“Take back control”? The eﬀects of supranational integration on party-system polarization

In this paper, we examine the relationship between supranational integration and domestic party-system polarization (extremism). We first construct a theoretical argument that uncovers the key trade-off between the “output legitimacy” of a supranationally integrated party system and the inevitable loss of “input legitimacy” caused by externally imposed policy constraints. This translates into a strategic tradeoff between responsibility and responsiveness at the party level of electoral competition. We hypothesize that while moderate supranational policy constraints can initially speed-up platform convergence, ever-closer political integration may reverse the trend towards higher levels of party-system polarization and party extremism. We apply our framework to the case of EU integration and test our key non-monotonic prediction both at the party-system level of polarization and at the party level of ideological extremism. Finally, we apply to synthetic control method (SCM) for causal inference in comparative case studies to study how political integration and supranational policy constraints have affected their overall level of party-system polarization over time. Our overall empirical analysis strongly corroborates our theoretical argument. into force. In corroboration


Introduction
Advanced liberal democracies have entered a period of political turbulence where the processes of economic globalization and supranational integration are affecting the core of domestic government and politics by creating deep fissures in cabinets, parties, and society at large. Evidence abounds of trends such as party-system dealignment and weakening of partisanship throughout the latest era of globalization (Przeworski 2019). "Brexit" in the UK, Trump's victory in the US, and the contemporaneous rise of right-wing nativist populism in the European "North" and left-wing redistributionist populism in the European "South" can all be viewed through the same prism of a populist backlash against the "straitjacket" of economic globalization and political integration (Algan et al. 2017;Autor et al. 2017;Colantone and Stanig 2018;Guiso et al. 2017;Rodrik 2017). Effectively, the common thread that connects all these electoral results is that of increasing party-system polarization and party extremism stemming from the increasing "sovereignty costs" of supranational integration and economic interdependence (Copelovitch and Pevehouse 2019).
Political scientists have shifted their attention to the study of political polarization because, even if policy outcomes remain the same, higher polarization implies heightened political uncertainty for citizens -albeit together with better defined political platform choices -as well as increased political country risk for foreign investors (Beazer and Blake 2018). With polarization, legislative politics become increasingly adversarial, (coalition) governments more unstable, and policy outputs potentially more volatile. What is more, domestic political instability may trigger a backlash against international unions and organizations -such as the European Union (EU) -as increased polarization along the pro/anti-integration axis may generate negative political spillovers for supranational governance: policy deadlock, the rise in the political implementation costs of policy reforms, the undermining of public support (Bearce and Jolliff Scott 2019), or even the political unraveling of the integration process itself (Hobolt and de Vries 2016;Obermeier 2018).
Political polarization may manifest itself at various stages of the domestic political process: in elections (Cox 1990), the government formation process (Indridason 2011), legislative bargaining (McCarty et al. 2006, and political communication (Prior 2013). 1 In the context of US politics, political polarization is tantamount to the ideological differentiation between the two main congressional parties (Layman et al. 2006). However, when it comes to parliamentary multi-party systems, the concept of polarization is not as straightforward in terms of measurement as it needs to account for both ideological differentiation (also in the form of political extremism) and party fragmentation, whereby niche groups of extremist ideologues may splinter off from larger office-seeking parties (Ezrow 2010). 2 In both contexts, much of the existing literature has studied the effects of constitutional rules and institutions, such as the (dis)proportionality of the electoral system (Calvo and Hellwig 2011;Curini and Hino 2012;Dow 2011;Ezrow 2008;Iaryczower and Mattozzi 2013;Matakos et al. 2016), gerrymandering (McCarty et al. 2009), and primary elections (Hirano et al. 2013), on party-system compactness, polarization, and extremism.
We focus instead on how extraneous policy rules and constraints -stemming from a country's level of political integration within regional blocs in addition to its economic integration within global markets 3 -affect party-system polarization and party-level platform divergence within the context of electoral competition. To illustrate our argument, we choose to focus on the EU as a preeminent example of a regional experiment in supranational integration that has given rise to a thick web of policy rules and regulations effectively constraining policy-making at the national level (sovereignty costs). In theorizing about this relationship, we highlight the tradeoff between the "inputs" of democratic representation and the "outputs" of material compensation (Frieden 2019) as the key determinant of the effects of supranational integration on domestic party systems. 4 More specifically, we show that a polarization backlash may ensue at the point where the "output legitimacy" of an open and supranationally bound domestic political system is not sufficient to counterbalance the inevitable loss of democratic responsiveness caused by externally imposed constraints on its domestic sources of "input legitimacy".
In our theoretical section, we provide an informal argument that first expounds this implicit trade-off between input and output legitimacy (or else representation vs. compensation) at the party-system level and then projects this trade-off onto the mixed incentives of parties strategically choosing a convergent or divergent policy platform in order to achieve the optimal balance between maximizing their chances of electoral success and minimizing their (ideological) "transportation costs". More specifically, we build on Schneider's (2019) findings with regard to the two distinct objects of electoral accountability in EU party systems: diffuse support for the project of European integration -which is arguably higher (lower) during times of growth and prosperity (recession and secular stagnation) -and spatial preferences for specific policies. 5 In particular, parties on the two sides of the ideological spectrum face the following trade-off: they can either tap into diffuse support for European integration by proposing platforms in line with the EU's policy constraints at the risk, however, of alienating their core ideological base (high transportation costs), or they can choose 3 In fact, Autor et al. (2017) find increasing support for nativist or extreme politicians in congressional US districts exposed to larger increases in import penetration. 4 Although we do account for the potential spillover effects of domestic-level polarization on supranational political processes in our empirics, our theoretical argument amounts to a partial equilibrium analysis that takes the process of integration deepening as exogenous and assumes that there is always a majority of voters in favor of integration so that in effect adherence to the imported policy constraints is never really in question. 5 Note that here we assume that diffuse support for the EU per se is uniform across the electorate. We effectively distinguish between pro-EU, Euroskeptic, and Europhobic voters with regard to their individual ideological preferences and whether those are compatible with the EU's policy agenda. In other words, diffuse support for the EU acts as an orthogonal valence factor on voter preferences.
to "stick to their guns" and propose Euroskeptic policy platforms in contravention of the EU's rules and regulations, thereby minimizing ideological transportation costs but also foregoing the electoral valence benefits of diffuse support for the EU. Therefore, we expect that, for low and moderate levels of supranational integration, i.e., slacker constraints on the domestic policy space and hence lower transportation costs, extreme parties will find it profitable to moderately converge and propose platforms that are tangentially congruent with EU policy rules and directives. However, as supranational integration intensifies and, hence, the permissible policy space shrinks, the rise in transportation costs will eventually trump the valence benefits of congruence with the European acquis. If that is the case, extreme parties will become better off pandering to their core supporters by proposing a Euroskeptic (non-congruent) platform that minimizes the distance away from their core ideology. As a result, our theoretical analysis generates the key prediction of a non-monotonic effect on aggregate polarization at the domestic level: the gradual deepening of supranational integration will at first lead to increasing platform convergence but, if it goes too far in terms of truncating the permissible domestic policy space, it can also generate a strong populist backlash in the form of extreme party-system polarization and fervent anti-globalism.
To corroborate our theoretical argument, we use time-series cross-section data to examine the effect of a country's legal status vis-à-vis the EU (formerly the European Community) -i.e., whether it is an associate member, a candidate member, or a full member -on its domestic party system. We control for the major treaties and time periods of European integration to account for the effect of deepening commitments, rules, and regulations as part of a shifting acquis. In accordance with our two-level theoretical argument, we conduct different empirical tests using two distinct panels: a first one at the party-system (country) level and a second one at the party level. Both of our empirical analyses highlight the non-monotonic effect of integration on party-system polarization and party-level extremism. In fact, the party systems in our sample tend to experience this "breaking point " before the Global Financial Crisis and the EU's "multiple crises" (i.e., debt, migration, populism, Brexit). In a similar vein as Frieden (2019), we show that the current state of affairs is simply the culmination of pre-crisis trends, while the recent crises simply exacerbate them further.
