Is democracy about redistribution?

Some scholars champion broad conceptualizations of democracy where distribution of economic resources is an integral part, whereas several prominent arguments drawing on narrower conceptualizations of democracy still assume that progressive redistribution is central to democratic politics. We empirically analyse individual opinions on whether progressive taxation and redistribution are among democracy's central characteristics. While many citizens around the world associate democracy with redistribution, we find that surprisingly few consider redistribution among the most central characteristics of democracy. We further analyse what factors affect individuals’ propensity to consider redistribution among democracy's most important features. Running multi-level models, we find that having lived under a communist regime and − although less robust – currently living under democracy make individuals less likely to hold this notion. However, individuals with more to gain from progressive redistribution (that is, little education and belonging to lower classes) are more likely to hold it. We discuss how our findings help shed light on two puzzles in comparative politics; (I) why do democracies not promote more redistributive policies than autocracies, and (II) why is there no net relationship between income inequality and democratization?


Introduction
Is democracy about redistribution? Many social scientists, at least, seem to think so. Some propose that redistribution is integral to the democracy concept, since political equality is impossible to ensure in economically inegalitarian settings. 1 But also, scholars conceptualizing democracy in narrower institutionalist terms expect democratic politics to centre on progressive redistribution. 2 Democracy is often considered a political system where the relatively poor 2 through their numbers at elections 2 influence decision-making and push through policies that tax the relatively rich and redistribute the resulting income to themselves. Accordingly, politics in real-world democracies should revolve around redistributive issues, and prominent contributions to comparative politics and political economy presuppose that progressive redistribution is vital for democratic voters. 3 For instance, drawing on this micro-assumption, Acemoglu and Robinson and Boix model how income inequality affects democratization prospects at the macro-level. 4 Since majority rule is expected to engender progressive redistribution, individuals support or oppose democracy according to how they benefit from such redistribution. These theoretical models have been influential in shaping the debate on inequality, redistribution, and regime change. But, empirical support remains surprisingly elusive, 5 and there are several plausible explanations for why this is so. 6 Adding to this literature, we empirically analyse the connection citizens make between redistribution and democracy. Specifically, we analyse whether individuals identify redistribution as an important characteristic of democracy and, if so, whether and under what conditions they consider redistribution to be among democracy's most essential aspects. Survey data show that although many individuals, globally, identify redistribution as an important characteristic, the vast majority of respondents do not include redistribution among the most important characteristics of democracy. We further investigate how contextual and individual factors relate to this notion. While system-level variables explain only a small share (328%) of the variance in our models, we find robust evidence that having lived under a communist regime and currently living in a democracy make individuals less likely to consider redistribution among the most important features of democracy. Further, individuals with more to gain from progressive redistribution (that is, little education and belonging to lower classes) are more likely to make the connection. However, we find no systematic difference between top-income and lowincome individuals, but some evidence that middle-income earners are more likely than low-income earners to consider redistribution among democracy's most important characteristics.

Redistribution, inequality, and democracy
Before presenting our empirical analysis, we discuss arguments and previous findings that motivate our study. Our micro-level empirical analysis may contribute to illuminate two quite puzzling macro-level patterns: the lack of a clear net effect of democracy on redistribution, and the missing relationship between income inequality and democratization. After reviewing these literatures, we survey relevant results on determinants of individuals' support for redistribution.

Democracy and redistribution
A large literature discusses how to define "democracy". While most scholars include elections and broad-based popular participation, 7 disagreement remains on what other aspects to include. Particularly, there is a debate on whether distribution of economic resources is within the conceptual boundaries of "democracy". The "mainstream position" is that democracy, analytically, only includes particular political institutions. 8 Yet, proponents of substantive democracy concepts argue that a minimum level of income is necessary for having the capacity and autonomy to exercise political rights properly, or that economic inequalities adversely impact on the functioning of democratic rights through generating unequal distributions of power resources. 9 The centrality of distribution of resources to many democracy scholars is exemplified by Coppedge et al. counting "Egalitarian Democracy" as one of six main democracy conceptions (alongside, for example, "Electoral" and "Liberal"). 10 According to such definitions, democracy is inherently about egalitarian distribution of resources and, in extension, progressive redistributive policies.
