The scale validity of trust in political institutions measurements over time in Belgium. An analysis of the European Social Survey,

The scale validity of trust in political institutions measurements over time in Belgium. An analysis of the European Social Survey, 2002-2010 Abstract Within the literature, there is an ongoing debate about the dimensional structure of trust in political institutions. While some authors assume this is a blanket judgment covering all institutions equally, others claim this form of trust is experience-based and therefore multidimensional. Based on five waves of the European Social Survey in Belgium, and using multiple group confirmatory factor analysis, we demonstrate that trust in political institutions is onedimensional, although it has to be acknowledged that some distinction between representative political institutions and policy implementing institutions should be made. We can demonstrate, furthermore, sufficient measurement stability over time, so that with some caution trends over time can be constructed. We close with some observations about what our results mean both for the theoretical status of political trust and for its evolution over time. Keywords: political trust, measurement, over time equivalence, multiple group confirmatory factor analysis, Belgium Introduction The evolution of political trust is a topic of ongoing concern in the literature. Since Easton (1965) it is assumed that that trust in political institutions functions as a form of diffuse support for the political system, and therefore is an important resource for the stability of democratic political systems. It is less clear, however, whether in the current era levels of political trust are declining, as some authors have argued (Inglehart & Catterberg, 2002), or whether they should rather be seen as stable (Marien, 2011). Analyzing this question is not just an empirical matter. In the literature, two competing views on the nature of political trust can be found. Some authors consider political trust as a form of blanket judgment, covering the entire political system. If this is the case, we can assume that political trust has a rather stable structure and changes only slowly over time. Looking at the pattern of geographical distribution of political trust levels, it can indeed be observed that these differences do not seem to change much over time (Hooghe, 2011; Uslaner, 2002). Other authors, however, are 1

more strongly inclined to see trust as an assessment of the actual functioning of political systems. The underlying logic is that institutions will be judged by public opinion on their own merits, and that as a consequence citizens will express specific trust in specific institutions (Bovens & Wille, 2008; Van de Walle, Van Roosbroek & Bouckaert, 2008). While the first conceptualization of political trust would imply that trust in political institutions is a solid one-dimensional latent concept, in the second, experience-based conceptualization, we do expect changes to occur in the relation between the individual items. Unless the performance of all government institutions would develop in exactly the same manner, one can expect that the satisfaction with some institutions will drop, while for others it will rise. In this paper, our aim is to analysis the scale validity of trust in political institutions over five observation points in the first decade of the 21 st century. The analysis is conducted in Belgium, a country that has known some substantial turmoil during this period. Belgium, therefore can be considered a conservative test: in a divided and volatile political system like Belgium, it is more likely that trust in political institution will not be a one-dimensional latent concept. Literature The importance of political trust has been widely recognized in political science literature. Most authors believe that a high level of political trust leads to acceptance of and compliance with political decisions (Marien & Hooghe, 2011). It is argued that trust in political institutions motivates citizens to assign a greater role to those institutions in providing collective goods for society (Hetherington, 1998). The theoretical status of this form of trust, however, is highly disputed. Hardin (2002) has even made the case that trusting political institutions for most citizens is not warranted at all, since they do not have the necessary information to arrive at a well-informed decision about the trustworthiness of those institutions. Within the literature on trust in political institutions, two main research lines can be distinguished: while some authors see this form of trust as diffuse and one-dimensional, others see it as experience-based and therefore multi-dimensional. Easton (1965) already argued that trust in political institutions first of all should be understood as a form of diffuse support, that is not related to the functioning of specific 2

