Trends in Market Inequality

Almost all of the recent literature by political scientists on market inequality, either as a dependent or an independent variable, relies on an unpublished, continuously updated OECD dataset on gross (pretax) earnings from employment for individuals who are employed full-time.⁵

Figure 1 summarizes trends in earnings inequality over the 1980s and 1990s in all countries for which the OECD reports comparable earnings data. The OECD dataset only enables us to measure inequality by decile ratios. As in most of the existing literature, the measure of inequality presented in figure 1 is the 90/10 ratio, that is, the ratio of the earnings of someone in the 90th percentile to the earnings of someone in the 10th percentile of the earnings distribution. The countries are ordered according to the absolute amount of change in inequality, measured as the most recent observation minus the earliest observation.

The OECD data on individual earnings cover different time periods for different countries. Note that the Danish time series ends in 1990, the Belgian time series in 1993, and the Canadian time series in 1994, and the Belgian, Danish, Italian, and Swiss data cover periods of ten years or less. Comparing countries in terms of total change in earnings inequality over time periods of different duration is obviously questionable. However, the country rankings based on average annual change in 90/10 earnings ratios are very similar to those in figure 1, and the correlation between total change and average annual change is 0.94 (see appendix). Keeping in mind the limits of some of the country data, figure 1 indicates that earnings inequality increased in nine out of fifteen countries in the 1980s and 1990s, but held steady or declined somewhat in the other six countries.

Consistent with conventional expectations, figure 1 suggests that rising earnings inequality is first and foremost a characteristic of liberal market economies as conceptualized by the “varieties of capitalism” school:⁶

the United States, New Zealand, the United Kingdom, Australia, and Canada. Though the extent of inegalitarian trends varies considerably among them, all liberal market economies have experienced significant increases in earnings inequality. The United States stands out as the country in which earnings inequality has grown most dramatically in absolute terms. (Relative to initial levels, the increase in earnings inequality in New Zealand from 1984 to 1997 was nearly as large as the increase in the United States from 1979 to 2000.) We also observe rising earnings inequality in the Netherlands, Sweden, Italy, and Germany; however, inegalitarian trends generally appear to have been more muted in Europe's “social market economies” than in the liberal (Anglo-American) market economies. This also holds for Japan.⁷

Many scholars have attempted to explain cross-national variation in earnings inequality trends since the 1970s. Labor economists typically attribute the rise in earnings inequality to changes in relative supply and demand for more- or less-skilled labor, but concede that such factors do not adequately explain the observed cross-national variation.⁸

Students of comparative political economy have recently explored the role of labor market institutions in a more systematic fashion. Some authors emphasize that more centralized or coordinated forms of wage bargaining generate wage compression;⁹ others point to the bargaining power of unions, measured by union density, as the key variable.¹⁰ David Rueda and Jonas Pontusson also show that countries with large public sectors tend to have more egalitarian earnings distributions.¹¹

All this might suggest that market forces have impelled earnings inequality to rise across the affluent OECD countries, but that centralized wage bargaining, strong unions, and large public sectors have muted the effects of these forces in some countries. Alternatively, changes in institutions might be seen as a source of rising earnings inequality. In particular, union decline and public-sector retrenchment (or restructuring) are rather consistently associated with changes in earnings inequality on a cross-national basis.¹²

Our goal is not to discuss the determinants of earnings inequality, but rather to point out that the individual earnings data presented in figure 1 capture only one dimension of labor market trends and do not provide the basis for a comprehensive assessment of the extent to which inequality has increased in different OECD countries. In particular, these data fail to capture the distributive effects of unemployment, underemployment, and labor force exit. The analytical problem that unemployment (or nonemployment) poses in this context is not simply that earnings-based measures of inequality exclude the unemployed segment of the labor force. The problem goes deeper because job losses during economic downturns are not distributed equally across the wage distribution. Evidence indicates that employers are more likely to fire unskilled (low-paid) than skilled (high-paid) workers during cyclical downturns. Employers are reluctant to fire skilled workers because it is difficult and costly for them to reacquire the skills that such workers embody when demand picks up again. Since low-wage workers disproportionately drop out of the employed labor force, increased unemployment tends to reduce earnings inequality among employed workers during economic downturns. Surely we ought not interpret this to mean that unemployment promotes equality.¹³

