Collective Identities, European Solidarity: Identification Patterns and Preferences for European Social Insurance

A degree of collective identity is often perceived to be a requirement for integration of core state powers, and even more so when redistributive measures are at stake. Existing research has shown that European identity is an important correlate of support for European social policy in general. This article explores the ways in which collective identities relate to the multidimensional nature of social policy at the European level. We explore in particular the features of European social policy that receive the most support from European publics, and how local, national and European identity moderates these preferences. We expect that people who have primarily sub&#8208;national or national attachments are less supportive of generous schemes and schemes that involve greater cross&#8208;country redistribution, and are more in favour of decentralized schemes administered by the member states than those with stronger European identities. We find support for these expectations in a conjoint survey experiment fielded in autumn 2018 in 13 EU member states.


Introduction
The Euro and migration crises have sparked increased efforts to Europeanize social policy-making. Policy-makers and academics have presented proposals for introducing EU-level social policy benefits, including policies providing unemployment benefit schemes and reinsurance mechanisms for the unemployed. These proposals, that can be collectively called European unemployment risk sharing (EURS), differ on many dimensions, including the degree of redistribution that they entail; the degree of individual and country-level conditionality; whether contributions are used to simply strengthen national unemployment benefits or to provide education and training subsidies and who will finance the scheme.
A commonly assumed political requirement for any given combination of such proposals is that a modicum of collective identity is needed to embrace new European-level provisions. This applies to any integration of core state powers (Genschel and Jachtenfuchs 2016), but even more so when redistributive measures or other forms of solidarity are at stake. A major problem, however, is that European identity formation is taking place at a much slower pace than European institutionbuilding, creating a tension between community and the scope of governance (Hooghe and Marks, 2009).
Against this backdrop, researchers have studied the relationship between European identities and European social policy in general (Börner and Eigmüller, 2018;Gerhards et al., 2016;Lahusen and Grasso, 2018;Verhaegen, 2018). They have also explored support for specific forms of solidarity, including European unemployment benefit schemes (Dolls and Wehrhöfer, 2018;Ferrera and Brunelli, 2019). However, we do not know what kind of European social policies receive public support or opposition, and which aspects of these policies are relevant to having a collective identity, indeed, collective identities, given the multidimensional nature of collective identities and European social policy (Baute et al., 2018).
In this article, we take such plural identities and policies seriously, asking how different ideal-types of local, national and European identitiesas laid out in the introduction of this special issuemoderate support for policy proposals for European unemployment benefits schemes that differ with respect to six dimensions: their generosity; country-level conditionality; individual conditionality; cross-country redistribution; financing and governance. We expect that questions of collective identity are most relevant to the dimensions that refer to national sovereignty, such as whether these policies are governed by the member states themselves, or are in the control of EU institutions, and the dimensions that touch upon questions of solidarity, such as cross-border redistribution.
We test our hypotheses using a conjoint analysis survey fielded in 2018 in 13 EU member states. Our findings show that the impact of collective identities on preferences for European integration of social security varies both across levels of identification and across dimensions of EURS. Individuals with national and more local identities are not necessarily opposed to EURS schemes but they do prefer packages that retain national oversight and avoid redistribution across countries. Conversely, individuals with a stronger European identity hold favourable views of packages that include European-level administration and cross-border redistribution.

I. European Unemployment Risk-sharing
Despite the many new policies and institutions introduced since 2010, the European Monetary Union remains vulnerable to asymmetric macroeconomic shocks. When asymmetric shocks hit countries with a limited fiscal capacity to adjust, their citizens may suffer, as the absence of external monetary adjustment (which is often impossible in a single currency zone) cannot be credibly and sustainably compensated for by domestic fiscal policy and automatic fiscal stabilizers. To address this structural problem, many overhauls of the European Monetary Union have been proposed. Most of them concern the financial aspects of economic stability. Proposals include introducing eurobonds (Von Weizsäcker and Delpla, 2010); establishing a European monetary fund and a European finance minister (Enderlein and Haas, 2015) and strengthening macroeconomic coordination (European Commission, 2013;European Council Conclusions, October 2014;Nicoli, 2016;Van Rompuy, 2012, p. 9). The goal of these proposals is not to provide directly a European layer of protection for vulnerable citizens in weaker economies but to ensure that domestic fiscal stabilizers remain effective while making asymmetric shocks less likely altogether. While they may be effective, such proposed solutions often fail to garner substantial public support.
