Introduction:
While corruption exists in both democracies and autocracies, its social consequences may differ fundamentally across regime types. Democratic norms of equality and impartiality make trust highly sensitive to institutional failure. We theorize two mechanisms—normative amplification and representative contagion—by which corruption erodes trust more in democracies. In democracies, corruption violates core fairness norms and implicates the citizenry that elected corrupt officials. In autocracies, corruption is expected and elites are seen as separate from ordinary citizens.
Methods:
To test this theory, we perform multilevel analysis of data from 62 countries combining individual-level survey responses with country-level democratic quality indicators.
Results:
We first demonstrate that perceiving corruption predicts lower generalized trust almost universally across individuals. We then show this individual-level psychological mechanism is considerably stronger in democracies than in autocracies, even controlling for inequality and country-level corruption.
Discussion:
These findings reveal an asymmetric vulnerability: the accountability structures that make democracies function also make their social capital fragile. This has important implications for understanding democratic resilience, as corruption threatens the social trust necessary for democratic cooperation differently across regime types.
Democracy may be uniquely sensitive to certain threats. Recent scholarship on democratic backsliding reveals how democracies can erode from within when norms decay and institutions weaken (Levitsky and Ziblatt, 2018). In this article, we identify a specific sensitivity: in democracies, social capital appears to be particularly responsive to corruption.
We theorize that this sensitivity arises from democracy’s foundational commitments to equality and impartiality. These commitments may create two psychological mechanisms that amplify corruption’s impact on social trust. First, normative amplification: in democracies, where universalism is the professed ideal, corruption may signal a breach of the social contract. Citizens may infer that if the institutions meant to embody fairness are compromised, the wider society is untrustworthy. In autocracies, by contrast, where particularism is expected, corruption confirms business as usual rather than signaling societal rot. Second, representative contagion: in democracies, corrupt officials are viewed as emanating from “the people” through elections, potentially implicating the citizenry itself. In autocracies, predatory elites are seen as a distinct class, quarantining interpersonal trust from elite malfeasance. If these mechanisms operate as theorized, then the individual-level psychological process linking corruption perceptions to social distrust should be regime-dependent—strong in democracies, weak in autocracies.
A study by You (2018) provides suggestive evidence for our thesis. Using country-level data on social trust and corruption, and studying democracies and autocracies separately, he demonstrated that more corruption is strongly associated with weaker social trust among democracies—but not among autocracies. This striking pattern is consistent with our theory. However, as the finding was obtained at the aggregate level, it leaves open whether it reflects genuine differences in how individuals psychologically process corruption, or whether it is an artifact of other phenomena.
The present paper aims to provide individual-level evidence for how trust among people in democracies may be especially sensitive to corruption. After replicating You’s country-level findings in more recent data from 62 countries—covering the full spectrum from autocracies like Russia and Iran to stable liberal democracies like New Zealand and Netherlands—we use multilevel modeling to test whether a corresponding individual-level pattern exists. We find that individuals’ perceptions of corruption are associated with lower generalized trust in democracies, while this same individual-level association is substantially weaker or absent in autocracies. These findings suggest an asymmetry in how corruption relates to social trust across regime types. While democracies foster high social trust through their institutions, they may simultaneously make that social capital more vulnerable to perceptions of institutional failure. This may be the price of accountability: the very norms that make democracies function—equality, representation, transparency—may also ensure that institutional failures resonate in citizens’ social worldviews.
Social trust—the belief that most people can be trusted—has long been recognized as a cornerstone of democratic societies (Putnam, 1993; Fukuyama, 1995). It facilitates civic cooperation, lowers transaction costs, and enables the collective action necessary for democratic governance (Ostrom, 2000; Knack and Keefer, 1997).
A dominant answer in the literature for what erodes this resource is corruption. When citizens perceive that public officials are acting dishonestly, they infer that the wider society is untrustworthy (Uslaner, 2002; Rothstein, 2011; Rothstein and Stolle, 2008). Rothstein and Uslaner (2005) argue that corruption and social trust are linked through perceptions of fairness: corruption signals that the system is rigged in favor of the connected, undermining the belief that others will play by the rules. Similarly, You (2005) emphasizes that corruption generates perceptions of unfairness that erode the foundation of generalized trust. This creates a “vicious circle” where corruption breeds distrust, which in turn facilitates more corruption by undermining collective enforcement of norms (della Porta, 2000; Rose-Ackerman and Palifka, 2016). Empirical research has documented this negative association across diverse contexts (Chang and Chu, 2006; Morris and Klesner, 2010; Richey, 2010; Seligson, 2002).
