studies investigating the relation of affective well-being (AWB) and physical activity (PA) in daily life.
AWB is assessed at multiple discrete time points across a day/multiple days. PA is measured continuously and aggregated over specific time intervals preceding or following the AWB rating. We have taken the chronological order into account by separately computing antecedent (PA preceding the AWB rating) and consequent (PA following the AWB rating) models.
Supplementary information
Supplementary Information (download PDF )
Supplementary Information Sections 1–19, which contain the respective Supplementary Tables and Figures (for example, Supplementary Information Section 4 contains Supplementary Table 4.1a–Supplementary Table 4.6e; Supplementary Information Section 6 contains Supplementary Fig. 6).
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Rehder, J., Timm, I., Berretz, G. et al. An individual participant data meta-analysis of how physical activity relates to affective well-being in daily life. Nat Hum Behav (2026). https://doi.org/10.1038/s41562-026-02427-2
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s41562-026-02427-2