institutions or office-holders. Rather, it should be seen as a general disposition toward the function of the political system as a whole, and as such it forms a stable base for the perceived legitimacy of that system. It has been argued therefore that trust in political institutions is not based on actual experiences, but rather amounts to a general judgment about the way the political system as a whole functions (Hetherington, 2005). This claim is further supported by the fact that among adolescents too, rather stable and well-structured attitudes with regard to political trust have already been found, even although they did not have any experience with any one of these institutions themselves (Claes, Hooghe, & Marien, 2012). Hooghe (2011) has made the point that such a blanket judgment, covering all forms of institutions in an equal manner, can still be rational. Research on quality of government has shown that there is a very strong correlation between the performance of various government agencies (Rothstein 2012). It can argued, therefore, that there is a general political culture, governing the conduct of political elites, no matter what specific function they have in the system. As Hooghe (2011, 275) summarizes it: Looking at studies on corruption, there are no examples of countries where Members of Parliament are corrupt but the government is clean. The degree of trustworthiness is therefore not an individual characteristic of a person, or even of a political party or an institution, but of the political system as a whole. As such it makes sense that my opinion on various actors loads on a single latent variable. The assumption in this line of the literature, therefore, is that citizens do not make any distinctions with regard to the way institutions as such perform. Other authors, however, claim more strongly that citizens do react to either the performance of institutions, or to their actual experience with some of these institutions (Bovens & Wille, 2008; Damico, Conway, & Damico, 2000). Furthermore, it is argued that the norms governing various institutions will differ: it is expected that e.g. Members of Parliament will behave in a different manner than e.g. police officers or judges (Fisher, Van Heerde, & Tucker, 2010). This experience-based conception of trust in political institutions therefore pays much more attention to rapid fluctuations in trust levels. It can be assumed, e.g., that a corruption scandal in Parliament will lead to lower levels of trust in Parliament, while it should not have an impact on the level of trust in e.g., the police or the courts (Bovens & Wille, 2008). This view also implies that trust measurements can be used to some extent as a proxy for satisfaction measurements: if some institutions develop a policy that will augment the quality of the services they provide to the public, this should, in the long or the short run, result in higher levels of trust in this specific institution (Van de Walle, Van Roosbroek & Bouckaert, 2008). 3

Rothstein and Stolle (2008) argue more specifically that citizens in this respect make a distinction between two groups of institutions. On the one hand, there are the institutions of representative democracy (Parliament, parties, politicians and ministers), where it is assumed that office holders here should be held accountable to public opinion. For the institutions of law and order (police, the courts ) on the other hand, we expect less accountability, but impartiality is here the main guiding norm. Rothstein and Stolle argue therefore that in stead of lumping all these institutions together into a diffuse measurement, both dimensions should be clearly distinguished. The theoretical discussion about the status of a trust in political institutions measurement is directly relevant to the ongoing debate about the trends in political institutions. If trust in political institutions is indeed a blanket judgment, the scale can be used in a rather straightforward manner to assess trends over time in the level of political trust. If however, trust in political institutions is multi-dimensional, this implies that it is not methodologically correct to make general statement about trust in political institutions. In that case, it is perfectly possible that e.g., the trust in Parliament would rise, while the trust in the courts would decline. Furthermore, it remains to be ascertained in that case whether the dimensional structure is stable over time, as some institutions might become more salient than others over time (Allum, Read, & Sturgis, 2011; Marien, 2011; Van de Walle, Van Roosbroek, & Bouckaert, 2008). Based on the literature therefore, we can arrive at the main research question, which is: can trust in political institutions be considered as a one-dimensional concept, and if so, is this measurement equivalent over time? Data and methods European Social Survey To examine our research question we use data form the European Social Survey (ESS). The original goal of the ESS was to design a survey that could chart and help to explain the interaction between Europe s changing institutions and the attitudes, beliefs and behavioural patterns of its population (European Social Survey, 2012). The first round of data collection was conducted in 2002. Since then the survey is held biannually. The battery on trust in political institutions has been included in every round of the ESS (Stoop, Jowell, & Mohler, 2002). This means we can use all current available data for our analyses, meaning data the years 2002, 2004, 2006, 2008 and 2010. 4

