What shapes generalized and particularized trust worldwide? Population, inequality, migration and media in a global perspective

Trust is a cornerstone of social capital and a vital determinant of democratic stability, effective governance, and sustained economic growth (Fukuyama, 1995; Knack and Keefer, 1997). High-trust societies tend to exhibit more inclusive institutions and robust markets, while low-trust societies often grapple with inequality, corruption, and political fragmentation (Rothstein and Uslaner, 2005; Uslaner, 2002). As such, understanding the sources of trust and whether they are universal or context-specific remains a central concern across the social sciences. Most research has focused on generalized trust, typically measured as trust in “most people.” This form of trust promotes cooperation among strangers, collective action, and democratic legitimacy (Rothstein and Stolle, 2008). Yet individuals also rely heavily on particularized trust, such as trust in family members and neighbors, which may either complement or substitute for generalized trust (Delhey and Newton, 2005; Yamagishi and Yamagishi, 1994). Despite its importance, comparative evidence on the determinants of particularized trust remains sparse, and studies rarely analyze both forms of trust systematically within the same empirical framework. The literature has identified several structural, cultural, and institutional drivers of trust. Economic inequality tends to erode trust by segregating social groups (Uslaner and Brown, 2005), while high crime rates, particularly homicide, undermine social cohesion by heightening insecurity (Rosenfeld et al., 2001). Modernization theory suggests that economic development strengthens trust by reducing existential insecurity and fostering institutional effectiveness (Delhey and Newton, 2005). Migration and ethnic diversity, however, remain contested. Putnam (2007) argues that heterogeneity reduces trust as communities “hunker down,” whereas others find that the effect depends on institutional inclusiveness and welfare support (Dinesen and Sønderskov, 2015). More recently, media environments have emerged as critical influences: social media use has been linked to polarization and lower trust (Allcott and Gentzkow, 2017; Bail et al., 2018), while traditional media may foster common narratives that sustain cohesion (Strömbäck et al., 2020).

Despite these advances, key gaps remain. First, few studies have examined how structural factors such as inequality, homicide, and fertility influence both generalized and particularized trust across continents. Second, comparative work on migration and media influences remains limited, especially beyond the European context. Third, most prior studies rely on individual-country analyses and rarely test the robustness of their findings across weighting schemes or regional samples. Our study addresses these gaps by applying unweighted OLS as well as models weighted by population and GDP, and by analyzing different regions such as Europe and advanced economies (AEs), Asia, Latin America (LATAM), and LAFMENA (Latin America, Africa, and Middle East and North Africa - MENA). The regional grouping strategy in this study is guided by both theoretical considerations and data constraints. Europe and other AEs, including countries such as the United States, Canada, Australia, and New Zealand ^[1], are grouped together. We group them following established cross-national research demonstrating that these countries converge across four key dimensions relevant to trust formation: institutional quality, as reflected in high governance indicators including rule of law and control of corruption (OECD, 2021); media environments, characterized by high journalistic professionalism and press freedom (Hallin and Mancini, 2004); levels of economic development, evidenced by comparable GDP per capita and welfare state structures (IMF, 2024); and social structures. These commonalities provide a coherent basis for treating Europe and AEs as a unified regional group. For developing regions, Latin America (LATAM) is analyzed separately where feasible, given its relatively larger and more consistent data coverage. However, Africa and the Middle East and North Africa (MENA) individually suffer from sparse observations across survey waves and key variables, which limits statistical power and reliability of estimates when treated independently. To address this limitation, these regions are combined with Latin America into a broader LAFMENA grouping in some specifications. This aggregation ensures a sufficiently large and balanced sample for robust econometric analysis while still capturing shared structural characteristics common across these regions, such as higher levels of informality, institutional constraints, and socio-economic volatility. Thus, the grouping reflects a pragmatic balance between conceptual comparability and empirical feasibility. Our study is based primarily on the joint European Values Survey–World Values Survey (EVS-WVS) time series, covering 107 countries from 1981 to 2023. To capture contextual conditions, we merge these survey data with external country-level indicators including homicide rates, inequality (GINI), GDP, and population sourced from the World Bank Data, Federal Reserve Economic Data (FRED), national statistical agencies, and IndexMundi. This yields a large and diverse sample of 441,481 individuals across 107 countries, spanning Europe, other AEs, Asia, Latin America (LATAM), Africa, and the Middle East and North Africa (MENA). This paper contributes to the literature in three main ways. First, it advances a comparative framework that jointly analyzes generalized and particularized trust, revealing important divergences in their determinants. Second, it incorporates underexplored structural and informational factors such as fertility, homicide, income inequality, and media use into global and regional analyses, showing how their effects vary across continents and regions. Third, by employing multiple weighting schemes and regional disaggregation, it tests the robustness of findings across different specifications and contexts, offering nuanced insights for both theory and policy.

