Social isolation and loneliness increase human mortality like known health risk factors such as obesity, alcohol consumption or smoking 15 cigarettes per day1. Lack of social contact also impairs momentary affective well-being2, impacts the structural and functional integrity of emotion regulatory brain networks3,4 and is a potent risk factor for mood disorders5. Social distancing directives during the coronavirus disease 2019 (COVID-19) pandemic have exacerbated this public health problem and highlighted the importance of finding remedial strategies6. One promising strategy to mitigate the negative affective consequences of lack of social contact is physical activity, a known protective factor for affective well-being and mental health7 with neural mechanistic links to emotion regulatory brain regions8. However, the everyday relevance and biological basis are unknown. In this study, we hypothesized that physical activity can compensate for the negative affective effects of lacking social contact in daily life and that individuals at increased neural4 and psychological9 risk for depression benefit most from this compensatory mechanism.


The physical activity of individuals (Fig. 1 and Table 1) significantly moderated the known relationship2 between momentary social isolation and decreased affective valence in everyday life (β = 0.01; 95% confidence interval (CI) = 0–0.02; P = 0.020; Supplementary Table 2). Specifically, higher physical activity significantly decreased the reduction in affective well-being associated with the lack of social contact (Fig. 2a–c). According to our data, about 349 milli-g (g/1,000) physical activity across 1 h (for example, walking approximately three miles per hour) are necessary to fully compensate for the lack of affective well-being in everyday life (Supplementary Results 1). We successfully replicated this effect in the second sample we studied during the COVID-19 pandemic (β = 0.03; 95% CI = 0.02–0.04; P < 0.001; Fig. 2c, study 2; Supplementary Table 8). At the neurobiological level, individuals with higher resting-state functional connectivity within the default mode network (DMN), a risk phenotype for loneliness4 and depression10, compensated best for this momentary ‘social-affective deficit’ through physical activity (β = 0.14; 95% CI = 0.01–0.26; P = 0.029; Fig. 3b and Supplementary Table 3). Moreover, we observed similar benefits of physical activity at the between-individual level and related it to established psychological risk factors for mental health. First, participants with small social networks and high habitual physical activity levels exhibited lower trait loneliness compared to those with low levels of habitual physical activity (β = 0.05; 95% CI = 0.001–0.092, P = 0.046; Fig. 2d and Supplementary Table 4). Second, individuals with a pronounced compensatory mechanism were less likely to frequently feel lonely during the first COVID-19 lockdown (odds ratio (OR) = 0.92; 95% CI = 0.85–0.99; P = 0.021; Supplementary Table 5). Further exploratory analyses showed that offsetting the social-affective deficit with physical activity was effective even under pandemic-like constraints (curfews, closed gyms), for example, when only light physical activity (β = 0.04; 95% CI = 0–0.8; P = 0.040; Supplementary Table 6) and physical activity at home (β = 0.08; 95% CI = 0.01–0.15; P = 0.032; Supplementary Table 6) are considered.


Our intensive e-diary and accelerometer-based longitudinal data suggest that physical activity can effectively and reproducibly compensate for the loss of affective well-being associated with lack of social contact in real life. While social contact and physical activity are well-known protective resources for mental health1,5,7, previous studies have predominantly examined these factors using questionnaires or individually in the real world2,8. Our naturalistic study extends the state of knowledge by showing a dynamic interplay of both factors impacting human affective well-being in everyday life. Our data further show that about 1 h of walking at a speed of three miles per hour can compensate for the ‘social-affective deficit’ in everyday life and that this beneficial effect even persists when physical activity is performed at lower doses and only at home. This indicates a considerable potential of physical activity to counteract the negative affective consequences of social isolation in everyday life. Importantly, the effect was larger in people at higher neural risk for affective disorders. These included people from the general population with risk-related changes in DMN brain connectivity4,10, smaller social networks11 and frequently perceived loneliness under the regulatory constraints of the COVID-19 pandemic. Thus, our data not only suggest an effective and accessible strategy to mitigate the negative effects of social isolation and loneliness in everyday life, but also contribute to the identification of probable responders and enrich existing evidence-based recommendations for the preventive management of affective dysfunction in the post-pandemic world6,9.