Against this backdrop, a number of scholars have suggested a form of European-level support for unemployed people across the continent (see for instance Beblavý and Lenaerts, 2017;Dullien, 2014). These proposals, initially floated by the President of the European Council Herman Van Rompuy in the midst of the Eurocrisis (Van Rompuy, 2012), are often named European unemployment benefit schemes or, more broadly, EURS schemes. EURS schemes have attractive features: they work as automatic stabilizers in cases of asymmetric shocks, and may also be perceived as being closer to the people. Furthermore, the introduction of a EURS scheme would complement existing tools for an EUwide social policy, such as the European pillar of social rights. Introduced in 2017, the European pillar of social rights aimed to reduce the fundamental asymmetry (Scharpf, 2002) between market integration and domestic social protection without harming national autonomy. Its functioning is based, on the one hand, on a series of common objectives to be achieved domestically (Zeitlin and Vanhercke, 2018), and on the other, on the progressive convergence of social rights standards through directives. The pillar lacks, however, a genuine European tier of shared social protection. The introduction of a EURS scheme would not only contribute to the Eurozone's macroeconomic stability but would also be an act of cross-border solidarity, building up a European-level layer onto national social policy. A EURS scheme constitutes, therefore, a suitable policy proposal to test the attitudes of European citizens towards European-level social policy and solidarity.
Of course, in theory and in existing political discussions, different EURS proposals have been described with very different policy characteristics. Beblavý and Lenaerts (2017) provide an overview of variants of EURS. Fundamentally, EURS schemes differ in whether unemployed citizens receive support directly from a centralized instrument, regardless of their residence (a genuine European unemployment benefit scheme), or whether funding is targeted to unemployed people in member states in distress (reinsurance schemes). The model of EURS on which we focus combines features of both models. While the framing of the experiment makes clear that the funding is tied to member state performance (see Annex 2, figure 2), we kept the door open for both direct support from the EU to citizens and more decentralized schemes managed by member states, in our administration dimension.
Capturing the full extent of all possible variations of EURS schemes is not suitable in a survey design. However, we build on the work of Beblavý and Lenaerts (2017) and focus on six dimensions of policy design: generosity, country-level generosity, individual-level conditionality, cross-country redistribution, financing and administration. The first dimensioncapturing the generosity of EURS schemesrefers to different replacement rates, that is, the maximum proportion of the last wage of the unemployed person that will be insured by the new scheme. The second dimension relates to country-level conditionality arrangements on education and training. The next addresses the variations in policy schemes with regard to individual-level conditionality. The fourth captures the fact that existing proposals vary with respect to whether and to what extent they involve crosscountry redistribution. The fifth dimension focuses on governance arrangements, asking what governance level should administer the scheme. Finally, the last dimension captures alternative options for domestic taxation, in other words, the fiscal impact of the EURS scheme on the national taxation system. The operationalization of these dimensions in this conjoint experiment is discussed in Section III. This allows us to test a very wide range of different policy options, ranging from pure reinsurance schemes (thus remaining very close to the actual policy debate) to schemes featuring more direct forms of support (although we fell short of testing genuine European unemployment benefit schemes whose activation is completely detached from employment performance at country level).

II. Collective Identities and Support for European Social Policy: Sovereignty and Solidarity
We aim to move beyond the nation versus Europe dichotomy and emphasize the different ways in which identities at the regional, national and supranational level can be combined and how these combinations play out with respect to support for European social policy. As Risse (2010) puts it, 'we need to know how the many "we's," the groups to which we feel attached, relate to each other'. The evolution of European identity and its relationship with the process of European integration has been well explored in recent research (Fligstein et al., 2012;Fligstein and Polyakova, 2015;Kuhn, 2019;Risse, 2010). For a review of this literature, see Kuhn and Nicoli (2020) in the introduction to this special issue.