The remainder of the paper proceeds as follows. Section 2 is for the literature review. Section 3 describes the methodology and data. Section 4 presents the findings. Section 5 discusses conclusions, implications, and limitations.

Trust is widely recognized as a cornerstone of social capital and a key determinant of democratic stability and economic performance (Fukuyama, 1995; Knack and Keefer, 1997). A central conceptual distinction in the literature separates generalized trust, defined as trust extended to strangers and unknown others, from particularized trust, which is confined to close-knit groups such as family, friends, and neighbors. Generalized trust enables cooperation beyond immediate networks and underpins inclusive institutions and market efficiency, whereas particularized trust reinforces local cohesion but may constrain broader social integration when it becomes dominant (Fukuyama, 1995; Knack and Keefer, 1997). Importantly, existing studies suggest that the determinants of these two forms of trust differ: generalized trust is more sensitive to institutional quality, fairness, and corruption, while particularized trust tends to persist even under weak governance and high inequality. Despite this foundational distinction, much of the empirical literature has examined trust as a unified concept. Two competing perspectives dominate explanations of trust formation. Culturalist approaches emphasize the persistence of historical legacies, civic norms, and associational life (Putnam, 2000; Guiso et al., 2008), while institutionalist accounts highlight the role of impartial and effective institutions in fostering fairness and reciprocity (Rothstein and Stolle, 2008; Rothstein and Uslaner, 2005). Uslaner (2002) further argues that trust has a moralistic component that is formed early in life but shaped over time by inequality and institutional performance, while Newton (2001) stresses the interaction between culture and institutions. Recent empirical work extends the relevance of trust to policy domains. Vu (2027) shows that both generalized and institutional trust significantly shape beliefs about climate change and support for environmental taxation, with institutional trust playing a stronger role in policy acceptance. Likewise, Vu and Phuong (2026) demonstrate that generalized trust promotes firms' adoption of climate mitigation practices in Vietnam. However, while these studies highlight the importance of trust, they largely do not systematically compare its distinct forms across global contexts. This leaves an important gap in understanding whether and how the drivers of generalized and particularized trust diverge: a gap this paper addresses through a unified comparative framework.

A large body of research links economic inequality to diminished trust. Inequality is argued to create segregated social environments that weaken reciprocity and shared norms (Uslaner and Brown, 2005), with empirical studies confirming a negative relationship between inequality and trust levels (Alesina and La Ferrara, 2002; Bjørnskov, 2007). Yet, the effect is not universally consistent: in some transitional settings, inequality may be perceived as temporary and thus less damaging to social cohesion (Norris and Inglehart, 2019). Similarly, crime, particularly violent crime, erodes trust by increasing perceptions of insecurity. Homicide rates are strongly negatively associated with trust (Rosenfeld et al., 2001), while higher trust societies tend to exhibit lower crime levels (Jing and Bond, 2015). Nonetheless, contextual factors such as kinship networks may buffer these effects (Delhey and Newton, 2005). Existing studies, however, typically assess these structural factors in isolation or focus primarily on generalized trust. Less attention has been given to how inequality and crime may differentially affect generalized versus particularized trust, or how these relationships vary across regions. This paper contributes by jointly analyzing these structural determinants in a global and regionally disaggregated framework.

Modernization theory posits that economic development fosters trust by reducing insecurity and strengthening institutions (Delhey and Newton, 2005). Empirical evidence confirms that higher GDP per capita is associated with increased generalized trust (Knack and Keefer, 1997; Stevenson and Wolfers, 2011). Demographic factors also shape trust dynamics. Urbanization may weaken localized trust due to anonymity, while potentially supporting broader generalized trust under strong institutional conditions (Putnam, 2007). High fertility rates and population growth are argued to strain public resources and reduce social cohesion, thereby lowering trust (Wilkinson and Pickett, 2009). Yet, demographic variables such as fertility remain underexplored in cross-country trust research, particularly in combination with other structural determinants. By incorporating these factors into a comprehensive empirical framework, this paper extends the scope of trust research beyond conventional economic indicators.