We captured affective valence via an established scale specifically developed and validated for investigating mood in everyday life12,13. Therefore, our study provides insights into mood changes provoked by physical activity and social interaction. However, given the ongoing discussions on mood assessments in the field, future studies should examine the effects of physical activity in the context of lacking social contact on specific emotions (for example, anxiety, anger). Moreover, although our real-life observational data have high ecological validity, they do not allow for causal inferences. In particular, our findings show correlations and the temporal directionality of effects, but we cannot rule out potential influences of undiscovered third variables. Future studies should address the causality question by incorporating experimental manipulations such as just-in-time adaptive interventions into their real-life investigations.


Our multimodal epidemiological cohort study shows that physical activity is reproducibly linked to better affective well-being in people lacking social contact in daily life, especially in persons at neural and psychological risk for affective disorders. These data suggest an effective and accessible strategy to mitigate the negative effects of social isolation and loneliness that can improve public health and enrich existing evidence-based recommendations for the preventive management of social isolation in the post-pandemic world.


The cohort study was conducted in accordance with ethical guidelines for medical research compliant with the Declaration of Helsinki 2013 version. All participants provided written informed consent for a study protocol approved by the institutional review board of Heidelberg University. Medical Faculty Mannheim (medical ethics committee II) at the Ruprecht-Karls-University in Heidelberg approved both studies (study 1: approval no. 2014-555N-MA; study 2: approval no. 2019-733N). Participants received monetary compensation for their effort. The flowchart depicts how the study size was arrived at in both the main (study 1) and the replication study (study 2); see Fig. 1.

Fig. 1: Participant flowchart.
figure 1

Participant numbers according to study stage. EMA, ecological momentary assessment.

Study population and measures

We studied a community-based cohort of 317 healthy young adults aged 18–28 years (57.09% females), recruited from September 2014 to November 2018, for 7 days during everyday life (Table 1 and Supplementary Table 1). We further studied a replication sample of 30 healthy adults aged 18–63 years, recruited from December 2019 to July 2022, for 6 months during everyday life during the COVID-19 pandemic in Germany (Supplementary Table 7). The biological sex of participants was determined using a questionnaire.

Table 1 Demographic and psychological characteristics, ambulatory assessment and neuroimaging parameters of study 1

Participants wore accelerometers on their hip (study 1) or wrist (study 2) to measure their physical activity, and repeatedly reported their real-life social contact and affective valence using smartphone-based e-diaries (Fig. 2a). Established multilevel reliability measures (Spearman–Brown14) yielded sound coefficients of ρ = 0.80 (within-individual level) and ρ = 0.94 (between-individual level) in our sample and for the two affective valence variables assessed (that is, unwell to well and content to discontent). Moreover, within and between person correlations of the two items applied yielded positive correlations (rwithin = 0.66; rbetween = 0.88), which indicates convergent validity for the affective valence assessment instrument applied. Participants additionally completed a battery of psychological questionnaires11,15, and we continuously tracked their geographical locations and situational contexts as described previously2 (Fig. 2b and Supplementary Information 1). A total of 175 participants from study 1 additionally underwent a resting-state functional magnetic resonance imaging (fMRI) scan after the ambulatory study week to quantify DMN connectivity (Supplementary Results 2), a neural risk marker for social isolation and depression4,10. In 76 participants from study 1, we additionally assessed individuals’ perceptions of loneliness during the ongoing first wave of the COVID-19 pandemic (Supplementary Results 4).

