Policy feedback and income targeting in the welfare state

Abstract

In light of ongoing debates about income targeting in the welfare state, this article explores how the design and outcomes of income targeting policies are related to popular targeting preferences. Based on the unique combination of fine-grained opinion and policy indicators in a multilevel analysis, the results show that targeting preferences are indeed empirically related to targeting policies. However, whether these preferences are affected more by the de jure targeting design or the de facto targeting outcome seems to vary between two very different policy domains. In the case of unemployment benefits, the results suggest positive policy feedback: support for high-income targeting increases when unemployment benefits are designed to benefit those with previously higher incomes. For income taxation, by contrast, the results suggest negative policy feedback. In that case, it is not so much the de jure design but rather the de facto outcome that matters: the more taxes effectively work to the advantage of higher-income earners, the less support there is for a tax that levies the same amount on everyone, regardless of income.

Introduction

Should your income determine what you get from the welfare state? Since the early days of social welfare, this has been a contentious and still unresolved question. While it is often argued that welfare state provisions should disproportionally benefit lower incomes to avoid wasting resources on people who are not really in need, others emphasise that expanding the scope of beneficiaries to include the non-poor contributes to the legitimacy of the welfare state and secures popular support for redistribution. Goodin and Le Grand (1987) famously argued that the ‘[non-poor] will defend those parts of the welfare state from which they see themselves as benefiting or likely to benefit, while supporting reductions in those parts from which they will not’ (1987) (see Korpi & Palme, 1998 for another prominent example). Today, more than 30 years after Goodin and Le Grand, there exists a rich body of literature studying the relationship between income targeting policies and welfare preferences, building on policy feedback theory. This theory postulates that popular preferences are shaped by public policies, either in a positive self-reinforcing way or in a negative, self-undermining way (see Campbell, 2012; Busemeyer et al., 2019 for reviews). Importantly, previous studies generally investigated how existing policies influence citizens’ opinions about quantitatively more or less welfare state, usually measured as redistribution preferences (for example, Brady & Bostic, 2015) or as support for more or less spending on (or government responsibility for) various social benefits and services (for example, Jordan, 2013). This article, by contrast, focuses on popular preferences regarding income targeting, which refer to qualitatively different ways of organising welfare provision. Despite their societal relevance, such targeting preferences have thus far received relatively limited attention in the literature (but see Reeskens & van Oorschot, 2013; Barnes, 2015; Sumino, 2016; Roosma, 2016; and Berens & Gelephitis, 2019 for notable exceptions).

While most scholarship has focused on the issue of positive versus negative feedback, this article argues that there is another highly relevant but hitherto underexplored question to address: are welfare preferences mostly influenced by the de jure design of welfare policies, their de facto outcomes, or both? The question is relevant because research has shown that the targeting design of particular policies does not always match well with the actual macro-level targeting outcomes, which are also influenced by external factors such as population composition and benefit take-up (Jacques & Noël, 2018; Marchal & Van Lancker, 2019). Despite its relevance, however, this issue has yet to receive the attention it deserves in policy feedback research. To our knowledge, not a single study has isolated and compared the relative impact of the targeting design and the (perhaps partially unintended) targeting outcome on welfare preferences. This is because these studies typically use policy indicators that either measure the de jure design only (such as legislative information; see for example Reeskens & van Oorschot, 2013) or the de facto outcomes only (such as concentration coefficients; see for example Brady & Bostic, 2015).

This article makes a major contribution to existing literature on social policy feedback, welfare preferences and income targeting by (1) separating targeting designs from targeting outcomes and examining their relative impact on targeting preferences; (2) distinguishing between positive and negative policy feedback; (3) for two specific welfare state policies, unemployment benefits and income taxation. The selection of these particular policies is informed by a most different systems design, which allows to test whether the feedback effects of targeting designs and outcomes are universal or play out differently across policy domains with a highly different redistributive logic and structure. The remainder of the article is structured as follows. In the first section, we zoom in on the targeting concepts used throughout the article. In the second, we build on policy feedback theory to formulate hypotheses about the relationship between targeting policies and targeting preferences (see Appendix I for an overview). More specifically, we first discuss whether policy feedback is expected to be positive or negative. After that, we use the theoretical framework of Soss and Schram (2007) to develop expectations about the relative impact of the de jure design and the de facto outcomes of targeting policies on targeting preferences. In the third section, we describe the data that were used to test this relationship empirically. To measure the targeting design, we rely on unique institutional indicators based on model family simulations, which are not dependent on outcome dimensions such as population composition. To gauge macro-level targeting outcomes, we calculate concentration coefficients, which measure the extent to which benefits (or taxes) actually end up among different income groups, based on the European Union Statistics on Income and Living Conditions (EU-SILC). In the fourth section, we investigate how the targeting design and outcomes are related to popular targeting preferences drawing on European Social Survey (ESS) data. The fifth section concludes and suggests some avenues for future research.

