Introduction

Trust in each other and our political institutions is argued to be at the heart of social cooperation. Yet, trust in core institutions and actors across much of the democratic world, with some exceptions, is stagnant or in decline, reflecting dissatisfaction with (perceived) policy performance and, more worryingly, how democratic politics is conducted (Bøggild, 2020). Does this apparent decline in trust have consequences for political life?

Political scientists seem to think so: in the foundational literature, Easton (1975) stated ‘no set of incumbent authorities in modern mass societies’ could make decisions or implement policy without trust—and ‘few systems [were] able to survive long’ without it. Miller (1974) similarly posited that low trust would lead to ‘radical change’ that could not be assuaged by ‘throwing the rascals out’. Consistent with these wide-ranging concerns, the vast and long-standing academic literature on the trends, correlates and causes of political trust (e.g. Citrin & Stoker, 2018; Cole, 1973; Easton, 1965) often claims that political trust has potent effects on outcomes such as voter voter turnout, policy preferences and compliance with public policy (e.g. Dalton, 2004; Hetherington, 2005; Norris, 2011; Zmerli & van der Meer, 2017).

Despite this elite and scholarly interest, there is surprisingly little systematic empirical evidence on these claims. Prominent scholars have repeatedly argued that the literature is without ‘reliable knowledge’ (Marien & Hooghe, 2011) or ‘systematic information’ (van der Meer & Zmerli, 2017) about these consequences, and that further research in the area is vital (Weinberg et al., 2021). Despite long-standing assumptions about the importance of political trust and concern about its perceived decline, we know very little about whether this importance and concern is warranted. Given that the literature on political trust has attracted at least 30,000 citations up to 2021,Footnote 1 this is a significant and meaningful gap in our understanding.

This paper addresses this gap by systematising and summarising what we know about the consequences of political trust through a comprehensive, quantitative meta-analytic review. I first conceptualise political trust and then provide a synthesis of the theoretical reasons we would expect trust to matter and for what outcomes. I then present a meta-analysis of the existing empirical literature. From an initial pre-defined literature search, 9139 papers are screened based on pre-defined inclusion criteria. The final analysis includes 61 papers published between 1995 and 2022 reporting 329 coefficients using 3,616,404 observations, all of which use trust as a predictor for a range of potential outcomes.

Summarising these standardised effect sizes indicates that trust has a small-to-moderate relationship with most included outcomes, such as preferences over environmental policy.

(rz = 0.9), immigration policy (rz = 0.09) and propensity to vote for challenger parties (rz = − 0.05). The only null effect is on informal participation such as protests (rz = 0.01). Alongside these estimates, the meta-analysis highlights key spatial and methodological gaps in the literature: approximately 78% of the data is from Europe or North America and 77% of the data is from cross-sectional (as opposed to experimental or panel) analyses. In a second step, I use meta-regression models to understand how these effect sizes vary across key moderators. The regressions indicate that research design features such as regions, response scales, the object of trust (e.g. Parliament, politicians) and modelling strategy (e.g. logistic, ordinal and multilevel) explain little variation in the effect sizes, whilst the choice of outcome does. This indicates that variation in effect sizes is likely due to some genuine, substantive differences. In the Appendix, I also test whether the literature is attenuated by reporting bias, in which significant or substantial effects are published to a greater extent than non-significant or small effects. The results suggest that there is likely some reporting bias in the data, but these seem unlikely to affect overall effect sizes. Overall, these results show that trust matters for a broad range of important outcomes related to what people want from their political systems and how they interact with it. The effects are widespread, from political participation to underlying policy preferences.

This analysis brings vital clarity to understanding the importance of political trust. If trust were to decline, as observed in the United States, United Kingdom, and other established democracies, the literature predicts declining turnout, greater support for challenger parties, reduced compliance with public policy, and more conservative preferences over government policy from the environment to public spending. The conclusions are therefore important to numerous fields of political science, such as political behaviour and public policy. Conceptually, the results suggest that trust is indeed a ‘reservoir of support’ (Easton, 1975) for governments, and that its decline does not signal the rise of an engaged but critical citizenry (Norris, 2011). Instead, it is more likely to result in an ‘exit’ from political life (Hirschman, 1970), as declining formal participation and refusal of government action is not replaced by informal participation. Whilst the importance of trust also lies in its normative role as a measure of political legitimacy, the empirical literature indicates that trust is also a vital resource for governments to maintain stability, oil the wheels of policy, and to facilitate a vibrant civil society.

Despite the long history of research on the causes and measurement of political trust, research into its consequences has been lagging, with strong claims about its relevance often not well supported. This paper lays out the theoretical mechanisms linking trust and its various consequences, provides a strong empirical and transparent evidence base for these claims, and charts a clear way forward for this disparate literature.