Moreover, by way of correcting for the potential endogeneity of integration decisions and post-treatment bias effects, 6 we perform robustness tests using the synthetic control method (SCM) (see Abadie and Gardeazabal 2003;Abadie et al. 2010;and Abadie et al. 2015) that relies on non-EU member-states as counterfactual cases. This quasi-experimental comparative case study method also allows us to illustrate the explanatory power of our argument vis-à-vis alternative explanations for long-term trends in party-system polarization. All in all, we identify the behavior of parties and 6 Note that in our theoretical analysis we do not seek to explain why countries integrate further per se. We surmise that higher levels of economic and political integration come about as a result of pro-active policies of market liberalization, openness to trade, and the extraneous pressures of globalization (e.g., capital flows, immigration flows, environmental externalities, etc.). the characteristics of party systems as intervening variables that capture the underlying relationship between the input and output legitimacy of national democracies in a interdependent political and economic environment.
With this paper, we also seek to contribute to the literature on the convergence of national party systems and the transformation of electoral competition as a consequence of regional integration (Dorussen and Nanou 2006;Hix 2003;Hobolt and Tilley 2016;Mair 2000). Nanou and Dorussen (2013), for example, find that, because EU legislation limits the set of policies that parties can pursue once in government, the distance between parties' positions has tended to decrease in policy domains where EU involvement has increased. We actually expand on their results by taking a systemic view of party-system polarization -thus obviating the need for an arbitrary distinction between mainstream and non-mainstream parties -and by deriving the conditions for both conditional convergence and divergence of party platforms. In this regard, the paper also contributes to the vast literature on spatial (voting) models of party competition, which tends to focus on the electoral consequences and longterm impact of parties' ideological shifts (see, e.g., Adams and Somer-Topcu 2009;Ezrow 2005Ezrow , 2008Hellwig 2012;Karreth et al. 2013;Williams 2015), but does not always address the question of why they move to the center or to the margins.
In what follows, we first provide a theoretical discussion of the trade-off faced by advanced liberal democracies between input and output legitimacy in the face of political (and economic) integration. We then delve into that trade-off at the party level by elaborating on the theoretical mechanism of ideological transportation costs and showing that binding policy constraints will induce platform convergence up to the point where internal pressures for representation and input legitimization by partisan militants and core voters will cause divergence back towards the party's intrinsic ideology. Subsequently, we apply our argument to the EU context, test the theorized effects of European integration on European parties and party systems using different measures of party-system polarization and party-level extremism, and present strong evidence confirming our hypothesized non-monotonic relationship between supranational integration and party-system polarization. Finally, we conduct some robustness tests using the synthetic control method and proceed to discuss some comparative case studies. Alternative robustness tests are presented in the Online Appendix available on the Review of International Organizations' website.

The input and output legitimacy of party systems
In the current era, politics is no longer "business as usual" and that applies not just to the supply side of electoral competition and policy formation but also the demand side of partisanship and electoral behavior (Guiso et al. 2017). Recent electoral contests in Western Europe, Latin America, and the US showcase the "political limits of globalization" (Acemoglu and Yared 2010;Rodrik 2012) and the fragility of the modern liberal world order. Arguably, the key factor that has introduced "noise" and unexplained volatility into the post-war paradigm of output legitimacy and "embedded liberalism" (Ruggie 1982) has been the corrosive effect of globalization on democratic politics in the form of heightened sovereignty costs (Copelovitch and Pevehouse 2019).
Output legitimacy effectively pertains to the notion that a democratically elected government should "care for the common good" by pursuing economic efficiency and material prosperity. As such, it can potentially come at the expense of so-called input legitimacy, which amounts to a government's need to "carry out the will of the people" through representative democratic processes and institutions (Scharpf 1999). The two may be at odds when a government decides to pursue a policy that in its own assessment would produce the best functional outcome for society as a whole -even if the electorate would opt for a different policy direction -, resulting in a tension between efficiency and democratic representation (Eichengreen and Leblang 2008). 7 Policy responses to the recent Global Financial Crisis and the Eurozone Debt Crisis accentuated the growing incongruence between the democratic functions of "representative" government "by the people" and those of "responsible" government "for the people" (Mair 2009). Executive dominance now seems to characterize the predominant current of politics in advanced democracies as incumbents have come to prioritize the pursuit of economic outputs over political inputs (Scharpf 2011). In contemporary democracies, responsible (or output-oriented) government has taken prevalence over representative (or input-oriented) government, thus fueling the rise of populist parties that claim to still represent the "will of the people" but rarely deliver, as opposed to mainstream parties that take responsibility but no longer seem to voice their voters' concerns (Caramani 2017).
The output legitimacy of the post-WWII liberal economic order emanated from a long period of prosperity and economic growth as well as the establishment of efficient compensation mechanisms of social welfare for those left behind (Burgoon 2001;Walter 2010). National politicians were able to steadily craft a well-entrenched consensus over the desirable levels of integration within supranational governance structures and openness to global markets whilst at the same time allowing enough room for traditional left-right political vacillation within moderate levels of polarization. As countries started to integrate further both politically and economically, the menu of feasible and credible policy platforms became increasingly constrained by the exigencies of economic openness and supranational integration (sovereignty costs), thereby limiting the sources of input legitimacy and democratic representation. In other words, the pursuit of efficiency by means of economic integration, open borders, and rule harmonization started to trump popular demands for democratic institutions of regulation, stabilization, and social protection (part of the traditional agenda of the socialist left) and undermine ethnic homogeneity and national selfdetermination (part of the traditional agenda of the nationalist right). The rise in capital mobility and the concurrent emergence of a certain market-enforced discipline and orthodoxy in macroeconomic policy in the form of the "Washington Consensus" and the "Golden Straitjacket" began to crystallize the informal (soft) constraints of economic interdependence on the mainstream policy spectrum (Mosley 2005). In addition, by joining (or aligning with) international organizations and unions like the EU, the European Monetary Union (EMU), and the World Trade Organization (WTO), countries committed to formal (hard) policy constraints that truncated their domestic policy space (sovereignty costs), limited what parties could credibly promise in elections and deliver in government and, thus, curtailed their "room to maneuver" (Hellwig 2016;Kosmidis 2018).
For as long as domestic political systems remained embedded within a broader liberal democratic consensus the enhanced output legitimacy of liberal democratic politics induced parties to keep converging towards a set of policy inputs as circumscribed by the extraneous policy constraints of political feasibility, economic orthodoxy, and formal international commitments -albeit at the expense of unfettered democratic responsiveness. Eventually, however, this failure of representation came to be compounded by a failure of compensation for a large part of the population, a trend that was further exacerbated by the devastating effects of the Global Financial Crisis and the Eurozone Debt Crisis on the economy and the welfare state (Frieden 2019). There came a point in a number of advanced industrial democracies where populist leaders (of both the redistributionist left-wing and the nativist right-wing types) inveighing against the globalism of liberal elites and supranational bureaucrats started to gain in popularity and a populist anti-globalization agenda started to emerge as a credible alternative to the "there-is-no-alternative" (TINA) politics of globalization.
In other words, there appears to exist a "breaking point" beyond which the overall level of output legitimacy of an open and liberal democratic system is not sufficient to compensate for the loss in input legitimacy and responsiveness, thus causing a democratic backlash in the form of increased party-system polarization. The implicit trade-off between political inputs and economic outputs gets reassessed, thus giving rise to a sort of political atavism by established left-and right-wing parties reverting to their ideological roots or the emergence of new parties on the extremes of the ideological spectrum. This is due to the heightened pressure for popular representation by an increasingly disenfranchised and disenchanted part of their core electorate and the unwieldy nature of policy compromises necessary to maintain a country's liberal orientation.
This threshold point may be brought about by stochastic shocks to the aggregate welfare benefits of integration (e.g., in the context of capital account liberalization and financial crises), excessive constraints on the core democratic functions and policy sovereignty of a country, or both. Countries further integrated within global markets (mostly due to the small relative size of their domestic markets) and systems of supranational governance (often due to the political spillovers generated by the unfettered flows of capital, people, and ideas) will be subject to narrower constraints on their domestic sources of input legitimacy. However, higher levels of integration also imply higher exposure to exogenous macroeconomic shocks and risks (Kim 2007), i.e., more noise in their political system's overall level of output legitimacy. Therefore, as countries integrate further both economically and politically, they become more likely to move beyond the aforementioned breaking point and experience heightened political volatility and democratic instability in the form of increased party-system polarization.