The battery measuring trust in political institutions as found in the ESS is as followed: Using this card, please tell me on a score of 0-10 how much you personally trust each of the institutions I read out. 0 means you do not trust an institution at all, and 10 means you have complete trust. The institutions are the country s parliament, the legal system, the police, politicians, political parties, the European parliament and the United Nations (UN). Unfortunately we cannot take trust in political parties into account during the analysis, because this item is not asked in the first round and given our research question it is important to include only items that were asked in an identical manner in all five waves of the European Social Survey. The Belgian case The ESS is a cross-country survey that is being conducted in over 30 European countries. This makes the data ideal for cross-national comparison (Allum et al., 2011). In this paper, however, we are not interested in cross-cultural measurement equivalence, as this topic already has been investigated in other studies, but rather in measurement equivalence over time. In order not to confound both forms of measurement equivalence, in the current manuscript we limit ourselves to just one country, where we have opted for Belgium. We have done so, first because with regard to the level of trust in institutions, Belgium can be considered as an average case for Western Europe. Second, however, it is a political system that is plagued by an endemic legitimacy crisis. Back in the 1990s, Belgium was first hit by a shock with regard to trust in political institutions when it turned out that the police and the courts had bungled the investigation toward a serial child murderer (Hooghe, 1998). During our observation period, Belgium experienced a number of political crises as a result of the ongoing debates between the Dutch and the French language group. Since 2007 Belgium has known governmental instability, and the 2010 elections led to the electoral breakthrough of a separatist party that wants to break up the country in three independent regions (Deschouwer, 2012). Belgium, therefore, can be considered as a conservative test. We can assume that public opinion in the country will express different judgments in the various branches of government, and judgments might differ between the federal and the regional level. 1 If, even in these circumstances, trust in political institutions still proves to be one-dimensional and equivalent over time, this offers firm support for the blanket judgment hypothesis. 5

Methods Initially we conducted a principal component analysis (PCA) using SPSS. The goal of this analysis is to simplify the correlation matrices. It tests whether different variables can be accounted for by a small number of factors (Kline, 1994). Subsequently we conducted a confirmatory factor analysis (CFA) per round of data collection using Mplus (Muthén & Muthén, 2012). This model will be used for the further analyses. The model will be evaluated by investigating the root mean square error of approximation (RMSEA) and comparative index (CFI). The RMSEA is used to test the model based on an absolute fitwhile the CFI is used to evaluate the fitted CFA solution. To test for over-time measurement invariance we conduct a multiple group factor analysis (MGCFA), again using Mplus software (Muthén & Muthén, 2012). This method entails five hierarchical levels (Steenkamp & Baumgartner, 1998). The first level is configural invariance, which basically means that the pattern of salient and non-salient loadings has to be the equal across all groups. This form of invariance states if the construct political trust can be measured using a single factor in all rounds of data collection. The next level is metric invariance. In this form of measurement invariance, factor loadings on the items are fixed to be equal among all years. When metric invariance is established, mean scores on items can be compared meaningful across observations. The third level is scalar invariance. This is an even stricter test because here also the intercepts are constrained to be equal across all years (Rensvold & Cheung, 1998). Scalar invariance is required to conduct latent mean comparisons across groups and time, which is more robust than mean score comparisons (Allum et al., 2011). As mentioned, measurement invariance test have a fourth and fifth level, which test for factor covariance invariance and error variance invariance respectively (Steenkamp & Baumgartner, 1998). These tests fall out of the scope of this article though, since we only want to compare latent means to detect trends and in practice these levels are never investigated in this kind of survey research. We will conduct the MGCFA according to the method suggested in the previous paragraph. This means that during a first round we will test for configural invariance, during a second for metric invariance and during a third for scalar invariance. Even though strictly speaking the indicators are ordinal, with 10 discrete points, we regard them as continuous for the analysis (Allum et al., 2011). As estimator we use maximum likelihood (ML) because the distribution of the data is approximately normal. ML is considered to be the most robust estimator when not much distortion occurs (Bollen, 1989). To test for model fit every round, we rely on three different fit indices (Bollen & Long, 1992; Bollen, 1989; Mulaik, 2007). The basic goodness- 6