In addition, the underlying data used to calculate the 90/10 earnings ratios presented in figure 1 are, except in the case of Norway, restricted to full-time workers. Part-time workers earn considerably less per hour than full-time workers in all OECD countries, but they tend to do better relative to full-time workers in countries where wages among full-time workers are more compressed.¹⁴

Including part-time workers in our measure of earnings inequality should not matter very much to a comparative assessment of earnings inequality at any point in time, but it is likely to matter a great deal to a comparative assessment of changes over time in earnings inequality because there is substantial variation in the growth of part-time employment across countries. Measured as the percentage of the employed labor force working fewer than 30 hours per week in the main job, the incidence of part-time employment declined from 1990 to 2003 in the United States, Denmark, Norway, and Sweden, but increased by more than 5 percentage points in Australia, Germany, Ireland, Japan, and the Netherlands.¹⁵ Assuming that part-time/full-time pay differentials remained unchanged, the 90/10 earnings ratios presented in figure 1 might be said, from the perspective of cross-national comparison, to overstate the growth of earnings inequality in the former group of countries and understate the growth of earnings inequality in the latter group.

For eleven OECD countries, data from the Luxembourg Income Study enable us to trace the evolution of Gini coefficients for gross market income among working-age households over the 1980s and 1990s, with “working-age households” defined as households headed by individuals aged 25–59. Figure 2 shows the earliest and most recent available Gini coefficients, with countries ordered by the magnitude of absolute change from the earliest to the most recent observation. Here too, the calculations refer to country-specific time periods.¹⁶

Figure 2 conveys a strikingly different picture from that of figure 1. The data on household income show an OECD-wide rise in market inequality. The Netherlands is the only exception to this general trend. Equally notable, figure 2 casts new light on the American experience of the 1980s and 1990s. Whereas the dramatic growth of earnings inequality in the United States over the last two decades (see fig. 1) appears to be exceptional, the United States does not stand out when it comes to trends in market inequality among working-age households. More generally, figure 2 provides no support for the proposition that the growth of inequality has been more pronounced in liberal market economies than in social market economies. Quite the contrary, Sweden, Finland, and Norway hold three of the top six positions when the affluent OECD countries are ranked by change in household inequality over this period.

Finland and the Netherlands illustrate the divergence between trends in individual earnings inequality and household income inequality most starkly. In Finland, earnings inequality among full-time employees declined marginally from 1980 to 2001, but household inequality increased sharply from 1987 to 2000. In the Netherlands, by contrast, earnings inequality increased from 1979 to 1999, but household inequality declined over roughly the same period (1983–99).

As we demonstrate in the appendix, the divergence between trends in individual earnings inequality and household income inequality represents a substantive puzzle rather than an artifact of the different properties of Gini coefficients and 90/10 ratios as measures of inequality. Analyzing patterns of labor force entry and exit provides one possible avenue to resolve this puzzle. In particular, the rise in market income inequality in countries that did not experience much increase in earnings inequality might be due to an increase in the share of households with little or no market income. We might reasonably look first to the youngest and oldest labor force participants. Among those aged 50 to 59, increases in early retirement may have contributed, though this form of labor force exit is most prevalent in continental European countries, which were not the countries experiencing the sharpest rise in income inequality.¹⁷

Expansion of higher education may have had a similar effect among households headed by individuals aged 25 to 29. It turns out, however, that if we restrict our analysis to households headed by individuals aged 30 to 49, the picture that emerges is essentially the same as that shown in figure 2. In each country the magnitude of increase in inequality is similar for this narrower age group, and the cross-country correlation between changes in market inequality for households with heads aged 25 to 59 and for households with heads aged 30 to 49 is 0.99.

