Unemployment risk-sharing in the EU: How policy design influences citizen support for European unemployment policy

This article explores public preferences for European unemployment programs explicitly discussed in actual policymaker debate. European policymakers have been considering European-level Unemployment Risk Sharing (EURS) to stabilize member-state economies and provide a safety net for the unemployed. Using a conjoint experiment conducted in 13 European member states, we analyze public support across six crucial policy dimensions of EURS. The findings reveal that (a) overall support for EURS policies is broad and substantial, but sensitive to particular policy mixes; (b) citizen support is conditional on the program being generous and on coverage being limited to countries providing education and training and individual beneficiaries looking for and accepting work; and (c) cross-country variation is modest and most prominent with respect to cross-country redistribution.


Introduction
A major debate in comparative political economy concerns the development of social protection in the European Union (EU). In the past decade, the European Commission and the Council have presented several proposals to establish Eurozone-level risk-sharing (European Commission, 2017;Van Rompuy et al., 2012). More recently, the European Commission agenda has prominently included proposals to establish European-level reinsurance of national unemployment benefits (Von der Leyen, 2019: 10), and the initial EU response to the COVID-19 pandemic has included the program of Support to mitigate Unemployment Risks in an Emergency to disburse benefits and 're-insurance' to citizens (European Commission, 2020: 3).
These ideas about EU unemployment benefitswhat can be generically called European-level Unemployment Risk Sharing (EURS)constitute an extensive form of international cooperation in socio-economic protection, among the most jealously defended prerogatives of national sovereignty. Not surprisingly, then, such politicaleconomic pooling of sovereignty remains controversial among scholars, policymakers, and publics. Support for EURS must overcome concerns about the level and distribution of costs, and about how EU-level provisions will relate to existing national welfare states (Baute et al., 2018a). More fundamentally, the development of EU-level social provisions is at the front lines of debate about whether European integration should be about not only market liberalization but also about meaningful social embedding by developing a 'social pillar' of integration (Scharpf, 2002).
Scholarship on public opinion about EU social policies has yielded diverging findings. Some uncover considerable support for EU cross-border solidarity (Ferrera and Pellegata, 2017;Gerhards et al., 2020), while others find modest support, especially as compared to support for national policies (Dolls and Wehrhöfer, 2021;Lahusen and Grasso, 2018). Bremer et al. (2020) suggest that public opinion depends on the policy domain and instrument: support is strongest for cross-national solidarity towards natural disasters and lowest towards struggles with debt, and precautionary ex ante instruments (like training or education) are preferred over ex post remedies (like income transfers). Significant differences across countries are driven by macroeconomic contexts and expectations as to what EU involvement promises for one's country (Vasilopoulou and Talving, 2019). Finally, scholars reach diverging conclusions about how much left-right ideology (Gerhards et al., 2020;Kleider and Stoeckel, 2019), knowledge and cosmopolitan orientations (Armingeon, 2021;Bechtel et al., 2014) structure support for European solidarity.
Yet, there is disconnect between empirical assessments of EU social policy preferences on the one hand and actual policy debate on the other. First, many studies focus on broad principles of social policy far removed from particular policy instruments on offer (Kuhn and Kamm, 2018). Second, more concrete studies often do the opposite: zoom-in on particular policy features rather than on dimensions characterizing policy debate. For instance, research on international bailouts in Germany (Bechtel et al., 2017) and Italy (Franchino and Segatti, 2019) does not explore the development of supranational social policy. Third, many studies emphasize European-level provisions without clarifying whether EURS will supplant or supplement national unemployment insurance. Hence, existing scholarship, while yielding extensive insight into solidarity and EU social policy, has not sufficiently clarified how policy design impacts support for EURS.
This article studies how public opinion responds to differences in EURS policy design reflecting important policy debates. We focus on six policy dimensions central to policy debate on EURS, about which we have theoretically informed and preregistered hypotheses on public preferences. 1 We conducted a population-based conjoint survey experiment in 13 EU member states in late 2018, which allows isolating the causal effect of each policy feature on support for EURS. It also provides leverage to inductively judge which policy packages yield the most and least public support. We focus on the experimental components of the survey, capturing how policy dimensions affect support for EURS, not on how country-or individual-level characteristics alter such effects.
We find that overall support for EURS is broad and substantial, contingent upon particular policy mixes. Support is stronger for policy mixes that are generous while requiring increased tax burden than for mixes that are less generous and entail no such burden, so long as benefits are made conditional upon labor-market activation and provision of education and training. Despite cross-country variation, such majorities are stable across our sample countries. Hence, EURS explicitly aiming to supplement national-level insurance can be expected to command substantial political support, contingent upon the assistance being generous and having clear conditionality.

