Political trust is an important indicator of political legitimacy. Hence, seemingly decreasing levels of political trust in Western democracies have stimulated a growing body of research on the causes and consequences of political trust. However, the neglect of potential measurement problems of political trust raises doubts about the findings of earlier studies. The current study revisits the measurement of political trust and re-examines the relationship between political trust and sophistication in the Netherlands by utilizing European Social Survey (ESS) data across five time points and four-wave panel data from the Panel Component of ESS. Our findings illustrate that high and low political sophistication groups display different levels of political trust even when measurement characteristics of political trust are taken into consideration. However, the relationship between political sophistication and political trust is weaker than it is often suggested by earlier research. Our findings also provide partial support for the argument that the gap between sophistication groups is widening over time. Furthermore, we demonstrate that, although the between-method differences between the latent means and the composite score means of political trust for high- and low sophistication groups are relatively minor, it is important to analyze the measurement characteristics of the political trust construct.

Political trust remains a prevalent topic in political science research. A vast amount of empirical research into consequences of political trust has shown that citizens with lower levels of trust in political institutions engage less often in institutionalized forms of political participation and more often undertake system-challenging political behavior (Hooghe and Marien 2013), and they are less likely to comply with the laws (Marien and Hooghe 2011). Having crucial implications for political participation and law-abiding behavior, political trust is considered to be an essential indicator of legitimacy in democratic regimes (Levi and Stoker 2000), and therefore seemingly declining levels of political trust in Western democracies over the last couple of decades have attracted considerable scholarly attention.

In this paper we address the question to which extent these trends, or the absence thereof, may be attributed to methodological artefacts, in particular shortcomings in the measurement of political trust. Cross-national and over-time studies attempting to chart and explain the trends in levels of political trust commonly employ additive indices of multiple survey items tapping on confidence in various political institutions. A major pitfall of these empirical studies is that they overlook basic measurement considerations of political trust. Firstly, the composite score model unrealistically assumes that trust in different political institutions equally weights and loads on the latent political trust construct. Secondly, the composite score model that is frequently employed in studies on political trust assumes indicators to be free from measurement errors. However, as Dekker (2012) illustrates, incidental measurements of political trust are highly sensitive to context. Measures from different simultaneous surveys and panel measures that are only 1 month apart from each other can still vary to a substantial degree. Measurement error and index artifacts should of course not be confused with real differences. Therefore, political trust measures need to be analyzed with statistical models that allow for different weights for each indicator and that control for random fluctuations in those measures.

The neglect of potential measurement problems of political trust raises doubts about the findings of earlier studies. Therefore, in the current study, we revisit the measurement of political trust through robust statistical analysis. We utilize European Social Survey (ESS) data for the Netherlands across five time points (2004, 2006, 2008, 2010, and 2012) and four-wave panel data from the Panel Component of ESS in the Netherlands. While panel data allows us to identify the measurement error, stable trait and over-time change components of political trust, the cross-sectional time series data from the ESS study allows us to explore the comparability of the political trust construct over a longer period of time.

We pay special attention to differences between political sophistication groups. Debates on the development of political trust have pointed to political sophistication as a key explanation (Bovens and Wille 2010; Dalton 2004). We start our analysis by testing measurement invariance of the political trust constructs among political sophistication groups in order to establish whether it is meaningful to compare the means of political trust constructs across groups and over time. We then utilize multi-group confirmatory factor analysis and autoregressive simplex model analysis to estimate and compare mean political trust scores for high and low political sophistication groups.

The empirical domain of this paper is restricted to the Netherlands, in the period 2004–2012. This restriction is substantively not important as the methods and—most likely—the results too can be generalized to other settings. First, the measurement items used in this study are common measures of political trust and are used in a variety of countries and settings. Secondly, although the Dutch case in the period 2004–2014 does exhibit some particularities with regard to the trend in political trust (see below), the structural relationships of political trust with other theoretical constructs are comparable to those in other countries (Catterberg and Moreno 2006; Hooghe and Marien 2013). Short-term deviations are attributed to idiosyncratic changes in political culture (Bovens and Wille 2008, pp. 301–303; van der Meer 2010).

The contribution of this paper to the literature is thus twofold. First, our analyses highlight the general importance of explicating and testing assumptions about the measurement properties of concepts. Taking such assumptions for granted may result in flawed conclusions about trends and relationships. Our findings show that composite score models that implicitly assume trust indicators to have equal validity and also to be free from measurement errors, tend to overestimate the level of trust for both high- and low-sophistication groups. Our analysis further illustrates that composite score models yield exaggerated estimates of the differences between the trust levels of high- and low-political sophistication groups when the condition of full scalar invariance is not met. Secondly, the current study revisits the relationship between political sophistication and political trust by employing statistically robust analyses. Our findings illustrate that, when compared to their less sophisticated counterparts, highly sophisticated citizens do display higher levels of trust in three key components of democratic regimes, namely; parliament, politicians, and political parties. However, in sharp contrast with earlier research these differences are found to be minor.

