Many contemporary accounts suggest that liberal democracies are under threat due to an erosion of citizens’ attachment to the norms and institutions of democratic politics (for example, Levitsky and Ziblatt Reference Levitsky and Ziblatt2018; Mounk Reference Mounk2018; Runciman Reference Runciman2018). A ‘crisis of democracy’ has been proclaimed regularly and in different formulations for decades (see Crozier, Huntington, and Watanuki Reference Crozier, Huntington and Watanuki1975; Ercan and Gagnon Reference Ercan and Gagnon2014; Merkel Reference Merkel2014), but those narratives have largely converged around a concern that citizens’ support for their political systems – and thus the legitimacy of those systems – has eroded in recent times (van der Meer Reference van der Meer2017; Thomassen and van Ham Reference Thomassen and van Ham2017). However, empirical studies disagree about whether any such erosion has in fact taken place, with many finding evidence of declining political support (Citrin and Stoker Reference Citrin and Stoker2018; Dalton Reference Dalton, Van der Meer and Zmerli2017; Foa, Klassen, Slade, Rand, et al. Reference Foa, Klassen, Slade and Rand2020), but others reporting ‘trendless fluctuations’ (van Ham, Thomassen, Aarts, and Andeweg Reference van Ham and Thomassen2017b; van der Meer Reference van der Meer2017; Norris Reference Norris2011, Reference Norris2022). In this study, we aim to resolve this long-standing empirical dispute.
First, we need to clarify the terms of the debate. It is well known that political support ‘is not all of a piece’ (Easton Reference Easton1975, 437): qualitatively different attitudes with different normative and empirical implications fit under the broad umbrella of that term. On the broadest, or ‘most diffuse’ level, empirical studies appear to agree that support for democratic principles has remained high, perhaps increased (Claassen Reference Claassen2019b; Dalton Reference Dalton2004; Norris Reference Norris2011; Stoker Reference Stoker2017). Similarly, most studies of trends in ‘satisfaction with the way democracy works’ (SWD) have found trendless fluctuations overall, albeit with short-term declines in some countries, especially after the 2008 financial crisis (Claassen and Magalhães Reference Claassen and Magalhães2022; van Ham, Thomassen, Aarts, and Andeweg Reference van Ham and Thomassen2017b; Valgarðsson & Devine, Reference Valgarðsson and Devine2022). As van Ham and Thomassen (Reference van Ham and Thomassen2017, 20) observe, ‘the real disagreement about empirical trends appears to lie at the level of support for political institutions and political authorities’, an observation that indicates a need to focus on political trust.
When it comes to political trust, several studies offer evidence of decline: Russell Dalton (Reference Dalton, Van der Meer and Zmerli2017, 376) finds that ‘Declining trust in government has spread across almost all advanced industrial democracies since the 1960s/1970s’, and Jack Citrin and Laura Stoker (Reference Citrin and Stoker2018, 53) conclude that ‘The majority, if not all countries, exhibit a loss of confidence in their legislature’. However, some contend that talk of a crisis is misplaced: Pippa Norris (Reference Norris2011, 241) argues that ‘public support for the political system has not eroded consistently across a wide range of countries around the world’, further suggesting that ‘the crisis myth, while fashionable, exaggerates the extent of political disaffection and too often falls into the dangers of fact-free hyperbole’ and ‘instead, trust and confidence in political institutions usually waxes and wanes over the years in these societies’ in a pattern of ‘trendless fluctuations’ (see also Norris Reference Norris2022, 133). Similarly, van Ham and Thomassen (Reference van Ham and Thomassen2017, 17) report ‘no consistent evidence for declining political support’ (see also Thomassen, Andeweg, and van Ham Reference Thomassen, Andeweg, van Ham, van der Meer and Zmerli2017), and van der Meer (Reference van der Meer2017) finds that ‘Despite evidence that political trust declines in many longstanding democracies in the 1960s and 1970s, the last few decades are characterized by trendless fluctuations in most countries’. The question thus remains a point of contention among scholars.
In this article, we resolve these ongoing debates about trends in political trust with a comprehensive empirical investigation that is unprecedented in geographic and temporal scope. Our analysis is organized into four parts. Firstly, we present our analytical framework, which highlights the need to distinguish between trust in different kinds of institutions: specifically, we distinguish between ‘representative’ or ‘political’ institutions (primarily parliaments, governments, and political parties) on the one hand and non-representative ‘implementing’ institutions (primarily the civil service, legal system, and police) on the other. Secondly, we review the existing literature, arguing that disagreement among researchers is partly due to focusing on different institutions as well as different periods, regions, and countries. Thirdly, we present our data and methods: a global analysis that collates findings from 3,377 surveys conducted by 50 cross-national and national research projects in 143 countries across the world between 1958 and 2019. Applying Claassen’s (Reference Claassen2019a) Bayesian dynamic latent trait models as well as Stimson’s (Reference Stimson1991) ‘dyad-ratios algorithm’ on the aggregate level and multi-level regression models (MLMs) on the individual level, we uncover underlying trends in these different kinds of trust. Finally, we present the results of these analyses by individual countries as well as aggregated by world regions and globally. Our analysis of country trends is based on the entire dataset while the regional and global analysis is based on a subset of the dataset (2,799 surveys conducted in 89 countries) covering countries that were liberal or electoral democracies in a majority of the period covered, according to the Varieties of Democracy’s ‘Regimes of the world’ classification. We identify important differences between regions and countries, but our overall conclusion is that trust in implementing institutions appears to have been stable or increasing around the world in recent decades, whereas trust in representative institutions has been declining.
What crisis?
