What kind of EU fiscal capacity? Evidence from a randomized survey experiment in five European countries in times of corona

Abstract

Based on a conjoint survey experiment, we explore the support among European citizens for a European Union (EU) budgetary assistance instrument to combat adverse temporary or permanent economic shocks hitting Member States. Suitably designed, there is substantial support for such an EU instrument generally and across the sample countries. Support is broader when budgetary support is conditional on debt reduction in good times and on monies being spent in specific policy areas, in particular healthcare and education. Support also increases when there is a role for the European Commission in terms of monitoring and providing guidance. However, there is little support for policy packages that terminate a programme and impose fines in the case of non-compliance. Further, there is broad acceptance of programmes that entail long-run redistribution towards poorer countries. Financing the assistance through a progressive tax increase is more popular than through a flat tax increase. In general, there is substantial scope for constructing assistance packages that command majority support in all sample countries, particularly if programmes have spending conditionality and progressive tax financing.

1. INTRODUCTION

During the first decades of its existence, Europe’s Economic and Monetary Union (EMU) has suffered from large and uneven swings in the economic performance of Member States. Some of the divergences have been caused by asymmetric shocks, but most can be attributed to severe common shocks that have propagated differently through the EMU. This has in particular been the case for the global financial crisis, the Eurozone debt crisis and, most recently, the Covid-19 crisis. The capacity to stabilize the common element of the dynamics is limited by the constraints on the ECB’s policy instruments, while some EMU Member States have effectively become unable to use fiscal policy to stabilize their economies.

It has long been argued that a viable EMU needs meaningful budgetary instruments to deal with the adverse shocks and, in particular, when they cause divergences.¹ As a result, EU level policymakers have presented proposals for further fiscal integration. The ‘Four Presidents’ Report’ (Van Rompuy et al., 2012) envisaged the gradual creation of a central fiscal capacity (CFC) to promote structural reforms and mitigate asymmetric shocks. The ensuing ‘Five Presidents’ Report’ by Juncker et al. (2015) described the path to completion of the EMU with a fiscal union as a major building block. The report emphasized that a Euro-area stabilization function should avoid permanent transfers, which requires preceding structural economic convergence, and be compatible with an incentive to conduct a sound fiscal policy. Such a capacity would aim at promoting resilience to temporary economic shocks. The European Commission’s (2017) ‘reflection paper’ described different concrete options for a euro-area macroeconomic stabilization function. These broad and long-horizon proposals have been followed by small-scale concrete initiatives. In 2018, the Commission proposed a European Investment Stabilization Function (EISF) to be embedded in the 2021–27 Multiannual Financial Framework (MFF). The EISF would provide for 30 billion of low-interest loans to Member States. However, the EISF proposal died because of lack of political support. Parallel the Commission worked on a so-called Budgetary Instrument for Convergence and Competitiveness (BICC), also to become part of the new MFF. The BICC would provide resources for structural reform. When stating her priorities at entry into office, the Commission President mentioned a European reinsurance of national unemployment benefit schemes (Von der Leyen, 2019). The Covid-19 crisis has led to several new Commission proposals. The European instrument for temporary Support to mitigate Unemployment Risks in an Emergency (SURE) is a 100-billion-euro facility backed by guarantees by Eurozone member states to provide cheap loans to countries to maintain employment during the Covid crisis. Most recent is the 750 billion euro ‘Next Generation EU’ to support the economic recovery from the Covid crisis. Its main component is a temporary Recovery and Resilience Facility (RRF), which provides grants and loans for investments and reforms and will replace the BICC (European Commission, 2020).

The support of national political decision-makers for expansion of EU budgetary assistance instruments appears to be limited. However, while politicians frequently express their position, claiming to have the support of their voters, how their populations really think about the introduction of new EU budgetary instruments is less clear, especially since such instruments can come in different forms and with potential conditions attached to them. In this paper, we therefore address the question what kind of EU budgetary assistance arrangement, if any, citizens from different European countries prefer. Existing public opinion data are mostly based on surveys that present policy elements in isolation, or with a very parsimonious amount of detail, in order to ‘protect’ survey respondents from complexity. However, the responses to questions on policies presented without any detail will not reveal much about the actual policy preferences, simply because respondents have no chance to express their position on realistic and completely formulated assistance packages. Nonetheless, knowledge of popular support is crucial, as politicians are accountable to their population and need the popular support for the long-run viability of such proposals. Pushing ahead with policy designs that are disliked by large parts of the population will likely cause a backlash in the longer run undermining further European integration.

While recent studies have explored popular preferences on a range of EU-level policies,² to the best of our knowledge, there is no analysis yet on detailed preferences towards alternative designs of an EU CFC. Consequently, this paper contributes to addressing this gap by using the results of an experiment that combines a framing experiment with a so-called ‘conjoint experiment’. The survey experiment was fielded in March 2020 to 10,000 representative respondents from five EU countries (France, Germany, Italy, Netherlands and Spain), in order to shed light on their support for EU budgetary assistance packages for countries facing temporary or permanent economic distress.

In the experimental part of the survey, we first randomly assign respondents two possible frames, building upon a distinction between temporary and permanent shocks that is conceptually important in the context of the optimum-currency-area theory (De Grauwe, 2018). In one, the distress is temporary, which would typically be the result of a dip in the business cycle. The other describes a permanent negative shock, which would for example result from a permanent decline in an important economic sector. We choose these frames not only because of the empirical relevance of the different conditions they describe, but also because they may call for different policy responses, which may in turn attract different levels of popular support. In particular, the response to a permanent negative shock might be perceived as generating long-term redistribution.

The experimental part of the survey then carries out a conjoint experiment to gauge respondent attitudes about a possible EU fiscal capacity programme to address such permanent or temporary economic shocks. In that conjoint experiment, respondents see side-by-side two different policy packages characterized by a number of ‘dimensions’ taking particular values, the random assignment of which are ‘treatments’ for each package that a respondent sees and judges. The dimensions for each package pertain to (1) the need for conditions on the support; (2) how the resources are to be spent; (3) how domestic taxation will be impacted; (4) whether long-run redistribution among countries is tolerated or even an objective; (5) what role the European Commission should have and (6) whether and how countries should be punished in the case of non-compliance with the programme’s conditions. A package consists of a full set of choices, one for each dimension. The dimensions characterizing the assistance packages are motivated by the main elements featuring in actual proposals made so far by officials and other experts as well as in the discourse in these circles. Respondents are shown three pairings of packages (six packages total, hence). For each of these pairings, respondents are asked to choose which of the two shown packages they like most (or dislike least), and also to indicate the extent to which they support each package (both the chosen package and the ‘rejected’ one). The most important advantage of this conjoint setup is that it allows for causal inference of treatment effects of policy design on preferences, that is, support or preference for a particular policy package, resulting from randomly varying the policy features of a specific dimension shown to respondents representative of their population, while holding the features or settings for all the other dimensions constant.

The results of the nested survey experiments paint a quite clear picture of citizen support for EU budgetary assistance programmes. First, we find that there is generally quite widespread support for such programmes in the face of both temporary and permanent shocks. In fact, respondents’ preferences appear to be essentially the same for both the temporary and permanent shocks scenarios. Second, there is a remarkable congruence in support when it comes to a programme’s allocation of resources. As regards the spending of the resources, interestingly, but not entirely surprisingly given the March 2020 timing of our survey experiment, respondents tend to strongly support health care spending, followed by support for education spending and respondents express little support for spending on the banking system and deposit holders. Respondents express only modest support for budgetary conditions (related to concern with stabilization), also in response to a temporary shock; also in this case, respondents are more strongly swayed by the allocation role of spending than budget conditionality. Third, there is support for an active role of the European Commission in terms of monitoring the implementation of the programmes and providing guidance. Fourth, based on support for programme rules on paying-in versus taking-out of the assistance facility by participating countries, respondents tend to prefer programmes that allow some long-term redistribution (where all countries can take-out more than they pay in), and particularly programmes that entail redistribution from rich to poor countries (where only poor countries can take out more than they pay in). Fifth, financing the budgetary assistance programme through progressive taxes is preferred to financing it with a flat tax increase for everyone. Sixth, in the case of non-compliance with the conditions of the programme, the preference is to examine the reasons for non-compliance, but not to terminate it and impose a fine.

There are also differences in the attitudes among the countries. Support for an assistance programme is on average highest for Spanish respondents and, depending on the measure of support used, lowest for French or Dutch respondents. The differences in average support among the countries are quite limited, though. In terms of the individual dimensions, however, the Dutch and Italian samples stand out somewhat. The Dutch are the only population against any cross-border long-run redistribution and they are the only ones supporting termination and imposing a fine in response to non-compliance. The Italian respondents are the only ones not strictly favouring budgetary conditions for financial assistance. These differences between the Italian and Dutch respondents in the survey mirror similar differences in the positions by their respective governments in EU-level discussions on how to respond to the Covid crisis. The preferences of the other countries’ respondents appear to reflect the more ‘middle-ground’ positions of their governments.

Still, there is overall rather substantial congruence among the preferences of the different populations. This opens the possibility of finding assistance packages that get majority support from all individual countries. A package that commands such unanimous cross-country support is characterized by a combination of budgetary conditions, mandatory healthcare spending, monitoring and guidance by the Commission, redistribution to poor countries, progressive taxation and no termination and fines following non-compliance. Finding unanimous support becomes more difficult when shifting to flat tax financing or requiring spending in other areas. Still even with these variations, unanimous support may be found if we relax our conservative measure of support somewhat.

How confident can we be that our results reflect the ‘true’ preferences of the respondents? It is important to realize that our survey is based on respondent views in a (partially) pre-political environment, that is, before any concrete policy proposals are debated by political parties that seek the edges of polarization. Hence, our survey gives respondents the opportunity to reason and form their own opinion about the assistance package, thereby providing the best possible guarantee of expressing their true views.

The remainder of this paper is structured as follows. Section 2 reviews the related literature. Section 3 describes the conjoint experiment in detail, while Section 4 reports and interprets both the aggregate and country-level results. In Section 5, we explore the support for selected policy packages. Finally, Section 6 concludes the main body of the paper.

2. LITERATURE ON THE POLICY DEBATE ABOUT EU FISCAL INSTRUMENTS

The debate on the EU-level policies distinguishes between instruments aimed at reducing structural economic differences among countries, which manifest themselves in systematic differences in welfare and competitiveness and instruments aimed at addressing the consequences of unforeseen shocks hitting EU economies. The need for the different types of instruments obviously depends on the empirical nature of the shocks. How large and frequent are the shocks? Do they affect countries symmetrically or asymmetrically? Are they temporary or permanent?

The original Optimum Currency Area theory emphasized the need for adjustment mechanisms in response to asymmetric shocks. Mundell (1961) studies the role of labour mobility, while Kenen (1969) explores the need for fiscal coordination. Contributions made during the run-up to EMU hypothesized the potential endogeneity of the degree of business cycle synchronization. One view argued that enhanced trade and investment flows in the EMU lead to geographical concentration of sectoral activity and, hence, to more specialization, implying that sector-specific shocks increasingly become country-specific shocks.³ The essentially opposite view hinges on the idea that intensifying intra-industry trade flows will cause country-specific business cycles to become more aligned (Frankel and Rose, 1998).

