‘All hands on deck’ or separate lifeboats? Public support for European economic solidarity during the Covid-19 pandemic

ABSTRACT
 What are the sources of public support for international aid in times of crises? This paper investigates the determinants of public support for the EU Covid-19 aid package (‘NextGenerationEU’, NGEU), the largest aid package in EU history. Using what is to the best of our knowledge the first EU-wide study of public opinion on the Covid-19 rescue package, we first establish that public support for within-EU redistribution is influenced by long-standing factors such as ideology and citizen identities, corroborating previous literature. Next, our vignette survey experiment shows that short-term shifts in support for the rescue package are driven mostly by elite endorsement cues rather than information about the precise terms of aid or amount. We also find that the role of elite endorsement cues varies across different subsets of the population depending on their level of political sophistication, national level attachment and country context. Conversely, the effects of policy details about the aid package are more mixed. Thus, in the face of multidimensional decision-making in times of crises, national institutions are a key catalyst of moving short-term support for policies of international solidarity. We discuss how our findings have implications for democratic governance and accountability in the EU.


Introduction
While the Covid-19 pandemic has led to a dramatic loss of human life and devastating economic and social disruptions across the world, 1 countries have been affected differently. Border closings and intra-EU restrictions, such as a ban on exporting medical equipment (Wolff and Ladi, 2020), put the future of European integration in question. In response to the crises, the European Union launched a strategic recovery package designed to aid countries that were hardest hit by the crises. This aid package, known as the 'NextGenerationEU' (NGEU), is the largest aid package in EU history 2 with more than €750 billion distributed to help countries recover from the immediate economic and social consequences of the pandemic. However, the launch of the package led to elite controversies over the degree to which EU members should respond collectively to the crises, not the least in light of the uneven burden that the crises inflicted on different member states. However, despite a growing body of literature on the importance of public support for EU integration (S. Hobolt and De Vries, 2016) and foreign aid (Milner and Tingley, 2013), we know less about how this massive health crises impacts public support for European integration.
This study provides the first comprehensive, EU-wide analyses of public opinion on the EU economic response to the Covid-19 crises. 3 We suggest that while long-standing determinants of support for within-EU redistribution in general guide citizens support for aid also in times of crises, the extent to which national governments endorse or oppose the package will have an important effect on public support for international fiscal solidarity. 4 Therefore, understanding what governments support the package and if (and by whom) they are trusted, is important to explain variations in public support for international redistribution and aid. Thus, much in line with studies suggesting that the visibility and salience of national level institutions increase in time of crises (Hetherington andHusser, 2012 Devine et al., 2020), as well as literature suggesting that a crises can lead to a 'rally around the flag' effect and a heightened perceived dependency on leaders (Mueller, 1970;Bol et al., 2021), we posit that the pandemic will lead to important elite effects on public support for international redistribution and aid.
We rely on newly collected observational and experimental data, including a total of 64,089 respondents in 25 EU countries. 5 The survey was administered online from October 2020 to January 2021, when NGEU was hotly debated in the public discourse. Thus, the timing of the fieldwork and ubiquity of the crisis for the everyday life of citizens provides this study with a unique degree of in-time, real-world validity. Using an original factorial vignette survey experiment embedded in the EU-wide online survey, we test whether elite endorsement cues and/or policy details sways support for the aid package.
We first demonstrate that slow-moving individual level characteristics that explain support for EU governance more broadly, such as partisanship, ideology, geographical attachment, socio-economic status, education and age, inter alia, also significantly explain patterns of public opinion on the Covid-19 aid package in the expected direction. Our findings also concur with previous studies on public support for EU economic governance in general (Bechtel, Hainmueller, and Margalit, 2014;Kuhn and Stoeckel, 2014;Charron, 2018, 2020a) and show that a majority (roughly 55 per cent, according to our measure) of EU citizens support an EU-wide aid package in response to the pandemic. Thus, there appears to be a relatively stable baseline of support/opposition for within-EU fiscal solidarity in crisis times.
We then move to test whether various information types can change support for the NEGU via our original survey experiment. The findings suggest that the extent to which national governments endorse or oppose the package significantly influence public opinion. For example, simply receiving a positive endorsement cue from one's home government increases one's likelihood to support a European-wide response to the crisis by roughly 2 per cent on average, while receiving an opposition cue from skeptical 'frugal four' countries reduces it. We also find that the effect of positive endorsement cues is contingent on which governments endorse the package. Specifically, providing information on the NextGen package being led by France and Germany reduces support for the package, in particular among newer member states. We also test for heterogeneous treatment effects and find that elite endorsement cues enhance support among the higher educated, those with higher European identity, and by country-level economic development. Conversely, providing policy details about the terms and conditions of the aid package, such as its (unprecedented) size or repayment conditions attached did little or nothing to sway public opinion in either direction. However, information about the distributional consequences leads to less support in several of our models, in particular among the higher educated and citizens in poorer countries that are not chief beneficiaries of NGEU.
We thereby seek to contribute to the literature in several sub-fields, including public attitudes on international aid, the effect of elite cues and public support for EU integration in several interrelated ways. First, the growing literature on EU citizen attitudes toward cross-border aid transfers demonstrates that public opinion on such transfers is strongly correlated with latent political attitudes, such as cosmopolitanism, altruism or perceptions of political institutions, inter alia (Bechtel, Hainmueller, and Margalit, 2014;Kuhn, Solaz, and Elsas, 2018;Kleider and Stoeckel, 2019;Bauhr and Charron, 2018). Our findings generally corroborate most of the expectations from this literature in terms of support for NGEU. Yet while informative about the contours of public debate on the degree to which citizens want the EU involved in economic affairs, many of the factors highlighted from the literature are latent, often inter-related, longstanding characteristics, not moved to ephemeral changes, and also less relevant for policymakers seeking to sway public opinion.
