Introduction
Rising economic inequality is one of the defining political issues of our time (Kenworthy and Pontusson, Reference Kenworthy and Pontusson2005), with the decade since the global financial crisis (2007–2009) seeing numerous warnings about rising inequality in advanced democracies (OECD, 2011; Piketty, Reference Piketty2014). Beyond its negative consequences for well-being (Neckerman and Torche, Reference Neckerman and Torche2007), inequality also has important social and political implications. Scholars have raised concerns that inequality may encourage disillusionment with democracy and drive ideological polarisation (Andersen, Reference Andersen2012; Iversen and Soskice, Reference Iversen and Soskice2015; Winkler, Reference Winkler2019). Addressing inequality and its consequences will undoubtably form a key part of political debates in years to come.
It is important to ask: what do citizens want from their governments in light of growing economic inequality? Specifically, how does inequality affect attitudes towards social policies such as redistribution? This focus on mass public opinion is important both on normative grounds (policy should reflect the wishes of the public), and for empirical reasons: studies from several countries have found that public policy preferences tend to be reflected in government decisions, welfare state size, and inequality (Aalberg, Reference Aalberg2003; Page and Shapiro, Reference Page and Shapiro1983; Soroka and Wlezien, Reference Soroka and Wlezien2005; Wlezien, Reference Wlezien1995).
Recent research on public attitudes towards redistribution has often focused on country-level context as a key determinant of preferences. Scholars have sought to explain differences in support for redistribution as a function of occupational risks of unemployment (Rehm, Reference Rehm2009), the prevalence of economic hardship (Blekesaune, Reference Blekesaune2013), different welfare state institutions (Jæger, Reference Jæger2006, Reference Jæger2009; Linos and West, Reference Linos and West2003; Svallfors, Reference Svallfors1997), policy feedback (Pierson, Reference Pierson1993; Soss and Schram, Reference Soss and Schram2007), and macroeconomic factors (Dallinger, Reference Dallinger2010; Finseraas, Reference Finseraas2009; Jæger, Reference Jæger2013; Lübker, Reference Lübker2007; Steele, Reference Steele2015). This research has demonstrated strikingly the influence of context – particularly macroeconomic context – on support for redistribution.
At the same time, scholars have long understood that support for redistribution is largely driven by individual characteristics, particularly a person’s core beliefs and socio-demographic position. Among these individual-level predictors, political ideology is widely thought to be the most consequential (Bobo, Reference Bobo1991; Feldman and Zaller, Reference Zaller1992; Iversen and Soskice, Reference Iversen and Soskice2001; Jacoby, Reference Jacoby1994; Sears et al., Reference Sears, Lau, Tyler and Allen1980). Not only is political orientation important at the individual level, but the competition between left and right plays a critical role in structuring domestic debates about redistribution (Feldman and Steenbergen, Reference Feldman and Steenbergen2001). Given increasing warnings about polarisation in advanced democracies (Iversen and Soskice, Reference Iversen and Soskice2015; Winkler, Reference Winkler2019), understanding the dynamics of contestation between left and right over specific policy questions has become a pressing issue within the study of politics.
While scholars have acknowledged the theoretical connections between individual political and country-level economic variables, this has rarely been tested empirically. For instance, while earlier research in the welfare regime tradition acknowledged how social policy affects both the level and distribution of public support for the welfare state (Jæger, Reference Jæger2009; Linos and West, Reference Linos and West2003; Svallfors, Reference Svallfors1997), such work did not specifically address how macroeconomic variables shaped individual policy attitudes, or how these variables interacted with individual political preferences in shaping attitudes. In this way, the question of whether macroeconomic context influences the degree of left-right divergence regarding redistribution, both cross-sectionally (across countries) and longitudinally (within countries), remains open.
In this article, we explicitly test the proposition that the effect of individual political orientation on redistributive preferences varies between and within states as a function of macroeconomic context – specifically, as a function of country-level wealth and income inequality. While research has explored the link between inequality and political polarisation (Iversen and Soskice, Reference Iversen and Soskice2015; Winkler, Reference Winkler2019), our focus here is specifically on polarisation understood as a divergence of policy attitudes – that is to say, larger differences in the redistributive preferences of the left and right (Lelkes, Reference Lelkes2016). To test this, we draw on a time series cross-sectional dataset from sixteen countries and four waves of the International Social Survey Programme (ISSP). We find that redistributive preferences exhibit greater divergence in wealthier and more unequal states – this is to say that greater country wealth and income inequality predict increased left–right divergence in redistributive preferences. Crucially, we also find that rising inequality within states over time implies larger differences in preferences between left and right. These findings constitute an important empirical validation of the interactive relationship between political orientations and inequality with relation to policy preferences, and reveal potentially serious challenges to traditional forms of redistribution in years to come.