Generosity
EURS can vary in the generosity of payments to unemployed beneficiaries, such as in replacement rates and payment duration (Beblavy and Lenaerts, 2017;Gallego and Marx, 2017). We hypothesize that individuals prefer more generous policies that maintain or improve the domestic situation in countries with either high or low benefit levels (e.g. Gelissen, 2000), though for varying reasons. Following De Vries' (2018) benchmarking theory, we can expect that citizens benchmark EU benefits against their own country's benefits. Respondents in countries with high benefit levels may fear that a relatively low floor in a common EU scheme creates downward pressure on their own country's benefits, even if countries can pay out more than the EU floor. Citizens tend to oppose social cutbacks (Häusermann et al., 2019), any EU dismantling of national-level social protection, and 'downward convergence' in European social policymaking (Sinn and Ochsel, 2003). In contrast, for countries with low benefit levels, we expect more unvarnished support for EU-level initiatives constituting easily recognizable improvement upon the status quo.
Although some studies articulate conditions under which retrenchment is publicly palatable (Giger and Nelson, 2011;Rehm et al., 2012), we expect support for more generosity. Recent results for single countries, such as Gallego and Marx's (2017) conjoint experiment in Spain, find that proposed benefit generosity strongly increases public support for labor-market reforms, while proposed pension retrenchment in Switzerland reduces support (Häusermann et al., 2019).
H1: Individuals will tend to lend more support to EURS packages that include more generous support for unemployed than to packages including less generous support.

National versus EU administration
European integration involves a fundamental tension between supranationalism and intergovernmentalism (Tsebelis and Garrett, 2001), namely whether power should mainly remain with member states or be moved to the supranational level. This applies particularly to 'core state powers', which include taxation and employment policies (Genschel and Jachtenfuchs, 2016: 44). Towards such competencies, polities are particularly hesitant to cede national power because doing so implies surrendering national sovereignty. Public opinion can be a 'constraining dissensus' (Hooghe and Marks, 2009) to integration because these issues are highly salient and politicized parts of national identity. Indeed, Bremer et al. (2020: 63) reveal European publics to be particularly skeptical of European integration in debt and unemployment policy.
Hence, an important discussion is whether EURS should be administered centrally by the EU or decentrally by member states. Centralized administration provides more EU-level control, more pooling of sovereignty, more supervision of national actions than a decentralized system. As Hooghe and Marks (2009) argue, functional pressures towards joint supranational rule are often in tension with citizens' preference for self-rule: citizens may have concerns with excessive oversight over national practices, seek to avoid regulatory harmonization or simply prefer decentralized systems. Hence, citizens likely want administration close to their domestic context, minimizing interference from supranational bureaucracies.
H2: Individuals will tend to lend more support to EURS packages that include national-level administration than to those with EU-level administration.

Between-country redistribution
A fundamental question in public debate around EURS is whether a policy should, ex ante, allow long-term between-country redistribution, or restrict itself to pure insurance with no net beneficiaries or net contributors to the scheme. The principle of 'pure insurance' can be imposed on EURS by stipulating that in the long-run countries can only receive as much as they have paid into the scheme (Arnold et al., 2018). This partly captures debate on distinct redistributive and insurance purposes of unemployment-related social policy (Rehm et al., 2012). In the EURS context, the intuition behind pure insurance is that redistribution can invite freeriding and risks moral hazard, like other areas of European fiscal pooling of sovereignty (Kanthak and Spies, 2018). Redistribution may presume more trust and commitment among member states than currently obtains. But EURS can be more fiscally stabilizing if long-term redistribution between net contributors and net beneficiaries is built into the scheme. To capture the real policy debate on EURS, one should distinguish a 'pure insurance' variant from two alternatives allowing redistribution: either a 'general redistributive' scheme where any given country can draw-out more than it pays in or a 'rich-to-poor redistributive' scheme where only poor countries can do so (Schmid, 2019).
We can anticipate many individual and, particularly, cross-country differences in attitudes, based on whether individuals are in countries likely to be net contributors or net beneficiaries of EURS (e.g. poor vs. rich countries, debtor vs. creditor, etc.). On average, however, we expect that individuals want to minimize cross-border financial transfers of any EURS scheme. Support for redistribution depends on trust, and Europeans generally trust other countries' citizens and governments less than their own (Deutschmann et al., 2018). In many policy discussions about European social arrangements, concerns about cross-border redistribution loom large (Bechtel et al., 2017). We therefore expect individuals to prefer policies where in the long run no participating country can be a net-beneficiary or net-contributor. Hence, individuals should prefer 'pure insurance' schemes where a country's long-term benefits are capped by its contributions to the scheme.
H3: Individuals will tend to lend more support to EURS packages that include the 'pure insurance' option than to those including either the 'general redistributive' option or the 'rich-to-poor redistributive' option.

Impact on taxation
Alternative EURS proposals have different implications on long-term national taxation. Many policy combinations may have no impact on net domestic taxation in the long run, while others could induce a long-term increase in taxation, borne by the population at large or by high-income earners only. Following studies of attitudes towards spending and taxation, we hypothesize that individuals want to minimize their own contributions, preferring to improve their welfare position without paying for it (Rueda and Stegmueller, 2019). This expectation conforms to findings from the study of international bailouts where citizens were highly sensitive to additional costs of bailout packages (Bechtel et al., 2017). However, if a tax is to be paid, most individuals should prefer that privileged households contribute more (Barnes, 2015;Cavaillé and Trump, 2014). This latter preference might be stronger or weaker depending on a country's existing inequality (Lupu and Pontusson, 2011).
H4a: Individuals will tend to lend more support to packages that include no increase in taxation than to packages including any form of increase in taxation.
H4b: Individuals will tend to lend more support to packages that include progressive taxation whereby only the rich contribute (e.g.) 1% of their income than to packages that include taxation where everyone must contribute (e.g.) 0.5% of their income.

