Abstract
Although the COVID-19 pandemic increased psychological distress globally, not all countries experienced the same mental health burden. This study investigates one potential driver of cross-national heterogeneity in pandemic mental health outcomes: country-level social capital. In particular, it explores whether pre-pandemic country-level social capital—generalized trust, personal/family relationships, and civic/social participation—buffered the mental health impacts of mixing/mobility restrictions, measured using high-frequency Google mobility data. Focusing on developed economies, two individual-level datasets are analyzed: a four-wave panel (n = 10 countries) and a cross-sectional survey spanning a near-complete sample of all developed economies (n = 34 countries). Hierarchical modeling shows increasing country-level immobility predicts heightened individual-level psychological distress, but this association is significantly weaker in higher-social-capital societies, particularly those with higher civic/social participation. Mediation analysis indicates that social capital attenuated two key pathways linking immobility to distress—greater loneliness and reduced satisfaction with governments’ handling of the pandemic—which accounts for nearly half of the apparent cushioning role of country-level social capital. Longitudinal modeling provides more causally robust evidence, while a cross-sectional replication across a near-complete sample of developed economies indicates generalizability. Findings suggest elements of country-level social capital may be a vital contextual resource fostering societal resilience during (trans-)national crises.
Introduction
The COVID-19 pandemic significantly harmed global mental health, with some estimates suggesting the worldwide prevalence of anxiety/major depressive disorders increased 25–30 percent (WHO 2020). A key driver of worsening mental health was the prolonged stresses associated with reduced social contact and decreased spatial mobility, stemming from measures designed to contain the virus (Chan et al. 2024; Laurence and Kim 2021). However, while the onset of the pandemic was global, not all countries experienced the same mental health burden (Chen et al. 2025; Cénat et al. 2022; Gloster et al. 2021). While a growing body of research has examined which country-level factors might explain cross-national differences in physical health outcomes during the pandemic, such as virus transmission and death rates (Bartscher et al. 2021; Elgar, Stefaniak, and Wohl 2020), to date, less attention has been given to the country-level factors that account for cross-national differences in psychological distress during the pandemic (although see Aknin et al. 2022; Yastrebov and Maskileyson 2022). One factor that may have played a key role in shaping how far the pandemic impacted mental health is average levels of social capital in a country; for example, the mean level of social trust or civic engagement across members of a society.
Social capital constitutes social networks and norms that facilitate cooperation for mutual benefit, which can exhibit protective (“buffering”) effects for mental health in the face of adversity (De Silva et al. 2005; Kawachi and Berkman 2000). Studies conceptualize social capital as both an individual- and contextual-level resource (Kawachi and Berkman 2001; Laurence and Kim 2021). When operationalized as an individual-level (or egocentric-network) resource, the protective benefits are accrued by individuals based on their own, self-reported level of social capital (e.g., how much they trust others, or whether they are civically engaged). However, when operationalized as a contextual-level resource (i.e., a macro-level shared property of all members of an area), social capital is believed to form a “public good,” where its protective benefits are accrued by all people in an environment (e.g., average levels of trust or civic engagement in an area) (Aminzadeh et al. 2013; Nakagomi et al. 2020). A growing volume of research has explored how individuals’ own level of pre-pandemic social capital (e.g., their frequency of civic engagement) buffered the impact of the pandemic on their mental health—an individual buffering effect (Laurence and Kim 2021; Sato, Kondo, and Kondo 2022). Meanwhile, studies have also explored how contextual, country-level social capital protected people's physical health during the pandemic, for example, reducing COVID-19 death rates (Elgar et al. 2020). However, what remains unclear is whether country-level reserves of social capital can also cushion the collateral harm of large-scale crises, such as the pandemic, on individuals’ psychological health—a contextual buffering effect of social capital for mental health.
This study aims to fill this knowledge gap. We argue that when uncertainty is high, and individual coping mechanisms are weakened, national levels of social capital may become critical for protecting mental health during large-scale crises. This question is explored by investigating whether country-level social capital (measured using aggregated levels of generalized trust, personal/family relationships, or civic/social participation in a country) buffered the impact of the pandemic on individuals’ mental health. In particular, we draw on high-frequency, country-level Google spatial mobility data (i.e., changes in the amount of time people were spending in residential settings) to measure how increases in spatial immobility across a country shaped individuals’ psychological distress, and whether this relationship was buffered (i.e., moderated) by levels of social capital in a country. In addition, we conduct exploratory analysis into two pathways through which any country-level social capital buffering processes might operate: loneliness and satisfaction with governments’ handling of the pandemic. We focus on the pandemic experiences of developed economies (United National Development Programme [UNDP] 2024) (given better data coverage during the pandemic), drawing on two cross-national datasets. The first is a four-wave panel survey of individuals (n = 10 developed economies) followed during April 2020–July 2021, to undertake more causally robust analyses. The second is a cross-sectional survey of individuals across n = 34 developed economies conducted at the start of the pandemic, to test the generalizability of findings in a near-complete sample of all developed economies.
The global onset of the COVID-19 pandemic presents a unique opportunity to explore how societies respond to large-scale crises and how macro-level (national) social structures, such as social capital, might protect individuals’ mental health. Through this approach, the study also aims to help understand cross-national differences in mental health outcomes during the pandemic, shedding light on what factors can generate societal resilience during global crises.
Theoretical Framework
Social Capital and Mental Health
Social capital constitutes social networks, norms, and trust that facilitate cooperation for mutual benefit (Almedom 2005; Borgonovi and Andrieu 2020). Social capital is often disaggregated into different dimensions, including cognitive social capital (social trust/norms, predisposing people toward collective action) and structural social capital (networks which provide access to resources/facilitate cooperation). Such networks are, in turn, disaggregated into bonding social capital (closer ties, such as friends/family) and bridging social capital (weaker ties across social groups, often proxied by civic/social participation) (Li, Pickles, and Savage 2005).
Social capital has long been linked to better mental health (Almedom 2005; De Silva et al. 2005). It provides access to emotional/informational/practical support networks (Aminzadeh et al. 2013; Louie, Upenieks, and Hill 2023). Social networks can also foster positive psychological states, such as belongingness or security, alongside shared identity and trust (in people and institutions), reducing stress and anxiety (Kawachi and Berkman 2000; Snel et al. 2022). However, studies have also demonstrated how social capital, and social networks more broadly, can negatively impact mental health. For example, social cost theory suggests that negative social comparisons, social obligations, and costs of social support, as well as conflict and strain across ties, can be detrimental to mental health (e.g., Song 2020; Song et al. 2021). While social capital can therefore be both beneficial and detrimental to mental health, studies demonstrate how the positive benefits of social capital can become particularly important for mental health when individuals face acute/chronic stressors, such as pecuniary hardship, widowhood, violent crime, or suicide, reducing their negative impact on mental health, that is, exhibiting a buffering effect (e.g., Aminzadeh et al. 2013; Nakagomi et al. 2020; Zeng and Wu 2022). More recently, studies have extended this buffering capacity of social capital to examine its protective role for mental health during large-scale crises (Aldrich and Meyer 2014). Indeed, within-country studies report how individuals’ own social capital cushioned mental health during the COVID-19 pandemic, for example, where more locally connected individuals saw smaller pandemic increases in distress (Han and Chung 2023; Laurence and Kim 2021; Laurence, Russell, and Smyth 2024; Mannarini et al. 2022; Sato et al. 2022).