Theory and hypotheses
Social scientists have long theorised that attitudes towards redistribution are driven by both economic interests and personal values. Political sociologists have demonstrated how lower and working classes tend to exhibit greater support for redistribution than the middle and upper classes (Andersen and Yaish, Reference Andersen and Yaish2018; Blekesaune and Quadagno, Reference Blekesaune and Quadagno2003; Dallinger, Reference Dallinger2010; Jæger, Reference Jæger2006, Reference Jæger2009; Linos and West, Reference Linos and West2003; Rehm, Reference Rehm2009; Svallfors, Reference Svallfors1997). Equally, research has shown left–right political orientation to be a major axis of conflict, with those on the political left expressing greater support for redistribution than those on the right (Bobo, Reference Bobo1991; Feldman and Zaller, Reference Zaller1992; Iversen and Soskice, Reference Iversen and Soskice2001; Jacoby, Reference Jacoby1994; Sears et al., Reference Sears, Lau, Tyler and Allen1980). Studies have shown that identification with the left or right has an exogenous effect on redistributive preferences, irreducible to other underlying variables (Jæger, Reference Jæger2008).
How ought we understand the connection between political orientation and social policy preferences? The psychological approach locates individual left–right (or liberal–conservative) orientations in the rejection or acceptance of hierarchy and inequality, as well as the social and economic status quo (Jost et al., Reference Jost, Glaser, Kruglanski and Sulloway2003b). As a group-based phenomenon, left–right orientation allows individuals to take positions on policy issues without obtaining detailed policy knowledge (Zaller, Reference Zaller1992). Via the mechanism of ‘elite cues’, individuals reflect the policy platforms promoted by the political elites and parties for whom they have an ideological affinity (Zaller, Reference Zaller1992). Through this, citizens who may not have detailed opinions on all policy matters remain capable of articulating policy preferences by calling upon mental shortcuts, and echoing the preferences of elites on their ‘side.’
Nonetheless, individual-level variables do not explain all variation in attitudes. Rather, context also has a crucial role to play. Building on research within the welfare regime tradition (Esping-Andersen, Reference Esping-Andersen1990), which emphasised how country-level characteristics influenced public debate over social policy, recent studies have found that macroeconomic factors exert a significant influence on aggregate policy preferences. For example, citizens of wealthier countries tend to be less supportive of redistribution than those elsewhere; in addition, as countries get wealthier, support for redistribution tends to decline (Blekesaune, Reference Blekesaune2007, Reference Blekesaune2013; Dallinger, Reference Dallinger2010; Jæger, Reference Jæger2013), with the upshot that welfare state retrenchment may become more likely.
Other research emphasises the effect of country-level income inequality on attitudes. Meltzer and Richard provide the classic statement on public demand for redistribution, where demand (based on the income of the median voter relative to the average income) increases with income inequality (Meltzer and Richard, Reference Meltzer and Richard1981). Some authors have found evidence supporting this contention, both between countries and within them over time (Finseraas, Reference Finseraas2009; Jæger, Reference Jæger2013). Others find that inequality has no effect, or only a small effect on preferences (Steele, Reference Steele2015). Further, both Dallinger and Lübker find that the impact of inequality depends on the regime-type of a country: the citizens of minimalistic welfare states are less sensitive to increased inequality than those habituated to high levels of social protection (Dallinger, Reference Dallinger2010; Lübker, Reference Lübker2007). Indeed, alternative models such as that of Bénabou predict the opposite of Meltzer and Richard: that with rising inequality support for redistribution will decline, driven by those whose relative economic standing is enhanced by inequality (Bénabou, Reference Bénabou2000). This conjecture equally finds support from some survey-based and field experiments (Sands, Reference Sands2017; Trump, Reference Trump2018).