The debate about trends in citizens’ support for their political systems is rooted both in normative concerns that democratic systems should have the support or consent of their subjects, and in the more practical concern that regimes may not last long, or function as well, if their members do not support them (Bollyky et al. Reference Bollyky2022; Claassen Reference Claassen2020; Easton Reference Easton1965; Hetherington and Rudolph Reference Hetherington and Rudolph2015; Marien and Hooghe Reference Marien and Hooghe2011). However, it has long been noted that there are different types of political support, and declines in one type do not have the same implications as declines in another type (Easton Reference Easton1975, 437).
In this respect, David Easton (Reference Easton1965, Reference Easton1975) crucially differentiated both between different objects (or levels) of support – the community, the regime, and the authorities – and between more diffuse and specific types of support; to simplify a nuanced discussion, these refer broadly to support for what the system is in the abstract and what it does at any given time, respectively (Easton Reference Easton1975, 444). Dalton (Reference Dalton and Norris1999, Reference Dalton2004, Reference Dalton, Van der Meer and Zmerli2017) and Norris (Reference Norris and Norris1999, Reference Norris2011, Reference Norris, Van der Meer and Zmerli2017) have since reformulated this framework by distinguishing between regime principles, norms, and institutions on Easton’s ‘regime’ level, resulting in five levels: the community, regime principles, regime norms, regime institutions, and the authorities. Their reformulations differ slightly but both classify support for individual political leaders and rulers on the first, most basic or proximate level, with trust in political institutions on the second level and satisfaction with the performance of democracy on the third.
Further, even at the level of regime institutions, support is not all of a piece. As Hibbing and Theiss-Morse (Reference Hibbing and Theiss-Morse1995, 23) observe: ‘There is no reason to expect each institution to be held in similar public regard (and they are not)’. For example, Dalton (Reference Dalton2004) and Norris (Reference Norris2011) agreed that, in the US, trust in the Supreme Court had fared much better than trust in Congress. More generally, Rothstein and Stolle (Reference Rothstein and Stolle2008, 444–445) argue that the branches of state charged with implementation should be separated from representative (or ‘representational’) institutions, and they, Schnaudt (Reference Schnaudt2019), and Schneider (Reference Schneider2017), find empirical support for this distinction: trust in parliament, government and political parties appear to capture one underlying dimension of trust, whereas trust in the police, army, and courts capture a separate dimension, with trust in the civil services somewhere in between. Some do, nonetheless, argue that political trust is a latent construct which underpins trust in each of these institutions (for example, Marien Reference Marien, Hooghe and Zmerli2011, Reference Marien, Zmerli Sonja and van der Meer2017; Newton and Zmerli Reference Newton and Zmerli2011; Zmerli and Newton Reference Zmerli, Newton, Zmerli and van der Meer2017; but see van der Meer and Ouattara Reference van der Meer and Ouattara2019). However, even if one latent attitudinal factor does influence all political trust evaluations, they may also be influenced (perhaps at the margins) by object-specific evaluations that produce important variations in levels and changes over time, especially on the aggregate level.
These prior studies highlight the importance of conceptual precision in any empirical analysis of trends in political trust: analysing different types of political support or trust in different institutions haphazardly is likely to confuse rather than clarify the crisis debate because some types may be rising when others are falling. These distinctions are not only analytically important but theoretically and normatively crucial as well. Easton (Reference Easton1975, 437) wrote that: ‘Transparently, not all expressions of unfavourable orientations have the same degree of gravity for a political system. Some may be consistent with its maintenance; others may lead to fundamental change.’ Dalton (Reference Dalton and Norris1999, 59) later elaborated that: ‘As the object of dissatisfaction becomes more general – the performance of the regime or attachment to the political community – the political implications increase’ (see also van der Meer Reference van der Meer2017; Merkel Reference Merkel2014). In this respect, trust in institutions has generally been considered a relatively diffuse type of support (Devine and Valgarðsson Reference Devine and Valgarðsson2023; Easton Reference Easton1975, 447), so declines in this type of support would be considered more serious than, for example, declining support for particular incumbents, but perhaps not as grave as declining commitment to democratic norms.
The implications of trends in trust between different institutions have received less attention in the literature, but the ‘crisis’ literature generally focuses on a potential crisis in the legitimacy of democratic or political authorities and institutions, rather than of non-representational institutions. While trust in, for example, the judiciary and police is likely to be important for various reasons (see Tyler Reference Tyler2001), we argue that interrogations of the ‘crisis of democracy’ thesis should focus on trust in representative institutions rather than non-representative ones, as these are the core elements of democratic government. Some authors are careful to highlight these classic distinctions in types of political support (for example, Dalton Reference Dalton, Van der Meer and Zmerli2017; van Ham and Thomassen Reference van Ham and Thomassen2017; van der Meer Reference van der Meer2017), but most prior studies of trends in trust use different measures in isolation: most commonly trust in the legislature but also in the police (Hutchison and Johnson Reference Hutchison, Johnson, Van der Meer and Zmerli2017), legal system (Torcal Reference Torcal, van der Meer and Zmerli2017), government and political parties (Bargsted, Somma, and Castillo Reference Bargsted, Somma, Castillo, Van der Meer and Zmerli2017) and/or various distinct views about politicians in general (Dalton Reference Dalton2004, Reference Dalton, Van der Meer and Zmerli2017). In this study, we systematically compare trust in representative institutions and implementing institutions, arguing that the former is more pertinent to debates about a potential crisis of democracy.
Our evaluation of the crisis of democracy thesis then depends on several considerations. First, does a crisis require an underlying, global trend of decline or is it enough that some type(s) of trust has been declining in many countries? If the latter, how many countries need they be – and would this be negated by a comparable number of countries experiencing the reverse trend? Should all countries count the same in that assessment or should larger, more established and/or geopolitically important democracies ‘count’ more than smaller or newer democracies? The answers to those questions will depend on perspective, but we nevertheless believe that explicitly addressing them is important. Underlying global trends of decline would raise serious questions about the overall direction of democracy in human history, but precipitous declines in many countries coupled with trendless fluctuations in others may still signify a crisis of democracy, especially if the former are large and geopolitically important countries that may strongly shape the trajectory of global society. Guided by these concerns, we seek to provide a more nuanced and comprehensive investigation of trends in different types of political support around the world.
