This study analyses how different institutional settings influence individual and cross-national differences in satisfaction with democracy (SWD). The dominant theoretical paradigm is that pluralistic or consensual features of representative democracies should foster higher SWD among the citizenry. As Lijphart (Reference Lijphart2012) argues, all the features that characterize consensual democracies which seek to maximize representation and the plurality of decision-making majorities should tend to produce more positive citizen evaluations of their functioning. However, more recent analyses find no empirical evidence of a positive effect of consensual democracies on SWD (Bernauer and Vatter Reference Blais2011), prompting some to argue that the “difference between majoritarian and consensus institutions is not particularly important for popular perceptions of a regime” (Peffley and Rohrschneider Reference Peffley and Rohrschneider2014, 16).
So, do institutions that promote consensualism have any positive impact on SWD? As this study will show, strong linkages exist but their relationships do not always work in the same expected positive direction. This article discusses and shows the consistency of these apparently contradictory results through a comprehensive cross-regional analysis, and also provides individual-level evidence for the logic behind aggregate-level results. Our empirical results support the view that countries with greater electoral proportionality tend to have higher levels of SWD, while at the same time party/government system fractionalization is associated with lower SWD. Our analysis further suggests that people are capable of valuing both good representation and also a concentrated government system where parties can be held accountable—a combination of electoral outcomes that has been described by some as an electoral “sweet spot” (Carey and Hix Reference Cheibub, Gandhi and Vreeland2011).
In the first part of this paper, we test these essentially aggregate-level arguments by relying on a time-series cross-sectional (TSCS) panel data set covering 58 democracies, 300 elections and 833 country years between 1990 and 2012 based on aggregate survey information. This is in stark contrast to the body of research whose analysis usually hinges on many fewer cases with a bias toward established Western democracies. This makes it not only hard to generalize the empirical findings but also difficult to disentangle the often highly collinear variables at the aggregate level. Our panel data set also allows us to conduct a more complex longitudinal analysis of the causes of changes in SWD at the national level.
In the second part of the study, we replicate the analysis at the individual level by employing survey data from the Comparative Study of Electoral Systems (CSES). For this purpose, we merge information from the four existing rounds of the CSES, which provides us with information from 38 democracies, 96 elections and 84,000 voters. Through analysis of this data set, we are able to reconfirm our initial aggregate-level findings. With regard to electoral proportionality, we are able to show that voters whose parties receive a lower seat share than their vote share are more dissatisfied, demonstrating and confirming once more that representational deficits have direct repercussions on SWD at the individual level (Blais, Morin-Chassé and Singh Reference Bormann and Golder2017). But at the same time, we can also observe that these individual effects of the representational deficits get amplified in highly fractionalized government contexts.
ARGUMENTS AND HYPOTHESES
In this study, we analyze how different institutional features are related to individual and cross-national differences in SWD. An influential discussion on this topic was framed by Lijphart (Reference Lijphart2012), who differentiated between consensual and majoritarian types of democracy. According to Lijphart, consensual democracies seek to maximize decision-making majorities and can be characterized in terms of inclusiveness, bargaining and compromise. Majoritarian democracies, on the other hand, concentrate political power and can be described as exclusive, competitive and adversarial. This divide is especially relevant to those features that belong to the “executive-parties dimension” of Lijphart’s classification: electoral proportionality, party system fragmentation and a concentration of executive power. These variables are expected to be interconnected and are therefore considered to have similar effects on SWD.
In our study, we unbundle the effect of this set of “institutions” and argue that electoral proportionality and party/government fractionalization are distinct outcomes of the electoral rules (Lijphart Reference Lijphart1994; Taagepera Reference Taagepera2003) that have different effects on SWD. In important aspects, this argument parallels a puzzle posed to research on electoral turnout. While proportional representation (PR)-systems and electoral proportionality are positively associated with turnout rates empirically, the presence of a higher number of parties appears to decrease participation in elections (Jackman Reference Jackman1987; Blais and Dobrzynska Reference Blais and Aarts1998; Blais Reference Blais and Dobrzynska2006; Blais and Aarts Reference Blais, Morin-Chassé and Singh2006). The positive effects of PR-systems are attributed to their ability to mobilize and to provide a more effective representation of small parties and of minority groups. Voting itself could also be more “satisfying” because fewer votes are wasted (Karp and Banducci Reference Karp and Banducci2008, 330). Yet, as Jackman (Reference Jackman1987) has pointed out, multiparty systems also tend to produce coalition governments. These in turn endanger the decisiveness of elections since electoral outcomes no longer determine the final composition of governments.