of-fit test is the chi-square test (χ²). Unfortunately this test is highly sensitive for large sample sizes. This means that it will indicate significant differences, even if in reality the differences are small (Bollen & Long, 1992; Rensvold & Cheung, 1998). Especially in cross-national and longitudinal research this can be problematic, since they usually deal with a large sample size. That is why, in addition to the chi-square test, we take two other model fits into account, the RMSEA and CFI. These three model fit indices should complement each other (Bollen & Long, 1992; Mulaik, 2007). Results The first test is a principal component analysis that allows us to establish whether or not political trust is a one dimensional construct. The results in Table 1 show that the items only measure one component or construct. This means that political trust is indeed, and quite strongly, a one dimensional concept. The results also give us a first indication of the items. They all load highly on the construct, which indicates a good model. But this has to be further examined by testing for invariance while conducting a confirmatory factor analysis. Table 1. Results of PCA for all rounds of ESS. Component round 1 Component round 2 Compontent round 3 Component round 4 Component round 5 Trust in Belgian 0.824 0.809 0.822 0.824 0.821 parliament Trust in legal system 0.774 0.783 0.767 0.789 0.786 Trust in police 0.690 0.668 0.653 0.701 0.691 Trust in politicians 0.846 0.821 0.795 0.810 0.799 Trust in European 0.845 0.827 0.840 0.840 0.865 parliament Trust in UN 0.751 0.737 0.753 0.768 0.780 Eigenvalue 3.750 3.613 3.596 3.745 3.766 Explainedvariance 62.49% 60.22% 59.93% 62.42% 62.76% N 1600 1627 1691 1667 1637 Entries are the result of a principal component analysis for the Belgian survey in all rounds of ESS. The next step is to test for configural invariance. We investigate the factor loadings and error variance for all rounds, which are summarized in Table 2. The factor loadings represent the relationship between the latent construct and the items. A factor loading is considered salient when it has a value above 0.30 (Brown, 2006). The error variance is the opposite,it is the part of the item that is not explained by the latent construct. As Table 2 shows, all items have a 7

factor loading above 0.30. This means that the item measurement contributes significantly to the latent concept of trust in political institutions. This allows us to proceed with our analyses. However, we can notice some interesting features when looking at Table 2 where we investigate configural invariance. The factor loadings on the measures of trust in representational institutions (i.e., parliaments and politicians) are higher than the factor loadings on the implementing institutions (i.e. legal systems and the police) and the United Nations. Correspondingly, the error variance is higher for the latter. Although these differences do not endanger the one-dimensional character of the concept, this does lend some support to the observation of Rothstein and Stolle (2008) that both dimensions should be distinguished. 8

Table 2. Configural invariance for trust in institutions (ESS for Belgium). Wave 2002 Wave 2004 Wave 2006 Wave 2008 Wave 2010 Factor loading Error variance Factor loading Error variance Factor loading Error variance Factor loading Error variance Factor loading Error variance Belgian 0.819 0.328 0.798 0.363 0.824 0.322 0.825 0.319 0.819 0.329 parliament Politicians 0.843 0.290 0.814 0.337 0.782 0.389 0.802 0.357 0.787 0.381 Legal system 0.677 0.542 0.695 0.517 0.660 0.565 0.700 0.510 0.700 0.510 Police 0.555 0.692 0.533 0.715 0.519 0.731 0.571 0.674 0.566 0.680 EU parliament 0.786 0.383 0.769 0.408 0.768 0.411 0.766 0.413 0.804 0.535 United Nations 0.634 0.598 0.607 0.632 0.628 0.605 0.647 0.581 0.661 0.563 Source: ESS 2002-2010 + own calculations.notes: standardized factor loadings and error variance from a MGCFA with no constraints on the full sample. Fit indices: χ² = 1839.844; df = 45; p-value = 0.000; RMSEA = 0.149; CFI = 0.927; N (wave 2002) = 1889; N (wave 2004) = 1776; N (wave 2006) = 1798; N (wave 2008) = 1760; N (wave 2010) = 1704. 9

