Linking Party Preferences and the Composition of Government: A New Standard for Evaluating the Performance of Electoral Democracy André Blais, Université de Montréal Eric Guntermann, Université de Montréal Marc André Bodet, Université Laval Abstract We propose a new standard for evaluating the performance of electoral democracies: the correspondence between citizens’ party preferences and the party composition of governments that are formed after elections. We develop three criteria for assessing such correspondence: the proportion of citizens whose most preferred party is in government, whether the party that is most liked overall is in government, and how much more positively governing parties are rated than non-governing parties. We pay particular attention to the last criterion, which takes into account how each citizen feels about each of the parties as well as the intensity of their preferences. We find that PR systems perform better on the first criterion. Majoritarian systems do better on the other two.

The authors would like to thank the Social Sciences and Humanities Research Council of Canada and the Centre for the Study of Democratic Citizenship for their support. This article will appear in Political Science Research and Methods. 1

In recent years, numerous studies have sought to assess how well the public’s preferences are reflected in government. Such studies have focused on ideological congruence (e.g. Powell, 2000, 2006; Blais and Bodet, 2006; Golder and Lloyd, 2014; Golder and Stramski, 2010) and the representation of policy preferences, variously defined (Hobolt and Klemmemsen, 2005; Hobolt and Klemmensen, 2008; Soroka and Wlezien, 2010; Erikson, MacKuen and Stimson, 2002). These are valuable contributions, but they rely on measures of preferences that, like all others, have limitations. Using ideological self-placement assumes that preferences can be represented along a single dimension, while policy opinions expressed in surveys may not be based on true attitudes (Zaller, 1992). Here we propose another approach to measuring preferences and their representation in government. We show that citizens have preferences for various political parties, which go beyond vote choice, and that we can evaluate the extent to which these preferences are reflected in the composition of governments. The closer the link between citizens’ party preferences and the composition of government is, the better the performance of electoral democracy. We show in this study how this standard can be used to ascertain the relative performance of proportional representation (PR) and majoritarian systems (non-PR). Ever since scholars began studying representation, much attention has focused on the impact of electoral systems on the quality of democratic representation. A number of studies emphasized the relationship between the proportion of votes received by the various parties and their seat share in the legislature (see, especially Rae, 1967; Taagepera and Shugart, 1989), with the purpose of determining how that relationship varies according to factors such as the electoral formula and district magnitude. Unsurprisingly, that research found a stronger correspondence between vote share and seat share in PR systems and more importantly in large district systems. Another stream of research was opened up by Powell (2000, 2006), who, in his groundbreaking work, argued that scholars should focus on the representation of voters’ preferences in government. Democracy is about allowing ordinary citizens to influence indirectly

2

(through their votes) the decisions that are made by their representatives. Such a perspective has two major implications. The first is that when one looks at the consequences of electoral systems, attention should shift from the legislature to the executive, where actual decisions are made in parliamentary systems. The second is that the relevant input is not ballots but rather citizens’ preferences. Powell’s perspective has been extremely influential. There has been an explosion of research, pioneered by Powell himself, on how well governments represent voters, and on how the quality of representation is affected by institutional variables such as the electoral system. Because of the availability of huge cross-national data sets such as the Comparative Study of Electoral Systems (CSES), it is now possible to determine how close governments are to voters on the left-right ideological scale. A number of studies suggest that PR systems produce greater ideological congruence between voters and governments (Budge and McDonald, 2007; Huber and Powell, 1994; McDonald and Budge, 2005; McDonald, Mendes and Budge, 2004; Powell, 2000, 2006; Powell and Vanberg, 2000). Other analyses find no significant difference between PR and non-PR systems (Blais and Bodet, 2006; Golder and Stramski, 2010; Golder and Lloyd, 2014). Powell (2009) explains this “ideological congruence controversy” by pointing to the atypical period covered by recent analyses (1996-2004). Golder and Lloyd (2014), however, using cluster-robust standard errors and cluster-robust bootstrap standard errors, find a significant difference between PR and non-PR systems in only one of the post-war decades and no overall difference during this period. They also find that, while PR systems produce more ideological congruence in legislatures, they produce less congruence in government, controlling for legislative congruence. Finally, Golder and Lloyd show that majoritarian systems are associated with greater variation in legislative ideological congruence but that PR systems have more variability in government congruence. They argue that the coalition-building process implied by proportional systems both decreases ideological congruence and increases its variability. The main limitation of research on ideological congruence is that it is entirely based on

3

the left-right scale as an indicator of citizens’ views. While it is the best available single indicator of citizens’ and parties’ overall ideological orientations, it is less valid in some countries and periods than in others, and it is obviously a very partial measure of people’s preferences, especially in multidimensional electoral markets. Other authors have focused on the representation of citizens’ policy priorities (Hobolt and Klemmemsen, 2005; Hobolt and Klemmensen, 2008), their preferences for spending on different areas of government policy (e.g. Soroka and Wlezien, 2010), and their “mood” for more or less government spending (Erikson, MacKuen and Stimson, 2002). These approaches also make important contributions to our understanding of representation. However, they are generally limited in the number of countries for which they have data, making it difficult to analyze how representation varies across institutions. We are only aware of two such studies that have looked at the opinion-policy link with a sufficient number of cases to be able to assess the impact of electoral systems. Wlezien and Soroka (2012) find that plurality systems have greater policy responsiveness. Kang and Powell (2010), on the other hand, find similar responsiveness in both systems. We propose in this paper a new perspective on how citizens’ preferences are reflected in government. While previous research is about citizens’ ideological or policy preferences, we focus on their party preferences. Voters vote mostly for parties1 and legislative elections in non-presidential systems are ultimately about which party (or parties) will form the government. We make the plausible assumption that voters want the parties they like the most to be in government and the parties they dislike the most to be out of government, and thus, at the aggregate level, democracy works best when governing parties are better liked than non-governing parties. We compare the capacity of PR and majoritarian elections to deliver 1