Country-level conditionality: training and education
The two final dimensions concern country-and individual-level conditionalityrespectively, the conditions that participating countries and individual beneficiaries must fulfill in order to be eligible for benefits. At the country level, conditionality often involves 'social investment' that includes training for labor-market participation. Our expectation follows studies finding strong support for training and education policies in welfare systems: European citizens have been found to prefer policy alternatives that, beyond mere subsidy, contribute to structurally lifting unemployed individuals through education and training opportunities ; see also Knotz and Lindvall, 2015). There are good reasons to expect the same for European-level assistance: trust is lower across than within countries (Deutschmann et al., 2018) and fears of moral hazard are higher. Thus, imposing tough conditions, particularly fostering labor-market adjustment through activation and education and training policies, can be attractive remedies to risks associated with EURS (but see Lengfeld and Kley, 2021, for opposing findings).
H5: Individuals will tend to lend more support to packages that include conditionality associated with the provision of education and training opportunities than to packages with no such conditions.

Individual-level conditionality: Job search effort
Individual-level conditionality should strongly shape support for EURS. The most common conditionality concerns labor-market activation of assistance-beneficiaries. An EURS scheme might offer benefits unconditionally, or instead make benefits conditional upon unemployed beneficiaries accepting any appropriate job offer, or more stringently still, also requiring frequent job application. The latter two options are often seen by policymakers to minimize moral hazard associated with unemployment income assistance (Vandenbroucke and Luigjes, 2016). Studies reveal widespread public support for work-related obligations on recipients of unemployment benefits (Buss et al., 2017), for active labor-market provisions , and for the deservingness of potential beneficiaries (Van Oorschot et al., 2017). Support for individual-level conditionality might also reflect expectations about program costs: while conditionality requires more monitoring and administration, it can shorten dependence upon state assistance that makes it less costly.
H6: Individuals will tend to lend more support to packages that include stronger conditionality with regard to activation than to packages including weaker or no conditionality.

Mixes of dimensions
How the aforementioned six features of policy design combine into policy mixes is central to public support. Can more popular dimensions compensate for less popular ones (Häusermann et al., 2019)? It is important to understand which dimension(s) are the strongest in swaying public attitudes towards EURS. It is also important to know which internally consistent policy mixes are most or least popularfor instance whether (internally-consistent) policy mixes combining less generous but lower tax options garner more or less support than (equally consistent) policy mixes combining generosity with higher taxes. We have no strong theoretical priors on such issues, so we treat these as inductive empirical questions: Our exploration, hence, not only tests the marginal benefits of particular features (deductively hypothesized) but unearths which mixes are most/least popular among Europeans (inductively revealed).

Research design
We conducted a conjoint experiment fielded among 19,500 respondents in 13 EU member states: Austria, Belgium, Denmark, Estonia, Finland, France, Germany, Hungary, Ireland, Italy, the Netherlands, Poland and Spain. These countries vary with respect to welfare-state models, economic performance, geographical location and Eurozone membership. Fieldwork was conducted by the survey company IPSOS in Fall 2018. Respondents accessed the survey via an online platform on personal computers and mobile devices. Surveys were translated into each country's main language(s) (e.g. Belgian survey fielded in French or Dutch; Estonia's in Estonian or Russian). In each country, we surveyed 1500 respondents. Quotas for age, gender, education and regional distribution ensured that the sample was representative of the population's demography.
Conjoint experiments have become a widely utilized method to study public preferences on policy and institutional design Hahm et al., 2019;Häusermann et al., 2019;Jeannet et al., 2021;Leeper et al., 2020). In our experiment, respondents first received general information on the aim of ensuring sustainable unemployment benefits in countries facing crises (see the Online appendix). The conjoint itself presents three pairs of two policy packages that vary in the six policy dimensions discussed above. Respondents assessed packages by indicating which of the two proposals they prefer in a given pairing, and by rating every package as something they (strongly or somewhat) reject or support. Hence, respondents assessed six proposals in total. The Online appendix provides information about randomization. Table 1 summarizes the dimensions, dimension-values and related hypotheses. The dimension-values in all cases reflect our attempt to maximize clarity of options to survey respondents on the one hand, and fidelity to actual policymaker discussion on the other. The first dimension (D1) concerns generosity, the percentage of the last wage of the beneficiary (or 'replacement rate') insured by the new scheme, featuring three possible levels (40%, 60%, 70%). The level shown might be higher than some countries' current levels, in which case the scheme would create a common floor for national schemes' generosity. In other countries, proposed EURS generosity may entail lower replacement rates than existing policy, in which case (as the framing specifies) countries could (continue) providing higher benefits at their own country's expense.
The second dimension (D2) concerns the level of administration, distinguishing national from EU-level administration. Although all schemes concern a European policy, the two levels of EURS administration capture whether the governance and implementation shall be provided by a national (i.e. country level) agency or by a dedicated EU-level agency.
The third dimension (D3) models cross-country redistribution. We model three possible alternatives. A first 'insurance option' entails no redistribution: each country's own long-term contribution limits the support that it can receive. A second 'general redistributive option' allows unemployment-impacted countries to receive more than they paid in. And a third 'rich-to-poor redistributive option' allows poor countries to pay less into the scheme than do rich countries, rendering the former long-run net beneficiaries and latter net contributors.
The fourth dimension (D4), concerning domestic taxation, involves three options of tax burden: a 'no impact' option (i.e. taxation remains the same in the long run); a 'flat-increase', according to which 'taxes will increase for everyone by 0.5% of their income'; and a 'progressive' option, where 'taxes will increase by 1% of income only for the rich'.
The fifth dimension (D5) captures country-level conditionality. The dimension focuses on two possibilities ('no conditions' and 'education and training'), describing whether respondents prefer a simple benefit scheme or instead demand a key form of social investmenteducation and training for the unemployed.
The sixth dimension (D6) concerns individual-level conditionality. Here, we focus on whether benefits are conditional upon a beneficiary's job search efforts: 'no conditions'; 'accept any suitable job offer or lose the benefit'; or 'apply for at least one job per week, and accept any suitable job offer, or lose the benefit'.