European identity should not be seen as synonymous with support for European integration (Kuhn 2019), itself a multidimensional phenomenon (Stoeckel, 2013). People may identify as European or feel attached to the European continent, and at the same time disagree with European integration or with particular policies. With respect to European social policy, European identity may be too superficial to translate fully into solidarity with other Europeans. While most of the literature suggests that identification patterns are increasingly important determinants of support for European integration (Carey, 2002;Hooghe and Marks, 2005), Hobolt and Wratil, (2015) found that while identities play a role, economic concerns remain paramount.
Either way, early research on European identity presupposed that national and European identity would compete with and exclude each other. For example, classical intergovernmentalism saw the persistent power of national identities as an obstacle to European integration (Cederman, 2001;Kuhn, 2019). More recent research, however, argues that regional, national and European identities can be compatible, and that individuals can have multiple identities, that is, 'the concurrent identification with multiple socially or geographically-defined groups' (Steenvoorden and Wright, 2018, p. 3).
Following the set-up of Kuhn and Nicoli (2020), we conceptualize the multi-level character of collective political identities by building on Díez Medrano and Gutiérrez's (2001) concept of nested identities. Individuals with multi-level identities can have many different attachments; local, national and European, that can be combined in several ways. Kuhn and Nicoli's (2020) introduction identifies six ideal-types of possible relationships among these identities. We focus here on the three general ideal-types ( Figure 1) (leaving out the three 'special' cases (see Kuhn and Nicoli, 2020). Individuals with what we shall call 'localist identities' feel most strongly attached to their intra-national region, and less so to their nation or to Europe. 1 Individuals with 'nationalist identities' have a national attachment that is stronger than both their regional and supranational attachments. Individuals with 'European identities' have attachments to Europe that are stronger than to other levels of political community.
Research exploring attitudes towards national welfare states identifies (in broad terms) ideologies, perceived outcomes and attitudes towards disadvantaged groups as key determinants of support for welfare states (Blekesaune and Quadagno, 2003;Svallfors, 2010;Van Oorschot et al., 2012). However, the inherently multi-level nature of European social policy invites a reflection not only on the impact of fundamental redistributive attitudes, but on the level of community in which individuals locate themselves. Indeed, a growing body of research has shown that individuals with a European identity are more likely to support European social policy than people who hold exclusively national identities. Lahusen and Grasso (2018) and Verhaegen (2018) report that there are significant relationships between European identity and support for fiscal solidarity with member states suffering economic difficulties. Furthermore, Gerhards et al. (2016) show that European identity is also linked to support for European-level welfare states. Similarly, Bechtel et al. (2014) find that Germans with more cosmopolitan outlooks are more supportive of international bailouts. Using laboratory experiments in Germany and the UK, Kuhn et al. (2018) found that people who identify as European are more likely to share resources with anonymous recipients from other European member states than people who don't identify as European. Only Hooghe and Verhaegen (2017) find no significant relationship between European identity and support for European social policy.
While these contributions have been essential in clarifying the relationship between collective identities and support for European social policy, they have studied only whether collective identity is related to European social policy support in general. However, attitudes towards European social policy are inherently multidimensional, similar to attitudes towards national welfare systems (Gelissen and Van Ooorschot, 2013). This depends on the multidimensional nature of European social policy itself, involving several dimensions, such as the generosity of the scheme, governance, and cross-border redistribution. Different schemes may vary across a range of features, some of which may foster support in some groups, while undermining support in others. It is thus important to disentangle the effects of collective identity on support attributed to different dimensions of policy design.