This is clearly the case in closed list elections. In open list elections, people may also

vote for candidates but their first vote is for the list. In non-list systems, citizens vote simultaneously for a candidate and a party. While some clearly vote for the local candidate (Marsh et al., 2008, chapter 8), most vote first and foremost on the basis of how they feel about the parties. 4

such an outcome. Party preferences are shaped to a large extent by voters’ ideological orientations as well as their opinions about specific policy issues. However, they are influenced by many other factors, in particular voters’ judgments about parties’ competence. In other words, elections are not only about issue positions. They are also about valence (Clarke et al., 2004). Because elections are not only about issues or ideologies, an important criterion for assessing the performance of electoral democracy is whether the parties voters like exercise power in government and the parties that they dislike have little or no power. Our standard of performance is not unrelated to ideological congruence and policy responsiveness. In order for ideological congruence or policy responsiveness to take place, parties citizens like more, which presumably are closer to their own positions, must occupy more seats in government. The difference between the party preference perspective, on the one hand, and the ideological congruence and policy responsiveness approaches on the other, is that the former focuses on how citizens feel about parties rather than on what drives those feelings. Obviously, party preferences are only one measure of preferences and, like other measures, they are imperfect. However, as we show below, citizens do have meaningful party preferences. It does make sense, therefore, to assess the extent to which they are reflected in government. We use party preferences instead of the vote as our reference point because vote choice is affected by strategic considerations that lead some citizens to vote for a party that is not their first choice, because their preferred party is perceived to be non-viable (Cox 1997). A vast literature has documented the presence of a significant amount of strategic voting, in PR as well as in majoritarian elections (Abramson et al., 2010; Alvarez and Nagler, 2000; Blais and Nadeau, 1996; Blais et al., 2001; Gschwend, 2007; Meffert and Gschwend, 2010). According to one study, in a typical election, about a quarter of voters vote for a party that is not their first choice (Blais et al., 2011). Vote choice is thus different from party preferences and from a normative perspective priority should be given to citizens’ preferences.

5

Another reason for preferring party preferences over vote choice is that vote choice does not fully reflect preferences. Voters frequently have second choice parties that they like almost as much as their first choices. People also often strongly dislike some parties. This is particularly important to keep in mind given the literature showing that negative evaluations are often more powerful than positive ratings (see for example, Soroka, 2014). Figure 1 shows the proportion of citizens in the elections covered by this paper who had distinct preferences for parties.2 We can see that the mode, the median, and the mean are for people to give four different ratings (on the 0 to 10 scale) to the various parties. Most people thus have a set of distinct preferences among the various parties, and their vote choice does not fully inform us about these preferences.3 2

6.3% of all respondents answered no party like-dislike questions. They were dropped

from all analyses. 3 If vote choice contained all information about respondents’ preferences, they would have at most two distinct party evaluations, one party they rate highly and one or more parties they rate less positively. Since the vast majority of respondents give three or more evaluations, vote choice is clearly less informative.

6

Figure 1: How many distinct like-dislike scores do respondents give?

Proportion

0.2

0.1

0.0 1

2

3

4

5

6

7

8

9

Number of Ratings

There are two key contrasting perspectives in the literature on what elections are supposed to accomplish (Lijphart, 1999; Powell, 2000). On the one hand, the majoritarian perspective involves concentrating all power in the single party citizens prefer over all others. This vision has generally been seen as entailing that the party that gets a plurality of votes should be in government and the others excluded. On the other hand, the proportional (Powell, 2000) or consensual (Lijphart, 1999) view involves representing as many citizens as possible in the government-formation and policy-making processes. In terms of votes, this means that as many voters as possible should find the party they voted for represented in the legislature and/or cabinet. Both perspectives have been formulated with respect to the translation of votes into legislative and cabinet seats. It is unclear what kind of translation of party preferences into cabinet seats proponents of each perspective would advocate. Nevertheless, we propose two criteria that we think advocates of proportionality and majoritarianism would subscribe to. 7

The first is the proportion of citizens whose most liked party is in government. Proponents of the proportional view would argue that as many citizens as possible should have a representative of the party they vote for in government. With party preferences, the norm would be that as many citizens as possible find their most liked party in government. The majoritarian vision is about giving power to the first choice party of the citizenry as a whole, in other words, to the top aggregate preference. If we only had access to first choice preferences, we might assert that the party that is the most liked party of the largest proportion of voters should be in government. However, we do have access to all preferences, so we can determine which party is the most liked overall, considering all citizens’ feelings about each of the parties, and observe whether that party is in government. Given that we have access to citizens’ sets of preferences for all parties, there is no reason to limit ourselves to criteria related to the first preferences of all citizens (criterion 1) or to the single most preferred party overall (criterion 2). We thus propose a third and richer criterion, which takes into account each citizen’s evaluations of each of the parties as well as the intensity of these preferences, and ascertains how liked the governing parties are, overall, relative to the opposition parties. According to this perspective, electoral democracy works best when the governing parties are better liked than the opposition parties. Why might we expect PR to perform better or worse than majoritarian systems with respect to these various criteria? Non-PR systems tend to reward large and regionallyconcentrated parties in the legislature, while harming small parties with small and dispersed support. PR generally produces a closer reflection of vote choice in the legislature. As a result of greater proportionality, there are more parties represented in the legislature under PR and the government is usually a coalition of two or more parties while in majoritarian systems the outcome is typically the formation of a single-party government (Katz, 1997).4 In short, political systems can perform better or worse on our criteria because of disproportionality 4

In our data set, 82% of the cabinets formed after PR elections were coalitions; none of

the cabinets formed after non-PR elections were coalitions.

