Multilevel blame games: Blame-shifting in the European Union

Who blames whom in multilevel blame games? Existing research focuses either on policymakers' preferences or their opportunities offered by the institutional structures in which policymakers operate. As these two strands of literature barely refer to each other, in this article we develop an integrated theoretical model of blame-shifting in multilevel governance systems and assess it empirically. In line with the first strand, we assume that policymakers have a preference for shifting blame onto actors on a different level from themselves. In line with the second, we suppose that opportunities for doing so depend on institutional responsibility for policymaking and policy implementation. We check the plausibility of our integrated model by examining policymakers' blame attributions in three cases where European Union migration policies have been contested: border control, asylum, and welfare entitlements. We find that our integrated model does better in explaining blame-shifting in these cases than the isolated models.


| BLAME-SHIFTING IN MULTILEVEL GOVERNANCE SYSTEMS
A core feature of modern politics-whether national or international, democratic or autocratic-is that actors play blame games. They try to "deflect blame by blaming others" (R. K. Weaver, 1986, p. 385) when policies are contested. 1 The question, therefore, is who blames whom in these blame games. The most common answer in the literature is that blame games are played between political parties, especially between parties in government and opposition (Gerhards, Offerhaus, & Roose, 2009;Hansson, 2017;Roose, Scholl, & Sommer, 2018;R. K. Weaver, 1986). For example, in the United States (US) Democrats blame Republicans for their tax cuts and health-care policies and in the United Kingdom (UK) the Labour Party blames the Conservative Party for its Brexit policy (Heinkelmann-Wild, Kriegmair, Rittberger, & Zangl, 2019).
Another answer provided by the literature is that blame games are played between governmental actors and their subordinate agents, especially the bureaucracy (Gilad, Maor, & Bloom, 2013;Hinterleitner & Sager, 2019;Hood, 2011). For instance, government representatives blame the agencies in charge of public transportation or health care for inadequate services, and agency managers respond by trying to shift the blame back onto the government (Mortensen, 2016;Nielsen & Moynihan, 2017). So, while Pentagon representatives blamed members of the Office of Strategic Influence (OSI) for planned disinformation campaigns targeted at allied countries in the course of the "War on Terror," OSI officials blamed the secretary of defense (Schwarzenbeck, 2017, pp. 94-107).
Moving on from blame games between government and opposition or government and bureaucracy, a more recent literature points to blame games involving governmental actors located at different levels of a multilevel governance system (MLGS). In the US, state authorities in Louisiana blamed the federal government, including President George W. Bush, for inadequacies in crisis management after Hurricane Katrina, while the federal government pointed to the state-level authorities' own responsibility for crisis management (Maestas, Atkeson, Croom, & Bryant, 2008). Moreover, when United Nations (UN) peacekeeping missions fail, member states repeatedly blame the UN, while UN officials shift the blame back onto member states for their failure to provide sufficient resources or mandates. In the European Union (EU), the European Commission blamed the authorities in the member states for violating European standards on particulate matter pollution, while, in return, some of these authorities criticized EU standards for being too restrictive. Somewhat stylized, two strands can be distinguished in this literature.
The first, preference-focused strand asserts that actors' preferences explain the direction of blame-shifting in MLGS. It holds that policymakers have a strong preference for shifting blame onto policymakers from another level of government than the one on which they themselves are located. This strand of literature can thus explain differences in blame-shifting strategies among actors located on different levels of a MLGS. 2 It can explain, for instance, why national policymakers tend to shift blame onto EU-level actors such as the European Commission rather than blaming each other and why the Commission tends to blame EU members rather than the European Parliament (EP), the European Court of Justice (ECJ), or EU agencies. However, the preference-focused strand has difficulty in explaining why policymakers on one level do not always shift blame onto policymakers on another. Why, for example, did German politicians blame the Greek government for the social disruption that occurred during the implementation of European rescue programs rather than the European Commission?
The second, opportunity-focused strand argues that institutional structures of policymaking shape actors' opportunities to shift blame in MLGSs. Thus, actors who are in positions of institutional responsibility for policymaking find it harder to avoid blame than actors who are not involved in policymaking (or, at least, are much less involved in it). 3 Consequently, this strand can explain why actors with responsibility for a given policy become the targets of blame attributions. For example, during the "Euro crisis," Greek politicians repeatedly blamed the EU for the strict conditionalities they had to fulfill under European rescue programs. However, the opportunityfocused strand cannot explain why responsible actors get blamed by some actors more than others.
Why, for instance, does the European Commission, as the watchdog of the European treaties, refrain from joining the choir of member states blaming the European Central Bank (ECB) for overstepping its mandate by purchasing deficit countries' government bonds?
