Goodbye to Meltzer-Richard: Testing major theories of redistribution

Abstract

We re-analyse four major explanations of redistribution including the Meltzer-Richard model, power resources theory, Iversen-Soskice’s political institutions explanation, and Lupu and Ponstusson’s skewness theory. For each of these, we reconsider the causal chain and test their assumptions using a comprehensive, original dataset on working-age income-inequality consisting of 589 country-years for affluent democracies in the period 1963–2019. We find that partisan governments are directly related to redistribution and have a strong effect on the generosity of social policy. Lupu and Pontusson’s skew measure has no effect on redistribution in models with controls but does have a positive effect on generosity of social policy. Finally, we find that the mean-to-median income ratio has a consistent, negative, and highly significant effect on redistribution, directly refuting the very premise of the Meltzer-Richard model.

Keywords

redistribution political economy social policy Europe welfare state power resource theory Meltzer-Richard unions inequality

Introduction

Experienced income inequality has risen in virtually all post-industrial societies. This trend can be attributed both to increasing market income inequality and to stagnant or declining redistribution by the state.¹ While differences persist across the established welfare-state types, this is a common trend to virtually all post-industrial nations in Europe, North America and elsewhere.

What does the literature have to say about this dynamic? The redistribution scholarship offers various theoretical explanations for the level of redistribution. The most prominent of these has been the Meltzer-Richard (MR) model (1981) which has informed many subsequent works on redistribution for four decades now. The model, which has an unmistakable intuitive appeal, lacks empirical support. In fact, studies have consistently failed to find evidence in support of MR’s main prediction: that redistribution will increase with income inequality. It is therefore bewildering that MR is continuously revisited as a point of departure for new studies of redistribution.

In this article, we examine MR along with three additional theories and offer a systematic comparison of the explanatory power of these theories of redistribution. To be clear, our concern is with actual redistribution rather than with public attitudes. We begin the next section by reviewing various explanations for redistribution and then we develop theory-specific models based on their ‘master variables’. We then operationalize and test the models using an original dataset consisting of 589 country-years. After presenting our main findings, we conclude this article by drawing out the main theoretical implications.

Literature review and theory

In this study, we conceptualize and measure redistribution as the difference between pre- and post-tax and transfer income inequality as a percentage of pre-inequality. Redistribution in this sense depends on the social risks affecting market incomes, or the level of need, the political aptitude to respond to these needs, and the state’s concrete policy effort to smooth income disparities. There is agreement that redistribution is closely connected to the magnitude of welfare state effort, including both social insurance and non-contributory social assistance (Atkinson, 2015; Bradley et al., 2003; Korpi and Palme, 1998). This relationship is not perfect, that is, the magnitude of effort and structure of spending both shape the extent of redistribution (Jensen and Van Kersbergen, 2017). If one adds the distributive impact of social services, redistribution is enhanced (Marical et al., 2006). In Wang and Caminada’s (2011) data for 19 post-industrial democracies, only 15% of redistribution happened through taxes, including mandatory payroll taxes, and 85% through transfers. Their data spanned the entire population, so pensions were included. Our dataset includes only working-age individuals, so we focus on unemployment and sickness benefits. Obviously, if taxes and social expenditures are low, not much income can be redistributed. On the other hand, as noted, not all social expenditures are allocated in an equally redistributive manner, which is why it is important to measure social rights directly. Since such measures have only become available in the past 20 years, earlier work on redistribution used social expenditure either as a proxy or as a measure of government effort.

Arguably the most influential theory in this literature is the model developed by Meltzer and Richard (1981). The MR model argues that a greater difference between the mean and the median income will increase the magnitude of government, and consequently redistribution, because the median voter is decisive under majority rule. They assume that the median voter will support further redistribution until its benefits are outweighed by the costs of lower per-capita income due to work disincentives. The model is simple, though as Iversen and Goplerud (2018) point out, it rests on a number of additional assumptions: a proportional tax, a flat-rate benefit, a balanced budget, efficiency-costs of taxation, full voter turnout, and the absence of other salient issues in politics. Meltzer and Richard further acknowledged that their model pre-supposes voters are well-informed on the efficiency costs of taxation. Here lies an implicit assumption that public opinion shapes public policy. As Gilens (2012) and Witko et al. (2021) have shown for the United States, though, preferences of business and the rich matter more for policy than the preferences of labour and lower income groups. Although Meltzer and Richard never test their theory, the assertion that higher inequality will result in more redistribution has become the starting point for many studies.

Moene and Wallerstein (2003) extend the MR model by adding the degree to which social policies mix insurance with pure redistribution. They find that inequality lowers spending on policies that provide support for unexpected income loss. In contrast, they find no relation between income-inequality and welfare programmes that benefit recipients regardless of their employment status. While this seems to negate the MR premise, they argue that their null-findings regarding inequality and spending are due to the absence of redistributive programmes open only to active labour market participants.

Lupu and Pontusson (2011) propose a theory similarly based on exclusively materialistic political competition, but go beyond the magnitude of inequality and focus on its structure. They propose that in the absence of cross-cutting ethnic cleavages, the extent of social spending is determined by the ‘skew’ of the market income distribution: the distance between the middle and the poor in relation to the distance between the middle and the rich. The intuition behind this model is that greater income proximity among classes will lead to an affinity with respect to their redistributive expectations. If the distance between the middle-income and the poor earnings ratio (50:10) is smaller than the distance between the middle-income and the rich (90:50), the middle class will have greater affinity to the poor and support redistribution, and vice-versa. In their original analysis, skew indeed has a significant effect on both redistribution and social spending.