Dependent variables: chose or support EURS
After being presented a side-by-side pairing of policy packages, respondents were first asked to indicate which package they preferred. This yields the binary-dependent variable The European Union National Governments H2: National-admin. Packages should attract more support than EU administration D3: Between-country Redistribution May some countries receive more support from the program than they pay into it?
No, in the long-run countries cannot receive more support from the program than they paid into the program. Yes, in the long-run countries can receive more support from the program than they paid into the program Yes, in the long run, poor countries will receive more support from the program than they paid into it, while rich countries will receive less support from the program than they paid into it.
H3: No-redistribution packages (no option to draw out more than country pays in) should attract more support than packages allowing between country redistribution.

D4: Domestic Taxation and Redistribution
What is the long-term impact on the taxes you have to pay for your unemployment insurance?
No impact in the long run: the level of taxes remains the same in country In the long run, taxes will increase with 0.5% of income for everyone in country In the long run, taxes will increase with 1% of income only for the rich in country H4a: No-tax packages should attract more support than packages with tax increases. H4b: 1%-tax-rich packages should yield more support than 0.5%-tax-all packages. (continued) Chose EURS Package. Second, respondents were asked to rate each package on a scale from 1 (strongly against) to 5 (strongly in favour). For our baseline models, we created a binary variable Support EURS, where 1 equal somewhat or strongly favouring a package and 0 equals somewhat or strongly opposing or being neutral towards a package. Not surprisingly, Chose and Support are quite highly correlated (covariance of .52). But the two outcome measures reflect different research-design features that independently gauge support for EURS. The summary statistics (see the Online appendix) harbor hints of support for EURS. With respect to Support EURS (binary), 46% of (randomly varying) EURS packages received 'somewhat' or 'strongly' support. If we exclude neutral answers, then 76% of the sampled population supports some European unemployment insurance. Only 12.9% of respondents supported none of the six packages they saw, while 21.9% of the respondents 'strongly' or 'somewhat' supported four or more of the six packages. Figure 1 focuses on the country-specific means for Support EURS (binary), where 1 equal somewhat or strongly supporting and 0 equals being neutral towards or somewhat or strongly opposing. The country variation is modest, ranging from 37% support in France to 54% support in Estonia. This descriptive visualization suggests that publics in those Northern and Western member states with the most developed welfare systems are less supportive of EURS than their counterparts in Southern and Central and East European states. Figure 2 summarizes the sample means of Chose EURS Package (binary) and Support EURS (binary) across the six policy dimensions. This captures the chance that packages including a given dimension value is either chosen as most appealing (least objectionable) in pairings (darker bars) or (somewhat or strongly) supported by respondents (lighter bars). Higher shares for Chose EURS (binary) than for Support EURS (binary) are expected, since the former is based on the 50% chance of being 'chosen' in a pairing while the latter is based on 1-5 ratings for all packages. The descriptive patterns broadly follow expectation, with respondents more likely to choose or support packages with more generosity, national-level administration, no new taxation, and country-and individual-level conditionality. The exception is the (very) modest preference for between-country redistribution.
Inferring support from such results requires caution. Although our framing is as neutral as practicable, the survey context differs from real political contexts where political parties campaign for or against EURS. Asking respondents which of side-by-side packages they prefer may cue respondents to somewhat 'support' the package that they chose as least objectionable. Our analyses consider such possibilities in robustness tests (e.g. excluding neutral answers), but any conjoint results face real inferential concerns (Abramson et al., 2019;Leeper et al., 2020). For instance, conjoint experiments might assign too much weight to the intensity of individual preferences on certain attributes, losing sight of the distribution in favour of or against each attribute. This may not be a problem, since the intensity of preferences is often key in guiding choices in multidimensional setups, which better reflect the complexities of the real world (Bansak et al., 2020: 21). Also alleviating this concern is that respondents' answer to an open question about which dimensions mattered most to their EURS judgments are highly correlated with the package characteristic they most or least favoured in the experiment.