Importantly, experimental evidence confirms that this relationship is causal: exposing individuals to information about institutional corruption reduces their generalized trust in others. Rothstein and Eek (2009) demonstrate that Swedish participants randomly assigned to scenarios depicting corrupt public officials subsequently express lower trust in strangers. Martinangeli et al. (2024) replicate this finding across multiple countries, showing that learning about poor institutional quality causally reduces generalized trust. These experimental studies establish that the corruption-trust link reflects a genuine psychological mechanism, not merely spurious correlation. Corruption perceptions also have broader psychological consequences: research shows that perceived corruption is associated with increased conspiracy beliefs (Alper, 2023; Cordonier et al., 2021; Cordonier and Cafiero, 2024), suggesting that corruption undermines not only interpersonal trust but also trust in official explanations and institutions more broadly.
However, this narrative is challenged by You's (2018) finding that country-level corruption is not associated with lower social trust in autocracies. To reconcile these findings, we propose that regime type moderates how individuals interpret and react to corruption. In other words, we suggest that regime type influences the very individual-level mechanism that links corruption perceptions to trust. This approach builds on previous work using cross-level interaction models to examine how country-level factors moderate individual-level relationships (Hakhverdian and Mayne, 2012).
2.2 The moderating role of democratic institutions
We propose two micro-level mechanisms whereby individuals in democracies should exhibit a stronger psychological link between corruption perceptions and generalized trust than individuals in autocracies.
2.2.1 Normative amplification
Democratic and autocratic regimes establish fundamentally different normative frameworks. Democracies are built on principles of equality before the law and impartial treatment of citizens (Dahl, 1998). The norm of impartiality—treating citizens equally regardless of their connections or status—is central to the legitimacy of democratic governance (Rothstein and Teorell, 2008; Mungiu-Pippidi, 2015). When officials engage in favoritism or bribery, they betray not just administrative rules but the core promise of democratic governance. This normative amplification means that for individuals living in democracies, corruption signals a fundamental breach of the social contract: if the institutions meant to embody fairness are compromised, why should strangers be trustworthy? (Warren, 2004). In autocracies, by contrast, particularism—the allocation of public goods based on personal connections rather than universal rules—is often the norm rather than the exception (Mungiu-Pippidi, 2006). Corruption is endemic and expected. When individuals perceive corruption in such contexts, it confirms business as usual rather than signaling a breakdown of social order. The psychological link between corruption perceptions and generalized trust is therefore attenuated: corruption is discounted as a survival strategy within a known system (Smith, 2007).
2.2.2 Representative contagion
In democracies, officials are selected through competitive elections and are supposed to represent “the people” (Manin, 1997). This creates what we term representative contagion: when individuals observe that elected officials are corrupt, they may infer that their fellow citizens, who selected these officials and whom these officials represent, are also untrustworthy. The corruption of representatives becomes psychological evidence about the represented (Rothstein and Eek, 2009). In autocracies, by contrast, predatory elites are typically viewed as a distinct class, separate from and often opposed to ordinary citizens (Acemoglu and Robinson, 2012). Their corruption is contained within the political sphere and does not implicate horizontal relationships among citizens. This quarantines interpersonal trust from elite malfeasance. Individuals can maintain trust in their neighbors while simultaneously acknowledging that the ruling class is corrupt (Ledeneva, 2013).
Assuming either, or both, of these mechanisms operate, we obtain three testable hypotheses that build progressively from aggregate patterns to individual-level mechanisms to cross-national moderation. The first hypothesis is that You’s (2018) finding is robust:
H1: At the country level, the association between perceived corruption and generalized trust is strong among democracies and weak or absent among autocracies.
The second hypothesis is that, consistent with the experimental literature, perceiving corruption is generally associated with lower trust in other people.
H2: Within countries, individuals who perceive higher levels of corruption have lower generalized trust.
The third hypothesis is that, consistent with the mechanisms outlined above, the effect of perceiving corruption on trust varies with regime type.
H3: Perceptions of corruption predict lower generalized trust more strongly in democracies than in autocracies.
We combine individual-level data from the most recent wave (2017–2022) of the World Values Survey (WVS; Haerpfer et al., 2022) with country-level indicators (averaged across the same period) of democratic quality from the Varieties of Democracy (V-Dem) project (Coppedge et al., 2025; Pemstein et al., 2025). Our analysis includes 62 countries for which we have complete data on all variables of interest. We use WVS Wave 7 (2017–2022) because it contains the corruption perception module required for our analysis. Although a Joint EVS/WVS dataset exists with 92 countries, the European Values Survey does not include the corruption perception items, making it unsuitable for our purposes. Our 62 countries therefore represent the full set of countries with complete data on perceived corruption, generalized trust, and democratic quality indicators.