The Conceptual Foundation of Political Trust

To develop expectations about what consequences political trust is likely to have, and interpret any results thereof, the conceptual boundaries need to be clarified. What is meant by political trust? What does it signify, and how is it relevant for political life? Though it is not an uncontested concept, the dominant approach is that ‘political trust’ is a dimension of political support, the latter of which consists of attitudes and beliefs that individuals have towards the (nation) state. This Eastonian conceptualisation (Easton, 1965) has been developed by many others, particularly Norris (2011) and Dalton (2004). This framework is bounded by ‘diffuse’ and ‘specific’ forms of support, in which the former includes attitudes such as patriotism and support for democratic ideals, and the latter is a targeted evaluation of specific incumbents—such as presidents and prime ministers—or other representatives. Within this spectrum, trust is a ‘mid-range’ level of support, linking the overarching principles of a regime and the principal actors within the system (Zmerli & van der Meer, 2017, p. 4). Political trust is almost always relational, in the sense that it is about trust in a specific actor or institution, usually the core representative institutions such as parliament, government, parties, politicians in general, and sometimes including non-representative institutions such as the legal system and civil service. Political trust may also be referred to as ‘institutional trust’ or ‘confidence’, though these may also refer to broader attitudes such as satisfaction with democracy, and there is a lively empirical and conceptual debate about the differences between trust, confidence, and satisfaction.

In recent years, there have been growing calls for a more multidimensional understanding of the broad concept of ‘political trust’. Bertsou (2019), for example, argues for a distinction between trust and distrust (see also van de Walle & Six, 2014), where the former is a relatively settled affective or evaluative positive evaluation and the latter an affective or evaluative perception of untrustworthiness of institutions. The implication therein is that these are not just the absence of one another—a lack of trust does not mean there is distrust—but rather that they are distinct concepts. Others have moved towards conceptualising and operationalising a ‘trust family’ which includes trust, distrust, and mistrust (which is characterised by monitoring of the political system and updating of trust beliefs), and the proposition that these attitudinal dimensions have different consequences (Jennings et al., 2021a; Norris, 2022). Still others have highlighted a tension between ‘latent trust’ and ‘explicit distrust’, where some individuals may harbour relatively trusting attitudes even if they provide ‘distrusting’ survey answers (Intawan & Nicholson, 2018). Yet, as this paper will go on to show, these conceptual and even empirical contributions are not yet reflected in the dominant survey measurement instruments, which almost entirely rely on questions narrowly measuring ‘trust’.

These conceptual and measurement ambiguities have consequences for debates over the normative standing, and desirability, of political trust. Foundational work held firm to the claim that it was positive for the political system and, therefore, its decline a concern (Easton, 1965; Miller, 1974). However, this has also been called into question, particularly in the philosophical literature (Warren, 1999), and most recently restated by Norris (2022). This is evidenced by the high levels of political trust in non-democratic, high-corruption states such as Russia, and nefarious actors in democratic states; a clear example of the two-faced nature of political trust is that trust in President Donald Trump increased belief in conspiracy theories and anti-vaccination during the COVID pandemic, though trust in core political institutions had the opposite relationship (Devine et al., 2023). Trust therefore is not only a positive: it may well have corrosive as well as constructive consequences for (democratic) societies.

Whilst this paper engages less with these conceptual debates given the empirical depth, I return in the concluding section to how the findings highlight the importance of these conceptual distinctions, and, primarily, the mismatch between the depth of these conceptual debates and the conceptual thinness of empirical work.

The Consequences of Political Trust: Theoretical Predictions

Political trust is a staple topic in political science research, a core indicator of political legitimacy, and is conceptually rich. As numerous review articles have pointed out, it is surprising that empirical research on whether it is related to outcomes of interest is surprisingly lacking (Carstens, 2023; Citrin & Stoker, 2018; Levi & Stoker, 2000; Uslaner, 2018; Weinberg et al., 2021). The empirical work that does exist is fragmented and rarely draws on similar theoretical frameworks. Individual studies explore potential outcomes from gun control in the United States (Ryan et al., 2020) to peace agreements in Colombia (Esparza et al., 2020) and how citizens react to political scandals (Dancey, 2012). Despite this diversity, scholars are converging on the most empirically studied outcomes. In a set of recent review articles and chapters, Zmerli and van der Meer (2017) identify compliance, participation, voting behaviour, and policy preferences as the most common outcomes; Citrin and Stoker (2018) similarly highlight policy preferences, participation, vote choice and compliance; Uslaner (2018) addresses policy preferences and electoral behaviour (see also, Levi & Stoker, 2000). I follow these studies and describe the core theoretical predictions for policy preferences, political participation, and compliance; the latter also includes the recent literature on trust, compliance and the COVID pandemic. Finally, I address potential moderators, including the object of trust (such as parliament or government), spatial and temporal variation, and study design, which may condition the relationship between trust and its outcomes.