Nevertheless, even scholars operating with purely political-institutional democracy concepts propose strong empirical links between democracy and redistribution. Nineteenth-century political thinkers from Marx to Ricardo anticipated that the poorer classes would use their numbers to induce substantial redistribution of wealth under democracy. 11 Enfranchised poor workers and small-scale farmers would push for collectivization of industrial plants and progressive land reforms, or highly progressive taxation systems. Relatedly, democratization should often open up for well-organized leftist parties favouring redistributive social policies. 12 Politicians from wealthy groups have taken this seriously, using different strategies 2 including reforming electoral rules 13 2 to mitigate the "redistributionist" threat from franchise extensions to the poor masses. 14 The same implication follows from the Meltzer-Richard model. 15 The median voter 2 who is poorer than the country's average income level 2 has incentives to set a high tax rate accompanied by progressive redistribution. Franchise extensions are therefore associated with increased government size and progressive redistribution, as voting rights are extended to poorer groups. Redistribution following democratization may also come through increased taxation and spending on public services benefitting much of the population. 16 Despite these plausible arguments, the empirical evidence is surprisingly mixed, casting in doubt the notion that redistribution is a central aspect of democratic politics. Mulligan, Sala-I Martin, and Gill find no evidence that democracies promote more progressive redistributive policies than non-democracies. 17 This corresponds to the mixed results on how democracy affects income inequality. 18 Indeed, tests showing the clearest association between democracy and low inequality employ proxies measured in gross income (before taxation and redistribution). 19 Despite the strong argument associated with the Meltzer-Richard model, there is little evidence that democracy furthers progressive redistribution.
What could account for this gap? First, autocracies redistribute more than what many theories suggest. Certain autocracies have strong incentives to pursue redistributive policies. Various autocracies in Latin America, Eastern Europe, and Southeast Asia have provided extensive social policies, 20 and particularly, former communist regimes pursued redistributive social policies, 21 reflecting their ideological underpinnings but also, potentially, their relatively large and poorsupporting coalitions.
Second, democracies vary substantially in redistributive outcomes; 22 the Meltzer-Richard model may, simply, often provide an inaccurate picture of democratic politics. 23 Money matters also in democratic politics, and wealthy citizens may employ their resources to skew policy outcomes to their benefit, generating less redistribution. 24 The rich have more resources for lobbying activities, and legislative malapportionment (and other institutional peculiarities) may provide "a tool for some groups to preserve their economic interests". 25 More generally, established elites employ different strategies to successfully undermine the redistributive impact of democratization. 26 Finally, in many democracies the group alliances and prevalent issues diverge from those prescribed by the Meltzer-Richards model (poor vs rich; focus on redistributive issues). Ansell and Samuels argue that democracy has often been (effectively) demanded not mainly by poorer groups to ensure redistribution, but rather by new economic elites 2 emerging with industrialization 2 to protect their property against expropriation (by old autocratic elites). 27 Hence, democratic politics is often complex with multiple issue dimensions and class-cutting alliances, possibly explaining the weak relationship between democracy and redistribution. 28 Democratic politics may often not mainly revolve around redistributive issues after all.

Inequality and democratization
The above discussion has implications for whether (and how) income inequality affects democratization. Inequality matters in many political-economic models of democratization because elites and non-elites alike consider democracy to induce progressive redistribution: 29 democratic politics centres on the struggle between a richer minority and a poorer majority, leading to redistribution since the poorer majority sets policy. Since "each political regime has different redistributive consequences, every individual supports the political arrangement that maximizes his welfare". 30 Hence, elites prefer autocracy 2 where no such redistribution is expected 2 and the masses prefer democracy. Consequently, inequality affects democratization prospects through altering the gains and costs of democracy for both groups, and thus their willingness (and capabilities) to fight for or against regime change.
These political-economic models are among the most well-developed theoretical explanations in the democratization literature. 31 Yet, they are characterized by a "relative lack of empirical corroboration"; 32 tests on a net effect of income inequality have provided only non-robust results. 33 Possibly, the mechanisms proposed by Boix and Acemoglu and Robinson work only in specific contexts -for example, when opposition groups are sufficiently organized to solve collective action problems. 34 Yet, citizens in many countries might not consider democratic politics to mainly centre on redistribution. 35 Most citizens might disagree with egalitarian conceptualizations of democracy. Further, they might not even expect democratic processes to revolve around progressive redistribution, possibly influencing their demands on democratic incumbents and their political behaviour. Section 3 analyses whether citizens consider redistribution a central characteristic of democracy, but we first review a different literature informing our empirical specifications.