We argue here that from a citizen’s perspective both electoral proportionality and government system fragmentation can have independent and contradictory effects on SWD. Theoretically, a case can be made in favor of a positive effect on SWD for both majoritarian and consensual systems: “If we argue that a consensual system is better for support […], we are using the representation argument […]. If we stress the accountability argument we would be more likely to argue that majoritarian systems would be better since such systems allow us to know whom we can reward or punish for performance in office […]” (Listhaug, Aardal and Ellis Reference Listhaug, Aardal and Ellis2009, 318). For many people, there is no contradiction in valuing both aspects at the same time: fair and pluralistic electoral representation but also concentrated party and government systems where single parties can be held accountable. Citizens may be especially happy with electoral “sweet spots,” characterized by a low-magnitude PR electoral system that tends to produce highly representative governments but limits party and government fractionalization (Carey and Hix Reference Cheibub, Gandhi and Vreeland2011). It is this duality of counteracting consequences of electoral systems that is likely to be responsible for a considerable degree of confusion in the literature on SWD.
Aggregate-Level Hypotheses
Findings with respect to electoral proportionality are far from uniform. There are a number of studies reporting that countries with greater proportionality tend to have higher levels of SWD (Berggren et al. Reference Bernauer and Vatter2004; Anderson, Blais and Bowler Reference Anderson, Blais and Bowler2005; Farrell and Mcallister Reference Farrell and McAllister2006), but studies that analyze the effects of electoral proportionality over time report no relationship between the two variables (Ezrow and Xezonakis Reference Ezrow and Xezonakis2011; Martini and Quaranta Reference Martini and Quaranta2014; Quaranta and Martini Reference Quaranta and Martini2016). This discussion leads us to our two first hypotheses:
Hypothesis 1a: Countries with a more disproportional electoral system tend to have lower levels of SWD.
Hypothesis 1b: Decreasing electoral proportionality leads to decreasing SWD within countries over time.
Somewhat paradoxically, other studies report that countries with majoritarian electoral systems (Karp and Bowler Reference Karp and Bowler2001; Berggren et al. Reference Bernauer and Vatter2004; Farrell and McAllister Reference Farrell and McAllister2006; Aarts and Thomassen Reference Aarts and Thomassen2008; Singh Reference Singh2014), and concentrated party and government systems (Weil Reference Weil1989; Karp and Bowler Reference Karp and Bowler2001; Anderson, Blais and Bowler Reference Anderson, Blais and Bowler2005; Martini and Quaranta Reference Martini and Quaranta2014; Quaranta and Martini Reference Quaranta and Martini2016) tend to have higher levels of SWD. Here, the accountability argument together with clarity of responsibility can serve as plausible theoretical explanations for these findings (Manin, Przeworski and Stokes Reference Manin, Przeworski and Stokes1999). Accountability is only possible if it is clear in citizens’ eyes which party is responsible for policies. Single-party government provides the most clarity, while coalition governments make it more difficult for voters to assign blame and responsibility or to vote incumbents out of office (Powell Reference Powell2000; Lundell Reference Lundell2011). This leads us to our next pair of hypotheses:
Hypothesis 2a: Countries with a more fractionalized government composition tend to have lower levels of SWD.
Hypothesis 2b: Increasing fractionalization of the government composition erodes SWD within countries over time.
Individual- and Cross-Level Hypotheses
There is already some evidence at the individual level for the beneficial/negative effects of representation/under-representation on SWD. First, it has been consistently reported that having voted for parties that won an election substantially increases SWD (Anderson and Tverdova Reference Anderson and Tverdova2001; Anderson, Blais and Bowler Reference Anderson, Blais and Bowler2005; Singh, Lago and Blais Reference Singh, Lago and Blais2011; Curini, Jou and Memoli Reference Curini, Jou and Memoli2012). More recently, Blais, Morin-Chassé and Singh (Reference Bormann and Golder2017) have demonstrated that SWD decreases if the seat share of the party that respondents prefer falls short of its vote share.
On the other hand, losers might be more dissatisfied when the policies implemented do not match their preferences. Following this logic, we can also assume that the positive effect on SWD of voting for the winner can be conditioned by the relative positions of each party in the cabinet. Thus, we further distinguish electoral winners between those who have voted for the party of the Prime Minister (PM) or president—therefore the party that leads the government—and those winners who have voted for the other government parties. Our expectation is that winners who have voted for the party of the PM/president derive much more satisfaction from their electoral victory than those whose party ends up as only a minor coalition partner:
Hypothesis 3: Electoral winners who voted for the party of the PM/president have more SWD than electoral winners who voted for a minor coalition partner and the latter have more SWD than electoral losers.
Additionally, we can utilize these individual findings to test our argument with the help of cross-level hypotheses, since the party and government systems can be expected to have important repercussions on the degree of representation of voters. This argument is inspired by a study by Anderson and Guillory (Reference Anderson and Just1997) that showed that the nature of democratic institutions—whether they are consensual or majoritarian in Lijphart’s terms—should mediate the effects of winning and losing. These same authors argued that electoral winners in majoritarian democracies will be more satisfied since there will be fewer obstacles against the winning parties enacting their policies. By the same token, we expect that fractionalization of the government should condition the effects of winning and losing an election. Although this argument was devised with the consensual–majoritarian dichotomy in mind, we find this cross-level interaction more plausible when applied to the government system since this variable directly captures the extent of power-sharing.