The ensuing backlash towards more extreme core electoral platforms -now effectively incompatible with the country's policy constraints -will render the pro-/antiglobalization dimension of electoral competition much more salient. Since the rise in platform polarization may run counter to the policies imposed by the "Golden Straitjacket" of economic globalization and political integration, then the expediency of continued membership in international organizations and the desirability of a country's overall liberal orientation may be called into question, thereby further politicizing the pro-/anti-globalization axis (Kriesi 2016). Therefore, even if actual policy outcomes and existing integration trends are not to be severely affected by the rise in platform polarization, they still become increasingly tenuous and fragile.
As we will proceed to demonstrate empirically, we also claim that political integration by means of rules harmonization and policy coordination generates quantitatively and qualitatively distinct effects on national party systems vis-àvis economic globalization, i.e., current and capital account openness writ large. Whereas economic globalization generates "narrow-but-strong" informal constraints -such as for example the Mundell-Fleming "impossible trinity" of monetary independence, financial openness, and exchange rate stability -, political integration produces "broad-and-strong" formal constraints on macroeconomic (monetary, fiscal, and trade) policy targets. There is often an explicit (hard) process of rule enforcement and policy censoring that comes about with political integration while, on the other hand, the constraints imposed by economic openness tend to be fuzzier (soft). Moreover, political integration often subsumes economic integration since supranational integration amounts to the deepening of economic relations among regional partners that are already assumed to be "natural trading partners" (De Nardis and Vicarelli 2003;Micco et al. 2003). Finally, with political integration one can actually put a face on those external sources of output legitimacy, which in turn also makes it easier for domestic actors to shift blame onto them (Schneider 2019); even so, political trust can be rather easily rebuilt among national politicians (and their successors). On the other hand, markets are impersonal and the loss of market confidence can be abrupt and very durable.

The responsibility-responsiveness trade-off of parties
In this section, we further develop our theoretical argument with regard to the relationship between the scope of supranational policy constraints imposed on multiparty parliamentary democracies and their party-system polarization (extremism). 8 In order to do so, we extrapolate from the party-system level onto the party level and elaborate on the strategic trade-off faced by parties in the electoral arena between responsibility and responsiveness.
Each party is associated with a core ideological position in the left-right spectrum and runs on a certain platform within a strategic game of electoral competition. On one hand, parties care about maximizing their vote share, which may be regarded as a rough -but fair -approximation of the probability that they enter government and, hence, their ability to acquire office rents. 9 On the other hand, parties also seek to minimize the costs incurred by proposing a platform away from their core ideological position. These are essentially "transportation" (communication) costs that parties must incur in order to re-brand themselves and sell their new policy program to their voters or otherwise to appease their more militant ideological factions,they are, thus, strictly increasing in the distance between the proposed platform and the party's core ideological position. 10 Based on the above, we effectively assume that party behavior is driven by both office-seeking incentives of vote-maximization and electability and ideological motives of responsiveness and rhetorical consistency with core party positions. 11 In an unconstrained policy environment non-centrist parties face the following trade-off: moving towards more moderate (centrist) positions increases not only their vote share and, hence, their expected office rents but also their transportation (communication) costs. That is, in the absence of any other changes in the institutional set-up (e.g., changes in the electoral rule), the degree of policy convergence towards the center will depend on the relative size of the spoils of office (centripetal force) and the transportation costs (centrifugal force). Thus, this trade-off will be reflected by some limited convergence towards the center of the policy space up the point where spatial transportation costs net of vote share gains are zero (see Fig. 1).
But what happens when additional policy constraints are imposed as a result of a country's participation in supranational integration schemes (in addition to its overall level of economic openness)? Are the incentives faced by extreme parties to converge towards the center of the policy spectrum amplified or mitigated? To answer this question, we need to shed more light on the mechanism at play. Membership in an international organization such as the EU can impose hard constraints on the domain of permissible domestic policies (e.g., in the form of harmonization of taxation and welfare spending) and effectively limit national governments' "room to maneuver" (Hellwig 2016). We assume that the constrained set excludes policies too much to the left or the right of the policy spectrum. How will then extremist parties behave if those restrictions start to bind (i.e., if the policy platforms that they would have proposed in an unconstrained environment fall outside the permissible policy space)? Will they further converge towards the center in order to abide by the constraints or will they continue to advocate their core policies in the knowledge that they are not compatible with the country's overall level of political (and economic) integration?  Fig. 1 How the responsibility-responsiveness trade-off shapes non-centrist parties' incentives to converge or diverge. The horizontal axis depicts the left half of the traditional left-right spectrum while the vertical axis depicts spatial and valence (orthogonal) costs and benefits of different ideological platforms for a left-wing party. Its optimal unconstrained electoral platform occurs at the point where its ideological transportation costs net of vote share gains are zero. When supranational policy constraints are relatively slack (left bound A), the valence benefits of respecting those constraints outweigh the net spatial costs, thus inducing the party to converge towards the left bound. On the other hand, when the supranational policy constraints are relatively tight (left bound B), the opposite is the case and the left party will revert back to its extreme unconstrained position. The threshold point between decreasing and increasing polarization in terms of the slackness of constraints (depth of political integration) occurs somewhere in-between and is a function of the party system's input and output legitimacy. Clearly, the higher the orthogonal valence benefits are, the tighter the policy constraints need to be for the non-monotonic effect to kick in (threshold point mores closer to the center) In accordance with our theoretical distinction between inputs and outputs, we purport that when the electoral policy space is constrained, parties effectively compete along two dimensions: the traditional left-right ideological spectrum (representative inputs) and a second (binary) dimension that captures whether a party runs on an electoral platform that is compatible with the supranational union's rules and regulations (responsible outputs). We assume that while positioning along the traditional input-related dimension of electoral competition generates spatial benefits and costs in terms of vote share and transportation costs, posturing along the second outputrelated dimension generates orthogonal valence (popularity) payoffs as a function of the overall output legitimacy of the party system and the country's extant level of political (and economic) integration. When output legitimacy is high, a party that proposes a responsible electoral platform that complies with the constraints imposed by supranational bureaucrats (and markets) will be rewarded with a high positive valence (popularity) payoff; when output legitimacy is low, the corresponding payoffs will also be low. By contrast, a party that chooses to ignore international commitments (and economic orthodoxy) by proposing extreme non-permissible policy platforms will forgo those valence payoffs. In other words, this formulation of preferences (a simplification of a standard valence model) introduces a potential trade-off between responsibility in terms of respecting the exigencies of political (and economic) integration -thereby enhancing the party's electability and international credibilityand responsiveness in terms of pandering to core voters' and militants' ideological preferences -thereby exhibiting rhetorical consistency and ideological purity.
Therefore, when it comes to their electoral strategy, parties face the dilemma of whether to propose a policy platform on the boundary of the constrained policy space or stick with their unconstrained platform proposal; anything in-between would effectively amount to a strictly dominated strategy. Clearly, non-centrist parties' decision to converge or not will depend on (a) the tightness of such constraints on the domestic sources of input legitimacy and (b) the overall level of output legitimacy of a party system subject to a given level of supranational integration. On one hand, the tightness of the constraints creates centrifugal incentives by raising the transportation cost of convergence while, on the other hand, higher output legitimacy creates centripetal incentives by raising the valence (popularity) benefits of advocating a moderate platform that is tangentially compatible with the country's international commitments. Therefore, any shift away from the optimized unconstrained tradeoff between vote share maximization and transportation cost minimization will only occur if and only if the orthogonal valence benefits of advocating integration-friendly policies (and the corresponding gain in votes) outweigh the additional ideological transportation costs imposed by binding supranational policy constraints. Figure 1 below provides a graphical demonstration of the argument in the case of a left-wing party.
