Is Redistribution Good for Our Health? Examining the Macrocorrelation between Welfare Generosity and Health across EU Nations over the Last 40 Years

Abstract

Context: Social determinants of health are finally getting much-needed policy attention, but their political origins remain underexplored. In this article, the authors advance a theory of political determinants as accruing along three pathways of welfare state effects (redistribution, poverty reduction, and status preservation), and they test these assumptions by examining impacts of policy generosity on life expectancy (LE) over the last 40 years.

Methods: The authors merge new and existing welfare policy generosity data from the Comparative Welfare Entitlement Project with data on LE spanning 1980–2018 across 21 countries in the Organization for Economic Cooperation and Development. They then examine relationships between five welfare policy generosity measures and LE using cross-sectional differencing and autoregressive lag models.

Findings: The authors find consistent and positive effects for total generosity (an existing measure of social insurance generosity) on LE at birth across different model specifications in the magnitude of an increase in LE at birth of 0.10–0.15 years (p < 0.05) as well as for a measure of status preservation (0.11, p < 0.05). They find less consistent support for redistribution and poverty reduction measures.

Conclusions: The authors conclude that in addition to generalized effects of policy generosity on health, status-preserving social insurance may be an important, and relatively overlooked, mechanism in increasing LE over time in advanced democracies.

Research has identified that as much as 90% of increased life expectancy (LE) over the last 100 years has come from nonmedical social factors, for example, improvements in environmental and sanitary conditions that brought about markedly improved living conditions (McGinnis, Williams-Russo, and Knickman 2002). This understanding of population health as largely nonmedically determined has contributed to a vast literature on the social and structural determinants of health spanning multiple decades (Daniels, Kennedy, and Kawachi 1999; Lynch and Kaplan 1997; Pickett and Wilkinson 2015; Williams 2003).

Scholars familiar with politics and political economy are often quick to point out that the social determinants of health are, in fact, mostly politically determined (Chung and Muntaner 2007; Navarro 2008). National and subnational governments make policy decisions about such matters as the distribution of resources, the amount and progressivity of taxation and redistribution, public goods provision, minimum wage laws, and social insurance protections, all of which shape living standards (Coburn 2004). A long literature in political economy has pointed to the rise of redistributive welfare states, in particular, over the course of the last century as an important cause of decreased income inequality in countries that are now high income (Lindert 2004, 2021; see Rehm 2016 for a review). Over time, unique clusters of welfare state policies have formed across countries, due at least in part to national political differences such as the success of left power resources (Korpi 2018; Reynolds 2022; Stephens 1979) and electoral institutions (Milesi-Ferreti, Perroti, and Rostagno 2002) fostering higher wages, better working conditions, and public goods (Esping-Anderson 1990).

In this article we contribute to the literature on welfare state policies as a political determinant of health. We share the perspective of past scholars in theorizing welfare states as influencing the social determinants that are crucial for population health (fig. 1). Our main contribution is to investigate more precisely three distinct mechanisms through which social policy outputs might influence population longevity over long periods: redistribution, poverty reduction, and risk protection. Thanks to the development of longer historical datasets on policy outputs extending more than 50 years in some cases (see Scruggs and Tafoya 2022), we can now better assess the long-run impacts of major social policy decisions on health. To make a case for the importance of these policies, we begin by reviewing the state of the literature on the relationship between welfare states and population health.

Background

Literature on welfare states and health has proceeded along three main areas, each overlapping in time but focusing on different measures and conceptualizations: (1) welfare state regimes, (2) aggregate social spending, and (3) specific welfare state transfers. (For an exhaustive review of the research across these areas, see Beckfield and Krieger 2009). Each stage has been marked by significant advances in knowledge but also critical limitations that we aim to address.

The greatest number of studies compare and contrast the health (and/or health inequalities) of nations with differing welfare state regimes (Bambra 2006; Cavelaars et al. 1998; Chung et al. 2013; Eikemo, Bambra, et al. 2008; Eikemo, Huisman, et al. 2008; Espelt et al. 2008; Grosse Frie, Eikemo, and von dem Knesebeck 2010; Hochman and Skopek 2013; Kunst et al. 2005; Muntaner et al. 2017; Navarro et al. 2006; Navarro and Shi 2001; Olafsdottir 2007; Sanders et al. 2009; Zambon et al. 2006). Pioneering work by Esping-Andersen (1990) presented a schema of three welfare regimes, each distinguished by a particular configuration of social protections: social democratic (universalistic benefits tied to citizenship); conservative (categorical benefits tied to social status); and liberal (targeted benefits tied to means). In their meta-analysis, Muntaner and colleagues (2011) conclude that nearly two-thirds of studies examining welfare state regimes and health find that social democratic welfare states have more salutary effects on population health and/or health inequalities, with evidence for the latter being slightly shakier than for the former (e.g., Brennenstuhl, Quesnel-Vallee, and McDonough 2012; Chung and Muntaner 2007).