While previous research implicitly acknowledges a potential connection between individual- and country-level variables, the predominant empirical approach used makes two (theoretically and statistically equivalent) assumptions: first, that macroeconomic context affects all individuals equally regardless of their political orientation, and second, that political orientation has the same influence on preferences across all economic contexts. These assumptions are unsatisfactory. A growing body of research finds that contexts (variously defined in terms of the socio-demographic and economic milieux individuals inhabit) interact with political predispositions in shaping policy preferences in a variety of domains spanning energy issues, foreign relations, border security, and immigration policy (Branton et al., Reference Branton, Dillingham, Dunaway and Miller2007; Gravelle, Reference Gravelle2014, Reference Gravelle2016, Reference Gravelle2018, Reference Gravelle2019; Gravelle and Lachapelle, Reference Gravelle and Lachapelle2015). In this way, it seems probable that individuals form their redistributive preferences based on their political orientations, but in ways that depend on their economic context. Therefore, we predict that the cleavage in the redistributive preferences of supporters of the left and right differs in size as a function of economic context. This is to say that we expect economic development and income inequality on the one hand, and individual-level left–right political orientations on the other, to interact.
Such interactive relationships are not entirely new to research on redistributive preferences. Earlier research found that different welfare regimes generate different attitudinal cleavages – between classes, genders, and occupations – within the population (Jæger, Reference Jæger2009; Linos and West, Reference Linos and West2003; Svallfors, Reference Svallfors1997). Such research was pioneering in its revelation that social policy itself influences the degree and location of political conflict. It paid insufficient attention, however, to the conflict driven by individual-level political orientation.
Scholars’ interest in these interactions is now increasing. For instance, Jaime-Castillo and Saéz-Lozano find that cross-country differences in taxation levels are associated with the size of the attitudinal gap between left and right regarding social policy (Jaime-Castillo and Sáez-Lozano, Reference Jaime-Castillo and Sáez-Lozano2016). Longitudinal analyses focusing on times of crisis have formed another strand of research: for instance, both Forma (Reference Forma2002) and Gonthier (Reference Gonthier2017) consider whether public opinion has historically been more divided during periods of economic downturn. Still, neither study directly connects changing macroeconomic context with the size of the gap in redistributive policy preferences between left and right. The question of how economic change influences political contestation over redistribution remains unanswered. The goal of this article is to test this interaction explicitly.
Concerning the expected interaction between country-level wealth and political orientation, our expectation is that as wealth grows – both across, and within states over time – conservative opposition to redistribution relative to those on the left will likewise increase. Individuals who identify with the political right are likely to prioritise ideals of hard work and merit in assessing the ‘deservingness’ of the beneficiaries of redistribution (Feldman and Zaller, Reference Zaller1992). They are more likely to see the welfare state as a technocratic instrument aimed at rectifying the failings of the market, rather than as a transformative egalitarian tool (Sloman, Reference Sloman2018). Therefore, as country-level wealth rises, those on the right are more likely to view redistribution as unnecessary. In other words, market outcomes in more developed country contexts (in comparison to less developed countries) are more likely to be seen as just by those on the political right (Jost et al., Reference Jost, Blount, Pfeffer and Hunyady2003a; Jost and Thompson, Reference Jost and Thompson2000). By contrast, those on the political left are likely to continue to press for a more equal distribution of resources.
The effect we predict reflects a form of ideological ‘divergence,’ as defined by Lelkes (Reference Lelkes2016), where supporters of left and right diverge in their policy preferences even though they may move in the same direction overall. While this form of divergence may seem intuitively less severe than if left and right were to move in entirely opposite directions, it still has significant implications for political contestation. Scholars have long established the reliance of successful welfare states on broad-based ‘coalitions’ of support (Esping-Andersen, Reference Esping-Andersen1990; Pierson, Reference Pierson and Pierson2001), which, it seems, would be equally undermined by increasing ‘divergence’ as they would be by increasing ‘consistency.’
Political elites in more economically developed countries are also likely to have more oppositional stances regarding economic and social policy, which is expected to be mirrored in mass opinion via the mechanism of elite cues (Huber and Inglehart, Reference Huber and Inglehart1995; Zaller, Reference Zaller1992). Given the longer time-frame implied by this mechanism, it is likely a better explanation of variation between countries than within them. This is supported by Thorisdottir and colleagues, who, using European Social Survey data, find that left–right orientation is associated with acceptance of inequality in the more economically advanced countries of Western Europe, but no association is found in the less advanced countries of Eastern Europe (Thorisdottir et al., Reference Thorisdottir, Jost, Liviatan and Shrout2007). We therefore test the following hypotheses:
H1a (between effect): With higher country-level wealth (measured as gross domestic product per capita), the effect of political orientation on redistributive preferences is larger.
H1b (within effect): As country-level wealth (measured as gross domestic product per capita) increases, the effect of political orientation on redistributive preferences increases.
