1. Introduction
Right-wing parties are more likely to reach office in countries with majoritarian electoral rules (Döring and Manow, Reference Döring and Manow2017). This pattern has been attributed to the impact of electoral systems on the behavior of middle-income voters. Contrary to their electoral choice under proportional representation (PR), middle-income voters are more inclined to support right-wing parties under majoritarian systems because they are afraid that a left single-party government would prioritize the preferences of poor voters and pursue a radical redistribution policy (Iversen and Soskice, Reference Iversen and Soskice2006). Despite the popularity of this theoretical proposition, this behavioral micro-mechanism remains empirically untested.Footnote 1
In this research note, we conduct the first empirical test of this argument and examine the role of an attitudinal factor that reinforces the posited effect. Using survey data, we find a consistent relationship between three elements: the electoral behavior of middle-income citizens, the electoral support for right-wing parties, and the type of electoral system. Our results demonstrate that part of the dominance of the right under majoritarian rule is due to middle-income voters' reduced inclination to support left-wing parties in these institutional contexts; our evidence comes from a considerable sample of established democracies between 1996 and 2016 and also appears when leveraging the electoral reform that took place in New Zealand in the mid-1990s. We also show that voters with higher concerns about taxes are particularly inclined to modify their vote when the electoral system changes. Hence, it may be argued that, as attention to redistributive policies increases among middle-income voters, their incentives to join the poor in squeezing the rich intensify under PR.
These findings contribute to three kinds of literature. First, by providing empirical evidence of differential voting behavior among middle-income citizens across electoral systems, they contribute to the political economy literature on electoral systems and democratic representation (Lupu and Pontusson, Reference Lupu and Pontusson2011; Becher, Reference Becher2016). Second, they offer insights into the broader academic debate on whether proportional representation fosters better ideological congruence between representatives and citizens (Blais and Bodet, Reference Blais and Bodet2006; Powell, Reference Powell2009; Golder and Stramski, Reference Golder and Stramski2010; Golder and Lloyd, Reference Golder and Lloyd2014; Ferland, Reference Ferland2016). Finally, the consequences of these findings also speak to the literature on the effect of electoral systems on redistribution (Austen-Smith, Reference Austen-Smith2000) and public expenditure (Milesi-Ferretti et al., Reference Milesi-Ferretti, Perotti and Rostagno2002; Persson et al., Reference Persson, Roland and Tabellini2007).
2. The ideological effects of electoral systems
Electoral systems do not perfectly translate votes into seats. The parliamentary delegations of big parties tend to be larger than the electoral support they obtain, particularly in systems with high representation thresholds (Cox, Reference Cox1997). Previous studies also postulate that the ideology of parties plays a role in overrepresenting them (Grofman et al., Reference Grofman, Koetzle and Brunell1997). Sources for this second distributive effect of electoral systems, sometimes labeled as partisan bias, are multiple and include, among others, malapportionment (Samuels and Snyder, Reference Samuels and Snyder2001), district magnitude variance (Monroe and Rose, Reference Monroe and Rose2002), and the relative proportion of a party's votes which are effective rather than wasted or surplus votes (Rodden, Reference Rodden2019). These explanations require a districted electoral system and a distinct distribution of parties' electoral support across geographical constituencies to operate.
An alternative explanation of the partisan effects of electoral systems is found in the model proposed by Iversen and Soskice (Reference Iversen and Soskice2006). Instead of focusing on the translation of votes into seats, this explanation is behavioral and provides a mechanism that explains vote choice variations across countries: middle-income voters select their ballot strategically depending on the electoral system as they anticipate the redistributive consequences of different government compositions under different electoral rules.
This approach departs from the previous ones in at least four key points. First, it does not require a districted electoral system and only relates to the electoral formula. Partisan effects can emerge in countries with a single national electoral district, providing a more general response as to why some electoral rules overrepresent specific ideologies. Second, it does not present a geographical component. In other words, partisan effects can exist irrespective of the territorial distribution of the vote for each party. Third, Iversen and Soskice explain the partisan effects of electoral systems in terms of the incentives they offer to different income groups to vote for parties with some particular ideology. They do not assume that the translation of votes into seats varies by political party, but rather that the share of votes obtained by each party varies under different electoral systems. Finally, the partisan effects described by Iversen and Soskice are unidirectional and always predict a representative advantage for right-wing parties over left-wing ones in majoritarian countries. Unlike other explanations of partisan effects, they theorize that specific parties will benefit from each electoral system.