Regime-based research suffered some strong critiques, perhaps because it was the inaugural approach to studying welfare states and health. First, some studies question the coherence of welfare state regimes (Scruggs and Allan 2006, 2008), suggesting that they are, at best, loose proxies for what they purport to measure. In fact, nearly all countries employ a mix of policy instruments that combine the features commonly portrayed as distinguishing these ideal types. For instance, the UK, one of Esping-Andersen's quintessential “liberal” welfare states, maintains an ideal-typical universalistic program in the National Health Service. Also, the United States has an old-age pension that fits the social insurance model much more than a liberal social assistance pension and is arguably more generous to nonworking spouses than pensions in erstwhile conservative welfare states. Second, national welfare regime types are too often treated as a historically fixed feature, although there is ample evidence of considerable changes over time in national benefit policies and their redistributive effect. Quintessential social democracies such as Sweden have become much more regressive and much less generous over time (Kato 2003; Sowula et al. 2023). In some countries, social protection for the aged has grown substantially but has shrunk for young and working-age people (Lindert 2021).

Perhaps most critically, measuring national welfare states as historically fixed does not allow us to examine whether and how policy change might affect population longevity via different pathways. Thus, rather than measuring national welfare policy by classifying countries as belonging to a regime type, we identify three policy mechanisms by which a country's social policy generosity can be conceptualized and measured over time, and we examine whether changes in the policies are associated with population health. Specifically, we examine if generosity is associated with improved health via variation in reduction of income inequalities (redistribution), improvement in economic conditions for the bottom of the income distribution (poverty reduction), and insurance against household income loss in the face of economic shocks (status preservation). We discuss these in greater detail below in the section on “Three Mechanisms Linking Welfare Policy Generosity and Health.”

The second set of studies on welfare states and health conceptualized welfare states in terms of overall social spending (Bradley et al. 2011; Chung and Muntaner 2006; Ng and Muntaner 2015; Raphael and Bryant 2004; Regidor et al. 2011). Despite its many contributions, this research has largely overlooked the fact that aggregate social spending is not an accurate measure of welfare state program generosity to particular income groups, that is, its redistributive profile. Expenditures alone tell us little about who or what is being invested in and how. The importance of this point is illustrated by the oft-cited paradox of the United States's exorbitant health spending but inferior health profiles (Bradley et al. 2011). Research to date has also failed to confront a major complication of using data on spending to assess the effects of welfare generosity on health—that is, poor population health may generate economic insecurities that push social expenditures upward. Thus, reliably isolating the effects of welfare states on health requires measuring the program structure of policy generosity, not simply traditional features like quantity of spending in a particular social sector such as health or pensions.

A third group of studies examines the impact of specific welfare programs on health. These include studies on family benefits, such as parental leave, child allowances, and subsidized child care (Arber and Lahelma 1993; Burstrom et al. 2010; Elstad 1996; Lahelma et al. 2002; Lundberg et al. 2008; Raphael and Bryant 2004; Whitehead, Burström, and Diderichsen 2000), unemployment benefits (Nordenmark Strandh, and Layte 2006; O'Campo et al. 2015; Renahy et al. 2018; Shahidi, Siddiqi, and Muntaner 2016; Shahidi et al. 2019; Shahidi et al. 2020; Siegel, Vogt, and Sundmacher 2014; Wennemo 1993), social security/retirement pensions (Kangas 2010; Lundberg et al. 2008), education (Filmer and Pritchett 1999; Grytten, Skau, and Sorensen 2014; Kim and Jennings 2009; Pampel and Pillai 1989), and/or some combination of programs (Ng and Muntaner 2015). A major limitation of domain-specific studies is that they tend to deal with a relatively narrow part of the much larger whole and do not combine multiple major welfare program areas; nor do they confront an impact of resource redistribution through the tax system on social equality and health. By narrowly focusing on the causal impact of specific programs or sectors, these studies lose sight of the interactive effects of welfare programs and their combined effects.