Second, we further these findings via our survey experiment that allows us to directly compare the effects of more short-term communication treatments along with the aforementioned latent factors. This allows us to not only study long standing determinants of public opinion, but also factors that may sway public opinion in the short to medium term. The long-standing interest in the relative importance of policy details compared to elite endorsement cues and heuristic (see i.e., Boudreau and MacKenzie, 2014;Bullock, 2011), can be understood in light of its important consequences for the room for political elites to maneuver and ultimately the functioning of democratic accountability. Although there are considerable complexities in the way political information makes its way to opinions and ultimately policy stances, dual-processing models divide these into two main kinds: either policy-relevant information is processed systematically or citizens rely on cues from trusted policy actors (Chaiken, 1980;Petty and Cacioppo, 1986). The potential dangers associated with the public readily conforming to the policy preferences of the elite has been well documented, at least to the extent that such conformity would also entail disregarding contradictory knowledge, values or 'reasoning for themselves' (Zaller 1992, p. 45). However, the general level of knowledge about policies at the international level can be expected to be lower than for domestic policy issues, and elite endorsement cues may, therefore, be particularly consequential in the EU. Yet most studies to date on the effect of elite cues use observational data (Gabel and Scheve, 2007;Pannico, 2017;Steenbergen, Edwards, and De Vries, 2007) or do not compare the effect of elite endorsement cues to policy details (Stoeckel and Kuhn, 2018;Torcal, Martini, and Orriols, 2018;Baute and De Ruijter, 2021). Experimental studies rely mainly on respondents in single/fewcountries and rarely tests the effects of policy details in relation to elite endorsement cues, which limits our ability to compare the relative importance of these sources of information as well as the scope and generalizability of the findings. Our experiment suggests that citizens utilize elite endorsement cues rather than substantive information (save distributional information) on complex, inter-EU economic policies. 6 Finally, while some previous studies have investigated questions related to a hypothetical 'crisis' (see i.e., Verhaegen, 2018;Bauhr and Charron, 2018), most citizens can be expected to have first-hand experiences of the effects of the Covid-19 crises in terms of lock downs, health or economic consequences, and responses to the crises may, therefore, be more salient to citizens Bobzien and Kalleitner, 2020;Cicchi et al., 2020;Bauhr and Charron, 2021). Also, the Covid-19 crisis is of a magnitude that can be expected to have implications for the broader EU integration project, and therefore, provides not only an excellent 'real-world' test case for theories on the sources of public support for within EU redistribution, but also for the prospects of future EU integration.
Public support for international aid in Europe: latent factors, elite cues and aid details There is considerable scholarly work on public support (or skepticism) for European integration (S. Hobolt and De Vries, 2016). Given previous economic crises and a set of resulting bailouts, a sub-field within this literature has focused on the political consequences of the EU stepping from 'regulation to redistribution' (Borzel, 2016: 14), with an emphasis on public support for inter-EU economic solidarity, and EU economic governance in general (for example, Daniele and Geys, 2015;Kuhn and Stoeckel, 2014;Bechtel, Hainmueller, and Margalit, 2014;Margalit, 2017, Bauhr andCharron, 2018). Empirical studies suggest that latent individual-level factors significantly explain patterns in public opinion, such as cosmopolitanism, altruism, egalitarianism (Bechtel, Hainmueller, and Margalit, 2014;Ciornei and Recchi, 2017) trust in institutions and perceptions of corruption (Bauhr and Charron, 2020b;Bauhr and Charron, 2018;Daniele and Geys, 2015), left-right ideology (Kleider and Stoeckel, 2019;Bansak et al., 2020), or various types of identity (Kuhn and Stoeckel, 2014;Verhaegen, 2018 Bauhr andCharron, 2020a), along with the nearly universal finding that higher education is positively related with greater support for within-EU fiscal redistribution. Interestingly, studies on public support for EU redistribution tend to point to more limited evidence for that factors associated with economic self-interest systematically explain attitudes support, yet some point to national-level self-interest as a key exception (Kuhn and Stoeckel, 2014;Bobzien and Kalleitner, 2020).