8

itself or because of the inclusion of more parties in government. We expect PR to perform better with respect to criterion 1 (H1). In a majoritarian system, the party that gets to form the government typically gets a plurality but not a majority of the vote (Blais and Carty, 1987, 1988). Furthermore, some of its support is likely to come from voters who prefer small parties but do not vote for them because they are not viable. PR elections, on the other hand, produce multi-party coalitions that are more likely to include a wider array of voters’ first preferences. Thus, a greater proportion of citizens should find that their most liked party is in government in PR systems, because more parties are in government. We expect majoritarian systems to perform better than PR systems with respect to the second criterion, according to which the most preferred party overall should be in government (H2). Since PR systems typically have multi-party coalitions and the formation of these coalitions is somewhat detached from electoral results, there is a higher probability that the coalition government will not include the most liked party. The more parties there are in government, the more likely a coalition of parties successfully manages to exclude the overall most liked party. Furthermore, in majoritarian systems, the most liked party, which generally is a large party, benefits from disproportionality, and we should thus expect non-PR to increase the chances of that party being in government. However, majoritarian systems, especially first past the post (FPTP) systems, sometimes produce anomalous outcomes, whereby the party with the most votes does not win a plurality of the seats and thus does not form the government (Shugart, 2008). Nevertheless, we still predict that, all in all, the goal of having the most preferred party in government is met more often in majoritarian systems. What about our third and most important criterion, that is, ensuring that the governing parties, overall, will be better liked than the non-governing parties? We expect majoritarian systems to perform better on this front (H3). The basic reason is the presence of coalition governments under PR. This has one good consequence, as it increases the number of citizens

9

whose most preferred party is in government. It also has an important negative effect, as it means the inclusion in the coalition government of parties that are not much liked by many people. We thus expect that the more parties there are in government, the less liked governing parties are compared to non-governing parties. Table 1 summarizes our expectations of how the electoral system and the number of parties in government affect performance on each of our criteria. Table 1: Expectations Criterion 1

Criterion 2

Criterion 3

PR (vs. non-PR)

+

-

-

number of parties

+

-

-

1 The Data We use the four modules of the Comparative Study of Electoral Systems that are currently available. Each of these contains a party like/dislike question in which respondents are asked to rate each party on a 0 to 10 scale where “0 means that you strongly dislike that party and 10 means that you strongly like that party”. This is a straightforward question that taps how much voters like or dislike each of the parties and is thus our indicator of citizens’ party preferences. This is in line with prior research on strategic voting that uses party feeling questions as indicators of sincere preferences (Blais et al., 2001; Abramson et al., 2010; Blais et al., 2011).5 CSES collaborators are asked to request evaluations of the six most relevant parties, or even more if necessary. Of the 87 election studies included in our analysis, 29 have questions about six parties, 38 cover more than six (12 have seven parties, 15 have eight, and 11 have 5

We take into consideration all preferences respondents expressed for parties, regardless

of whether they voted or of how many parties they actually evaluated. This is another advantage of focusing on party preferences rather than votes. 10

nine), and 20 include less than six. As in previous studies (e.g. Golder and Stramski, 2010), we focus on lower house elections in non-presidential democracies. We adopt the criterion used by Cheibub, Gandhi and Vreeland (2010) to distinguish between presidential and non-presidential systems: nonpresidential regimes are those in which the government is responsible to the legislative assembly. The only exceptions to our use of this standard are cases where the president plays such a predominant role in cabinet formation that legislative elections have essentially no influence on government composition (see below). Our sample includes 87 legislative elections held in 35 countries. Seventy-four of the elections were held under PR and 10 were under non-PR. The non-PR elections were in Australia, Canada, France, and Great Britain. There are also three elections held under mixed systems in Japan and Thailand (see Online Appendix for a list of elections).6 We include all lower house election studies in non-presidential democracies included in CSES modules 1, 2, and 3, as well as those available in module 4 as of April 2015, except the 1998 Ukrainian election and the elections in Taiwan, since these were held under the presidential-parliamentary variant of semi-presidentialism, in which presidents have more control over the cabinet and legislatures have less power than in the other variant, premierpresidentialism (Shugart, 2005). Austria is also considered presidential-parliamentary. In that country, however, parties effectively control the recruitment of presidents, and power is shared between two parties, all of which limit the power of the president (Müller, 1999, cited in Shugart 2005). Therefore, we have decided to retain the 2008 and 2013 elections in Austria in spite of the formal constitutional powers of the president. In all the countries that we have included, a new cabinet was formed within a few months of the legislative election.7 6

Mixed systems that are compensatory are classified as PR. Mixed parallel systems (Japan

and Thailand) are dropped from all analyses in which we simply compare PR and non-PR systems, since they are partly majoritarian and partly proportional. 7 We only included studies conducted after legislative elections. The 2002 and 2012 election studies in France, for example, were held after the presidential election. They were thus

11

For each party included in the CSES election studies, we have the distribution of ratings it gets from the electorate, on the one hand, and the proportion of seats it had in the cabinet that was formed after the election, on the other.8 With that information, we are able to compare the performance of majoritarian and PR systems with respect to the three criteria outlined above. Note that we only consider one government per election.9 We perform the following analyses. We first simply compare PR and non-PR systems. We then examine the relationship between various measures of performance and three indicators of proportionality: a simple dummy variable that equals 1 for elections conducted under PR and 0 for majoritarian elections10 , a disproportionality index (Gallagher, 1991), and mean district magnitude. Mean district magnitude was calculated by dividing the total number of seats by the total number of districts.11 The PR dummy variable is the simplest and most direct, but also the crudest indicator. The Gallagher index has the advantage of being an encompassing measure of the degree to which the distribution of seats in the legislature departs from the distribution of votes but the disadvantage of being affected by decisions made by the parties as well as by the rules themselves (Cox, 1997, Chapter 9). Mean district magnitude (see, for example Coppedge, 1997; Clark and Golder, 2006) has the advantage of being a purely institutional indicator but the disadvantage of tapping only one element excluded. 8 Data on cabinets are from the European Journal of Political Research with the following exceptions: data for Albania 2005, Bulgaria 2001, Croatia 2007, Romania 1996, Romania 2004, Thailand 2001, Turkey 2011 are from the Economist Intelligence Unit. Data for Slovenia 1996 are from http://www.vlada.si/en/about_the_government/governments_of_ the_republic_of_slovenia/government_of_the_republic_of_slovenia_1997_2000/ 9 That is, if more than one government was formed before the next election, we only retain the first. 10 As mentioned above, mixed parallel systems (Japan and Thailand) are dropped in the analyses based on the PR dummy variable. 11 Seats and districts are from all electoral segments. For example, for each election in Germany, we added the total number of seats elected in districts to that elected by lists and then divided this sum by the total number of districts in both segments.