De Grauwe and Mongelli (2005) arrive at moderately optimistic conclusions when exploring to what extent the process of monetary unification itself contributes to the fulfillment of the optimum currency area criteria.⁴ However, ensuing developments make clear that much of the divergence dynamics among the Eurozone member states is due to large common shocks that propagate differently or with a different intensity through the various parts of the area.⁵ This is in particular the case for the developments that were ignited by the global financial crisis, the Eurozone sovereign debt crisis and the current Covid crisis. The role of the ECB in combatting union-wide overcapacity has become impeded by the zero lower bound constraint, while its possibilities to address asymmetric developments are limited in any case. This task naturally lies with fiscal policy, which is constrained by the high levels of public debt in some countries severely hit by the Covid crisis.

A crucial element when designing facilities at the European level is how their deployment differs between temporary and permanent shocks. This is important for at least two reasons. First, the two types of shocks may call for different types of support policies. For example, De Grauwe and Ji (2016) favour a shift in emphasis from structural reforms to risk-sharing arrangements to stabilize business cycles. Second, support in response to a permanent negative shock might be perceived as creating long-term redistribution. Indeed, much of the resistance to setting up fiscal facilities at the European level appears to be driven by the fear that these lead to permanent transfers among countries, hence structural redistribution, instead of mere risk sharing. The need to avoid permanent transfers is spelled out in, for example, Juncker et al. (2015). In view of the potential concern with structural redistribution, one of the dimensions of our conjoint experiment addresses preferences concerning long-run redistribution.

A major concern with EU transfer programmes is the danger of moral hazard (potentially leading to the much-feared structural redistribution): aware of the fact that they will receive support in the case of an economic decline, a country’s policymakers may choose to cut back on politically costly economic reform or act in a fiscally less disciplined way than they would otherwise do.⁶ Concern with moral hazard is a reason why the debate on further EU budgetary integration has come to a stalemate. Some countries want to see risk-reduction first, before facilities for risk-sharing desired by other countries can be set up (Beetsma and Larch, 2018). Bénassy-Quéré et al. (2018) recognize the legitimacy of the concerns of both country groups and make a number of proposals for politically acceptable progress with the completion of the Eurozone architecture.⁷ Concern with moral hazard is also a reason why, for example, support from the ESM comes with conditions embedded in a macroeconomic programme intended to address deficiencies, such as weak tax collection, a bloated public sector, inefficient product and labour markets and the like. Fear of moral hazard, and the need for ‘conditionality’, also dominates much of the discussion about EU support for recovery from the Covid crisis.⁸ Hence, in our experiment, we will investigate the role of budgetary conditions on support for EU assistance programmes. We will also investigate support for Commission monitoring and guidance and the handling of potential non-compliance with the programme’s conditions.

Various concrete proposals, both by policy institutions and academic experts, have been made for some CFC to support countries experiencing temporary or more permanent economic hardship. Besides the initiatives discussed in Section 1, there have been pleas for a CFC from the European Fiscal Board (2017, 2018) and researchers of the IMF (Arnold et al., 2018).⁹  Claeys (2017) proposes a euro-area stabilization tool of limited size to manage the aggregate fiscal stance and to provide risk-sharing against large shocks hitting individual member states. Different designs can be envisaged. One would be a scheme that protects investment in a downturn – the Commission’s EISF could have been an embryo for this. Such a scheme could serve both a short-term role in keeping up demand and a longer-term role by improving a country’s productive capacity. Another design would be the reinsurance of national unemployment benefit systems. This option, which differs from a genuine European unemployment benefit scheme, has been examined in various publications (e.g., Beblavý et al., 2015; Beblavý and Lenaerts, 2017; Dolls et al., 2018). Because equilibrium unemployment differs across Eurozone countries, it has been proposed that transfers be triggered when a so-called ‘double condition’ is fulfilled: unemployment should exceed its historical average over a long period and it should have increased substantially in a short time period (see, e.g., Carnot et al., 2017).¹⁰ In view of these different possible designs, one of the survey dimensions concerns the question how financial support should be spent.

Finally, this paper is related to a strand in the literature investigating public support for European-level policies. While the number of contributions in this area is enormous, the papers using experimental methods comparable to ours are, as yet, very few. Bechtel et al. (2017) use a conjoint experiment to analyse how the likelihood that German respondents reject a bail-out plan of other Eurozone economies is affected by the features of the plan. They find that only a small minority of respondents are fundamentally opposed to a bail-out, while most respondents support a bail-out when it comes with certain features. Of particular importance are the costs of the programme, the degree of burden sharing among creditor countries and conditions on the specific austerity policies of the recipient country. More recently, Hahm et al. (2019) have looked into the role of institutional reforms in determining support for European integration. The design of the current experiment is partly led by the experience from an earlier project (Vandenbroucke et al., 2018; Kuhn et al., 2020; Nicoli et al., 2020; Burgoon et al., 2022). That project explores public attitudes towards the construction of a European unemployment reinsurance scheme. It finds substantial support for such an instrument, as long as a proposed policy mix includes sufficient generosity and conditions with regard to job search efforts by the unemployed and education and training efforts for the unemployed, preferably in combination with redistributive tax financing and national-level administration. The experiment in the present paper considerably enlarges the policy areas studied beyond unemployment benefit provision. Moreover, our experiment focuses on a number of other dimensions than those of previous studies.

3. DESCRIPTION OF THE FRAMING AND CONJOINT EXPERIMENT

Our research design relies on randomized survey experiments, particularly a combination of a framing experiment and a conjoint experiment. The framing experiment randomly assigns the way the EU fiscal programme is introduced to respondents – distinguishing a programme aimed at addressing permanent shocks or instead aimed at temporary shocks. With either framing, our research design then gauges respondent attitudes about different policy features of the proposed EU fiscal programme using our more important experiment, what is known as a ‘conjoint’ or ‘factorial’ experiment. This is a type of randomized survey experiment that needs to be distinguished both from regular survey questions and from simpler framing or survey experiments in which respondents are asked about their view on individual policy items. In a conjoint experiment, respondents are presented with policy packages, that is, combinations of measures on a set of policy dimensions (which we explain momentarily).

The fieldwork of our experimental survey (the conjoint experiment nested within the framing experiment) was carried out by the specialized firm IPSOS in late March 2020 in five European countries – France (FR), Germany (DE), Italy (IT), the Netherlands (NL) and Spain (ES). Respondents took the survey via an online platform in their own language, including Catalan. We selected these five EU Member States to cover a variety of economic performance and structure and to capture a balance of northern and southern European polities that are known to differ substantially in their views on EU budgetary integration. Moreover, these countries constitute the five largest euro-area member states. Hence, they are likely to have the largest weight in designing an assistance programme. In each country, we have 2,000 respondents, yielding 10,000 respondents in total.

The sample is selected to be representative of each country’s populations in terms of education, age, gender, profession, regional distribution and income. The sample is drawn from the IPSOS country panels, themselves based on address-based sampling (ABS) to ensure good representation of national populations (Fahimi, 2009). Based on these panel populations, respondents within the relevant sample tranches are invited to fill out the survey, yielding representative sample shares in terms of education, age, gender, profession and regional distribution and a hard quota was also applied to ensure representativeness of equivalized income distribution. Seeking representativeness – in general, and particularly on these individual characteristics of age, income and education – reduces potential selection effects, for instance by ‘pro-European’ or ‘anti-European’ individuals having a particularly strong desire the participate in or shun the survey. An overview table of the discrepancy between the full population and the sample with respect to these characteristics (available upon request) shows that the discrepancies are generally small. Further, IPSOS, like virtually all online-panel survey organizations, pays a small fee to the respondents in their panels, something that should reduce the effect of various attitudinal biases (again, such as innate pro- or anti-European feelings) on the decision to participate. Finally, the results of our analysis of the resulting sample, discussed below, do not exhibit unduly large fractions of individuals with extreme views on survey items that might relate to response (social policy and EU integration mentioned in the survey invitation to panellists). As with most online panels, IPSOS provides little information on non-response. However, empirical studies on non-response in online panels suggest that non-response tends not to yield significant bias (Lee, 2006).¹¹ The Roberts et al.’s (2014) study of item response and quality in online panels finds that those who ultimately respond (i.e., complete a survey request) are likely to yield higher quality item response and quality answers.

We confront respondents with two different descriptions of an economic policy problem that is to be addressed by an array of new EU policy proposals, which come in packages. This creates two different ‘framings’ for the survey experiment, of which each respondent gets to see only one in total. Appendix A presents the exact texts. The first frame describes a temporary decline in the economy, typically a worsening of the economy’s business cycle. The second frame describes a permanent decline in the economy. This could be a permanent decline in an important industry or sector or a permanent shift in consumers’ preferences away from certain national products.

In designing framing or conjoint experiments, we need to strike a balance among the following elements: the need to embed the dimensionality of the public’s concerns, the need for a sufficiently simple presentation, so that it can be understood by the respondents and the need to present policy packages that are as realistic as possible. Hence, in the design, we are guided both by practical concerns on the feasibility of the experiment and the need to be able to address our basic research question. Therefore, we confine ourselves to presenting respondents with pairs of randomly selected policy packages consisting of six dimensions. While some literature finds higher thresholds in dimensions, values and pairings to be possible (Sauer et al. 2011), we found that pairings of packages with six dimensions was the most that could fit on most computer screens (see discussion below) and, based on pretesting, was the rough maximum to avoid overloading respondents.

The dimensions shown for a given package constitute the actual treatments in the experiment, whose randomization thus allows for robust causal inference. To avoid biasing respondents into a particular direction, the introductory frames try to present the circumstances and the assistance packages as neutrally as possible. For the same reason, in the experimental part presenting the assistance packages themselves, we do not mention explicit ‘pros’ and ‘cons’ of packages. Instead, the potential benefits and costs are implicit in the possible answers to the dimensions of the policy packages presented to the respondents (see below). Finally, one may ask whether various policy design features of packages are credible and will be enforced. Enforcement of agreements is crucial to the actual policy development of the EU and can sometimes be problematic. However, such issues are not realistic to build into what we are trying to gauge with the survey experiment: the actual policy design features that ex ante define a policy package that potential voters embrace or eschew. Hence, the implicit assumption in the experiment is that packages will be implemented as posed to the respondents.

Table 1 presents the questions for each dimension and the possible answers. The first dimension concerns the question whether there should be budgetary conditions for receiving support. Such conditions are intended to alleviate potential moral hazard. As discussed above, conditionality is a major bone of contention in any discussion about European budgetary assistance packages. For example, when discussing potential emergency support in response to the Covid crisis via the ESM, the issue was raised whether countries that demanded help had a sufficient record in terms of fiscal discipline. An alternative to imposing budgetary conditions would be to require a country to conduct potential growth-enhancing structural reforms in return for assistance, as this might be a more durable way of dealing with the shocks it experiences. We have chosen for the former as they are more tangible and understandable to respondents. We leave an investigation of potential conditions on structural reforms to future research.

Table 1.