The relationship between migration, diversity, and trust remains contested. Putnam (2007) argues that ethnic diversity reduces trust in the short run, while Dinesen and Sønderskov (2015) find negative effects of immigration where integration is weak. Conversely, supportive institutional frameworks can mitigate or even reverse these effects (Kesler and Bloemraad, 2010). Goodhart and Vu (2025), using World Values Survey data across 52 countries, add further nuance by showing that migration can positively relate to trust in government up to a threshold, beyond which the effect reverses. Their findings also highlight the importance of economic growth and identify a “trust paradox” across regime types. Despite these findings, limited attention has been paid to systematic cross-regional comparisons of how migration interacts with broader structural and informational factors in shaping different forms of trust. This paper addresses this gap through regional disaggregation and robustness analysis across specifications.

The media environment has emerged as a crucial yet complex determinant of trust. Social media has been associated with polarization, misinformation, and declining trust (Allcott and Gentzkow, 2017; Bail et al., 2018; Barberá et al., 2018), while traditional media may foster shared narratives and social cohesion. At the same time, trust in traditional media has weakened significantly, driven by competition from partisan and non-mainstream outlets and by political attacks on journalistic credibility (Strömbäck et al., 2020). Empirical evidence highlights this fragility. Trust in mass media in the United States declined from 68% in 1968 to 32% in 2016 (Jones, 2018), with only a small minority expressing high confidence (Guess et al., 2018). Cross-nationally, about half of respondents report trusting news most of the time (Newman et al., 2019), though structural changes in media ecosystems continue to challenge credibility (Hanitzsch et al., 2018; Van Aelst et al., 2017). The proliferation of alternative media, political disintermediation, and disinformation further complicate the trust landscape (Ladd, 2012; Groshek and Koc-Michalska, 2017; Egelhofer and Lecheler, 2019). Recent large-scale evidence by Goodhart and Vu (2026), based on 97 countries, shows consistent patterns: traditional media consumption is positively associated with institutional trust, while social media use is negatively associated, particularly with political trust. Nonetheless, comparative research examining how different media types shape generalized versus particularized trust remains limited. This paper contributes by integrating media variables into a broader model of trust and testing their effects across regions and specifications.

The literature shows that the deleterious effects of corruption extend beyond institutional trust to erode the broader fabric of social capital, defined as generalized trust and norms of reciprocity (Putnam, 2000). Research shows that perceptions of corruption have a negative influence on generalized trust, a finding supported by cross-national studies and experimental evidence in which exposure to corruption information lowered trust levels (Stulhofer, 2004; Rothstein and Eek, 2006). This relationship is reinforced by insights from criminology, which demonstrate that victimization creates a psychological spillover effect; just as crime victims often generalize their distrust from the individual perpetrator to the wider society, citizens who observe corrupt political actors may irrationally attribute these negative behaviors to the general population (Rountree and Land, 1996; Blount, 1995).

Taken together, the literature identifies a wide range of determinants of trust: cultural, institutional, structural, demographic, and informational. However, it remains fragmented in three key respects. First, most studies do not jointly analyze generalized and particularized trust within a unified framework. Second, structural and informational variables such as fertility, homicide, inequality, and media use are rarely examined together, particularly in global and regional comparative settings. Third, limited attention has been paid to regional heterogeneity across model specifications.

This paper addresses these gaps by:

1. Developing a comparative framework that distinguishes and jointly analyzes generalized and particularized trust;

2. Integrating underexplored structural and informational determinants in a global and regional context;

3. Employing multiple weighting schemes and regional disaggregation to ensure robustness and uncover contextual variation.

In doing so, it offers a more comprehensive and nuanced understanding of the drivers of two distinct types of trust (generalized vs particularized), contributing to both theory and policy.

This study builds on and extends the empirical frameworks developed by Goodhart and Vu (2025, 2026), which integrate individual-level survey data with country-level structural indicators to analyze the determinants of trust. These studies provide a robust foundation for modeling trust as a function of both micro-level characteristics (e.g. demographics and behavior) and macro-level conditions (e.g. economic, institutional, and informational environments). In particular, Goodhart and Vu (2026) demonstrate that media consumption patterns are systematically associated with variations in trust, while Goodhart and Vu (2025) emphasize the importance of broader structural factors such as macroeconomic performance, migration, and political context. Building on these insights, this paper adopts a unified empirical framework that jointly incorporates informational, structural, and demographic determinants of trust. However, it departs from the existing literature in an important respect: rather than treating trust as a single aggregated concept, this study explicitly distinguishes between generalized trust and particularized trust, allowing a systematic comparison of their determinants across both global and regional contexts. This distinction enables a more precise test of whether the drivers of trust differ across its two core dimensions: an issue that has been largely overlooked in prior empirical work. To test the determinants of generalized trust and particularized trust, we use Model (1). We use OLS estimation with standard errors clustered at the country level. In addition, we weight the regressions by the GDP and total population of the respondent's country of residence.