While slow-moving latent values, identity, and enduring characteristics, such as education, are reliable explanatory factors in models of public opinion on most issues of EU integration, some studies suggest that citizen support is malleable, and that short-term communication effects can sway public opinion. Building on the literature of political communication, scholars mainly focus on the impact of policy details and/or elite endorsement cues, (e.g., who supports/opposes it). A cue has been defined as 'a message that people may use to infer other information and, by extension, to make decisions' (see i.e., Bullock, 2011, 497). An important literature investigate in particular the extent to which providing information about the party affiliation of a messenger or their policy position will affect opinions. In this study, we are particularly interested in endorsement cues by national governments, i.e the effect of particular governments supporting and opposing the aid package. Studies find that citizens use elite endorsement cues to help guide their opinion, and that this effect may be particularly relevant for citizens' opinion on the EU and its policies (De Vries et al., 2011;Hobolt, 2007;Hooghe and Marks, 2005;Steenbergen, Edwards, and De Vries, 2007;Kuhn, 2018 Pannico, 2020;Vössing, 2020). While the importance of elite endorsement cues may differ across different foreign policy issues (Guisinger and Saunders, 2017), or be strongly conditional on respondent characteristics (Nicholson, 2011), elites are often expected to play a role in structuring public opinion (Zaller et al., 1992). A positive endorsement by national level elites could be expected to strengthen public support for the aid package, not the least since recent studies suggest that the Covid-19 pandemic may have lead to an increase in institutional trust (Devine et al., 2020;Bol et al., 2021) and perceptions of quality of government (Charron, Lapuente, and Bauhr, 2021) However, the effect of endorsement heuristics may differ depending on which governments endorse or oppose the package Thus, while the messenger is typically important, and sometimes more important than the message itself (Kuklinski and Hurley, 1994), different actors can influence public opinion differently and citizens can be expected to be more influenced by elites they trust (Zaller et al., 1992;Kuklinski and Hurley, 1994). What elites endorse of oppose the package, whether domestic or from other EU countries, may thereby matter. Within EU politics, some have found that elite polarization across countries or endorsements from larger members can sway public opinion (Schmitt, S. Hobolt, and Popa, 2015;Del Ponte, 2021) Furthermore, opposition to the aid package by some governments could provide a strong signal of elite polarization around the policy proposal, which can be expected to decrease public support for the proposal. The NGEU encountered important initial opposition from the 'frugal four' countries (Austria, Denmark, The Netherlands, Sweden), which signals elite opposition and polarization. We, therefore, anticipate that support or opposition from domestic or international elites will influence public opinion: H1: Public support should increase (decrease) when provided with a positive (negative) elite endorsement cue.
In order to investigate the relative effect of elite endorsement cues on public support for within-EU redistribution, we elect to contrast these findings to more baseline utilitarian factors. Much in line with the broader literature we contrast this to more specific policy details, including the short term winners and losers of the policy, its size/prestige and its overall non-repayable benefits. Studies sometimes find that variations in the information provided on the EU aid expenditures, such as costs, distributive nature of EU spending (e.g., which countries benefit most), or re-payment structure affects attitudes toward fiscal solidarity (Bechtel, Hainmueller, and Margalit, 2017;Kuhn and Kamm, 2019;Kuhn, Nicoli, and Vandenbroucke, 2020;Beetsma et al., 2020). As per economic information, some studies suggest that citizens rely on a utility function in expressing support or opposition to international aid, whereby details concerning costs, burden-sharing, distribution or conditionality affect opinions (Malhotra and Margalit, 2010;Bechtel, Hainmueller, and Margalit, 2017;Vandenbroucke et al., 2018). Put simply, according to this framework, citizens seek to primarily maximize gains while minimizing their share of the burden. For example, relying on a conjoint survey experiment, Bechtel, Hainmueller, and Margalit (2017) find that German citizens are significantly more likely to oppose bailouts to other EU countries, when costs are higher and when Germany's share was greater, while Vandenbroucke et al. (2018) find that citizens in 13 EU countries prefer aid packages that contain certain conditionalities on average. Moreover, their study shows that citizens are swayed by the size and prestige of an EU aid package, which they find to increase its popularity among the public. Based on these findings, we anticipate that respondents will be affected by policy details about the historic size and scope of the NGEU, as well as distributional aspect of the package favoring most severely affected countries. Although policy details provide a different type of information than endorsement cues, they should not be seen as neutral 'facts' about the aid package, but are oftentimes highly directional. For instance, providing details about what countries will benefit from these funds (and by implication, what countries will not benefit from these funds, at least in the short term) is a statement that may direct opinions in particular ways, and prime considerations of distributional justice or simply utilitarian concerns about your own country not benefiting from the funds. Similarly, highlighting how much of the aid will be grants and consequently not repaid may influence the perceived fairness of the aid package among respondents and reduce support.
Furthermore, although the size of the NGEU may also lead to negative reactions among some, we believe, much in line with previous research, that highlighting this aspect will instead be perceived as highlighting prestige and the importance and historical scale of the crises and thereby trigger willingness to collaborate on common solutions. In sum, we anticipate that: H2a. (size/prestige): Framing the NGEU as the largest amount of aid in EU history will positively affect public support H2b. (conditionalities): Providing information that the NGEU allocates grants not to be repaid will negatively affect public support H2c. (distributional aspects): Providing information that the majority of funds goes to certain countries (e.g., Italy, Spain, Greece) will negatively affect support among citizens in other member states The extent to which elite cues are seen as reliable and persuasive proxies, should be at least partly related to citizens general endorsement of the elite communicating a policy position or message. Much in line with Zaller et al. (1992) RAS-model citizens would be more likely to accept and remember elite endorsement cues from elites they trust/identify with. Some studies show that elite identification across the multi-level governance system can be mutually reinforcing meaning that citizens can identify both with their nation state and the EU (Hooghe and Marks, 2005;Risse, 2015). However, the strength of citizens identification with their domestic institutions relative to EU ones is typically seen as consequential for their support for policies relating to EU integration, with strong national identifiers generally being more skeptical (Carey, 2002). If national level elites endorse a policy, however, this may partly mitigate skepticism against EU integration (including cross-boarder redistribution) among strong national identifiers. Furthermore, we expect those that identify strongly as Europeans to be more receptive of cues from EU elites.