One conceptual framework that has been drawn on to explain this buffering capacity of social capital is the stress-buffering model of social support, which posits that supportive relationships reduce the negative impact of stressors on health by providing emotional, informational, or practical resources that reduce the perceived threat of, or help cope with, stressors in everyday life (Cohen and Wills 1985). Research has documented that both the perception and actual availability of support operate important independent buffering effects for mental health (Cohen and Wills 1985; Rui 2022). Drawing on this model, social capital studies suggest that the social networks that underpin (both individual- and contextual-level) social capital, coupled with their constituent norms of trust/reciprocity, are linked with both greater access to social support and greater expectations that support would be available if needed (perceived support), driving social capital's stress-buffering capacity (Kawachi and Berkman 2001; Kim, Subramanian, and Kawachi 2006).
Further research, however, drawing on additional conceptual frameworks, has extended the pathways through which social capital's buffering processes might operate beyond support alone. For example, studies emphasizing the benefits of social capital for collective efficacy and institutional trust suggest buffering can occur by enabling better coordination within communities, and between communities and state institutions, to manage the impact of stressors, while also dampening their perceived threat when trust in institutional support is higher (Helliwell et al. 2021; Nakagomi et al. 2020; Zeng and Wu 2022). Meanwhile, social identity approaches suggest that wider social networks (which underpin social capital) can promote a sense of belonging and group-based identity, which reduces threat appraisals and feelings of isolation, while building resilience to stressors through cultivating psychological resources such as perceived control (Haslam et al. 2021; Thoits 2011).
Research into social capital's buffering capacity demonstrates how social capital as an individual-level resource, such as one's own ties, trust, or civic participation, can cushion mental health from the impact of stressors (Acevedo, Ellison, and Xu 2014; Robinette et al. 2021). Here, social capital is viewed as a “private good,” where the protective benefits are accrued by individuals based on their own, personal connectivity. However, studies demonstrate that contextual-level social capital (e.g., average levels of trust or civic engagement within an individuals’ neighborhood, school district, or census county) can also cushion mental health, even after accounting for their individual-level social capital (Aminzadeh et al. 2013; Laurence and Kim 2021; Sato et al. 2022). Here, social capital forms a “public good” where all members of an area receive protective benefits regardless of their personal connectivity. Importantly, therefore, contextual social capital is posited to confer protection beyond individuals’ own social capital. Applied to the stress-buffering model for example, contexts richer in social capital may foster greater actual/perceived social support given stronger social connectivity, civic/social networks, and prosocial norms across a context can facilitate more effective pooling of time and resources to organize support; improve information flows about who is in need, how resources can be accessed, and how best to provide support; and heighten expectations of support provision (Kawachi and Berkman 2001; Kim et al. 2006). Therefore, even more isolated individuals living in contexts where others are highly connected may experience protection if residents are better able to organize support and coordinate action to overcome problems.
During major crises, contextual-level social capital may be especially effective for protecting mental health as connectivity/prosocial norms among all members of an area may better enable cooperation, information flows, support, or adherence to governmental directives. The capacity/willingness of one connected individual to cooperate/act pro-socially in a place where most are disconnected during a crisis may be more limited (e.g., Wind and Komproe 2012). In addition, different dimensions of contextual social capital could confer different degrees of protection during crises. For example, research suggests denser networks of close friend/family ties may act as a burden (time demands, emotional strain, stresses of responsibility), which can harm mental health, especially in the initial stages of crises, although potentially facilitating speedier recoveries in the longer term (Hawkins and Maurer 2010; Weil, Lee, and Shihadeh 2012). Meanwhile, prior work has demonstrated that civic dimensions of social capital appear more robust to the impacts of large-scale crises and more effective at cushioning mental health than closer, bonding forms of social capital (Hall et al. 2023; Lim and Laurence 2015).
Research has therefore explored how localized forms of social capital (e.g., within neighborhoods) may protect individuals’ mental health during crises. Studies also examine how country-level social capital may have protected individuals’ physical health during the pandemic. However, to our knowledge, little research has investigated the capacity of country-level social capital to cushion how (trans)national crises harm mental health, nor whether different dimensions of country-level social capital confer differential protective effects. Using the pandemic as a natural experiment, we outline key reasons why country-level social capital might buffer the impact of major crises on psychological health.
Country-Level Social Capital, Mental Health and the COVID-19 Pandemic
Drawing on the framework above, we suggest country-level social capital may evince contextual buffering properties during large-scale crises via cushioning two key pathways of psychological harm. The first pathway is feelings of loneliness/social isolation. A key stressor-pathway during the pandemic was disconnection from many sources of social contact and regular support networks, which heightened loneliness/isolation, increasing distress (Chan et al. 2024; Laurence and Kim 2021). However, country-level social capital may have cushioned how far government mixing/mobility restrictions impacted loneliness and isolation in societies.
One possibility is that, despite the presence of mixing restrictions, people still found ways to provide social support during the pandemic. Informally organized mutual aid groups provided necessities (e.g., food, medicine), welfare checks, transportation, or emergency care, especially to more isolated or vulnerable populations (Carstensen, Mudhar, and Munksgaard 2021; Sitrin and Sembrar 2020). Over time, many groups developed into more formalized organizations (Fernandes-Jesus et al. 2021), provided national support (e.g., raising funds for key workers), and developed ties with/received Funding from state institutions, acting as brokers between governments and communities (Benton and Power 2021; Rendall et al. 2024). Similar support was provided via civic/voluntary networks already in place before the pandemic, which mobilized members, and undertook further recruiting, to provide help (Carstensen et al. 2021; Sitrin and Sembrar 2020).
In countries with higher social capital, pre-existing norms/networks may therefore have more effectively facilitated the collective organization of mutual aid networks (Helliwell et al. 2021). In particular, denser pre-pandemic networks of civic engagement, as a form of “institutionalized altruism,” may have been especially effective for mobilizing support given they are more resilient than informal helping behaviors during crises (Lim and Laurence 2015). As such, during the pandemic, individuals in countries with higher social capital (providing more optimal conditions for the emergence of prosocial behaviors) may have both received and provided more social support. This wider presence of helping behaviors across a country may have, in turn, fostered greater perceptions of social support in societies, even among those not directly receiving it (Han and Chung 2023). Given both perceived and actual support are tied to less loneliness/isolation (Zhang and Dong 2022), greater social support in higher social capital countries may have buffered the impact of mixing restrictions on feelings of loneliness and isolation, in turn reducing their impact on distress.
Another mode by which country-level social capital may have buffered the impact of mixing/mobility restrictions on loneliness and isolation is that national levels of social capital are also linked with a stronger sense of cohesion, shared unity, and belonging (Delhey et al. 2018; Keeley 2007). Stronger group identification and shared identity can foster psychological resources such as a sense of collective meaning, purpose, and self-efficacy, which reduce feelings of loneliness and isolation (Haslam et al. 2021; Robinette et al. 2021). Accordingly, in countries with higher social capital, increasing mixing restrictions may have resulted in fewer feelings of loneliness and social isolation, even with significant reductions in social contact, in turn, buffering the impact of restrictions on mental health.