Iversen and Soskice's (Reference Iversen and Soskice2006) theory of the partisan effects of electoral systems is based on the assumption that there is a unidimensional political scenario with three equal-sized classes—low (L), middle (M), and high-income (H) groups—and three redistribution policies—(A) no redistribution, (B) moderate redistribution, from H to M and L, and (C) radical redistribution, from H and M to L. The order of preferences for each of these groups is: L (C > B > A), M (B > A > C) and H (A > B > C). Given this order of preferences, the M group should be interested in coalescing with L to pursue a policy that redistributes from H to M and L. However, some electoral systems do not lead it to vote accordingly. PR systems foster the existence of one party representing each group (L [left], M [center], and H [right]) and coalition governments, either between L and M or between M and H. M can support the centrist party expecting a governing coalition with the left that will lead to a moderate redistribution policy. In contrast, majoritarian rules are conducive to the existence of only two parties: LM and MH. M is now split between the two existing parties and cannot avoid that each alternative will pursue its first preference: policy C (radical redistribution) in the case of LM or policy A (no redistribution) in the case of MH. Unable to guarantee its first preference (policy B), and fearing that, in the case of victory, the left will adopt its least preferred option (policy C), M allies with H under majoritarian rules, resulting in higher support for right-wing parties and lower levels of economic redistribution than in PR democracies.Footnote 2
The dominance of right-wing governments under majoritarian rules boils down to a credible commitment problem (Becher, Reference Becher2016).Footnote 3 Left-wing parties struggle to credibly commit to moderate redistribution policies if they govern on their own under a majoritarian electoral system. However, under PR rules the credible commitment problem diminishes as centrist parties gain representation and can influence left-wing parties to avoid engaging in excessive redistribution once in government. Voters anticipate the redistributive consequences associated with different government compositions and adjust their votes accordingly. Majoritarian systems, therefore, align the electoral preferences of middle-income voters with the affluent, while PR systems provoke middle-income citizens ally with the poor. Therefore, middle-income voters are expected to have a lower likelihood of voting for right-wing parties under PR systems than under majoritarian rules. Hypothesis 1 aims to test this implication.
H1: The probability of voting for a right-wing party by middle-income voters is higher under a majoritarian system than under proportional representation.
Fear of a radical redistribution policy is the argument proposed by Iversen and Soskice to explain the distinct vote of the middle-income group under different electoral systems. Although not developed in the original model, this fear of a radical redistribution policy has an observational consequence related to the salience of taxes for vote choice. Issue salience refers to “the degree to which a person is passionately concerned about and personally invested in an attitude” (Krosnick, Reference Krosnick1990, 60), and it has been extensively used as a weight of political preferences in the vote calculus (Downs, Reference Downs1957; Stokes, Reference Stokes1963). If middle-income voters' concern about a radical redistribution policy is the underlying motivation that leads to distinct vote choices under different electoral systems, we should observe that the salience of the taxation issue moderates this effect. Therefore, voters concerned with taxation should be more likely to vote differently depending on the electoral system. In the case of the argument we want to test, we should observe that those middle-income voters who pay more attention to tax issues to decide their vote should be particularly disinclined to support a left-wing party under majoritarian rules. By contrast, this fear should strongly mitigate under PR, producing a corresponding change in voting behavior. Hence, Hypothesis 2 reads as follows.
H2: The higher the salience of taxes in middle-income voters' electoral considerations, the higher the probability that they vote for a right-wing party under a majoritarian system in comparison to proportional representation.Footnote 4
3. Research design
This article tests the micro-level foundations of electoral rules' ideological effects as theorized by Iversen and Soskice, that is, it examines if middle-income voters support more right-wing parties under majoritarian rules due to the difficulties of left-wing parties to commit to centrist redistribution policies under this type of electoral systems. Although some of the findings of Iversen and Soskice have been contested by Höhmann and Tober (Reference Höhmann and Tober2018), the micro-level implications of the article have not been empirically tested.