We also expect citizens' political sophistication, measured as education levels, to moderate endorsement cues and other information. In Zaller's (1992) model citizens with a high level of political sophistication are more likely to resist framing effects (although they may also be more likely to receive them). Thus, much in line with the dual processing model (Chaiken, 1980;Petty and Cacioppo, 1986) citizens would base their opinions on cues when they have lower levels of ability (or motivation) to engage in more time consuming and complex forms of information processing (Kam, 2005;Hobolt, 2007). However, some studies also point at the opposite effect, i.e., that elite cues or policy details would have a stronger influence on swaying politically sophisticated voters. Stoeckel and Kuhn (2018) find that citizens with low levels of political sophistication are the least likely to be influenced by cues on international bailouts. This could potentially be explained by politically sophisticated citizens processing more information and that they, therefore, might encounter more contradictory information. They would consequently also be more aware of the complexities of multidimensional issues, which, in turn, make them resort to elite cues in order to deal with complexities. Thus, politically sophisticated citizens are not only more likely to know elite positions on policy issues (Gattermann, De Vreese, and Van Der Brug, 2016, see also Zaller et al., 1992), they may also be more likely to be influenced by them. As the NGEU is new, international and complex, we anticipate the following conditional hypotheses: H3a: Those that are strong national identifiers are more likely to be influenced by domestic endorsement cues, while strong European identifiers are more likely to be swayed by international endorsement cues.
H3b. Elite cues and policy details are more important to determine support for financial aid among politically sophisticated citizens, ceteris paribus Finally, building on the bench-marking literature on support for EU integration (for example, Sánchez-Cuenca, 2000;Rohrschneider, 2002), we also expect geographic differences in attitudes towards EU aid in response to the Covid-19 pandemic. In particular, a country's relative level of prosperity can have an effect on citizens' preferences for inter-state transfers and aid. In other words, respondents utilize cues from their own country's economic situation relative to others in the EU to form an opinion on if they support any contribution from their country to EU-wide aid. For example, Vasilopoulou and Talving (2020) find that citizens from poorer member states show less fiscal solidarity than do those from wealthier states, which they attribute to concerns about relative domestic scarcity and a competitive disadvantage in poorer areas of the EU. We expect this also to be the case, in particular since the poorest members in the Eastern part of Europe are not expected to be the main recipients of the Covid-19 aid package. With this in mind, we thus anticipate lower support, in general, and stronger negative effects of distributive information among the newer Central/Eastern EU members.
H4a: Public support for the EU aid package in response to Covid-19 will be lower (greater) in poorer (wealthier) areas in general H4b: Distributive aid information will reduce support more in (non-recipient) poorer EU areas.

Sample and data
To explore our hypotheses on public support for EU aid in response to Covid-19, we rely on the online sample from the latest European Quality of Government Index (EQI) survey (see Charron, Dijkstra, and Lapuente, 2014;Charron, Lapuente, and Bauhr, 2021), conducted in 25 EU member states. The survey was fielded in the months of October 2020 through January in 2021. 7 The online survey includes a total of 64,089 respondents. The respondents were sampled via a quota system according to age, gender and education levels. In addition, the survey includes a set of demographic questions and a battery of political questions used in our observational analyses. Appendix 1 provides more details on the sample, survey questions, administration and weights constructed for the analysis.

Vignette experiment
We designed a randomized factorial vignette experiment to investigate the effects of elite endorsement cues and policy details on public support for NGEU, and fielded our experiment as part of the larger survey. The vignette randomly exposed respondents to several separate pieces of information. The treatments provided concrete information about the current state of the aid proposal (at the time of question formulation) in non-technical language. The experiment distinguishes between our main expected explanations for variation in support for a common EU-responseeconomic/utilitarian policy details and elite endorsement cues (Table 1).
As you may have heard, European Union countries (treatment 1) have been negotiating an EU aid package (treatment 2 or 3) to provide economic support to those member countries most affected by the COVID-19 crisis (treatment 4). (treatment 5, and/or 6).
To test H1, treatments 1, 5 and 6 provide cues about the degree to which various elites initially support or opposed the plan during negotiations. First, treatment 1 provides information on the positive endorsement of the pan-EU leaders behind the deal, the so called 'Franco-German Plan', which may also contribute to signal legitimacy to respondents that the weight of the two largest economies facilitated the package. Second, we also capture endorsement/opposition cues via treatment 4 and 5, which signal to the respondents living in countries whose government were largely supportive of the deal according to news reports during the summer of 2020. At this time, the so-called 'frugal four' (Netherlands, Denmark, Austria and Sweden) were openly opposed to the size of the bill, and in particular the grant-aspect of the aid package going to mainly Southern European countries. Treatment 4 thus provides a cue of the opposition to the aid package during the end-round of negotiations in May-July, 2020. Respondents in these countries did not receive treatment 6 to render the information more realistic and to avoid directly conflicting and false information.