In Table 3 we investigate covariance between items, more specifically covariance between trust in the legal system and trust in the police. Because of the established covariance the error terms between these two items are allowed to correlate in our model. The final remark is about the high level of variance between trust in the European parliament and trust in United Nations. By allowing these items to correlate, we improve the measurement fit. Table 3. Allowed correlations to improve model fit. Wave 2002 Wave 2004 Wave 2006 Wave 2008 Wave 2010 Factor loading S.E. Factor loading S.E. Factor loading S.E Factor loading S.E Factor loading S.E Police with 0.247 0.024 0.292 0.024 0.314 0.023 0.318 0.024 0.305 0.024 legal system EU parliament with UN 0.392 0.024 0.338 0.026 0.432 0.023 0.406 0.024 0.467 0.023 Source: ESS 2002-2010 + own calculations. Notes: standardized factor loadings and standard error from a MGCFA with no constraints on the full sample. Improved fit indices: χ² = 136.644; df = 35; p-value = 0.000; RMSEA = 0.040; CFI = 0.996; N (wave 2002) = 1889; N (wave 2004) = 1776; N (wave 2006) = 1798; N (wave 2008) = 1760; N (wave 2010) = 1704. This leads us to the measurement model presented in Figure 1. After conducting a MGCFA, this model provides an excellent fit, which indicates configural invariance. The RMSEA equals to 0.040, which is according to generally accepted standards (Brown, 2006). Our model has a CFI equal to 0.996, which again indicates a very good fit. Therefore, this model will be used for the other measurement invariance tests Figure 1. Measurement model of political trust. Notes: standardized factor loadings and error variance from a MGCFA with no constraints on round 1 of ESS. The standardized factor loadings and error variance from a MGCFA with no constraints on the other four rounds are shown in Appendix A. 10

We proceed to test for metric invariance. We conduct another MGCFA but now we constrain all factor loadings to be equal across the five rounds. Figure 2 summarizes the factor loadings and error variance of this analysis. We can observe the same conclusions as before: all items still measure one latent variable and there is still some difference between the representational and implementing institutions. But more important are the model fit indices. The added constraints do not worsen the fit. The RMSEA even improves a little (RMSEA = 0.030) and the CFI decreases negligible (CFI = 0.995). Moreover the chi-square ratio is not significant (Δχ² = 45.800; df = 32). This means that the chi-square test shows no significant difference between the configural and metric model. We can thus conclude that the measurement of political trust is metric invariant. Figure 2. Metric invariance for the political trust scale in all waves. Notes: standardized factor loadings and error variance from a MGCFA with metric constraints on the full sample. Metric constraints: the factor loadings are constrained across all waves, while the intercepts, error terms and variance of the latent concept are free. Fit indices: χ² = 197.893; df = 75; p-value = 0.000; RMSEA = 0.030; CFI = 0.995; N= 8927. To test for scalar invariance, we constrain the intercepts to be equal across all rounds, in addition to the previous restrictions on the factor loadings. The results of this test are shown in Table 4. We were not able to detect full scalar invariance since the chi-square ratio was significant (Δχ² = 434.827; Δdf = 20). This means we cannot safely compare latent means across time. This result is, however, not surprising. While interpreting metric invariance we found a dichotomy between types of institutions and this makes it harder to establish scalar invariance. However, according to Byrne, Shavelson, & Muthén (1989), we can still test for partial scalar invariance. If this can be confirmed, we can still compare the latent means. We 11

followed Sörbom & Jöreskog's (1982) method to establish this. We first determine which round has the highest chi-square contribution, then, according to modification indices, we release the constraint on the intercept of a particular item. Table 4 summarizes the model build-up toward partial scalar invariance. We had to relax the constraints on 5 items before establishing scalar invariance. For round 2002, 2008 and 2010 we relaxed trust in politicians. Additionally, we relaxed trust in the Belgian parliament for round 2002 and 2006. From this we conclude that the measurement of political trust is partial scalar invariant (Δχ² = 84.349; Δdf = 15), especially since we were able to obtain an excellent model fit (RMSEA = 0.036, CFI = 0.992). The relaxed items can be explained by looking back at the score means (Table 5). First we notice that in almost every round the intercept of trust in parliament is relaxed (round 2002, 2008 and 2010). This is not surprising since the score means of trust in parliament has strong fluctuations between the rounds. The same explanation holds for trust in politicians. Trust in legal systems, trust in the police, trust in the European parliament and trust in the UN were less subjected to changes throughout all rounds. Therefore it makes sense that these item did not need to be relaxed because the intercepts were already more or less equal. 12