We expect that identities are linked to support for European social policy through two main channels: general concerns with national sovereignty and concerns with the Source: author's adaptation from Kuhn and Nicoli (2020). Note that Kuhn and Nicoli (2020) use a broader ideal-type labelling individuals with supranational attachment as 'Supranationals'. However this article operationalizes the original conceptualization by using survey questions on European identity, we therefore opt for replacing 'Supranationals' with 'Europeans' in our discussion. boundaries of solidarity. Firstly, social policy and the welfare institutions have been part of the core business of modern states (Genschel and Jachtenfuchs, 2016). The establishment of a EURS scheme moves decision-making power away from member states and towards a joint institution, thereby decreasing national sovereignty. It is therefore likely that people with localist or national identities see a European social policy as detrimental to national sovereignty. Secondly, social policy also relates to questions of solidarity and of sharing resources with other people. As captured by the concept 'parochial altruism', people's willingness to share resources is bounded, and often overlaps with, national or regional group boundaries (Kuhn et al., 2018). Hence, people with localist or national identities may be reluctant to share resources with other Europeans with whom they lack an emotional bond. Accordingly, of the six policy dimensions of the conjoint experiment, our attention focuses on those that directly express solidarity, such as the generosity of the scheme and country-level redistribution, and on the level of governance, which should capture fundamental concerns with national sovereignty.
Given the link observed in the literature between solidarity and group identification, we consider that schemes with more generous features may garner more support among those who see themselves as genuine Europeans; and even more so, when explicitly considering the cross-national redistributive implications of social policy. On these grounds, our first two hypotheses are: H1: Individuals with European identities are more supportive of more generous packages than are localists and nationals.
H2: Individuals with European identities are more supportive of cross-country redistribution than are localists and nationals.
Finally, it is important to single out the effect associated with the national versus the EU level of governance of a EURS scheme. Even individuals without a strong attachment to Europe may support a European-level social policy, because, for instance, they may see themselves as the net beneficiaries of such a policy. However, even when support for EUlevel social policy is high, individuals without a strong European identity may be concerned about potential losses in national sovereignty should the EURS be administered at an EU rather than a national level. Conversely, those embracing a European identity can be expected to be more comfortable with an EU-level administration, ceteris paribus. To disentangle the effects associated with the relationship between governance arrangements and identification from other factors affecting support for EU-level social policy, we hypothesized that: H3: Localists and nationalists are more supportive of the national governance of EURS schemes than are Europeans.
In other words, we expect that individuals with European identities are more likely to support stronger European-level solidarity and are likely to be less concerned about the loss of national sovereignty than localists and nationalists.

The Conjoint Experiment
To test our hypotheses, we use a conjoint experiment on EURS schemes. As introduced in Section I, we model alternative EURS schemes across six dimensions. The first dimension captures the generosity of the scheme by focusing on different replacement rates, that is, the maximum proportion of the last wage of the unemployed person who will be insured by the new scheme, featuring three levels (40, 60 and 70 per cent). The second dimension refers to country-level conditionality, that is, whether the EURS schemes should have no conditions attached or instead that participating countries should have education and training policies in place. Similarly, the third dimension captures individual-level conditionality, either attaching no conditions to provision of benefits, or requiring the individuals to accept 'any suitable job offer or lose the benefit', or requiring them to apply 'for at least one job per week, and accept any suitable job offer or lose the benefit'. The fourth dimension addresses the possibility of cross-country redistribution: (1) an insurance option, which does not have a redistribution option, and the support receivable by participating countries is capped by their own long-term contributions to the scheme; (2) a tolerant option, where any country can receive more from the scheme than what it contributed to it; and (3) a redistributive option, where only the poor countries are allowed to draw more out from the scheme than they pay in, while rich countries would as a result end up paying more than they receive. The fifth dimension is the level of governance at where the scheme is administered, either being solely managed by the respective national administrations, or instead mandating a dedicated EU body to administer it. A sixth and final dimension refers to the scheme's impact on domestic taxation: 'no impact', where taxation remains the same; 'taxes will increase for everyone of 0.5 per cent of their income', capturing a flat distribution of potential costs; and 'taxes will increase of 1 per cent of the income, but only for the rich', capturing a progressive spreading of the scheme's potential costs.