Conjoint experiment – questions for each dimension and the possible answers

(1) Are there budgetary policy conditions that countries must fulfil to get support?	No conditions Countries should reduce their public debt in good economic times; otherwise, they will not receive support in bad times.
(2) Are there restrictions on the spending areas on which the budgetary support may be used?	No restriction. Participating countries may use budget support to spend on any policy or purpose. Yes. Budget support must be used for spending on education. Yes. Budget support must be used for spending on unemployment benefits. Yes. Budget support must be used for spending on investment in transport and infrastructure. Yes. Budget support must be used to protect the banking system and depositors. Yes. Budget support must be used for spending on healthcare.
(3) What is the role of the European Commission in the management of the programme?	No role: monitoring is in the hands of national authorities The European Commission monitors the national implementation of the programme The European Commission recommends specific actions to national governments to address their economic problems, and it monitors the implementation of the programme.
(4) May some countries receive more support from the programme than they pay into it?	No, over the long-run countries cannot receive more support from the programme than they pay into the programme. Yes, over the long-run countries can receive more support from the programme than they pay into the programme Yes, over the long run, poor countries will receive more support from the programme than they pay into it, while rich countries will receive less support from the programme than they pay into it.
(5) What is the long-term impact on the taxes that people in your country have to pay?	No impact over the long-run: the level of taxes stays the same in your country Over the long run, taxes increase by 0.5% of income for everyone in your country Over the long run, taxes increase by 1% of income only for the rich in your country
(6) Are there any extra penalties for governments that violate the conditions of the European budgetary support programme?	No automatic termination of budgetary support, but reasons for non-compliance will be examined Budgetary support shall be terminated and countries pay an additional fine.

The second dimension addresses the question whether there should be a restriction on how the support is spent. The baseline is no such restriction, while the alternatives capture important spending areas. Mandatory spending on education captures the notion that this would strengthen a country’s long-run growth potential, enabling it to mitigate the consequences of a permanent adverse economic shock. This is also the case for spending on transport and infrastructure, an important component of public investment, which features prominently in the recent Commission’s ‘Next Generation EU’ plans. Respondents may also realize that governments under budgetary pressure find it politically easiest to cut public investment (European Fiscal Board, 2019) and it therefore needs to be protected. Or, they may be of the view that, since the EU already contributes substantially to infrastructure spending, there is little need for further infrastructure spending. Hence, the overall balance of respondents’ support for this option is a priori unclear. Spending on unemployment benefits intends to do justice to the various proposals for an European unemployment (re)insurance capacity. A priori one would expect it to play a larger role in dealing with the consequences of a temporary than of a permanent economic decline. Potential spending on the banking system and depositors is included in view of the fact that the banking union is still incomplete and that the stagnation on this front is largely attributable to fears about the bill associated with legacy costs of weak banks and with a European deposit insurance scheme that would be more likely to be tapped to support depositors from countries with troubled banks.¹² Hence, it is conceivable that this spending option commands systematically different support from different EU countries. We include healthcare as a spending option, because this is an increasingly important spending area, partially as a result of population ageing and also because it plays a central role during the Covid crisis. In fact, recently, it was agreed that the ESM will make resources available in the form of collectively guaranteed loans for health expenditures related to the corona crisis.

The third and sixth dimensions concern the role of the European Commission and the possibility to punish non-compliance with the programme’s conditions. In practice, one of the Commission’s tasks is to monitor whether spending through EU programmes is done in an appropriate way. Hence, the third dimension addresses preferences concerning a desired or acceptable degree of intrusiveness by the Commission, while the sixth dimension addresses how non-compliance should be handled.¹³

Dimension 4 turns to the key policy design issue of whether on average over time countries may receive more (or less) from the programme than they contribute. The importance of this dimension is obvious, because of the widespread fear of the governments of the economically and financially more healthy countries that they will have to structurally support other countries, reminiscent of the systematic resource flows often observed among regions within a country. These issues are not directly stated as such in the policy design, but are instead artefacts of the key policy design issue regulating between-country redistribution: namely, whether (particular) countries can receive more support from the programme than they pay into it. Hence, this dimension touches upon the distinction between pure insurance via risk-sharing versus redistribution. The distinction is not straightforward.¹⁴ Conceptually speaking, the second alternative, which states that potentially each country could benefit more than it contributes does not a priori entail ex-ante redistribution: resource flows prompted by the programme could by coincidence go more frequently towards a country than away from a country. When in expected terms, at the moment the programme is introduced, no country loses resources, there is no ex-ante redistribution. However, ex-ante redistribution is also not a priori excluded under the alternative. For example, some countries may be more frequently hurt by negative shocks than other countries. Importantly, even if the purpose is to design a scheme that is purely intended for risk-sharing of the consequences of shocks, stakeholders may still fear that it will be hard to avoid any ex-ante redistribution. Under the third alternative, it is ex-ante clear that poor countries will benefit more than rich countries.¹⁵ It should be noticed that the EU already features a number of redistributive programmes, such as its Structural and Cohesion funds, which make this alternative a potentially realistic one.

Finally, dimension 5 deals with the longer-run financing of the assistance programme, which may require a permanent rise in taxes. Taxes may go up in the long run if they are needed to service new debt issuance associated with the support programme. Moreover, if there is a structural redistribution between countries, this may have an additional impact on the tax level in the ‘net contributor’ countries. Indeed, an option frequently proposed to alleviate the immediate financial consequences of the Covid-19 crisis would be to issue very long-run debt, of which the repayment is spread over a number of generations. By stating the potential consequences in terms of the taxation of income, we try to be as concrete as possible about the potential cost of an assistance programme, without overloading the respondent. The key with our wording (for all dimensions but particularly for specific money or tax questions) is that the options be clearly distinct from one another (which is more important than the realism of detailed options offered). As is customary in this type of survey experiment, the baseline alternative of this dimension is to have no effect on long-run taxes. This allows us to investigate how support for the programme changes when respondents are confronted with the fact that the assistance programme comes with an individual cost. We consider the case of a long-run tax increase by 0.5% of income for each respondent, as well as an increase by 1% of income for the rich. Neither of the frames states the precise magnitude of the assistance package. Hence, we can implicitly assume that the tax revenues will suffice to finance the package. We introduce these two alternatives to see how support for a package changes if its cost is concentrated among the rich rather than the general tax-payer.

Each respondent is confronted with three pairs of randomly drawn policy packages. A policy package is a combination of six answers, one for each of the dimensions. Appendix B provides an example of a screenshot seen by respondents. For each pair, the respondent needs to identify the preferred package and indicate how much (s)he likes or dislikes each of the two packages, before moving to the next pair. Hence, for each package in the pair, we obtain binary choice information: 0 = judged as worse than the alternative, while 1 = judged as better than the alternative. We refer to this variable as ‘Choose’. In addition, to each of the packages the respondent sees (s)he assigns an absolute-level rating on a 5-point Likert scale ranging over ‘strongly in favour’, ‘somewhat in favour’, ‘neutral’, ‘somewhat against’ and ‘strongly against’. We refer to this variable as ‘Support’. Either way, we have package-level information and for each package, we know whether a package was chosen or not, its rating, and its composition in terms of dimensions, that is, the treatment.

One might wonder whether it is desirable to ask respondents beforehand about their general predisposition towards an assistance package before presenting them with the concrete packages. We have deliberately not done so, because this could undermine the analytical leverage of the experiment. In particular, it could induce respondents to assess packages on the basis of their general prior activated by this question, implying that the effects attributable to the dimensions would be reduced, potentially defeating the purpose of the experiment, which is about revealing the effects of varying the contents of the packages, rather than understanding support based on vaguely defined priors. For the same reason, we do not inform or cue respondents about whether the packages can achieve better outcomes than can purely national arrangements.¹⁶ Generally, we try to minimize the influence of factors we cannot experimentally control (e.g., providing information on the current state of economy) on the experiment. For the same reason, we also do not expose respondents to the idea of potential permanent transfers from one group to another group of countries, say from Northern to Southern European countries. We want respondents to judge packages on the information given via the experiment, without biasing their views about the packages by giving them additional non-experimental information. The effect of permanent transfers on support is experimentally established indirectly via dimension 5 (on whether and which countries can take-out more than they pay into the programme).

One might also ask whether one should provide explicit information on the status quo and allow respondents to also have the option of rejecting a policy package in favour of the status quo. However, also this would undermine the logic of the experiment. The experimental treatments are the dimensions of the alternative policy packages. Including the status quo as a separate option implies that no longer we can causally interpret the effects attributable to different treatments as the sole drivers of the choice between packages, so the interpretation of estimates of these effects would become meaningless. Indeed, to the best of our knowledge, there exists no conjoint experiment using the status quo as a choice option. Further, we do not see how to control knowledge of the status quo at the individual level, except for including a detailed description of what the status quo looks like for each of the dimensions. That said, notice that each of the frames refers to a new European programme, thus indicating a deviation from the status quo. Further, by allowing respondents to rate each package on the Likert scale, we allow them to compare the desirability of adopting the package with not having the package, that is, the status quo.

To each respondent we apply an attention check, which is failed by about 15% of the respondents. The attention check presents a question with potential answers, but asks the respondent to tick one specific answer. The question is asked along with a large number of individual-specific questions, ranging from socio-economic status, political preferences, concerns about future developments and about Covid to Europe-mindedness after the respondents judged and chose among the policy packages. The attention check is a powerful way to filter-out individuals who do not read the questions or give answers carefully. While much of the analysis carried out in the empirical part of this paper relies only on the subsample of individuals who pass the attention check, as we show later, those who failed the attention check do not differ in any meaningful way in the pattern of their preferences from those who passed the attention check.

4. EMPIRICAL ANALYSIS AND INTERPRETATION

4.1. Descriptive analysis

Before delving into the econometric analysis, we provide a descriptive overview of the main outcomes. Figure 1 shows the distribution of support/rejection scores pooled and by country, while Figure 2 shows the fraction of packages seen by the respondents that are supported by them, pooled and by country. Both figures are created from the Support variable, whereby the respondents could rate each package on the 5-point scale ranging from ‘strongly in favour’ to ‘strongly against’.

Figure 1.

Distribution of support pooled and by country

Note: 1, ‘strongly against’; 2, ‘somewhat against’; 3, ‘neutral’; 4, ‘somewhat in favour’ and 5, ‘strongly in favour’.

Figure 2.

Fraction of packages supported, pooled and by country

Figure 1 demonstrates that the fraction of packages judged as ‘strongly in favour’ exceeds the fraction judged as ‘strongly against’, while the fraction ‘somewhat in favour’ exceeds the fraction ‘somewhat against’. This pattern is seen for the pooled sample, as well as each individual country, even for countries that have a reputation for being skeptical about EU level budgetary assistance.

Of course, many individuals hold a neutral position on one or more packages they see.¹⁷ To better grasp the actual levels of support, it is therefore useful to differentiate between two different levels of support. In Figure 2, the dark bars indicate the share of packages, pooled and per country, that are supported, that is, receiving the verdict ‘strongly in favour’ or ‘somewhat in favour’, when neutrals are counted as not supporting the package. The light bars indicate instead the share of supported packages if neutral judgements are excluded. Note that these are both extreme views on support: those who have neutral views on certain packages end up being completely excluded or counted as against. Nonetheless, support is generally quite large: excluding neutrals, even in the most sceptic country, the Netherlands, almost 60% of the packages are supported, while in the country where support is highest, Spain, 70% of the packages are supported.

These findings are consistent with the fact that a sizeable fraction of individuals have a very positive view of an EU support programme: almost 15% reject none of the packages (not graphically shown), while about 20% of the respondents have negative views on only one or two packages of the six they were shown. Conversely, only about 11% of the respondents reject five or all six packages they have seen. Hence, also fundamental opposition to a programme is limited. These findings are consistent with previous studies (see Vandenbroucke et al., 2018) that identify a similarly low level of fundamental opposition to the construction of an EU-wide unemployment re-insurance scheme.