To solve these puzzles, more recent contributions to the literature combine insights from both strands, opportunity-focused and preference-focused (cf. Mortensen, 2012Mortensen, , 2013a. Drawing on this literature we developed a model of blame-shifting in MLGSs that integrates the opportunityfocused and preference-focused literature strands in a theoretically meaningful way (Section 2). Thus, our model is not formed simply by adding-ad hoc-insights of the two strands together but offers their theoretically coherent integration. Drawing on theories of bounded rationality, it suggests that blame-shifting, like any other social behavior, is shaped by a combination of the actors' preferences and their opportunities. Moreover, it refines core assumptions from extant literature. Going beyond the opportunity-focused literature, we assume that blame accompanies not only responsibility for policymaking, but also-and, in fact, mainly-responsibility for policy implementation. Likewise, going beyond the preference-focused literature, we theorize the standard assumption that multiple actors from the same level of government share a preference for shifting blame onto actors at the other levels of a MLGS. While the main contribution of our article is thus theoretical, we study blame attributions in the EU to offer an empirical plausibility probe of our model (Section 3). Using content analysis, we compare policymakers' blame attributions in three cases where EU migration policies have been contested: (a) the EU's external border policy; (b) the Common European Asylum System; and (c) EU citizens' entitlement to social benefits in their country of residence. We demonstrate that our integrated model fares better in explaining blame-shifting in these cases than the isolated models offered by the dominant literature strands (Section 4). In the conclusion, we summarize our contribution (Section 5).

| THEORY: AN INTEGRATED MODEL
To examine who blames whom in MLGS, we suggest a blame-shifting model that integrates the underlying assumptions from two strands in existing blame-shifting literature. The first strand focuses on policymakers' preferences as to whom they want to shift blame onto but tends to ignore the fact that the opportunities for shifting blame to these actors are typically constrained by institutional structures. The second strand, by contrast, highlights that policymakers' opportunities for shifting blame to particular actors are often constrained by institutional structures, but tends to ignore that different policymakers have different preferences with regard to whom they want to scapegoat. Drawing on the standard assumptions of theories of bounded rationality, we integrate these two strands in a theoretically coherent model. We suppose that, as in any social behavior, actors' choices regarding whom they shift the blame for contested policies onto are shaped by a combination of their preferences and opportunities (Elster, 2015, p. 190f.). In this view, actors assess their opportunities and opt for the one they expect to serve their blameshifting preferences best. Therefore, our model specifies the conditions that shape policymakers' preferences as well as the opportunities they have to shift the blame onto the actors in question.

| Blame-shifting preferences
Building on the preference-focused literature, we assume that policymakers always have a strong preference for avoiding blame for contested policies by shifting it onto other actors. If policies are publicly criticized, policymakers use presentational strategies. They engage in strategic communication that downplays their own responsibility for the said policies and emphasizes the responsibility of others. With regard to MLGSs, we further assume that-ceteris paribustheir position in the system determines their preference for blame-shifting (Gerhards et al., 2009;Maestas et al., 2008;Mortensen, 2012). We hold that policymakers on each level prefer shifting blame onto actors on the other level rather than their own. In other words, "where you stand depends on where you sit" (Alisson, 1969). We expect, for instance, policymakers from member states to prefer to point the finger at the EU as a whole or at EU actors such as the European Commission, the European Council, the ECJ, or the ECB rather than at other member state (MS) actors. In short, they prefer "playing the blame game on Brussels" (Schlipphak & Treib, 2017, p. 355).
There are two mutually reinforcing reasons why, in MLGSs, policymakers prefer to shift blame onto another level rather than blaming actors on their own level: loyalty and interdependence. Due to the frequency of their meetings-and other interactions-loyalty among actors on the same level is often much stronger than loyalty between actors at different levels of government (Bearce & Bondanella, 2007;Lewis, 2005;Taninchev, 2015). For example, in the EU, the loyalty among members of the European Commission or the EP can be assumed to be much stronger than their loyalty to MS representatives. Their frequent meetings in Brussels, Luxemburg, or Strasbourg create a mutual loyalty that deters them from shifting blame onto one another, a deterrence that does not exist to the same extent vis-à-vis policymakers from MS actors (Gerhards, Roose, & Offerhaus, 2013, p. 125). At the same time, the loyalty among policymakers from MS actors frequently exceeds their loyalty to EU actors. In particular, MS ministers typically respect each other more than, for instance, members of the European Commission, let alone the European Parliament. They consider themselves equals who deserve each other's loyalty (Bull, 1977;Roose et al., 2018, p. 70). This loyalty is further strengthened by their frequent meetings inside and outside the EU context. While this bond tends to deter MS policymakers from blaming each other for contested EU policies, there is nothing similar to deter them from blaming the EU itself and other actors representing EU actors. 4 While this is particularly true for the EU where the low level of loyalty between national-level and EU-level actors is not really offset by cross-level bonds, for instance within the same political party (family), it tends to be a common condition of many-albeit not all-MLGS. 5 Leaving aside mutual loyalty, in many MLGSs, policymakers from the same level often depend on one another more than on policymakers from other levels. This deters them from shifting blame onto each other. For both office-seeking and policy-seeking actors, support from policymakers at their own level is typically more crucial than support from any other level (Moynihan, 2012). For example, as MS actors depend on each other for decision making in the Council much more than on the European Commission (Heisenberg, 2005;Novak, 2013), they have to exercise more caution in blaming each other than in blaming the European Commission or other EU actors. They simply cannot afford to upset representatives from their fellow MSs. Of course, a similar dependence might also deter members of the Commission from shifting blame onto policymakers from MS actors. However, as Commission bureaucrats depend much more on their standing in Brussels than on their reputation among MS actors, they will be even more cautious about shifting blame onto other EU actors than about blaming MS policymakers when EU policies are contested.