Targeting concepts

Targeting can be regarded as the process of identifying and delineating groups of beneficiaries of and contributors to social policies. Although there are plenty of criteria for targeting (e.g. age, household type), the concept of targeting is most commonly associated with policies that focus on income (or more broadly, means) to define who should receive support and how much (i.e. the benefit side of the welfare state) (Saunders, 1991), and vice versa, who should pay for such support and how much (i.e. the tax side of the welfare state) (Morel & Palme, 2012). Two defining characteristics of income targeting are thus its direction, towards lower or higher income groups, and its degree, making large or small income distinctions. In this article, we zoom in on the targeting design of two specific policies: unemployment benefits and income taxes. In both policy domains, we find different income targeting rationales across countries, which makes them interesting test cases to explore to what extent the feedback effects of targeting policies on targeting preferences are domain-specific. In the following, we mostly conceptualise targeting of unemployment benefits in terms of benefit amounts and income taxes in terms of tax rates, as this represents the most common and intuitive interpretation of the respective policy domains in scientific literature and public debate.

In our conceptual framework (see Table 1), we argue that on the benefits side of the welfare state, ‘low-income targeting’ applies to a situation in which lower income groups receive higher amounts compared to higher income groups. In its most extreme form, low-income targeting consists of means-tested social assistance schemes that cater to the poor and the poor only. There are numerous alternative ways, however, in which welfare states can give more to the poor. One prominent example is ‘targeting within universalism’, which refers to policies that are universally accessible but at the same time provide supplements for low income categories (Skocpol, 1991; Jacques & Noël, 2021). In contrast, ‘high-income targeting’ entails that higher income groups are entitled to higher benefit amounts.

Table 1. Stylised overview of income targeting concepts

In most countries, unemployment benefits are organised as a workers’ social insurance. Within unemployment insurance systems, however, there are two dominant rationales. The Beveridgean approach, mainly found in the Anglo-Saxon countries, advocates a flat-rate benefit for everyone becoming unemployed. This rationale aligns most closely to the equal treatment ‘no targeting’ approach depicted in Table 1. The Bismarckian approach, on the other hand, which is common in the continental European countries, entails that those with higher (past) wages are entitled to higher benefits, so that acquired living standards can be maintained after the occurrence of a social risk. As such, social benefits closely mirror the equity-based logic of private insurance: higher earnings entail higher contributions, which lead (at least in theory) to higher replacement rates. Still, a design aiming to guarantee equal replacement rates across different income groups may have regressive outcomes in terms of benefit amounts (high-income targeting in Table 1), as equal replacement rates will result in higher benefits for those that were (previously) better off.

Following Musgrave and Thin’s seminal study (1948), tax policy research has developed a vocabulary of their own to consider tax design, focusing on the progressivity built into tax systems (e.g. Dingeldey, 2001; Morel & Palme, 2012). This term refers to the extent lower incomes face lower tax rates (rather than amounts), which translates to low-income targeting in our framework (see Table 1). As such, a purely proportional tax, in which everyone pays an equal share of his or her income, tends to be seen as non-targeted (Musgrave & Thin, 1948). Some might disagree with such a notion, arguing instead that a system is non-targeted if everyone is liable to pay the same tax amount. In the past, such systems existed in the form of a ‘poll tax’ (or head tax), and currently, some local or specific taxes are still organised according to this principle. However, usually, such targeting schemes are not considered non-targeted. Rather, as a poll tax represents a lower tax burden for higher incomes, it is recognised as being more beneficial for higher incomes (hence included under high-income targeting in Table 1).