It is important to emphasize that the substantial support we see so far is the outcome of randomizations over all possible treatments over the different dimensions, hence many of the packages seen by respondents may contain one or more less desirable elements. In particular, at this stage, we have not yet selected specific packages that can count on broader support than other packages. The substantial support for EU budgetary assistance programmes in general provides hope that it is possible to design programmes that can count on sufficient support in each of the sample countries, hence that a European deal can be struck that is acceptable to the populations of all countries potentially participating.

4.2. Econometric analysis of complete sample

Our model is a simple regression model, Equation (1), where the unit of observation is the package, the dependent variable is either whether the package is chosen or supported, and the independent variables are fixed effects, the components (i.e., dimensions) of the treatment, a set of individual-level control variables and interaction terms of the components of the treatment and the individual-level controls:

$${\mathrm{OUTCOME}}_{i,j,k,f}={\alpha }_j+{\phi }_f+{\sum }_f{\Delta }_{\mathit{f}}\left({\mathit{\beta }}_{1\mathit{f}}\mathrm{ }{\mathbf{BUDGETCOND}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}+{\mathit{\beta }}_{2\mathit{f}}^{\mathrm{\prime }}\mathrm{ }{\mathbf{POLAREA}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}+{\mathit{\beta }}_{3\mathit{f}}^{\mathrm{\prime }}\mathrm{ }{\mathbf{COMROLE}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}+{\mathit{\beta }}_{4\mathit{f}}^{\mathrm{\prime }}\mathrm{ }{\mathbf{REDISTR}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}+{\mathit{\beta }}_{5\mathit{f}}^{\mathrm{\prime }}\mathrm{ }{\mathbf{TAX}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}+{\mathit{\beta }}_{6\mathit{f}}\mathrm{ }{\mathbf{FINE}}_{\mathit{i},\mathit{j},\mathrm{ }\mathit{k},\mathit{f}}\right)+{\gamma }^{\mathrm{\prime }}{\mathrm{CONTROLS}}_{i,j,k,f}+{\delta }^{\mathrm{\prime }}{\mathrm{DIM}}_{i,j,k,f}⊛{\mathrm{CONTROLS}}_{i,j,k,f}+{\epsilon }_{i,j,\mathrm{ }k,f},$$

(1)

where ${\alpha }_j$ is a country-specific constant (⁠${\alpha }_1$ is set to zero), ${\phi }_f$ is a frame-specific constant (⁠$f=1$ refers to a temporary shock, $f=2$ to a permanent shock; ${\phi }_1$ is set to zero), ${\Delta }_f$ is a dummy which equals 1 if the frame is $f$ and zero otherwise, ${\Sigma }_f$ is the summation operator over $f$⁠, while $\left(i,j,k,f\right)$ indexes the kth package (⁠$k=1,.,6$⁠) presented to individual i of country j under frame $f$⁠. The regression in addition controls for the ‘conjoint pair’, the round (first, second or third) in which the policy package is shown to the respondent. Further, ${\mathrm{OUTCOME}}_{i,j,k,f}$ is the outcome of the judgement of the package, which can be either ${\mathrm{CHOOSE}}_{i,j,k,f}$⁠, a binary variable that indicates whether (from the presented pair) the package is chosen (⁠${\mathrm{CHOOSE}}_{i,j,k,f}=1$⁠) or not (⁠${\mathrm{CHOOSE}}_{i,j,k,f}=0$⁠), or ${\mathrm{SUPPORT}}_{i,j,k,f}$⁠, which takes on a value of 1 if the package is supported, that is, if it is rated ‘strongly in favour’ or ‘somewhat in favour’, and 0, otherwise, that is, if it is rated ‘neutral’, ‘somewhat against’ or ‘strongly against’. Hence, we are taking a conservative approach as we count ‘neutrals’ as against. Variable ${\mathrm{SUPPORT}}_{i,j,k,f}$ takes account of an overall satisfaction or dissatisfaction with the EU support packages presented. Further, ${\mathrm{BUDGETCOND}}_{i,j, k,f}$ is a dummy variable taking a value of 1 if budgetary conditions are present,${\mathrm{POLAREA}}_{i,j, k,f}$ is a vector of five dummy variables taking a value of 1 when spending is mandatory on the area indicated (the baseline being no condition on how support should be spent), ${\mathrm{COMROLE}}_{i,j, k,f}$ is a vector of two dummy variables taking a value of 1 when the Commission monitors, respectively, when it monitors and makes recommendations, ${\mathrm{REDISTR}}_{i,j, k,f}$ is a vector of two dummy variables taking a value of 1 when long-term redistribution to any participating country is allowed, respectively, when redistribution from richer to poorer countries is explicitly mandated, ${\mathrm{TAX}}_{i,j, k,f}$ is a vector of two dummies taking a value of 1 when taxes go up by 0.5% for everyone, respectively, when they go up by 1% for the rich only and ${\mathrm{FINE}}_{i,j, k,f}$ is a dummy which is 1 when non-compliance is punished with termination of the programme and a fine. Hence, for each possible answer to a question (dimension) in Table 1, except for the first answer, which is the ‘baseline’, there is a dummy variable. The dummy measures the effect on the outcome relative to this baseline when another option for this dimension is chosen. Further, ${\mathrm{CONTROLS}}_{i,j,k,f}$ is a set of individual-level controls and ${\mathrm{DIM}}_{i,j,k,f}$ stacks in one column vector the sets of dummies corresponding to the six dimensions in Table 1. We denote by $⊛$ the operator that takes the product of each element of ${\mathrm{CONTROLS}}_{i,j,k,f}$ with each element of ${\mathrm{DIM}}_{i,j,k,f}$ and stacks the resulting products into a column vector. Finally, ${\epsilon }_{i,j, k,f}$ is an error term. Further, ${\beta }_1,.,{\beta }_6, \gamma ,\delta$ and $\theta$ are scalars or column vectors of appropriate dimensions. Throughout the main text, we estimate Equation (1) with OLS and standard errors clustered at the individual level.

Equation (1) is the general formulation of the regressions that we conduct. In the next subsection, we start by studying its purely experimental version, while imposing that the dimensions have identical effects across the frames, that is, ${\beta }_{11}={\beta }_{12}$⁠, et cetera, and excluding the interactions between the dimensions and the individual controls, that is, we set $\delta =0$⁠. The coefficients of our six experimental treatments (in bold in the equation) can be interpreted as having a causal effect on support thanks to their random assignment.

Deploying a conjoint experiment has important advantages (Hainmueller et al., 2014) when compared with regular survey experiments. First, and foremost, the random assignment of policy packages to respondents allows for robust causal inference of the effect, in this case through regression Equation (1), of variations in the treatments along the different policy dimensions on preferences. The treatment effects that we estimate are the average of the responses across the different subgroups in society. Second, the model in Equation (1) allows to estimate the role of interaction effects, that is, what is the effect of a change along a specific dimension under alternative settings for another dimension.¹⁸ Third, a conjoint experiment reduces the risk that respondents simply provide socially desirable answers rather than expressing their true opinion. The reason is that potentially contentious elements are bundled in a larger policy package.

4.2.1. Aggregate baseline results

As discussed above, we present here the results from our baseline estimations; these are simple regression models where the dependent variable is either the binary choice variable or the binary measure of support; the independent variables are a constant, a dummy for each country (except France), a dummy for a permanent frame, dummies for the levels of the six dimensions and a set of controls (education, gender, age, income, conjoint pair and Covid-19 concerns). Since respondents score six packages each, we use panel-robust standard errors clustered at the level of the individual respondent. We restrict the sample to those respondents who successfully pass the attention check at the end of the survey, but a robustness-check on the full sample suggests that no differences exist (see below).

The most efficient way of showing the effects of the policy dimensions on the degree of support is by means of plots of the ‘average marginal component effect’ (AMCE). The AMCE measures the average causal effect of changing the treatment for a given dimension away from its baseline on the likelihood that a package will be supported or chosen, holding the treatments for all other dimensions the same. In Figure 3, we limit the graphic representation to the purely experimental elements of the analysis, that is, the treatments; information on the (mostly negligible) effects of the controls is found in Table C.1 in Online Appendix C, which reports the econometric estimations underlying Figure 3. Figure 3 depicts the AMCEs for the full sample of 10,000 individuals (i.e., 60,000 observations) across the five countries; the country-specific results are shown later.

Figure 3.

EU assistance programme – AMCE plot full sample of respondents

Note: Horizontal line pieces depict the 95% confidence intervals.

Overall, the results are the same, regardless of whether we look at choice or support for packages. As shown in Figure 3, all else equal, packages featuring budgetary conditions are about 7 percentage points more likely to be supported and about 10 percentage points more likely of being chosen out of a pair, compared with packages that feature no budgetary conditions.

Turning to the second dimension, the baseline is to have no conditions on how the budgetary support is to be spent. The absence of a condition on spending has significantly more support only when compared with mandatory spending on protecting the banking system and depositors. This outcome may not be too surprising in view of the fact that the banking system is widely blamed for being (at least partly) responsible for the global financial crisis and the fact that some banks had to be saved with tax-payers’ money.

Perhaps unsurprisingly, considering that the survey was fielded at the end of March 2020, the most preferred alternative to the baseline is a requirement to spend the budgetary support on health care. A package with health care is about 11% points more likely to be supported and about 13% points more likely to be chosen than the same counterpart with no spending condition. On average this is a substantial effect for a competence reserved for the member states. An obvious explanation would be that the survey is taken during the Covid crisis, although, as we show in Section 4.4, a pilot study fielded in the Netherlands in late October 2019 (well before the Covid-19 outbreak) shows consistent results.¹⁹ The next-preferred alternative is a requirement to spend the support on education. This alternative has about 3–4 percentage points more likely support than the alternative of no condition on spending. Including a requirement to spend the support on unemployment benefits only has a small and insignificant positive effect on support, in spite of the fact that unemployment spending would directly contribute to stabilizing the economy. Spending on infrastructure and transport commands more support than the baseline. The difference is significant but limited in magnitude. Still, it may suggest a preference for extending the role of the EU in this area.

Turning to the third dimension, we see that there is significantly more support for giving the Commission an explicit role, either in terms of monitoring or monitoring and making recommendations, than to give it no role at all. This is consistent with the idea that a degree of joint oversight is preferred, even more so when such oversight is coupled with instruments to coordinate and steer domestic action.

The fourth dimension tackles one of the politically most controversial aspects of the EU budgetary support debate, that is, whether the programme is designed in a way as to yield long-term redistribution between countries. This dimension tackles the issue by focusing on whether the programme allows (no, all or only poor) countries to draw-out more than they pay in the budgetary assistance programme. Such design has been debated and also resonates with similar recent debate on whether the EU-level recovery instrument in response to the Covid crisis should provide grants, which would be redistributive, or loans at potentially concessionary interest rates. However, policy design affects it, redistribution is a highly divisive issue, which has led to fierce clashes among Eurozone governments both at the height of the Eurozone debt crisis and recently during the negotiations about measures countering the negative economic effects of the Covid crisis. Interestingly, the aggregate results show that packages that (potentially) lead to long-run shifts in resources are between 3% and 5% points more likely to be supported than packages that do not have this feature. However, since our sample is built to include countries with very different perspectives, at least officially, on cross-border redistribution, this dimension requires further scrutiny of disaggregated country results, which we provide later.