Results and discussion: pooled and cross-country support for EURS
We summarize our findings in two steps. The first focuses on the pooled results for all country samples, testing our hypotheses on policy dimensions garnering support for EURS, and then examining the policy mixes that command the most (least) public support. Our second step focuses on whether attitudes towards EURS in different surveyed countries corroborate or nuance the pooled results, and on the political potential for EURS in intergovernmental EU decision-making.
Pooled results: support for EURS per dimension and as a policy mix We fit regression models of support for a given EURS package in the full sample: six packages, in three pairings, judged by 1500 respondents in each of the 13 countries. The unit of analysis, hence, is package-pairing-respondent-country. The baseline specifications of the dependent variable include either whether a package was chosen per pairing -Chose EURS (binary))or the per-package rating given to each package -Support EURS (binary). The Online appendix summarizes the various specifications for each as a function of the six policy-design dimensions. Following studies of conjoint-experimental design , our baseline models are Ordinary Least Squares (OLS) estimators with respondent-robust standard errors. Our specifications exclude respondents who failed the attention check and who were inconsistent more than once in their per-package ranking and choice-of-package (e.g. choosing package A over package B in a pairing, and rating B higher than A). This censures-out 20.6% of the sample respondents.
Our key explanatory factors, being experimentally derived, can be seen as empirically orthogonal to one another. Hence, EURS support can be estimated by linear OLS regression of the outcome variable on dummies for each level of each attribute (excluding a reference category for each dimension) (Abramson et al., 2019;Bansak et al., 2020;Hainmueller et al., 2014;Leeper et al., 2020). We include individual-level controls for unemployment, low income, age and gender, and 12 country dummies and dummies for the judge package's pairing-ordering. These controls address possible remaining omitted variable bias and mitigate heteroscedasticity and country-level composition or ordering effects. The important inferential point is that the baseline yields results very similar to alternative specifications in terms of sample, controls and estimator, as discussed below. 2 The main results are that respondents tend to prefer packages with 70% replacement rate for generosity (D1); national as opposed to EU-level administration (D2); betweencountry redistribution, either for any given country or redistribution from rich to poor countries (D3); no increased tax burden (D4); conditionality requiring countries to offer education and training to their unemployed (D5); and requiring individual beneficiaries to meet job-search conditions (e.g. accept a suitable job offer) (D6). These patterns are stable across specifications, except for between-country redistribution (D3), where the (always) positive coefficients lose statistical significance with multi-level random-intercept estimators (see the Online appendix). We consider counterfactual analyses to (a) clarify the substantive size and significance of specific policy features relevant to our hypotheses; and (b) identify the policy mixes (full EURS packages) commanding the most support.
Results per dimension: testing hypotheses 1-6. Figure 3 plots the average marginal component effects (AMCEs) across our six dimensions, based on the baseline models of both Chose EURS package (binary) (M2) and Support EURS package (binary) (M5). Such AMCEs gauge the causal effect of a given attribute-value on the probability that a package will be chosen, relative to the baseline attribute value. Our interest is in average preferences rather than interactions with (demographic or other) covariates and sub-samples, whereby marginal-effect results are less sensitive to the reference category than holds for smaller sub-samples involved in interactions (Leeper et al., 2020). Also shown are the 95% confidence intervals of these effects. The marginal effects can be interpreted as the increased or decreased chance of supporting EURS across each dimension's option. The AMCE for each attribute-level is equal to the estimated coefficient for that attribute-level and can be judged relative to the support level at a given dimension's baseline value. Figure 3 also shows how the experimental treatments yield stronger marginal component effects when focusing on Chose EURS (binary) than on Support EURS (binary). These variations should be judged in terms of the different baselines discussed above: .50 for Chose EURS Package (binary) and .46 sample mean for Support EURS (binary). In either specification, the results capture the increased (decreased) probability of supporting a given EURS package.
Such counterfactual analysis corroborates six of our seven preregistered hypotheses. Both specifications suggest that the strongest predictor is generosity: based on Support EURS (binary), respondents are about 12% more likely to support the most generous packages (70% replacement-rate) than the least generous ones (40% wage replacement). The second-most influential dimension is individual-level conditionality relevant to labor-market activation, where the most popular are packages where beneficiaries must accept a suitable job offer. In line with our hypothesis, respondents are 7.5% more likely to support packages requiring beneficiaries to accept a suitable job than packages lacking such activation requirement. The effects of long-term taxation burden are also substantial: The most popular packages pose no extra taxation burdenroughly 6% more support than packages where everyone bears some (modest) burden. Packages with a more progressive tax burden are also unpopular, but less so. Furthermore, packages requiring education and training benefits for the unemployed are about 7% more popular than packages without such a requirement. Again, this supports our hypotheses.
The dimension with the lowest effect is between-country redistribution: We see a small (and, in the baseline models, marginally statistically significant) preference for some redistribution (either for any contributing countries or from rich to poor countries). This goes against our hypothesis that Europeans should tend to look unfavourably upon between-country redistribution in EURS programs. But the substantive significance is lowperhaps unsurprising in light of plenty of research showing that between-country redistribution divides poor and wealthy polities. We address this issue in our per-country discussion below.
Results for full EURS packages: estimating 'votes' for EURS. Such marginal effects clarify support for particular features across dimensions, but not for the combinations of features in policy packages likely to be presented to voters. Although the estimations of marginal effects suggest that the most popular features might be combined to yield the packages with the most traction, such inference faces two problems. First, using the marginal effects to judge support for policy combinations does not capture how a given package feature might vary in salience and might command levels of support that vary depending on how it combines with other features. Second, some combinations are not politically viable and are likely irrelevant to actual debate on European social policy: for instance, a 'free lunch' combination of highly generous benefits (e.g. D1 = 70%) without any new tax burden (e.g. D4 = 'no extra taxes') is unlikely realistic. Indeed, 'free lunch' options are popular, where the combination of the marginally most positive features captured in Figure 3 is predicted by the model to command more than 60% support (80% should assume that neutrals stay home). This or any other fiscally problematic combination need not be a 'free lunch' if public spending on other programs gets reduced or governments incur higher public debt. But such trade-offs are prima facie implausible in actual EURS-policy discussion. We particularly want to consider politically viable mixes.
Gauging which policy mixes command the most support, hence, requires counterfactually exploring particular, politically meaningful packages. Most important are six modal policy packages that capture what determines support for EURS. These EURS packages vary by generosity (low 40% vs. high 70%); taxation (no taxation vs. tax burden concentrated upon the rich); between-country redistribution (country must pay in what it draws out vs. a scheme that includes redistribution from rich to poor countries); country-level conditionality (no country-level conditionality vs. conditionality upon provision of training and education programs); and individual-level conditionality (no conditionality vs. requiring recipients to accept suitable job offers). The only dimension our comparison holds constant is the generally more popular 'national' level of EURS administration (D2). Figure 4 summarizes predicted support for the six counterfactual EURS packages combining these particular policy features. It captures the predicted share of voter support for particular EURS packages should any be put to a vote in our European sample. Of course, predicting voter support based on survey data involves complex speculation. Moreover, responsiveness to voter positions often departs from one-to-one representation of expressed voter wants (cf. Soroka and Wlezien, 2010;Wratil, 2019). Instead, Figure 4 reports counterfactual results based on our baseline approach (see the Online appendix).
We focus on models yielding a 'low floor' and 'high ceiling' of predicted support for EURS. The low floor is based on results from our EURS Support (binary) (see the Online appendix)based on somewhat or strongly supporting EURS packages (vs. being neutral or somewhat or strongly against). The grey bars capture predicted means for such Support EURS (binary), with 95% confidence intervals. This is a low estimate in that the results presume that all neutrals would vote against the EURS package. The 'high ceiling' excludes the neutrals.
A second predicted level of support for each EURS package can be seen as a high-ceiling estimate, based on the assumption that neutrals stay home on an election Note: Bars show results of low-estimate support, based on Support EURS (binary) (1 = somewhat or strongly favour; 0 = neutral or somewhat or strongly against); Strips show results of high-estimate support, based on Support versus Oppose (1 = somewhat or strongly favour EURS; 0 = somewhat or strongly against; missing = neutral). day or as a group be split evenly among the somewhat or stronger pro-and anti-EURS camps. Our high-end estimate focuses on Support versus Oppose EURS, the share of respondents who somewhat or strongly support EURS (based on the rating-based Support EURS (categorical)) relative to those who somewhat or strongly oppose EURS in that same measure. In Figure 4, this high-ceiling prediction is captured by the solid-line 95% confidence interval strips of predicted support for each EURS package.
We order the hypothetical six packages as follows. The first three packages are EURS policy options without conditionality: no country-level training requirement (D5) and no individual-level activation requirement (D6). The second three packages are EURS options with such country-and individual-level conditionality. For each cluster we consider three combinations with respect to varying generosity (D1), tax redistribution (D4) and between-country redistribution (D3): (1) Low generosity, no between-country redistribution, no tax change and national-level administration; (2) High generosity, no between-country redistribution, .5% tax change for everyone, national-level administration; (3) High generosity, between-country redistribution, 1% tax change for the rich, national-level administration. Across these options, results can be judged relative to a vote threshold (.5) or averages across all package combinations (.46 for the low-floor prediction and .64 for the high-ceiling prediction). Figure 4 shows that (country-and individual-level) conditionality is key to undergirding public support for EURS packages. A package's generosity and tax level and between-country redistribution also matter aplenty: Packages that are more generous and that entail tax and between-country redistribution garner significantly more support than less generous packages and/or packages without redistributive tax burden or between-country redistribution. However, these features matter less than does conditionality. Without conditionality, support for EURS may fall below the 50% thresholdeven in the third counterfactual package (bar three in Figure 4), the more popular combination of generous, redistributive tax burden and between-country redistribution (where poor countries can obtain more than they contribute). Package three's low estimate of .45 and high estimate of .62 is less than the average for all shown packages (.46 and .64, respectively, for the low and high estimates).
In short, the EURS package garnering the most support combines high generosity, national-level administration, redistributive tax financing, between-country redistribution, and above all, country-and individual-level conditionality towards labor-market activation. The low-floor prediction of support is about 60% of the respondents, and the high-ceiling prediction of support approaches 80%. This is not much lower than the counterfactually predicted support for the most popular but likely nonviable 'free lunch' EURS package.