Fieldwork was conducted by IPSOS with a representative online survey fielded in 13 EU member states: Italy, the Netherlands, Belgium, France, Germany, Austria, Spain, Poland, Hungary, Finland, Denmark, Ireland and Estonia. 2 To ensure that the sample was representative, we included population-sample quotas for age, gender, education and regional distribution before drawing a random sample of 1500 individuals in each of the 13 countries. 3 Conjoint experiments are well suited to analyse preferences towards alternative policy proposals. Respondents can choose between different policy options (or packages) that vary across different dimensions; in turn, each dimension can take different levels, hence (1) leading to many possible alternatives, and (2) allowing a causal reading of the effect of dimensional change on preferences across policy packages . Recent applications of this method to solidarity research include the work by Bechtel et al. (2014Bechtel et al. ( , 2017 who investigated German attitudes towards Eurozone bailouts, and Gallego and Marx (2017) who investigated preferences for alternative labour market policies in Spain. Building on the methodological insights of these contributions, as well as on actual policy proposals, we set up a set of synthetic models of European unemployment risk-sharing that varies across six different dimensions. The aim was to capture the variety of EURS policy designs and to ensure that we could grasp the fundamental political attitudes (for instance, towards conditionality or generosity) of the respondents. To avoid cognitive biases as well as to maintain the attention of the respondents, the wording and framing of the experiment was kept as short as possible.
The respondents first received some general information on the context and aims of EURS (Annex 2, figure 1). The actual conjoint experiment then took place immediately after the framing. Respondents were confronted with three pairs of EURS policy packages consisting of the six policy dimensions discussed in Section I (Annex 2, figure 2). For each pair, the respondents were asked to provide the information that constitutes the main dependent variables of this analysis. They were asked to indicate which of the two packages they preferred (the choice variable) and how they rated each individual package (rating variable) on a scale ranging from being very much against to very much in favour of it. The six dimensions were presented in a random order to the respondents to avoid any ordering effect; but once the order was established for each respondent, it was maintained across the different conjoint iterations.
Operationalization Dependent variables. The data collected in the survey experiment allow two different dependent variables: package choice, on the one hand, asks respondents to identify which of the packages they prefer; package score, on the other hand, asks respondents to rate each package from very much against to very much in favour. We estimate our baseline models using the choice variable, as it is the variable with the clearest experimental nature since it forces the respondent to make a choice. For robustness purposes, however, we test our main hypotheses using alternative variables and specifications.
Independent variables. As it is customary for conjoint designs, the main independent variables are the six dimensions of the conjoint experiment themselves, that act as genuinely exogenous treatments on the respondents. We interact these six dimensions with the respondents' ideal-types. Collective identities (at regional, national and European level) are operationalized using the following question: 'on a score from 0 to 10, how much do you feel attached to … [your region/your country/Europe]'. To construct an indicator that captures not only the absolute level of identification with a certain level of territorial unity but also the relationships between them (see Kuhn and Nicoli, 2020), the attachment variable is treated to model the ideal-types of multi-level identification: localists, nationalists and European identifiers (Kuhn and Nicoli, 2020, Figure 1). To do so, three variables are constructed to sort individuals according to their predominantly local (localists), national (nationalists) or European attachment (Europeans). In each case, the value for the attachment question (regional, national or European, respectively), is to be higher than the two others. To operationalize this without losing the variation within each category, the three variables are created so to gauge the difference between the highest score (European, nationalist or localist attachment) and the second highest one of the two others. The localist score of an individual is positive when the local attachment is higher than both the European and national attachment; the national score is larger than 0 when the national attachment is higher than both the local and European attachment; the European score is larger than 0 when the European attachment is larger than both the national and local attachment. Higher scores capture more exclusive attachment to the polity in question with respect to the two other levels of community: For instance, if the localist score of an individual is 5, this means that there are at least five points of difference between it and the second-highest attachment score in the answer. Among respondents with a single predominant identity (about 60 per cent of the total sample), about 29 per cent have a predominantly localist identity, about 57 per cent have a predominantly nationalist identity and about 14 per cent have a predominantly European identity.
Control variables. Our models control for respondents' age, gender (operationalized by a dummy variable), and education (standardized by the nine ISCED-11 categories) and income (standardized by country-specific, family-adjusted income deciles). They include a dummy variable for whether the respondent is currently unemployed. We include country-fixed effects in all specifications without a multi-level structure to ensure that any country-level variability is taken into account. In certain robustness checks we account for the multi-level structure of our dataset.