Other studies seem to reach different conclusions. Schwabisch et al. (2006) examine the impact of the 50:10 and the 90:50 ratios of market household income on spending separately. They find that while the 50:10 ratio has a small positive effect, the 90:50 ratio has a far larger and strongly negative impact on social expenditures. This asymmetry, they argue, indicates that as the rich become more distant from other social classes, they begin to opt out of social programmes in favour of private alternatives, which negatively effects social spending. Elkjær and Iversen (2023) examine the structure of inequality with a model that shares MR’s assumption of income-based voting but diverges with respect to the composition of the electorate. Rather than an ordinally increasing distribution which privileges the decisive median-voter, their model consists of three distinct income groups such that no one class can fully determine the level of redistribution. This opens the door to a coalitional prism centred around the middle class’s preferred partner. Drawing on Rehm (2016), they propose that the middle class’s redistributive preferences partially hinge on a desire to insure against a given level of risk due to a loss of income. Since the stakes for such a scenario increase with distance between the middle class and the poor, they expect bottom heavy income inequality to result in more redistribution.

This three-class approach brings to the fore elements beyond mere income. Political institutions have figured prominently in many studies of redistribution. Persson and Tabellini (2003) for instance, argued that single-member districts induce geographically concentrated spending whereas PR leads to more universalistic spending, which is more redistributive. More to the point, in an earlier iteration of their three-class model, Iversen and Soskice (2006) argued that PR electoral systems redistribute more than majoritarian systems because they favour more frequent centre-left governments. Their argument assumes that partisan politics matters, but that the propensity for certain government coalitions varies as a function of the electoral system. While in majoritarian systems both parties have centrist platforms, they cannot credibly commit to sticking to them following the election. For middle-class voters, the consequences of post-election deviations are asymmetrical: a deviation under a centre-right government (too little taxation and spending) will result in insufficient transfers from the rich to the middle-class and the poor. A deviation to the left (too much taxation and spending) could result in transferring income from the rich and middle-class to the poor. As result, middle-class voters will lean toward the centre-right. Such credibility concerns do not exist in PR systems, where middle-class parties can form and dissolve coalitions after the elections. Not fearing excessive redistribution, middle-class voters can maximize their utility by allying with the working-class party and imposing the costs of redistribution solely on the rich. Their analysis indeed finds that PR systems are negatively associated with right-leaning governments and that right-leaning governments are negatively associated with redistribution. Döring and Manow (2017) confirm the association between majoritarian systems and conservative government but show that PR systems present a more nuanced picture. They also show that the voting behaviour of the middle class is only one possible explanation, with electoral geography being another one. Our current focus is on the association between electoral system and partisan government, rather than the mechanisms that connect the two.

We would like to propose a fourth alternative which is a variant of Power Resources Theory (PRT) focused on partisan politics. We argue that welfare state generosity is tied to partisan politics, since benefit structures and institutions are built up and entrenched by successive governments over long periods of time. The historical development of the welfare state demonstrates that incumbent social-democratic, Christian-democratic and liberal parties, left distinct legacies with respect to social legislation and redistributive patterns (Esping-Andersen, 1990; Huber and Stephens, 2001). Social-democratic parties, which have long been strong supporters of redistribution and egalitarianism, are associated with lower levels of poverty and inequality (Brady, 2009; Huber and Stephens, 2014; Nelson, 2012). While Christian-democratic parties too share historical links with the welfare state, recent accounts found that Christian-democratic governments were in-fact negatively associated with redistribution (Bradley et al., 2003; Huber and Stephens, 2014).

Beyond the four theories presented above, many additional factors have been proposed to attenuate or accelerate redistribution. Immergut’s (1992) veto-points framework draws attention to how institutional structures could delay social policy expansion at the height of the golden age. Federalism, presidentialism, bicameralism, and referenda all provide multiple access points for opponents of redistributive policy to slow down welfare expansion. By the time of retrenchment, the relationship grew more complex as the same mechanisms were now used to block cuts to popular welfare policies (Huber and Stephens, 2015; Pierson, 2001), but also facilitated drift, which reduced policy efficacy (Hacker, 2005).

Ethnic, linguistic, and racial diversity have long been acknowledged as social cleavages that can cut across class lines. This may carry negative implications for distributive politics in so far as it reduces solidarity among the would-be constituencies of redistributive parties. A decline in social solidarity may also have an indirect effect on redistribution, by harming unionization rates (Lee, 2005; Stephens, 1979). Alesina and Glaser (2004) showed that racial and ethnic fractionalization had a significant negative effect on social spending in the United States and Europe while Rueda (2018) shows that support for redistribution declines earlier in the income distribution in more heterogenous settings. Larsen and Harell (2023) shed light on the micro-foundations linking the presence of minorities to redistribution by leveraging variation in redistributive attitudes and perceptions of minorities across European and North American countries. Desmet et al. (2009) and Baldwin and Huber (2010) both add to this literature by demonstrating that the effect of social heterogeneity on redistribution or public-goods provision is quite sensitive to different measures of diversity. There is nevertheless a scarcity of studies on the long-term effects of racial and ethnic fractionalization on inequality because there are no time series data comparable across countries. Studies in this vein commonly rely on cross-sectional data or use immigration as a proxy for diversity.