We follow a twofold empirical strategy to test the argument. First, we use the Integrated Module Dataset of the Comparative Study of Electoral Systems (CSES) and examine evidence from 72 elections in 20 established non-presidential democracies between 1996 and 2016 that employ either a PR or a majoritarian electoral system. The countries included are Australia, Canada, France, and the UK as majoritarian, and Austria, Belgium, Denmark, Finland, Germany, Greece, Iceland, Ireland, Italy, the Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, and Switzerland as proportional.Footnote 5 For a list of the elections considered in our study, please refer to Figure C1 in the Online Appendix.
Second, we focus on New Zealand, which constitutes an ideal case to further corroborate the validity of our argument for two reasons. First, in 1993 the country moved from a first-past-the-post (FPTP) system that had been in place for almost a century to a mixed-member proportional (MMP) system. After this reform, the interparty dimension of New Zealand's electoral rules is typically proportional and, hence, perfectly fits our purposes (Vowles et al., Reference Vowles, Aimer, Banducci and Karp1998). In the context of established democracies, full overhauls of electoral rules at the national level are exceptional (Renwick, Reference Renwick2010). However, when they take place, they create the opportunity for researchers to conduct “crucial experiments”—that is, in Shugart's words (Reference Shugart, Gallagher and Mitchell2005, 34), “case studies in which the effects of specific electoral rules can be isolated from other variables”. In other words, by examining a case that has gone through a process of electoral reform, many factors affecting voting behavior (or other outcomes) aside from the electoral system can be held constant, and, as a result, they can be assumed not to have an impact on the phenomenon of interest. In this scenario, the effects can be interpreted as “causal”.
The New Zealand case has a second advantage related to the extraordinary quality and reach of the New Zealand Election Study. We use the three-wave panel that ran from 1993, the last election under FPTP, through 1996 and 1999, when the first two elections under MMP were held (Vowles et al., Reference Vowles, Aimer, Catt, Miller and Lamare2022a, Reference Vowles, Banducci, Karp, Aimer, Catt, Miller and Denmark2022b, Reference Vowles, Banducci, Karp, Aimer, Miller and Sullivan2022c). The surveys were mainly administered by mail through a self-completed questionnaire. Despite attrition rates, the dataset contains up to 761 individuals who participated in the three studies of interest. The existence of a measure of taxation salience in voters' minds allows us to test Hypothesis 2.Footnote 6
The main dependent variable is dichotomous: voting for a left- or a right-wing party. We classify the parties' ideology according to the family they belong to (Von Beyme, Reference Von Beyme1985; Mair and Mudde, Reference Mair and Mudde1998). Despite some blurring of the differences between party families, parties within them still present important ideological commonalities (De La Cerda and Gunderson, Reference De La Cerda and Gunderson2023), particularly regarding views on redistribution and welfare (Afonso and Rennwald, Reference Afonso, Rennwald, Manow, Palier and Schwander2018, 175). Thus, in the cross-national analysis, the dependent variable takes value 1 if the respondent has voted for a party that belongs to the conservative, Christian-democratic or radical right families, and value 0 if the respondent votes for either a center or a left-wing party.Footnote 7 In the New Zealand analysis, the dependent variable takes value 1 if the respondent votes for the National Party—the only right-wing party present in the three elections under consideration—and 0 if she votes for either a center or a left-wing party.
We also estimate a battery of models that, instead of focusing on vote choice, they use the left-right position of the party voted for in the last general election as dependent variable. More specifically, we use the left-right scale (RILE) from the Comparative Manifesto Project to measure the parties' ideological positions (Lehmann et al., Reference Lehmann, Franzmann, Al-Gaddooa, Burst, Ivanusch, Regel, Riethmüller, Volkens, Weßels and Zehnter2023). Unlike expert classifications, party manifesto data are time-variant measures, which make them particularly suitable for our purposes.
The main independent variables of interest are the electoral system (majoritarian vs. PR) and income groups. In the cross-national analysis, income is measured using the household income variable that classifies respondents into quintiles. We code as middle-income individuals those within the second, third, and fourth quintile groups, whereas those in the bottom and top quintile groups are classified as low- and high-income, respectively. This operationalization guarantees that the median voter is always within the middle-income group, as required by the model of Iversen and Soskice. For the New Zealand analysis, we classify respondents into three equal-sized groups, depending on their income. Middle-income voters are those comprised in the income categories that have approximately the same number of respondents who are richer and poorer than them.
