The interplay of economic vote dimensions: inequality, redistribution preferences and support for incumbents

Economic voting is multidimensional (covering valence, patrimony and positional dimensions), and a growing number of research contributions have explored the existence and strength of the effect of these dimensions on voting. However, we know comparatively little about the interplay between these dimensions. This article fills that void by focusing on how the interplay between the rise of income inequality (the valence dimension) and redistribution preferences (the positional dimension) influences support for incumbents. Using the Comparative Study of Electoral Systems (CSES) Modules 4 and 5, I find that preferences for income redistribution reduce the likelihood of voters supporting incumbent parties. Modest evidence demonstrates that this relationship is stronger in countries where inequality increased to a greater degree between elections. Voters who want more redistribution tend to re-elect left-wing governments and want to throw out right-wing incumbents. However, rise of inequality hurts both left-wing and right-wing incumbents.


Introduction
The state of the economy can help us understand voting behavior and explain the election performance of incumbent parties. Scholars have found a strong, positive, consistent effect is produced by the economy on incumbent support (Duch and Stevenson 2008;Lewis-Beck 1988;Lewis-Beck and Stegmaier 2007;Okolikj and Quinlan 2016). Testing the relationship between general economic performance, and voting for incumbents dominates economic voting research.
Nevertheless, the premise that citizens, seeing the economy performing poorly, will tend to throw the rascals out, has been contested, as a normatively appealing but theoretically limiting finding (Hellwig 2012). Another important point of valence economic model criticism is partisan contamination bias, meaning that "inferences about the direction of causality between perceptions of the economy and party support remain questionable" (Anderson 2007;Bailey 2019;Evans and Andersen 2006, p. 194). Others have criticized the unidimensionality of economic voting, saying that researchers only focus on valence issues .
In recent years two alternative strands of economic voting research have developed: patrimony and positional economic voting (Lewis-Beck and Nadeau 2011). Patrimony is the dimension that goes beyond perceptions of the economy, and takes into consideration the assets voters hold and how they predict voting behavior ). Positional economics assume that voters' positions on economic policy issues shape their vote (Lewis-Beck and Nadeau 2011;Lewis-Beck et al. 2013).
A review of the evolving literature 1 reveals that researchers have mainly focused on showing that the effect of patrimony and positional economic voting exists and that it is unrelated to valence. Additionally, the positional economics research strand has focused on demonstrating that people who have preferences for leftist (or rightist) policies are more likely to vote for left (or right) parties as a consequence.
However, we know little about the interplay between these dimensions and whether positional economics can influence support for incumbents. Lewis-Beck et al. (2013) showed that patrimony through the positional dimension affects the vote. But what is the role of the interplay between positional economics and valence on vote choice? Are we in danger of overlooking certain connections and findings by treating these dimensions independently, without examining their relationship? Stokes noted that "it is, of course, true that position-issues lurk behind many valence-issues" (1963, p. 373).
This article fills that void by looking at the interplay between valence and positional dynamics and its impact on the vote. For this, I take income inequality (as the valence dimension) and preferences for redistribution (as a positional issue). Dassonneville and Lewis-Beck (2020) show that inequality functions as a valence condition at both the aggregate and the individual level. Preferences for redistribution are an issue that polarizes voters and one where we observe a significant degree of macroheterogeneity (Quinlan and Okolikj 2018). Thus, I take preferences for redistribution as a positional economic issue that is activated, i.e., made more salient by rising inequality.
Apart from disentangling the relationship between the dimensions of the economic vote, studying income inequality is particularly timely, given its rise worldwide (Milanovic 2012). Previous research has shown the adverse consequences inequality can have. It can negatively influence the state of democracy (Scheve and Stasavage 2017), economic mobility (Piketty and Saez 2014), political and civic participation (Uslaner and Brown 2005), and even the stability of the economy (Stiglitz 2012). Connecting inequality and redistribution preferences to vote decisions is the growing trend (Pontusson and Rueda 2010;Rueda and Stegmueller 2015). This paper contributes to the ongoing debates on this question by providing a comparative analysis.
