Intra-Party Politics and Public Opinion: How Candidate Selection Processes Affect Citizens’ Satisfaction with Democracy

Yael Shomer received funding for this research from the European Community's Seventh Framework Programme (FP7/2007-2013) under Grant Agreement No. 276914, and from the Norwegian Research Council Grant No. 222442. Gert-Jan Put thanks the Research Foundation Flanders (FWO) and the KU Leuven Junior Mobility Programme for their generous support. All datasets used in the paper and code for replicating the models are available at: https://​dataverse.​harvard.​edu/​dataset.​xhtml?​persistentId=​doi%3A10.​7910%2FDVN%2F4B5AD1.

Of course, parties may sometimes have other reasons to undertake these reforms, such as severe electoral defeat (Hopkin 2001; Pennings and Hazan 2001).

Note that Hazan and Rahat (2010) actually advocate a curvilinear relationship whereby exclusive selectorates promotes the lowest levels of competitions, primaries promotes medium range competition levels, and party delegates correlates with the highest levels of competition.

Yet, others argued that decentralized procedures do not necessarily encourage disunity (see: Hazan and Rahat 2010, p. 158).

Combining datasets between various comparative surveys is a reasonable approach, however the reader should be cautioned that design effects do differ across international surveys. For example, while the CSES are post-election surveys taking place not long after the election, the other comparative surveys used in this analysis deal with a broader set of human values and research themes. Thus, for some of the country years used in the analysis, the data collection process took place longer after the elections and candidate selection processes under investigation occurred. Despite these differences, we argue that the similarities between the various studies are strong. First, the question wording and response scale for the satisfaction with democracy item (our dependent variable) is similar for all surveys (except for the Israel 1999 country session in ESS, where a 0–10 scale was recoded to 4 point scale). Second, studying the methodological notes and reports for each of the comparative surveys boosts confidence that their resulting datasets can be combined. In terms of sampling, mostly equivalent sampling plans have been followed where multistage sampling has been used: primary units were localities (e.g. counties, regions, municipalities), selected according to their population size. In a second step, secondary units (individual respondents) were randomly selected within the primary units (e.g. on the basis of electoral registers). In sum, the reported minor differences do not outweigh the advantages of combining these comparative survey datasets. As Norris (2009) puts it, “when large-scale multi-national surveys covering many societies are combined with systematic variations in institutional and societal contexts, this process is capable of providing powerful insights for the study of comparative politics.

In order to estimate the effect of institutional determinants and selection procedures on the attitudes of citizens, we first, needed to make sure that the surveys were taken after the election (for which the selection of candidate took place) and the start of the parliamentary term. At that moment, candidate selection procedures already took place and survey respondents’ attitudes could have been influenced by it.

For the first three Belgian country sessions, we used data from the General Election Study Belgium, carried out by the Pole Interuniversitaire Opinion public et Politique (PIOP) and the Instituut voor Sociaal en Politiek Opinieonderzoek (ISPO). For the last two country sessions, ESS data on Belgian citizens was downloaded. With regard to the Israeli case study we used the Israel Democracy Institute's Israeli Democracy Index surveys.

Additionally, we are able to anecdotally demonstrate that the positive relationship we found between democratic selections and satisfaction are not a product of a temporal dynamic such that democratic selections' negative effect on satisfaction levels are in the short run, but their positive effect is a longer term process. Hence, we use Israel's Labour party and Belgium's CVP to test the potential temporal dynamic (results are available from the authors upon request). We thank an anonymous reviewer for directing our attention to Lieberman's article.

The validity of the satisfaction with democracy measure has been established and it was verified that it does not simply measure support for the incumbent government (e.g., Kornberg and Clarke (1992).

While intra-party variation in selection process exists, we do not address it in this paper and thus code a party's selection process according to the manner by which the majority of its MPs were selected. Additionally, we emphasize that many parties use complex-multi staged procedures to select their candidates (see Hazan and Rahat 2010). While our three point scales are rougher measures of inclusiveness and decentralization, they at least allow us the cross-national comparisons. A more detailed scale, which takes into consideration the complexity of the selection (like the 25 points scale of selectorate presented by Hazan and Rahat (2010)) are less suitable for such cross-national comparisons.