Prior studies
The academic debate regarding whether political trust is in decline often results from researchers considering different countries, time periods, or data sources. Crisis claims tend to focus on a small number of major advanced democracies, particularly Anglosphere countries such as the US and Canada (Dalton Reference Dalton2004, Reference Dalton, Van der Meer and Zmerli2017; Hetherington Reference Hetherington2005), the UK (Jennings, Clarke, Moss, and Stoker Reference Jennings, Clarke, Moss and Stoker2017), and Australia (Dassonneville and McAllister Reference Dassonneville and McAllister2021; Stoker, Evans, and Halupka Reference Stoker, Evans and Halupka2018).
For example, Dalton (Reference Dalton, Van der Meer and Zmerli2017) primarily considers trends in the US and Canada, showing declines from the 1960s to the 80s in the US and between the 80s and 90s in Canada. Yet Norris (Reference Norris2011) argues that there has been no structural downward trend in political trust in the US, especially when it comes to the judicial and executive branches and especially after the 1980s. By contrast, Citrin and Stoker (Reference Citrin and Stoker2018, 52–54) find trust in the US government and legislature reaching historic lows in 2016 and also identify declines in several other democracies. Polling data further suggests that confidence in Congress and trust in the federal government in the US started declining sharply again after 2016 (Gallup 2021; Pew Research Center 2020).
Looking beyond North America, Norris (Reference Norris2011) reports a pattern of ‘trendless fluctuations’ in several countries and regions between 1981 and 2005, particularly EU countries between 1998 and 2009. Likewise, van Ham and Thomassen (Reference van Ham and Thomassen2017) conclude that there has been no universal long-term decline in trust in political authorities across Western European democracies between the 1980s and 2010s, concluding that ‘insofar as we observe decline at these lower levels of political support, the sharpest decline is observed in the last decade, coinciding with the global economic crisis. Hence, decline is certainly not long-term.’ (van Ham, Thomassen, Aarts, and Andeweg Reference van Ham and Thomassen2017b, 31). There is no consensus regarding trends in Western and Northern Europe: one study using Eurobarometer data found significant declines in trust in parliament between 2007 and 2011 (Armingeon and Guthmann Reference Armingeon and Guthmann2014) and Citrin and Stoker’s (2018) comparison between levels of political trust in the earliest (1981-1990) and later (2006-2013) waves of the WVS/EVS surveys suggested declines, whereas another study using data from the ESS found no decline in trust in representative institutions in the region between 2002 and 2012 (Torcal Reference Torcal, van der Meer and Zmerli2017).
Looking beyond Western Europe, it is widely agreed that in Southern Europe trust fell significantly following the global financial crisis (Armingeon and Guthmann Reference Armingeon and Guthmann2014; Citrin and Stoker Reference Citrin and Stoker2018; Torcal Reference Torcal, van der Meer and Zmerli2017), but it remains unclear if this represented a temporary drop or a more permanent shift. Finally, a study of trust in parliament in post-communist Central and Eastern Europe found consistent declines in most countries, from relatively high levels right after the fall of communism (Závecz Reference Závecz, van der Meer and Zmerli2017).
In other world regions, a study of trends in Latin America found that trust in institutions had grown from 1995 to 2011 (Bargsted, Somma, and Castillo Reference Bargsted, Somma, Castillo, Van der Meer and Zmerli2017), but there appear to be no discernible trends in either Africa or the Middle East (Hutchison and Johnson Reference Hutchison, Johnson, Van der Meer and Zmerli2017). In Asia and the Pacific, one study found gradual declines in trust in Japan, South Korea, and Taiwan and dramatic declines in Thailand and the Philippines but stability or even growth in other countries, including no decline in Australia and growing trust levels in New Zealand (Park Reference Park, van der Meer and Zmerli2017). Conversely, a recent study finds that political trust has declined in Australia following recent elections (Dassonneville and McAllister Reference Dassonneville and McAllister2021).
How can we synthesize these findings? In the case of the US, it seems that political trust (but, perhaps, not trust in non-representational institutions) declined from the 1960s to the 80s, fluctuated for decades after that, but declined again after the financial crisis, reaching historical lows after 2016. Meanwhile, trends outside the US appear to be diverse and ambiguous. Previous studies have suggested that trust has also declined in Eastern Europe, several Asian countries, Canada, the UK, and perhaps Australia and Southern Europe; yet it may have risen in Latin America and New Zealand while being broadly trendless across Western Europe and most of Asia and Africa. Those studies that have sought to distinguish between countries, measures, and periods have tended to find ‘trendless fluctuations’ instead of overall declines across the globe (van Ham, Thomassen, Aarts, and Andeweg Reference van Ham and Thomassen2017b; van der Meer Reference van der Meer2017; Norris Reference Norris2011), but there is still no consensus (Citrin and Stoker Reference Citrin and Stoker2018; Dalton Reference Dalton, Van der Meer and Zmerli2017).
As we have already noted, many of these studies reach different conclusions because they use measures of trust in different institutions (and sometimes of SWD). However, the temporal and geographic variation in the focus of studies reviewed above also hinders systematic analyses and coherent overall conclusions. In our view, it is not satisfactory to observe that some countries have experienced declining trust in one institution and others have not, and to conclude on that basis that we are simply experiencing ‘trendless fluctuations’ (Norris Reference Norris2011, Reference Norris2022). Instead, we need to ask what types of political trust have been declining, where have they been declining, and over what periods?