To address the research question, I use the Comparative Study of Electoral Surveys (CSES) Module 4 (2018) and Module 5 Advance Release 2 (2020). Analysis reveals that the effect of preferences for redistribution on incumbents' support is stronger in countries with greater (and thus more obvious) increases in income inequality. Furthermore, voters who want more redistribution tend to reelect left-wing incumbents and to replace right-wing ones. Rise of inequality hurts both left-wing and right-wing incumbents.
In the next section, I consider the definition of valence and position issues and then focus on the economy. After that, I proceed to discuss the interplay between valence and positional dynamics and defend my choice of valence issue. Subsequently, I discuss the mechanism through which these are connected with voting for incumbents and devise several hypotheses. Then, I present the relevant data and methods to test these hypotheses, followed by the empirical findings. Finally, I discuss the findings and offer potential new research avenues.

Theoretical framework
Defining valence and positional issues Stokes (1963) defines valence issues as "those that merely involve the linking of the parties with some condition that is positively or negatively valued by the electorate" (p. 373). Positional issues are "those that involve advocacy of government actions from a set of alternatives over which a distribution of voter preferences is defined" (Stokes 1963, p. 373).
However, it is not always easy to distinguish valence issues from positional ones. Two potential sources of confusion are (a) the focus of the policy and (b) the changing nature of the issues. Fiorina (1981) argues that the distinction between valence and positional issues depends on the focus of the policy, and whether that focus is primarily on means or ends. This dividing line nicely highlights the nature of the issues. For valence issues, people want similar things, such as less terrorism, a good economy, or a good educational system. The means to achieve those ends can be very different. Making a distinction between these issues can be challenging for voters, as this contrast is often not clear even in party manifestos (Curini and Martelli 2015).
On the other hand, the nature of issues can also change. Change can happen over time or between contexts/countries. Divisive issues in one country can be utterly uncontroversial in others. Moreover, voters converge on positional issues, so these issues over time change from positional into valence ones (Green 2007). The change in the other direction is also possible. Stokes (1963) suggested that this matter should be settled empirically. Valence and positional issues have been studied extensively in terms of the economy. The following section provides an overview of both valence and positional economic issues in prominent empirical studies.

The economy as a valence and a positional issue
In terms of economic voting, valence has been traditionally understood as an evaluation of a country's economic performance. The key assumption is that all citizens want to reap the benefits of a good economy (Tillman 2008). A wealth of literature relies on this assumption and tests whether governments performed well, through the lenses of the state of the economy. 2 Other valence issues include unemployment (Dassonneville and Lewis-Beck 2013); support for the welfare state (Blomqvist and Green-Pedersen 2004); or bailouts in crisis-affected countries (Magalhães 2012). Recently, Dassoneville and Lewis-Beck (2020) showed that income inequality is a new valence issue, a point to which I will return later.
Issue voting studies have a very long tradition in voting literature (Downs 1957;Rabinowitz and Macdonald 1989). Positional economic voting is an emerging strand of economic voting theory. The economy is becoming an increasingly positional issue, and voters have become more policy-oriented (Lewis-Beck and Costa Lobo 2017). Among the positional economic issues, progressive taxation has been most widely tested in empirical models, both at the national level (Fraile and Lewis-Beck 2013;Lewis-Beck et al. 2013;Lewis-Beck and Nadeau 2011) and comparatively (Paparo and Lewis-Beck 2017). Among other issues, in analyzing the Portugal 2011 election, Costa Lobo (2013) uses four items about the role of the state in the economy. Nezi and Katsanidou (2014) analyze how differences in pro-bailout policies affect voters in Greece, showing that positional economic voting gains strength during times of crisis. Talving (2017) confirms that particular finding utilizing fiscal austerity policies as an economic issue, in a comparative analysis at the European level. One of the most comprehensive tests of positional economic voting comes from Quinlan and Okolikj (2018), who used redistribution, welfare and defense spending as positional issues in their global analysis of the strength of positional economic voting. All of these authors find that positional economics, unrelated to valence, do contribute to explaining vote choice.