As for the hypothesised interaction between political orientation and income inequality, our expectation is that greater inequality, both between and within countries over time, will lead to a more pronounced gap between left and right. To the extent that the macro-economy embodies inequality, we should expect that conservatives would be more likely to accept (and indeed endorse) such inequality, and to justify its maintenance (Jost et al., Reference Jost, Glaser, Kruglanski and Sulloway2003b; Trump, Reference Trump2018). Conservatives are also likely to perceive revisions to an unequal economic status quo as more threatening relative to a less unequal one (Jost and Thompson, Reference Jost and Thompson2000). By contrast, as inequality rises, egalitarian values should lead to stronger support for redistribution among those on the left. As such, we expect the effect of political orientation to be amplified as income inequality increases. In a similar fashion, then, we expect to observe a pattern of ‘divergence’ in the policy preferences of the left and right (Lelkes, Reference Lelkes2016).
While some authors suggest that the shared experience of rising inequality may push those on the right and left closer together (Forma, Reference Forma2002), we are less confident in this proposed relationship, since it implies that those on the left and right perceive and interpret contextual cues identically – a notion contradicted by a range of evidence (Branton et al., Reference Branton, Dillingham, Dunaway and Miller2007; Gravelle, Reference Gravelle2016, Reference Gravelle2019). We propose that, other things (notably class and income) being equal, those on the left and right experience inequality differently: as fair or unfair, a result of individual differences or systematic injustice, and as sign of national economic success or failure. This is not to deny the possibility of social solidarity across political lines, but rather to predict that it will be weakened by the different reactions of political groupings to context. This leads to our second set of hypotheses:
H2a (between effect): With higher country-level income inequality, the effect of political orientation on redistributive preferences is larger.
H2b (within effect): As country-level income inequality increases, the effect of political orientation on redistributive preferences increases.
Data and Methods
To evaluate these theoretical propositions, we draw on time series cross-sectional survey data from the International Social Survey Programme (ISSP). Specifically, we draw on the 1990, 1996, 2006, and 2016 waves of the ISSP focused on the role of government, and include countries represented in at least three of the four waves. Earlier waves of the ISSP have been used to study redistributive policy preferences (Dion and Birchfield, Reference Dion and Birchfield2010; Jæger, Reference Jæger2009; Linos and West, Reference Linos and West2003; Steele, Reference Steele2015; Svallfors, Reference Svallfors1997). Our analyses include the recently available 2016 wave, which postdates the global financial crisis, and extends the time period covered by a further decade. The 16 countries included in our analyses are: Australia, the Czech Republic, France, Germany, Ireland, Israel, Japan, Latvia, New Zealand, Norway, Slovenia, Spain, Sweden, Switzerland, the United Kingdom, and the United States. Our data comprise 76,850 respondents clustered within 54 country–wave observations. While it would be desirable to increase the number of countries and years, data completeness prevents us from doing so – we discuss this limitation more in the conclusion. Nonetheless, on aggregate, these country–wave observations also exhibit widely varying levels of wealth and income inequality. They also exhibit a pattern of increasing wealth and increasing inequality over time (see Figure 1; see also the descriptive statistics in the Online Appendix).
Figure 1. Market Inequality and GDP per Capita (PPP) of Included Countries, 1990-2016
Our measure of support for redistribution from the ISSP asks: “On the whole, do you think it should or should not be the government’s responsibility to reduce income differences between the rich and the poor?” The possible responses were: “Strongly Disagree,” “Disagree,” “Agree,” and “Strongly Agree”. The question is pitched at a broad level and leaves it to the respondent whether the question relates to the ideal role of government or the need for government action in the present (Dallinger, Reference Dallinger2010). The question also does not direct respondents to think of specific policies meant to address inequality. It is, however, more evocative of redistribution from the wealthy than redistribution to the poor (Cavaillé and Trump, Reference Cavaillé and Trump2015). Nevertheless, we contend that the question serves as a broad-gauged measure of support for redistribution that is fit for the purpose of comparative research.
To capture political orientations, the ISSP asks respondents for the political party they voted for in their country’s last election. This can be considered as a measure of respondents’ latent political and ideological orientations in contrast to their self-identification with the “left” or “right” in politics as such, which was not asked in the ISSP survey instruments. To facilitate comparative analysis, country experts coded each party into the categories “Far Left,” “Left,” “Centre,” “Right,” and “Far Right” (ISSP Demographic Methods Group, 2017). Previous research has relied on these party codes as a measure of left–right position (Iversen and Soskice, Reference Iversen and Soskice2001; Jaime-Castillo and Sáez-Lozano, Reference Jaime-Castillo and Sáez-Lozano2016). From these codes, we create synthetic “Left,” “Centre” and “Right” categories, reflecting the fact that very few respondents voted for parties coded as far right or far left.