Policy feedback theory and the case of income targeting

Positive versus negative policy feedback

A growing body of literature emphasises that citizens’ opinions about welfare provision are influenced by the welfare policies they are exposed to (e.g. Jordan, 2013; Raven et al., 2011; Soss & Schram, 2007). Most scholars in the field argue that such policy feedback mainly works in a positive, self-reinforcing way: the more a welfare policy is pursued, the more likely citizens are to endorse it (Busemeyer et al., 2019; Pierson, 1993). The main argument is that people generally ‘learn to love’ the current state of affairs through complex processes of socialisation and interest formation (Arts & Gelissen, 2001; Mau, 2004).

Reeskens and van Oorschot (2013), for example, maintain that targeting preferences are likely to follow a welfare regime pattern. They expect support for equity-based high-income targeting to be greatest in conservative-corporatist regimes, because of the dominance of social insurance schemes. Need-based low-income targeting is assumed to be most popular in liberal welfare regimes, given their relatively strong reliance on means-tested social assistance. Equal treatment, finally, is understood as the most preferred option in social-democratic regimes, which are characterised by universal provision of welfare. Similarly, the authors argue that targeting preferences are also influenced by the institutional design of individual welfare programmes, which each follow their own policy rationale. Here, the expectation is that high-income targeting is most popular in case of earnings-related insurance benefits, while low-income targeting and equal treatment are most liked in case of means-tested and flat-rate benefits, respectively.

Building on the perspective of positive policy feedback, we expect that:

H1a: the more welfare policies are targeted at lower (or higher) incomes, the more citizens will support low-income (or high-income) targeting (positive feedback hypothesis)

It is increasingly recognised, however, that feedback effects may just as well operate in a negative, self-undermining way (Campbell, 2012; Fernandez & Jaime-Castillo, 2013). In such a scenario, citizens are more likely to react against a welfare policy the more it is pursued, presumably because they strongly dislike that policy in the first place. Roosma et al. (2016), for example, argue that in countries with less progressive tax systems, there will generally be a higher demand for progressive taxation. The underlying assumption is that most people dislike tax policies that achieve little redistribution from the rich to the poor and will therefore react with an increased demand for progressivity against any policy drift away from progressive taxation. Taking into account possible negative policy feedback effects, it seems equally plausible that:

H1b: the more welfare policies are targeted at lower (or higher) incomes, the less citizens will support low-income (or high-income) targeting (negative feedback hypothesis)

De jure policy design versus de facto policy outcomes

While a great deal of policy feedback scholarship has focused on the issue of positive versus negative feedback (Busemeyer et al., 2019), we argue that another question needs to be asked: are welfare preferences mostly influenced by the de jure institutional design of welfare policies, their de facto outcomes, or both? Most feedback scholars seem to assume that popular preferences are mainly driven by the institutional design of public policy. The policy design written in formal laws may however deviate significantly from the ‘on the ground’ policy outcomes. This also applies to income targeting in welfare provision: policies that are designed to target low-income (or high-income) groups may be less targeted in reality than they appear on paper (see Appendix II). This is because the actual targeting outcome of a welfare policy not only depends on its institutional design, but is also contingent on external factors (Jacques & Noël, 2018; Marchal & Van Lancker, 2019).

One important factor in this regard is the composition of the target population. If that population consists primarily of low-income people, it can be the case that the policy outcome will be targeted towards low-income people even if the policy design intends to target high-income categories. Imagine, for example, an unemployment benefit that is organised as a typical social insurance in which higher wages prior to unemployment give rise to higher benefits. Although the policy intention is to benefit higher incomes more, the outcome will be very different if the unemployed population consists predominantly of low-skilled workers with a low income. In such a scenario, the outcome will actually be targeted towards low-income categories. Another example pertains to the domain of income taxation, where targeting outcomes may also diverge from targeting designs. When policymakers devise a progressive system, they intentionally levy a higher tax rate on higher incomes. However, in countries with an extremely low level of market income inequality, the outcome of a progressive system might be closer to a poll tax, as most citizens will pay more or less the same amount because they are taxed on more or less the same income. A second important factor for understanding potential discrepancies between de jure policy designs and de facto policy outcomes is the non-take-up of benefits and services. It could well be, for example, that an unemployment insurance that rewards higher wages (prior to the social risk) with higher benefits will end up being targeted towards lower incomes if there is a great deal of non-take-up among high-income earners (for example because they fear the stigmatisation that comes with it). Similarly, opaque tax codes and complicated forms may lead to disproportionate take-up of tax advantages among the highly educated or by those who can afford professional assistance when filing their taxes.