12

(district magnitude) of the electoral system, leaving aside other aspects such as the presence of legal thresholds. We thus have three explanatory variables: a dummy for whether the election is held under a PR or a non-PR system, the degree of disproportionality, and mean district magnitude. In the latter two cases, as is usual in the literature, we use a logarithmic transformation of the variables. In order to assess the influence of the mediator variable mentioned above, we include a variable corresponding to the number of parties in cabinet.12 We expect the differences in the performance of PR and majoritarian systems to be explained in good part by the fact that the former usually lead to the formation of multi-party coalition governments. We add two control variables that may also affect the quality of representation. First, we include the level of economic development, measured by the GDP per capita of the country the year each election was held, in current U.S. dollars, from the World Bank (2014). Second, we control for the overall level of democratic development, measured by the average of the Political rights and Civil liberties scores, published by Freedom House (2014) in Freedom in the World, for each country the year the election was held. To analyze the relationship between our three indicators of proportionality and our three criteria of performance, we use Bayesian methods.13 Conventional frequentist methods assess how often a particular result would be at least as distant as an estimate from the value posited by a “null” hypothesis if we were to repeat a sampling procedure a large number of times. In order for frequentist hypothesis tests to make sense, we must assume that we can repeatedly sample from an underlying population. Bayesian approaches, on the other hand, give us 12

Readers should note that we considered Liberal-National coalition governments in Aus-

tralia single-party governments, since their alliance has lasted such a along time. 13 All models were run in JAGS using two chains of 100,000 iterations, a burnin of 50,000 and a thinning rate of 10. All models and convergence diagnostics are in the Online Appendix. We also performed non-Bayesian estimations (logit or OLS), and the findings are quite similar. See the Online Appendix. Of particular note, all findings for which we find a high level of confidence in our Bayesian analyses are significant in the non-Bayesian results.

13

the plausibility of a particular hypothesis given our particular sample of data. There is no focus on how often a particular result is to occur if we take samples a large number of times. This Bayesian approach is more reasonable when the assumption of replicability is unrealistic, as typically occurs in the social sciences (see, for example, Gill, 2008; Jackman, 2009). Given that our units of analysis are elections, it would be unrealistic to claim that repeated sampling is possible.14 Inference in Bayesian analysis does not involve confidence intervals or significance tests. Rather than using cutoffs, like 0.05, for null-hypothesis significance tests, with Bayesian methods we interpret the results of an analysis by describing the posterior distributions of the parameters of interest. This can be done by stating the proportion of a parameter’s posterior distribution that is in a given range, for example, greater than or less than 0. This is essentially how confident we can be that the coefficient is positive or negative.

2 Findings 2.1

Criteria 1: How many citizens have their preferred party in government?

We would hope that as many people as possible find that their preferred party is in government rather than in the opposition. As a matter of fact, on average, 49% of citizens are in such a fortunate situation in the 87 elections covered by our study. The proportion ranges from a low of 28% in Norway (1997) to a high of 70% in Finland (2011).15 14

We do not run hierarchical models (with countries as a second level), because our number

of elections is small. Other studies (e.g. Lupu, 2014) have also used single-level models with the same data. 15 Ties in individuals’ first preferences were randomly broken. The results are similar if ties are removed from the analyses (see Online Appendix). The analyses in this section include elections in both parliamentary and semi-presidential systems. Results are similar when analyses are limited to parliamentary systems (see Online Appendix). 14

Do PR systems do better on this criterion? The answer is yes. On average, 50% of citizens get their most liked party in government under PR, compared to 43% in non-PR elections (see Table 2 for main results). The linear regression results in Table 3 show that everything else being equal the percentage of citizens whose most liked party is in government is 7 points higher under PR.16 Likewise, that percentage is higher as the disproportionality index gets lower and as mean district magnitude gets higher, in other words, in more proportional systems.17 However, the level of confidence that we have in this conclusion is only high for the PR dummy, where it is 87%. This relationship is due to the number of parties in government. Table 4 shows the results of a model including this variable and shows that the coefficient on the PR dummy drops considerably and the coefficient on the number of parties in government is positive and the level of confidence in it being positive is very high (nearly 100%).18 As expected, then, more people have their preferred party in government under PR, and this is due to the fact that PR leads to the presence of more parties in cabinet. On this first criterion, PR elections perform somewhat better. 16

Note that we used linear models here even though the dependent variable is a proportion

since more complex models (such as Tobit) have been found to give similar results (e.g. Wooldridge, 2009, 589). Linear models are easier to interpret and have been used in other studies with dependent variables that only take on positive values (e.g. Gerring et al., 2015; Avdeenko and Gilligan, 2015). 17 We always expect a different sign with the Gallagher index than with the proportional dummy and mean district magnitude (MDM). 18 Note that our results are consistent with a mediating role for the number of parties and with common practice for establishing mediation. However, we do recognize that more complex approaches could make a clearer case for mediation. (see Imai, Keele and Tingley, 2010)

15

Table 2: Main Results

Proportion most liked in cabinet (%) Most liked party overall in cabinet Evaluation of governing vs. opposition parties N

Non-PR

PR

Overall

42.70

50.20

49.50

100.00

85.10

87.40

1.03

0.97

0.99

10

74

87

Table 3: Criterion 1. Proportion with Most Liked Party in Cabinet Model 1 Mean SD Intercept 0.39 Proportional

0.07

Model 2

Model 3

P

Mean

SD

P

Mean

SD

P

0.18

1.00

0.45

0.12

1.00

0.42

0.17

1.00

0.09

0.87 -0.01

0.04

0.65 0.01

0.07

0.73

Gallagher* MDM* Freedom House

0.03

0.06

0.73

0.04

0.07

0.89

0.04

0.11

0.89

GDP per capita

0.00

0.00

0.56

0.00

0.00

0.52

0.00

0.00

0.56

N 84

87 *We used the log of these variables

16

87

Table 4: Criterion 1. Proportion with Most Liked Party in Cabinet (with number of parties) Model 1 Mean

SD

P

0.13

1.00

-0.01

0.10

0.53

Number of parties

0.05

0.07

1.00

Freedom House

0.01

0.07

0.57

GDP per capita

-0.00

0.00

0.57

Intercept 0.37 Proportional

N 84 *We used the log of these variables

2.2

Criterion 2: Is the most liked party in government?

