Political trust and democracy: the critical citizens thesis re-examined

ABSTRACT This article empirically assesses competing perspectives of the relationship between democracy and political trust. We conduct multilevel analyses on a cross-national panel dataset of 82 countries for the period 1990–2020. The findings suggest that there is a strong, negative relationship between democracy and political trust that cannot easily be dismissed as an artifact of model misspecification or response bias. Moreover, we re-examine the critical citizens thesis by disaggregating political trust into trust in partisan and “non-partisan” institutions to test the claim that well-functioning democracies contain and channel distrust into the more partisan political institutions to keep distrust from generalizing to the entire political system. The results fail to find a statistically significant difference of the effect of democracy on trust between partisan and non-partisan institutions, suggesting that low political trust within democracies may be a more acute problem than much of the literature suggests.


Introduction
While scholars have long recognized political trust as fundamental for the effective exercise of state authority, 1 recent events have revealed the consequences of its absence among a critical mass of citizens.Indeed, examples are now easy to come by across liberal democracies, from the Capitol Hill insurrection to the vitriolic and sometimes violent protests against vaccine mandates in several countries.In the United States, low political trust has been linked to heightened partisan polarization, 2 which is becoming increasingly prevalent across many established democracies. 3Partisan polarization has facilitated the rise of disinformation, voter suppression and populist anti-system parties, leading to a weakening of both democratic norms and support for democracy. 4hile these trends suggest a potential connection between low political trust and democratic decline, it is unclear whether there is a direct link between the two.
This article empirically explores the connection between political trust and democracyor more specifically, the extent to which a country is democratic, inclusive of all regime types, ranging from established democracies (e.g.Australia, Denmark), partialdemocracies (e.g.Dominican Republic, Bangladesh) to autocracies (e.g.China, Vietnam).We draw upon cross-national time series data to address two primary objectives.The first is to assess competing perspectives by examining the extent to which the level of democracy predicts the level of political trust across countries.Few scholars have conducted empirical analyses of political trust across regime types with a critical mass of authoritarian regimes. 5Moreover, these analyses are often based on simple correlations and/or do not account for potential response bias in non-democratic states due to self-censorship out of fear of reprisals for expressions of dissatisfaction.Consequently, we undertake multilevel analyses that control for a variety of both country-level and individual-level variablesincluding a control for potential preference falsificationto confirm the nature of the association between political trust and democracy.
The second objective is to re-examine the critical citizens thesis by disaggregating political trust.Until recently, political trust was widely considered a single underlying factor, a view empirically supported by factor analyses suggesting that most individuals do not differentiate between political institutions. 6However, this view has recently been challenged.New studies suggest that there are both theoretical and empirical reasons to disaggregate political trust. 7Consequently, this study tests whether trust differs between partisan political institutions (such as political parties or the government) and the theoretically non-or less partisan political institutions (such as the civil service or the justice system) both within and across regime types, providing a unique contribution to the debate of whether low political trust in non-partisan institutions may ultimately be detrimental to democracy. 8he results suggest that there is a strong, negative relationship between democracy and political trust that holds in the presence of a comprehensive set of controls, including controls for response bias.Moreover, we find that democracy is the single most influential variable on political trust across all models, underscoring its relative importance to political trust as compared to other factors identified in the literature.In comparing trust between partisan and non-partisan political institutions, we do not find a statistical difference in the effect of democracythe relationship remains negative with democracy as the most impactful covariate across both categories of political institutions.These findings suggest that low political trust across-the-board may be more problematic to the functioning of democracy than proponents of the critical citizens thesis had previously thought.