In sum, as a result of their deeper loyalty to, and stronger dependence on, actors from the same level of government, we expect policymakers in a MLGS to prefer shifting blame onto actors on a different level of government.

| Blame-shifting opportunities
Drawing on the opportunity-oriented literature, we argue that policymakers can shift blame according to their preferences only when they are able to maintain the "illusion of objectivity" (Kunda, 1990, pp. 482-483). Whether or not a given policymaker can plausibly deny responsibility when shifting blame depends on the institutional structures of the MLGS in which the contested policy was made (Hood, 2011, 76, 146-147;Schwarzenbeck, 2015, p. 37). The institutional structure shapes the opportunities for shifting blame onto actors on another level. Policymakers' opportunities for blame-shifting are severely hampered if institutionalized policymaking responsibility is obviously located at their own level of government. In this case, policymakers lack the opportunity for plausible deniability. However, in MLGSs institutionalized policymaking responsibilities are hardly ever obvious; they are far more likely to be complex and thus difficult to comprehend. 6 Policymaking typically involves a multistage process, in which multiple actors across multiple levels share responsibility. The EU is a good illustration. Policymaking is divided between the Council and the EP, acting on the initiative of the Commission. Due to this institutional complexity, not just the general public but even experts struggle to assess the institutional responsibility for policymaking in the EU (Hobolt & Tilley, 2014, p. 45;León et al., 2018, p. 661;Rittberger et al., 2017, p. 912). Consequently, like any MLGS, the EU offers policymakers ample opportunities to shift blame onto actors on other levels (Gerhards et al., 2009;Gerhards et al., 2013;Hood, 2011, 83, 122;Roose et al., 2016).
However, even in MLGSs such as the EU, policymakers' blame-shifting opportunities are not unlimited. Not all policymakers have the same responsibility for policymaking. And, the more that policymaking responsibility is located on their own level, the more difficult it will be for them to shift the blame for contested policies onto another level (Hood, 2011, 76, 146-147;Roose et al., 2018, p. 48;Schwarzenbeck, 2015, p. 37). For instance, due to its responsibility for making agricultural policy, the European Commission will have more difficulty in shifting blame for policies in this domain onto MS actors than in other domains where its policymaking competencies are more restricted.
Moreover, as policymaking in MLGSs is always, at least to some degree, shared among actors located on different levels of government, policymakers' blame-shifting opportunities depend much more on their institutional responsibility for implementing a policy than on their responsibilities for making it (Heinkelmann-Wild, Rittberger, & Zangl, 2018;Rittberger et al., 2017). We suggest that, if a policy requires "on the ground" implementation, the implementing actors will almost automatically become focal. As those responsible for policymaking are difficult to identify in MLGSs, the implementer will be the actor most closely associated with the contested policy and thus will be hard put to avoid blame. Other actors' attempts to shift the blame onto the implementer are likely to be considered plausible; implementers' own blameshifting attempts, on the other hand, are unlikely to be seen as equally plausible. By the same token, policymakers from the same level as the policy implementer will find it difficult to shift blame to actors on another level. Thus, if, for example, an EU agency is responsible for the implementation of a contested EU policy, it will be difficult for EU actors such as the Commission to follow their preference for shifting blame onto MS actors. Only when contested policies do not require active implementation will policymakers be unconstrained in following their preferences for shifting the blame to an alternative level.
In sum, we suggest that policymakers' opportunities to shift blame according to their preferences are more constrained in MLGSs than the preference-oriented strand in the literature would have us believe. As policymakers are difficult to identify in MLGSs, actors' blame-shifting opportunities will be constrained much more by institutional responsibilities for policy implementation than by actors' policymaking responsibilities. 7 Beyond what the opportunity-focused strand in the literature suggests, they are only unconstrained when the policy in question does not require implementation.