Before beginning, it is worth highlighting that the literature has not fully grappled with issues of reverse causality. It is plausible that trust is (reciprocally) caused by most outcomes studied. For example, Hooghe and Dassonneville (2018) find that vote choice can influence trust, which in turn influences vote choice; Haugsgjerd and Kumlin (2020) similarly show how policy evaluations influence trust and trust, in turns, influences policy evaluations. Whilst this paper takes as a starting point the theories that posit trust as a predictor rather than outcome, it is also recognised that the reverse relationship is at play. More research using panel data and experimental designs should begin to tease out the causal direction.

Policy Preferences

The consequences of trust for citizens’ policy preferences over a broad range of policy areas is one of the most theoretically and empirically well-developed literatures in this area. Trust is seen as a heuristic (Rudolph & Evans, 2005; Rudolph et al., 2017) which citizens use to judge whether to support the expansion of government or not. The central hypothesis is that trusting citizens are more likely to support the expansion of government activity, and this has led Hetherington (2005, p. 3) to argue that ‘declining political trust has played the central role in the demise of progressive public policy’.

This trust-as-heuristic approach has been extended in two ways. The first is the cost hypothesis: trust only matters when citizens are required to make ideological or material costs (Hetherington, 2005; Rudolph & Evans, 2005). The second is the salience hypothesis: trust does not matter equally for all policies, but specifically those which are salient to citizens (Hetherington & Husser, 2012). To provide an example of these hypotheses for preferences over government spending, trust is expected to matter most for spending when someone is an ideological loser (against expanding public spending), a material loser (will face higher taxes or other policy trade-offs with no or little gain), and if they care about the policy area. Whilst this provides the fundamental hypotheses across policy areas, others have extended them. Fairbrother (2017) highlights the importance of trust for believing information about the environment, whilst other researchers emphasise the time horizon of the policy, in which higher trust facilitates long-term policy making (Christensen & Rapeli, 2021; Fairbrother, 2019; Jacobs & Matthews, 2012).

Despite this, the existing literature’s fundamental claim is that trust is related to policy that expands government power or spending. This is hypothesised to be conditional on material and ideological costs as well as salience, and other policy features such as its time horizon. The empirical evidence, however, is mixed, with some finding weak or null results (e.g. Peyton, 2020). There is also no clear idea of whether the relationship is a meaningfully substantial one, or statistically significant but ultimately substantively small. By pooling across numerous studies, this paper addresses this gap.

Participation and Vote Choice

Participation can be separated into formal (such as voter turnout) and informal (boycotts, protest). Unlike for policy preferences, the fundamental hypothesis about the effects of trust for participation are not clear and indeed are conflicting; Gabriel (2017) argues that speculation about the relationship ‘outweighs by far’ sound empirical evidence. For formal participation like turning out to vote, trust may engender or reflect a commitment to formalised politics (increasing turnout) or instead complacency and satisfaction (reducing it) (Belanger & Nadeau, 2005; Citrin, 1974; Dalton, 2004). Whilst some authors (Hooghe, 2018) believe it can ‘safely be stated’ that trust has a positive effect on turnout, a metaanalysis on the numerous predictors of voter turnout suggest that trust has a null effect and the expected effect in only a third of the studies (Smets & van Ham, 2013). Despite voter turnout and trust being two of the most studied topics in political science, there is theoretical and empirical uncertainty about their relationship.

The same debate presents itself with with even greater intensity with respect to informal participation. The rise of ‘unconventional’ participation coincided with the collapse of political trust in the United States (Barnes & Kaase, 1979; Miller, 1974) and the focus on tectonic generational change in shifting repertoires of political behaviour (Inglehart, 1997). The initial hypothesis, consistent with the growth of protest movements in the United States, was that low trust facilitated a critical citizenry (Norris, 2011). However, it may also produce an exit from politics; citizens may simply not believe the political system is worth engaging in. Dalton (2004) provides an example of these conflicting findings: trust was negatively related to informal participation in some countries, positive in others, and unrelated still in others. Typically, it is argued that trust is positively related to formal participation but negatively to informal participation (e.g. Hooghe & Marien, 2013), even if these relationships vary across contexts (Valgarðsson et al., 2021).