We include a series of individual-level controls in all our analyses to minimize omitted variable bias. In the cross-national analysis, we also include two aggregate-level controls.Footnote 8 On the one hand, the Gini index accounts for income inequality, a factor that could affect the salience of redistribution preferences for vote choice (Rueda and Stegmueller, Reference Rueda and Stegmueller2019). On the other hand, the position of the main left and right-wing parties accounts for a potential correlation between the parties' ideological position and electoral system type. The correlation between party system fragmentation and party system polarization (Sartori, Reference Sartori1976; Bartolini and Mair, Reference Bartolini and Mair1990) is nowadays contested (Norris, Reference Norris2024), but we should not exclude the possibility that the positions of the main left- and right-wing parties vary depending on the electoral system. In particular, the ideology of the main center-left party considers the possibility that the mainstream left could be more moderate under PR rules because the electoral system allows other left parties to take more radical positions. Similarly, the ideology of the main center-right party considers the possibility that the mainstream right takes more extreme positions under PR rules because it may need the support of a radical right party to form or keep the government. These two possibilities could confound the relationship we want to test between income groups and vote orientation. We use the aforementioned RILE index to control for these parties' ideological positions (Lehmann et al., Reference Lehmann, Franzmann, Al-Gaddooa, Burst, Ivanusch, Regel, Riethmüller, Volkens, Weßels and Zehnter2023).
Regarding model specification, dichotomous and continuous response variables are modeled using logistic and OLS regressions, respectively. In the cross-national analyses, standard errors are clustered at the election level and specifications include year-fixed effects. In the New Zealand analyses, all independent variables are measured in the first wave of the panel to avoid potential post-treatment biases and standard errors are clustered at the individual level.Footnote 9
4. Results
In the comparative analysis, we estimate three models to test Hypothesis 1 (Table B1 of the Appendix). In the first specification displayed in Figure 1, we only include the middle-income dummy in interaction with the type of electoral system while controlling for inclusion in the high-income group. Middle-income voters are more likely to support right-wing parties than low-income earners. However, this pattern disappears in the case of PR rules. This effect is robust to the inclusion of contextual controls, such as the ideology of the main center-left party in the country and the level of economic inequality, and individual-level controls, such as the respondents' left-right ideology, gender, age, and educational attainment. Unsurprisingly, belonging to the high-income group and leaning to the right has a positive effect on the probability of voting for right-wing parties. Middle-income voters show a different behavior across electoral systems even when we account for their ideological orientations, and the estimates obtained from the models highlight that the behavioral change observed in majoritarian and PR systems is substantial. As shown in Figure 2a, middle-income voters are approximately 8 percent more likely to vote for a right-wing party in a majoritarian (46 percent) than in a PR democracy (38 percent), even after controlling for contextual factors.
Figure 1. Determinants of voting for the right (cross-national analysis).
Note: Horizontal lines are odds ratios along 95 percent and 90 percent confidence intervals measuring the probability that the respondent votes for a right-wing party. Based on Model 2 of Table B1 of the Appendix.
Figure 2. (a) Probability of voting for the right of the middle-income group across electoral systems. (b) Left-right ideology of the party voted for by the middle-income group across electoral systems.
Note: Horizontal lines are predicted probabilities along 90 percent confidence intervals capturing the propensity that the respondent votes for a right-wing party and the RILE position of the party voted for conditional on the type of electoral system. Based on Models 2 of Tables B1 and B2 of the Appendix.
As mentioned, the previous model does not fully exclude the possibility that the pattern observed results from the different positions of the parties rather than from the different strategic considerations of middle-income voters. We produce two additional analyses to fully account for the effect of party positions. First, we replicate the previous model and control for the ideological position of the main right-wing party (Table B9). The main results hold, which suggests that the differential voting behavior of the middle-income group is not due to differences in the ideological stances of right-wing parties across systems. Second, we estimate a series of OLS models where the dependent variable is the left-right position of the party voted for in the last general election measured as the RILE index (Table B2). Results are displayed in Figure 2b. It shows that under majoritarian systems high- and middle-income voters support parties with similar ideological positions, whereas under PR rules middle-income voters support parties with similar ideological positions than low-income voters.