A second major limitation of domain-specific spending studies on health is their short time frames. The most common type of study in this area is either cross-sectional or a time series of a decade or two (Muntaner et al. 2011). While investments in poverty reduction or redistribution might be expected to have relatively immediate effects on some dimensions of health, such as infant or child mortality (Rodriguez 2019), the effects of these and most other welfare policies on key outcomes like LE likely accumulate over decades. Social investments in anyone but young children may take considerable time to manifest in LE tables (Goldstein and Wachter 2006). Thus far studies have largely failed to grapple with the question of when we might expect to see the longevity benefits of welfare state investments (or longevity costs of welfare state retrenchment). To address the possibility of a protracted period of unfolding, we look at the totality of the change in LE over several decades against change in state welfare generosity.

Three Mechanisms Linking Welfare Policy Generosity and Health

In addition to changes in attention to how to measure welfare generosity and its political origins, the literature on living standards and the social determinants of health suggests that different mechanisms may link generosity to health. Understanding which of these mechanisms is driving health improvements (or detriments) is consequential to designing social policies that promote health.

Income Inequality, Relative Deprivation, and Health

A large literature on social determinants has examined how income inequality affects health (Lynch and Kaplan 1997; Pickett and Wilkinson 2015). Proponents of the income inequality thesis note that inequality can undermine population health through both material and psychosocial mechanisms. For instance, some studies have shown that income inequality operates by reducing collective action and subsequent demand for public health care, education, and housing, contributing to deprivation in health-enhancing goods and services (Anderson, Mellor, and Milyo 2008; Kawachi and Kennedy 1999; Lynch and Kaplan 1997). Alternatively, high inequality can harm health by contributing directly to stress and reactivity to stress, both of which degrade health through psychobiological pathways (Marmot 2004; Sapolsky 2004; Wilkinson and Pickett 2009). The evidence for the income inequality thesis is bolstered by the fact that health forms a gradient across the economic distribution, rather than showing a sharp cutoff once basic needs are met, and that the steepness of the gradient appears to be responsive to the level of inequality in society (Marmot 2004). This literature suggests that reducing inequalities, and not merely raising up the bottom, is important for improving population health (Wilkinson 2020). In terms of welfare state policies, in line with Esping-Anderson's (1990) conception of social democratic welfare states, this points to the need for programs that are more universalistic over programs that narrowly target benefits toward those at the bottom of the income distribution (Korpi and Palme 1998).

Poverty, Absolute Deprivation, and Health

In response to the income inequality thesis, many scholars contend that individual income alone is sufficient to explain the aggregate-level relationship between inequality and health (e.g., Gravelle 1998; Mellor and Milo 2001; Wagstaff and Van Doorslaer 2000). This interpretation sees poor health as a result of the effects of poverty on poor people's health, and it considers findings regarding income inequality as a statistical artifact (Kawachi and Kennedy 1999; Lynch et al. 2000). Since poor health tends to exist more often at the bottom of the distribution, this literature sees poverty reduction, not inequality reduction per se, as the way to raise average population health. A large literature in the United States has examined whether and how specific means-tested safety net policies, which generally aim to reduce poverty and economic insecurity at the bottom of the distribution, promote health (Kim and Jennings 2009; Shahidi et al. 2019). If accurate, this literature regarding poverty reduction suggests that LE would be best improved by targeted poverty-reduction policies that improve living conditions among people at the bottom of the income distribution, even if those policies have little to no effect on inequalities or economic insecurities across the income distribution.

Status Preservation and Health

Finally, some scholars argue that one of the greatest threats to health lies not in either absolute or relative deprivation, but rather in the loss of social status from a given baseline. Linked to traditional welfare state concerns with social risk protection, there is an increasing interest in how economic insecurity may erode health, and likewise, in how income stabilizing social insurance can reduce status loss (Hacker, Rehm, and Schlesinger 2013; Kopasker, Montagna, and Bender 2018; Rohde et al. 2017; Watson and Osberg 2017). Countries vary in how “status preserving” their social insurance systems are. That is, countries vary in the degree to which their social insurance programs replace an equivalent proportion of income across unskilled (and low-paid) workers and skilled (and higher-paid) workers. Thus, status preservation may be an important additional mechanism through which social policy can increase health and longevity, but one that does not require significant reductions in either inequality or poverty. Accordingly, we examine the extent to which social insurance policies that reduce income loss across the income distribution and provide protection to higher-income earners are beneficial to health and longevity separate from how redistributive or poverty-reducing they may be.

In sum, each of these literatures implies something different about the mechanisms linking welfare state generosity and LE, as follows:

• Redistribution: Policies that flatten the income distribution through progressive financing and are universalistic in nature (bring everyone under a broad tent) should be associated with larger gains in LE over time.