Furthermore, we expect that the modifying effect of fractionalization will not be limited to conditioning the effect of winning and losing, but it should also apply more generally to the quality of electoral representation. Our expectation is that the assumed negative effects of a representation deficit on SWD are amplified in strongly fragmented systems with a multitude of parties and complicated coalition dynamics. Thus, it should be more important in such systems to be adequately represented than in concentrated two-party systems, where the losing side is in any case doomed to opposition:
Hypothesis 4a: Electoral winners, in general, tend to be more dissatisfied with democracy in fragmented government situations.
Hypothesis 4b: The individual negative effects of representational deficits are amplified in more fractionalized government situations.
DATA AND MEASUREMENT
This study only covers countries that fulfill a number of minimal democratic criteria. To approximate these standards, all the countries in our study need to be classified as “Electoral Democracies” and at least as “partly free” by Freedom House. In addition, they must be classified as democracies by Cheibub, Gandhi and Vreeland (Reference Clarke, Dutt and Kornberg2010).
Aggregate TSCS Panel Data set
We are able to retrieve data for 58 countries between 1990 and 2012 that match the above democratic criteria and can thus compile an encompassing TSCS panel data set. This empirical sample exceeds those of previous studies in a number of respects. First, its regional coverage extends to democracies in Europe, North, South and Central America and South-East Asia, and thus overcomes the “Western” democracy bias that is inherent in most SWD studies. Second, it covers 300 election periods and information from 833 country years, with an average of 14.4 observations per country, aggregating public opinion data from about a million respondents. We only include in our analysis those democracies where we can collect information on at least three consecutive elections. This data set not only allows for the first longitudinal analysis on SWD outside Western Europe but also increases our confidence in the cross-sectional results because we are able to compare country means for a longer period of time. Third, there is a clear temporal ordering in our data set. We make sure that SWD is always measured after an election. Thus, in our data set the electoral variables (the causes) precede SWD.
In order to construct the TSCS data set, we rely on opinion data from 13 different sources, most of them international survey programs. Detailed information on the used data sets can be found in Table A in the Online Appendix. We only include representative surveys in our sample that use the same question wording and employ the same four-point scale ranging from not at all satisfied with the way democracy works (1) to very satisfied (4). When aggregating individual survey data, we weight all the data according to the post-stratification, design or demographic weight as necessary. We choose to calculate the percentage satisfied with democracy. SWD at the aggregate level is normally distributed with a grand mean of 50.1 percent and a SD of 19.6.
Explanatory Context-Level Variables
The degree of electoral disproportionality is measured using the Gallagher Index.Footnote 1 As discussed, we expect higher levels of SWD in contexts with more proportional electoral outcomes (and therefore better representation and fewer wasted votes). All the information about the aggregate-level variables is summarized in Table B in the Online Appendix.
Government fractionalization measures the extent to which the executive power is homogeneous.Footnote 2 It reflects the level of party plurality in the composition of the cabinet. It ranges from 0 (every deputy from among the government parties belongs to the same party) to 1 (every deputy from among the government parties belongs to a different party). We expect that countries with single-party and small coalition governments will tend to have higher levels of SWD than countries with heterogeneous coalition governments.
Context-Level Controls
Currently, the most prominent alternative explanation links SWD with the economic outputs of the political system. For crisis-ridden countries such as Portugal, Ireland, Italy, Spain or Greece, the literature mainly attributes the declining levels of SWD to the Great Recession (Armingeon and Guthmann Reference Armingeon and Guthmann2014; Sousa, Magalhães and Amaral Reference Sousa, Magalhães and Amaral2014; Cordero and Simón Reference Cruz, Keefer and Scartascini2016; Quaranta and Martini Reference Quaranta and Martini2017). But also, in general, there exists comparative evidence that economic performance is strongly related with SWD (Halla, Schneider and Wagner Reference Halla, Schneider and Wagner2013; Quaranta and Martini Reference Quaranta and Martini2016). While economic growth might have a positive effect on satisfaction due to benefits from the improving economic situation, the erosion of disposable income might diminish people’s SWD (Clarke, Dutt and Kornberg Reference Cordero and Simón1993, 1000 f.). Closely related, the level of economic development has been shown to be positively associated with SWD as well (Anderson and Tverdova Reference Anderson and Tverdova2003; Schäfer Reference Schäfer2010), while income inequality and poverty appear to depress SWD (Anderson and Singer Reference Anderson and Singer2008; Schäfer Reference Schäfer2010; Lühiste Reference Lühiste2014). In our model, we control for these factors by adding gross domestic product (GDP) per capita and GDP growth rate and, finally, income inequality. Footnote 3
A second category of hypotheses links SWD with various aspects of the quality of governance. A number of comparative studies have shown that corruption, rule of law and effective public administration are strongly related with SWD at the national level (Anderson and Tverdova Reference Anderson and Tverdova2003; Ariely Reference Ariely2013; Dahlberg and Holmberg Reference Dahlberg and Holmberg2014; Peffley and Rohrschneider Reference Peffley and Rohrschneider2014). In our study, we employ a measure that taps into all three dimensions, the Quality of Government Index, provided by the International Country Risk Guide. It is based on the rescaled average of three component variables “Corruption,” “Law and Order” and “Bureaucracy Quality,” where higher values express higher quality of government. Data come from the Quality of Governance Standard Data Set (Teorell et al. Reference Teorell, Dahlberg, Holmberg, Rothstein, Khomenko and Svensson2016).