The second pathway of pandemic harm which country-level social capital may have buffered is individuals’ confidence with their governments’ handling of the crisis. Research shows how more severe pandemic restrictions generally undermined people's trust and confidence in their government's handling of the pandemic (e.g., Gollwitzer et al. 2021). Lower confidence in governments’ crisis management, and disapproval of the use of mixing restrictions, in turn, was linked with increased distress (Chen et al. 2021; Snel et al. 2022; Ye, Yu, and Zhang 2024). Potentially, country-level social capital might have positively shaped perceptions of the necessity of mixing restrictions, alongside their governments’ handling of the crisis more generally, moderating how mixing restrictions impacted confidence in governments’ crisis management. Living in countries with more social capital can engender greater trust in, and acceptance of, public health measures, posited to stem from a greater commitment to the common good or sense of shared membership and connection to society (Bartscher et al. 2021). This may be partly driven by the positive link between social capital and trust/confidence in governments/institutions, which could promote positive perceptions of governments’ crisis handling (Helliwell et al. 2021). If country-level social capital increased acceptance of restrictions, this may have also reduced anxiety and stresses associated with them (Borgonovi and Andrieu 2020). Accordingly, in high social capital societies, mixing/mobility restrictions may have had a weaker negative effect on individuals’ satisfaction with governments’ handling of the crisis, in turn, cushioning the impact of restrictions on their psychological distress.
In sum, we suggest pandemic mixing/mobility restrictions (partly) heightened distress via increasing feelings of loneliness/social isolation and reducing people's confidence in their government's handling of the pandemic. However, country-level social capital may have cushioned the impact of restrictions on these pathways of harm, in turn, buffering how restrictions impacted psychological distress. Figure 1 summarizes this conceptual model.
Conceptual model outlining the tested pathways through which country-level social capital moderates the impact of the pandemic on mental distress.
Current Study
This study explores if, and how, country-level social capital buffered the impact of reduced social connectivity (country-level increases in the amount of time people spent in residential settings, i.e., spatial immobility) on individuals’ psychological distress across developed economies during the COVID-19 pandemic. Three dimensions of social capital are tested: country-level averages of generalized trust, personal/family relationships, and civic/social participation. Drawing on the conceptual framework above (summarized in Figure 1), we derive hypotheses. First, increasing country-level spatial immobility will be associated with higher individual-level psychological distress (H1). Second, country-level social capital will be negatively associated with individual-level distress (H2). Third, the key hypothesis is that the positive association between spatial immobility and psychological distress will be weaker in countries with higher aggregate social capital, that is, country-level social capital will play a buffering/cushioning role (H3).
The study also examines two potential buffering pathways: that increasing immobility will have a weaker positive association with distress in higher social capital countries because it leads to less negative attitudes toward the government's handling of the pandemic and because it leads to fewer feelings of loneliness and isolation (Figure 1). We suggest that increasing country-level spatial immobility will be associated with higher individual-level experiences of loneliness (pathway a on Figure 1). However, this positive association will be negatively moderated by country-level social capital (pathway b, Figure 1). Immobility will therefore exhibit a significantly weaker positive association with loneliness in countries with higher social capital (H4a). Higher loneliness will also be positively associated with greater psychological distress (pathway c, Figure 1) (H4b). Taken together, increasing immobility will have a weaker positive association with psychological distress in countries with higher social capital because residents experience smaller increases in loneliness (H4c).
Similar mediating pathways will operate via attitudes toward the government's handling of the pandemic (government satisfaction). Increasing country-level spatial immobility will be associated with individuals reporting lower government satisfaction (pathway d on Figure 1), but this negative association will be positively moderated by country-level social capital (pathway e, Figure 1). Increasing immobility will therefore exhibit a significantly weaker negative association with government satisfaction in countries with higher social capital (H5a). Higher government satisfaction will be associated with less psychological distress (pathway f, Figure 1) (H5b). Taken together, increasing immobility will have a weaker positive association with psychological distress in countries with higher social capital because residents experience smaller decreases in satisfaction with their governments’ handling of the pandemic (H5c).
Data and Methods
Data
This study draws on two cross-national datasets. The first is the Citizens Attitudes Under COVID-19 panel survey (henceforth, CAUP) of 11 countries: Austria, France, Germany, Italy, Sweden, United Kingdom, Poland, Australia, New Zealand, USA and Brazil (Brouard et al. 2022). Five waves of data were collected: Wave 1 (March 20–30, 2020), Wave 2 (April 15–May 2, 2020), Wave 3 (June 15–28, 2020), Wave 4 (December 4–10, 2020), and Wave 5 (June 28–July 13, 2021). Sweden, Poland, and Brazil entered the survey from Wave 2 onwards. Respondents (aged 18+) were recruited through quota sampling to be representative of each country (see Brouard et al. 2022).
The average wave-on-wave response rate across countries is 53 percent (from 75 percent in France/UK to 34 percent in New Zealand) (Brouard et al. 2022). Sixty-eight percent of respondents participated in two or more waves. 1 In response to attrition, top-up samples were added each wave to reach a minimum of n = 1000 respondents per country-wave and correct the sample's representativeness. A total of n = 30,455 uniquely identified respondents were surveyed. Several sample restrictions were put in place. First, Brazil is excluded from the sample given it is not a UNDP developed economy. Second, Waves 4 and 5 of data from the USA/Australia are excluded, given no interview date is available to link COVID-19/immobility data. Third, data from Wave 1 are excluded given several key covariates were not measured in this wave (and Sweden/Poland were not present in Wave 1). Sensitivity testing shows these restrictions do not change the substantive results (see below). Accounting for within-case missingness on key variables (<7 percent) results in a final analytic sample of n = 41,346 person-observations across 10 countries. Descriptives can be found in Supplementary Appendix S.1.
We employ two approaches to address bias from unequal attrition probabilities and within-case missingness. First, probability weights (Random Iterative Method) supplied in the data weight the sample to be representative of the country populations. In addition, we construct inverse probability longitudinal weights using full model variables including distress and immobility scores to further adjust estimates for non-random attrition when analyzing the data longitudinally. Second, to further examine bias from missingness, we test multiple imputation (20 datasets) by chained equations. The results are highly similar (see below).
The second dataset is the cross-sectional Global Behaviors, Perceptions, and the Emergence of Social Norms at the Onset of the COVID-19 Pandemic data (henceforth, GB&P) (Hensel et al. 2022). Around n = 110,000 participants (aged 18+) in 175 countries were recruited online through snowball sampling, 2 between March 20 and April 7, 2020. The sample is restricted to countries categorized as “developed economies” under the United Nations Development Programme classification 3 (UNDP 2024). The GB&P contains a near-complete sample of developed economies, excluding Iceland/Cyprus given absent mobility/pandemic data for the period. Accounting for missingness on key variables (<2 percent), the final analytic sample is n = 64,218 respondents in 34 countries (see Supplementary Appendix S.2 for the countries/respondents per country). Post-stratification weights are applied to improve representativeness and account for country sample size. Further external validity testing demonstrates strong sample similarity on key variables of the GB&P data when compared with randomly sampled data from countries where comparisons were undertaken (Hensel et al. 2022). Descriptives are available in Supplementary Appendix S.3.
The GB&P/CAUP datasets have complementary advantages. The GB&P data contains more countries (n = 34) to test hypotheses in almost all UNDP developed economies and more robustly estimate country-level effects. However, the GB&P data relied on convenience sampling, while being cross-sectional increases potential bias from unobserved time-invariant heterogeneity. While the CAUP is a sub-sample of developed economies (n = 10), its sample is designed to be representative, while its panel structure allows for more robust modeling.