Support for EURS per-country
We turn now to the per-country patterns in our 13 sample polities which provide further insight on our hypotheses 1-6 but also clarify whether policy mixes defining EURS packages command support of member states relevant to EU Council decision-making.
Per-country support across dimensions of EURS. The per-country patterns (see the Online appendix) broadly corroborate the pooled analysis, with several notable exceptions. With respect to administration level (D2), the Spanish and Poles prefer EU-level administration to national-level administration. This is plausible and in line with benchmarking theory, which suggests that voters make comparisons between the (national) status quo and (European) alternatives (De Vries, 2018). The preference for EU administration could reflect scepticism about Spanish and Polish institutions and the expectation that European administration should be more effective than national administration. With respect to domestic taxation (D4), respondents in Poland are not significantly less supportive of higher tax packages. With respect to individual-level conditionality (D6), the Finns, uniquely, are less likely to support packages that include such conditionality. None of these cross-country differences, however, is significant enough to overturn the results for hypotheses 1-6, for instance through jackknife removal of any country from the analysis.
With respect to between-country redistribution (D3), however, the cross-country differences are larger and clarify puzzling features of the pooled results, including the lack of support for H3. Figure 5 clarifies the volatility across countries with respect to betweencountry redistribution. It summarizes the combined AMCE plots for all 13 sample countries with respect to 'between-country redistribution' (D3). The results reflect 13 separate estimations, one per country, following the specification discussed above (see the Online appendix). Figure 5 captures the roughly even split in whether support for EURS is positive or negative for packages including redistribution, relative to pure-insurance packages (where countries can receive no more than they contribute).
For some countries (darker, larger font), these results are statistically significant. Hungarians, Irish and Poles are significantly more likely to support EURS when it includes general, any-country redistributionthat is, when any country can draw out more than it pays into the EURS program (relative to no redistribution). Hungarians, Poles, Estonians, Italians and Spaniards are more supportive of EURS if poor countries can draw out more than they pay in (entailing rich-to-poor-country redistribution). In contrast, some Northern European counterpartsin Denmark, Finland, Belgium and the Netherlandsare more supportive of no between-country redistribution. These patterns are statistically significant in two cases: in Denmark, for no redistribution vis-à-vis the general redistributive option (where any country can draw-out more than it pays-in); and in the Netherlands, for rich-to-poor, between-country redistribution. Although identifying what underlies these divisions is beyond the scope of this article, there are many possible explanations, including modest versus generous welfare states, net-contributor versus net-recipient position in EU fiscal conditions or low versus high state capacity. It is noteworthy that populations of three of the so-called 'Frugal Four' (Austria, Denmark and the Netherlands; Sweden was not surveyed) tend to be against crosscountry redistribution, while the surveyed polities that suffered most in the sovereign debt crisis (Italy, Ireland and Spain) tend to support cross-country redistribution.
Importantly, the major cross-country differences on redistribution between states help explain why H3 was not supported by the pooled models: The weak but positive average support for such redistribution captures a combination of positive and negative effect of cross-country redistribution in the country samples. 'The whole', captured by the pooled results, appears to be less relevant than 'the sum of its parts'.
Per-country support for EURS packages. This cross-country pattern also has implications for the combinations that polities can be expected to endorse or eschew. Per-country patterns matter, particularly for judging support for EURS in individual country referenda or European-parliamentary choices relevant to major EU decisions. Figure 6 captures support for the same six exemplary EURS packages shown in Figure 4, but here for each of the 13 sample polities. The underlying models use the pooled specification but are run per-country. The six counterfactual alternatives, and the low-and high-estimate predictions, repeat those from Figure 4. Predictions should be considered in light of the 50% threshold or the (lower) average predicted support for all package combinations (see the Online appendix). Figure 6 broadly corroborates Figure 4's pooled pattern. In all but three countries, the combination of generous, nationally administered, redistributive and (country-and individual-level) conditionality commands the most support among European polities. Particularly, conditionality constitutes the tipping point in policy design, the only exception being Finland (where conditionality is less popular).
The clearest exceptions are Denmark and the Netherlands, where packages without redistribution (domestically and between-country) are slightly preferred to packages with redistribution. These exceptions comport with our speculation about H3, where those in the richest, most generous Northern Europe welfare states judge redistribution differently than their Southern and Eastern counterparts. These exceptions, however, Figure 6. Predicted country-by-country minimum and maximum vote for EURS packages without versus with conditionalities. Note: Six package options for each panel are same as Figure 4 above. Bars show low estimate based on Support EURS (Binary) (1 = somewhat or strongly favour; 0 = neutral or somewhat or strongly against). Strips show high-estimate support, based on Support versus Oppose EURS (1 = somewhat or strongly favour; 0 = somewhat or strongly against; missing = neutral). also prove the rule as to which packages of EURS would most likely survive a Europe-wide referendum or European Council deliberation. Even in Denmark and the Netherlands, predicted support for generous, redistributive and conditional EURS (package 6) surpasses the 50% threshold, always higher than average predicted support: for Denmark, a low estimate of .57 and high estimate of .75 for package 6, compared to .41 and .55, respectively, for low-and high-estimate support averaged across all packages (see the Online appendix); and for the Netherlands, a low estimate of.58 (compared to a mean of .42 for all packages) and high estimate of .77 (compared to .57). Also, the pooled predictions in Figure 4 do not change when one country at the time is removed from the sample.
In short, in two countries (France and Finland) our low-end estimate of support for generous and conditional EURS is just under the 50% threshold. Such patterns might well predict a politics that would defeat an EURS proposal. However, the broad pattern suggests majority support in the diverse European settings for generous, national-administered, redistributive and conditional EURS.