Table 4. Scalar invariance models for across time equivalence (ESS). Model χ² df RMSEA CFI Wave with highest χ² contribution contribution 1. Configural invariance 136.644 35 0.040 0.996 2. Metric invariance 182.444 67 0.031 0.995 3. Scalar invariance 617.271 87 0.058 0.978 Wave 1 parliament 107.824 4. Free wave 1 τ parliament 504.753 86 0.052 0.983 Wave 1 politicians 116.303 5. Free wave 1 τ politician 386.093 85 0.045 0.988 Wave 3 parliament 38.962 6. Free wave 3 τ parliament 374.893 84 0.044 0.988 Wave 5 politicians 26.894 7. Free wave 5 τ politician 347.658 83 0.041 0.989 Wave 4 politicians 37.706 8. Free wave 4 τ politician 266.793 82 0.036 0.992 Source: ESS 2002-2010 + own calculations. Test statistic for the entire sample.notes: results are the test statistics of a MGCFA with scalar constraints: the factor loadings and intercepts are constrained, error terms and variance of the latent concept are free. 13

By establishing partial scalar invariance, we are allowed to compare latent mean scores. However, we have to be careful with our conclusions since they are not based on full scalar invariance. The latent means indicate that political trust stays more or less stable over time. When calculating the chi-square ratio to determine whether or not the latent means in the later rounds are different from the initial situation, we found they do not differ significantly. By establishing metric invariance, we can compare mean scores (intercepts) on the items over the different rounds. The mean scores are summarized in Table 5, which shows some interesting observations. First, we see an overall decline in the trust in the Belgian parliament. Trust in the legal systems and police, on the other hand, slightly increase. These institution are not affected by the political turmoil. Trust in the European parliament increases first, but declines from 2008 onward. To Summarize, trust in the implementing institutions seems to increase, while trust in the representational institutions of Belgium seems to decrease during political instability. This could have important implications. It suggests that citizens to some extent evaluate these two types of institutions differently. However, to obtain an even better view on the evolution of political trust, we still have to establish scalar invariance. Table 5. Mean scores (intercepts) of metric model. Wave 2002 Wave 2004 Wave 2006 Wave 2008 Wave 2010 Trust in Belgian 2.215 2.104 2.282 2.113 1.992 parliament Trust in 1.949 1.951 1.974 1.867 1.738 politicians Trust in legal 1.743 2.029 1.997 2.076 2.079 system Trust in police 2.449 2.593 2.682 2.754 2.585 Trust in EU 2.109 2.231 2.311 2.297 2.286 parliament Trust in UN 2.096 2.153 2.286 2.359 2.380 Source: ESS 2002-2010 + own calculations. N (wave 2002) = 1889; N (wave 2004) = 1776; N (wave 2006) = 1798; N (wave 2008) = 1760; N (wave 2010) = 1704 Note: standardized intercepts of all political trust items, based on the results of the metric model. 14