Data and methods
Our analysis uses a novel dataset of unprecedented scope, based on data synthesis conducted by the authors with help from resources such as the HUMAN Surveys project (Klassen Reference Klassen2020), GESIS’s ONBound Harmonization Wizard (Bechert, May, Quandt, and Werhan Reference Bechert, May, Quandt and Werhan2020), and the ‘Survey Data Recycling Project’ (SDR) (Slomczynski et al. Reference Slomczynski, Tomescu-Dubrow, Kołczyńska, Wysmułek, Oleksiyenko, Powałko and Zieliński2020), in an effort to pool all existing survey measures of political trust available from cross-national survey projects and a variety of national surveys. Our dataset is based on 5.3 million survey respondents from 50 survey projects conducted in 143 countriesFootnote 1 and we aggregate it to 3,377 country-year-survey observations. The surveys were selected by the criteria that they were fielded by a recognized cross-national or national research project and based on a type of random sampling of the adult population in each country.Footnote 2 Table 1 lists the survey projects included in our dataset and the number of aggregate observations, countries, and years that each of them covers. Full information on the survey project, country, and year coverage of each measure of political trust is reported in the Supplementary Information (SI), Appendix B. Throughout, we use the survey weights provided by each project where available.Footnote 3
Table 1. Survey projects included in the global dataset
The formulation of the questions and especially the number of response options given in the survey measures underlying the data from these different projects vary considerably (for a detailed discussion of this issue, see Durand, Peña Ibarra, Rezgui and Wutchiett Reference Durand, Peña Ibarra and Rezgui2021; Kołczyńska Reference Kołczyńska2022; Kołczyńska and Schoene Reference Kołczyńska, Schoene, Timothy P. Johnson, Beth-Ellen Pennell, Ineke A.L. Stoop and Brita Dorer2018). Here, we dichotomize all measures so that a value of 1 indicates an expression of trust while a value of 0 indicates an expression of lack of trust (the same approach as in Claassen (Reference Claassen2019a, Reference Claassen2020)). In Appendix B, we present detailed information on these differences, how we used semantic equivalence tests to determine how to dichotomize measures that included response scales with mid-points, and how trends and country differences resulting from this approach are almost identical to those resulting from using the full original scales. In Appendix C, we present descriptive trends in each measure by survey project; these indicate that our method of dichotomization leads to a high degree of consistency in the trends and levels of trust estimated by different survey projects.
As discussed, when evaluating whether trust is ‘in crisis’ or has been ‘declining’ in some general way, it is important to consider what types of trust have been declining, where they have been declining and when (if at all). In terms of the what, we analyse trends in trust in six different institutions, where we expect some measures to reflect trust in representative institutions and others to reflect trust in non-representative or ‘implementing’ institutions. Conceptually, parliament, government, and political parties should reflect the former and the civil service, legal system, and police the latter; in the first part of our analysis, we will examine to what extent aggregated measures of trust in these institutions follow that conceptual distinction.
In terms of the where, we examine trends in trust in every country where we could acquire data from consistent measures of trust in one or more of these institutions in more than one year, structuring our analysis by global regions (in line with the existing literature, especially Zmerli and van der Meer (Reference Zmerli and van der Meer2017)). We start by estimating trends in all countries in our dataset to present a comprehensive reference for trends in political trust around the globe, but the regional and global analyses that follow will only include democratic countries. While trust dynamics in non-democracies could have important implications, our focus here is on democracies; on what trends in trust can tell us about the ‘crisis of democracy’ debate and on trust in institutions that are at least nominally susceptible to democratic mechanisms of accountability. To classify whether a country is a democracy or not, we use the ‘Regimes of the World’ measure from the Varieties of Democracy (V-Dem) project (Varieties of Democracy 2022) and only include countries that were coded as either ‘liberal democracies’ or ‘electoral democracies’ in a majority of the years in which they are included our dataset.Footnote 4 The regional and global part of our analysis includes about 4.3 million respondents from 2,800 surveys conducted by 49 survey projects across 89 democratic countries.
In terms of the when, our choice of methods also allows us to examine non-linear trends, identifying countries where trust may have risen in one period but declined in another. We use data going as far back as available, and up until 2019; data from more recent years is less consistently available at the time of writing and it is likely to be heavily influenced by the short-term effects of the COVID-19 pandemic, so we leave an updated investigation of later trends to future studies.
In presenting our analysis and conducting regional and global models, we classify countries into the six-category ‘politico-geographic’ classification provided by V-Dem and the Quality of Government project (Teorell et al. Reference Teorell, Dahlberg, Holmberg, Rothstein, Pachon and Axelsson2020), which groups countries of the world into (1) Western Europe and North America (including Cyprus) (WENA), (2) Eastern Europe and Central Asia (EECA), (3) Latin America and the Caribbean (LAC), (4) the Middle East and North Africa (including Israel and Turkey, excluding Cyprus) (MENA), (5) Sub-Saharan Africa (SSA), and (6) Asia and Pacific (AP). We modify that classification here by including Australia and New Zealand in category (6) instead of (1), for obvious geographical reasons (see also Park Reference Park, van der Meer and Zmerli2017). Table 2 presents information on the number of observations for each measure in each region and overall, and Table B5 in the SI lists the countries classified in each region.