Robustness
Importantly, our baseline results are robust to many alternative specifications. First, the results are robust to alternative measures of support for EURS, including Strongly Support EURS (where 0 corresponds to strongly or somewhat oppose, neutral or somewhat favour and 1 to strongly favour); Support versus Oppose EURS (which excludes neutral answers); or the full five-point Support EURS (categorical).
Second, the results are insensitive to adding (additional) individual controls, including citizenship, various party-political and substantive policy beliefs, religiosity, etc. The baseline results are robust to country controls, including welfare-state type/generosity, Gross Domestic Product (GDP) per capita, growth, aggregate unemployment, aggregate inequality, poverty and government debt. They are also insensitive to sub-national compositional controls like full per-country NUTS-1 or NUTS-2 regional dummies, and full individual-respondent fixed effects (thereby focusing only on within-respondent variation across packages).
Third, the results are insensitive to (differently) censoring the sample to address possible heterogeneity in how respondents view particular EURS combinations. Focusing on those who reject or express lower-than-average support across all six packages does not change the patterns reported in Figures 3 and 4. Nor does omitting respondents answering 'neutral' all or most of the time (reflecting possible laziness or shortcuts in answering). Nor does excluding observations involving packages with internal inconsistencies, like calling for no tax increase but also for more generous replacement rates and/or between-country redistribution. 3 Fourth, the results are robust to key alternative estimators, such as logit/probit estimates of the Support EURS (binary) or Chose EURS (binary) or ordered logit/probit for Support EURS (categorical). They are also robust to alternative multilevel models with varying embedding structures (e.g. two-level package-respondent models or package-country models), and to multinomial logit and rank-order logit estimation. These patterns suggest that the reported results capture key political sentiments rather than our econometric or measurement caprice.