Voter turnout has also figured prominently in many studies of redistribution. Given the negative association between income and voting, Kenworthy and Pontusson (2005) argued that turnout is simply a proxy for low-income mobilization, and thereby conditions the government responsiveness to market inequality. Based on the negative association between income and voting, voter turnout can be considered as a measure of PRT. Some researchers have measured the skew in turnout and found a negative effect on redistribution (Mahler et al., 2013), while others found that institutions associated with higher turnout increased support for left policies (Bechtel et al., 2016).

Since we intend to test the proposed theories of redistribution, it is incumbent on us to elaborate what we consider to be their essential elements and corresponding hypotheses. Beginning with MR, the essential element is that the difference between the mean and median income will be positively associated with redistribution. The causal link is that as the distance increases, the median-voter’s policy preferences should change leading to a policy response and subsequently more redistribution. An increase in redistribution due to need/risk rather than policy change will not support the MR intuition. To account for this, we include measures of social risk in our analysis. Unemployment and single-mother household rates are expected to have a positive effect on redistribution, and employment a negative one, irrespective of policy change.

PRT argues that strong unions and associated labour market configurations will result in low levels of wage dispersion and that strong unions and left government will result in more redistributive public policy. While these two are considered PRT’s ‘master variables’, it is interesting to note that the theory holds diametrically opposite predictions concerning market-inequality from MR.² Whereas MR claims that greater market inequality will lead to more redistribution, PRT expects that they will be negatively associated because economic inequality is related to political inequality unless counterbalanced by strong organization among the lower classes. Lupu and Pontusson’s model similarly incorporates wage-disparity as a quasi-MR measure, yet they do not present a clear and testable related hypothesis as they do with their other ‘master variable’, skew.

Iversen and Soskice’s theory and PRT share a common appreciation of left governments. For the former, left governments serve as an important mediating mechanism responsible for the positive association between redistribution and its other master variable, PR systems. While MR never mentions government partisanship, we argue that it is not incompatible with the model and thereby does not negate it.

Less disputed determinants include voter turnout which Iversen and Soskice, MR, and PRT, among others, all argue should have a positive association with redistribution. Arguably, based on Kenworthy and Pontusson’s (2005) argument that turnout is tantamount to the mobilization of low-income workers, one can treat voter turnout as an aspect of power resources. Nevertheless, the theoretical consensus regarding turnout means that it does not discriminate between the tested theories. Similarly, the causal power of veto points does not negate any of the theories above but rather complements them as a moderating element. For instance, it explains why similar partisan configurations fail to produce equivalent redistributive measures (Immergut, 1992). Because this effect has served to both stall new social policies and protect extant policies, PRT does not have a clear directional expectation from veto points.

We take a step-wise approach to the analysis of redistribution. We begin by analysing the relationship of the master independent variables in the various theories to redistribution, first by themselves and then with the control variables. Since all the theories assume that their master variables work through policy, we then add policy as an independent variable. All of the above hypotheses are summarized in Table 1.

Table 1.

Variable definitions, sources, and hypotheses.

	Definition	Original data source	Expectation per theory
Dependent variables
Redistribution	$(\frac{Market income Gini - disposable income Gini}{Market income Gini}) \times 100$	LIS (2023), Eurostat (2021), and OECD (2021b)	PRT	MR	Institutions	Skewness
Nonaged welfare state generosity	Index of generosity of sickness pay leave and unemployment benefits combining replacement rates, qualifying conditions, and duration	Scruggs and Tafoya (2022)*	PRT	MR	Institutions	Skewness
Independent variables
Left government	Seats of leftist parties as proportion of the seats of all governing parties, cumulative from 1945 to date of observation	Brady et al. (2021)*	+		+
Union density	Net union membership as a percentage of employed wage and salary earners	Visser (2019)*	+
Mean to median ratio	Mean to median ratio of household income in the working-age population. Missing years interpolated	LIS (2023) and Eurostat (2021)	[−]	+
Proportional representation	Index where 0 = single member district, plurality, 1 = mixed district, and 2 = proportional representation	Brady et al. (2021)*			+
Skew	Ratio of the 90-50 wage ratio, to the 50-10 wage ratio	OECD (2021a)*				+
Wage dispersion	Ratio of gross earnings of a full-time employee at the 90th percentile to that received by an employee at the 10th percentile	OECD (2021a)*	[−]	+		−/+
Controls
Voter turnout	Votes cast in the most recent election as a percent of registered voters	Brady et al. (2021)*	+	+	+	+
Veto points	Additive index of presidentialism, strong bicameralism, federalism, and regular use of referenda	Brady et al. (2021)*	−/+	−/+	−/+	−/+
Trade openness	Sum of exports and imports as a percentage of GDP.	Feenstra et al. (2015)*	−/+	−/+	−/+	−/+
Immigration	International migrant stock as a percentage of the population reported quinquennially. Missing years interpolated	World Bank (2021)*	−	−	−	−
Unemployment rate	The number of unemployed persons as a percentage of the civilian labour force	OECD (2021c)*	+	+	+	+
% Children in single mother households	Children living in single-mother households, as a percentage of total children. Missing years interpolated	LIS (2023) and Eurostat (2021)	+	+	+	+
Employment	Civilian employment as a percent of total working-age population	OECD (2021c)*	−	−	−	−

*Available in Brady et al. (2021).