Earlier opinion research did not take the distinction between targeting designs and targeting outcomes into account. Reeskens and van Oorschot (2013), for one, report positive feedback effects of old-age pension provisions, with high-income targeting being more popular in countries with earnings-related pensions, and equal treatment having greater support in countries with flat-rate pension benefits. However, the selected policy indicators were operationalised on the basis of legislative information which measures de jure designs but not de facto outcomes. There are also some studies that have explored how people’s support for progressive taxation is affected by the actual degree of tax progressivity (Barnes, 2015; Sumino, 2016; Berens & Gelephitis, 2019). These studies generally find no significant relationship between the two. However, because these studies tend to rely on targeting outcome measures (most notably the actual degree of tax concentration across different income groups), they are unable to separate the impact of targeting designs and outcomes. The same applies to the studies that examine how tax and/or benefit progressivity influence preferences for more or less redistribution, or preferences for more or less welfare state (e.g., Jordan, 2013; Brady & Bostic, 2015; Jacques & Noël, 2018; Sumino, 2018).

According to the work of Soss and Schram (2007), the strength of policy feedback effects is a product of two factors: visibility and proximity. The first, visibility, refers to the salience of a policy for the broader public, with the basic assumption being that higher salience is more likely to arouse feedback effects. The second factor, proximity, ‘concerns the direct-versus-distant form in which a policy is encountered: the extent to which it exists as a tangible presence affecting people’s lives in immediate, concrete ways versus existing as a distant object appraised for its effects elsewhere’ (Soss & Schram, 2007, p. 121). Here, the assumption is that policies which are closer to people (either geographically, temporally or psychologically) are more likely to produce feedback effects.

Applied to our distinction between the targeting design and the targeting outcome of welfare policies, opposing expectations emerge. On the one hand, one might expect outcomes to matter more than designs in shaping targeting preferences. It seems plausible that people have greater awareness of the actual outcomes of a policy, which they experience first-hand through their day-to-day interactions with welfare policies. This should make targeting outcomes highly salient and visible to the general public. The targeting design as it is written in formal law, by contrast, might be less visible and more distant to most citizens, thereby limiting its impact on welfare opinions. This also echoes one of the most fundamental critiques on policy feedback theory: how confident are we that people have sufficient knowledge about the institutional design of a policy for it to influence their opinions? ¹ On the other hand, there are also plausible arguments for assuming that targeting outcomes will matter less than targeting designs. It could well be that the institutional design is actually easier to understand and interpret than the policy outcomes, which come about through various complex processes (as we have seen above). The official ‘rules of the game’ might therefore be more visible and proximate to citizens than the actual outcomes that are being produced. For example, knowledge about the replacement rate of unemployment benefits or the progressivity of the tax system might be more common than knowledge about the actual distribution of benefits and taxes among different income groups.

Based on the proximity-visibility model of Soss and Schram (2007), we formulate two opposing hypotheses:

H2a: the relationship between targeting preferences and targeting outcomes is stronger than their relationship with targeting designs (outcomes-matter-more hypothesis)

H2b: the relationship between targeting preferences and targeting designs is stronger than their relationship with targeting outcomes (designs-matter-more hypothesis)

Data and methods

Dependent variables: targeting preferences

To measure popular preferences about income targeting, we use data from the European Social Survey (ESS), a well-established general population survey that is conducted biennially across a wide range of European countries. More specifically, we analyse two items included in Round 4 of the ESS, which was gathered in 2008/09 among a sample of 56.752 respondents living in thirty-one different European countries. ² The items ask about the desirability of income targeting in two welfare state domains: unemployment provision and income taxation. As for the first, respondents were asked to choose whether unemployment benefits should be (a) higher for lower earners (i.e. low-income targeting), (b) equal for high and low earners (i.e. no targeting), or (c) higher for higher earners (i.e. high-income targeting). For income taxation, respondents had to choose one of the following options: (a) higher earners pay a higher share of their earnings in tax, (b) high and low earners pay the same share (or percentage) of earnings in tax, or (c) high and low earners pay the same amount of money in tax. The first option represents a progressive tax system that works to the advantage of people with a lower income (i.e. low-income targeting). The second option represents a proportional (flat) tax system, which comes closest to an equal-treatment-policy because it levies the same tax rate on high and low incomes (i.e. no targeting). The third option embodies a poll tax, which actually benefits higher incomes because they pay a smaller proportion of their income in taxes compared to lower incomes (i.e. high-income targeting).