Patterns of entry and exit into the labor force are closely related to employment conditions, and employment performance is surely the key to the contrast between Finland and the Netherlands, which, in the 1990s, represent opposite ends of the OECD spectrum in this respect. In Finland the employment rate declined by 14 percentage points from 1990 to 1994, and the rate of unemployment rose from 3 to 17 percent, remaining above 10 percent through the rest of the decade. By contrast, the Netherlands outperformed all OECD countries but Ireland in terms of employment growth in the 1990s, with its employment rate increasing by 10 percentage points between 1990 and 2000 and its unemployment rate dropping from 6 to 3 percent.¹⁸

While low-income households in the Netherlands gained employment relative to high-income households over the 1980s and 1990s, offsetting the relative earnings losses experienced by members of low-income households, comparable households in Finland lost employment relative to high-income households, while relative earnings remained more or less unchanged in the 1990s. Unfortunately, the LIS data do not allow us to investigate systematically changes in the distribution of employment. But we can report that the share of households in the bottom earnings quartile that had no employed person decreased by 31 percentage points in the Netherlands, while it increased by 7 percentage points in Finland from the earliest to the most recent LIS observation. Low-income households in the United States, as in the Netherlands, have compensated for rising earnings inequality by working more. For married couples with children, the average number of hours worked per year by families in the bottom quintile increased by 16 percent from 1979 to 2000, while the average number of hours worked by families in the top quintile remained constant over the same period.¹⁹

Other factors must also be considered as we grapple with the differences between the inequality trends shown in figures 1 and 2. Rising household inequality across the affluent OECD countries is partly a result of changes in demographics and household structure, in particular, the growing incidence of single-adult households²⁰

and the fact that, increasingly, men and women tend to marry or live with partners of like income.²¹ In addition, the data on household inequality in figure 2 encompass sources of income other than dependent employment. Income from real estate and financial assets appears to be of secondary importance in this context (see appendix), but patterns of self-employment may account for some of the differences between trends in individual earnings inequality and household income inequality.

To decompose changes in household inequality along these lines is a complicated task, well beyond the scope of this paper. Suffice it to say that, aside from employment opportunities for low-income households, marital instability and its concomitant, single-headed households, constitute the most obvious factor for explaining the cross-national variation in inequality trajectories shown in figure 2. The Nordic countries are distinguished by high divorce rates.²²

With respect to timing, however, the employment crisis experienced by the Nordic countries, especially Sweden and Finland, in the early 1990s provides a more powerful explanation of the rise of household inequality in these countries.²³

Figure 3 illustrates the significance of employment for trends in household inequality by plotting changes in Gini coefficients for household market income against changes in the employment rate (the employed as a percentage of the population aged 15 to 64) for each country.²⁴

The Netherlands stands out as the country that experienced the largest increase in the employment rate as well as the only country that experienced a decline in household market inequality over the 1980s and 1990s. At the other end of the spectrum, Sweden, Finland, and Denmark are distinguished by declining employment rates as well as rising household inequality. The outlier status of the United Kingdom and the United States in figure 3 is noteworthy. In both countries, household market income inequality increased significantly more than we would expect based on changes in the employment rate. It surely is not a coincidence that, among the eleven countries covered by this analysis, these are the two in which individual earnings inequality rose most sharply in the 1980s and 1990s.