Fig. 2: Ambulatory assessment and behavioral study findings.
figure 2

a, Accelerometry was used to measure physical activity, while affective valence and social contact were assessed through ecological momentary assessment. b, Exemplified sampling scheme: geolocations were continuously tracked and assigned using an advanced day reconstruction method (for example, at home, work). E-diaries were either location-based or triggered at random times. c, Study 1 (n = 317; Table 1). Physical activity engagement (x axis) offsetting the social-affective deficit (y axis) associated with the absence of real-life social contact as illustrated by the gray-shaded area between the solid (in company) and dashed (alone) green lines. The regression lines, derived from the multilevel interaction analyses (outcome: affective valence; predictor: real-life social contact; moderator: physical activity centered within-individual), demonstrate that the more participants had been physically active before an e-diary assessment, the less affective loss they experienced when being alone. Physical activity values to the very left of the x axis refer to sedentary behavior such as sitting, while values to the very right depict moderate activities such as walking. Study 2 (n = 30; Supplementary Table 7). Replication of the compensatory effect of physical activity during the COVID-19 pandemic. P values for the beta coefficients are two-sided and were derived from the t-statistics of the multilevel model. The error bars indicate the s.e. of the respective estimated mean valence scores. d, Trait loneliness. Participants with small social networks (light green) who engaged in high habitual levels of physical activity reported lower trait loneliness compared to those engaging in low habitual levels of physical activity (Supplementary Table 4). P values are two-sided and were derived from the t-statistics of the multiple linear regression. The error bars indicate the s.e. of the respective estimated mean loneliness scores. Credit: a, smartphone icon, Elisa Riva, Map in b created using OpenStreetMap.

Power analysis

Because statistical power analyses of multilevel models strongly depend on a host of assumptions (for example, on random slopes, covariance structure) that cannot be drawn in the absence of the final dataset16, we estimated whether our final sample size of n = 317 was suitable to detect the expected effects referring to the most recent simulation studies17. According to these simulation studies, a sample size of n = 200 was necessary to detect the minimum detectable effect size (0.08) in a level-1 direct effect analysis given a level-1 sample size of at least 30 at a power of 80%, which provides evidence for the sufficient power of our analysis.

Data analysis

All statistical analyses were performed with SAS v.9.4. Brain imaging data were analyzed using the CONN toolbox v.19c in MATLAB v. 9.8 (R2020a). Study 1: within participants (main model), we analyzed the main and interaction effects of momentary social contact (predictor: alone versus in company) and momentary physical activity (moderator: mean of milli-g in the 60 min before an e-diary prompt) on momentary affective valence (outcome) using multilevel models with time of day, time of day squared, current location (level 1), sex, age and body mass index (level 2) as covariates. Between participants, we predicted trait loneliness (outcome) with the main and interaction terms of social network size11 (predictor) and habitual physical activity level (moderator: hours per week). In addition, we predicted the frequency of perceived loneliness during the first COVID-19 lockdown (outcome) by extracting random slopes from the multilevel interaction of social contact and physical activity on affective valence (predictor: from the main model) and fitting an ordinal logistic regression model assuming proportional odds. At the neural level, we computed DMN connectivity estimates from the participants’ resting-state fMRI data (Fig. 3) and introduced them as an additional moderator into our main model, resulting in a three-way multilevel interaction analysis. In study 2, we used the main model of study 1 to replicate the findings during the ongoing COVID-19 pandemic (see Supplementary Information for more details).

Fig. 3: DMN and study findings at the neural level.
figure 3

a, According to the neuronal signatures seen in ‘lonely brains’, we analyzed the within-network connectivity of the DMN based on the 100-region, 7-network Schaefer–Yeo parcellation atlas18. b, Participants with higher within-DMN connectivity, a neuronal signature repeatedly found in lonely individuals and associated with affective disorders, showed a pronounced compensation of the momentary ‘social-affective deficit’ through physical activity (Supplementary Table 3). P values for the beta coefficients are two-sided and were derived from the t-statistics of the multilevel model. The error bars indicate the s.e. of the respective estimated mean valence scores.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.