• Poverty reduction: Policies that raise up the bottom of the distribution and target the poor will be associated with larger gains in LE over time.

• Risk protection/status preservation: Policies that provide greater protection against economic insecurity and prevent large income losses will be associated with larger gains in LE over time.

Figure 1 summarizes the theorized relationships and mechanisms linking welfare states to health.

Study Innovations

To address the aforementioned limitations of previous generosity measures and test potential health-related mechanisms along these three dimensions, we use data from the Comparative Welfare Entitlements Project (CWEP) to calculate benefit generosity based on the social insurance income replacement rates derived from specific social and tax policy parameters that vary by country and by year. These measures are not based on program expenditures per se. For example, to address poverty reduction outputs, we calculate benefit replacement rates in a particular country-year for notional individuals with low market incomes, defined as earning 50% of the average wages in that country and year. We use specific program and tax rules (from the laws of each country) in each country year to derive net-of-tax benefits that the low-income person/household would receive if they became sick, unemployed, or (assuming a career of low-income employment) retired. We divide that by the net income that a worker earning 50% of the average wage would earn in employment. This ratio varies across countries and years depending on changes to the benefit system or tax system. These calculations are considerably more time-consuming to calculate than total expenditures, because many social programs have minimum and maximum benefits levels; national tax systems differ in progressivity; and some countries treat benefits differently from wage income for tax purposes. Yet these policy parameters are more meaningful measures of system generosity than aggregate spending. Moreover, notional comparisons like these can provide a much more apples-to-apples comparison across countries or time, for two reasons. First, aggregate spending data would treat spending the same whether it was spent equally on all or or if it was spent only on those with higher (or lower) income. Second, aggregate spending data ignores whether and how much benefits are taxed as regular income. Third, sectoral or aggregate spending levels are unable to assess, as we can here, how programs treat low, average, or higher earners in a particular program or how that changes over time. For example, if conventional measures of aggregate pension spending levels rise 1% per year over decades, we cannot discern whether benefit spending is relatively equal across beneficiaries or very skewed.

The use of program benefits has been widely applied in econometric analyses of policy effects to address the complication that actual expenditures are affected by many of the same factors that drive population health (e.g., Bound et al. 2004). In some cases, expenditures may even be driven by population health itself, as sick individuals are more likely to need (and/or need higher levels of) welfare state support. Replacement rates provide a measure of welfare state generosity that is driven by supply rather than by demand, thereby minimizing the bias introduced by confounding and reverse causation. With this more reliable measurement tool, we revisit the literature on welfare states and health to explore not only whether generosity matters for population health but also the broad mechanisms through which this might occur.

Two studies that employ similar approaches are important to note. Beckfield and Bambra (2016) use the CWEP total generosity measure to examine the association between welfare state generosity (unemployment insurance, sickness benefits, and pensions) and LE for the United States and 17 other high-income countries during the period 1970 to 2010. They find strong associations between welfare generosity and LE: a 1-unit increase in total welfare generosity produces a 0.17-year increase in LE at birth (p < 0.001). A second study, by Jacques and Noël (2022), examines effects on age-adjusted mortality of three measures of decommodification: income redistribution, labor market polarization, and the reduction of labor market risk incidence. Using time-series cross-sectional regression across 21 countries in the Organization for Economic Cooperation and Development (OECD) from 1971 to 2010, they find that lower labor market polarization (measured by the share of market income allocated to the richest decile relative to the share of the poorest decile) and higher risk reduction (the degree to which the welfare state reduces the prevalence of large income losses) are associated with a lower death rate. They find no effects for income redistribution's effect on mortality. We build on this work by extending the time series out by nearly an additional decade (1980–2018), including a more diverse set of welfare entitlements encompassing different domains (e.g., redistribution, poverty reduction, and status preservation), and expanding the set of time-varying covariates (i.e., obesity, health spending). We also take a two-pronged approach to modeling, using both a first-difference modeling approach and an autoregressive distributed lag (ADL) model.

To preview, our findings are in line with previous literature. We find consistent and positive effects for total generosity on LE at birth across different model specifications in the magnitude of an increase of 0.10–0.15 years in LE at birth (p < 0.05). We find less consistent support for the effect of redistribution and poverty reduction on LE, but we do find significant and positive effects for status preservation as an LE-enhancing mechanism. We conclude that, in keeping with recent literature, overall generosity and status-preserving social insurance may be an important, but relatively overlooked, mechanism in increasing LE over time. Below we outline our methods and approach before turning to our results and discussion.