where TRUST_i​,c,t denotes the level of interpersonal trust for individual i in country c at time t. In this study, we distinguish between two types of interpersonal trust: generalized trust and particularized trust. Estimating the model separately for these two dependent variables constitutes a key innovation of this study, enabling a direct comparison of how different factors shape distinct forms of social trust. $SMEDIAi,c,t$ captures whether respondents obtain daily information about events in their country and the world through social media sources (e.g. Facebook, Twitter). $TRDMEDIAi,c,t$ indicates whether respondents rely on traditional media such as daily newspapers, television, and radio for daily information on national and global events. The inclusion of social media (⁠$SMEDIAi,c,t$⁠) and traditional media (⁠$TRDMEDIAi,c,t$⁠) is directly motivated by Goodhart and Vu (2026), who show that media environments play a central role in shaping trust outcomes. Social media exposure has been linked to misinformation, polarization, and lower trust, while traditional media is associated with higher levels of institutional trust. Incorporating both variables allows us to test whether these informational channels differentially affect generalized versus particularized trust. By embedding media variables within a broader framework, this study advances the literature by examining whether the informational environment interacts with structural conditions to produce distinct trust outcomes across regions. $NETMIGRi,c,t$ represents the net migration level of the country in which the respondent resides. The inclusion of net migration (⁠$NETMIGRi,c,t$⁠) follows Goodhart and Vu (2025), who find that migration has a nonlinear relationship with trust and varies across contexts. Migration is particularly relevant because it captures changes in social composition and diversity, which have been shown to affect social cohesion and trust formation. $INDIVIDUALi,c,t$ is a vector of control variables capturing characteristics for individual i in country c at time t (e.g. age, employment status, rural vs urban, etc). Prior studies show that women tend to exhibit higher trust (Delhey and Newton, 2005; Foster and Frieden, 2017), trust increases with age (Christensen and Lægreid, 2005), and rural residents often report higher trust due to stronger local ties (Brinkerhoff et al., 2018; Wang and You, 2016). Employment status is included to capture economic insecurity and expectations of institutional performance (Christensen and Lægreid, 2005; Zhao and Hu, 2017). $COUNTRYi,c,t$ is a vector of country-level controls (e.g. the total population, GDP per capita, etc). $μi,c,t$ denotes the error term. Country fixed effects (FE) and year fixed effects (FE) are included to account for unobserved heterogeneity across countries and over time.

While the empirical framework builds on established approaches, its contribution is threefold. First, it explicitly differentiates between generalized and particularized trust, enabling a direct comparison of their determinants within a single unified model. Second, it integrates informational, structural, and demographic variables, including underexplored factors such as fertility, homicide, and media use, into a comprehensive global analysis. Third, by applying alternative weighting schemes and regional disaggregation, the study tests whether observed relationships are robust across different contexts and specifications. Together, these innovations allow the paper to move beyond existing studies and provide a more nuanced and globally comparative understanding of how trust is formed and sustained.

Our dataset is constructed mostly from the joint European Values Survey-World Values Survey time-series (hereafter EVS-WVS), covering 107 countries from 1981 to 2023. In addition, we also include control variables for countries where respondents live (e.g. homicide, GINI, GDP, Population, etc.). These data are obtained from World Bank Data, Federal Reserve Economic Data (FRED), National Statistical Agencies, and IndexMundi. Merging these sources yields a large and diverse sample of 441,481 observations across 107 countries in Europe, advanced economies (AEs), Asia, Latin America (LATAM), Africa, and the Middle East and North Africa (MENA), spanning 1981–2023 (see Table 1). We describe how we use those data sources to create the variables below. A description and descriptive statistics of the variables are presented in Tables 2 and 3, respectively.

Interpersonal trust is measured using four variables derived from the EVS–WVS. Generalized trust is captured through two indicators: TRUST_G and TRUST_P. In particular, TRUST_G measures whether respondents believe that most people can be trusted, based on the standard question contrasting trust with the need for caution; it is coded 1 for “Most people can be trusted” and 0 otherwise. TRUST_P reflects trust in strangers or people met for the first time, following existing literature that classifies such trust as a form of generalized trust. Respondents rate how much they trust people they meet for the first time, coded as 2 for “Trust completely”, 1 for “Trust somewhat”, and 0 for “Do not trust very much” or “Do not trust at all”. Particularized trust is measured through TRUST_F and TRUST_N, which capture trust in family members and neighbors, respectively. Both variables are based on respondents' reported level of trust in each group, using the same coding structure as TRUST_P.