Policy details are captured by providing information on the deal itself in treatments 2-4 to test H2. Treatment 2 splits into two separate treatments where the first of these treatments (2) provide policy details on total costs of the aid package in historical context and the second (3) provides policy details about the non-repayable grants in the package. The other utilitarian aspect (4) refers to the distributional 'winners' of the money to be allocated, thus implying not all members are 'winners' in this deal. 8 In addition, the expectations that the treatment effects will be enhanced among politically sophisticated respondents, along with geographic identity and level of development (H3 and H4) are expressed in the last column. At the time of constructing the survey question, the expenditure information from the July 21st deal demonstrated that three countries, in particular, would be favored in this package due to their considerable health and economic damages -Italy, Spain and Greece. 9 Randomization checks by country show only a few instances of certain demographic groups systematically receiving certain treatments (see Appendix, section 6). In a sense, this experiment could be seen as assuming that respondents have rather limited prior knowledge about both the package and elite positions on this aid package, which would be reasonable considering that previous studies show limited public engagement with and knowledge about fiscal redistribution in the EU. It is, however, conceivable that the experiment will serve the function of primarily priming information to which some participants already had access. 10 The dependent variable: measuring support for EU aid The ensuing question following the information vignette then gauges respondents' support for the idea of the NGEU aid. Measuring 'support' (or 'preference') for cross-border financial assistance is not a simple task and there is no uniform approach to measurement currently in the literature. Several red-flags have arisen from studies on support for international aid and bailouts. For example, studies have indicated that responses could suffer from social desirability bias, as respondents could over-estimate their generosity (Kuhn, Solaz, and Elsas, 2018). The foreign aid literature also indicates that respondents are prone to an immediacy bias during or directly following a crisis, and express greater support for donations than in other times (Huber et al., 2011). Our question asks respondents to indicate their own view on an EU economic response vis-a-vis an exclusively national response via a horizontal rating scale. We do this for two main reasons. First, the literature has shown that respondents can be affected by onesided question framing, in that alternative considerations to supporting an EU-wide aid package are less likely to come spontaneously to one's mind (Chong and Druckman, 2007). Thus, in a non-technical way, we present a realistic trade-off that serves to anchor the preferences for inter-EU aid, and establish that all citizens have 'skin in the game' on this issue. Admittedly, there could be several possible contrary frames to supporting an EU aid package. Yet as Bexit demonstrated, a core tension in multi-level governance is that between national sovereignty and EU authority (Hooghe and Marks, 2009). We observed this tension in real time in regard to NGEU aid specifically; where a poll in the Netherlands during the summer negotiations showed that 61 per cent of voters did not support such an international response, and instead preferred a more national-focused recovery; thus, it was clear to many observers that the Dutch opposition to the EU-wide aid was driven mainly by domestic pressures. 11 We found that skeptics to the NGEU signal clear preferences to 'keep our funds at home'. For example, Geert Wilders (party leader of PVV in the Netherlands) said the package was a 'madness' and that this 'donation to Southern Europe' 'threw away billions that could have been spent in our own country'. 12 Similarly, the leader of the AfD, Alice Wiedel, stated that 'Merkel is once again pursuing European politics against the interests of German citizens and taxpayers in order to be celebrated as a 'great European'. 13 We found many other such statements from critics that reflect this core tension. Second, we seek to reduce acquiescence bias that would most likely arise from a standard Likert-scale 'agree/disagree'type question, and presenting a trade-off also mitigates some issues with social desirability. (De Vaus and Vaus, 2013: 106). In other words, we believed that a straight question (without 'statement 2' below) would more easily allow for respondents to simply 'agree'. We address biases due to statement ordering via randomizing statement order.
On 1-10 scale, with '1' being in full agreement with the 1st statement and '10' in full agreement with the 2nd statement, where do you place yourself on this issue? We find that overall, respondents lean toward an EU-wide aid package rather than an exclusively national one (mean = 5.9), yet there is considerable variation in preferences among Europeans (standard deviation = 2.9). As we are most interested in a positive/negative response to EU aid vis-a-vis national responses, we elect to transform the variable into a binary variable for our analysis, following previous studies (Bechtel, Hainmueller, and Margalit, 2017). Via this measure, we find that 55.9 per cent lean toward a collective EU response, while 44.1 per cent prefer a more nationally-exclusive one. 14 We also include tests with the full range of the variable in appendix, section 5.

Estimation and results from observational data
We begin by presenting the correlates with our dependent variable from the observational data available in the EQI survey. In this case, we are interested in establishing a baseline of public support for this exceptional case of EU economic integration via existing factors from the literature. Moreover, these results serve as a test of construct validity (Adcock and Collier, 2001). As our outcome is a binary measure, we employ logistic models. In addition, we provide linear estimates from the full range variable (1-10) in Table A3, which are substantively similar. Respondents are nested in 25 EU countries and 200 regions, and we account for any structural level variation via hierarchical logistic regression, with random regional-level intercepts and clustered standard errors. 15 To account for demographic imbalances between the sample and population, we include post-stratification and survey design weights, based on respondents' age, gender, education, region of residence, along with partisanship, which is recommended for online surveys in particular (Mercer et al., 2018). Our findings from the observational data corroborate much of the common wisdom from the literature. 16 The estimates reveal that, in general, females are less supportive of EU wide aid, while respondents aged 30-49 and 50-64 are significantly less supportive than younger (or older) respondents as well. Our model offers even further support for the positive effect of higher education on support for EU economic integration, and we also find that higher income earners and more urban residents are more supportive than lower income ones. In terms of occupation, we find no differences between public and private sector workers, yet people not currently in the workforce are more supportive than those who are. Finally, our measure of 'question order' shows that all things being equal, people who read the positive EU alternative first (lefthand side), tend to be 1.9 per cent more supportive.
In addition to demographic factors, we explore the effects of a number of proxies for identity, political attitudes and specific opinions on the Covid-19 crisis. Overall, we find that there is an independent effect of personal health worry about Covid-19, yet not about economic concerns stemming from the pandemic. People who are more personally worried about the virus are more supportive of an EU wide approach to aid, which is consistent with recent findings on inter-EU risk-pooling for medicines (Baute and De Ruijter, 2021). We also observe that the proxies for geographic attachment (national and European) are in the direction expected and mainly support a host of previous findings that under control for European (national) attachment, higher national attachment leads to less (more) support for EU economic integration Marks, 2005: Risse, 2015;Kleider and Stoeckel, 2019;Bauhr and Charron, 2020a), with the effect of European attachment in this case having a larger effect, all things being equal. In addition, we find that all political variables are in the expected direction, with more left-leaning respondents who are more supportive of domestic redistribution being overall more supportive of a common EU fund. Finally, the results here confirm the expectations of partisanship (see Hooghe and Marks, 2009;Bakker et al., 2015), whereby compared to other party families, Green and populist radical right parties (PRRP) partisans are significantly more and less pro-EU governance respectively. Figure 1 summarizes the relative effects of the significant correlates in rank order of greatest magnitude by marginal effects.