Our analyses also include a set of country-level economic variables. The first is the standard measure of relative country wealth: gross domestic product per capita (logged) measured using international dollars in purchasing power parity (PPP) form for the appropriate years, which are obtained from the World Bank International Comparison Program (World Bank, 2018). Though we recognise that economic inequality is notoriously difficult to define and measure (Piketty, Reference Piketty2014), we opt to use the Gini index, a measure of income inequality widely used in the comparative welfare states literature. Our Gini index data are obtained from the Standardized World Income Inequality Database version 8.1 (Solt, Reference Solt2016), the only source we know of for which data are available for all relevant countries and years. The SWIID data provide both a ‘market Gini’ (the Gini index of a country before taxes and transfers) and a ‘disposable Gini.’ Following Meltzer and Richard’s argument that redistributive preferences are shaped by market (pre-transfer) inequality (Reference Meltzer and Richard1981), we employ the market index in our study.Footnote 1 Some research finds that welfare state size (measured by social expenditure as a percentage of GDP) also shapes support for redistribution (Jæger, Reference Jæger2006, Reference Jæger2013; Steele, Reference Steele2015). We therefore include social spending as a percentage of GDP in our analyses, which we obtain from the OECD’s Social Expenditure Database (OECD, 2019).
Previous research also identifies a range of socio-demographic characteristics that shape redistributive policy preferences. These include sex, age, educational attainment, income, occupation, and employment status (Andersen and Curtis, Reference Andersen and Curtis2015; Dion and Birchfield, Reference Dion and Birchfield2010; Jæger, Reference Jæger2009; Linos and West, Reference Linos and West2003; Schmidt-Catran, Reference Schmidt-Catran2016; Steele, Reference Steele2015; Svallfors, Reference Svallfors1997). We therefore include these variables as controls in our analyses.
The four ordered categories of the ISSP redistribution measure point toward ordinal logit models. The ISSP data also clearly exhibit a multilevel structure: individuals are nested within country contexts at particular points in time, which are further nested within a given country context. These considerations point us to multilevel models (specifically generalised linear mixed models) with respondents as level 1, country-waves as level 2, and countries as level 3 (Hox, Reference Hox2010; Raudenbush and Bryk, Reference Raudenbush and Bryk2002). Random intercepts are specified at both the country-wave and country levels. These random intercepts address (though cannot entirely eliminate) the issue of possible unmeasured variation between countries and years.
Further, countries are characterised by particular levels of aggregate wealth and income inequality (among other factors) across time. They are similarly characterised by relative changes in wealth and inequality over time. We therefore include separate cross-sectional (between) and longitudinal (within) effects for wealth and inequality. We do this by calculating country-level mean scores for both GDP per capita and market Gini. These capture enduring country-level differences in wealth and inequality. We also calculate country mean-centred values for GDP per capita and market Gini. These capture within-country changes in wealth and inequality over time (that is, longitudinal, or within effects). By construction, these country mean and country mean-centred scores are orthogonal, allowing us to decompose the cross-sectional and longitudinal effects of wealth and inequality on redistributive policy preferences (Fairbrother, Reference Fairbrother2014; Schmidt-Catran, Reference Schmidt-Catran2016; Schmidt-Catran et al., Reference Schmidt-Catran, Fairbrother and Andreß2019). This allows us to test our hypotheses relating to cross-level interaction effects between political orientation and each of the cross-sectional and longitudinal components of wealth (GDP per capita) and inequality (market Gini). Finally, by including the survey year of each response as a categorical variable in our longitudinal models, we control for any idiosyncratic variations between years.
Results
The results from the multilevel ordinal logit models yield several findings of note (see Table 1). Examining first the effects of the covariates, the results from Models 1–5 indicate that women express slightly greater support for reducing income differences between the rich and poor, as do older individuals. Unsurprisingly, measures of higher socio-economic status are negatively associated with support for redistribution: higher levels of educational attainment (particularly holding a university degree) and having a top-quintile income are associated with decreased support for reducing income differences. By contrast, having a bottom-quintile income is associated with increased support. Those who report being employed are also slightly less likely to support reducing income differences.
TABLE 1. Explaining Support for Redistribution, ISSP, 1990-2016
* p ≤ 0.05, ** p ≤ 0.01, *** p ≤ 0.001.
Notes: Models are generalized linear mixed models fit by maximum likelihood using Gauss-Hermite quadrature. All continuous variables are grand mean-centered unless noted otherwise.