As mentioned above, we employ ordinary least squares (OLS) as our primary estimation method. This is due to several reasons. First, OLS provides directly interpretable marginal effects for both our binary trust measure (TRUST_G) and ordinal trust measures (TRUST_P, TRUST_F, TRUST_N), facilitating straightforward interpretation and comparability of coefficients across all trust outcomes: an advantage not easily achieved with non-linear estimators that require transformations of coefficients (Angrist and Pischke, 2009). Second, OLS with country-clustered standard errors produces consistent estimates even when the dependent variable is binary or ordinal, provided the model is correctly specified (Wooldridge, 2010). Third, our empirical strategy incorporates country and year fixed effects to control for time-invariant country-level heterogeneity and common temporal shocks, which is considerably more tractable in OLS than in non-linear models, particularly given our large sample of over 60 countries where the incidental parameters problem may arise in fixed-effects logit/probit specifications (Greene, 2012). Fourth, OLS is widely employed in the existing social trust literature for similar outcome variables (e.g. Alesina and La Ferrara, 2002; Delhey and Newton, 2005; Putnam, 2007), facilitating comparability with previous studies.

Media and migration variables are also constructed from the EVS–WVS and World Bank data. TRDMEDIA identifies whether individuals obtain daily news about their country and the world through traditional media (e.g. newspapers, television, or radio), coded as 1 for daily use and 0 otherwise. SMEDIA captures daily news consumption through social media platforms such as Facebook or Twitter, coded similarly. The net migration variable, NETMIGR, is calculated using World Bank data by dividing the net number of migrants in a given country by its total population.

A set of personal characteristics from the EVS–WVS is included as individual-level controls. FEMALE indicates gender, coded 1 for female and 0 for male. LNAGE is the natural logarithm of respondents' age. IMIGRANT identifies whether a respondent was born in the survey country, coded 1 for immigrants and 0 for natives. MUSLIM captures religious identification, coded 1 for Muslims and 0 for others. The MUSLIM variable is included for data structure and theoretical reasons. First, in many countries covered by the EVS–WVS, particularly in MENA, parts of Asia, and Africa, Muslims constitute a large and policy-relevant share of the population. Using a Muslim indicator allows us to capture meaningful cross-country and within-country variation in religious identity that is large enough for statistical analysis. Second, the literature on social trust often highlights that religious identity can shape norms of social interaction, in-group versus out-group trust, and community structures. In this context, a Muslim indicator is frequently used as a parsimonious proxy to examine whether belonging to a major global religious group with distinct institutional and social characteristics is associated with differences in trust outcomes. Third, from an empirical standpoint, including a single, clearly defined religion variable helps avoid over-parameterization and multicollinearity that may arise if multiple religious denominations are included simultaneously, especially when some categories have small sample sizes or overlap strongly with regional fixed effects. Given that country and year fixed effects are already included, the MUSLIM variable captures within-country variation in religious identity while maintaining model parsimony. Importantly, using this variable does not imply that other religions are unimportant; rather, it reflects a balance between theoretical relevance, data availability, and empirical tractability within a global comparative framework. EMPLOYED indicates paid employment status. Perceived social class is measured by HIGHERCL, ranging from 1 (lower class) to 5 (upper class). Education level, EDU, is coded as 0 for lower, 1 for middle, and 2 for upper education. MARRIED denotes marital status, coded 1 for married respondents and 0 otherwise. Finally, URBAN identifies whether the respondent lives in an urban area.

Country-level controls are drawn from the EVS–WVS, World Bank, FRED, and national statistical sources. CORRUPT measures perceived corruption, ranging from 0 (“no corruption”) to 9 (“abundant corruption”). Higher values indicate higher levels of corruption. HOMICIDE captures the intentional homicide rate per 100,000 people. Income inequality is measured through the GINI coefficient, constructed using multiple official data sources. Demographic and economic variables include the crude birth rate (BIRTH), logarithm of total population (LNPOPU), and logarithm of total gross domestic product per capita (LNGDPPC), all sourced from the World Bank.