Experimental results
We now investigate the factors from the experimental vignette, which test H1-H2. As our observations are nested in (NUTS 2) regions and countries, we estimate the models via hierarchical estimation specifying random regional intercepts to account for structural differences in the outcome variable (Wampold and Serlin, 2000). In addition, as treatment effects on political issues such as these are found to be similar between population and convenience-based samples (Mullinix et al., 2015), given our large sample and number of countries, we employ post-stratification and design weights with our pooled sample to elucidate more generalizable findings to the population. 17 For the sake of presentation, we elect to estimate the model via a linear probability model (LPM) in order to elucidate marginal effects. Substantively, we find no differences compared with logistic estimation.  Table A2 in the appendix. Marginal effects taken from a linear probability model, and total effects imply the change in probability of '1' in the dependent variable as a results of the full range of a variable. For binary measures, these are the same. 'PRRP' stands for 'populist radical right parties'. The sign (−/+) implies the direction of the effect. Mean support from the model is 0.55. Figure 2 shows the average marginal component effects (AMCE's) from the pooled sample. In addition to the full sample (shown in black dots), we also present the results from several additional models where we exclude certain countries, namely those that are specifically mentioned in the vignette. 18 In general, we find that the randomized information provided on the aid amount or type (grants) do not affect the overall preferences for EU Covid-19 aid vis-a-vis a national response. Distributional information about the 'winners' does have a negative effect, yet significant findings are more dependent on specification choices (see appendix 5). Thus our expectations regarding H2 are mainly rejected. However, the effects of the three 'endorsement cue' variables all significantly affect the outcome, showing support for H1. In particular, the strongest effect is the information that one's national government supports the NGEU, which in turn leads to greater public support. For example, when a respondent sees that one's own government supports the European-led response, it increases the probability of support by roughly 2 per cent, on average. Conversely, hearing that there was initial opposition from the 'frugal four' countries reduces the probability of support by roughly 1.1 per cent. Interestingly, the 'elite cue' of Germany and France's leadership increases preferences for a national response to the pandemic by roughly 1 per cent. In more substantive terms, the effects of positive and negative 'elite' cues are just over half the size of high education and income (compared with low) respectively.
In appendix section 5, we replicate the AMCE's in Figure 2 with alternative specifications, namely estimates using the full range of the outcome variable (A7), unweighted (A8), and using country fixed effects rather than random regional intercepts (A9). In addition, we check the results with the removal of those that selected the middle category ('5' or '6') as it is less clear what this response implies (A10). Each sensitivity check includes the four samples presented in Figure 2. Overall, we find the effects of positive domestic endorsement cues most influential and robust to all specifications, and negative endorsement cues are significant in 80 per cent of the alternative models. In the unweighted estimates, the Franco-German endorsement cue falls from significance, while the policy details on recipients (Italy, Spain and Greece) significantly leads to less support for an EU-wide response. Adding control variables to account for any imbalances in treatment assignment does not affect the results, while removing the middle category responses enhance all endorsement effects and render recipient details significant.

Heterogeneous treatment effects
Figures 3-5 present the conditional effect of the treatments by sub-groups. We follow the recommended practice by Leeper, Hobolt, and Tilley (2020) and plot the marginal means of the outcome variable for treatment and control groups. We observe a significant conditional treatment effect in cases where the confidence intervals around the marginal means of treatment and control groups do not overlap. Figure 3 presents the interaction with the levels of national attachment and the randomized components. 19 Overall, as shown in Figure 1, we observe first a large gap in the dashed lines indicating a strong and negative effect of national identity on support for a collective EU economic response to the pandemic, consistent with the literature (S. Hobolt and De Vries, 2016). In evaluating H3a, we see that there is a significant treatment effect of domestic (positive) endorsement cues on support for EU Covid-19 aid among both low and high national identifiers. Substantively, the marginal effect among high and low national identifiers is a 3.1 per cent and 1.4 per cent increase in probability of support respectively, ceteris paribus, meaning positive domestic endorsement cues have a significantly greater effect among national identifiers than for those with relatively higher European attachment. 20 This implies that some of the negative effects of national identity on support Note: Average Marginal Component effects (AMCE), with 95% confidence intervals from regional clustered standard errors. Estimates are change in probability in supporting an EU-wide response vis-a-via an exclusively national one. Models estimated via hierarchical linear regression with random regional level intercepts specified as the level 2 variable. Results for full sample reported, along with three additional models in which the specific country samples, which are explicitly named in the vignette, are removed. Estimates include weights based on gender, age, education, partisanship and regional residence. Further results shown in Appendix, Table A5. for a common EU economic response can be partially mitigated by positive endorsement by the domestic government. Additionally, the figure shows that among those with relatively lower national attachment vis-a-vis Europe are more receptive to information in generalnamely, that there is a small increase in support (1.4 per cent) for an collective EU response when details about the historic amount is revealed, yet a decrease by 1.6 per cent when information about other Member States' skepticism is revealed, showing support for H3a. We also find that among those with higher attachment to Europe in general vis-a-vis one's home country show more support when told of the historic amount of inter-EU aid, as per previous research (see Vandenbroucke et al., 2018).