Figures 3 and 4 (derived from the models displayed in Table C1 of the Appendix) show the analysis before and after the introduction of PR in New Zealand. The results support the argument that middle-income voters adapt their behavior to the expected composition of governments depending on the electoral system. Figure 3 shows the negative effect of the PR reform on the probability of middle-income voters supporting the National Party.Footnote 10 The introduction of PR almost halved the general tendency of the middle-income group to vote for the right relative to that of the low-income group. Middle-income voters are 16 percent less likely to vote for the right under the PR system in place since 1996 than under the FPTP used until 1993 (see Figure 4).Footnote 11
Figure 3. Determinants of voting for the right (New Zealand analysis).
Note: Horizontal lines are odds ratios along 95 percent and 90 percent confidence intervals measuring the probability that the respondent votes for a right-wing party. Based on Table C1 of the Appendix.
Figure 4. Probability of voting for the right of the middle-income group conditional on electoral system type (New Zealand analysis).
Note: Horizontal lines are predicted probabilities along 90 percent confidence intervals measuring the propensity that the respondent votes for a right-wing party conditional on the type of electoral system. Based on Model 1 of Table C1 of the Appendix.
We now focus on the test of the second hypothesis, which argued that the electoral reform would produce a larger swing to the left among middle-income voters who considered taxation an important issue. The evidence displayed in Figure 5 (derived from the models displayed in Table C2 of the Appendix) confirms Hypothesis 2. The 1993 and 1996 samples are divided into two groups according to the salience they give to taxation. Middle-income voters who consider tax rates relatively unimportant show similar voting patterns across electoral systems. In contrast, middle-income voters concerned with tax rates display a lower probability of voting for a right-wing party in 1996 relative to 1993. This difference is robust to the inclusion of a long battery of individual-level controls.Footnote 12
Figure 5. Marginal effects of middle-income group on voting for the right conditional on tax rates salience (New Zealand analysis). Logit models (full results are displayed in Table C2 of the Appendix) capturing the effect (with 90 percent confidence intervals) of belonging to the middle-income group on voting for the right (National) in 1996 (MMP) controlling for the vote (FPP) in 1993 and conditional on tax salience (0 = moderately important, not very important or not at all important; 1 = extremely important or very important). The reference category is low-income group. These models include control variables at the individual level (specially, high-income group).
To evaluate the structural validity of the posited hypotheses, we conduct a set of robustness checks.Footnote 13 First, we analyze whether case selection could change our findings in the comparative analysis. The results are robust to (a) excluding most countries at a time (Figure B1), (b) including French and US presidential elections (Table B12), (c) considering only first-past-the-post systems as majoritarian (Table B8), (d) following the Iversen and Soskice's analysis and excluding Iceland and Switzerland, third-wave democracies, or both at the same time (Tables B13 to B15), and (e) coding Ireland's electoral system as majoritarian (Table B16).
Second, we try alternative measurement strategies of the dependent variable to consider nuances to the general picture that right-wing party families oppose redistribution to the same degree, a consideration that may underestimate the strength of the relationship between income groups and vote orientation. On the one hand, Christian-democrats have redistributive preferences to the left of the liberal and the conservative right (Esping-Andersen, Reference Esping-Andersen1990; van Kersbergen, Reference van Kersbergen1995) that have pushed them to form electoral and government coalitions outside the right-wing bloc (Iversen and Soskice, Reference Iversen and Soskice2015, 202). On the other hand, most radical right parties combine their traditional authoritarian stances with a recent move to more pro-welfare positions, an issue that has gained importance in their agendas (Schumacher and van Kersbergen, Reference Schumacher and van Kersbergen2016; Afonso and Rennwald, Reference Afonso, Rennwald, Manow, Palier and Schwander2018; Chueri, Reference Chueri, Biard, Bernhard and Betz2019). We replicate the models excluding the radical right (Table B5) and Christian-democrats (Table B6) from the right-wing parties. Both analyses show that the main posited relationship holds and is stronger when we focus on those party families with stronger anti-redistribution positions.