Spatial immobility data are derived from Google Community Mobility Reports (Google LLC 2021). Pandemic data and country-level age composition are drawn from the Oxford COVID-19 Government Response Tracker (Hale et al. 2021). Social capital data are provided by the Legatum Institute (2023). Data on the gross national income (GNI) per capita are provided by the UNDP, and inequality data come from the World Inequality Database.
Measures
Individual-level psychological distress
Psychological distress is measured in both datasets using the Patient Health Questionnaire (PHQ) capturing depression symptomology. The GB&P measures PHQ-8 scores (Kroenke et al. 2009). Respondents were asked how frequently (four-category Likert scale of “not at all” to “nearly every day”), over the last two weeks, they experienced eight depressive symptoms (see Supplementary Appendix S.4 for details). Scores across eight indicators are averaged for a mean depression symptomology score (Alpha score: .86) (ranging from 0 to 3). The CAUP contains data on respondents’ PHQ-2 score, a short-form of the PHQ-8, based on the mean score of two symptoms: “feeling down, depressed, or hopeless” and having “little interest or pleasure in doing things” (Alpha score: .85) (range 0–3). Both PHQ-8 and PHQ-2 have been validated as severity measures for depressive disorders in general populations (Kroenke et al. 2009; Löwe, Kroenke, and Gräfe 2005). The two are correlated at r = .85; however, we also replicate the GB&P analysis using the PHQ-2 measure to test for consistency across datasets.
Country-level spatial immobility
As outlined, social disconnection stemming from social distancing/spatial immobility was a key pandemic stressor. Country-level spatial immobility data is taken from Google Community Mobility Reports, using location data from internet-connected devices to provide estimates of how much time people were spending in residential settings during the pandemic (Google LLC 2021). For each day, they calculate the change in the amount of time people spent at home, compared to an equivalent pre-COVID-19 baseline day. A baseline day is constructed as the median time spent at home for each day of the week using a 5-week baseline period (January 3-February 6, 2020). Change in time estimates are provided as positive/negative percentages. Separate daily estimates are available for each country. For each day/country, we calculate rolling averages based on the previous 7/14 days. Immobility data is matched to respondents based on their interview date/country of residence. Of course, immobility may vary across smaller areas (e.g., counties, municipalities) within countries, reducing the accuracy of the country-level measure to capture contextual connectivity in respondents’ more immediate spatial environments. Immobility data at smaller spatial scales could not be linked to either the GB&P/CAUP data. However, testing demonstrated that country-level immobility was very highly correlated with spatial immobility at smaller, sub-national scales. 4 We also replicate analyses using a government stringency index of pandemic-linked restrictions/responses for each day/country (Hale et al. 2021) (see below).
Country-level social capital
Prior research has largely explored whether individuals’ pre-pandemic social capital cushioned mental health during the onset of the pandemic. In measuring pre-pandemic social capital, their motivation is to test whether individuals going into the pandemic with higher social capital had more resources to draw on to buffer their mental health. We seek to replicate this approach by analyzing the potential protective role of countries’ pre-pandemic levels of social capital. 5
Pre-pandemic country-level social capital is derived from 2019 Legatum social capital measures (Legatum Institute 2023), with indicators selected based on their alignment with the dimensions of social capital from the pandemic/health literature (Bartscher et al. 2021; Elgar et al. 2020). The measures are based on external, nationally representative survey data (e.g., Gallup polling, World Values Study, European Values Survey) and aggregated to the country level (see Legatum Institute 2023 for details on methodology). Factor analysis (Promax rotation) returned two indices of social capital (see Supplementary Appendix S.5 for results): an index of personal and family relationships (bonding social capital): “if you were in trouble, do you have relatives or friends you can count on to help?,”“are you satisfied with opportunities to meet people and make friends?,” and “thinking about your life in general, my family give me positive support?” 6 ; and an index of civic/social participation (bridging social capital): “have you volunteered time to an organization in the past month?,”“have you helped a stranger or someone you didn't know who needed help in the past month?,” and “in the past month, have you voiced your opinion to a public official?” 7 Lastly, the cognitive dimension of social trust formed a distinct indicator: “generally speaking, would you say most people can be trusted?.” Correlations between the measures range from r = .22 and r = .27 in developed economies. We tested alternative, non-loading measures of social capital available in the data (institutional confidence, voter-turnout, charitable donations, and inter-household financial support) to examine which dimensions are most salient (see below). All measures are standardized.
Pathways: satisfaction with governments’ handling of the pandemic and loneliness
Loneliness is measured using the three-item short loneliness scale (UCLA-3), asking respondents how frequently they have “felt a lack of companionship,”“felt left out,” and “felt isolated from others” (“hardly ever” to “often”—mean score across three items) (Hughes et al. 2004). Positive attitudes toward the government's handling of the pandemic are measured using three indicators: “Are you satisfied or dissatisfied with the action of [country’s leader]?” (0 = completely dissatisfied; 10 = completely satisfied); “Are you satisfied with the way that the government is handling coronavirus?” (1 = Not at all satisfied; 4 = Completely satisfied); and “How satisfied are you with the way democracy works in your country?” (0 = completely dissatisfied; 10 = completely satisfied). Factor analysis (Promax-rotation) shows measures load on to a single index of government satisfaction during the pandemic. 8 The analysis of pathways is restricted to Waves 2, 4, and 5 of the CAUP data given the availability of measures across waves.
Covariates
Model estimates are adjusted for available individual-level covariates. In both the GB&P/CAUP data, this includes age, gender, education level, and household structure. In the GB&P data, models additionally control for monthly household income (equivalized), number of comorbidities, self-rated health, and marital status. In the CAUP data, models additionally control for perceived change in household income since the pandemic and number of COVID-19 symptoms in the last few weeks (see Supplementary Appendices S.1/S.3 for variable structures). The GB&P data do not contain individual-level indicators of social capital. However, the CAUP data contain individual measures of bridging and bonding social capital (trust) as well as generalized social trust. The index of bonding social capital is composed of three measures of people's trust in “family,”“neighbours” and “people you know personally.” 9 The index of bridging social capital is composed of three measures of people's trust in “people met for the first time,”“with different religious beliefs,” and “with a different nationality.” 10 These measures of trust, alongside the indicator of generalized social trust, map on to concepts of, and are frequently applied as indicators of, bridging and bonding social capital (Villalonga-Olives and Kawachi 2015). The individual-level social capital measures are only available for a limited number of waves—2, 4, and 5—and limited number of countries—they were not asked in the USA/Australia. Accordingly, we perform additional analyses on this reduced sample using these individual-level measures of social capital to perform tests for ecological fallacy in the proposed buffering role of country-level social capital 11 (see below).
At the country level, models adjust for 7-day rolling average COVID-19 death rate. In addition, several covariates are selected based on their association with social capital, mental health, and pandemic severity, including proportion aged 65 years and above, GNI (2020), and country income Gini coefficient (GINI) (2020) (Barrios et al. 2021; Elgar et al. 2020). Sensitivity of models to country-level covariates selection is tested by introducing additional covariates, including population density, life expectancy, inequality in life expectancy, average years in education, the Human Development Index (UNDP 2020), the Fragile States Index (The Fund for Peace 2020), proportion of GDP spent on healthcare, and governance and healthcare quality (see below).