| RESEARCH DESIGN: TESTING THE MODEL
We evaluate the plausibility of our integrated model by studying the attribution of blame for contested EU policies. We focus on the EU because it is a textbook example of a MLGS, for which we derive from our model the following propositions concerning the shares of blame attributions targeting the EU. We expect the following rank order (see Table 1) 8 : • High: The share of blame attributions targeting the EU will be high when the institutional structures of policy implementation allow MS actors to follow their preference for shifting blame onto the EU. This will be the case when policies are either implemented by EU actors or do not require active implementation. • Medium: The share of blame attributions targeting the EU will be medium when either MS actors or EU actors are heavily constrained by the institutional structures to pursue their preferences. This will be the case for MS actors (EU actors) when policies are implemented by MS actors (EU actors). • Low: The share of blame attributions targeting the EU will be low when the institutional structures of policy implementation allow EU actors to follow their preference for shifting blame onto MS actors. This will be the case when the policy is either implemented by MS actors or does not require active implementation.
We assess the plausibility of these propositions by comparing policymakers' blame attributions in three cases of contested EU migration policies. We selected the following cases: • EU border control: The establishment of the Schengen Zone led to the abolition of internal border controls in the EU in the 1990s. As a result, a common external border control system was created and Frontex was charged with implementing it. The EU agency assists MSs operationally and coordinates joint operations to protect the EU's external borders (Lavenex, 2015, p. 381). The EU's border control policy was contested by the public. The main criticism focused on its inability to prevent the deaths of refugees trying to enter the EU via the Mediterranean. Furthermore, the agency was accused of dealing inappropriately with refugees. Critics saw EU border control operations as contributing to the death of migrants either by not doing enough-as a result of their restricted mandate and limited resources-or by doing too much-by attracting migrants to take the dangerous route across the Mediterranean. • EU asylum system: With the elimination of internal border controls in the Schengen Zone, the EU established the Common European Asylum System (CEAS). Its cornerstone, the "Dublin system," stipulates that the MS in which an asylum seeker first enters the EU is responsible for his or her asylum claim (Lavenex, 2015, p. 381). The CEAS was contested by the European public. Among other things, it was criticized for overburdening MSs on the EU's external border while privileging members that do not have external EU borders.
Another critique was that discrepancies in the treatment of asylum seekers persisted despite common EU standards. Moreover, "Dublin" was criticized for creating incentives for MSs receiving first entry asylum seekers to refrain from registering them and thus apply a "wavethrough approach." • EU welfare entitlements: According to the European common market programme, EU citizens have the right to take residence in any MS without being denied the welfare entitlements provided in their country of residence. Thus, MSs cannot limit social benefits only to their own nationals. The EU welfare entitlement policy therefore does not require active implementation, but policymakers are obliged to refrain from direct political intervention leading to discrimination against EU citizens (Leibfried, 2015, p. 279 f.). This policy of welfare entitlements within the EU was heavily criticized by the public for enabling so-called "welfare migration" and "social tourism." We selected the three aforementioned cases according to the logic of a most similar case design (Przeworski & Teune, 1982, p. 32f.). While certainly dissimilar in some respects, the three cases are similar inasmuch as they belong to the same policy field of the EU's migration regime. The cases are thus intimately linked to the EU principles of free movement across internal borders and the establishment of a common external border. Their underlying policymaking structures are also similarly complex. Following a proposal by the Commission, the Council decided in all three cases on a policy in which the EP would be involved either through co-decision or consultation procedures. While not amounting to a quasi-experimental design, the similarities between the cases allow these confounding variables to be controlled and thus help us in singling out the effect of the independent variables.
As required in a most similar case design, the scores of the independent variables-namely policymakers' blame-shifting preferences and blame-shifting opportunities-differ in the three cases. Blame-shifting preferences vary within each case because the three cases allow for the assessment of blame attributions by both EU and MS actors. Furthermore, blame-shifting opportunities vary across cases, because the cases differ systematically with regard to the structures of policy implementation: EU border control policies are implemented by an EU actor (i.e., Frontex); the EU asylum system is implemented by MS actors; and the EU welfare entitlement policy does not require any active implementation. While controlling for confounding variables, the variance in the cases' independent variables allows for the evaluation of their assumed effect on the dependent variable, that is, the direction of policymakers' blame attributions.