However, we know little about the interplay between these two dimensions of the economic vote, as previous research has not tackled these dynamics in detail.

The interplay between valence and positional economy dynamics
Following Stokes' notion that position issues lurk behind valence ones (1963), I argue that these two strands should be related to each other in studies of economic voting. Valence can be interconnected with positional economics through several mechanisms. The two under consideration here are that: (a) Valence issues can make citizens prefer one policy option over other(s); and (b) valence can increase the salience of the position issues. To demonstrate these processes, let us turn our attention to the empirical world.
Income inequality is correlated with preferences for redistribution. Voluminous literature on political economy has shown that inequality has a significant effect on actual levels of redistribution (McCarty and Pontusson 2011) as well as support for redistribution (Alt and Iversen 2017). Traditional models of political economy state that preferences for redistribution depend on an individual's income. The seminal work in the field provides the most simplistic version of this logic. In it, Meltzer and Richard (1981) claimed that all individuals who have an above-median income in society want less redistribution, and those who have an income lower than the median want more. Their model implies that there will be more redistribution in unequal societies, because a low median income relative to the society's income mean is a sign of significant inequality (Lind 2005).
The more inequality there is, the more salient the topic of inequality and redistribution becomes (Milanovic and Roemer 2016), and saliency exerts a particular effect on policy responsiveness (Franklin and Wlezien 1997). Lax and Phillips showed that when an issue is not salient it is hard to expect that it will turn into policy (2009). As argued above, income inequality became more salient over time.
Here, I argue that rise of income inequality has become a valence issue, which I defend in the following section.

Rise of income inequality as a valence issue
This section aims to provide evidence that inequality should be treated as a valence issue. Backing for that view lies in (a) previous research findings; (b) citizens' prevailing view that differences in income are too large; and (c) the attitudes of different groups toward income inequality.
First, Dassonneville and Lewis-Beck (2020) find that across time and space, on a sample of 318 elections in 37 countries from 1963 up to 2017, support for incumbents is influenced by economic growth (positive effect) as well as income inequality (negative effect). Furthermore, they find that both left-and right-wing incumbents are held accountable and have worse electoral results when the inequality is on the rise. From this compelling evidence, they conclude that income inequality does not function as a positional issue, but rather as a valence one.
Secondly, the majority of citizens think that differences in income are too large in their country. On a sample of 41 countries that participated in the International Social Survey Programme (ISSP) module on Social Inequality from 2009, we can see some country variation on the issue. However, in all countries, above 50% of the population agree/strongly agree that differences in income are too large in their country (ISSP 2019). Further, on a total sample, a very high percentage, 83.5%, of respondents share this view. While historically, some countries were known and strong advocates of market-oriented policies which generated higher inequality (USA or UK, to name a few), this seems to be slowly changing for some of them. It would not be a first time, nor a first issues which converted from positional to valence. The graph shows the mean value per country on a 5-point scale and illustrates the previous point ( Fig. 1).
Finally, inequality affects citizens from all walks of life. Poor people and the middle class are affected because inequality reduces economic mobility. But richer groups in society are also negatively influenced by the rising income inequality. Namely, Rehm et al. (2012) showed that inequality presents a considerable risk for wealthier citizens too, as economic disadvantage (a key problem of low-income people) and economic insecurity (a significant risk for people with higher incomes) are correlated. Furthermore, Rueda and Stegmueller (2016), analyzing the USA, found that affluent individuals are more supportive of redistribution in states with greater inequality.
I argue that inequality should be treated as a valence issue, but the policies intended to deal with it are positional. In the following section, I develop a model combining inequality, preferences regarding redistribution, and support for incumbents.

Inequality, redistribution and support for incumbents
This section develops a model showing how inequality and redistribution, affecting one another, has the ability to predict vote for incumbents. Here I build on the previous sections and connect the three concepts into a single model.