Analytical plan and estimation strategy
The study undertakes three sets of analysis. Stage 1 analyses the CAUP panel data to examine the associations between country-level social capital, spatial immobility, and individuals’ psychological distress. First, the four waves of CAUP data are pooled and analyzed using three-level multilevel mixed regression models (Level 1: observations; Level 2: individuals; Level 3: countries), alongside survey-wave fixed effects, with country-level random coefficients for spatial immobility and survey wave. The lower n of level-2 units (n = 10 countries) may affect the precision of the model estimates. However, testing was conducted regarding the validity of applying multilevel models and is reported in the text. Interaction terms between country-level social capital and spatial immobility are modeled to test for the posited buffering (moderating) role of social capital.
Pooling the longitudinal CAUP data in the first instance allows us to estimate how time-invariant country-level covariates are associated with distress and explore the associations between levels of immobility, social capital, and distress. We then apply individual fixed-effects modeling to account for time-invariant unobserved heterogeneity to examine how changes in spatial immobility are associated with changes in distress. Under such specifications, time-invariant individual-/country-level characteristics drop out of the models, including country-level social capital. However, country-level social capital and immobility interaction terms continue to test whether the former moderates the association between the latter and distress. Given CAUP observations are clustered in both individuals and countries, fixed-effects models report errors clustered at the country level, but clustering at the individual level is also tested.
Stage 1 therefore aims to perform a more causally robust test of the buffering role of country-level social capital, albeit on a sub-sample of developed economies. The second stage explores the generalizability of findings from the CAUP analysis to developed economies in general via the near-complete sample of UNDP developed economies in the GB&P data. Given the data is cross-sectional, we estimate two-level multilevel mixed regression models with robust standard errors (individuals within countries), including random intercepts for countries, alongside survey-date fixed effects. A random-coefficient for immobility is included at the country level.
The third analytic stage aims to explore the potential pathways which may account for any observed buffering role of country-level social capital in the relationship between spatial immobility and individual-level psychological distress. To do so, the analysis returns to examining the longitudinal CAUP data and applies individual fixed-effects modeling to explore pathways of loneliness and government satisfaction during the pandemic. This analysis does not seek to undertake causal mediation analysis but instead aims to take a more exploratory analysis of the potential plausible pathways that may account for any observed moderating role of country-level social capital. Robustness/sensitivity tests for all analyses are discussed throughout.
Results
Spatial Immobility, Psychological Distress, and the Buffering Role of Country-Level Social Capital
The first analytic stage draws on four waves of the CAUP panel data, initially modeled using three-level multilevel mixed regression models predicting PHQ-2 (Table 1), to test if country-level social capital cushions the impact of spatial immobility on distress. Models contain all individual-level covariates although not shown (see Supplementary Appendix S.6/S.8 for full results).
Multi-Level Pooled Cross-Sectional and Longitudinal Modeling of Country-Level Social Capital, Spatial Immobility, and Individual-Level Psychological Distress (PHQ-2) (CAUP Data).
Note. Standard errors in parentheses; ML = multi-level; “-” signifies variable dropped due to time invariance; models contain full individual-level covariates (full results in Supplementary Appendices S.6 and S.8); CAUP Waves 2, 3, 4, and 5 (except Models 9-10, restricted to Waves 2, 4, and 5).
p < .05. **p < .01. ***p < .001 (two-tailed tests).
Model 1 (Table 1) demonstrates that increasing country-level spatial immobility is associated with higher distress (evidence supporting H1). Model 2 includes the three indices of country-level social capital to examine whether they have an overall negative association with distress; however, none of the associations were significant (evidence against H2). Models 3–5 then explore the key aim of the study: examining whether country-level social capital buffered the impact of restrictions on distress, by adding interaction terms between country-level immobility and each dimension of country-level social capital. There is no evidence that generalized trust (Model 3) or personal/family relationships (Model 4) moderate the association between immobility and distress. However, the interaction term between civic/social participation and immobility is significant and negative. Immobility thus had a significantly weaker positive association with distress in countries with stronger civic/social participation (Model 5: support for H3). When all interaction terms are modeled together (Model 6), the immobility*civic/social participation interaction term remains unchanged. As noted, the multilevel modeling approach may affect the precision of estimates due to the lower n of level-2 units (n = 10) (although see Elff et al. 2021). However, testing suggests the models were largely insensitive to different specifications or to more robust modeling approaches accounting for the smaller number of countries. 12
We next seek to more robustly test the observed moderating role of country-level social capital by applying individual fixed-effects modeling to remove unobserved time-invariant heterogeneity from the estimates (Models 7 and 8). Model 7 replicates Model 1, demonstrating spatial immobility has a significant, positive association with distress. Model 8 replicates Model 6, testing the proposed buffering role of social capital. Given country-level social capital indicators are time-invariant, their main coefficients drop out of the model. However, the interaction terms again demonstrate a significant negative interaction between immobility and civic/social participation (stronger support for H3).
These results thus provide evidence that country-level social capital, especially civic/social participation, buffered the impact of the pandemic on mental health. To understand the substantive implications of this moderating role, Figure 2 (based on Model 8, Table 1) plots the marginal effects of spatial immobility on psychological distress across levels of country-level civic/social participation (between the minimum [−1.66] and maximum (.95) values in the data). In low civic/social participation countries, immobility has a significant positive association with psychological distress: a 1-percentage point increase in immobility is associated with a 0.01-point increase in PHQ-2 (95 percent CI: 0.006, 0.014). However, as country-level civic/social participation increases, the positive immobility/distress association becomes weaker, until the highest levels of civic/social participation where immobility has no significant association with distress: a 1-percentage point increase in spatial immobility is associated with a 0.002-point increase in PHQ-8 (95 percent CI: −0.001, 0.005). This is a substantively significant finding. An increase from the fifth to 95th percentile of immobility (+29 percent immobility) in countries in the 95th percentile of social capital (high civic/social participation) is only associated with a 0.06-point increase in PHQ-2 (outcome ranging from 0 to 3). However, in countries in the fifth percentile of social capital (low civic/social participation), it is associated with a 0.3-point increase; a difference of 0.24 points in PHQ-2. This difference in increasing PHQ-2 is larger than the change coming from someone reporting their income “increased a lot” to “decreased a lot” on the self-reported five-item household income score (a 0.2-point increase in PHQ-2).
Marginal effects of country-level spatial immobility on psychological distress at different levels of country-level civic/social participation across 10 developed economies; CAUP data
There is a risk that the moderating role of country-level social capital does not represent a “true” contextual effect of living in countries with higher social capital but is a compositional effect, driven by individuals in countries with higher social capital themselves reporting higher social capital, that is, an ecological fallacy. As mentioned, the CAUP data contains individual-level indicators of social capital (trust indices of bridging and bonding social capital, alongside generalized social trust), but in a limited set of waves (2, 4, 5) and among a limited set of countries (measures were not asked in the USA/Australia) (see Supplementary Appendix S.8 for full details on the measures). We can use this restricted sample to test whether the key civic/social participation*immobility association remains significant after including these individual-level indicators of social capital and their attendant interaction terms with country-level immobility (see Supplementary Appendix S.8 for full results). Model 9 (Table 1), applying fixed-effects models, includes the three individual-level measures of social capital alongside the key country-level social capital*immobility interaction terms. Increasing individual-level indicators of bonding social capital and generalized trust, but not bridging social capital, are associated with decreasing psychological distress. In Model 10, we include interaction terms between the individual-level indicators of social capital and country-level spatial immobility. However, none of the interactions are statistically significant while crucially the country-level civic/social participation*immobility interaction term remains strong and significant. These findings provide evidence that the moderating role of country-level social capital is not solely compositional, driven by its association with respondents’ own self-reported social capital.