Political trust and democracy
The literature is divided on the relationship between political trust and democracy.One school of thought, rooted in the classical literature, suggests that democracies require high levels of political trust for regime stability and the effective exercise of authority. 9ow aggregate levels of political trust are expected to undermine the democratic process and lead to democratic crisis. 10As van der Meer and Zmerli contend, widespread political distrust provides "fertile ground for the emergence of anti-system political parties" and calls into question "the very survival of representative democracy and its institutions." 11In this sense, some suggest that democracies require a citizenry that is broadly trusting of political institutions, with sustained, across-the-board low political trust potentially resulting in democratic instability and deconsolidation.
Another perspective recognizes that political distrust plays an important role in democracies. 12The critical citizens perspective maintains that comparatively lower levels of political trust can be interpreted as a sign of a healthy democracy.While "critical" or "assertive" citizens exercising their democratic right to publicly criticize and challenge political institutions may decrease trust in those institutions, these challenges are thought to be an important aspect ofand therefore not threatenthe democratic political system. 13Accordingly, while trust in political institutions among some democracies may have become perennially low, the threat of democratic deconsolidation or regime instability is deemed negligible as it is considered independent from support for the principle of democracy, which, it is argued, remains comparatively strong.However, recent research suggests that among established democracies, support for democracy has been decreasing, while acceptance of authoritarian alternatives has increased. 14Consequently, the independence between political trust and support for democracy has recently been called into question. 15ome scholars take the middle ground and acknowledge that both political trust and distrust are essential for democraciesthat is, while widespread distrust is considered detrimental for democracy, some distrust may be healthy or even ideal. 16Warren contends that while democratic rights provide the foundation for generalized trust, democracies also affirm distrust by institutionalizing conflict. 17Consequently, he expects political trust within democracies to be bifurcated, with citizens having lower levels of political trust in those institutions designed to manage conflict. 18hat is, well-functioning democracies contain and channel distrust into the more partisan political institutions (e.g. the elected executive, the legislature), to keep distrust from generalizing to the entire political system.This provides a certain level of insulation for those institutions responsible for providing public goodssuch as the civil service, the police and the judiciarywhich are predicted to have comparatively high levels of trust within democracies.This distinction is analogous to the Eastonian framework of specific and diffuse political system support, with trust in partisan institutions more closely aligned with the former and trust in non-partisan institutions with the latter. 19Warren suggests that should less partisan political institutions suffer from sustained low trust, the viability of democratic regimes would be challenged.Accordingly, he proposes a "bifurcated" trust thesis in suggesting that in healthy democracies, one would expect political trust to be higher in less partisan as compared to partisan political institutions.
The literature pertaining to the relationship between democracy and political trust therefore yields three distinct perspectives: 1) political trust is expected, on average, to be higher across all institutions in democracies, with low trust leading to instability and crisis (the high trust thesis); 2) political trust is expected, on average, to be lower across all institutions in democracies (the critical citizens or low trust thesis); and 3) within democracies, political trust is expected, on average, to be higher in less partisan institutions and lower in more partisan institutions (the bifurcated trust thesis).

Empirical research
There is some empirical support for the perspective that democracies require high levels of political trust.Past research suggests that citizens with low political trust tend to have lower levels of law-abiding attitudes and higher levels of civil disobedience, which are considered essential to the effective exercise of authority and regime stability. 20However, the high trust thesis is called into question by empirical evidence directly linking the level of democracy to political trust. 21ystematic cross-national comparisons of the relationship between democracy and political trust across regime types are few.Many studies of political trust are limited to, or place the emphasis upon, democracies or subsets of democracies, 22 or focus on particular non-democratic states or specific world regions. 23Nonetheless, some studies do examine political trust across regime types, with parallel findingsa negative association between democracy and trust in various political institutions. 24Almost all these studies rely upon bivariate correlations in observing a difference in the average levels of political trust between countries, without taking into account the various factors known to influence trust.Two exceptions are Cinar and Ugur-Cinar and Dawson. 25hile Dawson finds that the negative association between democracy and confidence in non-partisan political institutions holds when accounting for a variety of countrylevel controls, Cinar and Ugur-Cinar find a statistically insignificant association.However, as the literature suggests that individual-level factors account for most of the variation in levels of political trust, these mixed findings may be a statistical artifact of the exclusion of these factors (such as the respondents' demographic attributes), which may result in overinflated estimates for country-level effects such as democracy. 26As such, it is unclear whether democracy's negative correlation with political trust would hold, or be as pronounced, in multi-level models that take into consideration both individual-level and country-level factors.Furthermore, while some studies conduct a systematic comparison of political trust between partisan and non-partisan political institutions among democracies, 27 only Cinar and Ugur-Cinar conduct this type of comparison across regime types. 28However, due to concerns of preference falsification in authoritarian regimes, they exclude full autocracies from their analysis, thereby limiting the variation of democracy.Consequently, no study has compared trust between partisan and non-partisan political institutions across the entire spectrum of democracy, while controlling for potential bias resulting from preference falsification.