Theoretical Background and Previous Research

Trends in Political Trust

Political trust, generally defined as citizens’ confidence in political institutions, is an important indicator of political legitimacy—the belief in the righteousness of these political institutions and the regime of which they are part. A widespread belief in legitimacy is commonly regarded as a necessary condition for the survival of political regimes. Therefore, measuring the level and development of political trust may provide us with important information about the stability of political systems (Easton 1965, 1975). Trust in the political regime of a country constitutes a reservoir of good will for when the day-to-day performance of the regime fails to meet expectations. Given these crucial implications, political trust is often considered as an essential component of the civic culture that is necessary for stability of democratic systems (Almond and Verba 1963). Therefore, the seemingly decreasing levels of political trust in Western democracies over the last couple of decades (Dalton 2004, 2005; Klingemann 1999) have stimulated a growing body of research on the causes and consequences of political trust.

The presumed decline of political trust (and, by implication, the presumed decline of the legitimacy of political systems) is often attributed to long-term processes of modernization and globalization. At the level of the individual citizen, modernization implies among other things a rise in the level of education (Klingemann and Fuchs 1995). Modernization theory states that the increased level of education, in combination with increasing political interest and a decreasing respect for traditional authorities and institutions leads to a growing dissatisfaction of higher-educated citizens with the working of the political system (Aarts et al. 2014; Dalton 2004; Thomassen 2005). This growing dissatisfaction presumably translates into a decrease in the trust in political institutions, as these institutions apparently fail to do what modern citizens expect from them.

The effects of economic globalization, on the other hand, work in a different direction. Economic globalization primarily impacts on those citizens who in the process become less competitive on the labor market. In Western countries, these citizens are primarily workers in those production segments of the economy which can relatively easily be moved to other countries or other parts of the world where the costs of production can be optimized (see for example Kriesi et al. 2008). For these workers, replacement jobs are hard to find since their level of education is relatively low. Higher educated persons, in contrast, are much better suited for jobs in economic sectors that will survive the first waves of globalization, e.g. the service sector, research and development, and other jobs with relatively high qualifications. In short, economic globalization will likely have negative consequences for the lower strata of the labor market, which tend to be the lower-educated citizens. It is to be expected that these negative experiences will in turn lead to a decrease of political trust among this group.

Modernization and globalization are thus expected to have diverging effects on the development of political trust of different groups in society. The modernization process is assumed to lead to a decrease of political trust especially among the higher-educated citizens. This assumption is also known as “positive effects hypothesis”. The globalization process, on the other hand, is thought to lead to a decrease of trust among the lower-educated citizens. This is known as the “negative effects hypothesis” (Dalton 2005).

Recent research into the case of the Netherlands has stressed the central role of education in understanding and explaining attitudes towards politics, including political trust. The phrase ‘diploma democracy’ has been coined by Bovens and Wille (2010). ‘Diploma democracy’ primarily refers to the alleged disappearance of lower-educated citizens from political life (Bovens and Wille 2012; Hakhverdian et al. 2011). Looking at political trust specifically, diploma democracy suggests that political sophistication acquired through education is the single strongest explanatory variable in understanding varying levels of political trust. According to this view, on top of the important consequences of economic globalization, the political arena itself has increasingly become a domain where only highly qualified, politically sophisticated citizens can exert influence and hence the less educated citizens become more and more alienated from politics (Bovens and Wille 2009, 2010). Supporters of the diploma democracy thesis argue that this feeling of exclusion from politics is causing cynicism and distrust among the less educated and less politically sophisticated citizens. These feelings of political exclusion add to the threat of economic exclusion as a result of economic globalization (Bovens and Wille 2010, pp. 415–416).

The empirical studies investigating the differences in political trust levels between the higher and lower educated citizens suggest that there is a gap between the two. The extent of this gap is not fully clear. Bovens and Wille (2010, pp. 412–413) quote several studies that find a positive relationship between education and trust. Other studies point to rather weak relationships, especially when other explanatory factors are taken into account (e.g. Listhaug and Wiberg 1995). However, the neglect of potential measurement problems of political trust raises doubts about the findings of these earlier studies documenting the gap between the levels of political trust among high and low sophistication groups.