Further testing was undertaken to examine the robustness of the findings in Table 1. First, the civic/social participation*immobility interaction remained significant after testing alongside interactions between immobility and all individual-/country-level covariates in the model, including alternative country-level covariates, suggesting it is unlikely to be a spurious relationship (Supplementary Appendices S.9, S.10.1, S.10.2). The findings also remained consistent after including interaction terms between government pandemic-financial support and immobility (Supplementary Appendix S.11). Applying multiple imputation to account for within-/complete-case missingness returned substantively similar findings (Supplementary Appendix S.12). In addition, we examined whether sample-restriction decisions affected the findings, replicating key models but including Brazil and analyzing the full five-wave CAUP sample. Both tests returned substantively similar findings (Supplementary Appendix S.13). Consistent findings were returned when applying an index of stringency of government restrictions as well as contemporaneous/14-day rolling average measures of immobility (Supplementary Appendix S.13). Furthermore, when spatial immobility*social capital interactions are modeled alongside the government restrictions*social capital interactions, the former remains significant while the latter is rendered non-significant (Supplementary Appendix S.13). We also examined whether the buffering role of civic/social participation persisted after testing alongside alternative measures of social capital. The key moderating role of civic/social participation remained consistent (Supplementary Appendix S.14).
Generalizability of Country-Level Social Capital Buffering Across Developed Economies
The previous findings demonstrate evidence, using a more robust analytic approach, that country-level social capital moderates the positive association between increasing spatial immobility and psychological distress. However, the more limited sample of developed economies (n = 10) raises questions of how generalizable these findings are to developed economies as a whole. The second analytic stage seeks to test whether the associations between immobility, social capital, and distress demonstrated in the CAUP data can be observed across a near-complete sample of developed economies. To explore this question, we turn to analyzing the cross-sectional GB&P data (n = 34 developed economies) (Table 2). Table 2 reports the results from multilevel mixed regressions predicting mean PHQ-8 (models contain full individual-level covariates; see Supplementary Appendix S.15 for results).
Multilevel Cross-Sectional Modeling of Country-Level Social Capital, Spatial Immobility, and Individual-Level Psychological Distress (PHQ-8) (GB&P Data).
Note. Standard errors in parentheses; models contain full individual-level covariates (full results in supplementary-appendix: S.15).
p < .05. **p < .01. ***p < .001 (two-tailed tests).
Overall, the findings show a high degree of similarity to those using the longitudinal CAUP data. Higher immobility is associated with higher distress among respondents (Model 1) (support for H1). Country-level social capital has no overall association with distress (Model 2) (evidence against H2). Higher immobility does not have a significantly weaker association with distress in countries with higher trust or stronger personal/family relationships. However, it does have a significantly weaker association with distress in countries with stronger civic/social participation (Model 3) (support for H3). The substantive implications of this moderating relationship are also highly similar to the relationships observed in the CAUP data. Immobility has a strong positive association with distress in low social capital countries: a 1-percentage point increase in immobility is associated with a 0.02-point increase in PHQ-8 (95 percent CI: 0.01, 0.03). However, the positive immobility/distress association becomes increasingly weaker and less significant as social capital increases, resulting in no significant association with distress in high social capital countries: a 1-percentage point increase in immobility is associated with a 0.001-point decrease in PHQ-8 (95 percent CI: −0.009, 0.006).
These findings thus provide support for the generalizability of the buffering role of social capital, previously identified in the CAUP data, to developed economies in general. Robustness tests were again undertaken. The significant civic/social participation*immobility association remained after testing alongside interactions between immobility and all individual-/country-level covariates (Supplementary Appendices S.16 and S.17.1–S.17.3). Given PHQ-8 is positively skewed, additional analyses tested a logged PHQ measure and a binary measure (0 = PHQ-8 <10; 1 = PHQ-8 ≥9 points, indicating moderate/severe depression symptoms). The findings remained substantively consistent (Supplementary Appendix S.18). We also replicated the key findings using the PHQ-2 measure, for greater comparability with the CAUP analysis, which returned substantively identical findings (Supplementary Appendix S.18). The moderating role of civic/social participation also remained when tested alongside alternative measures of social capital (Supplementary Appendix S.19). The findings are also not sensitive to alternative specifications of developed economies, being observable among International Monetary Fund (IMF) advanced economies and World Bank high-income countries (Supplementary Appendix S.20).
Lastly, how country-level social capital moderates the relationship between immobility and psychological distress could depend on the mix of different types of social capital. We applied Ward's hierarchical clustering method to generate 2-, 3-, and 4-cluster profiles (n = 34 countries) of country-level social capital (based on levels of trust, personal/family relationships, civic/social participation), to examine whether different profiles of social capital exhibited different protective effects (Supplementary Appendix S.21). We also tested higher-order interaction terms between all three dimensions of country-level social capital to examine whether any cushioning role of civic/social participation was contingent on levels of social trust and/or personal/family relationships (Supplementary Appendix S.22). Both sets of tests suggest country-level civic/social participation is the most important form of social capital for moderating the positive immobility/distress relationship.
Pathways of Buffering
The findings thus far demonstrate more causally robust evidence that country-level social capital moderated the impact of increasing spatial immobility on psychological distress (first analytic stage), and that this moderating role appears generalizable to developed economies as a whole (second analytic stage). The third analytic stage undertakes an exploratory analysis of the potential plausible pathways which may account for why country-level social capital moderates the immobility/distress relationship. To do so, we return to analyzing the longitudinal CAUP data and examine two pathways (loneliness and attitudes toward the government's handling of the pandemic) through which country-level social capital might moderate the association between immobility and distress. As stated, analyses are restricted to Waves 2, 4, and 5. Table 3 reports the results of individual fixed-effects models, containing full individual-/country-level covariates (see Supplementary Appendix S.23 for results).
Longitudinal Modeling of Buffering Pathways: Individual-Level Loneliness and Government Satisfaction (CAUP Data).
Note. Standard errors in parentheses; “-” signifies variable dropped due to time-invariance; models contain full individual-/country-level covariates (full results in Supplementary Appendix S.14); restricted to CAUP Waves 2, 4, and 5.
p < .05. **p < .01. ***p < .001 (two-tailed tests).
Model 1 (Table 3) replicates the key finding that civic/social participation negatively moderates the positive immobility/distress relationship (as observed in Model 8, Table 1). We then tested whether immobility was associated with smaller increases in loneliness or smaller decreases in government satisfaction in countries with higher social capital. Model 2 demonstrates how increasing immobility is linked with increasing loneliness but has a significantly weaker positive association in countries with higher civic/social participation. Interestingly, immobility also has a weaker positive association with loneliness in countries with higher generalized trust or stronger personal/family relationships (evidence for H4a). Model 3 demonstrates how increasing immobility is also associated with more negative attitudes toward the government's handling of the pandemic but that this negative association is significantly weaker in countries with higher civic/social participation or with stronger personal/family relationships 13 (Model 3) (evidence for H5a).