Independent variables: targeting policies

Targeting outcomes

The most common way to measure the achieved extent of targeting is by looking where in the income distribution benefits end up, or alternatively, which parts of the income distribution are faced with the highest tax liabilities. Kakwani (1977) proposed the concentration coefficient as a parsimonious way to track the extent to which benefits end up with the poorer or richer entitlement units. Formally, the concentration coefficient is expressed as follows:

$${\rm{TI}}_{\rm{CC}} ( {B,Y}) = - 2{\rm{Cov}}\left({B \over {\mu ( B)}},\;( {1 - G( Y)} \right)$$

with benefit B, income distribution Y and G(Y) the cumulative distribution function of Y. As long as it is calculated on positive values, the concentration coefficient assumes values between −1 and 1. A value of −1 implies that the poorest entitlement unit in the income distribution receives all the benefits (‘low-income targeting’), whereas a value of 1 indicates that all benefits are targeted at the richest unit (‘high-income targeting’). A value of zero indicates that there is no association between the benefit amount and the entitlement unit’s place in the income distribution. This is the value that the concentration coefficient shows when every unit receives the same benefit amount (‘no targeting’). In order to assess who is liable to pay the highest tax rates, we replace benefits B in the above formula by the effective tax rates, calculated by dividing the taxes paid by an individual with his or her gross income.

We use the concentration coefficients of unemployment benefit amounts and of tax rates as independent country-level variables of targeting outcomes (see Appendix VI). These indicators are calculated for 2008 on the basis of information from the European Union Statistics on Income and Living Conditions (EU-SILC), a representative survey conducted in a broad range of European countries.

Targeting designs

To operationalise the design of unemployment benefits and income taxes, we use the targeting design indicator proposed by Marchal and Van Lancker (2019). This indicator builds on a stylised and fictional income distribution of hypothetical households, taken from the OECD Benefits and Wages Model (OECD, 2014) representing the situation in 2008. These households have the same characteristics, except for their market incomes which increase with fixed increments equal to 1% of the average wage. The net disposable income of these hypothetical households is based on running these market incomes through models of the respective national tax-benefit systems. This fictional income distribution is calculated for different household types (a single, a couple with two children and a lone parent with two children). The underlying households (and hence the underlying fictional income distribution) are equal in each country included in the analysis. On these fictional income distributions, we calculate for each country the concentration coefficient for the benefits and taxes we are interested in. In order to calculate the targeting design of unemployment benefits, we calculate the concentration coefficient of the unemployment benefit against previous incomes for each typical family. We take the average of the resulting three targeting design indicators in order to reflect possible differences related to different minima and maxima for families with children. For tax rates, we calculate the concentration coefficient of the applicable tax rates (calculated as payables taxes relative to gross income, at 1% of average wage increments) against current incomes, for a hypothetical single person household. These institutional data, that capture how a tax-benefit system is designed to work, regardless of contextual factors, allow to assess the extent to which benefits and taxes are designed to end up with richer or poorer entitlement units.

As for the targeting outcomes indicator, a value closer to −1 indicates low-income targeting, a value closer to 1 implies high-income targeting. Note that, in the case of taxes, this means that values closer to −1 show that low-income families are faced with lower tax rates, and high-income families with higher tax rates. It has previously been shown that this targeting design indicator calculated on a fictional income distribution meaningfully gauges the extent of income targeting in different benefit systems, and captures the difference in the strength and direction of targeting across countries (Marchal & Van Lancker, 2019).