As for the fifth dimension, we observe that a long-run increase in the tax burden by half a percentage point is strongly disliked compared with the ‘free’ option of no increase in taxes or to a progressive increase in the tax burden, that is, by imposing a 1% tax increase on the rich. Finally, the sixth dimension – pertaining to the consequences for abuse of the programme – shows that the support for termination of the programme combined with a fine for countries not complying is marginally smaller (about 1–2 percentage points less likely) than the baseline of an investigation into the reasons for non-compliance, but no automatic termination. While the aggregate effects are – once again – quite close to zero, individual countries display differences we discuss later.

4.2.2. Temporary and permanent shock framings

As discussed in Section 1, the academic debate on fiscal unification is in particular concerned with the temporal versus permanent nature of economic shocks. Many of those who are skeptical about introducing EU budgetary assistance programmes fear that these lead to structural redistribution, in particular when these programmes are aimed at combatting permanent shocks, because moral hazard discourages implementing politically costly structural reforms that would alleviate the economic decline. In addition, while temporary shocks can be addressed by discretionary fiscal policy measures that stimulate aggregate demand (complementing the effect of automatic stabilizers), such measures are less suited to handle permanent shocks, which require instead structural policies that strengthen potential growth. Respondents may be aware of these comparative advantages of spending areas in combatting the different types of shocks. Hence, we would a priori expect respondents who are provided with the permanent shock frame to express relatively more support for mandatory spending on education or transport and infrastructure, while we would expect those who receive the temporary shock frame to be relatively more supportive of mandatory spending on unemployment.

To investigate the relevance of these considerations, we interact in regression (1) the frame with the different treatments along the dimensions of the experiment. In other words, we allow for parameter vector $\left({\phi }_f,{\beta }_{1f},{\beta }_{2f}^{\prime }, {\beta }_{3f}^{\prime },{\beta }_{4f}^{\prime },{\beta }_{5f}^{\prime },{\beta }_{6f}\right)$ to differ between the two frames. The result is depicted in Figure 4, which shows the AMCEs on the support variable ${\mathrm{SUPPORT}}_{i,j,k,f}$ for the two frames. We observe that the specific frame respondents are confronted with has in most cases only a limited effect. There is little difference in the support for spending on unemployment between the two frames, despite the fact that unemployment spending would typically alleviate a temporary, but not a permanent decline.

Figure 4.

Comparison of AMCEs under temporary and permanent shock frames

Note: Horizontal line pieces depict the 95% confidence intervals.

There is also no statistically significant difference in support for spending on transport and infrastructure. A requirement to spend the support on education does command somewhat stronger support among those confronted with a permanent rather than a temporary shock, which suggests that respondents at least to some extent realize that permanent economic declines may be better addressed with structural policies. However, the difference in support is not significantly different from zero. In addition, there are some differences between countries, discussed in the next section.

Why are the differences in support for the same treatment between the two frames rather limited? The answer to this question requires some speculation. One possibility is that the differences in framing are made insufficiently explicit. However, this is unlikely, because the formulation of the two frames (Appendix A) repeatedly emphasizes the nature of the decline (temporary or permanent). It could also be that the framing is sufficiently explicit, but that respondents do not grasp the economic implications of this distinction in the way experts understand them – after all, each respondent gets to see only one of the two frames and we try to keep the description as minimalist as possible, to avoid bias and information overload. Further, it is possible that respondents do understand the economic differences between the two frames, but that the limited differences in support reflect their true preferences. Even if a permanent decline entails systematic cross-border redistribution, they could still be as supportive of assistance as in the case of a temporary shock.²⁰ Finally, respondents’ attitudes could be dominated by questions of allocation of public resources rather than by questions of economic stabilization, redistribution or strengthening the economic structure. Consistent with allocation as the driving force behind the limited differences between the AMCEs under the two frames is that support for mandatory spending on healthcare is the highest under both frames.

4.3. Country-level econometric analysis

The results discussed pertain to the full sample, in which the respondents from the different countries are pooled. While a larger sample improves the precision of the estimates, it may also hide important country-specific variations in support for given packages. Exploring these variations is important, as much of EU decision-making is intergovernmental or requires even unanimous support. Packages with substantial aggregate support evenly spread over the countries stand a much higher chance of being implemented than packages with identical aggregate support but substantial variation in support over the countries.

4.3.1. Country-level effects of treatments

We start by looking, once again, at the purely experimental component of the analysis in Figure 5, which reproduces Figure 3 for each individual sample country. Obviously, the confidence intervals around the estimates become wider, as the number of observations for each country is smaller than the number of observations for the aggregate analysis.

Figure 5.

(a) Effects on support by country – dimensions 1–3 and (b) effects on support by country – dimensions 4–6

Note: Horizontal line pieces depict the 95% confidence intervals.

When it comes to the first dimension, except for Italy, all countries give significantly more support to a programme that imposes budgetary conditions than one without conditions. This preference is strongest for Germany and the Netherlands, followed by France and then Spain. The ‘moral hazard argument’ suggests that respondents may view a budgetary condition for support as an instrument to encourage a more prudent fiscal policy, which in the end would reduce the likelihood that support from other countries is needed in a decline and, hence, that they would face a higher tax bill to pay for the support. Seen from the perspective of a respondent from a high-debt country, such a respondent would probably assess the likelihood of receiving support in the event of an economic decline as lower when there are conditions attached to such support than when there are no conditions attached. The observed relative support pattern across the sample countries is in line with this reasoning, because Germany and the Netherlands feature the lowest levels of public indebtedness, with populations that perceive themselves as more likely to be on the paying than on the receiving end, and Italy features the highest public indebtedness.

Regarding the use of the budgetary assistance, the support for healthcare spending is always highest, for most countries with a margin of 11–12 percentage points over the baseline, followed by education spending in all countries, except for Italy and Spain. For the latter country, the support for education spending is still significantly higher than the baseline of no restriction on spending. With the exception of Spain, which is plagued by high unemployment among the young in particular, no country features significantly higher support for spending on unemployment benefits than for no condition on spending. Transport and infrastructure spending receive more support than the baseline of no earmarking in France, Germany and the Netherlands. For this latter country, transport is a key economic activity; hence, this outcome is not surprising. Spending on protection of the banking sector and deposit holders is highly unpopular, except for France and Germany where respondents do not exhibit a significant difference in support compared with no condition on spending.

Next, all countries support a role for the Commission, in particular when this role comprises both monitoring and recommending specific actions. The strength of the support differs across the countries and is highest in Germany and the Netherlands. A potential explanation is that respondents from these countries expect to make transfers to other countries and want these resources to be well spent, which they do not trust to be the case without monitoring and guidance for the national authorities of countries receiving budgetary support. Next, allowing long-run redistribution across countries or mandating such redistribution towards the poorer countries can count on substantial support in Italy and Spain, and limited support in France and Germany (but in the latter two countries only when it comes to long-run redistribution to poorer countries). No support for long-run redistribution of either kind is found in the Netherlands.

Regarding the next dimension, taxation, we observe that the respondents in all the countries are strongly against a flat tax increase compared with the baseline of no change in taxes. How can this be compatible with the generally high support for a budgetary support programme, as it seems unlikely that respondents do not perceive some link between such a programme and the need to finance it? First, even if respondents dislike a long-run increase in taxes, they may not be against a temporary increase in taxes to finance the support programme. Second, and more plausible, respondents may be in favour of a support programme, but they are simply not prepared to pay for it themselves and prefer to shift the burden to individuals from other countries or individuals higher up in the income distribution of their own country. Progressive taxation, whereby the rich are taxed to finance the policy, is substantially less disliked than the alternative of a flat tax increase. The exception is Italy, where this variant is significantly less popular than the alternative of not raising taxes at all, an outcome that may be the result of a decennia-long campaign by Berlusconi demonizing the idea of taxation on the rich.

On the final dimension, the Netherlands is the only country that supports significantly more than the baseline the termination of programme assistance and that supports imposing a fine in the case of non-compliance. The Italian population is significantly less supportive of this alternative than the baseline and the Spanish population is close to being significantly less supportive. These patterns may not be surprising if the Dutch population expects the Netherlands to be mostly a net contributor rather than a net recipient, and the Italian and Spanish populations expect their countries to be net recipients.

Differences between countries on the six dimensions of policy design (summarized in Figure 5), and for that matter on average support for all shown packages of the EU budgetary-assistance facility (NL<DE<FR<IT<ES in Figure 2), have many possible origins. The role of macro-level political economic experiences like debt levels or net-contributor or net-debtor in EU membership has already been noted. But there are, of course, plenty of other possible factors – from social policy experience to socio-cultural patterns. For instance, various aspects of social trust can play a role – not just trust in the EU or government but also generalized trust. Social trust has been found to matter a lot to support for welfare states and related fiscal interventions, either simple positive links where trust breeds support for such interventions (Rothstein and Uslaner, 2005) or more complicated links where trust has implications moderated by civic attitudes (Algan et al., 2016). These and other possibilities deserve fuller investigation than our modest cross-country sample can support.²¹ What is most important for the current study, however, is that the cross-country differences nuance rather than negate the pooled patterns of support for EU budgetary assistance.

4.3.2. Differences in framing effects at the country level

In Section 4.2.2 on the pooled estimates, we already discussed how temporary versus permanent shocks impacted our respondents’ preferences, concluding that a significant difference was only found for support for mandatory spending on education. When looking at the disaggregated country level, these differences in the effect of the frame remain generally small, with a few exceptions (Figure 6).

Figure 6.

Effects on support, pooled and by country, temporary versus permanent economic decline

Note: Horizontal line pieces depict the 95% confidence intervals. Dx indicates dimension x in Table 1.

Italians are significantly (at 10% level) more likely to support budgetary conditions when a country is facing a permanent rather than a temporary decline. When switching from a temporary to a permanent shock, the desired role of the Commission providing monitoring and recommendations weakens significantly (at the 10% confidence level) for German respondents. For Spanish respondents, it is the opposite. Support for the possibility that each country can benefit structurally more or less than other countries or for structural redistribution from rich to poor countries seems to command more support from Italian respondents following a permanent than a temporary shock. A potential explanation could be their familiarity with structural economic problems and the expectation that they would likely be net receivers. Finally, the French are significantly more likely to support termination of the programme coupled with a fine for non-compliance in the case of a permanent shock. Overall, while we observe some variation in support levels between the two frames across countries, this variation is rather limited.

4.4. Did the covid-19 outbreak affect public opinion on budgetary support programmes?

As discussed above, this survey experiment took place at a very peculiar moment in contemporary history: the end of March 2020 was the moment when the first wave of the Covid outbreak was peaking, or about to peak, in most western-European countries. It is therefore legitimate to explore whether this historical development weighed on the minds and the opinions of the respondents. For this reason, the models we estimate include, among the controls, the respondent’s personal concern about the Covid-19 outbreak. However, since some of our experimental dimensions include treatment options that may relate directly to the pandemic, such as mandatory spending on healthcare, it is worth asking to what extent the results so far could have been influenced by the Covid outbreak. This survey, which was developed in the second half of 2019, was not specifically designed to measure support for policies in response to the pandemic. Hence, it cannot answer this question in detail, while furthermore we lack fully fledged data to properly assess public opinion dynamics before and during the pandemic.

However, we still have some scope to explore the general validity of our experiment, because a pilot version of this study had been run in late October 2019, a few months before the existence and possible spread of the corona virus started to appear in the news. This pilot, which was fielded on a representative sample of 400 Dutch respondents, features no significant differences with the survey fielded in March 2020. Hence, we are able to compare the pre- and post-pandemic results of our survey experiment at least for Dutch respondents. De facto, we built an additional ‘natural experiment’ on top of our survey experiment.