Table 6. Comparison of structural and latent wave means (all waves). Structural mean Latent mean µ Rank Ƙ Rank Round 2002 4.93 0.00 Round 2004 5.00 1 0.17 2 Round 2006 5.18 4 0.23 4 Round 2008 5.07 3 0.21 3 Round 2010 5.01 2 0.17 1 Source: ESS 2002-2010 + own calculations. Note: µ is the symbol for the structural mean of political trust, constructed by the means of the sum scale of the six political trust items. Ƙ is the symbol for the latent mean of political trust, as shown in the output of the multiple group CFA. We established metric and partial scalar invariance for the measure of the latent construct political trust. These equivalence tests are however not a goal by themselves. We now finally arrive at the main question: do these test contribute something to the understanding of the development of political trust over time? To answer this question we compare the structural means with the latent means, which were obtained by testing for measurement equivalence with multiple group SEM. To be able to interpret the means correctly, we have to note that the latent mean of round 2002 was fixed to equal zero and the other rounds has to be interpret in comparison. This is why the first round is not included in the ranking. Consequently, the structural mean of round 2002 is also not included in the ranking. For the first and second place we notice that round 2004 and 2010 switch positions, although their scores are very similar (they switched position since the latent mean for round 2004 equals 0.172 and the latent mean for round 2010 equals 0.167). This difference is however neglectable. The ranking of the structural means and latent means for round 2006 and 2008 do not differ. Following this, the impact of using latent means should not be overstated, since both types of means lead to more or less similar conclusions. Discussion In this paper the goal was to establish?? the dimensional character of trust in political institutions and the measurement equivalence over time. This was done by conducting a principal component analysis and a confirmatory factor analysis on Belgian data from 2002, 2004, 2006, 2008 and 2010. The conclusion has to be that trust in political institutions, even in Belgium, is a one-dimensional latent concept, although it has to be acknowledged that we 15

do find some distinction between on the one hand the institutions that have a representative function, and on the other hand institutions that implement policy. We found the political trust measure to be metric and partially scalar invariant. This could provide researchers with insights on how to handle political trust questions. Establishing metric invariance entails that factor loadings of the measurement items are equal over time and that the mean scores may be compared over time. Additionally, this is an indication that political trust remains stable over time. As we have seen, we could not establish full scalar invariance. Some items seem to be highly subjected to changes between rounds, such as trust in parliament and trust in politicians. However, after relaxing the constraints on these items, partial scalar invariance was established. This is a further indication that political trust is actually stable over time and not declining as some authors claim. When this was established, we wanted to look at the trends in political trust. When looking at the items separately, we noticed some variation. Trust in the representational institutions (parliament, politicians) is in fact declining over the years. However, trust in the implementing (legal systems, police) and international institutions (United Nations and European parliament) remains stable or is even increasing. The conclusion therefore is that, at least for the Belgian case, there is no overall decline in trust in political institutions, but that we do see variations across various institutions. Furthermore, we wanted to examine whether equivalence test actually contribute to the understanding of the development of political trust. When comparing the structural and latent means, we found that these do not differ in a significant manner. 16

References Allum, N., Read, S., & Sturgis, P. (2011). Evaluating change in social and political trust in Europe. In E. Davidov, P. Schmidt, & J. Billiet (Eds.), Cross-Cultural Analysis. Methods and Applications (pp. 35 53). New York, NY: Routledge. Bollen, K. A. (1989). Structural equations with latent variables. New York: Wiley. Bollen, K. A., & Long, J. S. (1992). Tests for Structural Equation Models: Introduction. Sociological Methods & Research, 21(2), 123 131. Bovens, M., & Wille, A. (2008). Deciphering the Dutch drop: ten explanations for decreasing political trust in The Netherlands. International Review of Administrative Sciences, 74(2), 283 305. Brown, T. A. (2006). Confirmatory factor analysis for applied research. New York: The Guilford Press. Byrne, B. M., Shavelson, R. J., & Muthén, B. (1989). Testing for the equivalence of factor covariance and mean structures: The issue of partial measurement invariance. Psychological Bulletin, 105(3), 456 466. Claes, E., Hooghe, M., & Marien, S. (2012). A Two-Year Panel Study among Belgian Late Adolescents on the Impact of School Environment Characteristics on Political Trust. International Journal of Public Opinion Research, 24(2), 208 224. Damico, A. J., Conway, M. M., & Damico, S. B. (2000). Patterns of Political Trust and Mistrust: Three Moments in the Lives of Democratic Citizens. Polity, 32(3), 377 400. Deschouwer, K. (2012). From Protest to Power: Autonomist Parties and the Challenges of Representation. West European Politics, 36 (1), 289-291. Fisher, J., Van Heerde, J., & Tucker, A. (2010). Does One Trust Judgement Fit All? Linking Theory and Empirics. British Journal of Politics & International Relations, 12, 161 188. Hardin, R. (2002). Trust and Trustworthiness. New York: Russell Sage Foundation. Hetherington, M. J. (1998). The Political Relevance of Political Trust. American Political Science Review, 92(4), 791 808. Hetherington, M. J. (2005). Why Trust Matters. Declining Political Trust and the Demise of American Liberalism. Princeton: Princeton University Press. Hooghe, M. (1998). Selectieve uitsluiting in het Belgisch politiek systeem. Innovatie en protest door nieuwe sociale bewegingen. Res Publica, 40(1), 3 21. Hooghe, M. (2011). Why There is Basically Only One Form of Political Trust. British Journal of Politics & International Relations, 13(2), 269 275. Kline, P. (1994). An easy guide to factor analysis. London: Routledge. 17