Regressing change in Gini coefficients for household market income inequality on change in employment rates and change in individual earnings inequality, from the earliest to the most recent observations reported in figures 1 and 2, yields the following results (t-statistics in parentheses):

We should not put too much stock in a regression analysis based on only eleven observations and poorly matched data for household and earnings inequality, but these results are certainly suggestive: trends in employment and earnings inequality together provide leverage on the question of why income inequality among working-age households has grown more in some countries than in others. The standardized coefficients are −0.82 for change in employment rates and 0.42 for change in earnings inequality, suggesting that employment is the more important variable. In a similar vein, Kenworthy has shown that from the mid-1980s to the mid-1990s the effects of changes in employment and, to a lesser extent, earnings inequality were larger and more consistent than the effects of changes in the incidence of single-adult households.²⁵

Employment and family structure need not be construed as competing explanations for rising household inequality. Single-headed households are significant in this context precisely because their ability to compensate for rising earnings inequality by increasing employment is constrained. As Jelle Visser and Anton Hemerijck have shown, much of the Dutch “jobs miracle” in the 1980s and 1990s involved an increase in part-time employment among married women.²⁶

The implication of the evidence presented above is that employment contractions disproportionately hurt the employment opportunities and hence the fiscal well-being of low-income households; conversely, job growth disproportionately benefited low-income households. Again, low-income households in the Netherlands and, to a lesser extent, the United States and other liberal market economies appear to have compensated for falling relative earnings by increasing their employment—more household members work, or those that do, work more. Low-income households appear to have been less able—or less willing—to engage in this type of compensatory behavior in those countries for which we observe little or no increase in individual earnings inequality but a significant increase in household inequality (Finland, Norway, Denmark, and Switzerland) or, as in the Swedish case, a larger increase in household inequality than in earnings inequality.

The existing literature points to two explanations for “compensatory employment” in some countries but not others. On the demand side, a combination of wage compression through centralized wage bargaining and high payroll taxes may weaken relative demand for low-wage workers in the more egalitarian social market economies of northern Europe.²⁷

On the supply side, continued real-wage growth for low-wage workers and the public provision of relatively generous income support for unemployed workers may have reduced the need for low-income households in these countries to seek compensatory employment.²⁸ Further analysis, beyond the purview of this paper, is required to determine the relative significance of these factors.

Trends in Redistribution

The LIS database also allows us to examine the redistributive effects of taxation and transfer payments by governments. At the end of the day, what matters to people is not market income, but rather disposable income, that is, income after taxes and transfers. Some analysts measure redistribution as the difference between Gini coefficients computed for disposable household income and Gini coefficients computed for gross market household income expressed in percent of Gini coefficients for gross market household income.²⁹

We believe that the absolute difference between disposable-income and market-income Gini coefficients represents a better and more easily interpretable measure of redistribution.³⁰

Figure 4 provides data on the reduction in Gini coefficients brought about by taxes and transfers in our 11 countries in the early 1980s (later for Denmark and Finland) and the late 1990s (earlier for Switzerland and Australia). As with figure 2, the data are restricted to working-age households. The largest government transfer program, pensions, is thus excluded from consideration. This analysis also does not take into account the redistributive effects of the public provision of services, such as education and health care, despite the fact that, services are a critical component of the welfare states of many affluent countries.³¹

Services do not alter the distribution of income per se, but to the extent they are provided at low or no cost and are universally available, services are equivalent to a flat-rate benefit given to each household, which reduces the degree of consumption inequality. Unfortunately, the LIS data do not include information on the value of such services.³² Thus, in terms of cross-national comparison, the data in figure 4 understate the degree of redistribution in relatively service-intensive welfare states, most notably the Nordic ones.

Our main interest concerns trends in redistribution. The countries in figure 4 are ordered by change over time (from larger to smaller increases in redistribution). Strikingly, almost all welfare states became more redistributive in the course of the 1980s and 1990s. As with the data on market inequality, the Netherlands stands out as the major exception to the general trend. The Netherlands is the only country in which we observe any reduction in the redistributive effect of taxes and transfers—and the reduction in this case is remarkably large.