Third, we consider an alternative classification of the income groups in the cross-national analysis. The income data from the CSES distinguish between five income groups: from the lowest quintile to the highest quintile of household income. There is no fully satisfactory way of collapsing the five quintile groups into the three equally-sized income groups proposed by the Iversen and Soskice's model. The logic of dividing society into three equally-sized income groups is that no group has a majority so the middle-income group is forced to ally with either the poor or the affluent to reach government. In our cross-national models, we have chosen what we think is the lesser of two evils: merging quintiles from second to fourth—making the middle class a majority in society—and observe how this group votes under majority and PR rules. Table B7 shows that the results do not change if we choose to merge the income quintiles in a way that no group has a majority in society: the poor (first quintile), the middle-income (second and third quintiles) and the affluent (fourth and fifth quintiles).
Fourth, we resort to alternative econometric techniques. We find that the results hold when we use OLS and hierarchical linear models (Tables B3 and B4), and when we use election-fixed effects rather than year-fixed effects (Table B11) for the cross-national analysis. For the New Zealand analysis, results also hold when we use OLS regressions instead of logistic ones (Table C3).
Fifth, the findings from the New Zealand analysis do not change when we (a) consider the new radical right party ACT (Table C4), (b) adopt the same operationalization of the middle-income group as in the comparative analyses (Table C5), (c) include additional control variables (Table C6), and (d) use alternative strategies to measure how salient redistribution for individuals is (Figure C2). The latter reinforces the evidence that supports Hypothesis 2.Footnote 14
Finally, to further examine the mechanisms at play, we conduct two last analyses. In the cross-national part, we specify a model where we add the effective number of electoral parties as a control and main results remain remarkably similar (Table B10). In addition, we follow the ideological preferences and the vote choice of middle-income New Zealanders between 1993 and 1996. We find that their left-right position does not change significantly over waves, suggesting that the electoral reform effect is purely behavioral (Table C7).
5. Conclusions
In this research note, we conduct the first test of the micro-level foundations of electoral rules' ideological effects. We show that middle-income voters change their behavior in response to the anticipated redistributive consequences of the electoral systems. We also provide evidence that middle-income voters concerned with taxation are keener to adjust their behavior depending on the electoral rules than unconcerned voters. Substantively, we establish that a traditional determinant of vote choice, such as income, leads to different behaviors under different electoral rules for the middle-income group. Methodologically, we offer evidence that the effect is robust to a twofold test: a cross-national analysis in 20 established democracies between 1996 and 2016, and a dynamic analysis in New Zealand between 1993 and 1999.
These findings contribute to the literature on the ideological effects of electoral rules. Rather than providing a full model of the partisan effects of electoral systems, we focus on one crucial explanation of this pattern: how middle-income voters anticipate the redistributive consequences of electoral rules through changes in governments' composition and adjust their electoral behavior accordingly. However, this is still relevant to other electoral systems studies that highlight the role of the territorial distribution of political parties' support in interaction with malapportionment, variation in district magnitude or turnout heterogeneity across geographic constituencies to explain the dominance of right-wing parties in majoritarian systems. Analyzing how the territorialization of the electoral support of political parties driven by the history of the countries affects whether they are overrepresented or underrepresented in institutions could be a fruitful way forward.
Our findings may also encourage further research in two other areas. First, although it is not surprising that social position matters for electoral behavior, it is noticeable that the vote choice of middle-income citizens differs across electoral systems. This finding is relevant to scholars who emphasize the role of class to understand electoral outcomes. Second, the implications of our work go far beyond the study of electoral systems and have an impact on a wide range of analyses of democratic representation and economic redistribution. More specifically, studies on ideological congruence may also find this article interesting because it suggests that majoritarian systems are not only worse at representing citizens' political orientations but also introduce a right-wing bias in the adopted policies.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/psrm.2024.18. To obtain replication material for this article https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/9NKW9U&version=DRAFT
Competing interests
None.
Funding
This work was supported by the Ministerio de Ciencia, Innovación y Universidades [PID2020-119465GB-C21], the BBVA Foundation [Leonardo Grant] and the Areces Foundation [Grant for Research in Social Sciences].