To assess these attributions in the three cases, we engaged, as is common in the literature, 9 in content analysis of the media coverage of the three contested EU migration policies. 10 We examined the coverage provided by nine European quality newspapers for the 6-year period between 2010 and 2016. We included in the analysis two leading quality newspapers in Austria (Die Presse, Der Standard), Germany (Süddeutsche Zeitung, Frankfurter Allgemeine Zeitung), Ireland (The Irish Times, The Irish Examiner), and the UK (The Guardian, The Daily Telegraph). On the basis of the authors' language proficiency and the accessibility of the newspapers, all those selected are from Northern and Central European countries. Therefore, blame attributions targeting Eastern or Southern European countries' migration policies may be overrepresented in our sample. To be able to control for this potential bias, we added one quality newspaper with a pan-European reach (The European Voice/Politico Europe). Overall, this selection of nine newspapers (not countries!) aims to approximate blame attributions in the European press.
To single out articles covering the three cases of contested EU migration policies, we conducted keyword searches in digital newspaper archives (see Appendix Table S1). Given that we wanted to identify, as far as possible, all articles covering our cases, we used fairly unrestrictive search strings based on broad keywords. The results of the keyword search were reviewed manually to sort out articles that did not address the respective policy as well as articles that did not hint at any contestation of the policy. Only in the asylum case, where the original search string led to a sample size unfeasible for qualitative content analysis, did we add the keyword "problem" in order to reduce the sample size while, at the same time, deselecting (irrelevant) articles that did not cover any contestation of the policy, before reviewing the articles manually.
In our final sample of 1.040 articles covering the contestation of the three policies, we searched for policymakers' blame attributions, which were identified based on the following criteria 11 : • Blame object, that is, a contested policy for which blame is attributed. For the purpose of this article, we define EU policies as contested if they are publicly discussed as failures. While the perception of failure does not have to be consensual, it has to be widespread among the general public, and thus frequently referred to as such in the respective media coverage. • Blame sender, that is, an actor who attributes blame for a contested policy. For the purpose of this article, we focus on policymakers, who we define as actors who are involved in the making and/or implementation of the respective EU policy. We differentiate between two types of policymakers, namely EU actors (the Commission, the Council, and the European Parliament) and MS actors (EU MS governments). • Blame target, that is, the actors to whom blame is attributed. For the purpose of this article, we focus on actors who participate in the making or implementation of the contested policy. Blame targets are categorized according to their position in the EU MLGS as actors at EUlevel or at national-level.
In our samples, we identified 480 blame attributions. 12 To assess whether they corroborated the expectations of our blame-shifting model, 13 we calculated for each blame sender (EU or MS actor) and for each case (border control, asylum system, welfare entitlements) the share of blame attributions that targeted EU actors and MS actors, respectively.

| EMPIRICS: BLAME ATTRIBUTIONS FOR CONTESTED EU MIGRATION POLICIES
To assess our blame-shifting model empirically, we studied the blame attributions by EU and MS actors in three cases of contested EU migration policies. In the following sections, we will discuss for each individual case whether the share of policymakers' blame attributed to EU-level and national-level actors conforms to the expectations of our model. We will then proceed by comparing the findings across cases.

| EU border controls
Whom do policymakers blame for the perceived failure of the EU border control regime? Corroborating our predictions, we find that the bulk of MS actors' blame attributions (n = 65) are targeted at the EU-level (77%) while only a small minority of statements attribute blame to the national-level (23%). Hence, MS representatives predominantly blame actors at the EU-level. For example, Italian Interior Minister Roberto Maroni claimed: "The EU is doing absolutely nothing and is too slow and bureaucratic" (cited by Agnew, 2011, p. 12). Similarly, former French President François Hollande reportedly noted that Frontex's missions "led more and more migrants to cross the Mediterranean" (Marlowe, 2015, p. 10) and the British Foreign Office Minister, Joyce Anelay, criticized Frontex for its operation Triton, which, she claimed, acted as "an unintended 'pull factor', encouraging more migrants to attempt the dangerous sea crossing and thereby leading to more tragic and unnecessary deaths" (Travis, 2014). By contrast, statements by MSs blaming national actors are rather rare. One of the very few examples is a comment made by Italian Prime Minister Matteo Renzi who "issued a harsh criticism of his European partners for not assisting Italy [in securing the EU's borders]" (Kirchgaessner, Traynor, & Kingsley, 2015).
In line with our expectations, EU actors' blame attributions are more divided: they attribute responsibility for policy failures to the EU-level in 36% of their blame statements (n = 47) while in 64% of them they target the national-level. Hence, in a comparatively more narrow majority of blame attributions EU actors' blame is targeted at MSs, as when Commissioner Dimitris Avramopoulos claimed: "Frontex is not a European border-protection system. If we want one we will have to build one. […] Building an EU coastguard is something which member states are not ready to countenance" (cited by Panichi, 2015). In a sizeable minority of cases, EU actors target the EU-level. For instance, Members of the European Parliament (MEPs) blamed the Council by stating that its "political will lags behind reality" 14 and describing its actions as "empty lip service and crocodile tears." 15 Similarly, High Representative Federica Mogherini noted: "With this latest tragedy […] we have no more excuses, the EU has no more excuses" (cited by Kirchgaessner et al., 2015).