Methods

Using data from the CWEP and the OECD Health Statistics database, we examine the relationship between three measures of welfare generosity (corresponding with poverty reduction, redistribution, and risk protection) as well as total generosity and LE at birth in 21 Western nations during the period 1980–2018. The main analysis examines 16 western European countries and 5 non-European OECD countries for which complete data is available over the time period. We considered including eastern European countries, but the data on many countries does not exist before they became independent states in the 1990s, and information for several of the welfare state policy measures is not available at all or is only available for a short part of the period we study.

We chose 1980 as our starting point because it is the year on which Esping-Andersen (1990) constructed his widely cited welfare regime classification. Additionally, CWEP's coverage of welfare program data is less complete and less reliable for the 1970s. We ended our time series with 2018 because it is the last year for which we have CWEP data for our sample.

Data

Dependent Variable: LE at Birth

The OECD Health Statistics database provides comparable measures of LE at birth for all European Union countries back to the 1960s. For this study, we focus on LE at birth because it is one of the most commonly used measures of well-being, and it reflects gains in longevity at older ages as well reductions in deaths in youth.

Independent Variables: Total Generosity, Tax Progression, Benefit Equalization, Low-Earner Wage Replacement Rate, Higher-Earner Wage Replacement Rate

To construct our independent variables, we adapt data from the CWEP, a cross-national database of welfare state program characteristics. It measures features of the structure and generosity of welfare state benefits (unemployment insurance, sick pay, and public pensions) in a large number of OECD countries (Scruggs and Tafoya 2022). Measures focusing on average benefit generosity, using an earlier version of the dataset, have been used in numerous studies across the social sciences (e.g., Annarelli 2022; Alper, Huber, and Stephens 2021; Barth, Finseraas, and Moene 2015; Easterlin and O'Connor 2022; Romer 2017). This includes aforementioned work on comparative health systems (Beckfield and Bambra 2016; Jacques and Noël 2022).

Our main analyses cover change over a nearly 40-year period (1980–2018) using both a previously used summary measure, “total generosity,” and four new measures derived from features of the CWEP dataset that allow for evaluation of a “benefits gradient” among low, average, and higher income earners. A benefit gradient has, to our knowledge, never been used in earlier comparative work on welfare states and health, let alone over such a long period.¹Table 1 summarizes the five different measures of welfare state policy outputs that we include and the corresponding theoretical construct that they capture.

Total generosity is a composite index of features relating to the scope and coverage of three important redistributive social insurance programs that exist in (almost) all advanced industrial democracies: unemployment compensation, sick pay, and public pension insurance.² It has been used in previous studies on welfare states and health as a general measure of welfare generosity (e.g., Beckfield and Bambra 2016; Jacques and Noël 2022). For each of the three programs it encompasses, generosity is a function of (a) the proportion of an average earner's after-tax earnings that is replaced by the social program for a long-term full time worker; (b) the maximum duration of the benefit (for pensions, LE at age 65); (c) the qualifying period of social contributions needed to receive the benefit; (d) any waiting period required to claim benefits; and (e) the portion of the working population insured by the program (for pensions, the share of the population older than 65 who are receiving a pension). To compute the total generosity score, the first three characteristics (replacement rate, duration, and qualifying conditions) were normalized and summed. For unemployment and sickness programs, this sum is weighted by the portion of the labor force covered by the program. For public pensions, the sum of the component scores are weighted 1, based on the assumption that all those not working and above retirement age receive at least a social pension. While total generosity is a useful measure of the overall munificence of welfare states, this measure on its own does not explicitly capture the distributive qualities of these programs, that is, to what extent their contributions and benefits have a redistributive, poverty-reducing, or status-preserving character.

Economic Redistribution Measures: Tax Progressivity and Benefit Equalization

Tax progression, our first measure of economic redistribution, captures the difference between the average tax rate for high-earner households and that for low-earner households. Here a high earner is defined as a single person earning 200% of the average national wage or a family collectively earning 300% of the average national wage. A low earner is defined as a single person earning 50% of the average wage or a two-earner family making 100% of the average wage. A score of .50 would indicate that the tax rate for high earners is 50 percentage points more than the tax rate for low earners (e.g., 60% vs 10%).

Benefit equalization is our second measure of economic redistribution, which is the ratio of (a) the after-tax benefit for an average-earner household (i.e., single earner with 100% average wage or family earning 150%) to (b) the after-tax wage of a high-income household (single person earning 200% of the average wage or a family earning 300% of the average wage). A ratio of .50, for example, indicates that the average social benefit for an average earner equals 50% of the net wage of a high-income earner.