Perhaps the simplest question of all is whether voters manage to get the party that they like the most overall in government. For this, we compute the mean ratings of all the parties in each election. We identify the party that is most liked overall, and we determine whether that party is in government or not. All in all, the most liked party was in government in 76 of the 87 elections, that is, 87% of the time. Is the criterion better achieved in majoritarian systems? The answer is yes. The most liked party was in government after each of the 10 non-PR elections, while that was not always the case following PR elections. In fact, in 11 PR elections out of 74 (15%), the most liked party found itself in the opposition. That was the case in the following elections: Denmark (2007), Estonia (2011), Israel (1996), the Netherlands (2006), the Netherlands (2010), New Zealand (1996), Norway (1997), Portugal (2002), Slovakia (2010), Spain (1996), and Sweden (2006). Table 5 presents the results of probit regressions of a dichotomous variable indicating whether the most liked party was in cabinet or not on each of our explanatory variables, with

17

the exception of the proportional dummy, which was dropped because it perfectly predicts the outcome19 , as well as the two control variables. The table shows that the most liked party was more likely to be in government in systems with higher disproportionality and with lower district magnitude. Our analysis leads us to be highly confident (at least 99%) in these differences. As anticipated, the probability of the most liked party being in government is higher under majoritarian elections. This probability increases by 8 percentage points when going from an election with MDM of 1 to one with MDM of 10. Table 6 shows models with the number of parties in government added. We can see that the relationship between proportionality and the probability of the most liked party being in government persists and is thus not due to the number of parties who join governing coalitions but instead appears to be due to the disproportionality of the electoral system itself. In all non-proportional cases, the most liked party has a higher seat share than vote share.20 19

The most liked party was always in cabinet in majoritarian systems. We have also

presented probit and logit models in the online appendix, including the PR dummy, run using the bayesglm function in the arm package, as suggested by Gelman et al. (2008) for complete separation. The coefficients are in the same direction as those presented here, but they are not significant. 20 On average, non-proportional systems give the most liked party a 31 percentage point higher seat share than vote share. The comparable bonus in PR elections is 9 percentage points.

18

Table 5: Criterion 2. Most Liked Party in Cabinet Model 1 Mean SD Intercept -1.63 Gallagher* 0.60

Model 2 P

Mean

SD

P

1.12

0.06

0.15

1.08

0.58

0.27

0.99 -0.45

0.28

1.00

MDM* Freedom House

0.96

0.61

0.96

1.21

0.92

0.98

GDP per capita

0.01

0.00

0.98

0.01

0.01

0.94

N 87 87 *We used the log of these variables Table 6: Criterion 2. Most Liked Party in Cabinet (with number of parties) Model 1 Mean SD Intercept -1.45 Gallagher* 0.61

Model 2 P

Mean

SD

P

1.56

0.11

0.03

1.01

0.52

0.42

0.99 -0.48

0.18

1.00

MDM* Parties in Govt -0.12

0.19

0.75

0.11

0.22

0.68

Freedom House

1.01

0.75

0.97

1.20

0.64

0.98

GDP per capita

0.01

0.01

0.98

0.01

0.00

0.93

N 87 87 *We used the log of these variables

2.3

Criterion 3: Are governing parties better liked than nongoverning parties?

The previous criteria focus on each individual’s single most liked party (criterion 1) or the single party that is most liked overall (criterion 2). The last criterion considers how all voters 19

rate all the parties in government, in comparison with all opposition parties. We developed a measure that captures the utility that each individual derives from having a given set of parties in government and another set of parties in opposition. The greater the overall differential in ratings given to governing parties relative to those given to nongoverning parties, the higher an individual’s net global utility is. We weight each respondent’s like/dislike score for each party according to its share of seats in the legislature, so that small parties count less than big ones. The more positive the difference between weighted mean ratings of governing and non-governing parties, the higher the individual’s net utility. We then calculate the citizens’ overall mean relative utility for each election.21 All in all, governing parties are better liked than opposition parties in 82 of the 87 elections. The mean rating given to governing parties is 5.2 while the opposition parties get 4.2 on average, which yields a mean net utility of + 1.0 . The mean differential is slightly more positive in non-proportional systems: a gap of 1.03, compared to 0.97 in PR systems. Table 7 shows that elections held under the majoritarian rule, where the result was more disproportional, and with smaller district magnitudes produce governments that are better liked, in comparison with opposition parties. The three coefficients point in the same direction though only in the latter case do we reach a high level of confidence (99%). There thus is some evidence that majoritarian systems perform slightly better on this criterion. The gap predicted by model 3 between the mean ratings of governing and non-governing parties is 0.28 lower when MDM is 10 than when MDM is 1. Table 8 shows the results from a model including mean district magnitude and the number of parties. We can see that the value of the coefficient on mean district magnitude drops to a third of its previous value, while the coefficient on the number of parties in government is negative and has a high level of confidence. Thus, the lower performance of more proportional systems on this criterion is 21

We believe that it is more meaningful to consider relative ratings, that is, the difference

between the ratings obtained by government and opposition parties because we construe voters’ utility to be relative. In this way we also control for the fact that citizens in some countries may be more or less prone to give high or low scores across the board. 20

mostly due to the greater number of parties in government in those systems. Table 7: Criterion 3. How Much More Liked Are Governing Parties Compared to Opposition Parties? Model 1 Mean SD Intercept 0.48 Proportional