Clearly, the larger the electoral benefit of conforming to the rules and regulations of the international organization, the slacker the responsibility-responsiveness tradeoff, allowing conditional platform convergence to be sustained even as integration deepens further and policy constraints become tighter. However, there is a threshold value of the slackness of supranational constraints (input legitimacy) as a strictly decreasing function of the valence benefits of pro-membership policies (output legitimacy) beyond which the costs of convergence to more moderate and responsible positions can outweigh the benefits for non-centrist parties (while moderate parties will converge towards centrist TINA policies even further). 12 If, as argued earlier, we take the valence (popularity) benefit to be stochastic, then we should expect countries subject to narrower constraints of integration (due to small size, explicit policy conditionality, or lack of political clout in the policy-formation process) and exposed to more volatile macroeconomic shocks or business cycles to be more prone to go past this breaking point. This result can be summarized in the following two hypotheses, one at the party-system level and one at the party level:

Hypothesis 1 (H1) For any given orthogonal valence benefits of responsible policy platforms that respect supranational policy constraints (output legitimacy), the relationship between a country's depth of political integration -captured by the tightness of supranational constraints on the electoral spectrum of domestic party competition (input legitimacy) -and party-system polarization (extremism) is effectively nonmonotonic.
For low initial levels of political integration, party-system polarization (extremism) is expected to decrease as a result of tightening supranational policy constraints; however, there is a threshold point of integration deepening beyond which tighter supranational policy constraints are not expected to induce further policy convergence but rather to trigger a rise in party-system polarization (extremism).
Hypothesis 2 (H2) As a political system starts to integrates within a supranational union that harmonizes common policies and limits the range of permissible policies, non-centrist parties are initially expected to converge towards a narrower range of electoral platforms compatible with the country's international commitments; however, there is a threshold point in a country's level of political integration beyond which tighter supranational policy constraints are expected to induce non-centrist parties to revert back to their unconstrained core positions (and possibly moderate parties to converge even further towards responsible centrist positions). 13 In sum, our results shed light on how the process of supranational integration and the relative levels of input and output legitimacy of a domestic party system expose parties to a substantive trade-off between responsibility in terms of abiding by the country's international commitments and responsiveness in terms of truly representing the ideological preferences of core voters and party activists. Relatively low levels of supranational integration with relatively slack policy constraints (i.e., low sovereignty costs) intensify the process of electoral platform convergence and reduce party-system polarization (extremism) since extreme parties' gain in valence benefits (and possibly vote share) is high enough to compensate for their loss in terms of transportation (communication) costs. Yet, as the level of integration exceeds a certain threshold value and these policy constraints become tighter (i.e., high sovereignty costs), the situation can backfire leading to a reverse trend of platform divergence and increasing polarization. Simply put, the corresponding loss in responsiveness (as proxied by the increase in transportation costs) associated with further platform convergence to the union's acquis becomes too high relative to the valence gains of responsible posturing so that non-centrist parties are better off proposing a policy platform that reflects their ideological roots and caters to their core electoral constituency. This dynamic in electoral party behavior gives rise to the following non-monotonic relationship: while at first the tightening of relatively permissive supranational constraints is expected to lower party-system polarization (extremism), eventually ever closer integration and extremely narrow constraints are all else equal bound to generate increasing levels of party-system polarization (extremism). This is the key prediction of our theoretical analysis, which we set out to test in the following section.
Although we have been focusing so far on the convergence and divergence of electoral party platforms, we still expect government policy to remain at the prointegration center of gravity of the left-right spectrum -whether that be the position of the median party, the parliamentary mean, or the expected dominant party selected to form a government (Merrill and James 2007). Nevertheless, from the point of view of moderate voters, market participants, and even supranational principals -and for any degree of risk aversion -, there is a clear welfare ranking among mean-preserving electoral outcomes with respect to other moments of the distribution such as policy variance. For any generic specification of the government and policy formation process in parliamentary multi-party systems, platform convergence by non-centrist parties will generally imply less policy variance, and vice versa. Therefore, platform convergence and decreasing party-system polarization as a result of moderately increasing levels of integration imply decreasing policy variance, increasing government stability, but also lower responsiveness to the ideological preferences of extreme voters. However, once the non-monotonic (positive) effect of ever deeper levels of integration and tighter supranational constraints on party-system polarization (extremism) kicks in, we would then expect higher policy variance, increasing government instability, but also higher levels of democratic responsiveness across the spectrum of voters.

Supranational integration and the polarization of European party politics
In this section, we seek to probe our hypotheses on the non-monotonic relationship between supranational integration -insofar as it comes with extraneous constraints on the set of permissible domestic policies -and party-system polarization in the context of European multi-party parliamentary democracies. No other region has integrated politically further than Europe -albeit exhibiting at the same time substantial levels of political and economic diversity -which renders it a most suitable and empirically rich setting for a cross-sectional and intertemporal study of the hypothesized relationship.
The European integration process progressed by leaps and bounds on the basis of its output legitimacy as a successful technocratic project generating growth and stability across Europe. Especially during the early years of the Single Market and the euro (late '80s to early '00s) governments seemed secure in the output legitimacy of the multi-layered European structure as it was perceived as a bulwark of economic stability and sustained growth. Output-driven notions of democratic accountability and economic voting allowed moderate pro-EU parties to reap the valence benefits of EU membership and trust in EU institutions.
At the same time, the broadening and deepening of the supranational integration process -through successive intergovernmental EU treaties -reshaped the contours of domestic electoral competition by tying a number of policy areas together as part of the deeper commitments of EU membership, shrinking the degrees of freedom of electoral competition, and constraining the range of permissible and internally consistent policy platforms. Such hard constraints on the traditional left-right policy spectrum stemmed from (i) formal or informal intergovernmental commitments of EU or EMU membership (e.g., the fiscal coordination mechanisms of the Stability and Growth Pact, the European Semester, and the Fiscal Compact), (ii) enhanced monitoring and enforcement mechanisms (e.g., the Six-Pack and Two-Pack regulations), (iii) the hard process of compliance, transposition, and implementation of EU rules and directives (e.g., the Excessive Deficit Procedure governed by Article 126 of the Treaty on the Functioning of the European Union, Article 7 of the Treaty of the European Union, or the European Court of Justice), and (iv) (more recently) highpowered bailout conditionality agreements, such as the Memoranda of Understanding between the European Commission (EC), the European Central Bank (ECB), and the International Monetary Fund (IMF) (collectively known as the European Troika) and the indebted countries of the European South.
As of late, however, this technocratic process of "integration by stealth" (Majone 2005) has come under intense scrutiny as a result of stagnating labor productivity, sustained austerity, rising unemployment and debts levels, Brexit, and growing levels of third-country immigration. These "multiple crises" have sapped the EU's output legitimacy without so far having triggered any significant counterbalancing measures enhancing the Union's so-called input legitimacy. The prevalence of "emergency politics" (White 2013) and the institutional power transfers towards national and supranational executives that took place in response to the Eurocrisis (Bauer and Becker 2014; Rittberger 2014) have prioritized economic outputs at the expense of political inputs. Schmidt (2013) captures this nicely by pointing towards a precarious political imbalance: whereas output legitimacy (or "governing for the people") has largely come to reside at the European level -i.e., the main locus of "policy without politics" -, input legitimization (or "governing by the people") still takes place primarily at the domestic level -i.e., the main locus of "politics without policy". This structural disconnect between those (supranational) institutions that determine policy and those (national) institutions that shape politics has given rise to the EU's so-called "second-order democratic deficit" as deeper political and economic integration has come to be associated with the populist backlash and the "hollowing-out" of national democracy (Konstantinidis and Treurniet 2018).
In the empirical analysis that follows we proxy for these extraneous policy constraints through countries' legal status vis-à-vis the EU (or else the degree to which they are bound by the European acquis), controlling for the overall depth of integration over successive EU treaties. We also expect the tightness of these constraints (or else the degree of sovereign policy discretion) to vary with respect to countries' economic size (GDP), their (formal or informal) influence in the EU policy-formation process (as captured by population size), and the output legitimacy of overall macroeconomic conditions (e.g., GDP growth). In addition, we control for countries' capital account openness in order to account for the softer imperatives of financial liberalization and economic globalization at large. If our theory holds water, non-centrist European political parties should have converged towards the center (decreasing party-system polarization) during the early to middle stages of European integration (e.g., Treaty of Rome, Single European Act) and diverged towards the extremes (increasing party-system polarization) during its latter stages (e.g., Maastricht Treaty, Lisbon Treaty, Fiscal Compact).