| EU asylum system
Whom do policymakers blame for the failure of the CEAS? In this instance (n = 59), as expected, EU actors attribute blame predominantly to MS authorities (93%), with only a very small minority of attributions directed at the EU-level (7%). For instance, Commission President Jean-Claude Juncker reportedly "complained that national governments were failing to observe agreements on asylum procedures" (M. Weaver, 2015a). EU Commissioner Cecilia Malmström "accused EU governments of […] failing to protect refugees from North Africa" (Watt, 2011, p. 10). MEP Sylvie Guillaume described certain MSs' asylum systems as a "cata- In line with our expectations, the share of MS actors' blame attributions targeting the EU is somewhat higher. While the majority of MS actors' blame attributions (n = 228) are targeted at MS authorities (80%), a significant minority still target actors at the EU-level (20%). For instance, taking aim at Hungary, the French Foreign Minister Laurent Fabius condemned the "scandalous" and "extremely harsh" conditions asylum seekers entering the country were subjected to: "Hungary is not respecting Europe's common values" (cited by McLaughlin, 2015a, p. 8). Furthermore, the Austrian and German Foreign Ministers reportedly "accuse[d] the government in Rome of committing a breach of the Dublin II agreement, according to which the country responsible for a refugee is the one in which he first sets foot on European soil: Italy […] would simply allow the refugees to move north instead" (Die Presse 2014b; authors' translation). Additionally, Hungarian Prime Minister Victor Orbán reportedly "accused the German chancellor […] of triggering a surge of migrants to Europe by making Syrians exempt from the EU's so-called Dublin rules" (McLaughlin, 2016, p. 4) and "blamed Germany for encouraging people to risk their lives coming to Europe with its promise of more places for refugees" (Khomami, 2015). By contrast, for MSs to blame actors at the EU-level is comparatively rare. For instance, Orbán's chief of staff criticized "the policies […] of the European Commission, according to which anybody should be allowed into the territory of the European Union" (cited by McLaughlin, 2015b, p. 9). Furthermore, UK Home Secretary Theresa May reportedly stated that "the failings of a 'broken European migration system' were exacerbated by passport-free travel through much of the bloc" (McLaughlin, 2015a, p. 8).

| EU Welfare Entitlements
Whom do policymakers blame for the allegedly excessive amount of "social tourism" facilitated by the EU's welfare entitlement policy? In line with our model, among MS actors blame (n = 55) was attributed predominantly to the EU-level (87%). For example, Horst Seehofer, the chairman of the German co-governing party CSU, reportedly "blamed the EU Commission for the migration problem" (Die Presse, 2014a; authors' translation). Similarly, UK Prime Minister David Cameron claimed that "he shared the deep concerns of many people in Britain at the EU's requirement to lift transitional controls on Romanians and Bulgarians in January" (Mason, 2013). By contrast, only a small minority of MS actors' blame attributions targeted the national-level (13%). German Chancellor Angela Merkel did, however, accuse MSs of bringing the problems of so-called "welfare migration" on themselves. She was cited as saying that "there was a need for legislation but […] this would mainly be at the national-level" (Sparrow & Owen, 2015).
Corroborating the expectations of our model, EU actors' blame attributions (n = 26) are, with only one exception, targeted at MSs (96%). For example, Commissioner Viviane Reding reportedly stated, "that only Britain was to blame for any abuses of the benefit system by European nationals […] [and that] any abuse of benefits by European Union migrants is the fault of the Government which pays out too much" (Waterfield, 2013, p. 12). MEP Manfred Weber likewise blamed the UK: "We need more honesty in Britain. In the course of EU enlargement, it has waived the transitional period of seven years. That is why there are so many citizens from Eastern Europe there today. That was a decision taken in London, not in Brussels." (cited by Süddeutsche Zeitung, 2015, p. 7; authors' translation).

| Comparing cases
Policymakers' blame attributions in the three cases of contested EU migration policies indicate that our integrated model provides a plausible account of multilevel blame games. The combination of policymakers' preferences and their institutional opportunities for attributing blame is not only able to explain the attributions we observed across blame senders within each of the three cases, but also those of each blame sender across the three cases (see Table 2).