Disaggregating the concept of trust
The first two perspectives linking democracy to political trust treat the latter as unidimensionalthat is, as a single underlying factor.However, the third perspectivethe bifurcated trust thesisrequires the disaggregation of political trust.There is a literature that explores the extent to which the concept of trust can be validly disaggregated.The first concern is whether trust is a general characteristic of individuals.That is, whether individuals are simply more (or less) trusting as a general personality trait, thereby expressing roughly the same level of trust across-the-board for both social and political objects of trust.Uslaner contends that political trust is fundamentally different from social trust with each based on different foundations, 29 an assertion supported by empirical research that uses factor analysis and finds clear distinctions between measures of social and political trust. 30oncerning political trust specifically, there has been a similar debate on whether it is a single underlying attitude encompassing a wide range of political institutions or whether the concept can be further disaggregated.While factor analyses find a distinction between social and political trust, they also suggest no such distinction can be made between objects of political trust.That is, factor analyses of citizens in democracies suggest that the levels of trust in various political institutions (i.e. the government, parliament, political parties, courts, police and civil service) are all indicators of a single underlying factor (or unobserved variable) of political trust and therefore the indicators can, to a certain extent, be considered empirically interchangeable. 31onsequently, these results suggest that many citizens in democracies do not make strong distinctions between political institutions. 32onversely, recent empirical research calls into question the assumption of the unidimensionality (or aggregate nature) of political trust.Marien finds that, among established democracies in Europe, the factor loadings of implementing (i.e. less partisan) institutions are weaker than the factor loadings of representative (i.e.partisan) institutions, a finding corroborated by Cinar and Ugur-Cinar in comparing Cronbach's alpha statistics. 33This, Marien notes, follows the literature in that citizens of democracies hold different expectations of implementing and representative institutions, with the former expected to be impartial and the latter partisan.While Zmerli and Newton suggest that distinctions between political (i.e.partisan) and impartial (i.e. less partisan) institutions are theoretically important, they acknowledge that factor analyses find that only one latent dimension appears to underlie objects of political trust. 34However, using Mokken scale analysis, they find that political trust is hierarchically ordered within democracies, with impartial institutions (the police, courts and civil service) trusted more.Van der Meer and Ouattara's study of democracies replicates Zmerli and Newton's findings in confirming the hierarchical nature of political trust within each country. 35Moreover, they find that political trust does not meet the more stringent criteria for unidimensionality using a Rasch model in that the hierarchy of trust in various political institutions is not identical across all countries.However, they do find that trust operates somewhat similarly in almost all countries with the more politicized institutionsparliament, government and particularly political partiesthe least trusted, and the more impartial institutionsthe civil service, the courts and particularly the policethe most, albeit with different rankings of institutions within each of the two groups. 36Using multiple group confirmatory factor analysis, Breustedt finds empirical support for a two-dimensional conceptualization of political trust within democracies. 37That is, she finds that political trust is best modelled as two latent factorsone that includes representative or partisan institutions (i.e.political parties, government, and parliament) and another that comprises implementing institutions that maintain order and implement the law (i.e. the courts and police). 38onsequently, Newton, Stolle and Zmerli conclude that "political trust is not a single entity" and that "[t]rust in the more neutral and impartial institutions of the state (the courts, police, and civil service) tends to go together and be stronger than trust in the organizations of government in the forms of cabinet, parliaments, and political parties." 39The recent literature therefore suggests there are both theoretical and empirical reasons to disaggregate political trust within democracies into a two-dimensional model corresponding to partisan and less partisan institutions.

Measurement equivalence and disaggregating trust in non-democracies
Most of the literature pertaining to the disaggregation of trust has focused either exclusively or primarily on attitudes within democracies.While a distinction between partisan and non-partisan institutions may be relevant within democracies, such a distinction may not be relevant to citizens in non-democratic countries.Should that be the case, it would be difficult to establish measurement equivalence of disaggregated political trust across regime types.The literature suggests that "political trust is likely to have a different meaning in autocracies and democracies" and therefore cross-regime comparisons may result in construct bias. 40Specifically, past research suggests that a unidimensional political trust model is likely more appropriate for non-democratic regimes, as the difference between partisan and non-partisan institutions is largely irrelevant given the difficultly of separating any political institution from the incumbent government. 41For instance, in a one-party state the distinction between the government (i.e. the ruling party) and the civil service, the police or the courts is effectively meaningless if these latter institutions lack independence and can be controlled by the regime without restraint.Park's empirical study of the Asia-Pacific region also finds that there is less variation in trust among political institutions in autocracies as, in general, all institutions enjoy comparatively stronger support. 42However, as opposed to the hierarchy found in democratic countries, Park also finds thatwith some exceptions -"partisan" institutions (the government, parliament and political parties) tend to enjoy the most trust in autocracies.Therefore, given the aforementioned two-dimensional nature of political trust in democracies, measurement equivalence of political trust across regime types is an assumption unlikely to be met.This lack of comparability presents a challenge to conducting unbiased cross-national studies of political trust in samples including both democracies and non-democracies.