Data and empirical model at the party-system level
In order to estimate the impact of supranational integration on party-system polarization (Hypothesis H1), we collected data measuring parties' policy positions from the most recent Manifesto Project data set (Volkens et al. 2018) for twenty-eight EU countries between 1957 and 2015. 14 This data set contains information on the leftright positions of political parties as stated in their electoral manifestos and relies on saliency theory by estimating parties' position on the basis of both positive and negative emphases of issues. Ideological scores range from -100 to +100 where higher values indicate a more right-wing orientation.
We use this data set to code the following measures of polarization as our main dependent variable: (a) the distance between the two most extreme platforms (MDP) (Budge and McDonald 2006;Andrews and Money 2009;Matakos et al. 2016) and (b) the Dalton index (2008). 15 In this regard, party-system polarization and extremism are positively and monotonically related. Higher values of our dependent variables denote higher levels of party-system polarization on the left-right dimension.
Our main explanatory variables are EC/EU Integration Status and EC/EU Integration Level, both coded from the Enlargement Database (ENLABASE) (Schimmelfennig 2002(Schimmelfennig , 2016. 16 EC/EU Integration Status indicates the legal status of a country vis-à-vis the EU at the end of each year (based on specific events such as the signing, ratification, and entry into force of treaties, etc.) and ranges from 0 (no special status) to 47 if a state is integrated to the highest possible degree at a given time, where the "highest degree" of integration changes with the evolution of the organization. In other words, it is a fine-grained measure of "graded membership" and captures countries' proximity to the EU's frontier of integration (Schimmelfennig 2002(Schimmelfennig , 2016. So, for example, a core EU member-state like Germany that has always been at the forefront of the integration process always gets the highest score of 47 in the data set. EC/EU Integration Level is a cruder categorical variable derived from EC/EU Integration Status that classifies countries' legal status vis-à-vis the EU into the following categories: "No Special Status" (0), "Collaboration" (1), "Association" (2), "Partial 14 Data availability for each country starts around the time they become full-fledged parliamentary democracies. 15 The formula proposed by Dalton (2008) to operationalize party-system polarization is as follows: where v is party j 's vote share, x is party j 's proposed platform,x is the weighted mean of all parties' position, and N is the number of parties. 16 For an alternative specification using a measure of Europeanization as the main independent variable see Table A.3 in the Online Appendix.
Membership" (3), and "Full Membership" (4). Furthermore, in light of how our independent variables are defined, we need to control for the deepening of the European integration process over successive EU treaties with the Depth categorical variable also coded from the ENLABASE data set. Depth takes the following values within given time periods The baseline empirical model that examines the (non-monotonic) impact of the level of political integration with the EC/EU on party-system polarization has the following functional specification: where i indicates countries, e denotes elections, and t time (years or decades), i.e., each observation is a country-election. All specifications include country and time fixed effects. We also use additional control variables to account for factors that are likely to affect party-system polarization. First, the Effective Number of Parties (ENP) is calculated based on the Laakso and Taagepera (1979) formula. 17 If a country has a high number of effective parties, party policy platforms are expected to be more dispersed, thereby increasing polarization. The Disproportionality of the electoral system is coded using the Gallagher index. The inclusion of both these variables is meant to capture the effects of changes in the electoral system on party-system polarization as discussed in the related literature.
Second, increased global interdependence forces governments to follow certain macroeconomic policies and is, therefore, expected to decrease polarization independently of EU membership. 18 This effect is captured by the Trade Openness variable, which is measured as the size of a country's exports and imports relative to its gross domestic product (GDP) and is coded from the 8.1 Penn World Tables.
Third, to capture the idea that the impact of EU integration on party-system polarization might be felt more strongly in member-states that have a lower voting weight in the European Council and thus exert less control over the design of EU policies, we control for the logarithm of a country's population (Population (log) variable) as a proxy for a country's political clout. This variable is also coded from the 8.1 17 It is as follows: 1/ v 2 i , where v i is the percentage of the vote received by party i. 18 Scholars have shown that globalization increasingly exerts pressure on governments to implement monetary discipline, deregulation, privatization, and labor market flexibility in order to enhance competitiveness and productivity (e.g., Mosley 2005). Penn World Tables. Fourth, as a proxy for output legitimacy and economic prosperity, we also regress on GDP and GDP Growth using data coded from the 2015 World Development Indicators.
Fifth, any polarization measure is sensitive to the distribution of voters. This is especially true for the Dalton index since it measures deviations from the partysystem mean and weights them by party vote share. For instance, if the voter distribution is very dense around the mean, a party's movement towards the extremes might have a minimal contribution to overall polarization as it will lose many votes to centrist parties whose small deviations will now be overweighted. To account for this effect, we include an additional control variable (Median Voter), which measures the position of the median voter using the formula of Kim and Fording (1998). Finally, we include country (λ i ) and year-or decade-fixed effects (γ t ). Table A.1 in the Online Appendix includes summary descriptive statistics of the data used in the subsequent empirical analysis.
We start our analysis by focusing on the fifteen oldest EU member-states (up until the enlargement of 1995), namely Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, and the United Kingdom. Table 1 shows the results of our party-system analysis covering a maximum of 208 cases in total.
Model 1 of Table 1 estimates the baseline model with decade fixed effects. We find that the marginal effect of a country's level of integration with the EC/EU on party-system polarization is negative and significant for low levels of integration, suggesting that the slacker constraints imposed on associate members, candidate members, or partial members lead to ideological convergence among parties. This could be construed as evidence of the EU's "soft power". As the level of EU integration increases, however, its marginal effect on polarization becomes both positive and significant (evidence of the EU's "second-order democratic deficit"). Model 2 also controls for GDP Growth as a proxy for the party system's output legitimacy. Model 3 replaces decade fixed effects with year fixed effects. Model 4 controls for GDP Growth in addition to year fixed effects. The results change minimally across these four specifications as they all exhibit a significant non-monotonic relationship between the level of integration with the EC/EU and party-system polarization.
Model 5 uses the Dalton index measure of Polarization. While the sign of the EC/EU Integration Level coefficient remains the same, it is marginally not significant. In fact, this conforms with our prediction in Footnote 13 that the non-monotonic increase in party-system extremism (as measured by the two-party MDP index) should be steeper than that in party-system polarization (as measured by the multiparty Dalton index) because with more parties the average policy shift should be relatively milder and the Dalton index should smoothen out the non-monotonic effect. Finally, in Model 6, we replace the EC/EU Integration Level variable with the fine-grained EC/EU Integration Status variable. Its marginal effect on party-system polarization is again negative for low values of integration with the EC/EU.
To get a meaningful interpretation of these results, we plot the marginal effects on party-system polarization (extremism) at different levels of integration with the EC/EU using the regression results from Model 4 in Table 1. In Fig. 2, the solid line indicates how this marginal effect changes with the level of integration with The dependent variable Polarization is measured as the distance between the two most extreme parties (MDP index) in columns (1) through (4) and in column (6), and as the Dalton index in column (5). Numbers in parentheses are standard errors clustered by country. a denotes statistical significance at the ten percent level, b at the five percent level, and c at the one percent level the EC/EU. The two-tailed ninety-five percent confidence intervals around the line indicate the areas in which this effect is statistically significant. Accordingly, party-system polarization (extremism) decreases up to the point where an association agreement between a country and the EU enters into force. In corroboration "Take back control"? The effects of supranational integration...