Table 4 examines the effects of media consumption and net migration on generalized trust using unweighted OLS and weighted regressions (WRP and WRG). The results indicate that the relationship between media use and trust is nuanced and sensitive to weighting. Social media use (SMEDIA) is generally associated with lower levels of generalized trust, particularly for trust in strangers (TRUST_P). In the OLS specification, it has a negative and statistically significant effect on both measures. However, once population or GDP weights are applied, the effect weakens and becomes mostly insignificant, except for a small negative effect on TRUST_P in the GDP-weighted model. This suggests that the adverse impact of social media on trust is more pronounced in smaller or less economically dominant countries. By contrast, traditional media (TRDMEDIA) exhibits a more mixed pattern. It is weakly negative or insignificant in the OLS and population-weighted models, but becomes positive and highly significant in the GDP-weighted regressions for both trust measures. This indicates that in larger or more economically developed countries, reliance on traditional media is associated with higher levels of generalized trust, possibly reflecting higher-quality information environments or stronger institutional frameworks. The effect of net migration (NETMIGR) is positive and statistically significant for trust in most people (TRUST_G) in the OLS and population-weighted specifications, but becomes insignificant in the GDP-weighted model. For trust in strangers (TRUST_P), the coefficients are consistently insignificant across all specifications. These results suggest that migration may be weakly associated with higher generalized trust, but the effect is not robust. Turning to control variables, female respondents exhibit significantly lower levels of trust across all models. Age is positively associated with both measures of trust, especially for TRUST_P, while immigrant status is consistently negative and highly significant. Socioeconomic variables such as education and perceived social class remain strong and positive predictors across all specifications. At the country level, perceived corruption (CORRUPT) is consistently negative and highly significant, reinforcing the importance of institutional trust. Income inequality (GINI) exhibits weak and inconsistent effects once weighting is applied: positive and significant in OLS models, but losing significance in weighted regressions, suggesting the relationship is not robust and driven by smaller or less economically significant countries. Similarly, the birth rate (BIRTH) is negatively associated with trust in OLS estimates, but this effect diminishes substantially and becomes insignificant in weighted models, indicating that the relationship is largely driven by smaller or lower-income countries.

Table 5 extends the analysis by examining regional heterogeneity. The results reveal substantial variation, particularly for migration and traditional media. Social media consumption is negatively associated with both generalized trust measures in the full sample, though this effect is primarily driven by Europe and AEs, where social media use is negatively connected with trust in strangers (TRUST_P). No significant associations emerge in Asia, Latin America, or LAFMENA. Conversely, traditional media consumption yields regionally divergent associations: in Europe and AEs, it is positively associated with trust in most people (TRUST_G), while in Latin America and LAFMENA, it is negatively connected with trust in strangers (TRUST_P), with an additional negative association with TRUST_G in LAFMENA. The most striking regional divergence emerges for net migration. In the full sample, net migration is positively associated with both trust measures. However, this positive connection is driven almost entirely by Europe and AEs, where higher net migration is linked with substantially higher trust in both measures, and by Asia, where a modest positive association emerges for TRUST_G only. In stark contrast, net migration exhibits strong negative associations with both generalized trust measures in Latin America and LAFMENA. Among individual characteristics (INDIVIDUAL), females exhibit significantly lower trust across the full sample and most regions, except Asia. Age is positively connected with trust in strangers in Europe and AEs, and LAFMENA. Immigrant status yields region-specific effects: negative for trust in strangers in Europe and AEs, but positive for TRUST_G in LAFMENA. Muslim identification demonstrates regional heterogeneity: negative in Asia, positive in LAFMENA, and marginally negative in Latin America. Perceived social class is consistently positive across specifications, while education shows positive effects in the full sample, Europe and AEs, and Latin America, but not in Asia or LAFMENA. Marital status produces mixed results, and urban residence shows no significant association in any specification. At the country level (COUNTRY), perceived corruption is a robust and universally negative predictor across all regions. Income inequality similarly demonstrates consistently negative associations, with particularly pronounced effects in Latin America and LAFMENA. The homicide rate exhibits divergent patterns: positively associated with trust in Latin America but negatively connected in LAFMENA. The crude birth rate shows strong negative associations in Latin America and LAFMENA, while GDP per capita exhibits a dramatic regional reversal—positive in Europe and AEs but strongly negative in Latin America and LAFMENA.