Turning to the effect of party support (as our working measure of political orientation), the effects of left and centre party support are highly significant and in the expected direction: centre and left-wing party supporters express greater support for redistribution compared to right-wing party supporters. This aligns with the results of previous single-country and cross-national studies (Bobo, Reference Bobo1991; Feldman and Zaller, Reference Feldman and Zaller1992; Iversen and Soskice, Reference Iversen and Soskice2001; Jacoby, Reference Jacoby1994; Sears et al., Reference Sears, Lau, Tyler and Allen1980). We can gauge the substantive importance of this effect by translating the coefficients into predicted probability terms (setting the year to 2016 and holding all other variables at their means or reference categories). Doing so, we see that the predicted probability of a left-wing party supporter strongly agreeing that it is the responsibility of government to reduce income differences is 0.57; this drops to 0.43 for centre party supporters; it drops further to 0.29 for right-wing party supporters.
As for the effects of country-level macroeconomic context, the results from Model 1 do not provide evidence of an association between the level of social spending and redistributive preferences. With respect to the cross-sectional (between) effects of interest, though, the results from Model 1 indicate that higher GDP per capita is associated with less support for reducing income differences between the rich and poor (Andersen and Yaish, Reference Andersen and Yaish2018; Dallinger, Reference Dallinger2010; Dion and Birchfield, Reference Dion and Birchfield2010). To illustrate again using predicted probabilities, a centre party supporter in Latvia (mean GDP per capita: 15,904) has a predicted probability of strongly agreeing with reducing income differences of 0.61; this drops to 0.43 in Japan (mean GDP per capita: 35,812), and 0.35 in Switzerland (mean GDP per capita: 52,592). Still, there is no evidence of an effect of increasing country-level wealth: the longitudinal (within) effect of changes in GDP per capita is not statistically significant. There is also no evidence of an overall cross-sectional (between) effect of inequality: residing in a more unequal country is not associated with greater (or lower) support for redistribution. By contrast, growing inequality within a given country context is associated with stronger preferences for redistribution: the longitudinal (within) effect of market Gini is positive and significant. The findings of no cross-sectional effect of inequality but a positive longitudinal effect align with the analyses of Schmidt-Catran using European Social Survey data (Reference Schmidt-Catran2016). To illustrate the effect of growing income inequality using predicted probabilities, a centre party supporter in any given country experiencing a 2 point decrease in market Gini has a predicted probability of strongly agreeing with greater redistribution of 0.38; it is 0.43 in a country experiencing no change in market Gini, and 0.48 in a country experiencing a 2 point increase in market Gini. These results indicate that increases in country-level inequality do not need to be severe to produce appreciable changes in public preferences for redistribution.
The results obtained in Model 1, of course, assume that the effects of party support and GDP per capita are additive, and not interactive. Models 2–5 test separate interactions between party support and the cross-sectional and longitudinal effects of both GDP per capita and market Gini score that correspond to our research hypotheses. In line with our theoretical expectations, the results from Model 2 confirm a highly significant GDP per capita (between) × party support interaction (χ2(3) = 46.54, p < 0.001). The lower-order coefficient for GDP per capita (between) is negative and significant, indicating that, cross-sectionally, support for redistribution among right-wing party supporters is lower in wealthier countries. By contrast, the higher-order order GDP per capita (between) × left-wing and GDP per capita (between) × centre coefficients are both positive and significant though smaller in magnitude than the lower-order coefficient for GDP per capita (between), signifying increasing divergence in preferences, where the negative effect of GDP on attitudes is attenuated among the political left and centre, but not increasing consistency (which would require that the interaction coefficient be larger in magnitude than the lower-order coefficient).
To explicate this complex pattern of results, calculating predicted probabilities is again useful, as is depicting the results graphically. Figure 2 makes clear that higher GDP per capita is associated cross-sectionally with diminished support for reducing income differences among all political segments of mass publics, but this relationship is most pronounced among the political right. To illustrate, a left-wing party supporter in Latvia has a predicted probability of strongly agreeing with reducing income differences of 0.68 while a right-wing party supporter has a predicted probability of 0.51; the cleavage between left and right widens for Japan to 0.57 and 0.29, and for Switzerland to 0.52 and 0.21. In sum, the ISSP data support our first hypothesis that greater wealth produces greater left–right polarisation on the question of redistribution.