Table 2. Number of aggregate country-year-study observations for trust in each institution by region
In our main analysis, we use Bayesian dynamic latent trait models (Claassen Reference Claassen2019a, Reference Claassen2019b, Reference Claassen2020) to uncover any underlying trends in these types of trust. This method models each measure of trust as a binomial distributed count, modelling probability parameters from the counts of trusting responses and the total number of valid responses in each observation (survey-country-year). As described in more detail in Claassen (Reference Claassen2019a), we use the version of his model which provides the best fit, using a beta-binomial prior (which allows for overdispersion due to any errors in survey estimates beyond sampling error) and accounting for potential idiosyncrasies in levels and trends in each survey item in each country by including item and item-country intercepts as well as item slopes.Footnote 5
We start our analysis by deriving these estimates of latent trends in each type of trust in each country and begin by exploring the extent to which these different trends are related and whether they tap the two distinct dimensions of trust (trust in representative and non-representative institutions) that we expect. For this, we explore correlations and use exploratory and confirmatory factor analyses on the Bayesian estimates for these trends, both overall and within-countries. We then present all country trends in these estimates, presenting in the main text the measures that appear most distinct and for which we have the most data, retiring other measures to the SI because of space constraints. Our main analysis then extends these country models to explore underlying trends in groups of countries: by world region as well as globally, when including only democratic countries. For these models, we pool the data for each group of countries and add an additional term to the models: a country parameter which accounts for the effects of different levels of trust between countries within the country group, thus focusing on any underlying trend in that group.
Finally, we run two types of sensitivity analysis for the main (regional and global) analysis. First, we use Stimson’s (Reference Stimson1991, Reference Stimson2018) ‘dyad-ratios algorithm’, which has been used in many previous analyses of latent trends, including several studies of trends in political trust and support (Chanley, Rudolph, and Rahn Reference Chanley, Rudolph and Rahn2000; Jennings, Clarke, Moss, and Stoker Reference Jennings, Clarke, Moss and Stoker2017),Footnote 6 but has recently given way to more sophisticated Bayesian alternatives (Claassen Reference Claassen2019a; McGann Reference McGann2014) and is less well suited to country group analysis, since countries with a greater number of observations may dominate the results. Second, we use multi-level linear regression models (MLMs) on the individual level, where respondents are specified as nested within survey sources which are in turn nested within countries, allowing the intercepts for year to vary on both levels.Footnote 7 This approach ignores non-linear trends in the data to provide us with a simpler, summary sense of whether there appear to be any statistically significant trends in the data, and the estimated extent of any such trends on a regional and global scale. The results of the three methods all lead to the same substantive conclusions.
Analysis
We begin by estimating Bayesian dynamic models of latent trends by every country in our dataset, separately for each measure.Footnote 8 We start by exploring how related the estimates for trust in each of these six institutions are in our dataset, calculating pairwise correlations and running exploratory and confirmatory factor analyses to build the foundation for our presentation of the country trends. Figure E1 in the SI presents the pairwise correlations between the Bayesian estimates of underlying trust by each measure, using the estimates for all country years where data is available.Footnote 9 This shows that there is a very strong correlation between aggregate underlying trust in parliament, government, and political parties (0.87 Pearson’s r or more), and also strong pairwise correlations between all of these three and trust in the civil service and legal system (between 0.6 and 0.82), but trust in the police is only moderately correlated with the first three (between 0.45 and 0.65) – our ‘representative’ institutions – while it is strongly correlated with the latter two – the ‘implementing’ institutions (0.76 and 0.83 Pearson’s r).
However, our primary focus here is on trends in trust, and these overall bivariate correlations may be confounded by various factors at the country level. A better approximation to the true relationship between these measures is to examine the correlations between the within-country variation of these estimates; that is, their change over time within countries (see, for example, Fairbrother Reference Fairbrother2014). Figure E2 in the SI presents the pairwise correlations between these variables after they have been centred around the mean for each respective trust measure within the country. This again shows a very strong correlation between trust in representative institutions (0.81 Pearson’s r and more), but here, their correlations with trust in the civil service and legal system are only moderate (between 0.38 and 0.67), their correlations with trust in the police are weak (between 0.11 and 0.17), and the relationship between trust in the police and the other implementing institutions is also only moderate (0.4 and 0.53).
The results from exploratory factor analyses (EFA) on each of these sets of estimates (overall estimates and within-country changes) are presented in Appendix E in the SI: both suggest that these aggregate trust measures are reflecting at least two separate, underlying factors: one for trust in representative institutions, another for trust in implementing institutions. Scree plots and parameter tests suggest that two factors are appropriate in all cases. The results of CFA also support these findings: the fit of a two-factor solution is much better than for a one-factor solution, whichever version of the estimates is used.Footnote 10
Based on the above analyses and due to space constraints, we present estimated country trends for three of these institutions together here and present trends for all six separately in the SI, Appendix D. Here, we present trends in trust in parliament (as this is the representative institutions for which we have most data and loads most strongly onto that underlying factor), the legal system, and the police: although trends in the latter two load onto the same ‘implementing’ factor, they are only moderately correlated (and we have much fewer observations for trust in the civil service). Figures 1–6 present the country trends in each of these trust measures by world region, where the lines represent the (unit-normal standardized) point estimates for the latent trait in each country year and the shaded areas represent the 95 per cent Bayesian credibility intervals. The point estimates represent the mean value in the posterior distribution and so the most probable value – or the ‘expected value’ according to the model – whereas the credibility interval portrays the dispersion of probable values and thus the extent of uncertainty in each case (for example, McElreath Reference McElreath2020, 54–57).
Figure 1. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Western Europe and North America’ (WENA) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Figure 2. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Eastern Europe and Central Asia’ (EECA) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Figure 3. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Latin America and the Caribbean’ (LAC) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Figure 4. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Middle East and North Africa’ (MENA) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Figure 5. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Sub-Saharan Africa’ (SSA) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Figure 6. Latent trends in trust in parliament, the legal system, and the police within each country in the ‘Asia and the Pacific’ (AP) regional category. Point estimates and 95 per cent credibility intervals from Bayesian dynamic latent trend models using harmonized survey data.