Pooling LIS observations for working-age households for 14 countries (for a total of 59 observations) and controlling for the size of the welfare state, Bradley and colleagues show that the share of cabinet portfolios held by Left parties had a significant positive effect on levels of redistribution, measured as the percentage reduction in Gini coefficients brought about by taxes and transfers.³³

They find that union density, an alternative measure of working-class mobilization, also had a strong positive effect on levels of redistribution. Their cabinet share measure is cumulative, based on all years from 1946 to the year for which redistribution is observed. Thus their results for government partisanship should be taken to mean that countries in which Left parties have participated in government over extended periods of time tend to have more redistributive tax and transfer systems. Though Bradley and colleagues' analysis is persuasive, government partisanship does not seem to explain the general tendency for redistributive effects to increase.³⁴ In a number of the countries included in figure 4, Left parties gained control of government (or increased their representation in government) in the second half of the 1990s, but the trends in redistribution predate this turn to the Left. Generalizing across the rich OECD countries, the participation of Left parties in government declined in the 1980s.³⁵ Also, several of the countries for which we observe the largest increases in redistribution over the 1980s and 1990s—Germany, the United Kingdom, and Australia—are not distinguished by a history of Left-party dominance or, alternatively, by significant political advances by Left parties.

More generally, the evidence presented in figure 4 provides support for Paul Pierson's emphasis on the resilience of welfare states in the face of globalization or fiscal and demographic pressures and calls into question the common notion that recent tax and social policy reforms in the affluent countries have been uniformly and extensively regressive.³⁶

It should be noted, though, that the increases in redistribution that we observe in the LIS data are almost entirely attributable to the effects of transfer payments. In most countries, the contribution of taxation to redistribution declined over this period.³⁷ Also, when we look at the benefits provided by various social programs, we do indeed observe significant cutbacks in many countries.³⁸ Of particular note, given that our analysis focuses on income inequality among working-age households, James Allan and Lyle Scruggs document cuts in net income replacement provided by unemployment insurance and sick pay insurance in most countries included in figure 4.³⁹

How can we reconcile such evidence of welfare state cutbacks with the data on trends in redistribution? The obvious explanation for this puzzle is that labor market developments—rising joblessness among unskilled workers and rising earnings inequality—have rendered more households eligible for unemployment compensation and other transfer programs, and that increasing claims on these programs by low-income households have rendered government spending more redistributive, even as eligibility criteria have been tightened and benefit levels reduced.⁴⁰

In both Sweden and Finland, for instance, the number of unemployment benefit recipients increased more than fivefold between the mid-1980s and the mid-1990s, and the number of social assistance recipients nearly doubled.⁴¹ The share of working-age households receiving unemployment compensation jumped from 12 to 25 percent in Sweden and from 10 to 32 percent in Finland.⁴²

The bottom line is that there are two different paths to increased redistribution. One path involves new policy initiatives or policy changes aimed at redistributing income, such as easing eligibility restrictions or increasing replacement rates. The second path involves a more or less automatic compensatory response by existing welfare states to rising market inequality. In the 1980s and 1990s, most affluent OECD countries seem to have been on the second path.

The Effect of Market Inequality on Redistribution

The median-voter model elaborated by Meltzer and Richard⁴³

informs much recent discussion of the relationship between market inequality and redistribution.⁴⁴ This model assumes that government spending funds a certain amount of consumption for all individuals (the same amount for each individual) and that this spending is financed by a proportional income tax (so that the tax amount paid by individuals rises with market income). The basic intuition behind the model is that low-income earners have more to gain and less to lose from increasing government spending than do high-income earners. Assuming a one-dimensional model of voting, Meltzer and Richard argue that support for government spending is a function of the distance between the income of the median voter and the average income of all voters. As market inequality rises, the distance between median and mean income increases, and support for government spending consequently also increases.⁴⁵