Political Trust: Critical Look at the Common Analytical Strategies

In the current section, we discuss the common analytical strategies in political trust research and their methodological shortcomings that we briefly presented already in the introduction section. In the study of political trust, two practices seem to be widespread. The first of these is single-item measurement. Individual items asking about trust in parliament, or trust in the government, are analyzed as if these are valid and reliable indicators of the concept to be measured (Hetherington and Rudolph 2008; Newton 2001; Rudolph and Evans 2005; van der Meer 2010). However, when a concept is measured by a single item, the validity and reliability of the measure cannot be assessed at all, as it is impossible to distinguish between the various components of the measurement (true score, systematic error, random error) through single occasion measurement designs.1 Single item measurement thus results in a poor form of operationalization. Not only is “trust” defined as what is measured by a survey item, but the survey item also deliberately refers to only one object of political trust.

The second, and more commonly employed practice is that researchers construct political trust indices by adding or averaging the levels of confidence that individuals have for a set of political institutions (Bovens and Wille 2008; Brewer et al. 2005; Catterberg and Moreno 2006; Hendriks 2009; Marien and Hooghe 2011; van der Brug and van Praag 2007). However, these sum or composite score models that are frequently employed in political trust research have often been criticized for making a couple of unrealistic assumptions that jeopardize the robustness of statistical findings (Saris and Gallhofer 2007, pp. 314–315). Firstly, in sum or composite score models, each indicator is assumed to be equally contributing to the underlying latent construct. As far as the political trust constructs are concerned, utilizing these models means that the measures of trust in different political institutions equally weight and load on the latent political trust construct. However, previous research illustrates that validity coefficients for individual trust indicators vary to a substantial extent, and therefore the regression coefficients for each indicator needs to be corrected for by applying appropriate weights while calculating sum scores for obtaining robust findings (Saris and Gallhofer 2007). Secondly, composite score models further assume constitutive measures of the construct (the survey items) to be free of measurement errors. In other words, in these models, indicators are treated as perfectly reliable measures of the underlying political trust construct. Yet, MTMM experiments2 conducted with ESS measures tapping on trust in institutions demonstrate that political trust indicators are susceptible to measurement errors.3 Furthermore, not taking measurement errors into account also means that the reliability of survey questions is assumed to be equal across groups. Previous research, however, demonstrated that the reliability of the survey items is significantly contingent upon respondents’ levels of political interest and education (Alwin and Krosnick 1991; Judd et al. 1981; Judd and Milburn 1980), and hence measurement errors are not invariant for low and high political sophistication groups. Therefore, when comparing political trust scores across sophistication groups the differences in reliability of trust indicators need to be factored in the statistical models.

Confirmatory factor analysis (CFA) models can overcome the shortcomings of the analytical approaches mentioned above by allowing different factor loadings and accounting for measurement errors. CFA models are frequently employed in the study of various attitudes such as attitudes towards immigration (Meuleman et al. 2009; Savelkoul et al. 2012), welfare (van Oorschot and Meuleman 2012) and democracy (Ariely and Davidov 2010) with the purpose of comparing latent constructs means across populations and over-time. However, there are only a few studies on political trust that actually employ these measurement models that takes validity and reliability considerations into account (Allum et al. 2011; André 2014; Davidov and Coromina 2013; Poznyak et al. 2014; van Elsas 2014).

Data

For the first part of the study, we utilize European Social Survey data from 2004 to 2012 for the Netherlands. The data used for the second part of the study is obtained from Panel Component of European Social Survey, a developmental project aiming at facilitating biannual cross-sectional ESS survey and funded by the Netherlands Organization for Scientific Research (NWO). The panel study started with ESS Round-5 in October 2010 with the participation of 1829 respondents recruited through probability sampling of households and representative of the Dutch population over the age 16. After completing the ESS survey, all respondents were asked to take part in the panel study and 1501 of those respondents agreed to take part in the panel study.4 The subsequent waves of the study were conducted between May 2011 and January 2013 with 8 month intervals in-between. The last wave of the panel study completed with the participation of 647 respondents. The response rates for the panel study were 0.60 for the initial ESS study and 0.72, 0.70 and 0.86 for the subsequent waves, respectively.

For the current study, we used list-wise deletion, which yielded total samples of 1831, 1841, 1745, 1785, 1812 respondents for five consecutive waves of ESS (2004, 2006, 2008, 2010, 2012), respectively, and a total sample of 543 respondents in the panel study. While the mean age was 49.2, 48.6, 49.0, 50.0, and 51.1, the average years of formal schooling is recorded as 12.3, 13.3, 13.3, 13.4, and 13.6 in the final samples of subsequent waves of ESS, respectively. In the panel study, the final sample had a highly even gender distribution of 273 male (50.3 %) and 270 female (49.7 %) respondents, with a mean age of 51.6 (SD = 16.24) and with an average of 13.9 (SD = 4.07) years of formal schooling.