Extensive literature, with a good proportion of it coming from economic voting studies, clearly shows that valence as an element influences political attitudes and voting (Clarke et al. 2015;Lewis-Beck et al. 2013;Lewis-Beck and Stegmaier 2000). These are issues where parties want to convince voters of their competence (Tillman 2008). Or, as Clarke et al. put it, "in the realm of valence politics, it is not 'what', but rather 'who', and 'how' that matters" (2011, p. 238).
The "who" part of the quote refers to-which party? However, I focus on incumbency, because governments are the actors that are held responsible for adopting and implementing policies (Leon 2011). Thus, I expect incumbents to be rewarded (or alternatively punished) for the policies they (fail to) implement.
When voters differ on the means (the "how") used to achieve a set goal, they differ on policies. Political economy literature shows that the greater the level of inequality is, the more redistribution people want (Lind 2005). Following this, I assume that voters hold governments accountable for fighting inequality and for implementing redistributive policies (Iversen 2011;Lind 2005). Thus, the effect of redistribution preferences should be reflected in government support, and because inequality makes redistribution more salient, I expect the impact of redistribution preferences on incumbent support to be stronger, where income inequality is greater (H1). (Fig. 2).
Not all governments will be evaluated in the same way by voters. Typically left (or right) leaning voters expect the parties they vote for to provide left (right) policies. If they fail to do so, there is often a perception that they should be punished. Here, the mixture of traditional incumbency-oriented economic voting with the partisan theory of economic voting comes into play. Powell and Whitten found evidence that voters actually hold governments to different standards depending on their own ideological preferences (1993).
The cleavage between left and right is easy to recognize in the economy, as most are aware of what the relevant position issues are concerning the economy advocated by the left and by right-wing parties, respectively (Dassonneville and Lewis-Beck 2013). Redistribution is one of those issues, "the core issue of contention between parties of the Left and Right" (Pontusson and Rueda 2010, p. 674). From the extant positional economic voting literature, we know that the different ideological orientations of incumbents mediate the role of citizens' position on specific economic issues. Thus, I expect that voters who want more redistribution will reward and re-elect left-wing incumbents (H2a) and will punish and throw out rightwing governments (H2b) (Fig. 3).
The remaining question is what kind of influence the change in inequality produces in terms of these links?
For right-wing incumbents, the answer seems to be straight-forward. We should expect that (a) people who want more redistribution will tend to punish and replace right-wing incumbents; and (b) that rising inequality will make the impact of preferences for redistribution on vote choice stronger. Hence, I assume that the negative effect of preferences for redistribution on right-wing incumbents will be stronger in countries with a greater rise in income inequality (H3a).
On the other hand, left-wing incumbents are expected to gain support from those who favor redistributive policies. The question then arises as to what kind of effect rising inequality will actually have. Dassoneville and Lewis-Beck (2013) showed that left-wing governments are more severely punished in periods of rising unemployment, meaning that the role of positional economics is dependent on valence. Left-wing parties are typically seen as the owners of this issue and the parties most capable of tackling problems related to levels of unemployment. The same argument can be made for redistribution. Thus, my assumption is that left-wing incumbents will be punished by voters who want more redistribution when inequality is on the rise (H3b) (Fig. 4).

Data and methods
To test the above set assumptions, I use the Comparative Study of Electoral Systems (CSES) Module 4 Full Release (2018) and Module 5 Advance Release 2 (2020). First, I subset the data to all available democracies in the dataset (based on Freedom House Rating, all countries that score 2.5 or lower or a Freedom House scale), since in non-democratic regimes, incumbents tend to perform better than economic conditions warrant. Second, I exclude studies for which data are missing on some crucial variables or for which there are substantive reasons for excluding them. 3 After that, I am left with the 42 studies from 33 countries. 4 Due to unit non-response, there are 35,864 respondents in the main models.