IV. Results
We use several techniques to estimate how collective identities moderate the effect of EURS policy dimensions on preferences over unemployment benefits (Table 1, Figure 2). In particular, models 1-3 in Table 1 test our underlying hypotheses H1-H3 by splitting the sample in subsamples that cluster individuals according to their idealtypes. As our conjoint conditions are experimentally derived and orthogonal from one another, these baseline models are simple ordinary least squares regression estimations , with clustered standard errors and the full set of controls (including country dummies). To ensure the maximal validity of results, we exclude respondents who failed an attention check at the end of the survey, together with those who failed twice or more to provide a consistent answer between package choice and rating (that is, who first preferred a certain policy package over another but then rated it more negatively than the other). The sample is split according to the three identity ideal-types; these models are simple regression, split-sample models on the binary choice variable. To measure the variation in intensity within each score, the score is interacted with the dimensions. 4

Levels of identification and ideal-types
Models 1-3 in Table 1 test our hypotheses on the various strength of traction that the conjoint dimensions could have on individuals with a localist, nationalist or European identification. We expect individuals with a strong European identification to prefer more generous packages (H1) with cross-border redistributive elements (H2) and European governance (H3), compared with those with strong localist or nationalist identification. The subsamples for each of the ideal-types of identification are split so as to only focus on those individuals who have a positive score in each ideal-type. Table 1 and Figure 2 are based on these models. A second set of models instead codes individuals who do not fit in a specific subsample as 0. Figure 3 (interaction effects) and Figure 4 (preferred packages) are based upon this alternative coding for clarity of presentation. Figure 3 shows the marginal effects of the interactions for the three dimensions that are relevant for H1-H3.
Table 1 and Figure 2 show that individuals with superior nationalist or localist attachment are, on average, far more supportive of the national governance of EURS than those Notes: coefficients of ordinary least squares estimates and individual-clustered standard errors in parenthesis. *P < 0.1; **P < 0.05; ***P < 0.01.
with a predominantly European attachment. Interaction effects here are telling (panel c in Figure 3): the more exclusive the national or European attachment becomes, the wider the gap between the two grows. Hence, we find substantial support for H3. Similarly, the models shown in Table 1 and Figures 2 and 3 show clearly that individuals with nationalist and localist identities view less favourably schemes with clear crosscountry redistributive features than those with European identities. European identifiers are about twice as supportive of cross-country redistribution than individuals whose identification is primarily national and about 3.5 times more supportive than localists. Furthermore, the more exclusive nationalist or localist attachment becomes (that is, the larger the variable becomes 5 ), the more critical individuals become of cross-country redistribution (panel b in Figure 3). This aligns with our expectations in H2.
The relationship between generosity and collective identities (H1) is less clear-cut. As shown in Figure 2, all individualsregardless of their identificationfavour generous packages over parsimonious ones. The confidence intervals of the average effects overlap to some extent (being significantly different only between European identifiers and nationalists), suggesting that identification does not play a pivotal role here. On average, we therefore did not find strong evidence in support of H1: generous packages are on average liked by all sorts of individuals, regardless of their identification. Nonetheless, panel a in Figure 3 shows that when identities become increasingly exclusive, the gap between European identifiers (who favour generous schemes more) and nationalists (who favour generous schemes slightly less) widens. This suggests thatat least for individuals with very exclusive identities-a generous scheme could make a difference in their evaluation of a package.
To ensure that our estimates are reliable and valid, we carry out an array of robustness checks. 6 These include (1) the dependent variables (using instead of the package choice variable, the scoring variableboth in its five categories formulation and its binary transformation); (2) the controls used (including additional controls and no controls at all); (3) the sample size (including inconsistent and inattentive respondents) and the estimators (including logit for the binary outcomes and ordered logit for the ordinal outcomes and the multi-level random intercept and random coefficient models). The baseline results presented in the tables and figures remain very stable across these alternative specifications.