-0.11

Model 2

Model 3

P

Mean

SD

P

Mean

SD

P

0.40

0.89

0.31

0.33

0.83

0.66

0.33

0.98

0.27

0.66 0.06

0.16

0.71 -0.12

0.12

0.98

Gallagher* MDM* Freedom House

0.32

0.18

0.96

0.28

0.19

0.98

0.29

0.25

0.98

GDP per capita

0.00

0.00

0.84

0.00

0.00

0.90

0.00

0.00

0.86

N 84

87 *We used the log of these variables

87

Table 8: Criterion 3. How Much More Liked Are Governing Parties Compared to Opposition Parties? (with number of parties) Model 1 Mean

SD

P

0.32

1.00

0.08

0.76

-0.26

0.08

1.00

Freedom House

0.44

0.25

1.00

GDP per capita

0.00

0.00

0.94

Intercept 0.82 MDM* -0.04 Number of parties

N 87 *We used the log of these variables As with criterion 1, the number of parties in government appears to be a key variable determining how well citizens’ party preferences are reflected in government. Figure 2 shows the relationship between differences in evaluations of government and opposition parties, 21

on the one hand, and the number of parties in government, on the other, along with 90 % confidence bounds. The three elections with the highest scores and the three with the lowest scores are labeled. We can see the clear negative relationship between the dependent variable and the number of parties in government. Figure 2: Criterion 3. How Liked Are Governing Parties, Compared to Non-Governing Parties?

Turkey 2011

3

Bulgaria 2001

Ireland 2011 ● ●

Criterion 3 scores

2

PR 1

Mean=0.99

●

● ● ●

Mixed Parallel

●

PR

● ●

● ●

0

Spain 1996

Israel 1996 Slovakia 2010

−1

0

Non−PR

2

4

6

Number of parties in government

3 Discussion We have reviewed three different ways of evaluating how well citizens’ party preferences are reflected in the composition of governments that are formed after legislative elections, and we have tested three hypotheses about the capacity of majoritarian and PR elections to produce good outcomes. The first hypothesis was that more voters would have their most

22

preferred party in government under PR. We have found some evidence that this is the case and that it is due to the creation of coalition governments in proportional systems. The second hypothesis was that the most liked party overall is more often in government in majoritarian elections. That hypothesis is clearly confirmed. This result may not be surprising since it is precisely what majoritarian elections are supposed to accomplish: put in government the party that has the greatest support in the electorate. But it is still interesting to see that the actual outcomes correspond in this case to the conventional assumption. We have also shown that non-proportional systems perform better on this criterion because of the nature of the electoral system and not because of the existence of multi-party coalition governments. But we have argued that the most meaningful criterion when using party preferences to assess representation consists in determining how much more liked governing parties are overall than opposition parties. This criterion has the advantage of taking into account how (strongly) citizens feel about each of the parties instead of considering only each person’s first preferences (first criterion) or only the single party that is most liked overall (second criterion). On this front, we find that majoritarian elections perform somewhat better. Why are governing parties less liked, overall, under PR? The main reason is simply that PR elections usually produce coalition governments. As mentioned above, the governments that were formed after PR elections were made up of at least two parties 81% of the time, while all the majoritarian elections included in this study led to single-party governments. This raises the question of why governing parties are less liked compared to opposition parties when there are coalition governments. The main reason is simply that larger parties tend to be better liked overall than smaller parties and that coalition governments are more likely to include smaller, less liked parties. The mean evaluations of the governing party in a single-party government and of the largest party in a coalition government are nearly identical: 5.48 and 5.41, respectively. The average rating of coalition parties with less than 20% of seats in cabinet is 3.87, not much higher than the mean rating of parties that end up

23

in the opposition under coalition governments, 3.80. The process of coalition-making entails the inclusion of small parties in government, and many of these small parties are not liked by many voters. Coalition governments, as we have shown, tend to increase the proportion of citizens whose most liked party is in government. However, this entails including in government some parties that are not liked by many other citizens. There is thus a trade-off between satisfying a greater proportion of first choices and satisfying overall preferences.

4 Conclusion There are many ways to evaluate the performance of electoral systems. We have proposed a simple and original standard in this paper: the degree of correspondence between citizens’ party preferences and the party composition of the cabinet. We have argued that this standard makes a lot of sense since elections are very much about which party(ies) will have the opportunity to form the government. Our approach is based on the rather plausible assumption that people prefer to be governed by parties that they like and that they prefer parties that they dislike to be in opposition. From that perspective, a good voting system should maximize the probability that governing parties are better liked, overall, than opposition parties. We have developed three criteria for assessing how well citizens’ party preferences are reflected in the composition of government. The first two are particularly simple and straightforward. They are about the capacity of an election to maximize the proportion of citizens whose preferred party is in government and the probability that the party that is most liked overall is in government. These criteria are interesting but they are also limited. The former only takes into account people’s first preferences while the latter considers only the single party that is preferred in the aggregate. We have argued for a more complete measure, which incorporates citizens’ feelings about all parties as well as their intensity. This criterion is the difference between the ratings given to the governing parties and the ratings obtained by