Support for redistribution
Few clear theoretical priors and no previous empirical modelling efforts exist on what factors determine the belief that redistribution is central to democracy. Yet, a vast literature studies preferences for redistribution, and we speculate in Section 3.3 that those favouring extensive redistribution 2 who according to the "Meltzer-Richard logic" should favour democracy 2 could expectedly more often count redistribution among democracy's most important characteristics.
Also, many studies on support for redistribution draw on the Meltzer-Richard model, which indicates that rising inequality induces more redistribution. 36 Finseraas finds that higher inequality indeed accompanies higher demand for redistribution, 37 and Cramer and Kaufman that it increases public disapproval of the existing wealth distribution. 38 Dallinger finds that support for redistribution decreases in income level and that "it is in the liberal world of welfare capitalism that the demand for redistribution is low despite high inequality". 39 Hence, welfare state institutions not only affect redistribution taking place, 40 but also individual support for redistribution. 41 Moreover, studies show lower support for redistribution in the United States than in Europe, 42 and having lived under a communist regime increases support for redistribution. 43 At the individual level, Alesina and Ferrara find that wealthier individuals support redistribution less than poorer, and that social mobility reduces support for redistribution. 44 Rehm shows that industry and occupational-level factors relating to risk of future income loss affect such support, 45 and Haggard et al. that unskilled and semi-skilled workers have stronger preferences for redistribution. 46 Luttmer finds a pattern of "racial group loyalty"; support for welfare is lower for prospective recipients from other racial groups. 47 Strong group identification, more generally, may weaken preferences for redistribution. 48 In Sections 3.2 and 3.3 we draw on the literature reviewed above when discussing our expectations and interpreting results on how contextual and individuallevel factors relate to considering redistribution essential to democracy. Before that, we present our measures and discuss overall patterns in the data.

Redistribution as essential characteristic of democracy?
Data on what the important characteristics of democracy are, come from the following World Values Survey (WVS) question: Many things may be desirable, but not all of them are essential characteristics of democracy. Please tell me for each of the following things how essential you think it is as a characteristic of democracy. Use this scale where 1 means "not at all an essential characteristic of democracy" and 10 means it definitely is "an essential characteristic of democracy": (i) Governments tax the rich and subsidize the poor. This allows identifying the extent to which individuals classify redistribution (item i) as an important characteristic of democracy. The first sentence makes clear that the question does not ask about the items' desirability. Social scientists trained in conceptual logic might interpret the question as one of defining and delimiting the democracy concept. Following classical concept structure, they would expectedly rate items tapping necessary conditions of democracy with high scores, and all others low. If so, they would rate item (i) high if they operate with a maximalist or egalitarian concept and low if they have a minimalist understanding. Further, they would rate aspects peripheral to most plausible definitions, such as (vii), (viii), and particularly (ii) and (v), low. However, most respondents rate items on, for example, "a prospering economy" fairly high. This could imply they operate with peculiar democracy definitions (or "acquiescence bias"; respondents score any item in the positive). 49 Or, many respondents may interpret the question differently. As psychologists have demonstrated, most people do not operate with clearly delimited (classical) concepts, 50 and experimental evidence suggests that beliefs about causal relationships strongly influence people's categorization decisions. 51 Thus, respondents may score an item high if they consider the feature an important part of democratic politics or a consequence of democracy. This could explain why 81.1% of respondents answer six or more when asked if they consider "a prospering economy" an important characteristic (mean score is 7.8).
Hence, we may not mainly study whether people have minimalist or broad/egalitarian democracy definitions in the classical concept sense, but whether they consider redistribution a central pattern of democratic politics. Still, we cannot directly test this, and therefore make clear below if making different assumptions about respondents'; understanding of the question critically affects the interpretation of our results.