Since I am testing the interplay of economic vote dimensions and the role positional dimension (taking stances on positional economic issues) have on governments' electoral fortunes, I take the same approach as the vast majority of economic voting research, which tests the reward-punishment mechanism. Thus, the dependent variable is support for incumbents, coded 1-voted for the outgoing government; 0-other. This goes in line with previous findings, as literature shows that voters perceive electoral contention in terms of winners and losers and more concretely in or out of the governing role (Anderson et al. 2005;Stiers et al. 2018). Further, Stiers et al. (2018) find evidence that voters perceive the party that won the most votes as clear election winners. These parties, though not always, tend to be the ones that hold the position of Prime minister. Thus, as part of the robustness checks, I also run the same models, with dependent variable coded 1-voted for the party of the Prime minister; 0-other, to account for the effects of inclusion/exclusion of minor coalition partners. Results do not change much with different specification of the dependent variable. 5 3 The following election studies are dropped from the analysis: • From CSES Module 4: Greece (2012) because the government at the time of election was caretaker, and all its members were independents; Latvia (2011) as data on redistribution preferences are missing for that study; Romania (2014) because study relates to presidential election, and incumbents could not be ascertained and linked to the party, since President Basescu technically ran on a Democrat Liberal Party (PDL) ticket in 2009, but Romanian presidents cannot be legally a member of a party and Basescu was also closer to the People's Movement Party (PMP); Peru (2016) as incumbents did not field the candidate for the 2016 elections; Taiwan (2012) because left-right is not relevant in Taiwan (Jou 2010); Switzerland (2011) due to the consociational nature of their government, making it impossible to distinguish between incumbents and opposition. Thus, due to the consociational system, the expectation is that the reward-punishment mechanism will not operate in conventional fashion; and South Korea (2012) because of the missing data on several explanatory variables used in the model. For redistribution preferences, I take the typical survey item, whether respondents agree with the statement, "the government should take measures to reduce differences in income levels." The variable responses range on a 5-point scale from "strongly disagree" to "strongly agree." Assessment of the economy over the previous twelve months prior to the elections is also included as part of the model. Respondents evaluated the economy using the 5-point scale, ranging from "the economy is much worse than it was 12 months ago" to "the economy is significantly better than it was 12 months ago." To address "concerns that partisan and/or ideological bias may pollute economic positions and perceptions" (Quinlan and Okolikj 2018), I include ideology and party identification as part of the model. For ideology, I take respondents' self-placement on an ideological scale from 0 (left) to 10 (right). The variable is first multiplied by − 1 (then + 1 added) if the government is left-wing . Party identification is coded as binary, with 1 representing respondents' identification with the incumbent party; 0 others.
The usual suspects, socio-demographic variables, also form part of the model: gender, age (mean-centered), education and income. A dummy indicator for union membership is added, given the interconnection between membership of unions and voting for left-wing parties, as well as redistribution preferences.
In relation to context, I use the change in inequality (GINI), the difference between the level of inequality in an election year and the level in the year of the previous election. There are three reasons for taking this measure, rather than just the simple level of inequality. First, I argue that it is not inequality itself that is problematic, but rather the rise of inequality. The second and third reasons explain the way the variable is constructed. I take changes of inequality between two elections to account for the rise/decrease of inequality levels in the whole governments' term, even though these terms are not always of the same duration. Finally, inequality is something that changes slowly over time. Taking only the most recent years into account (i.e., the election year and the year immediately before) would give us a variable with minimal variation. Thus, I take the change in the period between the two elections. During that period, two-thirds of the countries in the analysis witnessed a rise in inequality (Fig. 5).
To tackle the traditional economic macro-indicators and in an attempt to disentangle the rise in inequality from other economic problems, I control for-GDP growth (annual) and the change in unemployment (for the election year minus the year before). This approach avoids the problem of collinearity (Nedau et al. 2013). 6 A dummy indicator for left-wing governments is included. The distinction between left-wing and right-wing governments is based on the ideology of the party of the Prime minister. Hellwig showed that "voters assign most, if not all, credit or blame for economic performance to this party, even in cases of coalition governments" (2012). Parties are classified on a left-right scale using MARPOR data 7 (2020). Additionally, instead of dummy classification, I have used MARPOR's rile score (2020) for a party of Prime minister, to classify countries in the analysis, as a robustness check. I take this approach, acknowledging that binary classification potentially hides party differences within these two broad groups. Effect for the interaction between redistribution preferences and MARPOR's rile scale remains in the expected direction and statistically positive. Results are presented in Appendix 5.