Most Favoured Packages
The differences in preferences for a EURS scheme across our three identity groups can be clarified further by considering predicted level of support for a given EURS policy packagenot just the marginal increase in support. This is important to explore, not least because an actual political debate can be expected to combine a range of features of policy design where voters can embrace or eschew a full combination. Our hypotheses are that EU benefit schemes that provide more generous assistance, more between-country redistribution and EU level of governance should be more strongly supported by respondents identifying as Europeans than by nationalists or localists. We explore these arguments by analysing further whether respondents strongly identifying with one or another of the three identity groups embrace package combinations in line with H1-H3.
In such an exercise it is useful to rely on models that include the full samples and then, within those samples, to focus on the (putatively different) support patterns across the localist, nationalist and European identity groups. To do so, we again take the choice of package as the outcome variable of interest and we consider how a given identity score moderates the way that the characteristics of packages influence the probability of choosing any given package. Based on such regressions, we can then surmise counterfactually the predicted likelihood that those strongly identifying as localists versus nationals versus Europeans choose a EURS package. And we can predict how such support varies across packages that themselves vary with respect to their generosity (relevant to H1), to country-level redistribution (relevant to H2) and to a European or national level of administration (relevant to H3)averaged across all other aspects of packages in the experiment. Figure 4 summarizes the most important results of such models. Figure 4a-c (generosity) shows the predicted support across packages that vary with respect to generosity (40, 60 or 70 per cent of last wage). H1, for which our earlier analysis provided the weakest and most sensitive result, does get some modest support in the analysis of ideal typical packages. We see that all identity groups are more likely to support more generous packages than less generous ones, all other things equal. But we see that the sensitivity of European identifiers to variations in the generosity of packages is greater than that of either localists or nationalistsreflected in the steep increase as one moves from the least to the most generous packages to which the respondents were exposed. And more importantly perhaps, we also see that European identifiers are more likely to choose the most generous packages than are either localists or, particularly, nationalists. The differences are quite modest, to be sure, but they are in line with our expectations articulated in H1.
Figure 4d-f show how our identity groups vary with respect to different levels of between-country redistribution. Recall that our experiment presented respondents with three possibilities: no redistribution; redistribution where any country can draw more on the new benefits programme than they pay in; and redistribution where poor countries can do so but rich countries cannot. Here, we find stronger support for the H2 pattern, where Europeans are significantly more likely than localists and nationalists to support benefit packages that involve redistribution between rich and poor countries. The pattern supports this in that the increased likelihood of supporting more redistributional packages than less redistributional ones is much sharper and clearer for Europeans than for either nationalists or localists whose propensity to support a package does not increase at all as one moves from the less to the more redistributional design. The pattern also supports the view that Europeans are significantly more likely to support the most redistributive forms of EU-level benefit schemes than their localist and nationalist counterparts.
Finally, Figure 4g-i show the starkest difference in the stances of Europeans, localists and nationalists in their support for European benefit packages when the programme is administered at the national or the EU level of governance. The models unambiguously predict that support by both localists and nationalists for packages that ensure national-level governance is substantially higher than for packages that are administered at the EU level. This pattern is stronger for nationalists than for localists, but in both cases the distinction is stark and statistically significant. For European identifers, on the other hand, the pattern is the opposite: they are statistically and substantively much more likely to support EURS packages that are administered at the EU level than those administered at the national level. All told, the patterns provide substantial support for H2 and 3, and modest support for H1. Figure 5 pools this information and focuses on ideal typical packages that combine the features discussed above. Figure 5, based on the same modelling strategy as Figure 4, shows the counterfactual likelihood that each of the identity groups will choose or support EURS packages that are less or more generous, redistributive and EU-governed. For each identity group (localists in Figure 5a, nationalists in Figure 5b and Europeans in Figure 5c), we consider three characteristics. The first (first column in 5a, 5b and 5c) predicted support for a package that is the least generous with respect to replacement rate (40 per cent), does not redistribute between countries, and is administered at the national level. The second column shows support for benefit packages with medium replacement rate (60 per cent), some redistribution by allowing any country to draw out more than they put in, and EU-level administration. The third column is for a package with the most generous (70 per cent) replacement rate, rich-to-poor-country redistribution and EU-level administration. Our expectation, surmised from H1 to H3, is that the Europeansmore than their localist or nationalist counterpartswould become increasingly and most substantially supportive of the more generous, redistributive and EU-governed package. This is precisely the pattern that emerges from the analysis. One can clearly see that the propensity to support the proposed packages increases more for the European identifiers than the others as one moves towards the most generous, redistributive and EU-governed assistance. Those strongly identifying as localists also prefer the more generous, redistributive and supranational combination. But that difference is more modest than it is among those strongly identifying as European. Also, the likelihood of Europeans to support the most generous combination in EURS packages is between 10 and 20 percentage points higher than it is for those strongly identifying as localist or nationalists. We interpret such patterns as supporting our broad line of argumentationbeyond the more disaggregated patterns focused on either generosity, or redistribution, or level of administration.