As expected, the share of blame attributions targeting the EU is high when the institutional structures of policy implementation allow MS actors to follow their preference for shifting blame onto the EU: MS actors attributed the bulk of blame to the EU in the border control case (77%), where the policy was implemented by EU actors, and in the welfare entitlement case (87%), where the policy does not require active implementation. Furthermore, as expected, the share of blame attributions targeting the EU is medium when policies are implemented by actors located on the same level of government as the blame sender: EU actors' blame attributions in the border control case were moderate with 36% targeting EU actors. Even MS actors' blame attributions in the asylum system case, with 20% blaming EU actors, are generally in line with our expectations. Finally, as expected, the share of blame attributions targeting the EU is low when the institutional structures of policy implementation allow EU actors to follow their preference for shifting blame onto MS actors. Only a few blame attributions by EU actors targeted other EU-level actors in the asylum system case (7%), where the policy is implemented by MSs, as well as in the welfare entitlements case (4%), where the policy does not require active implementation.
However, the share of MS actors' blame attributions to the EU in the asylum system case (20%) is clearly lower than the EU actors' attributions in the border control case (36%). Nevertheless, blame attributions are higher than in any of the cases in which our model predicts the share of blame to the EU to be low. All observed shares of blame attributions targeting the EU thus corroborate the expected rank order. Our model might even offer an explanation of why EU actors' blame attributions to actors at the EU-level in the border control case were higher than those of MS actors' in the asylum system case. As our theory assumes that loyalty and interdependence shape policymakers' blame-shifting preferences, a lower level of loyalty and interdependence among MS actors than among EU actors would explain this difference: MS actors' preference for shifting blame onto actors on another level of government rather than their own might be less strong than in the EU actors' case.
We acknowledge that the inferences from these results are limited. However, the purpose of this plausibility probe was to assess whether the shares of blame attributions targeting the EU that we observed empirically matched the rank order predicted by our integrated model. And the empirical results do indeed corroborate our theoretical propositions. They also match our theoretical expectations when comparing the data from the eight national newspapers and the pan-European one (Politico Europe/European Voice) from which we drew our sample of blame attributions. Despite minor differences, the rank orders are the same and in line with our model (see Table S4 in the Appendix). We are thus confident that the results are not driven by our selection of predominantly North and Central European newspapers. Moreover, the shares of blame attributions targeting EU actors within the cells of Table 2 differ quite significantly from the null hypothesis of an equal distribution of shares across cells. Our confidence that the null hypothesis is wrong is further increased by a chi-square test and the contingency coefficient we calculated in the Appendix (see also Appendix Table S5). While the chi-square test demonstrates that the null hypothesis of a random match can be rejected (99% confidence level), the contingency coefficient implies that there is a meaningful relationship between the dependent variable and the independent variables.

| CONCLUSION
The analysis of policymakers' blame attributions in three cases of contested EU migration policies lends support to our blame-shifting model. This support is further strengthened as this integrated model does better at explaining the observed shares of blame attribution than the isolated models offered by the dominant strands in the literature.
The preference-focused literature expects the shares of blame attributions by EU and MS actors to vary mainly across the two types of policymakers: national governments shift blame onto the EU-level while EU actors blame actors at the national-level. Thus, the distribution of blame should be similar throughout the rows in Table 2. This is not what we observed: MSs' blame attributions targeting the EU vary between 20% in the asylum system case and 87% in the welfare entitlement cases (top row). Similarly, instances in which EU actors blame the EU range from 4% in the welfare entitlements case to 36% in the border control case (bottom row).
The opportunity-focused literature assumes that the structure of policymaking and implementation shapes the shares of EU and MS actors' blame attributions. Since policymaking was constant across the three cases, the implementation structure would, accordingly, be expected to shape policymakers' blame attributions. Hence, the distribution of blame should be similar throughout the columns in Table 2. This is not what we see: in the border control case, attributions blaming the EU vary between 77 % and 36% (column 1), and from 87% to 4% in the welfare entitlement case (column 3). At 20% and 7%, attributions targeting the EU are only roughly similar in the asylum system case (column 2).
In sum, our integrated model does better at explaining the shares of blame attributions in the three cases of EU migration policies than the explanations offered by the existing literature strands. The observed blame attributions in the three cases thus add plausibility to our theoretically coherent integration of the opportunity-and preference-focused literature strands. On the one hand, our model highlights that, due to their loyalty and interdependence, policymakers located on the same level of government tend to share a preference for shifting blame for contested policies onto actors on another level. On the other hand, the model underlines that, due to the complexity of policymaking, policymakers' opportunities to shift blame according to their preferences are typically constrained by the prevalent structures of policy implementation. Together, these specific preferences and opportunities explain why multilevel blame games deviate from the well-known blame games between government and opposition or government and the bureaucracy.