The data include responses from WVS Wave 5 (2005Wave 5 ( -2007 Figure 1 shows the distribution, across all countries, for how important individuals think the various item-features are for democracy, one indicating "not at all essential" and 10 indicating "[definitely] essential characteristic". Although 23.5% of respondents identified progressive redistribution as definitely essential (mean score is 6.6), it is far from the feature most strongly associated with democracy. Of respondents, 54.6% provide 10-scores for "people choose their leaders in free elections" (mean ¼ 8.6) and 57.6% for "women have the same rights as men" (mean ¼ 8.6). The two latter are the characteristics most people clearly attribute to democracy, followed by "Civil rights protect people's liberty against oppression" and "People can change laws in referendums". This is unsurprising; a majority identifies the elements common to most extant democracy definitions as clearly central characteristics. However, also theoretically less obvious candidates, such as "the economy is prospering" and "criminals are severely punished" have far more 10-scores than redistribution (37.8 and 38.4% compared to 23.5%, and mean scores are 7.8, 7.6, and 6.6, respectively).
This pattern is not strongly contingent on the particular measure or sample used: first, the distribution of item (i) resembles that of (iv) "People receive state aid for unemployment", although (iv) has a higher mean and more 10-scores. Unemployment benefits have an insurance function for individuals wanting to mitigate risks of job loss, 53 but also contribute to redistribution from those paying taxes to those losing their wage incomes. Second, WVS Wave 6 (2010 -2014) contains the same question, although three items differ from Wave 5 (see Appendix). Appendix Figure A.1 shows the distribution across all (52) countries; the redistribution item distribution shows a striking resemblance to that for Wave 5. Only 20.1% of respondents provide 10-scores for "governments tax the rich to subsidize the poor", compared, for example, to 44.1% for "women have the same rights as men". The Wave 6 question contains another redistribution item: (vii) "The state makes people's incomes equal", where only 17.9% attribute 10scores. Independent of measure, redistribution is not widely considered among democracy's most central aspects.
We construct different measures to more comprehensively investigate whether redistribution is deemed among democracy's most central features. We mainly use measures based on Wave 5's item (i) 2 data on all our independent variables are available for the starting year of Wave 5 (2005) but not Wave 6 (2010). We present models adding Wave 6 data in the Appendix (Table A. 7), and most results are robust.
As using item scores directly is problematic for different reasons, our main measure is based on whether redistribution is a "prioritized" characteristic of democracy. First, "acquiescence bias" is one reason why survey methodologists often prefer ranking to rating scales. 54 It may be aggravated here, as democracy (to many) is a positively loaded term. 55 Remembering the skewed item-distributions in Figure 1, drawing clear information from a respondent scoring redistribution seven or eight is potentially problematic if she also scores most other items in the heterogeneous battery nine or 10. Second, citizens of particular countries could more often use extreme values when rating (any question); providing 10scores in Sweden may not be equivalent to providing 10-scores in Egypt. Such measurement equivalence issues 56 are alleviated when using measures based on ranking items. 57 Third, the items are ordinal -a 10-score is higher than a fivescore, but does not imply that it is twice as essential.
Although we discuss analysis using the direct redistribution-item scores (thus analysing whether respondents consider redistribution an important characteristic of democracy, despite the methodological issues), our main measure is a dummy scored one if the individual provides a redistribution-item score higher than the median-item score over the battery. It thereby captures individuals considering democracy to be more strongly associated with redistribution than numerous other characteristics. For respondents answering all 10 items, redistribution must score higher than the mean of the fifth-and sixth-ranked items; an individual could rank redistribution below the four items on elections, rights and referenda and still score one. For brevity, we label people scoring one "democratic redistributionists". Figure 2 (chart 1) shows the share of democratic redistributionists (SDR) by country.
SDR in the pooled sample is 18.9%. 58 Few respondents consider progressive taxation and redistribution to be among the most important features of democracy. Yet, SDR differs widely between countries. It is highest in Malaysia (42.6%), and no other country has SDR of more than 30%. In six of the 45 countries (Argentina, Bulgaria, Burkina Faso, Georgia, Romania, and Vietnam) SDR is below 10%. In Romania only 4.3% are democratic redistributionists.