The literature is clearer about the expectations of what voters do when they arrive in the booth. There is a consensus that trusting individuals are less likely to vote for ‘challenger parties’—those that reject the mainstream. Clearly, those who are less trusting of the political system are less likely to support incumbents and more likely to support those attempting to reform the system. This relationship is particularly strong for right-wing (populist) parties (Hooghe & Dassonneville, 2018; Hooghe et al., 2011; Jiang & Ma, 2020; Orriols & Cordero, 2016). Beyond challenger parties, trust has been shown to be positively related to incumbent voting and negatively to third-party voting in the US (Hetherington, 1999), and anti-incumbent voting in Canada (Belanger & Nadeau, 2005). To paraphrase Gabriel (2017), speculation about trust and participation, formal or informal, outweighs its empirical precision. This paper aims to provide the empirical precision.

Compliance

The early literature characterised trust as important for compliance with the law (Scholz & Lubell, 1998). The mechanism is trust as a heuristic, in which trust is used to judge whether accepting some risk and cost in a collective good—such as paying taxes—is likely to be returned in the future. If citizens do not believe that this is the case, it makes sense not to comply (Scholz & Lubell, 1998). The existing research has largely studied (intention to) pay (or avoid) taxes, claim welfare payments that citizens are not entitled to, and avoid transport fares (e.g. Letki, 2006; Marien & Hooghe, 2011). As noted, a similar framework has been applied to understanding trust and policy preferences (Hetherington, 2005; Jacobs & Matthews, 2017; Rudolph, 2003).

This literature was central to explaining the role of trust during the COVID pandemic. High trust was seen as importance to ensure compliance with public health regulations (Bavel et al., 2020; Devine et al., 2023), as well as believing in COVID-related information from authorities and vaccine uptake (Jennings et al., 2021a; Lindholt et al., 2021). An outpouring of empirical research studied compliance with vaccination, public health orders, travelling and social mixing and much else, and its potential moderators (Goldstein & Wiedemann, 2020; Jennings et al., 2021b; Wong & Jensen, 2020). The literature is still not systematised, though the large claim that ‘COVID is less deadly where there is trust’ is almost established wisdom (Bollyky et al., 2022). This paper assesses the importance of political trust during the COVID pandemic as a ‘most likely’ case of testing the compliance mechanism.

Potential Moderators

The volume of research on political trust means the literature is heterogeneous, which may influence results in meaningful ways. Researchers make numerous substantive and methodological decisions which may bias results in an as yet unknown direction. By pooling dozens of diverse studies, meta-analyses can help answer whether these decisions moderate the relationship between trust and its hypothesised outcomes.

Of particular interest is the object of trust. Studies measuring political trust vary on whether they ask about parliament, government, a particular office like president, and often use summative indexes of multiple objects, including objects like the civil service or courts. Trust in different objects may have different consequences. The more ‘diffuse’, institutional objects—like parliament—may have greater consequences for outcomes like political participation than trust in a particular incumbent. Trust in more specific objects like the President may have a greater correlation with policy preferences as the partisan cues and preferences align the preferences of respondent and government. Moreover, with the growth of established anti-establishment parties, trust in specific actors may have different, or even opposite, consequences than traditionally hypothesised.

There is also likely to be variation across space and time. The effect of trust on turnout for example is unlikely to be the same across countries diverse in electoral systems, regularity of elections, and levels of inequality. Those few papers that have explicitly explored crossnational differences have indeed found that the effect of trust varies across countries (e.g. Marien & Hooghe, 2011). For similar reasons, we would not necessarily expect trust to have similar consequences over time. The relationship between citizens and the political system has changed, with a shift towards greater dealignment and volatility changing the basis of voter decisions (Dassonneville, 2016). The effect on trust is unknown; it could increase trust’s importance as citizens rely on a simplifying heuristic, or decrease it as citizens rely less on top-down cues. That there is no difference, however, seems unlikely.

Finally, researchers make numerous design decisions relating to measurement and modelling strategy that have unknown implications for results. Researchers may opt for logistic regression over OLS, or multilevel rather than single level models. They may also opt for different datasets which measure trust in different ways. The commonly used Eurobarometer asks respondents whether they ‘tend to trust’ or ‘tend not to trust’ a battery of institutions, whilst the European Social Survey has an 11-point scale from zero (no trust) to 10 (complete trust); other surveys ask about ‘confidence’ rather than ‘trust’; and the American and British Election Studies ask whether respondents trust government ‘to do what is right’. It is unknown whether these design decisions have implications for the empirical literature on trust regarding its consequences or otherwise. In the analysis that follows, I use meta-regression to understand whether these and other design decisions moderate the relationship between trust and its theorised outcomes. In doing so, the analyses make a substantive contribution—understanding whether trust has the outcomes it is theorised to have—and an empirical contribution—whether design decisions moderate these relationships.