Comparative strategy and hypotheses
Despite the challenge of establishing measurement equivalence, we maintain that it is defensible to compare political trust across regime types with a comparative strategy that takes this into consideration.Specifically, we contend that comparing factor loadings and mean trust scores of each political institution between democratic and nondemocratic regimes will yield valuable information.While these comparisons may indicate measurement non-equivalence across regime types, they nonetheless provide important insight into the specific differences in composition of political trust between democracies and non-democracies, thereby allowing for more accurate interpretations to be drawn from the results.In other words, although construct bias of political trust is expected to be present across regime types, this strategy enables us to know the manner in which this bias operates, assisting us in our endeavour to draw valid and meaningful conclusions.
To investigate the various perspectives of the relationship between democracy and political trust outlined above, we test three hypothesesone using the aggregated measure of political trust (encompassing all six political institutions in the analysis) and two using disaggregated measures of trusttrust in partisan and non-partisan political institutions.
H1: Democracy is negatively associated with political trust (aggregate measure).
Our first hypothesis (H1) examines the critical citizens (or low trust) thesis by testing whether democracy is negatively associated with the aggregated measure of political trust.In doing so, we simultaneously test the opposing perspective, the high trust thesis, that expects democracy to be positively associated to political trust.H2: Democracy is negatively associated with trust in partisan political institutions.
We then disaggregate political trust into measures of trust in partisan and non-partisan political institutions.As generalized political distrust across-the-board may not bode well for democratic stability, these disaggregated measures of political trust allow for a test of the bifurcated trust thesis to confirm whether lower levels of trust are concentrated in specific political institutions.As this perspective posits that political conflict is channelled into partisan institutions, we expect the association between democracy and trust in partisan political institutions to be negative.H3: Democracy is positively associated with trust in non-partisan political institutions.
The logic of the bifurcated trust thesis also suggests that trust in non-partisan institutions is expected to be higher in democracies, as conflict is theoretically channelled away from these institutions, thereby capturing the beneficial effects of democratic rights and freedoms on trust.

Political trust
Following the widely used conceptual framework of political support, 43 political trusta component of political supportis operationalized by the public's confidence in political institutions.In keeping with recent cross-national empirical research, we specifically focus on six political institutionsthe civil service, police, justice system/courts, parliament, government and political parties. 44These data are taken from the World Values Survey (WVS) and the European Values Study (EVS), with confidence scores reverse coded to range from 1 ("none at all") to 4 ("a great deal"). 45We draw upon all available surveys conducted between 1981 and 2020.

Democracy
We follow Dahl's conceptualization of democracy.He suggests that although no contemporary political system meets the ideal of democracy, some politieswhich he labels polyarchiesmore closely approximate the democratic ideal to the extent they meet the following criteria: elected officials, free and fair elections, inclusive suffrage, the right to run for office, freedom of expression and association, and access to alternative sources of information. 46Consequently, we use the Variety of Democracy Project's Electoral Democracy (or Polyarchy) Index to measure the level of democracy of each country. 47t is a continuous measure combining the above criteria, ranging from a low of 0 to a high of 1.While we contend that democracy is best measured as a continuousrather than a dichotomousvariable, we nonetheless use a threshold of 0.5 to distinguish democracies from non-democracies.Notably, as this value is somewhat arbitrary, we tested various thresholds, none of which substantively changed the results presented below.

Country-level controls
The literature posits that macroeconomic factors likely influence political trust and, relatedly, that high levels of political trust in non-democracies may be the result of strong economic performance. 48Consequently, we control for three macroeconomic factors: 1) the level of economic development as measured by the natural log of GDP per capita; 49 2) economic growth as measured by dividing the current GDP per capita figure by the figure five years prior; and 3) economic inequality as measured by the GINI coefficient. 50ertain characteristics of the state can also affect political trust.Corruption is identified in the literature as an important cross-national correlate of political trust. 51As such, we control for corruption using the Political Corruption Index from the Varieties of Democracy Dataset, which ranges from 0 to 1. 52 We also include a measure of state capacity from the same dataset, as it is considered a crucial determinant of political trust, particularly in some non-democratic states. 53Wimmer suggests that states that do not follow the modern nationalist principle of ethnic self-rule are likely to be perceived as less legitimate, which may result in less confidence in state institutions. 54As such, we draw upon the ethno-political exclusion indicator from the Ethnic Power Relations Dataset as a control, which measures the proportion of the population belonging to politically relevant ethnic groups excluded from state power. 55We also include the State Antiquity Index as a control, in the event that the number of centuries that a country experienced complex political organization has a cumulative effect on contemporary perceptions of the state. 56elf-censorship due to fear is often cited in the literature as a potential source of response bias (or preference falsification) 57 in studies of political trust in non-democracies. 58That is, citizens may be reticent to express a lack of confidence in political institutions in states (primarily non-democratic) that do not tolerate dissent.Consequently, in these states, higher levels of political trust may be an artifact of self-censorship, leading to some scepticism of the validity of past findings of a negative association between democracy and political trust.To account for this type of response bias, we include the Political Terror Scale as a control variable. 59This scale measures the level of political violence perpetuated by the state against its citizens, defined as "violations of physical or personal integrity rights carried out by a state (or its agents)," including "state-sanctioned killings, torture, disappearances and political imprisonment." 60The scale is based on coding data from the U.S. State Department Country Reports on Human Rights Practices and ranges from 1 ("rule of law") to 5 ("pervasive terror").As Kuran notes, measurement techniques that attempt to account for preference falsification are often imprecise, imperfect and crude. 61While we do not claim that our measure is an exception, we nonetheless contend that it is reasonable to assume that the falsification of self-reported levels of political trust due to concerns of potential reprisals would be more prevalent in countries with a high level of political terror. 62