24

the opposition parties. The greater the difference, the higher voters’ net utility, and the better electoral democracy works. We have shown that, on this crucial criterion, majoritarian systems perform somewhat better than PR, mainly because the latter produces coalitions in which some of the smaller partners are not very much liked by many voters. A key feature of our approach is that it is based on citizens’ ratings of parties. As mentioned above, these data, which are available in most election studies and most importantly in the largest election data set (CSES), have a clear advantage over vote choice. Conceptually, we are interested in citizens’ preferences, and the party like/dislike ratings have greater face validity than vote choice as an indicator of preferences, particularly since vote choice is also likely to incorporate strategic considerations. We have also shown that most citizens have more than two distinct party evaluations, which means that their vote choice does not fully account for how they feel about parties. Furthermore, such ratings provide richer information since they inform us about how much each individual likes or dislikes each party in contrast to vote choice, which only tells us which single party the person supports. Some readers may infer that this study demonstrates that majoritarian elections are superior to PR elections. Such a conclusion is not warranted. The verdict about which is the best electoral system depends on value judgments. The standard that we have explored in this study, the correspondence between citizens’ party preferences and the party composition of governments, is just one of many to consider in the debate over the merits and limits of electoral systems (see Blais (1991)). Moreover, we have shown that electoral systems have different effects on different criteria. While coalitions formed under PR lead to the inclusion of more citizens’ most liked parties in government, these same coalitions allow smaller lessliked parties to enter government, leading to overall lower ratings of governing compared to non-governing parties. Moreover, the proportionality in electoral results under PR makes the inclusion of the overall most liked party less reliable than in majoritarian systems. In short, before concluding that one system is better than the other, we must decide which aspects of citizens’ preferences matter the most.

25

Our claim is rather that the yardstick that we have proposed and examined in this study is a meaningful one that has been neglected in prior research. We hope to have contributed to filling an important gap in the literature and we also hope to spur further work, with additional measures of how well the composition of government reflects citizens’ party preferences.

5 Appendix: Elections in Dataset 5.1

Proportional

Albania 2005, Austria 2008, Austria 2013, Belgium 2003, Bulgaria 2001, Croatia 2007, Czech Republic 1996, Czech Republic 2002, Czech Republic 2006, Czech Republic 2010, Denmark 1998, Denmark 2001, Denmark 2007, Estonia 2011, Finland 2003, Finland 2007, Finland 2011, Germany 1998, Germany 2002, Germany 2005, Germany 2009, Germany 2013, Greece 2009, Greece 2012, Hungary 1998, Hungary 2002, Iceland 1999, Iceland 2003, Iceland 2007, Iceland 2009, Iceland 2013, Ireland 2002, Ireland 2007, Ireland 2011, Israel 1996, Israel 2003, Israel 2006, Italy 2006, Latvia 2010, Montenegro 2012, Netherlands 1998, Netherlands 2002, Netherlands 2006, Netherlands 2010, New Zealand 1996, New Zealand 2002, New Zealand 2008, New Zealand 2011, Norway 1997, Norway 2001, Norway 2005, Norway 2009, Poland 1997, Poland 2001, Poland 2005, Poland 2007, Poland 2011, Portugal 2002, Portugal 2005, Portugal 2009, Romania 1996, Romania 2004, Slovakia 2010, Slovenia 1996, Slovenia 2004, Slovenia 2008, Spain 1996, Spain 2000, Spain 2004, Spain 2008, Sweden 1998, Sweden 2002, Sweden 2006, Turkey 2011.

5.2

Non-Proportional

Australia 1996, Australia 2004, Australia 2007, Australia 2013, Canada 1997, Canada 2004, Canada 2008, France 2007, Great Britain 1997, Great Britain 2005.

5.3

Mixed

Japan 1996, Thailand 2001, Thailand 2006.

26

References Abramson, Paul R., John H. Aldrich, André Blais, Matthew Diamond, Abraham Diskin, Indridi H. Indridason, Daniel J. Lee and Renan Levine. 2010. “Comparing Strategic Voting Under FPTP and PR.” Comparative Political Studies 43(1):43–61. Alvarez, R. Michael and Jonathan Nagler. 2000. “A New Approach for Modelling Strategic Voting in Multiparty Elections.” British Journal of Political Science 30(1):57–75. Avdeenko, Alexandra and Michael J Gilligan. 2015. “International Interventions to Build Social Capital: Evidence from a Field Experiment in Sudan.” American Political Science Review 109(3):427–449. Blais, André. 1991. “The Debate Over Electoral Systems.” International Political Science Review 12(3):239–260. Blais, André and R. Kenneth Carty. 1987. “The Impact of Electoral Formulae on the Creation of Majority Governments.” Electoral Studies 6(3):209–218. Blais, André and Richard K. Carty. 1988. “The Effectiveness of the Plurality Rule.” British Journal of Political Science 18(4):550–553. Blais, André and Richard Nadeau. 1996. “Measuring Strategic Voting: A Two-Step Procedure.” Electoral Studies 15(1):39–52. Blais, André, Richard Nadeau, Elisabeth Gidengil and Neil Nevitte. 2001. “Measuring Strategic Voting in Multiparty Plurality Elections.” Electoral Studies 20(3):343–352. Blais, André and Marc André Bodet. 2006. “Does Proportional Representation Foster Closer Congruence Between Citizens and Policy Makers?” Comparative Political Studies 39(10):1243–1262.

27

Blais, André, Romain Lachat, Airo Hino and Pascal Doray-Demers. 2011. “The Mechanical and Psychological Effects of Electoral Systems: A Quasi-Experimental Study.” Comparative Political Studies 44(12):1599–1621. Budge, Ian and Michael D. McDonald. 2007. “Election and Party System Effects on Policy Representation: Bringing Time into a Comparative Perspective.” Electoral studies 26(1):168–179. Cheibub, José Antonio, Jennifer Gandhi and James Raymond Vreeland. 2010. “Democracy and Dictatorship Revisited.” Public Choice 143(1-2):67–101. Clark, William Roberts and Matt Golder. 2006. “Rehabilitating Duverger’s Theory: Testing the Mechanical and Strategic Modifying Effects of Electoral Laws.” Comparative Political Studies 39(6):679–708. Clarke, Harold D, David Sanders, Marianne C Stewart and Paul Whiteley. 2004. Political Choice in Britain. Oxford, New York: Oxford University Press. Coppedge, Michael. 1997.