Data and Methods

To understand if and how trust is related to its hypothesised outcomes, I leverage hundreds of empirical analyses derived from diverse research designs to provide a holistic understanding of the relationship between trust, core outcomes, and potential moderators. This is called a meta-analysis, which is ‘defined as a systematic literature review supported by statistical methods where the goal is to aggregate and contrast the findings from several related studies’ (Viechtbauer, 2010).

Meta-analyses are not commonly used in political science. Blair et al. (2021) identified just five in five leading political science journals between 1999 and 2018, and just one of these used observational research. The use of meta-analyses has been increasing, with recent studies focusing on topics such as social trust and ethnic diversity (Dinesen et al., 2020), terrorism and political attitudes (Godefroidt, 2021), natural resources and conflict (Blair et al., 2021; O’Brochta, 2019), effects of political incivility (Van’t Riet & Van Stekelenburg, 2022), and gender and candidate choice (Schwarz & Coppock, 2022). Despite this, they are relatively sparse, often do not use observational data, and employ different analytical strategies. In this study, I follow existing best practice in political science and other fields. I ensure this in data collection by following PRISMAFootnote 2 guidelines, indicated in Fig. 1.

Fig. 1
figure 1

PRISMA text selection process

Full size image

Conducting a meta-analysis requires three steps. First, the data must be collected and coded. Second, the data must be processed to form a common effect size and coherent categorisation. Third, the resulting effect sizes must be analysed. I describe the my approach to these in turn.

Data Collection and Coding

The analysis in this paper is based on a data set of 61 separate texts that report 329 coefficients. These were collected through two main methods. First, I searched Web of Science with specified search terms (see Online Appendix A2). This resulted in 5511 initial documents. Second, I searched 10 leading political science journals which follows the process of Smets and van Ham (2013). No time period was imposed on this search. This led to 3624 initial documents. I obtained four additional documents through informal searches (such as following citations and asking colleagues). The total is 9139 initial texts.

I then screened all returns by title and abstract based on inclusion and exclusion criteria (see Online Appendix A). A fundamental criterion was that political trust had to be the predictor variable, with some hypothesised outcome as the dependent variable, with or without controls. Thus, all effect sizes can be expressed by the generic formula (here, ignoring more complex modelling choices and the presence of control variables):

$$Y = \beta_{0} + \beta_{1} \,Trust + \in$$

where Y is the outcome of interest, β0 the intercept, β1 the coefficient on the key variable, Trust, and the error term.

8871 texts (including 21 duplicates) were excluded at this stage, leaving 247 texts sought for retrieval. I excluded two at this stage as they could not be obtained. I screened 245 texts based on their full texts. 184 were excluded leaving 61 texts included for analysis. In the first stage, texts were excluded primarily because they did not study political trust or did not include it as an explanatory variable. In the second stage, texts were excluded for a number of reasons, all of which are detailed in Fig. 1.

All texts were coded on 39 characteristics of the studies using a pre-defined codebook that captured key details of the papers and their estimates, statistical information, and potential moderators. Three additional coders double-coded a sample of the final texts using the same codebook, with inter-coded reliability and differences listed in the Appendix.

Data Processing

I standardised all associations to a correlation coefficient. For coefficients that are the result of regression models, this is the partial correlation coefficient, which takes into account the sample size and number of other variables (i.e. the degrees of freedom) in the model that produced the coefficient. Following this transformation, all effect sizes were z-transformed to approximate a normal distribution, hereafter called Fisher’s Z. This means all effect sizes are unit free and bounded between − 1 and 1. Finally, effect sizes are reversed in direction where necessary, such that effect sizes indicate that higher trust is related to greater participation, support for policy, etc. See Online Appendix A for more details.

The primary analyses are based on seven core outcomes: environmental, immigration and spending policy, voter turnout, voting for challenger parties, informal participation, and outcomes related to COVID. These are in part drawn from the theoretical discussion previously, with the COVID category encompassing the ‘compliance’ studies due to the vastness of the recent literature. However, it is also important that each category has enough observations to mean meta-analyses are possible, and therefore all of these have 10 or more studies and observations. Results and details of others are available in the Online Appendix and replication material.

Modelling Approach

To estimate the average effect sizes, I use a random effects model which takes into account the study (such that estimates from the same study will be correlated) and, where relevant, the data source (such that observations from the same data source will be correlated). This can be represented by the following formula:

$$yi\, = \,\beta 0\, + \,bji\, + \,bki\, + \,eijk$$

where yi is the observed effect size, β0 is the intercept (the pooled effect), bji represents the random intercept for each data set, and bki represents the random intercept for each study. ei represents the error around each estimate derived from sampling error. This general model can be expanded by adding variables to explain the effect sizes, to which I return to.