Lipset, S. M., & Schneider, W. (1983). The Decline of Confidence in American Institutions. Political Science Quarterly, 98(3), 379 402. Marien, S. (2011). Measuring political trust across time and space. In S. Zmerli & M. Hooghe (Eds.), Political trust. Why Context Matters. Colchester: ECPR, University of Essex. Marien, S., & Hooghe, M. (2011). The Effect of Declining Levels of Political Trust on the Governability of Liberal Democracies. Paper presented at the Annual Conference of the Political Studies Association. London. Miller, A. H., & Borrelli, S. A. (1991). Confidence in Government During the 1980s. American Politics Research, 19(2), 147 173. Mulaik, S. (2007). There is a place for approximate fit in structural equation modelling. Personality and Individual Differences, 42(5), 883 891. Muthén, B. O., & Muthén, L. K. (2012). Mplus user s guid (7th ed.). Los Angeles: Muthén & Muthén. Rensvold, R. B., & Cheung, G. W. (1998). Testing Measurement Models for Factorial Invariance: A Systematic Approach. Educational and Psychological Measurement, 58(6), 1017 1034. Rothstein, B., & Stolle, D. (2008). The State and Social Capital An Institutional Theory of Generalized Trust. Comparative Politics, 40(4), 441 459. Sörbom, D.., & Jöreskog, K. G.. (1982). The use of structural equation models in evaluation research. In C. Fornell (Ed.), A Second Generation of Multivariate Analysis: Vol.2. Measurement and evaluation (pp. 381 418). New York: Praeger. Steenkamp, J.-B.., & Baumgartner, H. (1998). Assesing measurement invariance in crossnational consumer research. Journal of Consumer Research, 25(1), 78 107. Stoop, I., Jowell, R., & Mohler, P. (2002). The European Social Survey: one survey in two dozen countries.?? Plaats? European Social Survey (2012). Norway: Service, Norwegian Social Science Data. Retrieved from http://www.europeansocialsurvey.org/ Uslaner, E. M. (2002). The Moral Foundations of Trust. New York: Cambridge University Press. Van de Walle, S., Van Roosbroek, S., & Bouckaert, G. (2008). Trust in the public sector: is there any evidence for a long-term decline? International Review of Administrative Sciences, 74(1), 47 64. 18

Appendix A Figure 1, Appendix A: Solution Confirmatory Factor Analysis (wave 2004). Figure 2, Appendix A: Solution Confirmatory Factor Analysis (wave 2006). 19

Figure 3, Appendix A: Solution Confirmatory Factor Analysis (wave 2008). Figure 4, Appendix A: Solution Confirmatory Factor Analysis (wave 2010). Endnotes 1. It has to be noted, of course that in Belgium two different language groups have to be distinguished. If we split the sample however between Dutch and French speaking respondents, it can be observed that the structure of the attitudinal scale is the same in both language communities. While the level of political trust tends to be lower in the French speaking part of the country, this does not have an effect on the measurement qualities of the scale. 20