“District Magnitude, Economic Performance, and Party-

System Fragmentation in Five Latin American Countries.” Comparative Political Studies 30(2):156–185. Cox, Gary W. 1997. Making Votes Count: Strategic Coordination in the World’s Electoral Systems. Cambridge, UK: Cambridge University Press. Erikson, Robert S., Michael B. MacKuen and James A. Stimson. 2002. The Macro Polity. New York: Cambridge University Press. Freedom House. 2014. “Freedom in the World 2014.” http://www.freedomhouse.org/ sites/default/files/Country%20Status%20and%20Ratings%2C%201973-2013%20% 28FINAL%29_0.xls. January 12, 2014.

28

Gallagher, Michael. 1991. “Proportionality, Disproportionality and Electoral Systems.” Electoral studies 10(1):33–51. Gelman, Andrew, Aleks Jakulin, Maria Grazia Pittau and Yu-Sung Su. 2008. “A Weakly Informative Default Prior Distribution for Logistic and Other Regression Models.” The Annals of Applied Statistics 2(4):1360–1383. Gerring, John, Maxwell Palmer, Jan Teorell and Dominic Zarecki. 2015. “Demography and Democracy: A Global, District-Level Analysis of Electoral Contestation.” American Political Science Review 109(3):574 – 591. Gill, Jeff. 2008. Bayesian Methods: A Social and Behavioral Sciences Approach. 2nd ed. Boca Raton: CRC Press. Golder, Matt and Gabriella Lloyd. 2014. “Re-Evaluating the Relationship between Electoral Rules and Ideological Congruence.” European Journal of Political Research 53(1):200–212. Golder, Matt and Jacek Stramski. 2010. “Ideological Congruence and Electoral Institutions.” American Journal of Political Science 54(1):90–106. Gschwend, Thomas. 2007. “Ticket-Splitting and Strategic Voting under Mixed Electoral Rules: Evidence from Germany.” European Journal of Political Research 46(1):1–23. Hobolt, Sara Binzer and Robert Klemmemsen. 2005. “Responsive Government? Public Opinion and Government Policy Preferences in Britain and Denmark.” Political Studies 53(2):379–402. Hobolt, Sara Binzer and Robert Klemmensen. 2008. “Government Responsiveness and Political Competition in Comparative Perspective.” Comparative Political Studies 41(3):309– 337. Huber, John D. and G. Bingham Powell. 1994. “Congruence between Citizens and Policymakers in Two Visions of Liberal Democracy.” World Politics 46(3):291–326. 29

Imai, Kosuke, Luke Keele and Dustin Tingley. 2010. “A General Approach to Causal Mediation Analysis.” Psychological Methods 15(4):309–334. Jackman, Simon. 2009. Bayesian Analysis for the Social Sciences. Chichester, U.K: John Wiley & Sons. Kang, Shin-Goo and G Bingham Powell. 2010. “Representation and Policy Responsiveness: The Median Voter, Election Rules, and Redistributive Welfare Spending.” The Journal of Politics 72(04):1014–1028. Katz, Richard S. 1997. Democracy and Elections. New York: Oxford University Press. Lijphart, Arend. 1999. Patterns of Democracy: Government Forms and Performance in Thirty-Six Democracies. New Haven; London: Yale University Press. Lupu, Noam. 2014. “Party polarization and mass partisanship: A comparative perspective.” Political Behavior 37(2):331–356. Marsh, Michael, Richard Sinnott, John Garry and Fiachra Kennedy. 2008. The Irish Voter: the Nature of Electoral Competition in the Republic of Ireland. Manchester; New York: Manchester University Press. McDonald, Michael D. and Ian Budge. 2005. Elections, Parties, Democracy: Conferring the Median Mandate. Oxford; New York: Oxford University Press. McDonald, Michael D., Silvia M. Mendes and Ian Budge. 2004. “What Are Elections For? Conferring the Median Mandate.” British Journal of Political Science 34(1):1–26. Meffert, Michael F. and Thomas Gschwend. 2010. “Strategic Coalition Voting: Evidence from Austria.” Electoral Studies 29(3):339–349. Müller, Wolfgang C. 1999. Austria. In Semi-presidentialism in Europe, ed. Robert Elgie. Oxford: Oxford University Press. 30

Powell, G. Bingham. 2000. Elections as Instruments of Democracy: Majoritarian and Proportional Visions. New Haven; London: Yale University Press. Powell, G. Bingham. 2006. “Election Laws and Representative Governments: Beyond Votes and Seats.” British Journal of Political Science 36(2):291. Powell, G Bingham and Georg S. Vanberg. 2000. “Election Laws, Disproportionality and Median Correspondence: Implications for Two Visions of Democracy.” British Journal of Political Science 30(3):383–412. Rae, Douglas. 1967. The Political Consequences of Electoral Laws. New Haven: Yale University Press. Shugart, Matthew S. 2005. “Semi-Presidential Systems: Dual Executive and Mixed Authority Patterns.” 3(3):323–351. Shugart, Matthew S. 2008. Inherent and Contingent Factors in Reform Initiation in Plurality Systems. In To Keep or to Change First Past the Post? The Politics of Electoral Reform, ed. André Blais. Oxford; New York: Oxford University Press pp. 7–60. Soroka, Stuart N. 2014. Negativity in Democratic Politics: Causes and Consequences. Cambrige: Cambridge University Press. Soroka, Stuart N. and Christopher Wlezien. 2010. Degrees of Democracy: Politics, Public Opinion, and Policy. Cambrige: Cambridge University Press. Taagepera, Rein and Matthew S. Shugart. 1989. Seats and Votes: The Effects and Determinants of Electoral Systems. New Haven: Yale University Press. Wlezien, Christopher and Stuart N Soroka. 2012. “Political Institutions and the OpinionPolicy Link.” West European Politics 35(6):1407–1432. Wooldridge, Jeffrey M. 2009. Introductory Econometrics. Mason, OH: South Western Cengage Learning. 31

World Bank. 2014.

“GDP per capita (current US$).” http://data.worldbank.org/

indicator/NY.GDP.PCAP.CD. January 12, 2014. Zaller, John. 1992. The Nature and Origins of Mass Opinion. Cambridge; New York: Cambridge University Press.

32