Data and measurement

Dependent variables

Table 1 describes the measures we use in this analysis. Our main dependent variable, relative redistribution is based on the market and disposable Gini indices for working-aged households. Focusing on the working-age population (those aged 18 to 64) is important, as populations outside of this interval may distort our estimates of redistribution. For instance, countries with generous public pension systems will greatly lower retirees’ market income and inflate redistribution. Since we cannot identify which household members are the ‘head’ of their household, we eliminate all households with elderly members. This problem does not apply to children, so we do not eliminate households based on their presence.

Market income is defined as all income from labour (wages and salaries as well as self-employment income), capital (financial interest and dividends and real estate income), and private transfers (inter-household transfers and transfers from non-profit institutions). Although the LIS harmonization process is very thorough, we included two unique dummies to account for cases in which the definition of market income still varied: one for when market income consisted of pre-transfer but post-tax income (50 observations), and the other indicating a mixed case where market income was pre-transfer but combined both pre- and post-tax data (12 observations). Disposable household income includes market income plus all public transfers and less all direct taxes.³ Following LIS convention, market income is bottom-coded at zero while disposable income is top and bottom coded via the interquartile range as $Q_{3} \times {(\frac{Q_{3}}{Q_{1}})}^{3}$ and $\frac{Q_{1}}{{(\frac{Q_{3}}{Q_{1}})}^{3}}$ , respectively. In the case of both market and disposable income, the standard ILO equivalency scale is used for both LIS and SILC-Eurostat data, in which a household’s income is divided by the square root of the number of household members $(\frac{H o u s e h o l d I n c o m e}{\sqrt{# H o u s e h o l d M e m b e r s}})$ .

We constructed our initial income measures using LIS microdata which yielded 465 observations. SILC microdata provided an additional 71 country-years which were not available through LIS. At this point, the data skewed heavily to the post-2000 period. We then added 53 final observations based on OECD estimates for disposable and market income inequality which helped improve (though not resolve) the periodic imbalance. The SILC/LIS and OECD measures were highly correlated (.92 and .94 for market and disposable Gini, respectively).

Our second dependent variable is welfare state generosity for the working-age population. We use the index of sickness and unemployment benefits from Scruggs and Tafoya’s (2022) Comparative Welfare Entitlements Project (CWEP). The index combines measures of replacement rates for the first six months, qualifying conditions, and duration; higher values indicate greater generosity.

Independent variables

Most of the independent variables in our analysis come from the Comparative Welfare States Dataset (CWSD) and from a variety of original sources. The full list of variables, along with their definitions, sources (and related hypotheses) are described in Table 1. When needed, we used existing data to create measures that adequately operationalized the theoretical arguments which we test. For instance, our MR measure was produced by estimating the mean and the median of pre-tax and transfer household income and then dividing the former by the latter. While we believe that the majority of our measures are self-explanatory, we would like to briefly clarify a few of our modeling choices. ‘Left government’, for instance, was operationalized using a cumulative rather than temporal measure of left-incumbency, as the effects of party incumbency on institutions build up over the long term (Huber and Stephens, 2001). Among our controls, we operationalized globalization via trade openness and used the international migrant stock within the total population as a proxy for diversity.

Overall, we have collected data for 22 countries for the period 1969 to 2019. The countries are Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Luxemburg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom, and the United States. Certain variables are not available for the entire series, so the number of country-years slightly varies across models. There are very few observations before 1980 and most of the observations pertain to the post-2000 period.

Statistical estimation

Pooled time-series data present special challenges for the statistical analyst. The non-independence of observations in pooled time-series produces errors that are (1) serially correlated, (2) cross-sectionally heteroskedastic, (3) often correlated across units due to common shocks, and (4) often autocorrelated and heteroskedastic simultaneously. We handle serial correlation by correcting for first order auto-regressiveness rather than with a lagged dependent variable. Beck and Katz (2004, 2011) have shown that correcting for first order auto-regressiveness (AR1 models) include a lagged dependent variable on the right-hand side of the equation. This statistical setup, PCSE and AR1 corrections, is known as Prais-Winsten estimations. It deals with the problem of serial correlation without, as our results show, suppressing the power of independent variables.

Gaps in our data create an additional challenge, since Prais-Winsten estimations incorporate the dependent variable at t_-1 on the right-hand side of the equation. We thereby use Vernby and Lindgren’s (2009) dvgreg package,⁴ which, following Iversen and Soskice’s (2006) lead, developed a method to deal with gaps in the dependent variable. Dvgreg is specifically designed for dynamic panel data with gaps in the dependent variable but nearly complete data on the independent variables. For each gap, it generates estimates for the DV at t_-1 based on its previously observed values, thus making it possible to use AR1 corrections. Instead of using panel corrected standard errors, Vernby and Lindgren (2009) address heteroskedasticity via weighted least squares. They argue that their Monte-Carlo experiments suggest that this method produces accurate estimates and standard errors given reasonably high ρ and R² values and absent of prolonged gaps. Vernby and Lindgren illustrate their statistical package with Bradley et al.’s (2003) data on redistribution comprising 61 observations, far less than the current 504 country-years.