Starting with trends within the Western Europe and North America (WENA) region, presented in Figure 1, we see that trust in parliament appears to have declined in Cyprus, France, Greece, Iceland, Italy, Portugal, Spain, the US, and, perhaps, the UK. This trend is gradual in France, but in most other countries there are clear non-linearities, with a sharp drop around the 2008 financial crisis. Conversely, trust in parliament appears to be rising in Denmark, Norway, Sweden, and Switzerland, and mostly trendless in other countries. When it comes to trust in the police, this has barely been declining in any country (perhaps Cyprus), but rising in Austria, Belgium, Denmark, Finland, France, Germany, Iceland, Italy, Portugal, Spain, Sweden, and Switzerland. Trust in the legal system is generally somewhere between the other two institutions; in some countries, it appears closer to trust in parliament but in others, it is closer to trust in the police. Regardless, the overarching pattern there is one of stability, although it has been rising in Finland and Germany.
Turning to trends in the Eastern Europe and Central Asia (EECA) region, presented in Figure 2, trust in parliament has been declining in Albania, Bosnia and Herzegovina, Bulgaria, Croatia, Czechia, Hungary, Latvia, Lithuania, Moldova, Mongolia, Romania, Slovenia, and Ukraine. Notably, trust in parliament has only been rising among non-democracies in this region: Belarus, Kazakhstan, Russia (since 2000), and Uzbekistan. Conversely, trust in the police has only declined in Moldova and Ukraine, while it has increased in Czechia, Estonia, Georgia, Latvia, Lithuania, Mongolia, Serbia, and Slovenia. Trust in the legal system again trends more in parallel with parliament in some countries but more with police in others: it has been declining in Albania, Bosnia and Herzegovina, Bulgaria, Croatia, Moldova, and Ukraine, but rising in Czechia, Estonia, Latvia, and Lithuania.
In Figure 3, we see that trends in the three measures appear to be more similar to each other in Latin America and the Caribbean (LAC) than in the WENA and EECA regions, and most of these trends are non-linear: in particular, many countries appear to have seen increases in trust until the early 2010s, with relatively sharp declines since then, especially when it comes to trust in parliament. Nevertheless, there appears to have been an overall decline of trust in parliament in Argentina, Brazil, Chile, Colombia, Costa Rica, El Salvador, Honduras, and Peru – and a rise only in Ecuador. Generally, trends in trust in the legal system and the police appear similar to trust in parliament here, but are usually less pronounced.
Only four countries are classified as democracies in our analysis in the Middle East and North Africa (MENA) region presented in Figure 4 – Israel, Lebanon, Tunisia, and Turkey – and here we see declining trust in parliament in Lebanon (and perhaps Tunisia), declining trust in the legal system in Israel and Lebanon, and rising trust in the police in Lebanon and Turkey. Interestingly, we see steep declines in trust in some of the non-democracies in this region: especially in Egypt, Jordan, and Palestine. Similarly, our data for Sub-Saharan Africa (SSA), presented in Figure 5, is much sparser and generally covers a shorter period than in other regions. Here we see apparent declines in trust in parliament in Mali, Tanzania, and in recent years in South Africa, and perhaps a slight decline in all trust measures in Madagascar and rising trust in the legal system in Kenya, but other trends are unclear.
Finally, Figure 6 presents trends in Asia and the Pacific (AP) and here we generally see stark differences between institutions as in the WENA and EECA region: trust in parliament has been declining in Australia, Hong Kong, South Korea and Taiwan but rising substantially in New Zealand. Trust in the police, however, has been rising in India, Indonesia, Japan, the Philippines, and Taiwan – and the same goes for trust in the legal system in Australia. Trust in these implementing institutions does not appear to have declined in any of the countries in the AP region. Country trends in trust in government, political parties, and the civil service are presented in the SI, Appendix D: trends in the former two are broadly very similar to trust in parliament, although here we see clearer declines in trust in government in Canada and the UK (for which we have more observations there). We have fewer observations for trust in the civil service and those trends are generally less dynamic but similar to those for trust in the legal system.
Turning to our analyses of any underlying trends in each region, Figures 7 and 8 show underlying trends within each world region from the hierarchical Bayesian models, first for representative institutions and then for implementing institutions. These models only include countries that were coded as democratic in a majority of years in our dataset by the V-Dem classification, as explained above. Footnote 11 Figure 7 shows that trust in parliament has been declining overall in the WENA region as well as SSA and, especially, the EECA in the 1990s. In the LAC, we see non-linear trends of rising trust until the early 2010s and a steep decline since then, resulting in overall declines since 1990. The same applies to trust in government and political parties, except that the latter does not appear to have been declining in the WENA region. The other regions show few clear trends but trust in political parties appears to have been rising in the four countries included in MENAFootnote 12 and trust in government rising in the AP region. By contrast, Figure 8 shows that trust in the civil service has been rising in the WENA region as well as the AP region, but largely trendless in the others. In WENA and the EECA, trust in the legal system declined until the 1990s but has been rising since, especially in WENA, whereas we see a more linear trend of decline in LAC and a rise in the AP region. Finally, the contrast is most clear for trust in the police: this has been rising substantially in the WENA, EECA, and AP regions, and trendless elsewhere.
Figure 7. Bayesian estimates for latent regional trends in trust in representative institutions, from regional models using harmonized survey data, including only democracies.
Figure 8. Bayesian estimates for latent regional trends in trust in implementing institutions, from regional models using harmonized survey data, including only democracies.
Figure 9 depicts the results of our models for estimating underlying trends in each measure globally, using all of our data for democratic countries to get a sense of the general trajectory of trust in each of these institutions worldwide in recent decades. Here we see that trust in parliament, government, and political parties has been declining across the world in recent decades, whereas trust in the civil service and the police has been rising and trust in the legal system has been broadly trendless. We do see some non-linearities here as well: while trust in parliament and government clearly took a dive following the global financial crash of 2008, trust in political parties apparently did not – but the latter had declined even more dramatically in the 1990s.