An alternative measure uses pairwise comparisons with the other items: we construct nine dummy variables scored one if redistribution scores strictly higher than the relevant item. We then summarize these dummies, resulting in a variable ranging from zero to nine. We classify as "democratic redistributionists" individuals rating the redistribution item higher than (at least) five other items. Figure 2 (chart 2) shows the resulting SDR, and for another, "more restrictive", measure, scoring individuals that rated the redistribution item higher than seven other items (chart 3). The SDR and ranking of countries for the second measure closely resemble those for the median-based. But, the ordering of countries changes when applying the third (restrictive) measure. Malaysia still has the highest SDR (24.8%). However, Sweden now shows the lowest (0.6%), despite having an intermediate share for the other measures (about 20%). Similarly, Norway shows the second highest SDR according to the first two measures (about 30%), but SDR is only 3.3% for the third. Yet, this third measure is perhaps overtly strict, requiring individuals to consider redistribution more important to democracy than at least one item pertaining to elections and political rights.  Figure 2 (chart 4) also shows the share of individuals attributing 10-scores to redistribution. In three countries (Egypt, Jordan, and India 59 ) more than 50% do so (these countries also show relatively high shares for the other measures). Countries where few citizens identify redistribution as definitely important include Thailand, the United States, and Canada. Although less than 10% in these countries attribute 10-scores, they have close to (global) average SDR on our other measures. Thus, whether we analyse the correlates of citizens considering redistribution an important characteristic of democracy or among the most important could matter. We investigate this possibility below, but mainly focus on the latter specification.

3.2.
Under what contexts do people consider redistribution among the most important characteristics of democracy? Although our preferred measure identifies few democratic redistributionists, country-level and individual characteristics may still affect the probability of considering redistribution among the most important features of democracy. This section presents an initial discussion of country-level correlates of theoretical interest. Section 3.3 presents a more thorough analysis studying relevant individual and contextual factors simultaneously. We mainly rely on the median-based measure using data from WVS Wave 5, and independent variables measured in 2005.
Regime type could systematically relate to SDR. Citizens in dictatorships may know less about how democracy works. Consequently, they may over-or underestimate its progressive-redistributive capacities (here and directly below, we must assume respondents apply the "causal" and not the classical concept interpretation). Yet, citizens of democracies 2 if they have not undergone recent democratization 2 may also lack a proper contrast with which to compare the redistributive features of democracies. Hence, some democratic citizens might overlook that democracy could have progressive features, and rather respond to the question, for example, based on comparisons of their own modestly unequal society with some ideal egalitarian society.
Quite different mechanisms could also generate a regime type -SDR correlation: Acemoglu and Robinson and Boix highlight how anticipated demands for redistribution affect elites' incentives to fight against or concede democratization. 60 Where citizens strongly associate democracy with progressive redistribution 2 and potentially would vote for class-based parties and politicians promising redistribution, rather than vote according to other policy dimensions 2 elites have higher costs of conceding democratization. Hence, they may fight for the current autocracy and invest in repressive capacity to avoid redistribution. But, where citizens do not consider redistribution essential to democratic politics (which, in turn, could affect their political behaviour), elites may allow democratization, anticipating only minor redistribution costs. Countries with low SDR could therefore be exactly those that turn democratic, whereas those with higher SDR remain autocratic. Indeed, our 35 democracies have mean SDR of 16.9% whereas our 10 dictatorships have a mean of 19.9%. As shown in Appendix Table A.4, this difference is robust to controlling for other (potentially relevant) country-level variables in ordinary least squares (OLS) regressions.
Also, mean SDR is only 15.6% for the 12 countries with recent (post-1989) democratization, 4.3 percentage points below the average for the 10 (remaining) dictatorships. One could thus speculate that democratization is more easily "allowed" where citizens do not strongly anticipate the new regime to be associated with progressive redistribution. However, this interpretation faces two important issues. First, the difference is not robust to controlling for other country-level covariates. Second, SDR is not necessarily static. We lack the time series for investigating this more closely, but the beliefs citizens hold about democracy's characteristics could be endogenous to personal experiences. Low SDR in recently democratized countries may thus rather reflect disappointing experiences with democracy not delivering redistribution.
Likewise, some citizens, notably those previously ruled by communist regimes, have experiences with highly redistributive dictatorships, and where income inequality has increased post-democratization. 61 Indeed, mean SDR was 8.5% in the five countries with communist regime experience, 62 which is less than half the share in other countries (18.7%). This could reflect "learning mechanisms"; citizens in ex-communist countries have experienced democratization processes not being followed by increased redistribution. The cross-country relationship between communist regime experience and SDR is robust to controlling for different factors such as income level, inequality, recent democratization, and regime type. Below, we analyse these relationships more carefully, running multi-level models incorporating individual-level characteristics.