In Online Appendix B, we test the robustness of our findings first via random-effects models and then by including cubic time-polynomial terms. In Appendix C, we compare different estimators and find evidence of serial correlation in our data which would render fixed-effects models inappropriate for this inquiry (Plümper and Troeger, 2019). Fixed-effects or country dummies are further incompatible with our data because the variation in redistribution primarily manifests between countries and not through time. The R² when regressing redistribution on country-dummies is .81, compared to .04 when regressed on year-dummies. The ANOVA test in appendix Table A2 similarly demonstrates this with countries accounting for over 20 times the amount of variance explained by years. Given that the vast majority of the variation is between countries, using simple country fixed effects will attribute the majority of variation in distributive outcomes to country differences and drown out important predictors that correlate with shifts in the level of redistribution.

We hypothesize that most of our causes operate over the long term and changes in the dependent variables occur gradually, a case of cumulative causes in Pierson’s (2003) typology of causes and effects. Thus, it is appropriate to measure the dependent and independent variables as levels. Moreover, in almost all pooled time-series studies of the determinants of inequality, regardless of whether it is measured by wage dispersion, household-income Gini, poverty levels, or top income shares, the DV is measured as a level.⁵ In order to test for sluggish effects of some of our variables, we performed a second robustness check by re-estimating our models with linear, squared, and cubed time terms. Much like splines, these polynomial models help smooth non-linear and gradual outcomes, with the added benefit of controlling for autocorrelation.

Results

Table 2 displays the results of our tests of the various theories of redistribution one by one, without any controls, starting with power resources model. As predicted, both the strength of left representation in government and union density are highly significant, and those two variables together explain 42% of the variation. The Meltzer Richard model, in contrast, shows the opposite from the predicted effect. The ratio of the mean to the median income is highly significant but negative. So, the more unequal the income distribution, the less redistribution happens. Iversen and Soskice’s argument about proportional electoral systems leading to greater redistribution both on its own and indirectly, by encouraging left executives, receives support from the ‘Institutions’ model. The highly significant left effect is as expected and is joined by an equally significant PR effect. The Skewness model shows some support for Lupu and Pontusson’s contention that a greater skew of the income distribution generates more redistribution. We entered the skew and the extent of inequality together, as they do in their analyses, to separate out the effect of the shape of inequality from its overall magnitude. The coefficient for the skew is significant and positive, and the coefficient for the 90:10 ratio is highly significant and negative, as in our test of MR.

Table 2.

Determinants of redistribution.

	PRT		MR		Institutions		Skewness
Left government	0.168	***			0.182	***
Union density	0.176	***
Mean to median ratio			−43.755	***
Proportional representation					3.103	***
Skew							4.213	**
Wage dispersion							−5.794	***
Market income: Post tax	−3.562	***	−4.862	***	−4.935	***	−12.934	***
Market income: Mixed	2.263		1.93		0.588		−1.534
Constant	17.149	***	76.729	***	18.831	***	40.407	***
Common ρ	0.9		0.9		0.9		0.9
Adjusted R²	0.42	***	0.17	***	0.36	***	0.36	***
Observations	545		542		562		414

*Significant at .05; **significant at .01; ***significant at .001; ^ significant opposite hypothesized direction.

Figure 1 shows the predicted level of redistribution for each of the master variables when we reevaluate these models while controlling for social risks (unemployment, employment, children in single mother households), globalization (trade openness), diversity (immigration), and political features (voter turnout and veto points).⁶ The variables in the PRT and Institutions models remain positive and significant, and the MR model continues to defy its theoretic expectations. The only ‘master variable’ affected by the controls is the income-skew measure which becomes insignificant. The skewness’ model’s other predictor, the 90:10 income ratio, remains significant and negative.

Figure 1.

Predicted redistribution per level of master variables (partial residuals in circles).

Since it is possible that the effect of certain master variables may be period-dependent, we replicated the analysis in Figure 1 for the pre- and post-2000 periods, separately. As Figure 2 shows, the only master variable that changed direction is skew. Before 2000, skew had a positive effect on redistribution, and after 2000 a negative one. The second master variable that changed is PR, moving from a strong positive association before 2000 to insignificance after 2000. In the models controlling for PR, left government moved from insignificance to positive and significant in the post-2000 period. The results for the PRT models and the MR models remain the same.

Figure 2.

Master variable effect by period with 0.9 (thick) and 0.95 (thin) confidence intervals (based on a two-tailed t-test).

In Table 3 we explore the intermediate step between the theoretical master variables and redistribution, the relationship between these variables and welfare state generosity for the working-age population. Here too, we test the master variable models with controls, with two changes: the dependent variable is now welfare generosity rather than redistribution and our measures of social risks are not included. The reason to include them in the analyses of redistribution but exclude them here is a theoretical one: with welfare state generosity held constant, greater risk or need will result in more redistribution. When it comes to explaining welfare state generosity, there is no clear theoretical reason to expect an impact of risk. Functionalist theories would argue that greater risks will generate a response to counter these risks in the form of new or more generous social policies; theories of the fiscal crisis of the state would argue that great problem pressure or high levels of risk will result in cuts to generous social policies.

Table 3.

Determinants of social policy (nonaged generosity) with controls.