We control for two important institutional characteristics that might affect the analysis as well: type of government and structure of the state. Type of government is measured as a categorical variable distinguishing between parliamentary, semi-presidential and presidential regimes.Footnote 4 Second, we control for the structure of the state, that is whether there exist independent sub-national tiers of government with certain areas of autonomy which are formally guaranteed, commonly in a written constitution (1) or not (0).Footnote 5
Party fractionalization is measured using the effective number of electoral parties (ENEP).Footnote 6 We include this variable to distinguish the effect of plurality of party supply from government fractionalization. Another potentially relevant control variable is ethnic fractionalization (Alesina et al. Reference Alesina, Devleeschauwer, Easterly, Kurlat and Wacziarg2003) since social diversity can be expected to impact on party fractionalization, probably in combination with the country’s electoral system (Ordeshook and Shvetsova Reference Ordeshook and Shvetsova1994). Finally, democratic elections might enhance people’s feelings about their political institutions and the political process (Banducci and Karp Reference Banducci and Karp2003; Esaiasson Reference Esaiasson2011; Blais, Morin-Chassé and Singh Reference Bormann and Golder2017). We, therefore, include in our model a variable temporal distance to elections, which is the difference between the year of observation and the election year for a given country.Footnote 7
CSES
The second part of our research analyses individual-level data from 96 post-electoral surveys from the CSES. We only include those surveys that cover parliamentary elections for the lower house—although they might have taken place in presidential or semi-presidential systems. For this data set, we merge all four existing waves of the CSES. It covers 38 countries between 1996 and 2013 that match the democratic criteria noted above. The sample includes information from all over the world, although most cases come from Europe. Outside Europe, it covers Australia, Brazil, Canada, Chile, Israel, Mexico, New Zealand, Peru, South Korea, Taiwan, Turkey, the United States and Uruguay. At the individual level, the database includes cross-sectional information on 84,000 respondents.
Explanatory Individual-Level Variables
Winning elections matters to voters when evaluating their democracy and its institutions. Previous studies usually rely on a categorical variable that distinguishes between electoral losers and electoral winners. Yet are all winners alike? Especially in the fragmented party and government systems, voters are often faced with the situation in which they have voted for a party that is part of a coalition but does not lead the government. For this reason, we further distinguish electoral winners between voters who have voted for the party of the PM or president and voters who have voted for another party in government.
We measure representation deficit at the individual level as the difference between the vote shares minus the seat shares of the parties respondents have voted for.Footnote 8 Thus, positive values reflect under-representation while negative values reflect over-representation of the respective party. For example, a value of 5 on the representation deficit indicator implies that the proportion of seats in the legislature is 5 percentage points lower than the proportion of votes for a given party.
Individual-Level Controls
At the individual level, we control for important sociological variables such as age (in years), gender (reference category: male), education level (primary, secondary or tertiary) and household income (constructed as income quintiles for each country). We also include left-right self-placement since there is documented evidence of a relationship with SWD (Anderson and Singer Reference Anderson and Singer2008; Schäfer Reference Schäfer2010; Anderson and Just Reference Anderson and Guillory2013; Lühiste Reference Lühiste2014). Furthermore, we control for respondents’ perceptions of political efficacy or accountability. For this, we rely on two survey items from which we create an additive index (Huang, Chang and Chu Reference Huang, Chang and Chu2008).Footnote 9 Our expectation is that greater political efficacy is associated with greater SWD.
Another factor compounding with the winning effect on SWD is ideological proximity with the ideological content of the policies adopted by the government. To test this possibility, we calculate a measure of respondent’s left-right proximity to the parties in government. Here, we expect ideological congruence with the government to increase SWD (Dahlberg and Holmberg Reference Dahlberg and Holmberg2014). Departing from Curini, Jou and Memoli (Reference Curini, Jou and Memoli2012), we calculate congruence as:
$$ideological\,congruence_{{ij}} {\,\equals\,}\left( {\left| {x_{{ij}} {\,\minus\,}\bar{p}_{j} } \right|} \right){\times}\left( {{\minus}\!1} \right),$$
where x ij is respondent i’s left-right self-placement in country j and p the left-right position of the cabinet. More in detail, p is calculated as the mean position of government parties weighted by the vote share each party has received. Information about the individual-level variables is summarized in Table C in the Online Appendix.