Because of the structure of the dataset, to avoid the violation of the independence of observations, and due to the binary dependent variable, I fit a multilevel logit model, 8 with several interactions to test the effect of redistribution preferences and levels of inequality changes, and redistribution preferences and indicator for leftwing incumbents. I add random slope for redistribution preferences, as lower-level variable in the interaction, following Heisig and Schaeffer (2019).

Analysis
First, it is worth mentioning that the empirical models provide evidence in support of a long-lasting incumbency-oriented hypothesis about the relationship between the perception of the economy and support for incumbents. Respondents who  considered the economy to be better than it was year ago, were more likely to vote for incumbents (Table 1). The main independent variable in the model, redistribution preferences, is connected with support for incumbents. The more a voter wants redistribution, the less likely they are to support incumbents. To further explore the relationship between redistribution preferences and incumbents support, and to make better sense of the logit coefficient, I have calculated the expected change in incumbents' vote share if preferences for redistribution were to change by one standard deviation. Since the distribution of the dichotomous dependent variable is 0.6 (0) and 0.4 (1), I used a linear probability estimate with OLS to calculate the vote share by multiplying the regression coefficient by the variable mean absolute deviation (Lewis-Beck and Quinlan 2019). With an increase in redistribution preferences of one standard deviation, incumbent vote share is expected to decrease by 6.4%.
The effect of the interaction between redistribution preferences and the change in Gini is in the expected direction. Voters who prefer more redistribution tend to punish governments more in countries that experience a greater rise in inequality. The effect provides evidence for H1. Additionally, the coefficient for preferences for redistribution in Model 2, after including the interaction, is positive and significant. This coefficient (Brambor et al. 2006) indicates that voters who want more redistribution are more likely to reelect incumbents in countries which have low scores on the variable changes of inequality (so the countries where inequality levels reduced or stayed the same in-between election years).
For the substantive effect, I follow Williams (2012) and examine the marginal effect of the variable of interest at representative values of the other variable. Thus, I plot the effect of redistribution preferences over the change in inequality. Figure 6 reveals weak substantive evidence for H1. The marginal effect of redistribution preferences slightly changes with a rise in inequality, but the change of the interaction effect is in the expected direction. However, the effects are small and the confidence intervals large, thus there is not insufficient evidence to substantiate H1.
Model 3 shows that voters who want more redistribution, in a country with leftwing incumbents, are more likely to reelect them (H2a), and voters wanting more redistribution in countries with right-wing incumbents are more likely to support the opposition (H2b). To allow readers to more intuitively grasp the meaning behind this interaction, I have recalculated the same model (Model 3), with the rescaled variable 9 included in the interaction, so that it is easier to present the marginal effects at the two levels of the left-wing government dummy (Williams 2012) (Fig. 7). The graph shows that voters align with what parties usually stand for. Those who want more redistribution tend to vote for left-wing governments and punish the right-wing ones.  Fig. 7 Marginal effect of preferences for redistribution over the government's ideology with a 95% CI For H3a and H3b, I present marginal effect of redistribution preferences (for respondents who want more redistribution on the same, above-mentioned, three-category redistribution variable) over different values of inequality changes in between elections, separately for left-wing incumbents and right-wing incumbents. The results are in the following figure (Fig. 8).
As we can see from both the models and the plotted effects, there is modest support for H3a and H3b. Both left-wing and right-wing incumbents are punished by voters who want more redistribution in countries that have experienced a rise in income inequality. However, the effects are not that pronounced.
I conclude by summarizing and discussing these findings and their implications.