	PRT		MR		Institutions		Skewness
Left government	0.309	***			0.256	***
Union density	−0.006
Mean to median ratio			−38.536	***
Proportional representation					2.098	***
Skew							5.858	***
Wage dispersion							−4.519	***
Voter turnout	0.033	*	−0.007		0.006		0.001
Veto points	1.441	***	0.086		1.265	***	0.703	***
Trade openness	0.084	***	0.069	***	0.063	***	0.089	***
Immigration	−0.403	***	−0.124	***	−0.305	***	−0.173	***
Constant	8.833	***	61.158	***	8.944	***	21.314	***
Common ρ	0.9		0.9		0.9		0.9
Adjusted R²	0.56	***	0.38	***	0.59	***	0.52	***
Observations	946		668		966		638

*Significant at .05; **significant at .01; ***significant at .001; ^ significant opposite hypothesized direction.

The results in Table 3 mirror those in Table A3,⁷ with two notable exceptions. Union density in the PRT model loses statistical significance. This is likely due to voter turnout, which is highly correlated within union density (.49) absorbing some of its effect. We do not observe the same change when we run this model without controls.⁸ Second, skew regains statistical significance in the predicted direction even in the presence of controls. It seems that an income distribution where the distance between the top and the middle is large compared to the distance between the middle and bottom may not be associated directly with the degree of redistribution, as Lupu and Pontusson originally suggested, but is associated with more generous social benefits. The remaining variables are left unchanged. Left incumbency remains significant and positive in the PRT and Institutions models and PR remains significant and positive in the latter. The mean to median ratio in the MR model remains highly significant and negative. This is perhaps the most consistent finding across our different models and specifications, which again underlines that greater inequality results in less generous policy.

Our robustness tests for redistribution and policy generosity are displayed in Appendix B. The inequality measures have a consistent and negative effect in every single model. Moreover, wage disparity is always significant and the mean to median ratio is highly significant in all but one of the models. By contrast, one or more of our power resources variables is always significant and positively signed across the different models. The PR variable is correctly signed and significant in all but one model whereas skew is mostly significant in the models with welfare state generosity as the dependent variable.⁹

Conclusions

Our main findings are that the size and the shape of income inequality are either irrelevant for redistribution or work in the direction opposite to that which the extant theories postulate. The needs, wants, attitudes of the median voter are by no means automatically reflected in policy. Rather, higher levels of income inequality translate into higher levels of political inequality through various channels. Money speaks in politics in various ways and inequality seems to only aggravate this dynamic. It does so by supercharging the donor class’s war-chest or by suppressing political engagement and interest at the bottom of the distribution (Solt, 2008). Bartels (2008), Gilens (2012) and Witko et al. (2021) have demonstrated for the United States, the country with the most unequal income distribution among post-industrial societies, that policies systematically respond to interests of the wealthy and business. Greater inequality is also associated with heightened national pride among all classes (Solt, 2011), which can divert voters from distributive to nationalist/populist parties.

The main antidote to economic and political inequality is organization among the have-nots in unions and political parties. Unions can serve as lobby groups and they are able to mobilize members into political participation, thus reducing differential turnout across income groups. Unions have traditionally been the main support base of left parties, and they remain important forces pushing for redistributive policy even as many of their members have moved to support populist or nationalist parties. Christian unions were also an important force pushing Christian democratic parties towards more pro-welfare state policies. As this base has weakened, these parties have become more supportive of neo-liberal policies. We find a consistent and substantively important effect of union density on redistribution, which compounds on unions’ previously established effect on market inequality (Huber and Stephens, 2014). This means that in the social battle for redistribution, resources may be less unequally distributed than previously thought.

Redistribution through taxes and transfers happens through legislation. Therefore, it matters who controls the government and since such systems take years to build, it matters over time. Long-term incumbency of governments committed to redistribution will result in more effective and generous tax and transfer systems. There is a vast literature demonstrating this relationship (Amenta, 2003; Campillo and Sola, 2020), and our results here show this relationship yet again. To the extent that PR electoral systems result in more frequent left incumbency, they produce more redistributive policy. In this sense, Iversen and Soskice’s argument complements rather than rivals PRT.

All of the theories assume that the master variables work through policy to bring about redistribution. And indeed, we do find a substantively strong effect on our policy variable, even though it only measures a part of the welfare state, essentially benefits for people in the labour force who are temporarily absent from the labour force due to sickness or unemployment. Left incumbency, a mainstay of PRT and Iversen and Soskice’s theory is a strong predictor of generous social policy. PR remained a significant predictor, even controlling for left incumbency, which supports Iversen and Soskice’s contention that PR systems incentivize more broad-based and universalistic policies than electoral systems that prioritize more narrow local support bases. The overlap in the determinants of redistribution and social policy offers a clue regarding the trend towards greater inequality in disposable income over time: in response to rising levels of unemployment, many countries reduced the generosity of their unemployment benefits and consequently their capacity to redistribute. Prominent examples include the 2004 Hartz reforms in Germany and changes to the Swedish unemployment system since the 1990s and more recently under the bourgeois governments of 2006–2014.

Unlike in our analysis of redistribution, income skewness turned out to be a good predictor of the generosity of policy. Since our generosity measure is targeted at full-time wage earners rather than households, the different results suggest the following: wage distribution where the middle is closer to the bottom than the top is favourable for a political alliance to protect wage and salary earners from loss of income but not for broader redistributive policies. This makes sense when considering that as household members, individuals often share economic position and corresponding preferences with members with different earning capacities. Their common interests in protecting the household against income loss via sickness and unemployment insurance do not extend to other redistributive programmes.