METHOD AND MODEL SPECIFICATION
TSCS Aggregate Panel Model
For the TSCS aggregate panel data set we estimate a three-level multilevel regression where country years (k) are nested within election cycles (i), which in turn are nested within countries (j):
$$y_{{kij}} {\,\equals\,}\beta _{0} {\plus}\beta _{1} t_{{kij}} {\plus}\beta _{2} x_{{kij{\rm }{\plus}}} \beta _{3} x_{{ij}} {\plus}\beta _{4} x_{i} {\plus}\mu _{i} {\plus}\mu _{{ij}} {\plus}e_{{tij}} ,$$
where y kij is the response variable of country j measured at election i, on occasion k. x kij is a time-varying covariate such as GDP growth, while x ij refers to a variable that varies between elections such as ENEP but does not vary within a given election cycle. x j denotes time- and election-cycle invariant covariates such as the type of the executive or the degree of ethnic fractionalization. Finally, t kij refers to a linear time trend variable that captures the measurement occasion.Footnote 10
The above model is also referred to as a random effects (RE) model. It makes the exogeneity assumption that the errors µ j are uncorrelated with the explanatory variables for all time periods. For this reason, it is sometimes argued that a fixed effects (FE) model should be preferred when dealing with time-series data since it allows for a correlation between the residuals and the explanatory variables. However, with a FE model it is impossible to test the effects of time-invariant variables. A similar problem arises in the context of rarely changing variables (Plümper and Troeger Reference Plümper and Troeger2007). In consequence, a FE model makes use of only a part of the variation in a time-varying variable since any higher-level cross-sectional variance is eliminated.
Building on the work of Mundlak (Reference Mundlak1978), Bell and Jones (Reference Berggren, Fugate, Preuhs and Still2015) and Schmidt-Catran and Fairbrother (Reference Schmidt-Catran and Fairbrother2015) solve this problem by simultaneously modeling the cross-sectional and longitudinal relationships by adding a group mean and a de-meaned term together in the model. Fairbrother neatly summarizes the procedure thus: “Separate longitudinal and cross-sectional associations between xtj and y can be identified by calculating the mean of xtj across all relevant years for each country. The coefficient on the country mean j captures the effect on y of enduring cross-national differences in xtj. To capture the effect on y of variation over time within each country, j can then be subtracted from xtj. The resulting longitudinal component xtjM (a country-year level variable) is group mean centred, and is orthogonal to j, such that the two coefficients can be estimated separately” (Reference Fairbrother2014, 124). This leads to the following “within-between” RE model:
$$y_{{tij}} {\,\equals\,}\beta _{0} {\plus}\beta _{1} t_{{kij}} {\plus}\beta _{2} x_{{kijM}} {\plus}\beta _{3} \bar{x}_{j} {\plus}\beta _{4} x_{{ijM}} {\plus}\beta _{5} \bar{x}_{j} {\plus}\beta _{6} x_{j} {\plus}\mu _{i} {\plus}\mu _{{ij}} {\plus}e_{{kij}} ,$$
where the original time-varying variable x
kij
and the election-varying variable xij are included twice in the model, decomposed into
$\bar{x}_{j} $
and x
kijM
and x
kijM
, respectively. A benefit of this approach is that the “within” coefficients will return the same results as in an FE model. Of equal importance is that this approach allows estimation of the cross-sectional association between a time-varying variable x on and y, while it enables us to include time-invariant variables simultaneously in the model.
CSES Individual-Level Model
The model with which we analyze the CSES survey data is similar to the previous one. We again estimate a three-level model, but this time individuals (k) are at the lowest level. The respondents are nested within elections/surveys (i), which in turn are nested within countries (j). At the aggregate level, there also exists some longitudinal variation (less than 3 percent of the total variance) since some countries are repeatedly observed over time. Nevertheless, there are also many countries that are only covered once or twice, and so we choose to disregard this limited longitudinal information and focus on a cross-sectional comparison. For this reason, we only include the group mean component
$\bar{x}_{j} $
of variables x
ij
that vary between elections/surveys, such as ENEP or GDP per capita. This leads us to the following “between” RE model:Footnote
11
$$y_{{tij}} {\,\equals\,}\beta _{0} {\plus}\beta _{1} x_{{kij}} {\plus}\beta _{2} \bar{x}_{j} {\plus}\beta _{3} x_{j} {\plus}\mu _{i} {\plus}\mu _{{ij}} {\plus}e_{{kij}} ,$$
where x
kij
is an individual-level covariate such as gender,
$\bar{x}_{j} $
the cross-sectional term for a time-varying variable x
ij
such as ENEP, and x
j
a time-invariant variable such as the type of executive. Second, we also estimate a number of models with cross-level interactions between individual-level covariates and contextual variables, which take the following form:
$$y_{{tij}} {\,\equals\,}\beta _{0} {\plus}\beta _{1} x_{{kij}} {\plus}\beta _{2} \bar{x}_{j} {\plus}\beta _{3} x_{{kij}} \,{\times}\,\bar{x}_{j} {\plus}\beta _{4} x_{j} {\plus}\mu _{i} {\plus}\mu _{{ij}} {\plus}e_{{kij}} .$$
Estimation
In order to analyze the TSCS aggregate panel data set, the first step is to estimate a null model, which serves as a point of reference (model 1). The second model adds the Gallagher Index and government fractionalization, while model 3 replaces government fractionalization with a measure for party system fractionalization (ENEP), along with all the economic, cultural and institutional control variables. To facilitate interpretation, we standardize all continuous variables before estimating our models so they have a mean of 0 and a SD of 1. We choose not to fit a model which includes the terms for ENEP and government fractionalization jointly since the latter is a direct outcome of the former.