Poverty Reduction Measure: Low-Earner Wage Replacement Rate

Our poverty reduction measure focuses on replacement rates for low earners. A score of 100, for example, would indicate that low-income households could expect to get as much income in after-tax program benefits as they earn in work (after taxes), averaging across three low-earner household types and for unemployment, sickness, and pension programs.

Status Preservation Measure: Higher-Earner Replacement Rate

Higher-earner replacement captures middle-class risk protection by representing benefit replacement rates for higher earners. It is constructed similarly to the low-earner replacement rate except it measures replacement for a single person earning 200% of the average national wage and a family earning 300% of the average. A score of 1 means that a higher earner is made whole by wage replacement from unemployment, sick pay, and pensions.

Controls

To account for other countervailing trends over the time period, in multivariate analysis we control for gross domestic product (GDP) per capita at 2015 Production Power Parity, the share of GDP spent on all health services, and the obesity rate. Data on obesity and share of GDP spent on health come from World Health Organization global health observatory data.³

Analysis

We first present descriptive results showing changes over time in LE and changes in the replacement rate measures. Next, we examine bivariate scatterplots of the 38-year change in LE against the 38-year change in each of the five welfare state generosity measures. In addition, we examine the bivariate scatterplots of the pooled country-years of LE against the level of welfare generosity.

We take two separate modeling approaches to the analysis. The first is a long-run differencing model in which we regress the full-period differences in LE (LE _I,2018—LE _i,1980) against the full-period differences of the independent variables and the level of LE in 1980. This approach captures the effect of changes in generosity during the full time frame under investigation, controlling for the period changes in the controls.

The second set of analyses is based on the long-run effect estimates using the annual-level information for each country and a more intricate ADL model. While the effects of welfare state changes may take a long time to fully manifest and may impact LE with differential lags depending on country or period idiosyncrasies, it is plausible that our long-run results are period specific. To evaluate this, we also report effects estimated using the panel structure of the dataset. Using the panel data also permits us to evaluate the independent effects of each welfare state output controlling for others. An important challenge here is properly specifying a pooled time-series regression model composed of individual variables that are trended and change slowly. We use an ADL model on the panel dataset. We include country and year fixed effects, two lags of the dependent variable, and three lags of the explanatory variables (lags 1, 2, and 5). This estimation method permits us to obtain long-run average effects for welfare generosity and control variables. These long-run effects are reported in table 3 for the full 21-country sample.

These models are estimated with panel-corrected standard errors, and they take the following general form:

Results

Descriptive Results

Changes in LE

Table 2 and figure A1 show the change in LE between 1980 and 2018. All countries experienced an increase in LE over the time period, with an average increase of 7.8 years and an increase range of 5–10 years. While these increases may not seem large, to put them in perspective, the COVID-19 pandemic, the largest pandemic in the last 100 years, is estimated to have reduced LE by about 2 years on average in countries that experienced LE declines at all (Kuehn 2022).

Changes in Welfare State Outputs

Table 2 summarizes the change in the generosity measures between 1980–2018, and figures A2–A6 show national trends in welfare generosity over time. Relative to LE, generosity scores show more variation in path and trend. Table A1 summarizes the overall trajectory of change in generosity measures over time. Sweden was the only country that experienced declines across all five generosity measures.

Empirical Scatterplot Analysis

Changes to broad features of welfare state policies in a particular year are generally expected to have effects on population health over the long run. This implies that abrupt changes will take time to have a large effect, and a series of small changes may take even longer. Thus, to begin our evaluation of the effect of changes in welfare state benefits we analyze changes over the long period (X₂₀₁₈—X₁₉₈₀) among our sample countries, giving us one observation per country.

Figure 2 shows that over the last four decades, for all five of the welfare program measures there is a positive association between increased redistribution, poverty reduction, and status preservation and increased LE. This association appears to be weakest in the case of low-income replacement rates and the strongest for benefit equalization.

Multivariate Regression Analysis on Long-Run LE Changes

Our multivariate regression models include the control variables described above as well as one additional covariate: higher initial LE at the start of the period (in 1980) plausibly limits the subsequent gains that countries can make. This “regression to the mean” is particularly plausible in the EU context, because health innovations can more easily diffuse across borders in that region. Thus, we add LE in 1980 to the models.⁴ Results in table 3 present ordinary least squares multivariate estimates of this long-run differencing approach (subtracting 1980 rates from 2018).