Our most consistent result, however, is the highly significant and negative effect of inequality on policy-generosity and redistribution. This is true whether we measured it as the MR mean-to-median ratio of household income, or Lupu and Pontusson’s 90:10 ratio of wage earners. High economic inequality obstructs the formation of coalitions favouring generous social protections and weakens the political power of those who have the most to gain from redistribution. Given the consistency of these results, both in this study and in previous studies, it is simply amazing that so many scholars studying redistribution still refer to Meltzer Richard as a seminal contribution and take it as a starting point for their analyses. For all of its intuitive and heuristic appeal, it is time to lay this theory to rest.

Supplemental Material

Supplemental Material - Goodbye to Meltzer-Richard: Testing major theories of redistribution

Supplemental Material for Goodbye to Meltzer-Richard: Testing major theories of redistribution by Itay Machtei, Evelyne Huber and John D Stephens in Journal of European Social Policy

Footnotes

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iDs

Itay Machtei

Evelyne Huber

Supplemental Material

Supplemental material for this article is available online at .

Notes

References

Alesina

Glaser

(2004) Fighting Poverty in the U.S. And Europe: A World of Difference. New York: Oxford University Press.

Amenta

(2003) What we know about the development of social policy: comparative and historical research in comparative and historical perspective. In: Mahoney

Rueschemeyer

(eds) Comparative Historical Analysis in the Social Sciences. New York: Cambridge University Press, 91–130.

Atkinson

(2015) Inequality: What Can Be Done? Cambridge, MA: Harvard University Press.

Baldwin

Huber

(2010) Economic versus cultural differences: forms of ethnic diversity and public goods provision. The American Political Science Review 104(4): 644–662.

Bartels

(2008) Unequal Democracy: The Political Economy of the New Gilded Age. Princeton, NJ: Princeton University Press.

Bechtel

Hangartner

Schmid

(2016) Does compulsory voting increase support for leftist policy? American Journal of Political Science 60(3): 752–767.

Beck

Katz

(2004) Time-series-cross-section issues: dynamics, 2004. In: 21st Annual Meeting of the Society for Political Methodology, Standford, CA: Stan ford University, 29–31.

Beck

Katz

(2011) Modeling dynamics in time-series–cross- section political economy data. Annual Review of Political Science 14(1): 331–352.

Blasco

Guillaud

Zemmour

(2023) The inequality impact of consumption taxes: an international comparison. Journal of Public Economics 222: 1–26.

10.

Bradley

Huber

Moller

, et al. (2003) Distribution and redistribution in postindustrial democracies. World Politics 55(2): 193–228.

11.

Brady

(2009) Rich Democracies, Poor People: How Politics Explain Poverty. New York: Oxford University Press.

12.

Brady

Huber

Stephens

(2021) Comparative Welfare States Data Set, Multiple Countries; 2021. University of North Carolina at Chapel Hill and WZB Berlin Social Science Center, dataset.

13.

Campillo

Sola

(2020) Power resources theory: a critical reassessment. Revista Española de Investigaciones Sociológicas 170: 19–50.

14.

Desmet

Ortuño-Ortín

Weber

(2009) Linguistic diversity and redistribution. Journal of the European Economic Association 7(6): 1291–1318.

15.

Döring

Manow

(2017) Is proportional representation more favourable to the left? Electoral rules and their impact on election, parliaments and the formation of cabinets. British Journal of Political Science 47(1): 149–164.

16.

Elkjær

Iversen

(2023) The democratic state and redistribution: whose interests are served? American Political Science Review 117(2): 391–406.

17.

Esping-Andersen

(1990) The Three Worlds of Welfare Capitalism. Princeton, NJ: Princeton University Press.

18.

Eurostat (2021) Eu-Statistics on Income and Living Conditions (Eu-Silc) Microdata, Multiple Countries; 2021, Luxembourg, dataset. https://ec.europa.eu/eurostat/web/microdata/overview.

19.

Feenstra

Inklaar

Timmer

(2015) The next generation of the penn world table. American Economic Review 105(10): 3150–3182.

20.

Gilens

(2012) Affluence and Influence: Economic Inequality and Political Power in America. Princeton, NJ: Princeton University Press.

21.

Hacker

(2005) Policy drift: the hidden politics of US welfare state retrenchment. In: Streeck

Thelen

(eds) Beyond Continuity: Institutional Change in Advanced Political Economies. New York: Oxford University Press, 40–82.

22.

Huber

Stephens

(2001) Development and Crisis of the Welfare State: Parties and Policies in Global Markets. Chicago, IL: University of Chicago press.

23.

Huber

Stephens

(2014) Income inequality and redistribution in post-industrial democracies: demographic, economic and political determinants. Socio-Economic Review 12(2): 245–267.

24.

Huber

Stephens

(2015) Postindustrial social policy. In: Beramendi

Häusermann

Kitschelt

, et al. (eds) The Politics of Advanced Capitalism. New York: Cambridge University Press, 259–281.

25.

Immergut

(1992) The rules of the game: the logic of health policy-making in France, Switzerland, and Sweden. In: Steinmo

Thelen

Longstreth

(eds) Structuring Politics: Historical Institutionalism in Comparative Analysis. Cambridge: Cambridge University Press, 57–89.