Table 6 reports the determinants of particularized trust—trust in family members (TRUST_F) and neighbors (TRUST_N), using unweighted OLS and weighted regressions (WRP and WRG). Overall, the results reveal consistent patterns across specifications. Social media use (SMEDIA) shows a nuanced effect: positively associated with trust in family under OLS but insignificant under weighted specifications. For trust in neighbors, however, SMEDIA is consistently negative and highly significant across all models. In contrast, traditional media use (TRDMEDIA) is positively and statistically significant for both TRUST_F and TRUST_N in every specification, indicating robustness. Net migration (NETMIGR) is generally negatively associated with particularized trust. Under OLS and population-weighted regressions, higher migration significantly reduces both trust in family and neighbors. However, under GDP weighting, the effect weakens and becomes insignificant for trust in neighbors, suggesting the negative relationship is more pronounced in smaller or less economically dominant countries. Regarding individual characteristics (INDIVIDUAL), females report significantly lower trust in both family and neighbors across most specifications. Age has no effect on trust in family under OLS but becomes positive in weighted models, while it is strongly and positively related to trust in neighbors in all specifications. Immigrant status does not affect trust in family but is consistently associated with lower trust in neighbors. Muslim identification shows positive associations with trust in family, while employment status is associated with lower trust in neighbors. Perceived social class is one of the most robust predictors, positively associated with both trust types across all specifications. Education increases trust in family in all models, while its effect on trust in neighbors is weaker. Marital status is positively associated with trust in family, and urban residence is associated with lower trust in neighbors. At the country level (COUNTRY), perceived corruption is consistently and negatively associated with both trust forms. Homicide rates are positively associated with trust in neighbors only under OLS, but this result is not robust. Income inequality is positively related to trust under OLS but loses significance in weighted regressions, suggesting sensitivity to country composition.

Table 7 examines regional heterogeneity in the determinants of particularized trust. The results reveal both common patterns and substantial regional variation.

Social media consumption is positively associated with trust in family members in the full sample, but negatively connected with trust in neighbors. Regionally, social media is negatively associated with trust in neighbors in Europe and AEs, while positively linked with trust in family members in Latin America and LAFMENA (though only marginally significant). No significant effects emerge in Asia. Traditional media consumption demonstrates consistently positive associations with both particularized trust measures across the full sample and most regions, suggesting it generally fosters particularized trust through reinforcing shared social norms.

Net migration exhibits strikingly divergent associations across regions. In Europe and AEs, net migration is negatively linked with trust in family but positively associated with trust in neighbors. In Asia, it shows no connection with family trust but is positively associated with trust in neighbors. In stark contrast, in Latin America and LAFMENA, net migration exhibits strong negative associations with both trust in family members and trust in neighbors, suggesting migration erodes both intimate and community-level trust in developing regions. For individual characteristics, females exhibit lower particularized trust across the full sample, with regional variation—no effect in Europe and AEs, negative for neighbors only in Asia, and negative for both in Latin America and LAFMENA. Age is positively connected with trust in neighbors across most regions. Immigrant status yields limited effects, negatively associated with trust in neighbors primarily in Europe and AEs. Muslim identification demonstrates consistent positive associations with particularized trust, particularly in Asia and LAFMENA. Employment status is negatively associated with both trust measures, especially in Latin America and LAFMENA. Perceived social class is consistently positive across regions, while education shows positive associations with trust in family but not neighbors. Marital status is positively associated with trust in family across most regions, and urban residence is negatively associated with trust in neighbors across all regions. At the country level, perceived corruption demonstrates consistent negative associations with trust in neighbors across all regions, but no significant association with trust in family. Income inequality is positively associated with trust in family in Europe and AEs but negatively linked with trust in neighbors across most regions, with strongest effects in Latin America and LAFMENA. The homicide rate shows positive associations in Europe and AEs, and Latin America, but negative associations in LAFMENA. The crude birth rate exhibits strong negative associations in Latin America and LAFMENA. GDP per capita shows strong negative associations with trust in family across most regions, suggesting economic development may weaken traditional family bonds. The markedly lower adjusted R² values for particularized trust compared with generalized trust models, particularly in Asia and Latin America, suggest that unobserved cultural, historical, and institutional factors play an important role in shaping trust outcomes across different regional contexts.