The WVS provides our key individual-level measures. Generalized trust is measured by the standard question: “Generally speaking, would you say that most people can be trusted or that you need to be very careful in dealing with people?” Responses are coded 1 for “most people can be trusted” and 0 otherwise. Perceived corruption is measured by asking respondents how widespread they believe corruption to be among public officials, on a scale from 1 (there is no corruption in my country) to 10 (there is abundant corruption in my country). While corruption perceptions may not perfectly align with objective corruption levels (Charron, 2016), perceptions are presumably what directly affect individual trust judgments.
We include standard individual-level controls: age (five categories: 18–29, 30–39, 40–49, 50–59, 60+), gender (male/female), education (three levels based on ISCED categories: low, medium, high), household income (three levels based on the WVS 10-point scale: low [1–3], medium [4–7], high [8–10]), and employment status (three categories: employed [full-time, part-time, or self-employed], not in labor force [retired, homemaker, or student], and unemployed/other).
From V-Dem, we use two measures of democratic quality: the Regimes of the World (RoW) classification and the Liberal Democracy Index. The RoW (Lührmann et al., 2018) is a categorical measure distinguishing closed autocracies (no multiparty elections), electoral autocracies (multiparty elections that are not free and fair), electoral democracies (free elections but limited liberal protections), and liberal democracies (free elections with strong liberal protections). Following our theoretical framework—which emphasizes that the mechanisms of normative amplification and representative contagion require genuine electoral accountability—we create a binary classification: democracies (electoral and liberal democracies, RoW = 2–3) versus autocracies (closed and electoral autocracies, RoW = 0–1). Electoral autocracies are classified as autocracies because, despite having multiparty elections, these elections lack the competitive integrity necessary for the representative contagion mechanism to operate.
In contrast to the categorical RoW measure, the Liberal Democracy Index is a continuous measure, which captures both electoral and liberal dimensions of democracy, including the quality of elections, checks on executive power, equality before the law, and individual liberties. This index ranges from 0 (least democratic) to 1 (most democratic). We use the Liberal Democracy Index rather than the Electoral Democracy Index (also known as Polyarchy) because our theoretical mechanisms—normative amplification and representative contagion—depend on features beyond electoral procedures. The liberal component of the Liberal Democracy Index captures the rule of law, checks on executive power, and equality before the law, which are central to our argument that corruption in democracies violates norms of impartiality. In robustness analyses, we also test whether results hold using the Electoral Democracy Index as an alternative moderator.
3.2 Analysis plan
Our research design tests three progressively refined hypotheses outlined above.
3.2.1 Country-level analysis (H1)
To test H1, we calculate country-level aggregates of perceived corruption and generalized trust, then examine whether their correlation differs when calculated separately among democracies and autocracies as defined by the RoW categorization. H1 predicts a strong negative correlation between perceived corruption and generalized trust among democracies but a weaker correlation among autocracies.
The above dichotomous analysis matches the original approach of You (2018). As a continuous alternative, we also examine the interaction between perceived corruption and the Liberal Democracy Index in a multiple regression analysis of country-level generalized trust. H1 predicts a negative interaction, representing a stronger negative effect of corruption in more democratic countries.
3.2.2 Multilevel analysis (H2, H3)
To test whether the aggregate pattern reflects genuine individual-level mechanisms (H2) and whether these mechanisms vary by regime type (H3), we estimate a random-intercept, random-slope multilevel logistic regression model. This approach models the hierarchical structure of the data, with individuals nested within countries. Standard errors appropriately reflect uncertainty at both levels.
At Level 1 (individual), generalized trust is modeled as a function of perceived corruption, controlling for demographic characteristics (age, gender, education, income, and employment status). At Level 2 (country), we allow both the intercept and the slope for perceived corruption to vary across countries. Crucially, we include a cross-level interaction between perceived corruption and the Liberal Democracy Index (treated as a continuous variable), which directly tests H3: whether the individual-level corruption-trust relationship varies with democratic quality. In other words, the cross-level interaction estimates whether the psychological mechanism linking corruption perceptions to trust operates differently depending on institutional context.
Formally, the model can be expressed as follows:
Level 1 (Individual):
Level 2 (Country):
where is the probability of expressing trust for individual i in country j; is perceived corruption (grand-mean centered); is a vector of demographic controls; is the Liberal Democracy Index (grand-mean centered); is the cross-level interaction coefficient testing H3; and , are country-level random effects assumed to follow a bivariate normal distribution.