Figure 2. Cross-sectional Interaction of GDP Per Capita with Political Orientation
By contrast, Model 3 yields no evidence of an interaction between party support and rising country-level wealth longitudinally. The GDP per capita (within) × party support interaction is not significant (χ2(3) = 4.55, p = 0.208). To illustrate, a left party supporter in a country experiencing no GDP growth has a predicted probability of strongly agreeing with greater redistribution of 0.55. By comparison, the predicted probability of a left party supporter in a country experiencing an increase in logged GDP of 0.1 – equivalent to raising GDP from 30,000 to 33,155, or from 50,000 to 55,258 – is not appreciably different at 0.56 (see Figure 3). For supporters of parties on the right, the predicted probability is 0.28 in both instances. So, while greater wealth cross-sectionally leads to greater polarisation in redistributive preferences, rising wealth longitudinally does not exert an effect among the left or the right.
Figure 3. Over-time Interaction of GDP Per Capita with Political Orientation
What of country-level income inequality and redistributive preferences? While Model 1 indicated no main (or additive) cross-sectional effect of income inequality, Model 4 yields a significant Gini index (between) × party support interaction (χ2(3) = 57.30, p < 0.001). Thus, income inequality moderates the effect of party support on preferences for redistribution, confirming our second hypothesis. Calculating predicted probabilities and plotting them in Figure 4 makes clear that, in cross-sectional terms, greater income inequality is associated with greater polarisation in the redistributive preferences of the political left and right. To illustrate, a left party supporter in Japan (mean market Gini: 42.6) has a predicted probability of strongly agreeing with reducing income differences of 0.60 while a right-wing party supporter has a predicted probability of 0.34. The cleavage between left and right widens for Australia (mean market Gini: 46.6) to 0.57 and 0.29, and for the United Kingdom (mean market Gini: 52.2) to 0.53 and 0.23. Supporters of left-of-centre parties thus appear to ‘resist’ the cross-sectional contextual effect of inequality, while greater income inequality amplifies the negative effect of right-wing party support on redistributive preferences.
Figure 4. Cross-sectional Interaction of Market Income Inequality with Political Orientation
Considering income inequality longitudinally instead of cross-sectionally, a different pattern emerges. While Model 1 yielded a significant main (within) effect of increasing income inequality, Model 5 adds nuance in yielding a significant Gini index (within) × party support interaction (χ2(3) = 30.23, p < 0.001). The lower-order coefficient for market Gini (within) is positive and significant, which indicates that support for redistribution among right-wing party supporters is higher as inequality increases over time. Further, the higher-order order market Gini (within) × left-wing coefficient is positive. This indicates that the longitudinal effect of increasing income inequality on support for redistribution is amplified among those on the left compared to the right – a pattern made clear in Figure 5. These findings can again be expressed as predicted probabilities. A left party supporter in any given country experiencing a 2 point decrease in market Gini has a predicted probability of strongly agreeing with greater redistribution of 0.51 while a right-wing party supporter has a predicted probability of 0.26. The redistributive preferences of the left and right further diverge where there is no change in market Gini to 0.57 and 0.29, and further still when a country experiences a 2 point increase in market Gini to 0.63 and 0.33. These results demonstrate that while increasing within-country inequality leads both the left and right to shift their preferences toward greater redistribution, those on the left are more responsive to such a change in economic context than those on the right.Footnote 2
Figure 5. Over-time Interaction of Market Income Inequality and Political Orientation
Discussion and Conclusion
The motivating question of this article has been: how do macroeconomic context and political orientation interact in shaping public demand for redistribution? Drawing on time series cross-sectional data from the ISSP between 1990 and 2016, we have endeavoured to push the empirical literature past the simple, additive effects of social class, political orientation, and country-level macroeconomic indicators to consider cross-level interaction effects – something that few studies have examined empirically (Andersen and Curtis, Reference Andersen and Curtis2015; Dion and Birchfield, Reference Dion and Birchfield2010; Finseraas, Reference Finseraas2009; Jæger, Reference Jæger2009). Our key findings are that both between-country variation and within-country change over time in macroeconomic context influence the size of left–right cleavages in redistributive policy preferences.
In investigating interactions between left–right political orientations (on the one hand) and GDP per capita and Gini scores (on the other), our analyses confirm that both country-level wealth and income inequality amplify the ideological cleavage in redistributive preferences between left and right. Concerning country-level wealth, our results show that, cross-sectionally, there is increased left-right divergence in redistributive preferences in wealthy states compared to less wealthy states, while longitudinally there is neither an additive effect of increasing country-level wealth nor an interaction with left–right orientations. Concerning levels of income inequality, our cross-sectional results indicate more unequal states have greater left–right cleavages in attitudes toward redistribution. Critically, our longitudinal results also indicate that increasing inequality within states increases support for redistribution on average, though this effect is more pronounced among the left. The divergence in attitudes between left and right therefore widens as inequality grows.