Next, we report the findings for the test of H3b, which anticipates heterogeneous effects based on political sophistication, proxied by education levels in Figure 4. First, we see that compared with those with less than tertiary education, those with higher levels of education are nearly 5 per cent more supportive of EU aid on whole (shown via the two dashed lines). Second, with respect to the heterogeneous treatment effects, we find indeed that the treatment effects are highly contingent on levels of political sophistication (education), as anticipated in H3b. For, example, the effects of both negative and positive cues significantly sway higher education respondents in the Estimates from a hierarchical linear regression model (random regional intercepts), with 95% confidence intervals from regional clustered standard errors. Heterogeneous effect shown via split-sample models. Dashed lines represent the average marginal mean for the low and high educated samples. Design and post-stratification weights included. Sample size for high and low national identifiers is 27,389 and 36,700 respectively. Further results shown in Appendix, Table A4. anticipated direction (−2.4 per cent and 2.9 per cent respectively), yet do not yield a significant effect among lower educated respondents. We also observe a significant, negative effect of information on the distributive aspects of the aid package; namely when higher educated respondents are presented with information on the 'winners' of the aid, their support is 1.6 per cent lower than those without receiving such information, which suggests some conditional evidence for utilitarian factors. Yet even among the higher educated, we do not see that aid details regarding size and structure matter. Interestingly, we find that none of the treatments significantly affects opinions among respondents with lower levels of political sophistication (education). 21 These findings largely corroborate H3b, and lend additional support to previous studies (see for ex. Chong and Druckman, 2007: 112), which argue that for some issues, greater prior sophistication enhances informational treatment effects.
Finally, to test H4a and H4b, we take a parsimonious binary proxy for relative wealth, using as a metric whether or not a Member State belongs to the Cohesion Fund (CF) from the last budget cycle (2014)(2015)(2016)(2017)(2018)(2019)(2020). 22 First, we see that there is an overall difference in means between support for a common Note: Predicted Marginal means (x-axis), averaged across all other factors, by Treatment and control groups. High (low) education respondents are on the right (left). Estimates from a hierarchical linear regression model (random regional intercepts), with 95% confidence intervals from regional clustered standard errors. Heterogeneous effect shown via split-sample models. Dashed lines represent the average marginal mean for the low and high educated samples. Design and post-stratification weights (by gender, age, education and partisanship) included. Sample size for high and low educated is 31,532 and 7642 respectively. Further results shown in Appendix, Table A4.
EU-response aid package between respondents in CF countries (black) and non-CF countries (blue), with the former showing greater support on average, which runs counter to H4a.
Second, in testing H4b, we compare the effects of the 'recipient' treatment in the two groups. As anticipated, the effect is negative among CF countries (p = 0.08), and the effect increases when Greece is excluded (−0.017, p = 0.05). Thus, regarding H4, we find that there is greater overall solidarity among poorer regions, yet such support is weakened by distributive information that does not signal these areas as primary beneficiaries of the aid.
Other noteworthy findings in Figure 5 include a consistent and positive effect of domestic endorsement cues regardless of relative level of development with the EU. However, we find that effects of the other two cues depended on whether one is in a wealthier or poorer country. Specifically, we find that the effect of negative cues reduces support among respondents in wealthier, non-CF countries by roughly 1.6 per cent on average. Yet on the cue regarding Franco-German leadership, we see a strong, negative impact in CF countries; decreasing support for an EU response by roughly 2.1 per cent. Estimates from a linear hierarchical regression model (random regional intercepts), with 95% confidence intervals from regional clustered standard errors. Heterogeneous effect shown via split-sample models. Dashed lines represent the average marginal mean for the CF and non-CF samples. Design and post-stratification weights included. Further results shown in Appendix, Table A4.

Discussion
In response to the Covid-19 pandemic, the EU launched the largest aid package in its history, the Next Generation EU (NGEU) to aid the countries hardest hit by the pandemic. This study is to the best of our knowledge the first, comprehensive EU-wide investigation of public support for the EU Covid-19 aid package. Our data shows that a slight majority is in favor of cross-border redistribution (55 per cent) to mitigate the effects of the pandemic, yet public opinion is polarized within the EU. We thus observe similar levels of support for EU economic integration efforts as do past related studies. When public opinion is polarized around a political issue, it is perhaps particularly important to investigate the factors that shift public opinion as well as the consequences this may have for democratic accountability and the legitimacy of the aid package. Using a unique experiment in 25 European countries we investigate the determinants of public support for this aid package, in the midst of the pandemic in Europe. Our results show that while long-standing determinants of support for within-EU redistribution in general guide citizens support for aid also in times of crisis, the extent to which national governments endorse or oppose the aid package will have important shorter-term effects on public support for international fiscal solidarity.
Our findings provide evidence that there is a relatively stable baseline of public support for aid in times of crises. Variation in this support can chiefly be explained via enduring explanatory factors from social identity models of public opinion (Hooghe and Marks, 2005), such as ideological orientation/partisanship and geographical identity (Bechtel, Hainmueller, and Margalit, 2014;Kuhn and Stoeckel, 2014;Charron, 2018, 2020b). However, even with relatively brief access to cue respondents, we also show the importance of short-term communication and framing effects. Particularly, we show that elite cues are on average more consequential than providing information about the precise terms and size of the aid package. We find that domestic elite endorsement increases public support for the aid package, while both EU leadership endorsement (France and Germany) and elite polarization (opposition from the 'frugal four') decreases public support. Conversely, we find that the details on the aid package itself, such as costs or structure does not affect public support in either direction. We find some evidence, however, (depending on weighting choices/ conditional variables) that specifying the 'winners' leads to less overall support.