26.

Iversen

Goplerud

(2018) Redistribution without a median voter: models of multidimensional politics. Annual Review of Political Science 21: 295–317.

27.

Iversen

Soskice

(2006) Electoral institutions and the politics of coalitions: why some democracies redistribute more than others. American Political Science Review 100(2): 165–181.

28.

Jensen

van Kersbergen

(2017) The Politics of Inequality. Basingstoke: Palgrave Macmillan.

29.

Kenworthy

Pontusson

(2005) Rising inequality and the politics of redistribution in affluent countries. Perspectives on Politics 3(3): 449–471.

30.

Korpi

Palme

(1998) The paradox of redistribution and strategies of equality: welfare state institutions, inequality, and poverty in the Western countries. American Sociological Review 63(5): 661–687.

31.

Larsen

Harell

(2023) Are Muslims in Europe and Black Canadians Evaluated Like Black Americans? Majority Perceptions of Out-Groups and Redistributive Attitudes. Unpublished manuscript.

32.

Lee

(2005) International migration, deindustrialization and union decline in 16 affluent OECD countries, 1962–1997. Social Forces 84(1): 71–88.

33.

LIS (Luxembourg Income Study) (2023) LIS Microdata, Multiple Countries; 2021. Luxembourg: LIS, database. https://www.lisdatacenter.org.

34.

Lupu

Pontusson

(2011) The structure of inequality and the politics of redistribution. American Political Science Review 105(2): 316–336.

35.

Mahler

Jesuit

Paradowski

(2013) Political sources of government redistribution in high-income countries. In: Gornick

Jäntti

(eds) Income Inequality: Economic Disparities and the Middle Class in Affluent Countries. Stanford, CA: Stanford University Press, 145–172.

36.

Marical

d’Ercole

Vaalavu

, et al. (2006) Publicly-provided services and the distribution of resources. In: Social, Employment and Migration Working Paper 45. Paris: OECD.

37.

Meltzer

Richard

(1981) A rational theory of the size of government. Journal of Political Economy 89(51): 914–927.

38.

Moene

Wallerstein

(2003) Earnings inequality and welfare spending: a disaggregated analysis. World Politics 55(4): 485–516.

39.

Nelson

(2012) Counteracting material deprivation: the role of social assistance in Europe. Journal of European Social Policy 22(2): 148–163.

40.

OECD (2021a) Employment and Labour Market Statistics, Multiple Countries; 2021, database. DOI: 10.1787/data-00302-en.

41.

OECD (2021b) Income Distribution Database (IDD), Multiple Countries; 2021, database. https://www.oecd.org/social/income-distribution-database.htm.

42.

OECD (2021c) Main Economic Indicators, Multiple Countries; 2021, DOI: 10.1787/data-00046-en.

43.

Persson

Tabellini

(2003) The Economic Effects of Constitutions. Cambridge, MA: MIT Press.

44.

Pierson

(2001) The New Politics of the Welfare State. Oxford: Oxford University Press.

45.

Pierson

(2003) Big, slow-moving and… invisible. In: Mahoney

Rueschemeyer

(eds) Comparative Historical Analysis in the Social Sciences. Cambridge: Cambridge University Press, pp.177–207.

46.

Plümper

Troeger

(2019) Not so harmless after all: the fixed-effects model. Political Analysis 27(1): 21–45.

47.

Rehm

(2016) Risk Inequality and Welfare States: Social Policy Preferences, Development, and Dynamics. Cambridge: Cambridge University Press.

48.

Rueda

(2018) Food comes first, then morals: redistribution preferences, parochial altruism, and immigration in Western Europe. The Journal of Politics 80(1): 225–239.

49.

Schwabisch

Smeeding

Osberg

(2006) Income distribution and social expenditures. In: Papadimitriou

(ed) The Distributional Effects of Government Spending and Taxation. London: Palgrave Macmillan, 247–288.

50.

Scruggs

Tafoya

(2022) Fifty years of welfare state generosity. Social Policy and Administration 56(5): 791–807.

51.

Solt

(2008) Economic inequality and democratic political engagement. American Journal of Political Science 52(1): 48–60.

52.

Solt

(2011) Diversionary nationalism: economic inequality and the formation of national pride. The Journal of Politics 73(3): 821–830.

53.

Stephens

(1979) The Transition from Capitalism to Socialism. London: Macmillan.

54.

Vernby

Lindgren

(2009) Estimating Dynamic Panel Models when There Are Gaps in the Dependent Variable. Sweden: Uppsala University. Department of Government Working Paper.

55.

Visser

(2019) ICTWSS Database, Multiple Countries; Version 6.1, Amsterdam: Amsterdam Institute for Advanced Labour Studies (AIAS). University of Amsterdam. dataset. https://www.ictwss.org.

56.

Wang

Caminada

(2011) Disentangling Income Inequality and the Redistributive Effect of Social Transfers and Taxes in 36 Lis Countries. Department of Economics Research Memorandum. Netherlands: Leiden University. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1909941.

57.

Witko

Morgan

Kelly

, et al. (2021) Hijacking the Agenda: Economic Power and Political Influence. New York: Russell Sage Foundation.

58.

World Bank (2021) World Development Indicators (WDI), Multiple Countries; 2021, database. https://data.worldbank.org/indicator/SM.POP.TOTL.ZS.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.30 MB