Individual-level controls
Following the literature, we include a number of individual-level demographic control variables to better predict the true effect of the country-level variables on political trust. 63Drawing upon the WVS and EVS survey data, our models include controls for sex, age, education, income, religiosity and interest in politics.We code sex as 0 (male) and 1 (female) and education as a binary variable with 0 (no university education) and 1 (some university education or higher). 64While postsecondary education has been found to be positively associated with political trust in some countries, 65 some expect it to be negatively related to political trust through cultivating the capacity of critical analysis of politics. 66We use the income level scale common across both the WVS and EVS that relies on a self-reported relative income ranking as either 1 ("Low"), 2 ("Medium") or 3 ("High").Following research suggesting that atheists tend to have less trust in political institutions, 67 we create a religiosity index ranging from 1 to 4, with 4 representing the highest level of religiosity. 68We also include a control for respondents' reported interest in politics, which ranges from 1 ("very interested") to 4 ("not at all interested"), about which the literature is divided in its expectations of either a positive or negative effect on political trust. 69Following past research suggesting that conservative value orientations/ideological self-placement on the right are found to be positively correlated with political trust, 70 we conduct a robustness check controlling for ideological disposition (i.e.respondent self-placement on the left-right spectrum, see Table S2).While the results are consistent with expectations, the inclusion of the ideological disposition variable does not significantly alter the effect of democracy.We do not include the control in the models presented below, given that it's inclusion substantially reduces the number of observations.
Beyond demographic variables, we also control for respondents' general trust profilesa factor that may influence political trust.First, we control for generalized social trust.We draw upon the standard dichotomous question of whether "most people can be trusted," which is considered a reliable and valid measure of trust across linguistic and cultural divides. 71Following the recent literature, 72 we create an alternative measure of generalized social trust comprising the average scores of "trust in people you meet for the first time," "trust in people of another religion," and "trust in people of another nationality," each of which range from 1 ("trust completely") to 4 ("do not trust at all").However, as there is no substantive difference in the results between the two variables, and as the alternative measure has less country coverage, we use the standard binary measure of social trust.
Second, following Wu and Wilkes, we control for the response pattern on questions related to trust. 73That is, we include a measure of individual variability in trust responses across institutions.This provides a measure for the extent to which respondents distinguish between political institutions, which tends to occur more frequently in democracies.This variable therefore also concurrently controls for potential acquiescence or dissent response biasthe tendency of some respondents to express the same level of confidence across all institutions. 74Consequently, our "Trust Variation" variable is the standard deviation of each respondent's trust scores across the six political institutions under investigation.Refer to Table S3 for the descriptive statistics of each variable and Table S4 for the zero-order correlation matrix.

Methods
We merge these data sources and construct a panel dataset includes 247785 observations from 82 countries over the period 1990-2020, with 217 country-years in the models. 75Given that our dataset includes both individual-level and country-level observations, we use multilevel models to assess the relationship between political trust and democracy. 76