First, we look at overall levels of support (Figure 7). The results are remarkably stable between the two periods. Average support for the packages presented very marginal increases from October 2019 to March 2020, but well within the margin of error. The number of neutral judgements also remains largely the same across the two periods.

Figure 7.

Pre- versus post-Covid outbreak – Dutch respondents

Notes: ‘In favour’ aggregates the cases of ‘strongly in favour’ (5) and ‘somewhat in favour’ (4); ‘against’ aggregates the cases of ‘strongly against’ (1) and ‘somewhat against’ (2) when neutrals (3) are excluded, while ‘against’ adds to these also the neutrals when the latter are included.

Next, we look at the specific effects of the dimensions, where changes should in principle be more visible (Figure 8). The econometric specification is again the baseline specification. Figure 8 shows that already before the Covid crisis the Dutch subsample exhibited a strong support for mandatory spending of European assistance on healthcare. While this preference inches forward during the post-Covid-19 outbreak, perhaps as a result of the estimates becoming more precise thanks to the larger sample size, the new results are well within the margin of error of the pre-Covid estimate. However, what is noticeable for this dimension is the concentration of the Dutch support for mandatory healthcare intervention after the outbreak. While before the outbreak the AMCE of transport and infrastructure was close to that of healthcare and that of education even exceeded that of health care, post-Covid the AMCEs of transport and infrastructure and education shrink and that of healthcare spending rises.

Figure 8.

AMCE plot for the Dutch pre- versus post-Covid outbreak, support as outcome variable

Note: Horizontal line pieces depict the 95% confidence intervals.

The Dutch respondents exhibit a difference in their support for European Commission oversight, which becomes stronger in the post-pandemic period, potentially reflecting that, since a pandemic-related EU assistance programme had by the end of March 2020 become an eventuality, respondents felt a stronger need for a role of the Commission as a guardian of the proper use of EU assistance.

Finally, the interaction effects analysis also shows a small improvement in the Dutch respondents’ attitude to support a programme with potential long-term redistributive benefits to poor countries: while before the pandemic the AMCE associated with this treatment was negative and significantly different (at the 10% level) from zero, during the pandemics the AMCE became insignificantly different from zero.

All, in all, the comparison of the Dutch subsamples before and during the pandemic suggests that the outbreak has had only a limited influence on their attitudes towards EU-level budgetary assistance. Of course, since the pre- and post-Covid comparison is limited to one country only, we should be cautious about generalizing the relative stability of the results to other countries. However, even before the Covid-crisis, there was substantial support for European assistance packages from one of the allegedly most sceptical countries, suggesting that such support may have been comparably forthcoming pre-crisis in other sample countries. Also, the strong preference for healthcare spending pre- and post-Covid for the Netherlands, combined with the strong post-Covid preference generally found in the sample, suggests that such strong preference for spending assistance on healthcare was generally present before the Covid crisis.

Finally, since we know the dates on which respondents completed the survey, spanning the period 24 March–7 April 2020. Hence, we ran regressions controlling for the completion date. The estimates of the date-fixed effect are generally not significantly different from zero. The two dates that exhibit some variation relative to the base date 24 March are 25 March and 1 April. Notably, before 1 April almost 92% of the respondents had already completed the survey. Leaving out these dates from the sample yields AMCEs close to those reported in Figure 3. The estimates are shown in Online Appendix G. While the time span for filling out the survey is limited, these findings provide modest corroboration of the time-wise internal and external validity of our results.

4.5. Further robustness checks

This subsection discusses a number of further robustness checks. The underlying econometric estimates are found in Table C.1 in Online Appendix C. They are based on direct variations on the baseline regression which is reported in Column (4) of the table. First, excluding the individual controls has no effect on the results (Online Appendix Table C.1, Column (2)). Second, including inattentives also leaves the results unchanged (Online Appendix Table C.1, Column (3)). Third, we replace the linear model with a logit specification for both outcome variables. The estimates are reported in Column (5) of Online Appendix Table C.1 for ${\mathrm{SUPPORT}}_{i,j,k,f}$ as dependent and in Column (7) for ${\mathrm{CHOOSE}}_{i,j,k,f}$ as dependent. Again the results are unchanged: significance and insignificance are preserved in each case. Fourth, we drop the neutral answers from the sample. The sample size obviously shrinks. However, it also means relaxing the conservative approach in measuring support. Indeed, the sizes of the coefficient estimates almost all increase in absolute magnitude, strengthening the effects found before. Qualitatively the results are unchanged, expect for the finding that progressive taxation (relative to the base-value of no new taxation) now features a negative AMCE significantly different from zero. Next, with proper randomization, control variables such as individual-level characteristics should not affect the estimated AMCEs. The estimates reported in Table F.2 in Online Appendix F conform that this is the case. Finally, we re-estimate the model with fixed effects for all 10,000+ respondents [Online Appendix Table C.1, Column (9) for support and Column (10) for choice], dropping the individual-level control variables. This is the maximum degree of heterogeneity we can control for, focusing hence only on the within-respondent variation across judged packages, and provides the strongest test of the quality of the randomization in our experiment. The coefficient estimates are virtually identical to those under the baseline and, hence, significance is always unaltered.

Table 2.

Support for individual countries’ most-supported packages

	Most-supported package in country in first column	Support in percent in:
Country		Pooled sample	DE	ES	FR	IT	NL
DE	Budgetary conditions, healthcare spending, monitoring and recommending, poor countries redistribution, progressive taxation, no fines	60.3	65.1	66.8	54.8	54.6	60.6
ES
FR	Budgetary conditions, healthcare spending, monitoring and recommending, poor countries redistribution, no taxation, no fines	60.3	63.4	65.6	55.3	58.0	59.5
IT	Budgetary conditions, healthcare spending, monitoring and recommending, all countries redistribution, no taxation, no fines	59.2	61.3	64.9	53.4	58.1	58.3
NL	Budgetary conditions, healthcare spending, monitoring and recommending, poor countries redistribution, progressive taxation, fines	58.9	64.2	64.8	53.9	48.7	63.4

Note: DE, Germany; ES, Spain; FR, France; IT, Italy; NL, Netherlands.

4.6. The role of individual characteristics

Throughout our analysis so far, we have focused on the purely experimental components of the research design: the dimensions of the experiment itself. The reason is that, if the sample is representative and the treatment assignment is random, looking at the dimensions in isolation is the best way to gauge the effects that are attributable to the treatments. The treatment effects can be seen as reflecting the average reactions among possible individual subgroupings. However, even though this goes beyond the space and thematic constraints of the present paper, and does not alter our core findings, in this subsection we do briefly touch upon the role of individual-level characteristics, as such an analysis may provide leads for measures to increase the support for assistance packages. In particular, we extend the baseline specification by adding the interactions of the different dimension dummies with the income level, the education level and Covid-19 worries. The motivation to include the interactions with socio-economic status variables is that these may be important drivers of an individual’s position on the various elements of an EU support package, because socio-economic status may to a large extent determine an individual’s benefits and costs associated with a package (at least in her own perception). The interactions with Covid-19 worries are motivated by the timing of the experiment end of March 2020.

The regression model is again Equation (1). However, we no longer constrain $\delta$ to zero, but we estimate elements of this vector of coefficients along with the other parameters we estimated in the purely experimental version. Concretely, we add to each regression the interaction term of the dimension dummy and one of the respondent-level variables of income, education or Covid-19 worries. Online Appendix Table D.2 presents the estimates.²² The results do not in any way vitiate the patterns reported in the baseline models of the main text, but they do suggest that socio-economic position and Covid-19 worry can moderate how different policy characteristics influence respondent support for EU assistance. With respect to a respondent’s education level, we observe that the direct effect of mandatory spending on education on support loses significance. However, its interaction with the education level of the respondent itself has a significant positive effect on support: the more highly educated the respondent is, the more she likes packages containing mandatory spending on education. The direct effects of mandatory spending on transport and infrastructure and banks and deposits lose significance. Turning to the role of the Commission, this is also the case for the direct effects of a role for the Commission. However, the interaction effects are positive and significantly different from zero: more highly educated people are more in favour of Commission monitoring and even more strongly in favour of the Commission combining monitoring with recommendations. Turning to the role of income, we see that the only change relative to the purely experimental analysis is a significant interaction with a role for the Commission: the higher the respondent’s income, the stronger her support for a package containing monitoring or monitoring and recommendations on the side of the Commission. Finally, we observe that more Covid-19 worries reduce the support for packages with budgetary conditions, mandatory spending on education and banks and deposits, monitoring by the Commission, a flat tax increase to pay for the support instrument and termination and a fine in the case of non-compliance, while they increase the support for packages with redistribution from rich to poor countries.

These patterns suggest the value of continued analysis of subsamples and the moderating role of individual characteristics. We leave analysis and discussion of such issues, however, to later work, and emphasize again that the experimental design is particularly suited to causal inferences about policy design rather than the more observational-basis or sub-sample-basis of moderating effects by individual correlates.

5. CONSTRUCTING POLICY PACKAGES WITH WIDESPREAD SUPPORT

So far, we have been mainly studying the effects on support of variations in individual treatments along the dimensions. However, policy packages consist of a combination of attributes. Which combinations are the most supported and which are the least supported? In this section, we explore the support for various policy packages. Since our study features 648 alternative policy packages, it is not possible to assess all of them in detail. Instead, we select a number of packages that are of specific interest for us. To this end, we estimate counterfactually the expected level of support, should these packages be submitted to the respondents once again. Support for a package is estimated as the sum of the estimated fixed effects in regression Equation (1) with ${\mathrm{SUPPORT}}_{i,j,k,f}$ on the left-hand side, plus the estimates of the coefficients on the dummies of the relevant treatments contained in a package. Unless noted otherwise, we impose that the coefficients for the two frames be identical and that neutrals are counted as being against the package, implying that we maintain our conservative approach in assessing support.

Figure 9 depicts for specific packages support in the overall sample, that is, pooling respondents from all countries. Package (a) is the least supported. It combines all the features that were most disliked by the respondents: no budgetary conditions, support spending earmarked for banks and depositors, no role for the Commission, excluding systematic redistribution among countries, a flat tax increase and termination and a fine in the case of non-compliance. It is expected to be ‘strongly supported’ or ‘somewhat supported’ by less than 25% of the respondents who would be confronted with it. This package contrasts with the most-supported package on the basis of our estimates, Package (b), which combines budgetary conditions, mandatory spending on healthcare, a maximum role for the Commission (monitoring and recommendations), redistribution from rich to poor countries, no change in taxes and no termination and fines in the case of non-compliance. It commands about 60% support. The drawback of this package is that it presents a combination of measures that are infeasible for at least some respondents. Some groups of respondents may (legitimately) expect that for them the package will not come at any costs in terms of a higher personal tax burden. This could, for example, be the case for those with very low or no labour income or living in a country more likely than others to be hit by a severe negative shock. Hence, it is relevant to investigate the support for packages that come without an effect on taxes. This may well mean, however, that for a representative survey sample, at least some nontrivial fraction of respondents would be confronted with the bill of the package. Therefore, Package (c) is the most-supported package that also features a tax increase: it imposes a long-run increase in taxes on the rich. The amount of support essentially equals that on the previous package. The final Package (d) replaces the progressive taxation of Package (c) with a flat rate tax increase. Estimating support for Package (d) allows to verify whether there is still sufficient support among the broad population of respondents if the average population member knows she has to pay for the programme. We observe that this package still has more than 50% support, but the support is less than that of the package with a tax increase only for the rich.