Based on this approach we find that larger increases in individual welfare state measures are associated with larger LE gains. Increasing benefit program generosity for average workers by a standard deviation (about 4.5 points) is associated with an LE gain of about half a year. Policies that raised average earner benefit equality, raised replacement income among higher earners, and increased the progressivity of the tax rate all had similarly sized effects in raising LE (0.04–0.08 years). The one welfare policy that was not clearly associated with increased LE during this period is the replacement rate for those with low income.

Estimates for our control variables provide mixed results vis-à-vis our expectations. LE at the start of the period does have a comparatively large impact on change in LE in the period: 1 year above the sample mean in 1980 lessens LE growth by between 0.4 and 0.6 years. Increases in obesity are likewise strongly associated with lower LE growth. We do not find the usual relationship between increased per capita income growth or increased health spending and growth in LE. These results are robust to restricting the sample to the 16 European countries (table A2); indeed, the estimated impact of a unit change in various welfare state policy measures is somewhat larger in this smaller group of countries, especially in the case of the low-earner replacement rate (with p values between .05 and .10).

ADL Model Results

Table 4 presents the results from the ADL models. Examining the results from the ADL models generates different conclusions than those from the long-run differencing model. When we enter each welfare variable individually, that is, without controlling for other welfare state measures, only program generosity exerts a clear positive impact on LE. Estimates for low-earner replacement rates (column 2) and high-earner replacement rates (column 5) are much smaller, very close to 0. Estimates for benefit equalization and tax progressivity—the first related to compressing disposable incomes between average benefits and higher earner income, the second related to the effort to compress disposable incomes in the labor market at the top and bottom—are associated with LE reductions.

In column 6 of table 4, in which we enter all welfare outputs simultaneously, most of the individual effect estimates become more consistent with expectations (with an important exception noted below). We see that raising program generosity for average earners (total generosity) and raising high-earner replacement rates have sizable positive effects on LE: raising either of these by 10 points would raise LE by more than 1 year. The measure that is most related to poverty reduction strategies, low-earner replacement rates, is not clearly related to raising national LE. Unexpectedly, however, benefit equalization (compressing the ratio of average earner benefits to high-earner disposable income) is estimated to reduce LE in the long run.

Among the control variable estimates we see that higher incomes are good for LE (doubling income raises LE by about 1.75 years), while higher obesity lowers LE (by .2 years for each percentage-point increase in population obesity). Somewhat unexpectedly, we see that higher health spending is associated with lower LE in the long run. The obesity and health spending results are sensitive to the sample selection; if we omit the non-European countries, specifically the United States and Japan, there appears to be little or no effect of higher obesity rates or more health spending on LE. These effects appear to be driven by the combination of very high obesity, very high health spending, and comparatively low LE in the United States and the very low obesity, modest health spending, and very high LE in Japan.

Discussion

We examined the effects on LE of changes in welfare state generosity over a 40-year time span in 21 high-income OECD countries. Overall, we find substantial improvements in LE and mixed patterns in welfare state expansion and retrenchment. While much of the previous literature on welfare and health has found social democratic regimes to be the most redistributive and to have the most substantial positive effects on health and health inequalities, those studies could not clearly identify the effects of specific pathways on those outcomes. Nor have they sufficiently accounted for gradual stagnation and retrenchment in generosity, although some have nevertheless concluded that the Nordic model continues to have an important mediating effect on class-based inequalities in health despite downward pressures from globalization and neoliberalism having blurred welfare regime typologies (Muntaner et al. 2017). The evidence presented here provides much stronger evidence based on changes in policy generosity that overall generosity of social insurance does indeed play an important role in extending LE at birth in the magnitude of 0.10–0.15 years. Several paired examples can bear this out. Considerable social insurance retrenchment in two social democracies, Sweden and Denmark, produced lower LE growth (+6.8 years) than that experienced in Finland (+8.1 years) and Norway (+7 years), both of which expanded social insurance in this period. Meanwhile, there was considerable expansion of various measures of welfare generosity in Italy and Portugal, two countries with relatively high gains in LE during the last 40 years.