However, as we have provided only a plausibility probe based on three cases of EU migration policies, more rigorous empirical testing of our model is needed. The model still has to demonstrate its explanatory power in policy fields beyond migration such as trade, finance and the environment. Moreover, the assumption should be tested that EU actors' preference for shifting blame onto the national-level holds independent of their nationality (Southern vs. Northern), function (Commissioner or MEPs) and party orientation (left vs. right). In addition, it might be interesting to explore the conditions under which actors blame specific EU actors such as the Commission and when they target "the EU" more generally. Perhaps even more importantly, the model still has to prove its applicability to the MLGSs of federal states such as Germany and international organizations such as the UN. Is it true beyond the EU that loyalties and interdependencies among actors of the same level tend to be stronger than loyalties and interdependencies that link actors across different levels of government (for instance when they are members of the same political party)? While the application of the model to other issue areas and other MLGSs might prompt its modification, we are, nevertheless, confident that it contributes to a better understanding of blame attributions in MLGSs.

ACKNOWLEDGMENT
We wish to thank the three anonymous reviewers, the editor, as well as the participants of the research colloquium at LMU Munich, the ECPR SGEU Conference in Paris in June 2018 as well as the biannual conference of EUSA in Denver 2019 for their constructive comments. We are specifically grateful to Felix Biermann, Catherine de Vries, Sara Hobolt, Maria Mexi, and Lisa Kriegmair. We also would like to thank Louisa Klein-Bölting, Juliane Glovania, and Severin Süß for excellent research assistance.

ORCID
Tim Heinkelmann-Wild https://orcid.org/0000-0003-2699-6048 ENDNOTES 1 Sometimes political actors even create new actors to whom they delegate policy tasks so that they can shift the blame onto them in cases of policy contestation (Hood, 2011;Mortensen, 2016;Tallberg, 2002).  Roose, Scholl, and Sommer (2016), and Schlipphak and Treib (2017). 3 Alcañiz andHellwig (2011), Anderson (2000), Arceneaux (2006), Cutler (2008), De Vries, Edwards, and Tillman (2011), Gailey (2013), Gailey and Lee (2005), Hamilton (1986), Hellwig and Samuels (2008), Hobolt and Tilley (2014), Hobolt, Tilley, and Banducci (2013), León (2011León ( , 2012León ( , 2018, León, Jurado, and Garmendia Madariaga (2018), León and Orriols (2016), Malhotra and Kuo (2008), Mortensen (2013a, 2013b), Nollkaemper (2018, Powell and Whitten (1993), Rittberger, Schwarzenbeck, and Zangl (2017), Rudolph (2003), and Wilson and Hobolt (2015). 4 Moreover, EU actors are the ideal scapegoats for MSs' blame-shifting strategies, as they are both less willing and less able to advocate for themselves than their fellow MS actors. On the one hand, as EU actorsespecially the Commission and the Council, though not the Parliament-are much less dependent on voter support than national governments or parliaments, they are less likely to try to shift blame back onto MS actors. Thus, in contrast to their national counterparts, they are more willing to accept the blame shifted onto them. On the other hand, as they suffer from the absence of a European public sphere, EU actors are less able to shift blame back onto MSs. National governments are usually able to make their positions clear to their national publics; EU actors typically have a harder time communicating to the publics of the MSs (Gerhards et al., 2009, p. 550). 5 In domestic MLGSs party loyalty across levels is typically stronger than in the EU and might, at times, overwrite loyalty among actors from the same level. We thus acknowledge that for domestic MLGSs the resulting loyalty conflicts should be theorized when adding complexity to our-so far-parsimonious model. 6 See endnote 3. 7 Implementing actors-or actors on the level that has the responsibility to implement policy-might still try to follow their preference for blaming another level of government. Yet, they are likely to learn that their blame attributions lack the required plausibility and to adjust their attributions accordingly. 8 Nota bene, conceptually our propositions are concerned with relative ranks (and thus a rank order) rather than absolute shares of blame attributions targeting the EU. We thus do not expect high, medium and low shares (which could be defined in fixed percentages) but comparatively higher and lower shares (which are defined in higher and lower percentages). Yet, for reasons of convenience and comprehensibility, we use substantive labels for the ranks and dub them "high," "medium" and "low." 9 For example, Gerhards et al. (2009), Rittberger et al. (2017, and Schwarzenbeck (2017). 10 We opted for an analysis of blame attributions in quality newspapers rather than in policymakers' speeches (or press releases) to make sure that we studied attributions that made it to the general public. 11 In the Appendix, we provide the detailed coding rules (see also, Appendix Table S2). To ensure reliability, we also conducted an intercoder test between the two coders and calculated their concordance (see Appendix  Table S3). 12 The coded data is available at: https://data.ub.uni-muenchen.de/169/ 13 Nota bene, we only speak of blame-shifting, if the distribution of blame attributions (across actors and cases) follows the expectations of our model. If our model is falsified, policymakers would still attribute blame, but they would not be following a blame-shifting logic. 14 Othmar Karas, MEP, cited by Gabriel (2015); authors' translation. 15 Rebecca Harms, MEP, cited by Bremer (2013, p. 1); authors' translation.