Figure 9.

Level of aggregate support for packages of interest

Notes: Package (a) is least supported: it includes no budgetary conditions, support for banks and depositors, no role for the Commission, no long-run redistribution, flat taxation and termination and fine for non-compliance. Package (b) is most supported: it includes budgetary conditions, mandatory spending on healthcare, Commission monitoring and recommendations, long-run redistribution to poor countries, no taxation and no termination and fine for non-compliance. Package (c) is same as package (b), replacing no with progressive taxation. Package (d) is same as package (b), replacing no with flat taxation.

Next, we explore support for packages at the level of individual countries. Figure 10 puts side-by-side the most-supported packages with taxes of each of the individual countries in the sample and assesses their support by the overall pool of respondents. For both Germany and Spain, the most-supported package with taxes is the first package in Figure 10. The most-supported package of France differs from this package in that the progressive tax increase for the rich is replaced by a preference for no change in taxes. It gathers about 60% overall support. The most-supported package by Italians differs further by allowing for long-run redistribution to go into any possible direction. It gathers slightly less than 60% support from the full pool of respondents. Finally, the most-supported Dutch package is the same as that for Germany and Spain, except that it favours fines for non-compliance. It also obtains slightly less than 60% support from the full sample of respondents.

Figure 10.

Aggregate support for most-supported packages at the country-level

Given that intergovernmental bargaining is key in forging any agreement on an EU budgetary assistance programme, Table 2 lists for the aggregate set of respondents and each individual country’s respondents, including those of the own country, the support of a country’s most-supported package. Start with the most-supported package of both German and Spanish respondents. The package receives more than 50% support in all individual countries, although the support of French and Italian respondents is quite a bit lower, less than 55%, than the support from the respondents of the other three countries, which is more than 60% for each of these three countries. The most-supported package by the French replaces progressive taxation with no taxation. This reduces the support from Dutch, German and Spanish respondents, while it raises support from (by definition) French and Italian respondents. Remarkably, the package most supported by the French still receives less support from the French than from any other country’s respondents, which is a reflection of the generally relatively low level of support of the French respondents to Eurozone support packages.

We see a similar, though less extreme, effect for Italy as well. The package most supported by Italians respondents, which replaces long-run redistribution to poor countries with potential redistribution to any country, receives more support from Dutch, German and Spanish respondents than from the Italians themselves. Finally, the package most supported by the Dutch, which only differs from the one most supported by the Germans and the Spanish by introducing fines for non-compliance, receives more support among the latter than among the Dutch themselves, which is in line with the generally high level of support among the Germans and the Spanish for EU assistance programmes. This particular package receives relatively little support among the French and the Italians.

An important question is whether there exist packages that receive majority support in each of the sample countries. Since this would be a package on which all sample countries can in principle agree if politicians align with the preferences of their own populations, it would stand a good chance of being politically implementable in the EU. Because our survey covers only a subsample of EU countries, we cannot be sure that such a package would be acceptable to all EU or all Eurozone countries. However, since there is substantial dispersion among our sample countries in terms of their structural economic situation and the positions that their governments have taken in the past when it comes to further budgetary integration, a package that is politically feasible in each of our sample countries could stand a good chance of being politically feasible at the EU or, if not in the complete EU, at least for each Eurozone member state. Table 2 shows that each of the nationally most-supported packages can count on more than 50% support in each of the sample countries, except for the package most supported by the Dutch, which receives less than 50% support among the Italians. However, not all of these packages may be feasible, because an assistance programme cannot be installed if taxes remain unchanged in all the participating countries. Therefore, we are also interested in packages that receive sufficient support in each country and that are feasible in the sense that respondents are willing to pay for the support. One package that fulfils all these criteria is the third package in Figure 9, that is, the package most supported by the German and Spanish respondents, which contains budgetary conditions, mandatory healthcare spending, monitoring and recommending by the Commission, redistribution to poor countries, progressive taxation and no termination and fines for non-compliance.

One might still ask whether this package is realistic, because it may not be politically feasible to shift the entire burden of the programme on a relatively small fraction of a country’s population. Therefore, in Figure 11, we show the national support levels of the package most supported by the Germans and Spanish, but with progressive taxation replaced by a flat tax increase. We observe that support in France drops to <50%, while support in Italy drops to marginally >50%.

Figure 11.

Support for selected flat tax package by country

Notes: (i) The bars indicate the support in respective countries for the flat-tax package (d) in Figure 9. The package includes budgetary conditions, mandatory spending on healthcare, Commission monitoring and recommendations, long-run redistribution to poor countries, flat taxation and no termination and fine for non-compliance. (ii) DE, Germany; ES, Spain; FR, France; IT, Italy and NL, Netherlands.

In the final step of our analysis, we explore the support for some variations on the package with progressive taxation most-supported by the Germans and the Spanish and which receives >50% support in each sample country. We do this by varying the area of mandatory spending, by considering flat taxation instead of progressive taxation and by calculating a less conservative support measure. The latter is achieved by dropping the neutral answers from the sample, hence in this case ${\mathrm{SUPPORT}}_{i,j,k,f}=1$ if the package is rated ‘strongly in favour’ or ‘somewhat in favour’, and ${\mathrm{SUPPORT}}_{i,j,k,f}=0$ if it is rated ‘somewhat against’ or ‘strongly against’. The results are reported in Table 3. Switching from progressive to flat taxation always reduces aggregate support.²³ Varying the spending area, as expected, we find that the package with mandatory spending on banks and deposit always receives the least support, followed by the one with no conditions on the spending area. Most supported is always the package with mandatory healthcare spending. Obviously, dropping the neutrals always raises aggregate support. The effect is often substantial. In fact, it is so substantial that with neutrals excluded each of the proposed packages receives more than 50% aggregate support, irrespective of whether it imposes flat taxation and irrespective of a potential condition on the spending area.

Table 3.

Aggregate support (in %) varying spending area, type of taxation and support measure

Fixed package features: Commission monitoring and recommending, poor country redistribution, budgetary conditions, no fines		Calculation support measure	By type of expenditure
			No conditions	Education	Unemployment benefits	Infrastructure	Banks and deposits	Healthcare
	Pooled frame
Type of taxation	Flat taxation	Neutrals against	43.4	47.8	44.7	46.2	39.3	54.6***
	Progressive taxation		49.1	53.6*	50.4	51.9*	45.0	60.3***
	Flat taxation	Neutrals excluded	63.5***	70.5***	66.0***	68.7***	58.1***	77.2***
	Progressive taxation		70.2***	77.2***	72.7***	75.4***	64.7***	83.9***

*

More than 50% support in three countries.

**

More than 50% support in four countries.

***

More than 50% support in all countries. No stars, more than 50% support in at most two countries.

Table 3 also indicates for each case the number of countries in which there is more than 50% support for a package. Importantly, when neutrals are excluded from the definition of support, there is always more than 50% support in each sample country, irrespective of whether the tax increase is flat or progressive and irrespective of any potential condition on the spending area. Even if a fraction of the neutrals would be against the presented packages when forced to make a choice whether to support or not, there seems to be substantial scope for constructing packages that receive more than 50% support in all sample countries, for example, by including some tax increase for everyone, but more for the rich, and by including at least some mandatory healthcare spending.

6. CONCLUDING REMARKS

Experts have long voiced strong doubts about the long-run viability of the euro in absence of supranational budgetary instruments to support economies hit by adverse economic shocks. However, the political consensus for such budgetary instruments has been missing so far. Some Eurozone member states fear that they may lead to structural redistribution. Hence, until recently the debate on further budgetary integration was stuck in a stalemate between countries wanting to increase risk sharing and those who want risk reduction. However, one of the priorities of the current Commission President is a European unemployment re-insurance scheme and, maybe more importantly, the current corona crisis has revived the discussion about the need for expanding budgetary support for countries in need.

The country-specific positions that we usually observe are those expressed by their political leaders, claiming to represent the views of their voters. However, we have only limited information on how these countries’ populations really think about EU budgetary support packages. The conjoint experiment in this paper intended to shed light on exactly that. It suggests that on average there is substantial support across our sample countries for European-level arrangements to help countries in temporary or permanent economic needs. The general level of support seems higher among our respondents than among politicians: it is even present for countries with political leaders normally opposing further budgetary integration in Europe.²⁴

Adequate design of policy packages can command substantial support. Most populations prefer to condition support on countries reducing their debt in normal times. There is also general support for imposing conditions on how support money should be spent: spending on healthcare comes first, followed by spending on education. Respondents generally see a role for the Commission in terms of monitoring and providing recommendations. However, the support for terminating a programme and imposing fine in the case of non-compliance is small. Further, there is even a general acceptance that programmes lead to long-run redistribution to poorer countries. This is an important observation, because it is extremely difficult to design ‘pure risk sharing’ programmes, that is, programmes that only share shocks, but do not lead to redistribution. One reason is that in reality, it is difficult to distinguish temporary and permanent economic shocks – many shocks are a mixture of the two extremes. The overall rather substantial congruence among the preferences of the different populations opens the possibility of finding packages that get majority support from all individual countries. A package that fulfils this condition is characterized by a combination of budgetary conditions, mandatory healthcare spending, monitoring and recommendations by the Commission, redistribution to poor countries, progressive taxation and no termination and fines following non-compliance. Unanimous support is more difficult to obtain when shifting to flat tax financing or requiring spending in other areas. Still, unanimous support may be available in these cases, for example, by introducing some tax progression and earmarking part of the budgetary support for healthcare spending. It is also important to notice that we have always been very conservative in our measure of support. Assuming that, say, half of the neutrals become supportive when forced to make a choice makes the unanimity criterion substantially easier to fulfil.

Obviously, one has to interpret our findings with caution. Although we use expressions such as ‘majority support’, one cannot interpret the support for our selected policy packages as the prediction of a real vote after a political campaign. The support we find represents genuine individual preferences, but it is also to some extent ‘pre-political’, that is, captured on the basis of a framing that may be different from the framing that comes to dominate after a political campaign on the issue of EU support instruments. Our respondents had to answer the following question: what do you think about a series of alternative policy proposals that are discussed at the European level, with a view to launching a new European initiative? Notwithstanding, the fact that we clearly told our respondents that this was about a new European-level initiative, creating a European scheme of mutual assistance, and that we made them think about conditions that might be imposed on countries, it is plausible that the responses focused mostly on the social content of the proposals and their concrete specification, and less on the fact that this would constitute a new European initiative that could open up conflict-lines among countries; or, less on the fact that the initiative might involve the temporary creation of EU-level debt. Imagine, for instance, that the central question of a public debate would be ‘are you for or against a new EU initiative?’, with a virulent campaign of some political parties against the EU; or, ‘are you for or against issuing new debt at the EU-level?’ Then, the outcome of a real vote might be different. We write ‘to some extent pre-political’, because the question whether the EU should support countries in need because of corona was obviously already being discussed at the moment of fielding the survey, although in vague terms, and we do observe some congruence between the country-level differences in public attitudes and the public positioning of national governments on the issue. In addition, we cannot rule out the possibility that our results are partially driven by how our survey was fielded when the corona crisis was already acute at least in large parts of Europe, when populations could be coming to expect net financial support. The central conclusion must not be that public support for European social initiatives is readily available; it should be that, depending on the orientation and framing of the debate and on the specific policy design that is proposed, widespread support from individual Member States for an EU support programme is possible. The broad support from the Dutch and German respondents at a moment when the acuteness of the corona crisis was higher in Southern Europe attests to this.²⁵ The actual political conflict is, therefore, in part a conflict about the way in which the relevant proposals are framed. Naturally, durable support for an assistance package requires that respondents understand the contents and the implications of the package. Obviously, we cannot be sure that this is the case for all our respondents. Yet, the fact that the far majority of them pass the attention check and that the survey results are generally plausible should give some reassurance in this respect.