For computational efficiency with large sample sizes (>85,000 individuals), we use an aggregated binomial approach. Observations are grouped by country, corruption level, and demographic categories, and trust incidence is modeled using a binomial distribution. This yields estimates identical to individual-level analysis but with substantially improved computational performance. Both perceived corruption and the Liberal Democracy Index are grand-mean centered to facilitate interpretation of main effects.
We also conduct robustness checks including: (1) adding competing cross-level moderators to test whether these factors can account for the democracy moderation; (2) testing press freedom, the Electoral Democracy Index, and state resilience (Travaglino et al., 2025) as alternative moderators in separate models (as their high correlations with liberal democracy, r = 0.90 and 0.78 respectively, preclude simultaneous estimation); and (3) leave-one-out analyses to ensure no single country drives the results.
For competing moderators, we include economic inequality (Gini coefficient from SWIID; Solt, 2020), political polarization (from V-Dem), and measures of digital information access. We include both social media use as a self-reported news source (country-level mean from WVS item on frequency of obtaining political information from social media) and internet penetration (percentage of population using the internet; World Bank, 2024).
If our theory is correct, we should observe a negative main association between corruption perceptions and trust at the individual level (H2) and a negative cross-level interaction, indicating that the corruption-trust relationship is stronger (more negative) in more democratic countries (H3).
Table 1 presents the 62 countries, ordered by the Liberal Democracy Index, with their results for generalized trust and perceived corruption.
| Country | Regime type (RoW) | Liberal democracy index | N | Generalized trust (%) | Perceived corruption M (SD) |
|---|---|---|---|---|---|
| New Zealand | Democracy | 0.83 | 1,057 | 59.5 | 5.52 (2.37) |
| Germany | Democracy | 0.83 | 1,528 | 46.0 | 5.58 (2.22) |
| Netherlands | Democracy | 0.82 | 2,145 | 61.2 | 6.20 (2.16) |
| Uruguay | Democracy | 0.81 | 1,000 | 14.9 | 7.70 (2.29) |
| Australia | Democracy | 0.80 | 1813 | 54.0 | 6.65 (2.28) |
| United Kingdom | Democracy | 0.79 | 3,056 | 45.8 | 7.10 (2.25) |
| Chile | Democracy | 0.79 | 1,000 | 14.3 | 7.10 (2.11) |
| Korea South | Democracy | 0.78 | 1,245 | 32.9 | 6.51 (1.59) |
| Canada | Democracy | 0.76 | 4,018 | 49.5 | 6.73 (2.00) |
| Japan | Democracy | 0.75 | 1,353 | 35.6 | 6.88 (2.06) |
| United States | Democracy | 0.74 | 2,596 | 39.7 | 7.83 (2.10) |
| Slovak Republic | Democracy | 0.74 | 1,200 | 21.6 | 7.81 (1.91) |
| Czech Republic | Democracy | 0.73 | 1,200 | 37.3 | 7.06 (2.00) |
| Taiwan | Democracy | 0.72 | 1,223 | 31.0 | 7.61 (2.10) |
| Greece | Democracy | 0.70 | 1,200 | 8.4 | 8.37 (1.70) |
| Cyprus | Democracy | 0.70 | 1,000 | 8.0 | 8.23 (1.91) |
| Peru | Democracy | 0.68 | 1,400 | 5.3 | 9.51 (1.21) |
| Argentina | Democracy | 0.64 | 1,003 | 20.7 | 8.51 (1.63) |
| Brazil | Democracy | 0.57 | 1762 | 6.6 | 9.45 (1.57) |
| Romania | Democracy | 0.56 | 1,257 | 11.9 | 8.73 (1.85) |
| Tunisia | Democracy | 0.53 | 1,208 | 14.2 | 8.16 (2.41) |
| Colombia | Democracy | 0.53 | 1,520 | 4.5 | 9.48 (1.48) |
| Mongolia | Democracy | 0.51 | 1,638 | 27.5 | 7.60 (2.20) |
| Armenia | Democracy | 0.47 | 1,223 | 8.1 | 7.55 (2.70) |
| Ecuador | Democracy | 0.45 | 1,200 | 5.9 | 8.88 (1.84) |
| Indonesia | Democracy | 0.44 | 3,200 | 5.2 | 8.38 (2.51) |
