Democracy and Bureaucracy: How their Quality Matters for Popular Satisfaction

Abstract

This study focuses on how citizens’ satisfaction with democracy is affected by system factors related to the input side of the democratic system, i.e. the electoral dimension, as well as to factors related to the output side, i.e. the quality of government. The study investigates the extent to which different election system characteristics and degrees of policy representation (in terms of ideological congruence between voters and representatives) and the presence of effective, professional and impartial governmental institutions are related to citizens’ satisfaction with the way democracy works. The study uses data from CSES on individual voters and party system characteristics together with data on governmental institutions from the QoG Institute. Overall, the results indicate that factors such as government effectiveness are of greater importance for citizens’ satisfaction with the way democracy functions, compared to factors like ideological congruence on the input side. Impartial and effective bureaucracies matter more than representational devices.

Notes

1. A plausible explanation for the different results is that levels of satisfaction with the working of democracy depend on the sample of studied countries. Compared to Klingemann’s study from 1999, Aarts and Thomassen’s more recent study (2008) covers several more newly democratised countries in eastern Europe, where most of them have proportional electoral systems at the same time as the levels of satisfaction with the working of democracy are lower (see Figure 1 for varying levels of satisfaction with the working of democracy in different countries).

2. For example Aarts and Thomassen (2008) used data from CSES module 2, while we are using data from CSES modules 1, 2 and 3.

3. One reason for using this two-step approach is mainly practical since prior research together with our hypotheses derived from the QoG research field leave us with too many variables related to the political system. The problem is thus purely statistical since we can include only a limited number of variables on the system level since we have a restricted number of cases of 32 countries.

4. The data can be obtained from CSES Secretariat (http://www.cses.org), Center for Political Studies. Institute for Social Research. University of Michigan. The data can also be downloaded from: http://www.umich.edu/~cses.

5. Freedom in the World, Freedom House’s annual global survey of political rights and civil liberties’ (http://www.freedomhouse.org).

6. The case selections are thus not based on a random sampling procedure but rather on which cases are available in terms of data. This implies that there are limitations in the generalisability of the results.

7. Golder and Stramski (2010) also propose an alternative measure of congruence labelled relative congruence. We have also tested this measure but, in relation to our dependent variable, satisfaction with the way democracy works, the absolute congruence measure is superior, in terms of both effect and variance explained.

8. We have also tested other measures and operationalisations for electoral systems such as Gallagher’s least square index for the degree of disproportionality, but the relationships are highly similar.

9. Since we in this part of the study only use cross-section data, we are talking about correlation rather than causation in a strict sense.

10. The QoG-related variables in model 2 are highly correlated with an alpha of 0.91 and a VIF value of 5.37, which is an indication of strong multicollinearity. The schoolbook procedure would be to construct an index or a factor of all four variables. However, we are not primarily interested in maximising QoG but rather in measuring which aspects of QoG are most relevant in terms of satisfaction with the working of democracy.

11. According to Aarts and Thomassen (2008), there are substantial differences between the levels of satisfaction with democracy between older and more established democracies compared with more recently democratised countries (satisfaction with the way democracy works is in general lower among more recently democratised countries, which also could be seen in Figure 1). All models in Table 1 have been run with a dummy variable measuring the age of democracies. Our conclusions are not affected but the results are not robust. The reason for this is that the age of democracies is closely related to the quality of government and economic performance and the models become sensitive to which indicators of economy are included. By including a variable for old and new democracies, we are thus inducing too much multicollinearity in the model as we are running out of cases, which makes the OLS estimations inefficient. For this reason we do not control for the age of the democracies in the final models in Table 1.

12. It should be mentioned that in the CSES data some countries are represented with up to three elections, which means that the data are based on 32 countries in 71 elections. A problem is that for some of the QoG variables we have data for only one time point (impartiality and professionalism) at the same time as these variables, theoretically, may vary over time. For the two other QoG variables the available time points often do not coincide with the CSES time points. In order to keep the cases as comparable as possible, we have here pooled the data in the sense that we are using average values for countries so that the unit of analysis is not country/year but only countries, so even if we can measure congruence, for example, for every election we are here using an average congruence for each country. An alternative approach, since we in any case end up with government effectiveness as the strongest predictor among the QoG variables and government effectiveness is measured over time (although not for the specific election year in all countries), would be to interpolate government effectiveness from the years that are closest to the actual election years. Such an approach gives us 67 country/year observations. If the data are structured in this way the results from the final models of Table 1 are a bit mixed, but still it is only government effectiveness that yields a significant effect. However, with such an approach we give greater weight to some countries (which often tend to be countries with institutionalised election studies) at the same time as we actually are interpolating government effectiveness between different time points, which does not give a correct estimate. A third approach is to only use the last available time point for each country where we have data for the relevant variables. Here again the unit of analysis is countries but now only measured at one specific time point. When rerunning the models in Table 1, the effects of both ideological congruence and government effectiveness decrease, but still government effectiveness outperforms ideological congruence in terms of impact on satisfaction with the working of democracy. The drawback of such an approach is that we are throwing data away. For this reason, we argue that the most reasonable approach is to keep countries as the unit of analysis and base the estimations not on interpolations or specific years, but on average of years within countries as long as the analyses are run on the aggregate level.

13. For the democratic variables the results are quite robust in the sense that it is only ideological congruence that matters when it comes to satisfaction with the working of democracy.

14. In country comparative studies on voters, individuals are clustered within countries and persons from the same country often share common experiences, which makes them more similar compared to voters from other countries. This leads to an underestimation of standard errors due to an overestimation of the extent of independent information contained in the data. Multilevel models capture this kind of country variation by re-parameterising the intercepts and the coefficients of ordinary regression models into country-level equations and will therefore provide more accurate coefficients, standard errors and significance tests (Hox 2002; Snijders and Bosker 1999). For this reason we have let the intercept vary between the countries, a so-called random intercept model with fixed coefficients. Also, by controlling for the election waves (i.e. repeated cross-section observations for some countries), we do not need to consider the fact that the design of the CSES study implies that some countries are represented up to three times while other countries are only represented once.