Using conjoint experiments of the type deployed here is a promising avenue for further research on the preferences regarding European policies. One such avenue would be to distinguish between a frame with normal business cycle fluctuations and one with extreme shocks, in which case moral hazard considerations associated with rescue packages may be dominated by issues of urgency. A second is to include into the sample countries from other blocks, such as Central and Eastern Europe and Scandinavia. A third is to study the support for alternative spending areas, such as climate protection and digitization. Finally, it would be interesting to explore how the support for assistance programmes is affected when they are financed by issuing EU-level debt.

SUPPLEMENTARY DATA

Supplementary data are available at Economic Policy online.

Footnotes

1

Early proposals include the ‘MacDougall Report’ (Marjolin et al., 1975), Padoa-Schioppa et al. (1987) and Italianer and Van Heukelen (1993). The latter propose a capacity outside the general EU budget for grants to countries suffering from shocks that raise their unemployment rate. For a recent plea in favour of a central fiscal capacity, see Buti and Carnot (2018). For further discussion, see Bilbiie et al. (2021).

2

For instance, Bechtel et al. (2014) on bailouts, Vandenbroucke et al. (2018) and Burgoon et al. (2022) on European unemployment reinsurance schemes and De Ruijter et al. (2020) on the joint procurement of medicines.

3

Krugman and Venables (1995), although not specifically referring to EMU, describe the mechanisms.

4

They consider among other things the endogeneity of financial integration, symmetry of shocks and flexibility of labour and product markets.

5

De Grauwe and Ji (2017) demonstrate a high degree of business cycle synchronization among euro-area economies over the period 1999–2014. That is, correlations of the business cycle component of GDP growth are generally high. However, the amplitudes of the business cycles differ substantially across countries, which would still confront the ECB with the problem that it can only imperfectly stabilize national economies. The countries hit hardest by a common negative shock would legitimately need support from other countries. Differences in business cycle amplitudes and their consequences are also highlighted in Belke et al. (2016).

6

There is a fear of moral hazard associated with an EU level macroeconomic stabilization function, for example, see Koester and Sondermann (2018) and Burriel et al. (2020). Some authors, such as Heijdra et al. (2018) argue that there is no need for EU fiscal support arrangements if countries adhere to following the responsible fiscal policies they have committed to.

7

Various proposals have been made to mitigate moral hazard in relation to budgetary support arrangements. Beetsma et al. (2021) present a mechanism based on asymmetric sectoral shocks coming from changes in world trade. Transfer flows are driven by cross-country differences in sectoral structure. Because shocks to world trade can be considered largely exogenous, moral hazard considerations should be relatively minor. Institutional moral hazard can also be mitigated by means of minimum standards with regard to the quality of domestic policies in the participating member states, which constitute ‘conditions’ for receiving support. Linking central support to quality assurance of the policies implemented by sub-central entities is a well-known strategy to fight institutional moral hazard in multi-layered welfare states (Vandenbroucke and Luigjes, 2016; Luigjes and Vandenbroucke, 2020).

8

Wyplosz (2020) acknowledges the possibility of moral hazard, but views the emergency created by the pandemic as more important.

10

Enderlein et al. (2013) propose a CFC based on national output gaps, to which countries with a better-than-euro-area-average cyclical position contribute and from which countries with a worse-than-average position receive support. Furceri and Zdzienicka (2015) explore transfers based on country-specific GDP shocks.

11

Specifically, Lee’s (2006) analyses of web-based panels like IPSOS’s finds that ‘nonresponse error measured by the differences between the estimates from the respondents and the known full sample values [based on additional survey and interviews with respondents] was not found to be large, implying that nonresponse error in these web survey data may not be critical.’

12

In fact, the President of the European Banking Authority recently argued for devoting part of the EU recovery funding for a preventive recapitalization of the banking sector resembling the US Troubled Asset Relief Programme deployed during the financial crisis of 2008 (Reuters, 2020).

13

Guttenberg and Nguyen (2020) argue that the governance of the decisions on the Recovery and Resilience Plans, which contribute to the recovery of the EU economies after the Covid-19 pandemic, lack democratic accountability. They envisage that the national parliament has a say in the Plan that its country submits and that the European Parliament has a veto over the Plan. This would enhance both national and political ownership of the Plans. Our experiment does not address the role of these institutions in the policy packages that our respondents see. However, this could be a topic for further research.

14

See Vandenbroucke (2020) for an account of pure insurance and redistribution and the normative connotations of these concepts in the EU context.

15

Still rich countries may benefit on net, because they will be eligible for assistance when they are hit by a negative shock, enabling them to reduce cutting back on other programmes. However, assistance could be more generous for poor countries for a given negative shock. In addition, ex-ante redistribution from richer to poorer countries may be in the economic self-interest of the former, for example, because poorer countries do not cut imports from the richer countries or because the poorer countries’ financial stability would be better guaranteed. We ignore such ‘second-order’ aspects in discussing redistribution.

16

An important question is how to design optimal arrangements under the restriction that they muster sufficient political support. Answering this question is beyond the scope of the present paper, though. The findings in Section 5 may provide leads for addressing this question.

17

This is partly a phenomenon known as ‘mid-pointing’ that reflects respondent laziness, bias or convenience. The Wang and Krosnick (2020) study of alternative specifications suggests that research designs including middle points (the neutrals, in this case), and hence allowing for mid-pointing, still are less biased and more efficient than research designs forcing pro- or anti-options (see also Krosnick, 1991). Nonetheless, our robustness analyses below include estimates based on samples excluding neutrals.

18

Online Appendix E expands the baseline specification by adding dimension–dimension interactions. However, only few of these interactions are significant and, hence, we do not further analyse the effects of dimension–dimension interactions on preferences.

19

The strong effect of earmarking assistance spending for healthcare on the appreciation of a package may beg the question why we see so little reshuffling of existing public spending towards healthcare. Alternatively, it raises the question whether respondents understand the concept of a government budget constraint. However, in our view, the aforementioned effect is not informative in this regard. First, our description simply states what happens with the additional resources received, without making any assumptions about existing spending. Second, reshuffling spending may not be so easy legally or politically, because much of the existing spending has already been committed to existing programmes and taking away resources from current beneficiaries is likely to cause resistance from this group.

20

Interestingly, maybe not surprisingly, for people who identify themselves as right-wing the fixed effect for the permanent frame is significantly negative (at the 10% level) in the regression with our support variable on the left-hand side. However, in the regression with the full sample, this effect is diluted by the presence of left-wing and centre-oriented people.

21

Our data can consider how conditions relevant to cross-national comparison might show up in variations in individual attitudes – such as individual-level variation in trust in one’s national government and in the EU. Addressing such factors is beyond the scope of our present experimental analysis of broad population support for policy designs. However, it is worth noting that controlling for trust, or for that matter interacting trust with package-specific dimension values, does not change the baseline AMCE results in either the pooled or the country-specific specifications. See Online Appendix D, Table D.1, for detail on trust, and the discussion below on other individual-level covariates (Section 4.6).

22

Caution is warranted, as the summary of the moderating role of individual correlates is confined only to the two-way interaction between the dimensions and individual-specific variables.

23

This is in line with other empirical research. For example, a quasi-experiment by Alpina et al. (2020) investigating the responses of mayors of Italian municipalities to austerity measures prompted by the imposition of local budgetary constraints suggests that mayors try to preserve popularity by putting most of the increased tax burden on high-income earners.

24

A recent survey among national parliamentarians of France, Germany and Italy by Blesse et al. (2020) on EU budgetary support instruments, in this case a European unemployment insurance, seems to confirm the ‘stereotype’ that German politicians are less in favour of such instruments than Italian politicians.

25

Another matter of relevance is whether proposals for an EU assistance scheme are made behind a veil of ignorance. It is unlikely that at the moment our experiment was fielded the veil of ignorance was completely intact, as forms of support were discussed early on and it would be unlikely that no support would be available for the hardest-hit countries.

9

The need for a CFC in the Eurozone is often motivated by a lack of cross-border private sector risk sharing, such as through diversification of asset portfolios. For recent estimates, see Cimadomo et al. (2020). However, even substantial cross-border risk sharing of this type does not a priori obviate the need for a CFC, because common shocks may be very large (as with Covid-19) and monetary policy constrained. In the context of a New-Keynesian model of a currency union, Hettig and Mueller (2018) demonstrate that, with monetary policy at the effective lower bound, fiscal coordination leads to higher government spending than in the absence of coordination. In the latter case, governments hold spending back, because higher spending produces a terms-of-trade appreciation, which depresses demand for home-produced goods in the presence of economic slack. Coordination internalizes the fact that in equilibrium the terms of trade are unaffected under a coordinated expansion. A CFC may likewise internalize this fact.

APPENDIX A: FORMULATION OF THE FRAMES

Framing 1 (Temporary shock)

Framing 2 (Permanent shock)

European governments spend money on policies such as infrastructure, education, social assistance, military defence, housing, etc. When a country is hit by a severe but temporary economic downturn, it can be difficult to maintain these policies during the downturn. We would like to hear your opinion about a new European programme discussed by European governments to address such difficulty. This new programme would provide temporary budget support to countries in need. Such support would never be larger than 1% of the receiving country’s GDP. The budget support would help governments maintain their policies during the economic downturn and stabilize the economic situation. This mutual assistance programme would be financed by the participating countries. This European assistance programme can be organized in different ways. Different conditions can be imposed on countries that benefit from the support. Therefore, in the next pages you will be shown alternative options. You will be asked to indicate which options you prefer (or dislike the least), and how much you are in favour or against these proposals.

European governments spend money on policies such as infrastructure, education, social assistance, military defence, housing, etc. However, when a country is confronted with long-lasting economic problems (such as a permanent decline in an important industrial sector), it can be difficult to maintain these policies. We would like to hear your opinion about a new European programme discussed by European governments to address such difficulty. This new programme would provide budget support to countries in need. Such support would never be larger than 1% of the receiving country’s GDP. The budget support would help governments in maintaining their policies and to address these long-lasting economic problems. The mutual assistance provided by this programme would be financed by the participating countries. This European assistance programme can be organized in different ways. Different conditions can be imposed on countries that benefit from the support. Therefore, in the next pages you will be shown alternative options. You will be asked to indicate which options you prefer (or dislike the least), and how much you are in favour or against these proposals.

APPENDIX B: EXAMPLE OF A SCREENSHOT WITH THE QUESTIONS AND A PAIR OF POLICY PACKAGES

Author notes

*

We thank for very helpful comments an editor, our discussants Edouard Challe and Moritz Kuhn, two anonymous referees and seminar participants at Bruegel (15 June 2020), ZEW Mannheim (1 July 2020) and CEPS (6 July 2020). Special thanks go to our discussant Marco Buti at the seminar at CEPS. We gratefully acknowledge funding of this project by the Amsterdam Centre for European Studies (ACES). The Managing Editor in charge of this paper was Tommaso Monacelli.