Statistical modelling

To test our hypotheses, we estimated a series of multilevel models (with individuals nested in countries), in which targeting preferences act as individual-level dependent variables, and targeting policies as country-level independent variables. Because our dependent variables have three unordered answer categories, we analyse them by means of multinomial regressions. In both cases, the ‘no targeting’ option is chosen as the reference category, so that the regression coefficients express the likelihood that respondents prefer low-income or high-income targeting over equal treatment. Our analytical procedure is structured as follows. First, the (standardised and grand-mean centred) country-level variables are introduced separately in the models (M1–M2 and M5–M6 in Table 2), as is generally recommended in multilevel analyses with a relatively small higher-level sample size. ³ Second, we include them simultaneously to examine their relative importance in explaining popular targeting preferences (M3 and M7 in Table 2). In line with recommendations from prior research (for example Sumino, 2016; Berens & Gelephitis, 2019), we also estimated models with country-level controls for the degree of income inequality, the level of economic development and the size of spending/taxation. Even after including these controls, the main effects of the targeting indicators remain largely unchanged, adding to the robustness of our findings (see Appendix IV). All country-level variables reflect the situation as it was in the same year as the ESS was conducted, i.e. 2008. Third, we control for the composition of the sampled population, by including respondents’ age, gender, educational level, employment status, and subjective income (M4 and M8 in Table 2). Subjective income was preferred over income because the latter has a high-item nonresponse, which would exclude a large group of respondents (i.e. 15.632) from the analyses. The item we use probes people’s subjective feelings about their household income and has four response categories: (a) living comfortably, (b) coping, (c) finding it difficult, and (d) finding it very difficult. As a robustness check, however, we re-ran the models with income included instead, measured as the self-reported total household net income. The main findings prove to be robust across these different model specifications (see Appendix V).

Table 2. Multilevel regressions estimating the effects of targeting policies on targeting preferences

Notes: The figures reported in the table are standardised linear regression coefficients. Analyses weighted by: post-stratification weights provided by ESS. AIC = Aikake Information Criterion; BIC = sample-size adjusted Bayesian Information Criterion.

*p < 0.10, **p < 0.05, ***p < 0.01.

Results

Cross-national variation in targeting preferences

We start our analysis with a description of Europeans’ preferences regarding income targeting in unemployment benefits and income taxation (see also Appendix III). With regard to unemployment provision, the left-hand panel in Figure 1 shows that in most European countries, equal benefits for higher and lower earners (no income targeting) is the most preferred policy option. This is especially the case in countries such as Croatia (80.3%), Denmark (78.1%), Ireland (73.1%), the United Kingdom (71.6%), Greece (71.1%), Estonia (69.3%), Poland (67.1%), and Belgium (61.8%). By contrast, in a number of other countries – including Portugal (62%), Germany (59.2%), Spain (57.1%), Latvia (51%), and Slovakia (50.1%) – the most popular choice is to grant higher benefits to higher earners (high-income targeting). In two countries, the Czech Republic and Israel, the public is about equally divided between no targeting and high-income targeting in unemployment provision. Low-income targeting (i.e. giving higher benefits to lower earners) is generally least popular, except in Turkey (38.6%), Hungary (24.1%), Slovenia (20.2%), and Greece (19.6%), where it is (joint) second choice.

Figure 1. Targeting preferences for unemployment benefits and income tax, per country.

With regard to income taxation, the right-hand panel in Figure 1 shows that in all countries, most citizens either favour a progressive tax system (low-income targeting) or a proportional system (no targeting). In 17 out of 29 countries, a progressive system is preferred over a proportional one (with a difference of at least five percentage points). The preference for progressive over proportional taxation is most outspoken in Israel (+51.3%) and Ireland (+39%). By contrast, in Latvia (+5.7%), France (+8%), and especially Croatia (+15.8%) and Portugal (+21.3%), a proportional tax is preferred over a progressive one. In a number of other countries, such as Germany, Romania, and Slovakia, there is hardly any difference in support for a progressive and a proportional tax (difference is less than five percentage points). A poll tax, in which higher and lower earners pay the same amount of money (high-income targeting), is clearly least popular. In most countries, it is a rather marginal choice, with percentages below 10%. Acceptance of a poll tax is highest in the United Kingdom (11.9%), Estonia (11.5%), and Denmark (11.3%).

Multilevel analyses

Model 1 demonstrates that the targeting design of unemployment benefits is positively related to popular preferences for high-income targeting (but not for low-income targeting). The more unemployment benefits are designed to benefit previously higher incomes, the more citizens support the idea that unemployment benefits should be higher for higher earners (compared with equal treatment). This suggests positive policy feedback (H1a). By contrast, the actual targeting outcome of unemployment benefits, as measured by their concentration coefficient, has no significant effect on popular targeting preferences (M2–4). This shows that the targeting design of unemployment benefits has a stronger impact on public opinion than the targeting outcome, which confirms our design-matters-more hypothesis (H2b).