We next explore whether loneliness and attitudes toward the government's handling of the pandemic might help account for the weaker positive association between immobility and distress in countries with higher social capital. Model 4 replicates Model 1 (predicting distress) but includes loneliness to test how far its inclusion reduces the immobility*civic/social participation interaction term. Loneliness predicts more distress (evidence for H4b). In addition, the immobility*civic/social participation interaction term is reduced by 21 percent (comparing Model 4 to Model 1) (evidence for H4c). Model 5 then removes loneliness but includes government satisfaction. Positive attitudes toward governments’ handling of the pandemic are associated with less distress (evidence for H5b). Furthermore, the immobility*civic/social participation interaction term is reduced by 26 percent (comparing Model 5 to Model 1) (evidence for H5c). Finally, Model 6 includes both loneliness and government satisfaction, which reduces the immobility*civic/social participation interaction term by 44 percent and from p < .001 to p < .05 significance. These findings suggest nearly half of the weaker positive relationship between immobility and distress in higher social capital countries can be accounted for by its association with loneliness and government satisfaction. However, a significant immobility*civic/social participation interaction term remains.
Given both the loneliness/government satisfaction pathways and psychological distress are self-reported contemporaneously, the observed relationships could solely be a consequence of reverse causality. 14 Additional analysis sought to provide some evidence of temporal precedence by replicating the above pathways analysis but using a lagged (t-1) measure of immobility, contemporaneous (t0) measures of loneliness/government satisfaction, and a leading (t+1) measure of psychological distress. The findings closely mirror the findings reported earlier, providing some initial support for the study's posited causal ordering (see Supplementary Appendix S.24 for full details). However, challenges remain preventing strong claims of causal identification (discussed further below).
Discussion and Conclusion
This study used the COVID-19 pandemic as a lens through which to explore whether different dimensions of country-level social capital can protect psychological wellbeing during large-scale, (trans)national crises. The findings demonstrate that as spatial immobility increased in a country, individuals’ psychological distress also increased. However, this positive association was significantly weaker in countries with higher average levels of social capital, in particular levels of civic/social participation, suggesting country-level social capital exerted a buffering role on individuals’ distress. The findings also demonstrate the existence of two plausible pathways that help account for this moderating relationship. Loneliness and dissatisfaction with governments’ handling of the crisis were linked with higher distress. However, under conditions of increasing spatial immobility, individuals in higher social capital countries reported smaller increases in loneliness and smaller decreases in satisfaction with their government's handling of the pandemic (although this path analysis does not identify strong causal evidence—see below). Yet not all dimensions of social capital mattered equally. Higher civic/social participation especially moderated the immobility/distress relationship. However, neither personal/family relationships nor generalized trust moderated the immobility/distress relationship (although they did variably moderate the association between immobility and loneliness/government satisfaction). Longitudinal analysis (in a sub-sample of developed economies) provides more causally robust evidence, while cross-sectional replication in a near-complete sample of UNDP developed economies demonstrates the generalizability of findings.
To our knowledge, this is the first study to show that dimensions of national-level social capital may foster societal resilience for mental health during major crises. In fact, levels of civic/social participation appear more important for moderating how the pandemic was associated with worsening mental health than other country-level factors, for example, GNI, share of GDP spent on healthcare, levels of human development,governance and healthcare quality, or inequality (Supplementary Appendix S.9; Supplementary Appendix S.17.1–S.17.2). These findings thus demonstrate that, alongside individual social capital (Han and Chung 2023; Laurence 2025b; Laurence et al. 2024; Sato et al. 2022), indicators of country-level social capital could also play a buffering role for mental health during the pandemic. Importantly, the moderating role of country-level civic/social participation does not appear to be solely a compositional effect stemming from moderation by individual-level social capital. The immobility*civic/social participation moderation remains strong and significant even after accounting for potential buffering by indicators of individual-level social capital. Accordingly, social capital among all members of a society appears to matter (a contextual effect). Indeed, in this cross-national study, individual-level indicators of social capital did not appear to moderate the relationship between peri-pandemic changes in spatial immobility and psychological distress. 15 The study's findings also complement research demonstrating how country-level civic/social participation in particular appeared to protect physical health during the pandemic, such as fewer COVID-19 deaths (e.g., Elgar et al. 2020), suggesting a particular effectiveness of civic/social participation for protecting health in general during the pandemic.
This study contributes to the wider health/social capital literature by demonstrating that, alongside individuals’ own levels of social capital, macro-level social structures (national levels of social capital) may also foster protection during crises and could be important for societal resilience against future (trans-)national emergencies (e.g., climate change). However, the study also highlights how different dimensions of social capital can matter differently at different levels. The study demonstrates how contextual civic/social participation in particular is associated with less psychological distress under conditions of rising immobility. This provides further evidence that during large-scale crises, civic dimensions of contextual social capital may be especially effective for protecting health and wellbeing, while dimensions of closer, bonding social capital or trust appear less effective (at least when measured at the country level) (Elgar et al. 2020; Hawkins and Maurer 2010; Lim and Laurence 2015; Weil et al. 2012). This contrasts with evidence that individual-level stronger-tie (family/friend) social capital was also important for protecting mental health during the pandemic (Laurence 2025a). One possibility for this difference across levels is that even in a country with high levels of bonding social capital, network closure will likely be high, inhibiting wider forms of collective organization or shared belonging, which may be more feasible with higher country-level civic/social participation and bridging social capital.
The study also presents novel theoretical insights into the pathways of social capital buffering. While pathways of social support/isolation have been previously documented (e.g., Aldrich and Meyer 2014; Louie et al. 2023), we identify a key political-attitudinal pathway: satisfaction with governments’ handling of crises. Potentially, country-level social capital engenders a greater degree of trust in, and acceptance of, public health measures, stemming from a commitment to the common good or shared feeling of responsibility to society, which may reduce stress/anxiety, in turn, protecting mental health. Interestingly, the moderating role of country-level civic/social participation was not driven by greater country-level confidence in government/politicians in general (see Footnote 9), demonstrating the particular importance of the civic/social participation dimension of social capital. One explanation is that during crises, more formal social networks, developed through civic/social participation, are more resilient and better able to mobilize help/sustain involvement, forming a more robust “institutionalised altruism” (Lim and Laurence 2015). Alternatively, civic engagement may engender a more superordinate, civic identity, in which adherence to, and support for, public measures is stronger. An additional analysis examined whether a country's mix of different dimensions of social capital had different implications for buffering (Supplementary Appendix S.21). However, levels of civic/social participation remained the key driver of buffering (although more nuanced profiles of social capital mix may emerge in more global samples of countries).
Despite observing that country-level social capital moderates the positive association between immobility and distress, it exhibited no overall (main) association with psychological distress. This may be partly a consequence of the mental health outcome examined (PHQ). Prior cross-national research suggests country-level social capital is positively associated with subjective wellbeing, such as life satisfaction (e.g., Almakaeva, Moreno, and Wilkes 2021), but few cross-national tests have analyzed clinically validated outcomes. Meanwhile, studies measuring sub-national contextual social capital generally show mixed evidence of main effects on validated mental health outcomes (see Kemppainen and Timonen 2024), and several show evidence of a buffering association even with absent main effects (e.g., Stafford et al. 2008; Takagi et al. 2013). Another explanation is that during the pandemic, widespread anxiety and disruption may have elevated baseline distress across countries, dampening any overall positive effect of national social capital on mental health. Instead, it was only during more severe-stress conditions (large-scale reductions in mobility) that the buffering quality of social capital activated to protect mental health. Indeed, the only other cross-national, pandemic study to look at country-level social capital found it was weakly but positively associated with distress, suggesting country-level social capital may have operated differently to non-pandemic times (Lee et al. 2021).