The different other country-level variables that we tested are not systematically related to SDR. In brief, income (log Purchasing Power Parity-adjusted gross domestic product (GDP)/capita from Maddison) and SDR correlate only weakly (0.11). 63 SDR is also unrelated to country size (0.01 correlation with log population from Maddison), and to ethnic fractionalization (from Alesina et al.; 0.08 correlation). 64 Finally, the data show basically no country-level correlation between SDR and Gini coefficients measuring income inequality (both market and disposable; from Solt), 65 and the income redistribution measure from Solt has a negative, but weak, correlation with SDR (20.04). None of the above variables are robust predictors of SDR in multivariate OLS regressions either (see Appendix  Table A.4).

Multi-level models
The multi-level regressions employ as dependent variable our measure scored one if the individual rates redistribution above the median item. Some models only include the statistically significant country-level variables from the Appendix OLS regressions, regime type and communist history. More extensive models add income level, disposable income inequality, redistribution, population size, ethnic fractionalization, and recent democratic transition.
Not much guidance exists on what factors determine whether individuals consider redistribution essential to democracy. Yet, we reviewed the literature on factors determining preferences for redistribution. According to the Meltzer-Richard logic, people preferring redistribution should also favour democracy the strongest. This does not guarantee they are democratic redistributionists, but well-documented cognitive mechanisms such as "wishful thinking" (preferences determine beliefs) 66 and "self-serving biases" (self-interest determines fairness norms) 67 may contribute in this direction. Individuals preferring redistribution might be more "optimistic" on redistribution being among democracy's essential characteristics. We acknowledge that this is speculation (respondents could, for example, associate democracy and redistribution, but normatively oppose both), but there is no other clear theoretical model for identifying individual-level determinants of our dependent variable, and we thus take this speculation to the data.
Personal income, educational background, age, and social class relate to individual-level differences in attitudes towards redistribution. We therefore suspect them to also relate to our dependent variable. For instance, since low-income earners should benefit from redistribution under democracy, they may be more susceptible to view democracy in this light. Similarly, those with little education likely have (relatively) lower expected future market income, controlling for current income, and may thus more likely be democratic redistributionists. Drawing on WVS data, we thus include educational level, income, and subjective social class, alongside age and gender. 68 Table 1 shows random intercept logistic regressions. 69 The intra-class correlations indicate that only 328% of the variance is due to country-level factors; the contextual level thus explains relatively little, and individual differences explain most of the variance. M12M2b only include individual-level variables. M1 includes age, gender, class, and education, whereas M2 adds income, modestly reducing the sample size. M2b adds two items capturing preferences for redistribution.
We expected poor citizens to more often consider redistribution among democracy's most important characteristics. However, middle-income citizens are more likely to agree with this assessment than poorer-income citizens. Acemoglu et al., for instance, note that democratization, in practice, often transfers power to middle-income rather than poor groups, and that redistributive policies therefore often favour the former but not the latter. 70 This might contribute to explaining why middle-income citizens more often are democratic redistributionists. Although the point estimates suggest that also the rich are more likely to be democratic redistributionist, there is no statistically significant difference between the richest and the poorest. Figure 3 visualizes the "effect" of income (chart 1), displaying predicted probabilities based on M2 for male respondents having completed secondary education and with all other variables at their mean or median (this applies for all simulations discussed; Figure A.2 shows predicted probabilities based on M5). Each circle represents the predicted probability (mean prediction is marked by a dense circle) from a set of simulated model estimates. 71 Notably, the substantive differences in mean predicted probabilities are fairly small (below 2%). High-income Table 1. Logistic multi-level regressions on "democratic redistributionist" dummy.
Individual level Age 20.00 * * *  Notes: Random intercept logistic regression on (median-based) democratic redistributionist dummy. Standard errors in parentheses. All independent variables are measured in 2005. The reference categories for class is "upper class", whereas "low" is reference category for education and income. * * * p , 0.001. * * p , 0.01. * p , 0.05.
citizens are, on average, predicted to resemble middle-income citizens, but there is more uncertainty associated with predicting the former. While many politicaleconomic models treat income as the crucial determinant of whether individuals gain from (and thus support) democratization, it does not critically explain whether individuals consider redistribution among democracy's most important characteristics.