We include both the cross-sectional and the longitudinal terms of the aggregate variables in the models whenever sensible. While government fractionalization and the social and economic covariates vary at the country-year level, the institutional and cultural control variables are time- and election-period invariant. ENEP and the Gallagher index, on the other hand, vary only at the election level but not at the country-year level. For this reason, we estimate their cross-sectional terms by calculating the means of all the election periods in a given country
$\left( {\bar{x}_{j} } \right)$
, weighted by the number of observations for each election so as to not give one election more weight in the estimation. The temporal (within) terms are calculated by subtracting the country means
$\left( {\bar{x}_{{ij}} {\minus}\bar{x}_{j} } \right)$
. Put simply, the differenced terms capture electoral fluctuations around each country’s long-term average.
For the individual models, we rely on the CSES survey data. We again estimate a null model for comparison (model 4). Second, we add all the individual-level covariates (model 5). In model 6 we add all the contextual control variables, the Gallagher Index, and government fractionalization. We choose to only include the controls that we have found significant in the analysis of the TSCS aggregate panel data set. Finally, model 7 adds the cross-level interactions between government fractionalization on the one hand and winning an election and representational deficits on the other.
RESULTS
We begin our discussion by presenting a scatter plot of the ENEP and the Gallagher Index for the 300 elections in our TSCS aggregate data set (Figure 1). Its first purpose is to show that, contrary to common wisdom, a low number of political parties and high electoral proportionality is indeed a somewhat frequent outcome. Second, it illustrates the fact that citizens appear to be most satisfied after elections that produce low party fragmentation but also low disproportionality. Third, the empirical distribution shows that the “sweet spot” hypothesis is not an interactive argument (compare also Tables F and G in the Online Appendix). This is illustrated by the considerable number of cases with high levels of SWD lying close to either of the axes. It appears that reasonably high levels of SWD can still be obtained as long as an electoral system successfully limits either party system fragmentation or electoral disproportionality.
Effective number of electoral parties and Gallagher Index by satisfaction with democracy (SWD)-quartiles (N=300 elections)
Analysis of the TSCS Data set
Table 1 shows the results of the multilevel analysis of the TSCS aggregate panel data set of SWD. The table is divided into four sections. At the top, the “within” coefficients are presented. This is followed by a section with the cross-sectional predictors. Below this is a section with the RE of the models (variance components). As can be seen from model 1, the null model, about 69 percent of the variance can be attributed to the country level, 14 percent to time variation (the election level) and 16 percent to the country-year level (occasions). These figures not only tell us that there is a sizeable amount of variation at every level but also that the largest part of the variation in SWD lies between countries and not within countries over time.
Multilevel (ML) Analysis of Time-Series Cross-Sectional Data Set
Note: ML regression with ML integration; standardized β for continuous variables; standard errors are in parentheses.
ENEP=effective number of electoral parties; GDP=gross domestic product; AIC=Akaike’s information criterion; ICC=intraclass correlation coefficient.
Significance (two-tailed); ***p<0.001, **p<0.01, *p<0.05.
Model 2 includes all the economic, cultural and institutional control variables together with government fractionalization and the Gallagher Index of disproportionality. As hypothesized in Hypothesis 1a, we find that countries with a more disproportional electoral system tend to have lower SWD.Footnote 12 The strength of the cross-sectional effect is considerable, comparable with that of GDP growth. Interestingly, we find no longitudinal effect (Hypothesis 1b). We believe the most plausible explanation for this is that there is very little time-varying information in the Gallagher Index for the great majority of countries in our sample, meaning that the same electoral rules tend to produce similar levels of proportionality within a country over time (see the time trends in the Online Appendix). Nevertheless, the absence of a longitudinal finding does not undermine the cross-sectional results. In the case of the Gallagher Index, one can reasonably expect to find much larger effects when comparing between countries (different electoral systems).
With respect to the fragmentation of the government system, we find that countries with greater fractionalization tend to have lower levels of SWD cross-sectionally (Hypothesis 2a).Footnote 13 This negative effect is substantial: an increase of 1 SD in government fractionalization results in a decrease of 3 percent in SWD.Footnote 14 However, although the longitudinal coefficient points in the same direction, it fails to reach significance (Hypothesis 2b). As for the Gallagher Index, we believe that there might not be sufficient longitudinal information in the measure to find an effect.
There is a surprising finding regarding fragmentation of the party system. Although both the cross-sectional and the longitudinal terms of ENEP are strong and highly significant predictors, they point in the same negative direction. First, we find that countries with greater legislative party fractionalization tend to have lower levels of SWD. Second, increasing party fractionalization leads to decreases in SWD within countries over time. Both effects are substantive, although the cross-sectional predictor appears to be stronger. An increase of 1 SD in ENEP results in a decrease of 4.06 percent in SWD.