The Meltzer-Richard model implies that countries with more unequal distributions of market income should exhibit higher levels of redistributive spending than countries with less unequal distributions. As commonly noted, this prediction is not borne out by the data for the OECD countries. Plotting levels of total government spending on social transfers and services against levels of earnings inequality among full-time employees (fig. 5), we observe an associational pattern quite the opposite of what the Meltzer-Richard model leads us to expect: countries with more egalitarian earnings distributions tend to have larger welfare states than countries with more inegalitarian wage structures. Torben Iversen and David Soskice refer to this pattern as the “paradox of redistribution.”⁴⁶

Does figure 5 invalidate the Meltzer-Richard model? As we have seen, it is problematic to use individual earnings inequality as a proxy for market inequality. Also, Meltzer and Richard present their model as an explanation of the size of government, but their model is really meant to explain redistribution. Meltzer and Richard in effect assume that all government spending takes the form of transfers to individuals or households and that all spending is (equally) redistributive. To test their model empirically, it would be more appropriate to examine redistribution rather than spending. The LIS data on market income inequality and redistribution among working-age households are in fact more consistent with the predictions of the Meltzer-Richard model than are the data presented in figure 5. Before turning to the LIS data, let us consider some objections to and extensions of the Meltzer-Richard model.

Clearly, the Meltzer-Richard model rests on a simple, perhaps simple-minded, view of politics. First, it assumes that elections determine redistributive policy and that electoral politics are only about redistributive policy. Voters are assumed to have well-defined, single-peaked preferences over redistribution that derive from their income relative to the mean income. Voters are also assumed to be fully informed about the policy choices before them. The “real world” of politics rarely, if ever, conforms to these assumptions. Second, the Meltzer-Richard model views political parties as motivated simply by the desire to win elections. The extensive literature demonstrating that the partisan composition of governments affects the extent and character of the public provision of social welfare calls this view into question.⁴⁷

Third, the median-voter logic may not apply as well to multiparty proportional representation electoral systems as to two-party winner-take-all systems. Finally, perhaps the most problematic aspect of median-voter models like this one is the assumption that all income earners are more or less equally represented in the political process, that is, that voting provides them with equal influence over the extent of redistribution.⁴⁸

The core propositions of the Meltzer-Richard model alone do not provide an adequate basis for understanding the politics of redistribution. The more interesting question is whether, in the context of other relevant considerations, they shed some light on the politics of redistribution. In addition, the model can easily accommodate inequality of political influence insofar as such inequality derives from the fact that low-income earners are less likely to vote than high-income earners. As Phillip Nelson points out, the income of the median voter should not be confused with the median income.⁴⁹

For the United States, the ratio of the median household income to the mean household income averaged 0.82 in the 1970s and 1980s. With household income weighted by the number of adults in each household, however, the ratio of the household income of the median voter to the mean household income averaged 1.02.⁵⁰ These figures reflect the comparatively low rate of voter turnout in American elections. By definition, the discrepancy between the median income and the income of the median voter will diminish as voter turnout approaches 100 percent. As voter turnout increases, the median voter typically becomes poorer relative to mean income.⁵¹

With regard to the motivations of political parties, the observation that parties cater to the distributive interests of their core constituencies and possibly also have ideological commitments to certain policies surely does not mean that they do not care about winning elections. Median-voter logic and partisanship need not be construed as mutually exclusive. In Geoffrey Garrett's formulation, we should expect governing parties of different political persuasions to pursue distinctive distributive policies so long as their pursuit of such policies does not threaten their prospects of reelection.⁵²

Also, median-voter logic does not necessarily require voters to be fully informed. As Karl Ove Moene and Michael Wallerstein put it, “voters may know little or nothing about the policy choices facing legislators, but if voters vote retrospectively, rewarding the incumbent government if their welfare has increased and punishing the incumbent otherwise, the parties in government have a strong electoral incentive to adopt policies that raise the welfare of a majority of voters.”⁵³

Proceeding from the same basic assumptions as Meltzer and Richard, Moene and Wallerstein attempt to resolve the “paradox of redistribution” by pointing out that government spending not only redistributes income, but also provides insurance.⁵⁴