By examining three pathways through which social insurance generosity may exert an influence on health, we demonstrate strong linear associations between changes in various measures of generosity and LE, and we find the most clear and consistent support for our measure of status preservation. A larger replacement rate among upper-middle income earners was associated with improvements in LE in our first differencing and in our fully adjusted ADL model. Remarkably, focusing only on increased benefits for the lowest paid generally produces the weakest association with improvements in LE. Perhaps even more surprising, though, was the lack of apparent effects of redistribution and inequality reduction on LE. In the long-run differencing models, redistributive tax structures and benefit equalization predict higher LE, but in the ADL models there is no effect. We also find in the ADL models a counterintuitive deleterious effect for benefit equalization between benefits for average earners and net income for higher earners. This seems to imply that raising risk protections at the middle and higher earner levels is more effective than full-fledged retrenchment or large-scale improvements in social insurance entitlements for lower income earners. These findings are quite surprising given the vast literature linking both poverty and inequality with health.

The story that emerges is consistent with recent research demonstrating the importance of preserving and increasing social insurance protections for the labor market across the wage spectrum when it comes to population health improvements (Beckfield and Bambra 2016; Jacques and Noël 2022). The finding regarding the positive impacts of status preservation on increases in LE comports with the growing attention of researchers (Benach et al. 2014; Hacker and Rehm 2022; Muntaner et al. 2020; O'Campo et al. 2015; Shahidi, Siddiqi, and Muntaner 2016). Despite its centrality to notions of decommodification, status preservation is a relatively neglected category in the literatures on welfare states and health. As Rohde and colleagues (2017: 307) conclude from their analysis of Australia's Household Income and Labour Dynamics in Australia survey, “it is mostly the prospect of loss rather than deprivation that impacts upon wellbeing.” The finding is also important given the increasing policy attention to the role of status loss in the growth of resentment politics and the resurgence of right-wing nationalism, as previously economically secure individuals are experiencing reduced economic security (Greer 2017). From a policy perspective, the finding is important because it helps distinguish between competing models (i.e., Beveridge vs. Bismarck) of social insurance (Rohde et al. 2017). Additionally, our measures recognize that program financing redistribution may vary across segments of the income distribution as well as over time, even within countries that generally fit one welfare regime type or another. Based on our descriptive analysis, we see that all measures of welfare state generosity have been trending in different directions (mostly flat or upward) depending on the country, thereby making it difficult to discern a single overall trend with regard to welfare state expansion or contraction.

Limitations and Future Directions

As with all research in this area, our analysis is limited largely by constraints on our sample size. Focusing our analysis on established welfare states in western Europe and the OECD with available and reliable data over the time period produces a sample of only 21 countries. Many present-day Eastern European countries did not come into existence until after the 1980s, and reliable estimates of our dependent and independent variables is only available for a limited period following the dissolution of the Soviet Union. Future analysis could expand this approach to examining welfare states’ effects on health by focusing on non-Western democracies.

Additionally, identifying the right model specification strategy is not straightforward. In keeping with recent previous literature (see, e.g., Jacques and Noël 2022), we employed an ADL model. However, this approach raises important questions about the time frame during which we might anticipate welfare policy to have an impact on health in rich countries. We chose to examine LE at birth because it is a commonly used metric that responds to mortality levels in a given year. Extending out to a 10-year time lag may be a more reasonable assumption, but this poses methodological challenges requiring data on all independent variables to be available over a significantly longer time span. Moreover, we uncover conflicts in findings depending on the choice of modeling approach. Our long-run differencing models over the full 40-year period find substantial positive effects on health of all welfare generosity measures except for poverty reduction. However, this approach commands less statistical power and may be sensitive to initial values. This suggests that results are sensitive to different modeling approaches and that more work should be done to examine how findings, interpretations, and policy implications might differ according to modeling strategy.

Conclusion

We conclude that while the study of welfare states and health face many challenges, the questions that this line of study pose could not be more urgent. As the world emerges from the COVID-19 pandemic, there is an opportunity to reassess the orthodoxies that have governed economic and social policy for the past 40 years. There is growing interest in alternative economic models that better balance growth against human well-being, along with a resurgence of interest in more redistributive and universalistic social policies. It is therefore timely to reexamine the relationship between welfare states and health.

We find that western European and OECD countries have experienced broad improvements in LE over the last 40 years, with LE increasing by up to nearly 8 years on average. However, using enhanced measures of welfare state generosity that represent changes in policy commitments by states, we find that countries that have stagnated or reduced their welfare generosity have seen less progress in growing LE than those who have made more generous commitments. We also show greater growth in LE for specific welfare pathways but not others. While additional research is needed, this finding supports research that suggests that raising up the bottom without reducing inequalities in income distribution and supporting the middle class is insufficient to substantially improve human well-being and advance population health.

Acknowledgments

Funding for the social welfare data collection was provided by the National Science Foundation (SES-1756905).

Notes

References