Figure 9. Bayesian estimates for latent global trends in trust in six institutions, from global models using harmonized survey data, including only democracies.
Sensitivity analysis
To explore whether alternative methods for trends analysis yield different results than the above, we also ran Stimson’s (Reference Stimson1991, Reference Stimson2018) ‘dyad-ratios algorithm’ and multi-level linear regression models (MLMs) on our dataset. As detailed in the SI, Appendix F, the former is a type of weighted factorial analysis of the aggregated data, whereas the latter is run using our individual-level data, nesting respondents within survey sources and countries in three-level models with random intercepts. We present the full results for regional and global trends from these models in Appendix F. These results are highly consistent with our main findings and with each other: they all suggest that trust in parliament and government has been declining across the world in recent decades, whereas trust in the civil service and police has been rising, with more ambiguous trends for trust in political parties and the legal system. This is true in the global models and WENA and the EECA regions, with the non-linear trend of rise followed by decline apparent in the LAC region. Tables with output from the multi-level linear regression models show that all of these trends are significant at the 99.9 per cent level (p < 0.001) (Stimson’s dyad-ratios algorithm does not provide uncertainty estimates); including in the LAC (although there, there is also a significant but small trend of declining trust in the civil service and legal system).
These models also bring the benefit of providing the estimated extent of these trends in each case, whereas the Bayesian models produce unit-item standardized values, which are difficult to interpret substantively. The coefficients for the year variable from the global and regional MLMs are presented in Table 3. These are directly interpretable as predicted changes in the proportion of trusting respondents in each case (Gomila Reference Gomila2021; Hellevik Reference Hellevik2009), so the global coefficient of −0.0029 for trust in parliament can be interpreted as a predicted decline of 0.29 percentage points each year, which translates into a predicted decline of about 8.4 percentage points over the 29-year period covered by these models (1990-2019).Footnote 13 Meanwhile, the predicted linear decline for trust in government is about 0.25 points per year globally (about 7.3 points over the whole period), but the models predict a rise in trust in the police of about 0.43 points (about 12.5 points in the whole period) and 0.18 for the civil service (about 5.2 points). More dramatically, Stimson’s algorithm predicts that trust in parliament is about 18 percentage points lower globally in 2019 than in 1990 (34.5 per cent compared with 52.5 per cent), trust in government about 13 points lower, and trust in the police about 15 points higher.Footnote 14
Table 3. Coefficients for the year variable from multi-level regression models where each trust variable is the dependent variable, from global and regional models. Full output is reported in Appendix F in the SI
Standard errors in parentheses.
+ p < 0.10, *p < 0.05, **p < 0.01, ***p < 0.001.
Discussion
‘It appears that democracy is always in crisis’, wrote Ralf Dahrendorf (Reference Dahrendorf, Pharr and Putnam2000) over twenty years ago. Since then, claims of democratic crises have only grown in volume and urgency (Ercan and Gagnon Reference Ercan and Gagnon2014; Plattner Reference Plattner2015). This narrative has in recent times converged on a conception of crisis as low and declining citizen support for their political systems, especially their trust in political institutions and actors, and thus a crisis of the legitimacy and potential stability of those institutions (Citrin and Stoker Reference Citrin and Stoker2018; Dalton Reference Dalton2004, Reference Dalton, Van der Meer and Zmerli2017). However, these claims have been contested on empirical grounds (van Ham, Thomassen, Aarts, and Andeweg Reference van Ham and Thomassen2017b; van der Meer Reference van der Meer2017; Norris Reference Norris2022) and the issue has remained unresolved. In this study, we have argued that the dispute is partly because of limited and varied coverage of survey sources, countries, and periods in prior studies, and because some have tended to conflate trust in representative institutions (parliaments, governments, and political parties) on the one hand and non-representative implementing institutions (the civil service, legal system, and police) on the other in their analysis.
To resolve this dispute, we have provided what we believe is the most comprehensive and sophisticated analysis of trends in political trust around the world to date. We harmonized data from practically all available survey projects which have fielded measures of trust in any of six institutions cross-nationally and nationally up to 2019, and applied Classen’s (2019a, 2020) Bayesian dynamic latent trends model, Stimson’s (1991) dyad-ratios algorithm, and multi-level regression models (MLMs) on the individual level to this dataset. All of these methods converged on the overarching conclusion that trust in representative institutions has been declining across the democratic world in recent decades, while trust in non-representative ‘implementing’ institutions has been stable or rising.
Our Bayesian models and Stimson’s algorithm suggest a non-linear but clear global trend of declining trust in parliament and government globally – coinciding with a more linear increase in trust in the police and, more modestly, the civil service. More specifically, our MLMs suggest that among democratic countries of the world, there has been an underlying trend of declining trust in parliament by about 9 percentage points in the period from 1990 until 2019, but a rise in trust in the police by about 13 points in the same period. Our Bayesian estimates by individual countries suggest that trust in parliament has been declining in about thirty-six democracies around the world, including large and geo-politically important countries such as Argentina, Australia, Brazil, France, Italy, South Korea, Spain, and the US (Canada and the UK are borderline, but show clearer trends of declining trust in government) – but rising in only six. In addition to this, several countries in Latin America have seen levels of trust collapse since 2010, following increasing levels in the preceding years. These trends are generally clearest in Eastern and Southern Europe but unclear – or clearly absent – in most of Asia and the Pacific, Africa, and the Middle East. These findings thus highlight the importance of being mindful of non-linear trends and regional differences when analysing trends in political support.
Of course, there are many nuances to these overall conclusions. First, it is true that, by this count, a majority of the individual countries in our analysis do exhibit ‘trendless fluctuations’ in trust in representative institutions (although in some cases it is difficult to determine whether this is because of truly trendless fluctuations or because of the short time period covered). This is important to note but does not change the fact that the overall trend across (democratic) countries is one of decline, as trust is declining in many more countries than it is rising. By way of analogy, we would consider it an important trend if trust in a particular electorate had declined overall by about 10 percentage points in a period, even if most individuals within that electorate had not changed their trust judgements.