Results
Before examining the relationship between democracy and political trust, we assess whether social and political trust are empirically distinct in our sample.The results of the principal component factor analysis (see Table S6) suggests that, consistent with past studies, there are two clear latent factorsone corresponding to social trust and another to political trust. 77Narrowing our focus to political trust, the Intraclass Correlation Coefficient (ICC) of the unconditional model is 0.192, indicating that slightly under 20% of the total variance of political trust in the sample is explained by betweencountry variation, thereby supporting the use of multilevel modelling.
Figure 1 presents a visualization of the relationship between the two focal macrolevel variablesdemocracy and political trust.Consistent with past research, the scatterplot indicates a negative bivariate association.Notably, China (plotted in red) is not alone among the low-democracy, high-trust countries and is not the sole driver of the negative relationship. 78To examine whether this relationship holds while accounting for a comprehensive set of both country-and individual-level controls, we turn to multilevel models.
Coefficient plots of the full multilevel models of (aggregated) political trust, partisan political trust and non-partisan political trust are provided in Figure 2, while Table 1 presents the detailed numerical results for the aggregated political trust models.The results indicate that, accounting for a variety of country-level and individual-level factors, democracy has a substantial and statistically significant negative effect on political trust.Model 1 of Table 1 suggests that, after controls, the political trust score of a completely democratic country will be on average 0.54 points lower (on the 3-point scale) than a pure autocracy.This effect size increases marginally after adding additional controls in Models 2 and 3. To ease interpretation of the relative effect sizes of each variable, we also include the standardized beta coefficients of each model.The beta coefficients suggest that democracy consistently has the largest effect on political trust, followed by income inequality and interest in politics.

DEMOCRATIZATION
In Model 2 (Table 1), we add the control for response bias due to self-censorship out of fearthe political terror scale.Somewhat unexpectedly, the political terror scale has a significant negative relationship with political trust, rather than the positive relationship that one would expect in the presence of self-censorship (expressing more positive responses than would otherwise be the case) in states that do not tolerate dissent.However, this result does reflect the intuitive interpretationthat there is less political trust in states that terrorize its citizens.The political terror scale becomes the fifth most influential variable in the model, as measured by the beta coefficients.Given these results, controlling for the political terror scale surprisingly amplifies, rather than decreases, the negative effect of democracy on political trust. 79n Model 3, we add the two controls measuring respondents' trust profiles.While both social trust and trust variation have a significant positive relationship with political trust, the influence of social trust is more consequential, surpassing the political terror scale as the fifth most influential variable modelled.While the addition of these control variables slightly reduces the coefficient of democracy as compared to Model 2, the effect of democracy remains substantial.Additionally, following expectations, the mean trust variation score is higher in democracies (0.479) than nondemocracies (0.421).Concerning the other controls, the results are consistent with expectations, with the exception of income inequality. 80Another noteworthy finding is that corruption is statistically insignificant and has little explanatory power, contrary to the literature cited above. 81Two individual-level variables of particular note are interest in politics and religiosity.The standardized coefficients suggest that these two individual-levels variables are much more influential than the "standard" demographic variables (i.e.age, sex, education and income).As longitudinal multilevel models incorporate both within and between effects for each coefficient, we breakdown the effect of democracy into its component parts.In the model, not shown, democracy has a within effect coefficient of -0.446 and a between effect coefficient of -0.133, suggesting that the effect of democracy is primarily driven by within effects, thereby significantly reducing the likelihood that its negative effect is due to omitted variable bias at the country-level.These results are consistent with the expectations surrounding the "post-honeymoon" effectthat is, after the initial transition to democracy, political trust tends to decrease and then stabilize at lower levels with democratic consolidation. 82In a separate model, not shown, we add the year as a covariate to test whether political trust has decreased over time.The results suggest that, ceteris paribus, political trust is, on average, lower in more recent years; however, the impact of year is small (B = -0.002,β = -0.016)and its inclusion does not substantially alter the effect of democracy.Interestingly, while the impact of year is smaller in the model limited to democracies, it is substantially larger in the model limited to non-democracies (B = -0.009,β = -0.097).
While the results in Table 1 examine political trust in aggregate, we disaggregate political trust into measures of trust in partisan (the parliament, government and political parties) and non-partisan (the civil service, police, justice system/courts) political institutions to test the bifurcated trust thesis.We run principal component factor analysis on our six measures of trust in political institutions to determine whether this disaggregation is empirically justifiable.The results (see Table S7) are consistent with past findings, namely that trust in the various political institutions comprise only one latent factor; however, that the factor loadings suggest a hierarchy between partisan and non-partisan institutions, which does not differ much among democracies and non-democracies.While factor loadings, particularly among democracies, suggest that political parties' contribution is similar to that of the courts and civil service, Table 2 indicates a clear distinction between partisan and non-partisan institutions, with the exception of government in non-democracies.We therefore examine the effect of democracy on trust in partisan and non-partisan institutions separately.
Figure 2 (see Table S8 for detailed numerical results) presents the results of the full multilevel model of trust in partisan political institutions.Similar to those of the aggregated political trust measure, they indicate that democracy has a substantial and statistically significant negative effect on trust in partisan political institutionsagain led by the within effectand is the most influential variable across all models. 83These results are consistent with the bifurcated trust thesis.The influence of the other variables in the models are similar to the models of the aggregate measure of political trust, with two notable exceptions.First, trust variation surpasses social trust as the fourth largest standardized coefficient.This result suggests that, after controls, higher individual variability in trust responses across partisan institutions is associated with a higher average level trust in those institutions. 84However, the variation is calculated across three, rather than six, political institutions and therefore is not fully comparable to the results in Table 1.Second, consistent with the literature, corruption is statistically significant across all models and has a substantially larger effect size.However, corruption is positively, rather than negatively, associated with trust in partisan institutions, suggesting that partisan political trust is, on average, higher in more corrupt countries.
While these results are consistent with the bifurcated trust thesis, the results of the non-partisan political trust models are not.Figure 2 (see Table S9 for detailed numerical results) suggests that democracy has a significant negative effect on non-partisan political trust, rather than the predicted positive effect, driven by a comparatively large within effect, and is again the most consequential variable in the model. 85In fact, the difference between the two democracy coefficients in the partisan and nonpartisan models is statistically insignificant. 86Concerning the other covariates, corruption has a statistically significant negative effectas opposed to the partisan models, which is consistent with the expectations in the literature.Moreover, the impact of political terror is more consequential in the non-partisan models, while the influence of trust variation is comparatively minimal, with a mean trust variation score of 0.261 in democracies and 0.231 in non-democracies.