The other individual-level determinants follow our expectations more closely: 72 Those with the least education are more likely to consider redistribution among democracy's central features. Figure 3 (chart 2) shows the mean predicted probability of being democratic redistributionist is 21.3% for the lowest education level, and 16.0% for the highest. This difference is significant at 1%. Moreover, individuals classifying themselves as belonging to the lower class are more often democratic redistributionists than upper-class respondents. Figure 3 (chart 3) shows a difference in mean predicted probability between these groups of 3.1%, and middle-class respondents are closer to upper-class respondents (ca. 1% difference).
The individual-level results (with some exceptions for income) remain fairly stable when adding country-level variables (M32M5). Thus, when controlling for, for example, GDP/capita, regime type, and inequality, lower-class respondents and those with little education are more often "democratic redistributionists". These results are robust to using our pairwise comparisons measures (Appendix Table A.6), and to including Wave 6 data (Appendix Table A.7). Interestingly, results hold also when including two WVS questions tapping preferences for redistribution in M2b. Hence, the results for education and social class are not solely driven by these groups being more positive to redistribution. These results indirectly serve as a (discriminant) validation test; our democratic redistributionist variable is not simply capturing some underlying preference for redistribution. As expected, respondents thinking that incomes should be made more equal are more often democratic redistributionists (significant at 1%), and so are respondents thinking that the government should take more responsibility to provide for people.
Returning to country characteristics, M3 retains the negative coefficients for democracy and communist regime experience from the cross-country regressions. However, only communist experience is significant (0.1%). Citizens living in countries with communist histories are systematically less likely to consider redistribution among democracy's most important characteristics.
This relationship could mainly be driven by older people with extensive firsthand knowledge of the redistributive schemes these regimes put up. 73 To test this, M3b includes a multiplicative interaction term (age * communist regime history). The estimate turns out statistically insignificant, and the sign actually indicates a larger difference (between countries with and without a communist past) for younger citizens, many of whom may not even remember life under communism (although they have likely heard many stories about it). Figure 4 displays predicted probabilities for this interaction effect. Each line represents the predicted probability based on a simulated set of model estimates. 74 While the difference is larger for younger citizens, it is significant (1%) for all ages. The probability of being democratic redistributionist is typically lower than 0.1 for those living in (ex)-communist countries, and just below 0.2 in non-communist countries. M4 includes all country-level variables. Both democracy and communist experience show negative significant effects (1%). The other country-variables are insignificant. 75 Figure 5 again shows the predicted probabilities of being democratic redistributionist in countries with and without a communist past, but now for citizens of current democracies and dictatorships. Our hypothetical individual living under democracy has a 9.4 percentage point lower probability (significant at 1%) of being democratic redistributionist if his country has a communist history. For current dictatorships, the difference is 14.3 percentage points (significant at 1%).
To investigate more nuanced patterns, M4b employs a tripartite regime measure (democracy, anocracy, dictatorship) from Epstein et al., based on the Polity Index. 76 These results indicate that citizens in anocracies are significantly more likely to be democratic redistributionists than autocratic citizens in autocracies, and there is no clear difference between citizens in democratic and autocratic countries. The anocracy result is further strengthened when including WVS Wave 6 (Appendix Table A.7).
Regarding other nuances, one theoretically interesting interaction should be explored: there is (non-robust) evidence that regime type is correlated with being a democratic redistributionist, but that society-level income inequality or actual redistribution is not. Yet, the regime type-democratic redistributionism correlation could vary over inequality and redistribution. If learning mechanisms are important for whether democracy is considered to entail redistribution, we would expect numerous democratic redistributionists in democracies with much redistribution, such as the Scandinavian countries, but fewer in democracies without progressive-redistributive policies. Moreover, a refinement of Acemoglu and Robinson's argument on income inequality and incentives for elites to concede democratization would indicate that elites only dare concede democratization in unequal societies if the future electorate do not consider redistribution essential to democratic politics. Democracy may thus go together with democratic redistributionism after all, but only in egalitarian countries. Hence, M4c adds interaction terms (democracy * inequality; democracy * redistribution). However, we find no evidence of any interaction between democracy and redistribution/income inequality.