The longitudinal effect of ENEP on SWD can also be clearly observed when looking at the time-line plots of these two variables by country (Figure 2). As we can see, many of the time trends of these two variables appear to run parallel and are almost perfectly correlated. At this point, we should also point out that the regression coefficient in our model is likely to be dampened by the fact that in about half of the countries there was little or almost no variation in the number of political parties, despite which we are still able to detect a substantive relationship with SWD. This is a robust but controversial finding and it shows that increasing the plurality of the party supply does, in fact, have negative effects on SWD.
Time evolution of effective number of electoral parties (ENEP) and satisfaction with democracy by country
We also find evidence that democratic elections temporarily cause SWD to increase. This finding is consistent with democratic theory, which posits a link between electoral participation and the legitimacy of the political system. In principle, democratic elections might temporarily enhance people’s feelings about their political institutions and the political process (Banducci and Karp Reference Banducci and Karp2003; Esaiasson Reference Esaiasson2011; Blais, Morin-Chassé and Singh Reference Bormann and Golder2017). Furthermore, our analysis reinforces recent findings, connecting greater levels of quality of governance with higher levels of SWD (Anderson and Tverdova Reference Anderson and Tverdova2003; Ariely Reference Ariely2013; Dahlberg and Holmberg Reference Dahlberg and Holmberg2014; Peffley and Rohrschneider Reference Peffley and Rohrschneider2014). Finally, we find a substantial longitudinal and cross-sectional effect of economic growth on SWD, confirming once more the importance of economic explanations (Clarke, Dutt and Kornberg Reference Cordero and Simón1993; Halla, Schneider and Wagner Reference Halla, Schneider and Wagner2013; Armingeon and Guthmann Reference Armingeon and Guthmann2014; Quaranta and Martini Reference Quaranta and Martini2016). Consistently, we find the level of economic development to be one of the most important predictors to explain lasting differences in SWD between countries.
It is also noteworthy that models 2 and 3 can explain a huge amount of the variation in the dependent variable, especially at the country level but to a lesser extent also at the election level. While 69 percent of the variation in the empty model 1 is due to differences between countries, the intraclass correlation coefficient for the country level decreases to a mere 0.34 in model 2—indicating a huge effect of the independent variables included in the model.
Individual-Level Analysis with the CSES Data
Table 2 shows the results of the analysis of the CSES post-electoral survey data. As model 4 shows, about 82 percent of the variance belongs to the individual level, 14 percent to the country level and only 4 percent to the time-varying election level. This is a very sizeable degree of clustering and underlines the necessity of a method that models these variances distinctively. We again make use of the multilevel toolkit and specify our models as shown in Equations 4 and 5.
Multilevel (ML) Analysis of the Comparative Study of Electoral Systems Data Set
Note: ML regression with ML integration; standardized β for all continuous variables; standard errors are in parentheses.
PM=Prime Minister; GDP=gross domestic product; AIC=Akaike’s information criterion; ICC=intraclass correlation coefficient.
Significance (two-tailed); ***p<0.001, **p<0.01, *p<0.05.
To summarize, we have estimated a series of multilevel models where respondents are clustered within surveys/elections, which are clustered within countries. Furthermore, we have decided to discard any longitudinal information at the aggregate level by including only the cross-sectional terms in our analysis. As we can see from model 5, which includes the individual-level variables, all the variables are highly significant, which is hardly a surprise given the large sample size.
As discussed above, we have two measures of electoral support to test our arguments at the individual level. The first one captures the difference between the vote- and the seat-share of the parties respondents have voted for, labeled as representation deficit. For this variable, we find once more (Blais, Morin-Chassé and Singh Reference Bormann and Golder2017) that voters whose parties are under-represented in the legislature tend to have lower SWD, although the substantive effect is only moderate in comparison. Second, we have further distinguished between electoral winners who have voted for the party of the PM or president and those who have only voted for a minor coalition party. Once more, we find that being an electoral winner is a very strong predictor and its substantive effect is comparable with that of political efficacy (winning an election is a categorical variable and therefore not standardized). However, what is more important for our argument is that we find that voters who have cast their ballot for the party that leads the government have twice as much satisfaction as voters who have only voted for a minor coalition partner (Hypothesis 3). Finally, party identification, again a categorical variable, also has a sizeable effect on individual SWD. Similarly, ideological congruence with the government parties substantially increases respondents’ SWD, although its effect is a little weaker.
In models 6 and 7, we have added only the most relevant aggregate variables to our analysis. In terms of the type of executive, we control for presidential and semi-presidential systems since winning a legislative election might have a different importance in these systems to that in parliamentary systems. However, we find the type of executive does not have a significant effect on SWD. The results reproduce all our previous findings for the TSCS data set. The relative strength of the coefficients on the Gallagher Index and government fractionalization are very similar. These effects are substantial and are highly significant, and when taken together these political variables are even more important than the level of economic development (even though GDP per capita is the most important single predictor). Model 7 adds a cross-level interaction between government fractionalization and the preceding significant individual-level variables that tap individual political representation. As previously discussed (Hypotheses 4a and 4b), differences in the level of fractionalization might diminish the positive effect on SWD of voting for the winner or exaggerate the negative effect of representational deficits. This is what can be seen from model 7.