Conclusion
Existing research has repeatedly shown that European identity is an important correlate of support for European social policy (Bechtel et al., 2014;Gerhards et al., 2016;Kuhn et al., 2018). While these studies have explored support for European social policymaking in general, this article emphasizes the multidimensional nature of social policy (Gallego and Marx, 2017). More concretely, it asks what kind of European-level social policy Europeans might prefer, and how having a predominantly local, national or European identity moderates these preferences. To do so, the study draws on an original conjoint experiment conducted in 13 European member states in the autumn of 2018.
Supporting H1-H3, our analyses show that, compared with exclusive nationalists, European identifiers are more likely to support policies that involve cross-country redistribution and that are managed by a European agency, and even more so as their identification with Europe strengthens. Conversely, individuals with primarily national identities support generous packages that do not involve cross-country redistribution and are administered at the national level; these patterns hold more strongly the more exclusive their national identity is. Localists, in turn, are situated in between, supporting generous packages, but do not express such a strong preference for country-level administration and are indifferent to cross-country redistribution.
This research certainly falls short of exhausting the study of determinants of preferences towards EURS schemes. While we focus here on collective identification patterns we acknowledge that on certain dimensions (for instance, generosity of the packages) the individual attitudes towards equality or the unemployed may play an even more important role (see for instance van Oorschot et al., 2012). While such extension falls outside of the scope of this article and special issue, future research will investigate such issues further. Nonetheless, the results presented above probably underestimate the degree of polarization brought about by identification patterns. While respondents were given the choice of assigning a negative rating to all the packages (and indeed, about 6 per cent of the respondents did so), the experiment lacks the option to simply reject any sort of EUlevel social policy (an option that some strict nationalists may have found attractive). In other words, the experimental design includes a built-in attenuation bias, in the absence of which the cleavage between European and nationalist or localist identifiers would have likely been even stronger.
While our findings have shed light on preferences for a specific form of European social policy -EURSthe evidence also suggests that the design features of EU-level social-policy proposals are likely to play a major role in determining support. However, different policies may have very different detailed designs. For instance, while our data and analysis gauge a very wide range of EURS alternatives, they do not experimentally evaluate fullyfledged European unemployment benefit schemes linked to individuals' employment status alone, not their country-level employment performance. Nonetheless, our findings have important implications for research and policy-making. We corroborate the long-standing view that collective identities matter in social policy preferences and we show that collective identification can have very different implications for the popularity of different EU-level social policy dimensions.
Thus, while it may be unsurprising that European identifiers are generally more supportive of European social policy than exclusive nationalists, our findings also suggest that even individuals with exclusive national identities are ready to support EURS as long as it does not involve cross-country redistribution and as long as the funds are managed by the member states rather than by an EU body. This suggests that there may be two pathways to European social policy: one based on fullyfledged, cross-country solidarity and one based on EU regulatory powers in the field of social policy backed by a common insurance, the latter possibly being supported even by those citizens who might be reluctant to approve EU integration. Hence, our study reveals that room for manoeuvre is available to policy-makers to address the fundamental asymmetry of the EU between market integration and national social policies (Scharpf, 2002), by developing European unemployment insurance that can enjoy substantial and broad public support.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.

Annex 1: Conjoint matrix & Conjoints Examples
Annex 2, figure 1: framing (as it appears in the online survey) Annex 2, figure 2: examples of a randomly selected pair of packages (as they appear in the online survey) Annex 3: sample distribution statistics Annex 4: list of variables Annex 5: robustness checks