| Kenya | Autocracy | 0.41 | 1,266 | 9.6 | 8.46 (2.36) |
| Mexico | Democracy | 0.41 | 1741 | 10.3 | 8.87 (2.05) |
| Guatemala | Democracy | 0.39 | 1,229 | 18.0 | 9.14 (1.67) |
| Nigeria | Democracy | 0.36 | 1,237 | 12.7 | 8.74 (2.18) |
| Maldives | Democracy | 0.34 | 1,039 | 21.3 | 9.27 (1.45) |
| Singapore | Autocracy | 0.33 | 2012 | 34.0 | 3.52 (1.99) |
| India | Autocracy | 0.32 | 1,692 | 17.7 | 7.77 (2.27) |
| Bolivia | Democracy | 0.32 | 2067 | 8.6 | 8.63 (1.94) |
| Philippines | Autocracy | 0.30 | 1,200 | 5.3 | 6.73 (2.71) |
| Malaysia | Autocracy | 0.30 | 1,313 | 19.6 | 8.00 (2.00) |
| Ukraine | Autocracy | 0.28 | 1,289 | 30.7 | 8.41 (1.88) |
| Lebanon | Autocracy | 0.28 | 1,200 | 9.9 | 7.83 (2.03) |
| Kyrgyzstan | Autocracy | 0.28 | 1,200 | 11.8 | 8.90 (2.09) |
| Serbia | Autocracy | 0.27 | 1,046 | 16.6 | 8.39 (1.92) |
| Pakistan | Autocracy | 0.25 | 1995 | 23.5 | 8.70 (2.06) |
| Jordan | Autocracy | 0.24 | 1,203 | 16.0 | 8.20 (2.27) |
| Iraq | Autocracy | 0.24 | 1,200 | 11.2 | 8.78 (1.73) |
| Morocco | Autocracy | 0.24 | 1,200 | 16.5 | 7.70 (2.00) |
| Hong Kong SAR | Autocracy | 0.22 | 2075 | 39.5 | 5.44 (2.03) |
| Zimbabwe | Autocracy | 0.20 | 1,215 | 2.1 | 8.55 (2.57) |
| Myanmar (Burma) | Autocracy | 0.17 | 1,200 | 15.1 | 7.38 (2.56) |
| Thailand | Autocracy | 0.16 | 1,500 | 31.4 | 6.97 (2.35) |
| Libya | Autocracy | 0.14 | 1,196 | 9.3 | 9.12 (1.55) |
| Ethiopia | Autocracy | 0.14 | 1,230 | 11.9 | 8.65 (2.21) |
| Egypt | Autocracy | 0.12 | 1,200 | 7.4 | 8.52 (1.81) |
| Kazakhstan | Autocracy | 0.12 | 1,276 | 23.9 | 6.98 (2.27) |
| Vietnam | Autocracy | 0.11 | 1,200 | 27.7 | 7.37 (2.13) |
| Turkey | Autocracy | 0.11 | 2,415 | 14.3 | 6.57 (2.22) |
| Iran | Autocracy | 0.11 | 1,499 | 14.8 | 6.77 (3.13) |
| Bangladesh | Autocracy | 0.10 | 1,200 | 12.9 | 7.75 (2.06) |
| Russia | Autocracy | 0.09 | 1810 | 23.9 | 7.66 (2.00) |
| Uzbekistan | Autocracy | 0.07 | 1,250 | 34.7 | 6.71 (2.56) |
| Venezuela | Autocracy | 0.06 | 1,190 | 14.2 | 8.66 (1.80) |
| Nicaragua | Autocracy | 0.05 | 1,200 | 4.2 | 7.87 (2.79) |
| China | Autocracy | 0.04 | 3,036 | 65.4 | 6.49 (2.37) |
| Tajikistan | Autocracy | 0.04 | 1,200 | 20.6 | 5.62 (2.58) |
Country-level summary statistics.
Countries are ordered by Liberal Democracy Index (V-Dem). Regime Type is based on the Regimes of the World (RoW) classification: democracies include electoral and liberal democracies; autocracies include closed and electoral autocracies. N represents the number of WVS respondents from each country. Generalized Trust represents the percentage of respondents answering “most people can be trusted.” Perceived Corruption shows the mean and standard deviation on a 1–10 scale.
Figure 1 tests H1 by showing how country-level generalized trust varies with perceived corruption, separately for democracies and autocracies. In support of H1, the pattern strikingly differs between regime types. Among democracies, there is a strong negative relationship: countries with higher perceived corruption have substantially lower generalized trust. Among autocracies, this relationship is considerably weaker—replicating You's (2018) finding in more recent data and with a theory-driven operationalization of regime type based on the Regimes of the World classification. The alternative analysis using the continuous Liberal Democracy Index as a moderator of the effect of perceived corruption on generalized trust confirms this pattern: the country-level interaction between perceived corruption and liberal democracy is negative (B = −12.07, 95% CI [−22.36, −1.77], p = 0.022).