In order to better grasp the interactive effects, it is more informative to look at the marginal effects plots in Figure 3. These fully confirm the negative conditional effects of government fractionalization. However, they also show that fractionalization does not affect all winners equally. It only reduces satisfaction for those who have voted for the party that leads the government. For electoral winners who have voted for a minor coalition partner, there appears to be no effect. This is an interesting situation but fully compatible with the usually employed power-sharing explanation of the linkage. In fact, the whole argument only makes sense for voters who have voted for the party that leads the government, since larger party and government fractionalization inevitably means coalition government and the sharing of power. On the other hand, if there is a minor coalition partner for whom people could have voted, this already implies some degree of government fractionalization.
Marginal effects plots of cross-level interactions Note: PM=Prime Minister.
CONCLUSIONS
This comparative cross-regional study has provided evidence that countries with higher electoral proportionality tend to have citizens that are more satisfied with the way democracy works. On the other hand, we have also found strong evidence that countries with a highly fractionalized government system tend to exhibit lower levels of SWD than those with concentrated party and government systems. Our longitudinal analysis has given additional support to the notion that increasing party fractionalization causes SWD to decline over time. These findings might seem paradoxical, especially if one takes Lijphart’s dichotomy between majoritarian and consensual democracies as a starting point since we would expect that all these measures should point in the same direction.
We did not finish our analysis by demonstrating aggregate-level relationships but also went on to ask if electoral proportionality and government fractionalization also affect respondents at the individual level. In an analogy to the Gallagher Index at the aggregate level, we have been able to demonstrate that voters whose parties have received fewer seats than their respective vote share are less satisfied with the way democracy works in their country. Second, we have been able to show that the satisfaction voters receive is more than twice as high when they have voted for the government party that leads a coalition as compared to electoral winners who have only voted for a minor coalition partner. Finally, we have found that the positive effect of winning an election on SWD is much diminished in highly fractionalized government systems, while the negative effects of representational deficits are amplified.
It, therefore, seems that people want to be represented adequately and have their votes counted equally and not wasted. However, citizens seem to dislike government fragmentation. This paradox seems to make sense since people might value both at the same time: good representation but also concentrated party and government systems where parties can be held accountable. It is this duality of counteracting consequences of consensual democracies that is likely to have produced the mixed results in the literature since the effects partially cancel each other out when they are not included jointly in a model or when they are combined into a single index (compare Tables F and G in the Online Appendix). For the same reason, we should not be able to detect any substantial relationships between the type of the electoral system or the average district magnitude with SWD, since PR-systems and higher district magnitude are related not only to higher levels of electoral proportionality but also to a more fractionalized party/government system (compare Tables D and E in the Online Appendix).
Nevertheless, what is the mechanism behind the effect of government fragmentation? Is it just due to the resulting lack of accountability—as we have mainly argued—or just the perceived greater instability and inefficiency of such governments? After all, research on government instability has repeatedly shown that the risk of breakup of government increases with the number of parties in government (Taylor and Herman Reference Taylor and Herman1971; Somer-Topcu and Williams Reference Somer-Topcu and Williams2008) and it has profound negative effects on economic outputs such as growth rates (Alesina et al. Reference Alesina, Ozler, Roubini and Swagel1996; Aisen and Veiga Reference Aisen and Veiga2013). These alternative explanations might deserve more detailed attention.
This study also poses some problems and opens new questions. For instance, we found no longitudinal linkages between SWD and the Gallagher Index or our measure of government fractionalization. A partial explanation for this could be that these two variables carry too little time-varying information. Future research should focus squarely on the countries where there are actually sizeable changes—this might be due to electoral reform or a changing party system—and analyze them over a longer period of time.
Another unexpected finding has to do with the relationship between party supply and SWD. Miller and Listhaug (Reference Miller and Listhaug1990) argue that multiparty systems should increase system support in the long term since they provide more choices, handle discontent among the electorate better and increase the possibility of the emergence of new parties that can channel new demands. However, regarding SWD, we have been surprised to observe the opposite effect, not only cross-sectionally but also longitudinally. But why is this the case? How can fragmentation of the party supply decrease SWD? Does too much offer hurt citizen’s perceptions of the party system? Where is the threshold? All these questions deserve further attention in the future; so far what our analysis has shown is that the effects of party system fragmentation are very similar to those of government fractionalization.
Acknowledgements
Pablo Christmann acknowledges support from the Ministerio de Educación, Cultura y Deporte (ES) [grant number FPU12/05193]. Mariano Torcal has received funding from the Institució Catalana de Recerca i Estudis Avançats (ICREA). We thank Ignacio Lago, Sergio Martini and Guillermo Cordero for their valuable and insightful comments on the manuscript.