Results

Describing the Field

The comprehensive systematic review provides a rich overview of existing studies, highlighting existing gaps in the literature. Five variables of interest are plotted in Fig. 2: the country of study, year of publication, the type of dependent variable used, the measurement of trust, and the object of trust. The countries of studies are shown in panel A. A plurality of the data is cross-national (two or more countries) (38%), whilst a fifth of the data comes from studies from only the United States (20%). 7.5% of the sample is from UK-only studies. There are a handful of studies from other countries, notably Canada, Denmark, Spain, and the Netherlands. Despite much of the data being from cross-national analyses, 51% of the data is from only European countries (likely due to the existence and availability of cross-national data), 25% is from North America (i.e. Canada or the US), and just 10% is genuinely global (data from three or more regions). Unsurprisingly, existing studies are largely confined to Europe and the United States.

Fig. 2
figure 2

Summary of data on key variables

Full size image

Panel B shows the growth in publication over time. This growth could be caused by multiple factors, such as the interest in trust over time, the amount of papers being published in general, or the volume of analyses reported in each paper. The earliest publication is from 1995, and whilst 2020 is an anomaly (likely due to COVID-related research), there has been a steady stream of research since 2010.

Panel C displays the broad clustering of dependent variables. 43% of the studies concern the effect of trust on some policy, such as preference, intensity, salience, and so on, across all policy areas. 17% concern voting, such as turnout or voting for challenger parties. 21% were classified as ‘other’; this includes outcomes such as perception of corruption.

The measurement of trust is typically via an index (panel D). This includes measures that are produced from a factor analysis of multiple trust measures, simple additive measures of multiple trust measures, or similar. The second most common measure are those that directly mention trust; a handful do not mention trust or confidence but interpret the measures as measures of trust, which are denoted as ‘neither’ in the graph (for instance, ‘would you say that the government is pretty much run for the benefit of most of the people?’). Panel E shows that the object of trust—what the respondent rates their trust in—is generally ‘other’, which is broadly an index of objects (for instance, a summative index of different institutions). ‘Government’ is the second most common, with ‘Parliament’ and ‘Politicians or parties’ used equally. Notably, almost all of the existing literature uses rather blunt measures of ‘political trust’, that are unlikely to capture the multidimensionality of the concept, or be able to tell us much about the differences between trust, distrust, or other subcomponents, which highlights a substantial mismatch between existing conceptual and empirical work.

This reveals a relatively limited geographical scope, a diverse range of measures of trust, and a diverse set of outcomes. In addition, there is a relatively limited range of research designs; a full 77% of the data comes from observational-correlational analyses, 17% use panel studies, and just 6% use either survey (4%) or lab (2%) experiments. There is quite a range of survey sources. 36% of the data points are from original data fielded by the authors, whilst 9% come from the ANES and 16% from the European Social Survey. The remaining come from a variety of data sources, such as data sets like the World Values Survey, Latin America Barometer, and the Icelandic National Election Survey (ICENES).

Finally, 96% of the observations are from individual-level data, with the remaining from country-level data (other aggregates were excluded). Of those individual-level analyses, the median average sample size is 2520, with a minimum of 256 and maximum of 194599.

The Consequences of Trust

Moving to estimating whether trust is related to its hypothesised outcomes, Fig. 3 displays the estimated ‘effect’ of trust derived from the random-effects meta-analysis on core outcomes. The X axis indicates the estimated correlation coefficient (rz, Fisher’s Z). The underlaid points are the individual observations, with the size indicating the weight they provide to the final estimate. The ‘overall’ row indicates the absolute effect size of trust across all outcomes in the dataset and therefore this coefficient can only be interpreted in terms of its magnitude rather than direction.

Fig. 3
figure 3

Effect of trust on outcomes of interest

Full size image

These results indicate that that trust has a small-to-moderateFootnote 3 and statistically significant (p < 0.05) effect on most of the outcomes studied. With regard to participation and vote choice, trust has a positive relationship with individual turnout (β = 0.06, CI = [0.01, 0.11]) and voting for challenger parties (β = − 0.05, CI = [− 0.06, − 0.03]). There is therefore more evidence that trust engenders a commitment to formalised politics—i.e., increases turnout—than it does complacency—i.e., reduces it. Meanwhile, there is clear evidence in support of the expectation that trust reduces support for challenger parties (or, conversely, higher distrust increases support for them). The results also suggest why there are competing expectations for informal politics: across the collected studies, there is an insignificant effect of approximately zero on informal participation. The competing findings may reflect that the true effect is zero, and we find relationships either side due to chance, or that the effect is highly conditional on political context.