The core proposition of their extension of the Meltzer-Richard model is that demand for insurance increases with income, holding risk (for example, the threat of unemployment or disability) constant.⁵⁵ People with higher incomes will choose to buy more insurance against income loss than people with lower incomes. From this follows a prediction concerning the relationship between inequality and welfare spending opposite to that proposed by Meltzer-Richard. Assuming that the mean income remains unchanged, the income of the median voter declines as inequality rises, and, consequently, the median voter's demand for social insurance declines as well. Moene and Wallerstein argue further that the contradictory logics of insurance and redistribution play themselves out differently for different types of social spending programs.⁵⁶ For policies that target those who have lost income due to lay-offs, sickness, or accidents, the demand for insurance dominates the demand for redistribution. Empirically, Moene and Wallerstein show that market inequality, measured by OECD earnings ratios for full-time employees, has a negative effect on spending on these types of policies. However, for health insurance and other policies that provide benefits for all workers, including employed workers, the reduction in demand for insurance and the increase in demand for redistribution associated with rising market inequality essentially cancel each other out.

The power resources approach advanced by John Stephens and Bradley and colleagues represents an obvious alternative solution to the “paradox of redistribution.”⁵⁷

In the power resources view, the negative association between earnings inequality and welfare spending across countries derives from the influence of working-class mobilization on both. Strong unions and Left parties compress the wage distribution and also boost redistributive social spending. In other words, there is no direct causal relationship between market inequality and welfare spending.

As indicated above, our main goal here is to recast the empirical basis of this debate by measuring inequality on a household basis—thus incorporating the effects of differential access to employment—and by focusing directly on the redistributive effects of public policy (tax policy as well as spending policy). To be consistent with our previous discussion and to side-step the “distortion” of redistributive effects created by generous public pensions (see appendix), our analysis remains restricted to households headed by people of working age. Excluding the retired population is obviously problematic from the point of view of testing median-voter theory, since retired people do vote, often more than working-age people. This is also a problem, however, for those who use individual earnings inequality as a proxy for market inequality.⁵⁸

Based on the most recent LIS observations available for eleven countries, figure 6 plots redistribution, measured as the (absolute) difference between Gini coefficients for gross market income and for disposable income, against levels of market inequality, measured by Gini coefficients for gross market income. We see a positive association between market inequality and redistribution among the Nordic and continental European countries, but, overall, figure 6 clearly does not support the Meltzer-Richard model. On the other hand, we no longer observe a pattern of association that runs counter to the Meltzer-Richard model. The “paradox of redistribution” effectively disappears when we measure market inequality and redistribution in the fashion proposed here.

Figure 7 plots changes in redistribution against changes in household market inequality over the 1980s and 1990s. Based on figures 2 and 4, figure 7 again uses the earliest available circa 1980 observation and the most recent available observation for each country. Here, we do observe the pattern of association that the Meltzer-Richard model predicts: in general, redistribution increased more in countries that experienced larger increases in market inequality in the 1980s and 1990s.⁵⁹

The exceptional nature of the Dutch experience again stands out and clearly influences the fit between increases in market inequality and redistribution. The American experience also appears to be exceptional. The United States stands out as the one country in which increased market inequality did not produce any increase in redistribution. Even if we disregard the Dutch and U.S. data points, figure 7 still shows a reasonably consistent positive association between rising market inequality and increasing redistribution. This positive association poses a challenge for power resources theory to the extent that it holds that trends in market inequality and redistribution are both attributable to changes in the distribution of power between labor and capital. (As the evidence presented in figures 6 and 7 pertains strictly to the redistributive effects of taxation and transfers, these figures do not bear directly on the insurance model of the welfare state proposed by Moene and Wallerstein.)