Conclusion
Do Europeans support social policy beyond the nation state to deal with unemployment shocks? This question is particularly relevant in the context of struggles over EU solidarity in recent crises. Although previous research has revealed support for European social policy in general (Beaudonnet, 2013;Gerhards et al., 2020), our study clarifies the 'devil in the policy details' of public opinion, focusing on how the design of European unemployment insurance impacts support for EURS. We move beyond understanding abstract preferences, instead exploring the multidimensionality of preferences for social policy (Gallego and Marx, 2017). This sheds light on the conditions under which citizens endorse European social policy and on how citizens deal with policy trade-offs such as higher generosity at the cost of a tax increase. Crucially, those conditions concern policy dimensions that comport with actual policymaker discussions about supranational unemployment risk-sharing.
Our general finding is that hypothetical EU-level social policies buttressing national insurance provisions garner majorities of support in the 13 member states surveyed. However, the key findings relate to how such support is conditional, thereby advancing the exploration of linkages between institutional design and public opinion (Hahm et al., 2019). Each of the six policy dimensions that we expect to influence such support turns out to be statistically and substantively significantand all but one in line with our (preregistered) hypotheses. Respondents tend to prefer packages with more generous assistance, more modest or progressive tax burdens, between-country redistribution, national-level administration and above all country-and individual-level conditionality focused on social investment and activation. Finally, the modest preference for between-country redistribution (contrary to our expectation)averaged across the thirteen sample countriesmasks divisions across polities, where some Southern and Eastern sample countries strongly prefer such redistribution while Northern European member states (more modestly) prefer pure insurance.
Compared to other studies, these results side with the empirical appraisals finding substantial potential for European solidarity (e.g. Ferrera and Pellegata, 2017;Gerhards et al., 2020), rather than with more negative empirical appraisals of such potential (e.g. Dolls and Wehrhöfer, 2021;Lahusen and Grasso, 2018). The contrast with our findings is, we suspect, partly bourn of our domain-focus compared to the more negative appraisals. Rather than focus on debt-stricken countries, we focused on the broader realm of unemployment-stricken countries. Indeed, other research shows the place and domain specificity of attitudes to solidarity (Beramendi, 2007). And rather than asking about onesided transfers from better-performing countries to countries facing structural difficulty (Dolls and Wehrhöfer, 2021;Vasilopoulou and Talving, 2019), we ask about a policy potentially beneficial to any unemployed citizen in any participating country.
The distinctiveness of our findings, however, may also reflect how our survey design encourages respondents to think about concrete packages with policy ingredients that can address key worries commonly shaping attitudes, notably moral hazard. The policy ingredients also include a link to social investment policiestraining, education and activationthat have traction with significant parts of the population. Although our current research design does not allow us to explore the matter fully, we suspect that these surveyed features of EURS explain why our results show public attitudes (in Germany and elsewhere) more open to cross-border solidarity than the results obtained by Dolls and Wehrhöfer (2021). But the key role we find for conditions attached to support is in line with Lahusen and Grasso's (2018: 260-261) findings and Gerhards et al. (2020: 247) suggestion that Europeans are not ready to unconditionally support other EU countries in trouble.
Our analysis has important limits that deserve highlighting. First, our study has (deliberately) not focused on identifying individual or country-level conditions that plausibly moderate or shape support for EURS. Our discussion of country-level patterns included only preliminary exploration of key contextual factors, such as the economic situation of the country or region of a respondent, or the existing social protection in national settings. Our own and others' future work will delve in more detail into the moderating or other roles of such contextual or, for that matter, individual-level, factors.
Second, as in any survey experimental design, we have to strike a balance between making the experimental dimensions comprehensible to respondents and making dimensions faithful to the actual policy debate we seek to capture. This constrains the number of attributes of the experiment. In our case, this meant that some policy features are left implicit; for instance, our exploration of attitudes towards financing possibilities does not include intersectoral transfers (cutting public expenditure elsewhere) or intertemporal ones (increasing public debt). This simplifying omission complicates our interpretation of programs combining increased expenditure with no tax increases as 'free lunch' packages. More fundamentally, our survey design can do little to gauge prioritization by respondents of a given design attribute. The design can only roughly approximate details of policy debate and framing on EURS, and what respondents and voters actually understand about such details. Future study of public opinion should actively explore such issues of internal and external validity and consider other research designs to gauge how framing and policy subtleties affect attitudes towards European social policy.
In the meantime, our analysis yields findings crucial to understanding 'Social Europe', suggesting a clear answer to the broadest question: Do European publics support EURS? 'Most likely yes' is our answer, so long as programs are sufficiently generous and made conditional upon countries providing social investment training and upon beneficiaries accepting employment activation.