Due to missing data on some country-variables, M424c only include 32 countries. Therefore, M5 omits redistribution and income inequality to include 40 countries. The dichotomous Democracy-Dictatorship measure is now only Figure 4. Probability of being "democratic redistributionist", by age group, in countries with and without communist history -based on Model 3b. Notes: Simulations of predicted probabilities are shown for a male individual with an education level of 4, an income scale of 4, and subjective class of 3, living in a democracy with no recent democratization history and no communist past. All other variables were held at their mean. weakly significant (10%). Also, the anocracy result from M4b is sensitive (see M5b). Hence, we do not have robust evidence that regime type relates to our dependent variable. In contrast, communist experience is robustly associated with considering redistribution not to be among democracy's key characteristics.
We tested equivalents to Model 4, but using our alternative dependent variable operationalizations, and report them in Table A.6. Model I is identical to M4 in Table 1; Model II uses the weak pairwise comparisons measure (that is, redistribution higher than five other items); Model III employs the stricter pairwise comparisons measure (seven other items). These models, especially Model II show fairly similar results to the baseline. We also included data from Wave 6, expanding the sample to more than 100,000 respondents for 71 countries. Table A.7 shows models for Waves 5 and 6 combined, using a "median-based" measure adjusted for the seven common items of both waves. 77 Again, the education and class results are robust. Similarly, communist regime experience and currently living under democracy make citizens less likely to consider redistribution among democracy's most important features. Above we measured whether citizens consider redistribution among democracy's most important characteristics. Thus, individuals who operate with extremely broad democracy concepts or associate democratic politics with numerous outcomes need not be classified as democratic redistributionists. To check the correlates of considering redistribution one important feature of democracy, we tested a linear multi-level model using the redistribution item score (1210) directly. Model IV, Table A.6 2 otherwise similar to M4 in Table 1 2 shows some differences to our main analysis, but there are many similarities. Regarding the similarities, the negative associations with communist history (significant at 1%) and democracy (10%) hold up, and there is no systematic association with, for example, income level or redistribution. Although slightly weakened, higher education is negatively associated also with this dependent variable. Yet, there are also differences. For instance, not only lower-, but also middle-class members are more likely than upper-class members to associate democracy with redistribution. 78 Since the item-score is ordinal, we also tested a hierarchical logit model (Model V, Table A.6) employing a dummy scored one for redistribution item-scores of six or more. Some results are sensitive to this specification change; for example, the point estimates still suggest the least-educated citizens are more likely to connect democracy and redistribution, but the education coefficients are no longer significant at 5%. Nonetheless, people living in (ex-)communist countries and democracies remain less likely to associate democracy with redistribution, and lower-class respondents are more likely to do so than upper-class respondents.

Conclusions
While the available data suggest that several citizens consider redistribution a fairly important feature of democracy, they also make clear that most citizens around the world do not count redistribution among democracy's most important characteristics. This is an interesting and for many perhaps a surprising result given that prominent theoretical arguments and numerous social scientists link democracy and redistribution. Some, operating with egalitarian or "maximalist" democracy definitions, make a conceptual link. Others expect an empirical link. Even if democracy is defined minimalistically, democratic politics is often theorized to centre on redistribution from the rich to the poor, as for instance in the widely employed Meltzer-Richard model.
We have studied whether ordinary citizens make such a link, focusing on whether redistribution is considered among democracy's most important characteristics by using WVS data. Most citizens 2 in rich or poor countries, or in egalitarian or inegalitarian societies 2 do not consider progressive redistribution among democracy's most important features. This is particularly the case for upper-and middle-class citizens, educated citizens, citizens of countries with communist regime experience (which tended to pursue quite progressive redistributive policies), and citizens of current democracies (although this is less robust).
That people currently living under dictatorship (may) have a stronger tendency to associate democracy with redistribution could be due to them not having learned the "disappointing reality" that democracy, in practice, often does not induce massive redistribution as predicted, for example, by many political-economy models. Still, these models also highlight that democratization is costly for rich elites in countries where democratic politics will be organized around redistributive issues. Hence, it may paradoxically be the societies where democracy is anticipated to yield the strongest redistribution that remain autocratic, due to elites in these countries fighting hard to preserve the old regime, as suggested by, for example, Boix. 79 Without time series data, however, it is difficult to separate between the two explanations, and we hope that future data collection will allow investigating these intriguing hypotheses more thoroughly.