Discussion
The economy is still a strong predictor of vote choice. The effect of the evaluation of the economy is one of the strongest predictors of the level of vote for incumbents across the countries in the analysis. Is there anything left for economic issues in this equation? Positional economic voting research, we might argue, is sometimes trapped into showing the tautological relationship between the variables and the ideological leanings of political parties. I take a different approach, trying to show how Fig. 8 Marginal effect of preferences for redistribution over the change of inequality with a 95% CI, for left-wing and right-wing governments taking a position on economic issues affects performance evaluation conditional on the incumbents' ideology. Voters care about redistribution; models provide evidence that preferences for redistribution influence both vote choice and support for incumbents. Additionally, I find very modest support that the effect of preferences for redistribution is stronger in countries where inequality increased more significantly between elections. Further disentangling this relationship with vote could provide more robust findings. This modest finding for H1 in this paper, and potentially similar more robust findings in the future, could be expected and explained by the fact that in the period after the Global Financial Crisis, inequality increased worldwide, which made redistribution more salient as an issue.
The effect of redistribution interacting with an indicator for left-wing governments showed that voters cast their votes in accordance with the longstanding positions of the parties and vote for those who are close to their ideological position (Ferland and Dassonneville 2019). "In general, the `partisan theory' of political competition assumes that left-wing parties represent lower-income voters and right-wing parties, high-income voters. Hence, if a leftist government is in power, it should opt for higher spending on items which redistribute to the poor" (Borck 2007). Borck's explanation seems to capture the essence of the effect-those who want more redistribution, tend to keep left-wing incumbents and remove right-wing ones. In this case, voters opt not to sanction, but to choose based on competency (Hellwig 2012).
Further exploring the role of inequality in the relationship between preferences for redistribution and vote for left-wing and right-wing incumbents highlighted the fact that voters who want more redistribution punish both types of incumbent when income inequality is on the rise. However, the effect, across different levels of increases in inequality, is slightly more pronounced for left-wing governments. The strength of these effects is modest.
The analysis provides very modest support (at best) for the view that changes in inequality negatively influence support for incumbents. It seems that inequality influences attitudes (such as preferences for redistribution), which further affect incumbent support. Scholars trying to tie inequality to government support should be aware of two further issues: (a) The relationship should be tested against the change in inequality, rather than actual levels of inequality; and (b) the period for capturing the change in inequality should be longer, not just focused on the change in the previous year.
Unfortunately, the data do not provide us with the measure of perceived inequality, nor any indication of how much blame citizens place on government for rising inequality. Who citizens actually blame for rising inequality would be an interesting research question but is not in the scope of this article. Here, I make the argument that inequality indirectly affects incumbents through preferences for redistribution.
Future research on the topic should consider panel data (when possible) in an attempt to better understand the dynamics of changes in both inequality and redistribution preferences over time. Potential fertile ground for the next step in this direction would be to include, not just income inequality, but also wealth inequality, since research shows that wealth inequality is even more pronounced than income inequality (Alvaredo et al. 2017). Researchers could explore the mechanisms between vote choice and wealth inequality and the policies that combat wealth inequality. Further, talking about the theoretical foundations of this paper, future research should explore the role of patrimony, the third dimension of economic vote, as Lewis-Beck et al. (2013) scratched the surface of that topic.
Lastly, this research could be an added argument in the very heated discussion about the "crises of the left," especially in Europe and the future of left-wing parties more generally, since it shows that they are still winning on one of their traditional stronghold issues-preferences for redistribution.

Appendix 5
See Table 4.   Please note that higher values on MARPOR rile scale mean that party is more right. Additionally, ideology variable for models presented in Appendix 5 has been created from left-right self-placement which is first multiplied by − 1 (then + 1 added) if the MARPOR rile scale value is smaller than 0  Standard errors in parentheses; Source: CSES Module 4 (2018) & CSES Module 5 AR2 (2020); * p < 0.05, ** p < 0.01, *** p < 0.001