Alongside the insights gleaned, this study has limitations. As discussed, despite being longitudinal, the CAUP data contained only n = 10 countries, potentially affecting the precision of model estimates. The GB&P data, despite sampling all developed economies, was also not based on random sampling, potentially affecting external validity. While these were the best available data gathered during the pandemic to test our hypotheses, there remains the risk of generalizing from these data to all developed economies. Furthermore, while the CAUP longitudinal modeling addressed bias from time-invariant unobserved heterogeneity associated with immobility and distress (time-variant measures), country-level social capital (a time-invariant measure) remains susceptible to such bias. The CAUP analysis may also be biased by time-variant unobserved heterogeneity. Both sources of bias are present in the GB&P cross-sectional analysis.
The analysis of potential pathways also has several important limitations. First, there is a risk of endogeneity in the mechanisms analysis, as both the mediators and psychological distress are self-reported and measured contemporaneously. Second, post-treatment bias may arise from the inclusion of control variables (e.g., financial stability) that are themselves influenced by immobility. Third, the pathway analysis may be affected by intermediate confounding (i.e., unobserved post-predictor variables related to both the mediators and the outcome) which could render the mediators as colliders and introduce bias. Fourth, direct/indirect associations may be biased if the posited mechanisms (loneliness and government satisfaction) also interact with immobility in shaping distress (although tests of their contemporaneous relationships suggest they do not). While some initial testing provided partial evidence in support of the framework's posited causal ordering (Supplementary Appendix S.23), future research, using data that would allow for more causally robust approaches (e.g., cross-lagged panel modeling), can undertake stronger tests of causal ordering. Ultimately, the pathway analysis is largely exploratory and designed to illustrate plausible pathways that may be in operation rather than suggest evidence of causal mediation.
More broadly, part of the spatial immobility/distress relationship may be driven by worsening mental health increasing levels of immobility (reverse causality). The core findings held when using the country-level stringency of government pandemic ordinances (Supplementary Appendix S.12), which is likely less shaped by societal mental health, while testing of the temporal precedence of the relationships provides somewhat stronger evidence that at least part of the immobility/distress relationship stems from the former shaping the latter (Supplementary Appendix S.24 for full details). Still, an effect of distress on immobility is likely present, which may bias the findings. Caution should therefore be applied to the causal identification and ordering of all results. In addition, future research will seek to test the individual-level mechanisms (e.g., perceived/actual social support, sense of belonging/shared identity, etc.) posited to account for why living in a country with higher social capital buffers how immobility shapes loneliness and government satisfaction, in turn, shaping psychological distress. These measures were unavailable in the current data.
A further limitation is that measuring pre-pandemic (2019) social capital, while reflecting current approaches in the pandemic literature, may also introduce bias if country-level social capital shifted unequally across countries during the pandemic. However, consistent cross-national estimates of country-level social capital based solely on data gathered during the pandemic (2020–2021) for the countries analyzed here were unavailable. The country-level social capital measures themselves are also based on aggregated survey responses (e.g., Gallup data), which themselves may contain biases (e.g., social desirability). In addition, the CAUP data did not contain a pre-pandemic measure of distress. Thus, the study only tests whether social capital cushioned changing patterns of immobility during the pandemic (although significant changes in immobility did occur over the survey period).
The immobility data also has limitations given it estimates population movement based on aggregated, anonymized location data, gathered from people with internet-connected devices. First, this may bias the mobility data toward certain demographic/socioeconomic groups, for example, younger, urban, higher socio-economic status groups, who possess such devices. Second, there may have been significant variation in patterns of spatial immobility across smaller areas within countries, which weakens the ability of this country-level measure to capture individuals’ personal experiences in their more immediate environments. Third, there are challenges to how accurately these data reflect actual patterns of mobility given not everyone owns such devices, and not everyone carries their devices when leaving the home. In addition, these data are only a proxy for social contact and do not provide aggregate levels of actual contact across areas (Prestige et al. 2025). Despite these limitations, mobility data did predict transmission rates during the pandemic (Nouvellet et al. 2021). In addition, the mobility data are highly correlated with the stringency of government pandemic ordinances (r > .75 in the data), while replicating analyses using the government policy stringency index returned consistent findings (Supplementary Appendix S.12).
As discussed, there is also the potential risk that the findings do not represent a “true” contextual effect but are a product of ecological fallacy, where country-level social capital is acting as a proxy for individual-level social capital. Additional analysis (using a more limited set of CAUP waves/countries), in which we included three indicators of individual-level social capital, demonstrated that the key country-level social capital/spatial immobility interaction term remained significant even after accounting for interaction terms between immobility and the individual-level proxies of social capital (Supplementary Appendix S.7). This provides somewhat stronger evidence that the country-level social capital moderation represents a contextual “buffering effect.” However, given we do not have identical individual-level measures of our country-level social capital indicators, in particular, the absence of an individual-level measure of civic/social participation, there remains a risk the findings are compositional. Furthermore, it should be noted that levels of contextual social capital can differ significantly across space within countries, which may introduce measurement error to the estimates. Unfortunately, no longitudinal or large-coverage cross-national data were available during the pandemic that contained sub-national spatial identifiers to examine the potential buffering role of regional social capital. Future work that is able to compare the potential protective role of country-level social capital with sub-national, regional levels of social capital will shed light on potential differences in the efficacy of social capital at different spatial scales.
Lastly, due to country coverage in the CAUP longitudinal data, this study focused on developed economies. These findings may not necessarily be generalizable to developing economies. Potentially, weaker healthcare infrastructure and greater economic instability in developing countries may have been a stronger driver of mental health outcomes than public health measures during the pandemic. Developing countries may also rely more on closer ties/familial bonds for support, while developed economies generally have more established civic networks, in which case family/friend dimensions of social capital may be more important in developing economies during crises. A full treatment of these potential differences is beyond the scope of this study (especially given the lack of longitudinal pandemic-period data for developing economies). However, future research will seek to test the generalizability of findings to developing economies and examine potential country-level drivers of heterogeneity in the buffering role of country-level social capital.
In sum, this study provides novel insights into the role of macro-level social structures for protecting mental health during crises, shedding light on the factors which might foster societal resilience during (trans)national emergencies. In particular, the results suggest that building stronger levels of civic/social participation could be a key means of safeguarding mental health during future crises. For example, governments could support and fund initiatives that cultivate volunteering, mutual aid, and citizen involvement in decision-making processes to help foster the development of civic organizations and networks across society. Increasing involvement of residents in local policy-making and implementation, and devolving greater power to communities, may also build stronger civic engagement across countries. Civic organizations could also be included in crisis-preparedness strategies to support their involvement in information dissemination, providing targeted support to vulnerable populations, and to mobilize networks to better coordinate collective action (e.g., organizing volunteers, pooling resources) (Aldrich and Meyer 2014; Putnam 2000). The results also demonstrate the important nuance in the role played by social capital for protecting mental health and that it can matter differently in different contexts and at different levels.
Supplemental Material
sj-docx-1-smh-10.1177_21568693251405194 – Supplemental material for The Protective Role of Country-level Social Capital for Mental Health During Large-scale Crises
Supplemental material, sj-docx-1-smh-10.1177_21568693251405194 for The Protective Role of Country-level Social Capital for Mental Health During Large-scale Crises by James Laurence and Bill Calvey in Society and Mental Health
Footnotes
Acknowledgements
The authors would like to thank Elie Michel for their help with access to the Citizens Attitudes under COVID-19 data.
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Economic and Social Research Council: [Grant Number ES/W00349X/1].
Supplemental Material
Supplemental material for this article is available online.
Notes
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