Fig. 2
Average marginal effect of level of political integration with the EC/EU on party-system polarization of Hypothesis H1, once a country reaches the "Partial Membership" level, further integration with the EC/EU increases party-system polarization. 19 In Table 2, we run several robustness tests. In Model 1, we use the wider data set, which includes data on all twenty-eight EU current member-states. The results are almost identical to the results obtained with the smaller data set (see Model 3 in Table 1). Model 2 restricts the analysis to the EU-15 and the pre-crisis period (up until 2009) in order to verify that this is not simply a story about the effects of the Global Financial and Eurozone Debt Crises. The results are robust to this restricted sample. In Model 3, we introduce the Treaty Period variable, which codes seven different big treaty events by month-year, 20 and interact it with the EC/EU Integration Level variable to estimate how proximity to the EU's integration frontier interacts with the process of integration deepening to affect party-system polarization at the domestic level. Model 4 includes separate interaction terms with each treaty dummy variable, while Model 5 is the same as Model 3 with EC/EU Integration Status as the main explanatory variable. Finally, in Models 6 and 7, we estimate the model with the 19 Again note that the empirical non-monotonic effect on aggregate party-system polarization is rather smooth despite the use of the two-party MDP measure of party-system extremism; this is arguably so because the breaking point should occur at different times in different countries and, therefore, the overall effect is averaged down. 20 The seven milestone treaties of the EU are the Treaty of Rome (signed in March, 1957), the introduction of the European Monetary System (entering into force in February, 1979), the Schengen Agreement (signed in June, 1985), the Maastricht Treaty (signed in February, 1992), the Schengen Regime (entering into force in March, 1995), the launch of the third stage of European Monetary Union (in January, 1999), and the Treaty of Lisbon (signed in December, 2007). As such, the Treaty Period variable builds on the Depth variable by providing more fine-grained data on treaty dates.  The dependent variable Polarization is measured as the distance between the two most extreme parties (MDP index) in columns (1) through (5) and as the Dalton index in columns (6) and (7). All models include the Median Voter, Disproportionality, Effective Number of Parties, Population (log), Trade Openness, and Trade Openness (squared) as control variables. Numbers in parentheses are standard errors clustered by country. a denotes statistical significance at the ten percent level, b at the five percent level, and c at the one percent level Polarization variable derived from the Dalton index. In Model 7, while the average marginal effect of EC/EU Integration Level on Polarization is almost always positive, it seems to decrease once the Schengen Agreement gets signed in June, 1985, then increase substantially with the signing of the Maastricht Treaty in February, 1992, and finally continue to be positive after the Maastricht Treaty but without going beyond its 1992 level (see Fig. 3).

Data and empirical model at the party level
If our theory holds water, then we should expect the effects detected in our partysystem analysis above to carry over to the party level of electoral behavior. In what follows, we proceed to test (party-level) Hypothesis H2 of our theory and seek to provide further support to the robustness of our empirical results above by thinking of party-system polarization as the result of extremism at the party level. The more extreme positions parties take, the more polarized the national party system becomes. 21 In this set of tests, we expect to find that during the earlier stages of European integration EC/EU membership mitigated parties' tendency towards extremism whereas during the later stages it reinforced it. In this case, our dependent variable Extremism (Mean) is operationalized as the absolute distance between a party's policy position and the mean party position. We created a party-level data set by collecting data for the same fifteen EU countries (Austria, Belgium, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Luxembourg, Netherlands, Portugal, Spain, Sweden, and the United Kingdom) for the years between 1958 and 2013. Our expanded party-level data set now includes 1420 cases. In the analysis to follow, we re-scale parties' position to range between one and ten. 22 Each observation in this model is a country-election-party. We estimate the following equation: where i indicates countries, e indicates elections, and j denotes parties, t denotes time (years or decades), and EC/EU Membership is a binary variable.
To account for the deepening of the EU, we estimate the equation using a Trend variable that starts in 1957 and increases by one unit in every subsequent year. We expect EC/EU membership to impose more constraints as the European integration deepens. 23 At the same time, we expect countries to face tighter supranational constraints by dint of a thicker policy acquis if they join the EC/EU at a later stage. To capture these effects, we include an interaction term (Trend*EU Membership). 24 We expect the marginal impact of EC/EU membership on party-level extremism to be negative during the earlier years of European integration and positive during the later years when the policy constraints imposed by the EU become tighter. Summary descriptive statistics can be found in Table A.2 in the Online Appendix.
We estimate an ordinary least squares (OLS) model with country fixed effects and robust standard errors clustered by countries. Our specification assumes that observations are not independent within clusters although they may be independent between clusters. The first column in Table 3 shows the estimates of a model where the main dependent variable is Extremism (Mean) and standard errors are clustered by country. 25 The coefficient of the EC/EU Membership variable is negative and significant 21 Note that in a multi-party system some parties may converge further towards the center while others diverge. See Stanig (2011) for an account of the non-monotonic properties of aggregate polarization measures in multi-party systems. 22 To re-scale the index, we add 100 to each party's position and divide the sum by 20. 23 As a robustness check, we run the model replacing the Trend variable with the Treaty Period categorical variable and dropping decade fixed effects. Our key results remain qualitatively unchanged. 24 The inclusion of this interaction term mitigates any potential effects of spurious correlation (see, e.g., Hix 2003). 25 As a robustness check, we estimate the same models with the dependent variable Extremism (Median) operationalized as the absolute distance between a party's position and the median party position weighted by vote share. The results (Table A.4 in the Online Appendix) are similar to those reported here.  The dependent variable throughout is Extremism (Mean). Numbers in parentheses are standard errors, clustered by country in Models 1, 2, 3, and 8 and by election in Models 4, 5, and 6. Model 7 uses bootstrapped standard errors. All models control for country fixed effects, decade fixed effects, and the Disproportionality variable, all not reported here. Model 8 only includes non-centrist parties. a denotes statistical significance at the ten percent level, b at the five percent level, and c at the one percent level at the five percent level while the coefficient of the Trend variable is negative but not significant. The interaction term coefficient is both positive and significant at the one percent level. Model 2 shows the regression results after controlling for GDP Growth. Model 3 includes the additional controls EMU and Single Market, which are coded as dummy variables taking the value of one only in years when EMU and the Single Market respectively were in effect. Of these two variables, only Single Market is found to exert a negative and significant effect on party-level extremism. Models 4-6 replicate the earlier regressions but cluster standard errors by election. Estimated coefficients as well as their significance levels are similar to those in the previous models. Model 7 estimates Model 2 using bootstrapped standard errors. The results again remain unchanged. Finally, Model 8 estimates Model 1 but only for non-centrist parties, i.e., those with an average re-scaled policy position strictly less than four or strictly greater than six in each election-year, thus limiting the number of observations to 530. The fact that the coefficient estimates on our key variables of interest remain qualitatively similar implies that our results are primarily driven by the behavior of non-centrist parties. This very much corroborates our theoretical predictions by showing that the previously asserted non-monotonic effect of integration on aggregate levels of partysystem polarization can be mostly attributed to the initial convergence and subsequent divergence of inherently non-centrist parties. Figure 4 shows the marginal effect of EC/EU membership on party-level extremism at different stages of the European integration process starting from 1957 (Model 1 in Table 3). 26 The two-tailed ninety-five percent confidence intervals around the line indicate the areas in which this effect is statistically significant. As expected, EC/EU membership exerts a negative effect on party-level extremism during the earlier years of European integration. More specifically, the effect is negative and statistically significant for the first eighteen years since the Treaty of Rome (1975). As time passes, the effect becomes positive and statistically significant after the first forty-three years from the creation of the European Communities (EC) and continues to increase with time. This also confirms our intuition that the domestic trends of rising political extremism and polarization were already underway well before the economic crises of the late '00s.
Put differently, EC/EU membership induced political parties to move towards the center until 1975 and then started pushing them towards the extremes in the early '00s. This finding lends support to Hypothesis H2 that during the early years of European integration non-centrist political parties moved towards the mean party position but then started to diverge towards the extremes as European integration deepened further.

Comparative case studies
By way of putting more flesh on the bones of our preceding regression analyses, we now turn to discussing specific country cases and how political integration and supranational policy constraints have affected their overall level of party-system polarization over time. In order to estimate this dynamic effect, we employ the "synthetic control method (SCM) for causal inference in comparative case studies" developed by Abadie and Gardeazabal (2003) and Abadie (2010Abadie ( , 2015. SCM makes use of a weighted average of control units (i.e., the "artificial country") selected from an untreated "donor pool" of countries to construct a counterfactual scenario for the treated country (i.e., the "actual country"). In the European context, the event (or intervention or treatment) whose effect we are interested in amounts to a country's accession to the European Community (EC)/European Union (EU) or the European Monetary Union (EMU) and with it the transposition of the full acquis of rules and regulations. SCM allows us to address the endogeneity of the binary treatment (EC/EU/EMU accession of country i at time t) and the potential post-treatment bias (Campos et al. 2018).