In the case of income taxation, however, findings are more mixed. On the one hand, we find that the less tax systems are designed to benefit lower incomes, the less people favour a progressive tax system (compared to a proportional one) that primarily benefits low-income groups (M5). This finding, again, seems to confirm the positive feedback hypothesis (H1a). On the other hand, we also find evidence for the negative feedback hypothesis (H1b): support for a poll tax – which mostly benefits higher incomes – is lower in countries where the tax system is designed to benefit higher incomes to a greater extent (M5). It must be noted, however, that both effects turn statistically insignificant once we also include the respective targeting outcome (M7) and control for population composition (M8). In the domain of taxes, it is actually the targeting outcome that is most strongly and robustly related to targeting preferences, thereby confirming our outcomes-matter-more hypothesis (H2a). We find that the more taxes benefit higher income groups, the less support there is for a poll tax levying the same tax amount on everyone regardless of their income. Since a poll tax favours higher earners (who would pay a lower tax share), this suggests support for the negative feedback hypothesis (H1b). No such relationship was found between the targeting outcome and preferences for a progressive tax (see also Barnes, 2015; Sumino, 2016; Berens & Gelephitis, 2019).

Taken together, the findings demonstrate that targeting outcomes have a stronger impact on targeting preferences in the case of income taxes, while targeting designs have a stronger influence in the case of unemployment benefits.

Conclusion and discussion

This article examined how targeting policies are related to targeting preferences, in two policy domains: unemployment benefits and income taxes. Based on the unique combination of high-quality opinion data and fine-grained policy indicators, it makes three important contributions to the growing literature on (social) policy feedback. The first has to do with the distinction we draw between de jure policy designs and de facto policy outcomes. It is broadly recognised in social policy literature that outcomes very often deviate from designs, because they are also influenced by external factors (Jacques & Noël, 2018). This also proves to be the case for income targeting in unemployment benefits and taxes (see Appendix II). For some reason, however, this important insight has not been given the attention it deserves in the (social) policy feedback literature, which continues to rely heavily on policy indicators that are unable to separate designs from outcomes. The fact that our targeting indicators are constructed to distinguish between designs and outcomes provides us with a unique opportunity to scrutinise their relative impact on popular welfare preferences.

The article’s second contribution is that it demonstrates the existence of both positive and negative policy feedback effects. From its early days, the dominant assumption in the literature has been that public policies tend to reinforce themselves: the more a policy is pursued, the more people will adhere to it (Campbell, 2012; Pierson, 1993). It was only later that also the notion of negative, self-undermining policy feedback received greater attention in the literature (e.g. Fernandez & Jaime-Castillo, 2013). Most scholars in the field now acknowledge that public policies may just as well be supported less the more they are pursued, if citizens disagree with or feel dissatisfied about those policies (Busemeyer et al., 2019).

The third contribution of the article is that it illustrates how policy feedback may play out differently across different policy domains. Our analyses clearly show that the nature of policy feedback differs between the two policy domains under examination: while design matters more than outcomes in the case of unemployment benefits, it is the other way around in the case of income taxation. These findings suggest that it is tax outcomes and unemployment benefit design that are more visible and proximate to citizens, thus leading to stronger feedback effects (Soss & Schram, 2007). Europeans might indeed have a fairly good sense of how taxes are actually distributed among different income groups, because most of them pay taxes and thus have some first-hand experience with the system. Furthermore, given the complex structure of most existing tax systems – which are full of special rules, exemptions, and deductions – it seems plausible that citizens feel more acquainted with and knowledgeable about the actual outcomes of the system rather than its institutional design. Unemployment benefits, by contrast, tend to have a programmatic structure that is somewhat easier to comprehend and are – because of its social stratification – outside the everyday lives of the many citizens who lack direct experience with these benefits. This might explain why the ‘official rules of the game’ embedded in the de jure policy design exert a stronger influence on targeting preferences in the domain of unemployment benefits than in the domain of taxes. Additionally, our analyses indicate that whether policy feedback is predominantly positive or negative also depends on the policy domain under consideration. In the case of unemployment benefits, we find evidence of a positive feedback effect: the more these benefits are designed to advantage higher earners, the more people support high-income targeting. For income taxation, by contrast, the most robust finding points towards negative feedback: the more taxes effectively work to the advantage of higher incomes in their achieved outcomes, the less support there is for a poll tax, which ultimately benefits mainly higher earners.