In terms of policy preferences, there is evidence that trust does indeed boost support for more generous policy. There is a moderate-to-small effect on preferences over immigration (β = 0.09, CI = [0.04, 0.14]), spending (β = 0.057, CI = [0.017, 0.13], and environmental policy (β = 0.09, CI = [0.03, 0.15]). Whilst there are not enough studies to identify the mechanisms, there is suggestive evidence that trust acts as a heuristic mechanism for citizens: the policy area that is most likely to be driven by other strong cues, spending policy, has a substantially weaker relationship with trust than that less likely driven by other stronger cues, like environmental policy and COVID-related compliance and belief.

Consistent with the broader work on trust and compliance, there is a moderate effect of trust on COVID-related behaviour and attitudes (β = 0.13, CI = [0.03, 0.23]), providing substantial support for this expectation. However, it should be noted that there is substantial heterogeneity underlying this central estimate. The average magnitude of the effect of trust on all outcomes in the dataset is 0.08 (CI = [0.06, 0.10]).

To put these effect sizes into context, other meta-analyses in political science have found that terrorism has an effect of 0.09 on rally effects to 0.132 for a conservative shift (measured in Fisher’s Z, the same unit as here) (Godefroidt, 2021); that ethnic diversity has an effect of -0.025 on social trust (Dinesen et al., 2020); and that globalisation has an effect size of − 0.1 on public spending (Heimberger, 2021). According to these results, the effect of trust is considerably larger than that of ethnic diversity on social trust and equivalent to, or slightly smaller than, the effect of terror attacks on a range of attitudinal outcomes. As such, whilst these relationships are moderate, statistically speaking, they are also meaningfully large relative to other relationships identified in political science.

Meta-regression Analyses

Researchers use different objects of trust, apply different modelling strategies, and opt for different spatial and temporal coverage. How does this moderate the effects of trust? Statistically, there is considerable heterogeneity in the included studies, suggesting this may be the case.Footnote 4 In this section, I use absolute effect sizes from all studies (i.e. I do not separate by dependent variables, as doing so would reduce sample size substantially). This means the analysis here shows whether the moderators influence the size of the effect; if it was not absolute values, results may be influenced by the direction of the relationship in the original studies. Building on the formula presented in the modelling approach section, the meta-regression can be represented as:

$$yi\, = \,\beta xi\, + \,bji\, + \,bki\, + \,eijk$$

where all parameters are the same as that previously estimated, except a (vector of) ‘moderator’ variables is included (βxi) which are expected to influence the effect size yi. Of note here is that the intercept is excluded. This means all levels of a (categorical) predictor can be included, and the estimates are essentially sub-group estimates for each level of the predictor. Figure 4 presents the effects of the core potential moderators: the object of trust, response scales of dependent and independent variables, and the statistical model of choice. Year of publication (or being made available) and number of control variables are added as descriptive graphs.

Fig. 4
figure 4

Effect of moderator variables on the overall effect of trust

Full size image

All of the models explain a statistically significant amount of variation (p < 0.0001). However, these differences are substantively small. Regarding the object of trust, there are no meaningful differences between the effect sizes, though questions with politicians and parties as the object have slightly weaker effects than government or parliament. This suggests that trust is reflecting some underlying trait rather than being meaningfully different depending on what the respondent is asked to judge. Regarding the choice of statistical models, whilst multinomial models (N = 26) have an insignificant effect around zero, the largest effect is for ordinal (N = 16) and vector auto-regression (VAR) (N = 3) models. The most commonly used models of logistic (N = 70) and OLS (N = 130) have similar effects. Similarly, response scales for either dependent or independent variables do not seem to affect effect sizes; when the independent variable is binary, the effect is approximately zero, but this has a low number of coefficients (N = 11). Researchers often have little choice when using secondary data, and little to guide them when designing original surveys; these results suggest that, at least as far as the consequences of trust go, these decisions don’t make a substantial difference. Finally, there is some suggestion that effect sizes are increasing over time, though this difference is not significant. Finally, effect sizes decrease with the number of control variables (p < 0.01).

Discussion

Despite fifty years of major concerns about the effects of low or declining political trust (Easton, 1975; Miller, 1974), we have lacked a systematic understanding of whether these concerns were justified. The literature on political trust is voluminous and a staple of political science; it is normatively significant as an indicator of legitimacy in democratic states. That we lack a clear idea of whether trust has consequences on important aspects of political life is a surprising gap in a fundamentally important literature.