In addition, our tests of heterogeneous treatment effects demonstrate that some European citizens are more strongly affected and persuaded by elite cues than others in general, but also that the persuasive appeal of different elites varies across different groups of citizens. The effect of elite cues are strongest among citizens with high levels of political sophistication. We also find that domestic endorsement cues are more persuasive among citizens with a high level of national attachment. However, endorsement of the package by Franco-German elite, and specifying that Southern European countries will be the main recipients of these funds, clearly depresses support for the aid package among highly educated citizens and those in relatively poorer member states from Central/Eastern Europe.
Studies to date disagree about the extent to which political elites follow the will of the people in democracies or if the elites instead shape citizens opinions on political issues in a more top down fashion (for ex., see Lenz, 2013;Steenbergen, Edwards, and De Vries, 2007). We show evidence for the latter, yet further work is needed to investigate the former. Also, the crisis may have heightened the perceived dependency on national political elites and the persuasive appeal of these elites may potentially be somewhat lower in other contexts of aid and international financial transfers. However, in all, we believe that situating this study in the midst of a pandemic provide more reliable estimates on citizens responses to a crises than previous studies asking about hypothetical crises.
While our study makes a number of empirical and theoretical contributions for several research fields, there are in addition clear implications of our findings for democratic governance and accountability in the EU. Ultimately, the extent to which citizens rely on elite endorsement cues affects how office holders are held into account. When citizens rely more or less exclusively on elite cues in forming opinion about large and complex policies, such as the NGEU aid package, the evaluation of the policy details may simply be less relevant. Instead, citizens' identity with, and trust and distrust in, both domestic and EU elites become consequential in supporting policies rather than the actual policies themselves. Notes 7. The survey was fielded as a hybrid between telephone (CATI) and online interviews. Yet for the purposes of the experiment used here, the question was asked exclusively to the online group. 8. Comparing the effects of different forms of information is of course not always entirely straight-forward. One issue that potentially arises is the extent to which different types of information differ in directionality (we thank one of our anonymous reviewers for pointing this out to us). However, although some elite endorsement cues may be highly directional, i.e., domestic endorsement cues, endorsement and opposition by other countries (i.e., France/Germany support or opposition from 'the frugal four') would most probably provide less clear information to citizens, and be dependent on citizens perceptions of these elites or sensitivity to the level of elite consensus. Details on i.e., the extent to which the funds will be channeled to countries other than your own in the short term may potentially provide a clearer and more strongly directional form of information. Thus, although endorsement cues may potentially on average be more strongly directional, which could partly influence our results, this is only one potential difference between our endorsement cues and policy information. 9. See for example: https://www.dw.com/en/eu-leaders-reach-deal-on-coronavir us-recovery-package/a-54242834 10. Since treatment effects are generally larger among highly educated respondents, our results also indicate that treatment effects are not contingent on respondents having little or no prior knowledge, at least to the extent that education levels provide a reasonable proxy for access to information on the aid package. 11. see for example: https://www.atlanticcouncil.org/blogs/new-atlanticist/whythe-netherlands-opposed-unconditional-european-coronavirus-aid/ 12. https://twitter.com/geertwilderspvv/status/1285433774998790144?s=20 13. https://twitter.com/AliceWeidel/status/1285484898111959040 14. Although measured differently, we find that this level of support is generally in line with previous research on EU fiscal solidarity in Europe. For example, Vasilopoulou and Talving, 2020 and Bauhr and Charron, 2018 find that roughly 53% of citizens EU-wide support helping struggling member states 'in times of crisis'. Baute, Abts, and Meuleman, 2019 finds that Belgians lean toward favoring fiscal solidarity 'rich countries should support poor EU countries' (mean 2.95 of 1-5 scale in favor), while Bechtel, Hainmueller, and Margalit, 2017 shows that a baseline of 47% of Germans support bailouts to struggling member states. Moreover, in a smaller pilot survey in 3 member states (June 2020), we find remarkably similar support (Appendix 1) 15. We elect to use the NUTS 2 regional level because this is the sampling unit of the EQI survey, which also provides 200 second level units rather than just 25 and greater spatial variation (see Figure 1 in the appendix for regional level variation of the dependent variable). In addition, we tested alternatively a country fixed effects model and the individual level estimates were statistically indistinguishable in nearly all cases, thus for efficiency, we elect the hierarchical model. 16. see Appendix, sections 2-3 for further variable description and Table A2 in for full results. 17. See appendix, section 5 for unweighted effects. 18. In particular, we are interested in the treatment effects for the positive endorsement cue minus the 'frugal four', and the distribution treatment, which specifies the three 'winner' countries by name. We also exclude France and Germany in the last model mainly to check the sensitivity of the effects on the first cue treatment. 19. We build on Bauhr and Charron (2020a) to construct our measure of relative national identity (see appendix, section 3 for more details). 20. A formal post-test of the interaction between national identity and positive domestic cue treatment is significant (p = 0.003). 21. In comparing the relative effects of the treatments by sub-groups, post-tests show the effects of both cues are significantly greater among high educated respondents at the 95% level of confidence greater, while the effect of recipient information is significant at the 90% level of confidence (p = 0.08). 22. CF members are at the 90% average of GDP per capita or lower within the EU, see Table A1.