These findings support our overarching theoretical argument that the effect of macroeconomic context on individual preferences is not direct, but is instead moderated by pre-existing political orientations. Consequently, scholars wishing to understand how context affects policy preferences would do well to theorise and model public opinion not as an undifferentiated bloc but as different political segments – not only starting with different preferences, but also reacting to context in different ways. This finding, while explored in other domains of public opinion research (Gravelle, Reference Gravelle2016, Reference Gravelle2018, Reference Gravelle2019; Gravelle and Lachapelle, Reference Gravelle and Lachapelle2015), has been underappreciated in large-N studies of social policy preferences.
There remains room for scholars to build on the findings presented here. Firstly, comparative political behaviour scholars are all-too-familiar with the challenges of data availability for longitudinal or time series cross-sectional research. While a larger sample of countries would have been desirable, we remain confident in our findings, within the relevant geographic contexts. Similarly, we acknowledge the limitations of the Gini index as a measure of inequality: while alternative measures of inequality (such as the ratio of top-decile to bottom-decile incomes) were not easily found for the relevant countries and years, such data may become more available in future, at which time their incorporation into public opinion research will be highly desirable. Finally, taking inspiration from the economic voting literature (Anderson, Reference Anderson2000), we propose that the use of panel data and more subjective measures of inequality holds the prospect of extending our results by showing how individual citizens on the left or right respond when they perceive the macro-economy to be changing.
Another important question raised by our research is whether the gap between left and right is more or less volatile across country contexts. Scholars have long understood that publics perceive inequality through the prisms of their respective political cultures and histories (Lübker, Reference Lübker2007). Consequently, we would not be surprised if rising inequality were more divisive in some historical and geographic contexts than others – for example, in states where average support for redistribution was weaker and belief in ‘social rights’ less engrained, compared to contexts where support was higher and ‘social rights’ were accepted by both left and right. This is to propose a three-way interaction between macroeconomic context, political culture or history, and individual-level political orientations. Such research (that is, research into the relative resilience of different publics to ideological polarisation on social policy) would make a valuable contribution to current discussions regarding the rise of extreme populist parties in advanced democracies.
Our findings have a number of practical implications. One is that rising inequality may drive increased polarisation – understood as ideological divergence – between left and right, at least when it comes to social policy. Scholars have long understood that the maintenance of the welfare state depends not merely on sufficient aggregate support – but on broad-based coalitions of support from different groups within society (Esping-Andersen, Reference Esping-Andersen1990; Pierson, Reference Pierson and Pierson2001). Our findings show that while rising inequality is associated with a slight increase in support for redistribution overall, it erodes the kind of inter-group consensus on which redistributive policy relies. Though this result leaves us less than sanguine about the short-term prospects of addressing the challenges of rising economic inequality and declining support for democracy, it nevertheless lays bare the challenge faced by those who would endeavour to do so.
This is not to conclude that our findings point inexorably to continued welfare state retrenchment. Rather, we are left asking: could changing economic circumstances not instead promote the development of new forms of redistribution? A burgeoning field of research addresses what might be called “new” social policy approaches (Vaalavuo, Reference Vaalavuo2013), such as Social Investment and Entrepreneurship, and Active Labour Market Policies (ALMPs) (Fossati, Reference Fossati2018; Jenson, Reference Jenson2017; Van Vliet and Wang, Reference Van Vliet and Wang2015). Scholars have tended to portray these policies as resulting from shifts in discourse regarding the relative responsibilities of individual and state. Our findings suggest the possibility that, beyond their discursive foundations, such alternative policies may also be the result of changing economic circumstances. That is to say that rising inequality may be driving increased divergence on some kinds of social policy, while promoting agreement on others. Research into this dynamic will be crucial to understanding the practical implications of changing economic context for social policy preferences, and for social policy itself.
Acknowledgements
This article originated with Raja Noureddine’s Honours thesis at the University of Melbourne, supervised by Tim Gravelle, and later presented at the 2019 annual conference of the World Association for Public Opinion Research in Toronto, Ontario, Canada. Raja thanks Fadwa and Fadi Noureddine for their unfailing support throughout the research process. Tim thanks Carol, Bryan and Sadie Gravelle (as always).
Supplementary material
To view supplementary material for this article, please visit https://doi.org/10.1017/S0047279420000537