Since we only get to observe party-system polarization outcomes during election years, we choose to operationalize our dependent variable as an annual five-year moving average of party-system extremism as measured by the MDP index. In order to deal with time-varying omitted variables, we match the actual and artificial countries on a yearly basis during the pre-event phase along the following set of predictors: the Effective Number of Parties (ENP), Disproportionality, the position of the Median Voter, GDP per Capita, Unemployment, GDP Growth, the KOF Index of Globalization (general), the KOF Index of Economic Globalization, the KOF Index of Political Globalization, 27 Cultural Proximity, Population, Population Density, the Status variable, and Trade Openness. 28 Our "donor pool" of countries draws on a sample of (European and non-European) advanced industrialized democracies that have never been full members of either the EU or the EMU within the given time span. 29 In our choice of matching variables and the donor pool, we attempt to satisfy two key assumptions of SCM: (i) the set of predictors of the outcome should include variables that can approximate the path of the treated country ("parallel trends") but should not 27 All KOF Globalization data have been drawn from Dreher (2006) and more recently from Gygli et al. (2018). 28 All new variables used here have been drawn from the International Political Economy Data Resource (Graham and Tucker 2019). 29 We do include for example EU associate countries, such as Turkey, or European Economic Area (EEA) members, such as Norway and Iceland. include variables that anticipate the effects of the intervention, and (ii) the countries used to construct the synthetic control group should not be affected by the event. 30 As in Campos et al. (2018), the choice of non-European countries for our donor sample clearly satisfies assumption (ii) as the membership treatment in one country should not affect polarization in countries that are otherwise ineligible for full formal membership. 31 Moreover, despite the prima facie implausibility of approximating European with non-European political systems, below we provide evidence that the pre-treatment parallel trends assumption (i) is satisfied by the SCM's endogenous choice of counterfactual countries based on a given array of (economic, political, and institutional) predictors detailed above.
In what follows, we present several comparative country cases that help illustrate the fact that political integration by way of formal accession to a supranational union (viz., the EC/EU and the EMU) can have a substantial non-monotonic effect on party-system polarization over and beyond the general trends induced by economic globalization or other omitted factors. The endogenous selection of synthetic controls is mostly driven by the degree of pre-event balance in the observables and the chosen time span is a function of the availability of appropriate donor countries. 32 Figure 5 presents six comparative cases including at least one country from the 1973, 1981, 1986, 1995, and 2004 The EC/EU membership treatment effect is captured by the discrepancy between the post-treatment polarization trends of the treated country (solid lines) and those of its synthetic controls (dashed lines) as it differences out any other common party-system polarization trends due to other factors. A widening negative difference or a narrowing positive difference between these trend lines is interpreted as a negative effect of supranational integration on party-system polarization; on the other hand, a narrowing negative difference or a widening positive difference is interpreted as a positive effect of supranational integration on party-system polarization.
In most cases, we observe a strong non-monotonic (U-shaped) pattern, which is consistent with our theoretical and empirical results: compared to the artificial country (consisting of a weighted average of non-EC/EU members at the time), initial 30 Although we are quite confident that both these assumptions are generally satisfied by our identification strategy, we should point out two key caveats: (i) European integration is a "moving target" and amounts to multiple interventions over time in the form of successive treaties of integration deepening within several policy areas, and (ii) treatment effects may very well be felt during the pre-event stage of accession negotiations, where candidate-members are expected to satisfy a given set of accession criteria while they remain in the union's ante-chamber. The latter effect was especially more pronounced in the latest waves of EU enlargement (2004,2007,2013), where the acquis was thicker and accession countries had to make up a lot more ground. 31 That being said, non-European countries are potentially subject to a softer (graded membership) treatment effect in the form of trade and association agreements. 32 Admittedly, we are not able to achieve perfect pre-treatment matching in many cases for two reasons: (1) most of our cases consist of countries that either joined the EC/EU early on (e.g., the UK) or had just recently become multi-party parliamentary democratic systems before joining the EC/EU (e.g., Greece), thus leaving little time within our time series to fully satisfy the parallel trends assumption, and (2) our donor pool of non-EC/EU/EMU countries is relatively limited due to data restrictions. 33 We are not able to present results for the newest EU members -namely, Bulgaria, Romania, and Croatia -as the post-event amount of observations in our data set is very limited. entry is associated with a decline in party-system polarization (or extremism as best captured by the MDP measure) -where, in fact, this process may have started earlier during the pre-accession phase -followed by a subsequent rise. 34 By way of probing our hypothesized relationship between the input-output legitimacy trade-off and political polarization, we also present a three-year moving average of annual GDP growth rates (as a proxy for output legitimacy) for each country in our sample in Figure A.2 in the Online Appendix.
To some extent, each country's party system follows its own idiosyncratic trajectory as conditioned by its particular electoral system and the relationship between its overall macroeconomic environment (output legitimacy) and its decision to join or not to join subsequent supranational integration treaties. The UK, for example, the largest of the first countries to join the EC in 1973, exhibits a double-dipped pattern, where two inflexion points of increasing polarization seems to occur in the run-up to the European Monetary System (EMS) in 1979 and the Maastricht Treaty in 1992 with a brief interlude of convergence in the aftermath of the Single European Act (SEA) of 1986. Note that both inflexion points fall within a prolonged recessionary period (see Figure A.2 in the Online Appendix). Following the early consensus around membership generated by the 1975 Referendum on the European Communities, the Labor and Conservative Parties started to exhibit increasing ideological divergence under the leadership of Michael Foot and Margaret Thatcher respectively in the early '80s followed by a brief period of convergence. The Maastricht Treaty negotiations coupled with the EMS Crisis of 1992 eventually brought to light widening intra-party rifts over Europe causing another anti-EU backlash.
Our findings have been nowhere more so evident than in Greece, a country that has exhibited a similar double-dipped polarization pattern centered around the major treaties of European integration. While the early optimism of EC accession (in 1981) -that came with a substantial injection of funds -, the ratification of the Maastricht Treaty, and the periods of adjustment and reform that ensued respectively led to significant ideological convergence between the main center-left (PASOK) and center-right (Nea Dimokratia) parties towards pro-EU policies, periods of economic recession (see Figure A.2 in the Online Appendix) under the tightening constraints of EC/EU membership (in the aftermath of the SEA in 1986 and the Lisbon Treaty in 2007) triggered prolonged episodes of party-system polarization (extremism). In other words, Greece had a sort of "love affair" with the EU during the pre-Olympicgames spending spree of 1996-2004, which bought plenty of output legitimacy and material compensation; since the beginning of the Greek Debt Crisis (2009-18), however, it has experienced severe political volatility and painful fiscal consolidation whilst teetering on the brink of bankruptcy and Eurozone exit.
Spain and Portugal, on the other hand, the next two countries to join the EC in 1986, exhibit smoother U-shaped post-accession patterns of polarization compared 34 Some of the inconsistencies in the patterns may be due to three reasons: (i) as we have argued before, the binary nature of EC/EU membership does not capture the complexity of the deepening process of political integration as existing treaties get updated and other treaties come into force, (ii) the time span of our panel data set varies from one country-case to another both in terms of pre-event and post-event data availability, and (iii) the donor pool of permissible counterfactuals is in some cases rather limited (e.g., Austria). to synthetic control trends. While both experienced periods of political turbulence and ideological divergence during the early years after the collapse of their dictatorial regimes, in both cases EC accession seemed to make a substantial contribution to the process of democratic consolidation in the form of ideological convergence and de-escalation of aggregate levels of party-system polarization. Yet, while the inflexion points of increasing polarization both seem to occur during recessionary periods (see Figure A.2 in the Online Appendix), they happen at different moments in the process of European integration (around the time of the Lisbon Treaty and the Global Financial Crisis for Spain, 35 and the Maastricht Treaty and the EMS crisis of 1992 for Portugal).