Second, trust in representative institutions has in fact been rising in a few countries: Denmark, Ecuador, New Zealand, Norway, Sweden, and Switzerland. These countries are important exceptions that could provide lessons to other countries dealing with declining trust, but the most populous among them is Ecuador, and their geopolitical influence probably pales in comparison with the decliners listed above. As per our analytical framework, any analysis of the ‘crisis’ thesis should consider such factors, as well as simply counting the number of countries experiencing declining trust.
Finally, most of our estimates do come with substantial uncertainty, as the Bayesian credibility intervals often overlap at the beginning and end of a given period, and this is especially true for all estimates of trends for global regions. Nevertheless, those 95 per cent intervals are largely arbitrary and the point estimates are considered the most probable value (McElreath Reference McElreath2020). The linear trends reported here are significant at the 95 per cent confidence level (although this is also an arbitrary threshold) in more traditional regression models, and all of our methods point in the same direction. Thus, we choose not to take the perspective of Sartori’s (Reference Sartori1970, 63) ‘over-conscious thinker, the man who refuses to discuss heat unless he is given a thermometer’ and instead conclude that in all likelihood, trust in representative institutions has generally been declining across established democracies, whereas trust in implementing institutions has been rising.
What should we make of these divergent trends in trust in different types of institutions? Most prior studies have treated ‘political trust’ or ‘institutional trust’ as a singular attitude and many have found empirical support for this approach at the individual level (for example, Marien Reference Marien, Hooghe and Zmerli2011, Reference Marien, Zmerli Sonja and van der Meer2017; Newton and Zmerli Reference Newton and Zmerli2011; Zmerli and Newton Reference Zmerli, Newton, Zmerli and van der Meer2017), although others have disputed this finding (van der Meer and Ouattara Reference van der Meer and Ouattara2019) and found separate factors for trust in representative institutions and implementing institutions (Rothstein and Stolle Reference Rothstein and Stolle2008; Schnaudt Reference Schnaudt2019; Schneider Reference Schneider2017). Our analyses strongly support the latter perspective on the aggregate level: trust in these two groups of institutions form two separate underlying factors, and this is especially true when it comes to aggregate changes in trust over time within a country; within-country changes in trust in representative institutions on the one hand and implementing institutions on the other is only moderately correlated (and weakly when it comes to trust in the police). This is in line with a long-standing conception of trust as relational and evaluative, where people form trust judgements at least partly relative to the particular object in question and its performance according to particular criteria (see also van der Meer and Ouattara Reference van der Meer and Ouattara2019).
From this perspective, the simplest interpretation of these divergent trends is that declining trust in parliaments and governments is not about people’s lack of trust in state institutions in general, but their lack of trust in elected representatives in particular. What we are experiencing is not a crisis of trust in the state per se, but an emerging crisis of trust in political and democratically mandated institutions. While increasing trust in implementing institutions is likely to be generally considered a positive development (see Tyler Reference Tyler2001), it seems to us worrisome from a democratic perspective that citizens are growing more trusting of non-representative institutions at the same time as they have grown less trusting of representative institutions, as this might pave the way for non-democratic political leaders to use the state in more autocratic ways. Other studies have demonstrated that people’s support for democratic ideals (Claassen Reference Claassen2019b; Dalton Reference Dalton2004; Stoker Reference Stoker2017) and their ‘satisfaction with the way democracy works’ (Claassen and Magalhães Reference Claassen and Magalhães2022; Valgarðsson & Devine, Reference Valgarðsson and Devine2022) remain steady at high levels, but eroding trust in the institutional mechanisms intended to bring democracy about may nevertheless, directly or indirectly, threaten the future of democratic governance (for example, Easton Reference Easton1965; Hetherington Reference Hetherington2005).
But a crisis of trust may also lead to a renewal of democracy (Dalton Reference Dalton2004) – a crisis leading to fundamental transition or ‘decisive intervention’ (see Hay Reference Hay1996) in favour of more trustworthy democratic governance. If it is something about the way democratic politics is practised that citizens distrust, perhaps those politics need to change. Given citizens’ continually high support for democratic ideals, those changes may well be in the direction of more democratic governance rather than less (Dryzek et al. Reference Dryzek, Bächtiger, Chambers, Cohen, Druckman, Felicetti, Fishkin, Farrell, Fung, Gutmann, Landemore, Mansbridge, Marien, Neblo, Niemeyer, Setälä, Slothuus, Suiter, Thompson and Warren2019; Smith Reference Smith2009; Ouattara and van der Meer Reference Ouattara and van der Meer2023). At the very least, declining citizen trust in parliaments and governments among so many established democracies are warning signs that need to be addressed in further analysis, as well as in the sphere of politics and public debate. Future research would do well to investigate the causes of declining political trust and its practical and normative implications: this may demand different research approaches, including mixed-methods research that illuminates the nature of citizens’ trust judgements and their lack of trust, as well as experimental methods that identify the causal mechanics of political trust. This would enable us to better understand how citizens understand and respond to the practices of politics, and the role of political distrust in shaping our democracies.
Supplementary material
Our online ‘Supplementary Material’ file includes Appendices A–H. To view supplementary material for this article, please visit https://doi.org/10.1017/S0007123424000498
Data availability statement
Replication data and code for this paper can be found at: https://doi.org/10.7910/DVN/GRARW6.
Financial support
This work was supported by the UK Economic and Social Research Council (ESRC) (VOV, WJ, GS, HB, DD, LM, grant number ES/S009809/1); and the Leverhulme Trust (VOV, grant number ECF-2022-795).
Competing interests
None.
Open access