Discussion
The results suggest that there is a strong and robust negative relationship between democracy and political trust.This negative relationship is consequentialdemocracy consistently has the largest effect of any of the variables modelled.These findings therefore lend support to the first hypothesis, that democracy is negatively related to political trust, thereby supporting the critical citizens thesis.This finding makes an important contribution to the literature in confirming that the negative association between democracy and political trust holds in the presence of a variety of countryand individual-level controlsincluding controls for potential response bias (i.e.preference falsification) and individual trust profilesand is driven primarily by withincountry effects.Therefore, the negative relationship between democracy and political trust cannot be easily dismissed as an artifact of model misspecification or response bias.
While the association is clear, its interpretation is less so.Whereas the findings lend support to the view that expects democracies to have comparatively lower levels of trust, they do not necessarily fail to support the alternate perspectivethat democracies require a high level of political trust for regime stability and the effective exercise of authority.Indeed, while the negative relationship lends some support to the critical citizens thesis, it is also compatible with the high trust thesis.That is, democracies may, in fact, require a high level of political trust and the current negative association between the two suggests that democracies may be going through a period of instability and decline.Without a crystal ball to see the future, each of the two different interpretations of the results are tenable.Consequently, the finding of a negative association between democracy and political trust, no matter how robust, does not resolve the debate.
The empirical analysis of the bifurcated trust thesis, however, provides further insight into whether the negative association between democracy and political trust is potentially problematic for democracy or a symptom of its vibrancy.The results support the second hypothesis that democracy is negatively associated with trust in partisan political institutions.These findings are consistent with the expectation that trust in these institutions is lower in democracies, as their function is to institutionalize political conflict, thereby becoming a focal point for distrust.While low trust in partisan political institutions should therefore pose few problems for democracy, generalized political distrust across all political institutionsspecifically across non-partisan institutions (arguably a proxy of diffuse political system support)is potentially more worrisome and may lead to democratic instability and deconsolidation.That is, if confidence in non-partisan institutionsthose institutions that provide public goods and enforce the lawis significantly lower in democracies, this could indicate a potential problem that may not augur well for the future.The third hypothesis, which posits that democracy is positively associated with trust in non-partisan political institutions, is not supported by the data.For both partisan and non-partisan political institutions, there is no statistical difference in the effect of democracyit is negative and the most consequential predictor of all the covariates.For proponents of the bifurcated trust thesis, these results are concerning for democracy.
Beyond the influence of democracy, the analyses provide additional insight into the factors that influence political trust.First, multilevel models allow for a variance analysis using the ICC statistic, which suggests that individual-level variation explains the lion's share of the variance of political trust (roughly 80%), with country-level variation accounting for roughly 20%. 87Consistent with the literature, the data therefore suggest that, as a whole, microas opposed to macrofactors are much more influential in determining a respondent's level of confidence in political institutions.Nonetheless, of the variables modelled, two among those with the largest effect sizesdemocracy and income inequalitypoint to the importance of country-level variables.Second, the analyses suggest that the effect of corruption differs across different types of political institutions.Namely, the effect of corruption has a positive significant effect on trust in partisan institutions, with a nearly identical negative effect on trust in non-partisan institutions.These effects are likely countervailing, resulting in the comparatively small effect of corruption on the aggregate measure of political trust.These findings provide a potentially interesting avenue for further investigation.Regardless of the cause of the differential effect of corruption, the findings further underscore the value of separately analysing trust in partisan and non-partisan political institutions.
The empirical analyses also provide insight into the political trust institutional hierarchy among non-democracies.Drawing upon a larger sample of non-democratic regimes, the findings corroborate the main thrust of Park's findings, that trust is higher in partisan institutions in non-democracies. 88In fact, the police is the only institution of the six with higher mean trust in democracies as compared to non-democracies.However, with the exception of the government, the mean score hierarchy across both democracies and non-democracies is similar, with democracies having more variation, as expected.Interestingly, mean trust in the courts in non-democratic countries is slightly higher than in the government, although the difference between the two is negligible.While these results indicate a certain degree of measurement non-equivalence (or construct bias) of political trust across regime types, they also shed light on the precise differences in the institutional trust hierarchy between democracies and non-democracies (and the variation within each of the two groups), which is fundamental to take into consideration when interpreting results from cross-regime comparisons.
Another critical component of drawing valid conclusions from comparing political trust across regime types is accounting for potential response bias (or preference falsification) resulting in the overreporting of political trust out of fear of reprisals by the state.It is difficult to account for this type of bias, as respondents who knowingly engage in it would likely not openly admit it.Our control for this type of preference falsification, the political terror scale, has a robust negative association with political trust.There are two different interpretations of this result.First, a positive association would have arguably provided clear evidence of preference falsification, while validating the use of the political terror scale as a proxy for this form of bias.Thus, the negative association may indicate that the political terror scale is a poor measure of response bias and simply reflect that respondents express less political trust in states that terrorize its citizens.Another interpretation is that the political terror scale is an appropriate measure of response bias, but this type of bias is less frequent than anticipated.This would be in line with research in countries such as China, which indicate that the high levels of reported political trust are not due to fear. 89Consequently, it could be that the political terror scale is an effective control for bias, but its effect is outweighed by a higher frequency of respondents providing truthful responses in states that do not tolerate expressions of dissent.While both interpretations are plausible, we contend that the inclusion of political terror scale does, at a minimum, provide some additional confidence in the findings.