This paper has provided a systematic review and meta-analysis of 61 papers and 329 coefficients to fill this gap. On the basis of existing literature, the answer is that trust matters for outcomes as diverse as turnout, vote choice, and policy preferences. Whilst these relationships are moderate at best, they are comparable to other existing metaanalyses in political science; it is perhaps unfortunately the case that we live in a world of small effects and, with that considered, it is encouraging that we even find moderate relationships across studies. Encouragingly, the analysis also suggests that these results are not purely the product of research design choices (such as response scales or estimation methods). The literature does have reporting bias, where large or significant effects are over-represented in published work, but this does not seem to be a critical problem; it is possible that the true relationship between policy preferences (specifically, over government spending and environmental policy) is weaker than reported.

Aside from these empirical conclusions, what have we learned about the nature of political trust? I’d like to highlight two conceptual conclusions. First, the hope that the decline of trust actually signals a rise of a ‘critical citizenry’ is not borne out (Dalton, 2004; Inglehart, 1997; Norris, 2011): there is little evidence that trust decreases turnout, and the estimated effect of trust on informal participation is approximately zero. This provides little evidence that a lack of trust stimulates voicing discontent outside of the political system; instead, citizens seem to exit (Hirschman, 1970). This isn’t to say that trust is unambiguously positive; a lack of trust increases the chances of voting for challenger parties, which may provide a valuable accountability mechanism. Second, the results suggest that trust is likely a political cultural concept more at the ‘diffuse’ end of political support than the ‘specific’ (Almond & Verba, 1963; Easton, 1965). As discussed, trust is seen as a mid-range indicator of legitimacy in the Eastonian conception, yet the lack of difference in effects between different objects of trust (such as Parliament versus politicians) suggests that people do not substantially distinguish between political institutions, at least as far as consequences are concerned; however, the number of studies using an index (therefore potentially obscuring these different mechanisms) far outweigh the number differentiating between objects, and so it would be fruitful for studies to more carefully distinguish between different objects of trust. Collectively, these suggest that trust is a political-cultural concept that stimulates loyalty to the political system and acts as a heuristic for citizens’ decision making.

The systematic review also highlights key gaps in the literature. The literature is heavily based on survey-observational data. However, this raises very legitimate concerns about reverse causality, and endogeneity more generally. It may be that spending policy preferences and trust are related, for example, because right-leaning individuals are both less trusting and in favour of lower spending; or that low trust and voting for challenger parties are related because voters choose challenger parties on the basis of policy but then adopt the low-trust positions of the elites. Whilst the overall estimate excluding observational studies (rz = 0.073) is almost identical to the estimate if only observational studies are included (rz = 0.077), the relatively few experimental studies highlight the necessity for more studies to adopt research designs with greater causal purchase. Some research is moving to experimental methods (e.g. D’Attoma, 2020; Fairbrother, 2019; Macdonald, 2021; Peyton, 2020) which is encouraged given the relative paucity of such designs. In addition to this, the literature could utilise panel datasets in which the same individuals (or aggregates) are measured over time and use both instantaneous and lagged trust variables as predictors. This would improve causal identification and also encourage us to theorise about the temporal relevance of trust: does it only matter instantaneously, or does it have consequences for subsequent months and years? What are the reciprocal effects of trust and outcomes? It would also help to understand whether the observed differences are driven by trust changing within individuals or differences between individuals with different levels of trust; some studies find that trust explains differences in policy preferences between but not within individuals (Goubin & Kumlin, 2022).

In addition to these design decisions, the literature is homogenous in other respects, revealed by Fig. 2: there is a strong geographical bias to the UK, US, and cross-national European studies. Contra the substantial and rich conceptual literature, the empirical measures are rather blunt and broad, consisting of single questions or an index which aggregates different objects of trust. As discussed, however, trust is multidimensional, and existing questions likely fail to distinguish concepts like distrust and mistrust that exert different effects on different outcomes; and may, in the case of latent trust, miss the mark entirely (Intawan & Nicholson, 2018). Survey research should attempt to distinguish these concepts with more fine-grained survey instruments, at the very least to understand to what extent widely-available survey questions address this multidimensionality and which dimensions they capture.

This paper has provided a systematic, robust empirical foundation to develop our understanding of the important but understudied consequences of political trust through a meta-analysis of the existing literature. Doing so provides insight into the nature of political trust as a concept, as well as it’s widespread effects on other important political behaviours and attitudes. As suggested by the opening quotes, these supposed consequences have long attracted elite and scholarly interest; the existing evidence, summarised here, suggests this interest is well placed.