It would be a mistake to interpret figure 7 as evidence for the political process posited by the Meltzer-Richard model. It would surely be a stretch to say that rising market inequality led median voters in the affluent countries to opt for more redistributive parties and policies in the 1980s and 1990s. As suggested above, the observable increases in redistribution should be seen mainly as a more-or-less automatic response to inegalitarian labor market developments by existing welfare states. Nevertheless, we also caution against an overly path-dependent view of compensatory redistribution. In many countries, income taxation became less progressive, and various social transfer programs were in fact scaled back over this period. But neoliberal politicians often advocated more extensive tax reforms and cutbacks than were in fact undertaken. It seems quite reasonable to suggest, in the spirit of Meltzer-Richard, that in the context of rising market inequality the policy preferences of median voters constrained this neoliberal offensive.

Taken together, figures 6 and 7 suggest that the logic of the Meltzer-Richard model captures a dynamic that liberal democracies have in common, but also that there are important cross-national differences in “tastes for equality” or beliefs about the proper role of government⁶⁰

that cannot be explained in terms of the effects of income distribution on the policy preferences of the median voter. Figure 8 further illustrates both of these points. The LIS database provides at least four different observations for ten countries and includes observations from the 1970s for some of these countries. For each of these ten countries, figure 8 presents a scatterplot of the over-time relationship between household market income inequality and redistribution. It also shows the coefficient (b) obtained by regressing redistribution on market inequality.⁶¹

With the exception of the United States, we observe a relatively strong positive association between market inequality and redistribution in every country. But the slope of the regression line differs markedly across countries. For the United States, the sign of the coefficient is positive, but the line is nearly flat. Followed by Germany, the Netherlands stands out as the welfare state that has been most responsive to—in other words, compensated most fully for—increases (and reductions) in market inequality. Contrary to what the comparative welfare-state literature might lead us to expect,⁶²

the Nordic welfare states do not, as a group, stand out as particularly responsive to market inequality, though we again emphasize that these data do not take into account the redistributive effects of publicly provided services.

What accounts for the “redistribution elasticities” shown in figure 8? Why are some welfare states more responsive to market inequality than others? From the perspective of the Meltzer-Richard approach, voter turnout would seem to be the most obvious determinant of cross-national variation in compensatory redistribution. As Walter Korpi argues, voter turnout can be seen at least partly as a measure of the capacity of unions and labor-affiliated parties to mobilize workers politically.⁶³

Voter turnout thus emerges as an issue on which median-voter theories and power-resource theories converge. Against this backdrop, figure 9 plots the regression coefficients presented in figure 8 against average turnout in general elections over the country-specific periods covered by our LIS data. Though the fit between the variables is not terribly good, this figure provides some tentative support for the proposition that high turnout produces institutionalized policy commitments, which in turn produce greater responsiveness to increases in market inequality. Perhaps the most important point to take away from figure 9 is that low turnout offers a potentially compelling explanation for why the American welfare state has been so much less responsive to rising market inequality than other welfare states.

Figure 9 suggests that voter turnout, treated here as a proxy for the electoral mobilization of low-income workers, conditions the responsiveness of government policy to market income inequality trends. Alternatively, voter turnout and market inequality might be seen as variables that affect redistribution independently. Pooling all available LIS observations, we regressed redistribution on market inequality and voter turnout (in the election immediately preceding each LIS observation).⁶⁴

The results of this exercise are quite promising:

Market inequality and voter turnout both appear to be associated with higher levels of redistribution. Clearly, more systematic analysis (controlling for other variables that might affect redistribution) is needed to confirm these results, but our preliminary analysis suggests that the Meltzer-Richard model accurately identifies the distribution of market income and voter turnout as key variables in the politics of redistribution, as distinct from the politics of social insurance. At the same time, our discussion points to a potential synthesis of median-voter and power-resource theories: the median-voter theorem helps us to understand changes in redistribution to the extent that unions, Left parties, or other actors mobilize low-income workers to participate in the political process.