The on-line appendix contains a model with an additional aggregate-level measure of governmental performance: corruption levels. Unfortunately, while the literature clearly identifies citizens' micro-level subjective governmental performance evaluations as a key determinant of their satisfaction with democracy, this type of data is simply not available for many of the country-years we study, or are not comparable across surveys. Table d in the on-line appendix contains a model that controls for economic development (in addition to the single measure of GDP per capita): a 3-years GDP average growth in percentages (data was calculated using the variable rgdpna from the Penn tables). In order to calculate the 3-years average for each country-session, we took 3 years prior to the election year (+ the election year), and calculated the difference between the GDP for each year and the previous year and divided it by the previous year (three times for each country-session in the study). Thus, for example, for Argentina 2005, we calculated (GDP2003-GDP2002) / GDP2002, and did the same for 2003–2004 and 2004–2005. That gave us the annual change in GDP. We then calculated the average of the three of them to obtain a 3 year GDP average growth (in percentage terms). The table in the appendix reveals that the main results of the paper did not change, namely the positive relationship between democratic selectorate and satisfaction with democracy. Neither did results for GDP/capita or new-democracies, or any party and individual level covariates change. Interestingly, and as expected, controlling for all other covariates, economic development is positively correlated with satisfaction with democracy. Nonetheless, the inclusion of the growth measurement did alter the effect of some of the electoral system variables on satisfaction levels. Thus, in the current model neither citizens in SMD systems nor citizens in Mixed systems exhibit differing levels of satisfaction with democracy compared to respondents who live in CLPR systems. Nonetheless, while the two coefficients lose their statistical significance, their sign remains the same as the one presented in the paper. All in all it seems the results concerning the effect of selectorate and decentralization on satisfaction are robust to the inclusion of the economic development indicator.

We were unfortunately unable to control for sophistication levels, or strength of partisan attachments. We were able to collect respondents' ideological self-placement for all country-sessions in our study except for Japan 1996. Thus, the total number of country-sessions is reduced to 27 and the total number of parties is reduced to 128. Consequently we lose 472 respondents. Moreover, while most surveys use an 11 point scale (0–10) when asking the respondent to place themselves on a left-right scale, the WVS and EVS, however, use 10 point scales (1–10). To enable comparison we merged the 0 and 1 categories from the 11 point-scale into a single category, effectively creating a 10 point scale. Table c in the on-line appendix presents the results. The substantive results are similar to the ones presented in the paper, except the effect of Mixed Member electoral systems loses its statistical significance (although the sign of the coefficient remains the same). More importantly, the effect of selectorate and decentralization is similar to the ones presented in the paper. Interestingly, we find that more right wing respondents exhibit higher levels of satisfaction. These results corroborate previous findings in the literature (Schäfer 2013; Anderson and Singer 2008). We further note that some researchers find a positive relationship between ideological congruence between voters and parties and voter's satisfaction levels (Myunghee 2009) but we simply do not have either object or subjective measures of parties' ideological position data for all the parties in our study.

The models presented are OLS models. We also ran Hierarchical ordered logit models, and obtained similar results, substantively, to the ones presented here. For ease of interpretation we present the Hierarchical Linear models. The ordered logit models are available in the appendix.

The standard deviation of satisfaction with democracy is 0.75.

We re-ran the analysis controlling for Gallagher Index of disproportionality. While the direction of the effect was negative, indicating that indeed voters in disproportional systems exhibit lower levels of satisfaction, the effect fails to reach statistical significance at a conventional level. Moreover, while the negative sign of the SMD coefficient remained, it, too, failed to reach statistical significance. Corruption levels have been hypothesized to affect political attitudes towards democracy (Van der Meer 2010). The coefficient of political corruption measured by Transparency International's Corruption Perception Index is positive, indicating that in cleaner countries citizens enjoy higher levels of satisfaction, but the results are insignificant in a conventional level. All other results in the model remained the same to the ones presented above. Lastly, regime type has been hypothesized to affect satisfaction with democracy (Huang et al. 2008). We ran the models while controlling for whether a country is a presidential or a parliamentary system. The regime type's coefficient was not statistically significant, while all other results remained similar to the ones presented in Table 1. All models can be found in the appendix.

Ideally we would have liked to verify the results using sub-sample analyses that ascertain the aggregate findings are not the product of a differing distribution of satisfaction with democracy and selection processes across old and new democracies. Unfortunately, only a small segment of our sample constitute new democratic countries (8), which result in perfect multi-collinearity among many of the covariates. Nonetheless, we re-ran the analysis excluding countries for which data was available during their transitioning to democracy period. Thus, individuals and parties were excluded if their data came from an election cycle which occurred during the first 10 years since democratization/independence. This effectively excluded Croatia, The Czech Republic (both 1996 and 1998), Hungary, Poland, Russia and Taiwan. The result are similar to the ones presented in the paper, and support the same substantive conclusion. Future research that extends the analysis to additional new-democratic countries will help shed light on this issue in the future.

We verify party switching does not cause the association we find to be spurious, by adding the covariate intra-legislative party switching to the Israeli analysis. The results (presented in Table e in the on-line appendix) demonstrate that party switching is not correlated with satisfaction with democracy and the substantive results concerning the positive relationships between selectorate and satisfaction remains.

For the Israeli models we could not include treatment contrast for both selectorate and decentralization in light of perfect multicollinearity and singularity.

We of course stress that these examples are mere illustrations of the theory and not an official test of it.

To further continue exploring whether candidate selection processes affect citizens' political attitudes and behaviors we are going to extend the analysis and examine other outcome variables. To begin with, we will focus our attention on institutional trust variables, such as citizens' trust in parties, the parliament and the government. We also want to examine whether and how democratic intra-party candidate selection processes affect citizens' interest levels in politics, their tendency to talk about political events with their relatives and friends, and their levels of political efficacy.