These findings add important insights to our knowledge on the feedback effects of targeting policies generated by prior research, which has mostly focused on attitudes towards progressive taxation, or towards redistribution and the welfare state more generally. Regarding the first, we can confirm the finding from previous studies that tax progressivity has no impact on preferences for progressive taxation (Barnes, 2015; Sumino, 2016; Berens & Gelephitis, 2019). However, we do find that greater high-income targeting decreases preferences for a tax system that mostly benefits higher incomes, which has not been part of prior research. With regard to the second, there is also some (albeit limited) evidence that a higher degree of high-income targeting in the benefit system as a whole increases support for redistribution (Brady & Bostic, 2015). These negative feedback effects stand in stark contrast with what we observe in the domain of unemployment benefits, where we see that greater high-income targeting is associated with a higher level of support for such high-income targeting (see also Reeskens & van Oorschot, 2013 for old-age pensions). The most likely explanation here lies in the fact that the two policy domains adhere to highly diverging logics: while unemployment benefits are first and foremost designed to allow people to maintain their acquired living standards and social status, income taxes usually aim to achieve greater redistribution from high-income to low-income earners. In that sense, it seems logical that people want more high-income targeting of unemployment benefits when these are already higher for higher incomes, and less high-income targeting of taxes when these currently work more to the benefit of higher incomes. ⁴

All in all, our analyses provide a fruitful basis for future policy feedback research to build on. One potential avenue is to consider a broader set of feedback types than the ones examined in the present article. As suggested by a recent typology, policy feedback effects can be classified according to three dimensions: (1) the direction of feedback, (2) its broadness/scope, and (3) its time lag (Busemeyer et al., 2019). Although a distinction was made between positive and negative feedback in terms of direction, we did not consider the possibility of so-called accelerating feedback, which ‘occurs when the expansion of a given policy creates further [italics added] support for an expansion of the provision’ (Busemeyer et al., 2019, p. 145). To study this, future research could use dependent variables that explicitly assess preferences for increasing (or decreasing) the extent of income targeting relative to the status quo. Additionally, we see merit in investigating how targeting preferences are influenced by the degree of real or perceived (in)congruence between de jure designs and de facto outcomes; both at the macro level of countries or regions and the micro level of individuals or households. With regard to the dimension of broadness/scope, we analysed feedback effects of targeting policies in two specific policy domains, i.e. unemployment benefits and income taxes. Although future research should definitely continue to examine feedback effects of specific targeting policies in other areas of the welfare state (including comparisons of policy domains that are more similar in terms of their targeting logic than the domains considered in this article), it should also look for more-general feedback effects from larger welfare constellations and regimes. In terms of the time dimension, future studies should inspect whether the feedback effects we observe are driven by longer-term institutional dependencies, more-recent policy dynamics, or a combination of both.

Relatedly, the cross-sectional nature of our data leaves open the possibility that the causal mechanism actually runs in the opposite direction, from targeting preferences to targeting policies (Raven et al., 2011). From the perspective of policy responsiveness theory, this scenario is plausible, especially when it comes to the de jure targeting design, which is (partly) driven by the choices and intentions of political actors that have electoral reasons to respect the wishes of the public (Brooks & Manza, 2006). It is less obvious with regard to de facto targeting outcomes, which are also influenced by external factors. This underlines the crucial importance – also for policy responsiveness scholars – of using indicators that can separate policy designs from policy outcomes. Unfortunately, we currently lack the longitudinal data needed to uncover the causal direction of the relationship between targeting policies and targeting preferences. Future research should aim at tracking targeting policies as well as targeting preferences over time, both between and within countries.

Finally, we encourage future studies to also take stock of potential heterogeneity in popular targeting preferences. In our theoretical framework, we made the assumption that the targeting designs and outcomes of social policies are equally visible and proximate to all members of the population. This is, however, a rather unlikely scenario. It seems plausible, for example, that unemployment benefits are more proximate and visible to people who have (recently) been unemployed, or face a disproportionally high risk of becoming unemployed. This should make policy feedback effects especially likely among these particular groups (Soss & Schram, 2007). In a similar vein, we should expect to see stronger feedback effects among people who pay taxes compared to those who do not (such as most full-time students). Future research should delve deeper into this heterogeneity by examining whether the effects of policy designs and outcomes vary across different groups in society. ⁵