Conclusion
Are these results worrisome for democracies?Proponents of the critical citizens thesis suggest that political distrust plays an important role in democracies.Therefore, the finding of comparatively low levels of trust across all political institutions trust should not threaten democratic stability.These findings are also compatible with Rivetti and Cavatorta's claim that political trust is not an inherently democratic trait. 90They contend that an evaluation of trust or confidence in political institutions does not necessary mean that these institutions are perceived as positive or beneficial, rather it may be an assessment that these institutions act in a predictable and routine manner, even if their actions are largely considered unethical or illegal.Consequently, these views suggest that the findings are unsurprising and do not pose an existential threat to democratic political systems.
Proponents of both the high trust and the bifurcated trust theses would suggest otherwise.The former group submits that democratic political institutions require relatively high levels of public trust.While the latter group acknowledges that political distrust plays a critical role in democracies, scholars such as Warren nonetheless posit that political trust is indeed an inherent feature of democraciesmore specifically, of non-partisan political institutions. 91Accordingly, the finding of lower levels of political trust across both categories of political institutions in democracies suggests there is cause for concern.While these results do not definitively resolve the debate of whether the negative association between democracy and political trust is problematic in the long-term, they do provide an important contribution by separately analysing political trust in partisan and non-partisan institutions across the entire spectrum of regime types.In doing so, these findings provide additional support to the view that perennially low levels of political trust, particularly in non-partisan institutions, may be more detrimental to democracy than much of the literature suggests.

Notes Funding
This work was supported by the MITACS Research Training Award.

Notes on contributors
Andrew Dawson is an Associate Professor of Sociology at York University, Canada.His research interests include political sociology, democracy, political and social trust, state legitimacy, violence and the rule of law.
Isabel L. Krakoff is a PhD candidate in the Department of Sociology at York University.She completed her MA in international affairs with a concentration in global gender policy in 2018 at George Washington University in Washington, D.C. before moving to Canada for her PhD.She is currently working on her dissertation exploring the intersection of right-wing populism and human rights claims.Her research interests include political sociology, the study of race and racism, critical sexuality studies, global gender policy, and mixed methods research.

Table 2 .
Mean trust score by political institution.