Austria, one of the 1995 EU enlargement countries, also exhibits the same Ushaped pattern. Austria is an interesting case of a net contributor country, developed enough at the time of accession that it did not have to go through an extended preaccession period of adjustment and transposition of the European acquis; yet, it also experienced an early period of convergence during a very productive period in the process of European integration in the late '90s (e.g., Amsterdam Treaty of 1997, launch of the euro). In Austria's case, the inflexion point of the populist backlash occurs around the time of the Nice Treaty (2003) and has arguably much to do with the growing influx of immigrants (but also with an economic slowdown as shown in Figure A.2 in the Online Appendix). Finally, it might probably be Poland, one of the newest EU member-states, that exhibits the clearest U-shaped polarization pattern, with the early years of economic adjustment and ideological convergence immediately followed by a strong backlash of increasing polarization in the aftermath of the Lisbon Treaty and the Global Financial Crisis (and a certain deceleration of GDP growth). 36 In the remainder of this section, we study the impact of EMU membership on domestic party systems as we expect distinctive effects generated by an ever deeper form of supranational integration and policy coordination. The EMU effectively amounts to a subunion of the EU as it consists of EU member-states that chose to adopt a common currency, thereby relinquishing control over their monetary policy and further coordinating their fiscal policies. Figure 6 below presents five comparative cases of the effects of EMU membership on aggregate party-system polarization: four are among those countries of the Southern periphery substantially impacted by the recent crisis and subjected to harsh bailout agreements (Greece, Ireland, Italy, Spain), while one belongs to the EMU subgroup of Northern creditor countries (Austria). The related annual GDP growth rate time series are presented in Figure A.3 in the Online Appendix. In all five cases, where matching of pre-event trends against those of artificial countries seems to work well, we once again detect a U-shaped effect of EMU accession on party-system extremism (as captured by the MDP measure). To a large extent, the non-monotonic effect reflects the Eurozone Debt Crisis 35 It should be noted, however, that the left-right polarization measure does not necessarily capture Spain's worsening center-periphery relations and the ideological divergence between Spanish parties on the pro-/anti-centralization axis; this trend presumably started much earlier. 36 Results from other Visegrad countries, such as the Czech Republic and Slovakia, exhibit equally clear and well-balanced trends and are available from the authors upon request.  political backlash both in terms of increased conditionality for debtor countries and bailout disbursements for creditor countries. While the preventive arm of the Maastricht Treaty's Stability and Growth Pact (SGP) forced countries to make macroeconomic adjustments in order to meet the Maastricht convergence criteria before acceding to the final stage of EMU, it was only until the corrective arm came into effect -in the form of the Excessive Deficit Procedure -that EMU's policy constraints started to have more bite and suppress party-system polarization. Once the preventive and corrective arms of the SGP proved to be inadequate, policymakers decided that the EMU's monitoring, prevention, and correction capabilities had to be complemented by policy innovations in the field of fiscal and macroeconomic governance consisting of new treaties (e.g., the Treaty on Stability, Cooperation, and Governance and its Fiscal Compact), fiscal coordination schemes (e.g., the European Semester), ad hoc mechanisms (e.g., the European Financial Stability Mechanism and the European Stability Mechanism), and revised sets of regulations (e.g., the Six Pack and Two Pack) (Closa 2015). In most of our cases, the non-monotonic inflexion point of increasing polarization occurs around the time of the turbulent crisis years of the late '00s and slightly before some of the indebted countries of the European South had to accede to high-powered conditionality agreements in the form of Memoranda of Understanding between the European Commission (EC), the European Central Bank (ECB), and the International Monetary Fund (IMF). 37 Clearly, the Eurozone Debt Crisis triggered rising levels of polarization both within and across countries.

Concluding remarks
Throughout the post-WWII era and all the way into the early 2000s the inherent tensions between globalization and national democracy (Rodrik 2012) appeared to be bottled up within a broader liberal democratic consensus with regard to the efficiency merits of economic interdependence, political integration, and technological advancement. In recent times, however, many advanced liberal democratic systems seem to have experienced a breaking point beyond which many of these trade-offs have become politicized, thereby juxtaposing the previously unchecked forces of globalization and integration against the counter-forces of nativism and illiberalism and polarizing public discourse around stark political dilemmas between populism and pragmatism, democracy and technocracy (Habermas 2015;Caramani 2017). Such tensions manifest themselves in the widening spatio-temporal gap between capital markets and democratic politics, the growing disparities between sources of input and output legitimacy, and the tightening trade-off between responsibility and responsiveness.
In this paper we postulated that one of the principal manifestations of this degenerative process is party-system polarization in the electoral arena. We theorized the relationship between integration and polarization both at the party-system level with the trade-off between political inputs (representation) and economic outputs (compensation) and at the party level of electoral competition with the responsibilityresponsiveness trade-off between net spatial transportation (ideological) costs and orthogonal valence benefits. We showed that while rapid integration in the form of tightening policy constraints may initially speed up ideological convergence across the spectrum, there is a threshold level in the slackness of supranational constraints beyond which they can have the opposite effect and backfire in terms of extreme platform polarization. This non-monotonic relationship was borne out by our empirical analysis of party-system polarization and party-level extremism in European parliamentary multi-party systems. We found a negative and statistically significant effect of integration on party-system polarization and party extremism during the earlier phases of the European project, when the integration process centered around areas of "low politics" (e.g., energy, trade, common market regulation). However, we also found strong evidence of a reversal of that process of ideological convergence during those later stages of European integration marked by the broadening and deepening of the EU's scope of policy competences.
Our empirical results also seemed to suggest that this "paradigm shift" had been underway well before the deeply recessionary effects of the Global Financial and Eurozone Debt Crises were felt. As a result of the persistence of austerity policies, rising debt burdens, stagnating labor incomes, bank bailouts, and the augmenting incidence of taxation on unskilled labor among other factors, issues such as income inequality, social justice, national identity, and the design of economic, financial, and political globalization have long been part and parcel of the core political agenda of democratic politics. Moreover, the surge in migration flows has for some time now rekindled the rise of nativism and identity politics against the multiculturalism of most advanced liberal democracies (Rodrik 2017).
Our work helps better understand the potential repercussions of unfettered crisisridden globalization coupled with political volatility and polarization. As long as the locus of democratic inputs remains at the national level, then basic questions of job (in)security, social protection, income inequality, and national identity will keep roaring back to the forefront of politics in advanced liberal democracies. We should, therefore, no longer be surprised by the political success of populist candidates advocating trade protectionism, immigration restrictions, and retrenchment from global policy coordination agreements on issues such as financial regulation, tax policy, defense, and climate change. Moreover, we should expect national elections to function less effectively as a way of translating citizens' preferences into government policy as a result of the growing incongruence between input-oriented representative and output-oriented responsible government.
While other papers in this Special Issue study the implications of the rise of populism for the stability of the liberal international order, we reverse the question and show how the design and overall effectiveness of institutions of global governance and supranational integration create "second-image-reversed" effects upon domestic democratic politics (Gourevitch 1978). Although we do not explicitly derive the conditions for withdrawal from a supranational organization with high sovereignty costs, we argue that increased party-system polarization in the form of ideological extremism, nativist populism, and nationalist anti-globalism (especially in key member-states) may irrevocably sap the political consensus underpinning the liberal international order.
Hence, our findings have wide-ranging normative implications about the political feasibility of the policy centralization process and the sequencing of economic and political integration especially in the European context. The EU's "multiple crises" (namely, the euro crisis, the migration crisis, Brexit, and the populist backlash) have highlighted the complex nature of economic interdependence and cross-border policy spillovers. On one hand, the emergency politics of crisis management, predominantly driven by considerations of output legitimacy and policy coordination, have led to a power shift towards unelected and unaccountable executive bodies both at the national and European levels. On the other hand, there is a backlash of growing popular demand for input legitimization in the form of national "beggar-thy-neighbor" solutions, often voiced (and in some cases put into effect) by populist and nationalist politicians. 38 Europe, therefore, finds itself at a crossroads whereby any further pursuit of efficiency-oriented policy centralization will have to be cushioned by institutional structures of democratic responsiveness and representation. However, this conundrum cannot be overcome as long as these mechanisms of input legitimization reside at the level of the nation-state. As Habermas (2015) argues, the only way to break free of the current input-output trade-off seems to be by reinstating democracy at the European level and infusing supranational institutions with direct mechanisms of democratic legitimization.