This paper provides new cross-country evidence on the determinants of interpersonal trust using a uniquely large and long-run dataset spanning more than four decades. Several key insights emerge from the empirical results, which together underscore the complex, multidimensional, and context-dependent nature of trust. First, the role of media consumption appears central but fundamentally heterogeneous. Social media is generally associated with lower levels of generalized trust, particularly trust in strangers, yet this effect is neither universal nor robust across specifications and regions. The attenuation of the negative relationship in GDP-weighted models suggests that the adverse influence of social media is more pronounced in smaller or less economically dominant countries. One interpretation is that weaker institutional environments, lower media literacy, and higher exposure to misinformation may amplify the trust-eroding effects of social media in these contexts. By contrast, in larger and more developed economies, stronger regulatory frameworks, more diversified media ecosystems, and higher levels of digital literacy may mitigate these risks. While early research suggested that online platforms might expand networks and promote trust (Valenzuela et al., 2009), more recent work has emphasized their role in polarization and the fragmentation of shared informational environments (Theocharis et al., 2015). Our findings align with this later view, but also show that effects depend critically on institutional context. Traditional media exhibits the opposite pattern: while weak or insignificant in unweighted models, it becomes strongly positive in GDP-weighted regressions and in AEs. This suggests that the trust-enhancing role of traditional media depends critically on the quality and credibility of information environments. In settings where traditional media are perceived as reliable and relatively independent, they may reinforce shared narratives, reduce uncertainty, and foster generalized trust. However, the negative or insignificant effects observed in some developing regions indicate that when media systems are politicized, fragmented, or less credible, they may fail to perform this coordinating function.

Second, the impact of migration on trust is highly context-dependent and emerges as one of the most heterogeneous findings. While aggregate results sometimes suggest a positive association between migration and trust, regional disaggregation reveals sharply divergent patterns. In Europe and AEs, higher net migration is positively associated with both generalized and particularized trust, consistent with theories of intergroup contact and diversity benefits. In Asia, modest positive effects emerge for generalized trust. However, in Latin America and LAFMENA, net migration exhibits strong negative associations with trust across all measures. This reversal underscores that the relationship between migration and social trust is not inherently positive or negative but is fundamentally shaped by the broader socioeconomic and institutional context in which migration occurs. This heterogeneity helps reconcile conflicting evidence in the literature, which has long debated whether diversity promotes tolerance through contact (Allport, 1954; Pettigrew and Tropp, 2006) or undermines trust through competition and status threat (Putnam, 2007). Our findings demonstrate that both processes can operate, with institutional quality, absorptive capacity, policy frameworks, and integration capacity determining whether migration is associated positively or negatively with trust. Migration similarly has a stronger and more consistent negative effect on particularized trust than on generalized trust, underscoring the multidimensional nature of trust. Third, the analysis highlights important distinctions between generalized and particularized trust. While some determinants such as institutional quality and socioeconomic status affect both forms of trust, others operate differently. Social media, for instance, tends to reinforce close ties (family) while weakening broader social trust (neighbors and strangers), suggesting a potential “bonding vs bridging” trade-off. This supports Uslaner's (2002) claim that bridging and bonding should be distinguished analytically rather than treated as substitutes.

The findings of this study carry several important implications. First, the divergent effects of media consumption suggest that efforts to strengthen media quality and independence may be important for fostering social trust. In developing regions where traditional media is negatively associated with trust, investments in media professionalism, transparency, and editorial independence could help transform media into a trust-building institution rather than a source of cynicism. In developed contexts where social media erodes trust, policies aimed at reducing misinformation, promoting digital literacy, and fostering constructive online discourse may help mitigate social media's negative effects on social cohesion. Second, the regional reversal in migration's effects carries important implications for migration policy and integration strategies. In high-income contexts, the positive association between migration and trust suggests that well-managed migration and effective integration policies can contribute to social cohesion rather than undermining it. In developing regions, however, the negative effects of migration on trust highlight the need for humanitarian and development policies that address the root causes of forced displacement, provide adequate resources for receiving communities, and support social integration in contexts of resource scarcity. International cooperation and burden-sharing are essential to prevent migration from becoming a source of social fragmentation in already vulnerable regions.

This study has several limitations. First, the cross-sectional nature of the EVS–WVS data limits our ability to establish causal relationships. While our models include country and year fixed effects to account for time-invariant unobserved heterogeneity at the country level and common temporal shocks, these specifications do not overcome the fundamental endogeneity concerns that plague observational studies of trust determinants. Our estimates should therefore be interpreted as associational rather than causal. The key regressors of interest, such as media consumption, net migration, etc., are potentially endogenous, as they may be jointly determined with trust or influenced by unobserved factors that also shape trust outcomes. Therefore, all policy implications derived from our findings should be considered tentative and suggestive rather than prescriptive. In addition, while employing OLS for the reasons outlined above, we acknowledge certain limitations of this approach. Future research employing instrumental variables, lagged exposure designs, shift-share instruments for migration, or panel data with within-respondent variation where the EVS–WVS structure permits would be valuable for establishing causal relationships.

The author sincerely thanks the editorial board and reviewers for valuable comments and suggestions to help improve this research paper.