Do Intergovernmental Organizations Have a Socialization Effect on Member State Preferences? Evidence from the UN General Debate

The question of whether intergovernmental organizations (IGOs) have a socialization effect on member state preferences is central to international relations. However, empirical studies have struggled to separate the socializing effects of IGOs on preferences from the coercion and incentives associated with IGOs that may lead to foreign policy alignment without alter- ing preferences. This article addresses this issue. We adopt a novel approach to measuring state preferences by applying text analytic methods to country statements in the annual United Nations General Debate (UNGD). The absence of interstate coordination with UNGD statements makes them particularly well suited for testing socialization effects on state preferences. We focus on the European Union (EU), enabling us to incorporate the pre-accession period—when states have the strongest incentives for foreign policy alignment—into our analysis. The results of our analysis show that EU membership has a socialization effect that produces preference convergence, controlling for coercion and incentive effects. cuestión

The question of whether intergovernmental organizations (IGOs) have a socialization effect on member state preferences is central to international relations. However, empirical studies have struggled to separate the socializing effects of IGOs on preferences from the coercion and incentives associated with IGOs that may lead to foreign policy alignment without altering preferences. This article addresses this issue. We adopt a novel approach to measuring state preferences by applying text analytic methods to country statements in the annual United Nations General Debate (UNGD). The absence of interstate coordination with UNGD statements makes them particularly well suited for testing socialization effects on state preferences. We focus on the European Union (EU), enabling us to incorporate the pre-accession period-when states have the strongest incentives for foreign policy alignment-into our analysis. The results of our analysis show that EU membership has a socialization effect that produces preference convergence, controlling for coercion and incentive effects.

Introduction
Do intergovernmental organizations (IGOs) have a socialization effect on member state preferences? Some scholars of international relations (IR) argue that IGO membership leads to a convergence in states' foreign policy preferences through a socialization process (e.g., Wendt 1994;Checkel 2005;Bearce and Bondanella 2007). The regular and sustained interactions that occur within IGOs are said to lead members to adopt similar preferences in world politics. From the constructivist perspective, in particular, international institutions are social environments that not only create external constraints for members, but can also transform their identities and interests.
Others disagree with this view. Proponents of the rationalist approach argue that state preferences are formed within national boundaries and any shifts in observable behavior are the result of bargaining among states to maximize gains from these organizations, rather than changes in underlying preferences (see, e.g., Mearsheimer 1994;Moravcsik 2013). Resolving this debate requires us to empirically verify that IGOs have a socialization effect on member states that leads to preference convergence. Yet, there are two challenges to doing so. The first is deriving convincing measures of state preferences from observable behavior. The second is ensuring that any observed preference convergence results from socialization rather than alternative processes, such as coercion, material incentives, or coordination.
This paper contributes to this debate by providing new empirical evidence. We use a new source of data on state preferences-countries' annual statements in the United Nations General Debate (UNGD). In their UNGD statements, governments discuss the major events of the past year and the issues in world politics they consider most important (Baturo, Dasandi, and Mikhaylov 2017). Importantly, UNGD statements are not institutionally connected to decision-making processes in the United Nations and there is no coordination among states on producing these statements. The absence of coordination makes UNGD statements especially well suited to assess whether IGO membership leads to preference convergence through socialization.
We focus on the case of the European Union (EU). This enables us to address the argument that preference convergence may be due to the incentives that states have to align their preferences with those of an IGO to gain membership. By focusing on the EU-with its multiple-step accession process-we can account for any incentive effects that occur during the accession process and explanations based on IGO coercion. Furthermore, the EU has long been viewed as having a distinct normative and value-oriented foreign policy approach (Manners 2002). This allows us to more precisely identify whether EU members adopt these normative values in their foreign policy preferences through a socialization process. We also focus on the EU because it is the most institutionalized IGO in world politics, and hence the most likely case for socialization processes to foster state preference transformation (Checkel 2005). As such, if we find no evidence of socialization effects here, then we are unlikely to observe this with any other IGO.
We test the effects of EU membership, and the different stages of EU accession, on preference convergence using measures of preference similarity derived from UNGD statements through text analytic methods. Our findings demonstrate that EU membership leads to preference convergence with the EU, controlling for incentive effects. We find no evidence to suggest that states diverge from the EU position after they become members, as might occur if alignment was driven primarily by the incentive of gaining membership. We also examine the 2004 EU enlargement in greater depth to show that following accession, the new EU member states increasingly emphasized themes associated with the values and norms of EU foreign policy in their UNGD statements. Therefore, our analysis indicates that IGO membership fosters state preference convergence through a socialization process.

Socialization and Intergovernmental Organizations
Since the rise of constructivism as the main rival to rationalism in IR, a principal dividing line in the rationalistconstructivist debate has been the nature of state preferences. Central to constructivism is the idea that international interactions not only change the behavior of states, but can also transform states' identities and interests (Wendt 1992(Wendt , 1994Finnemore and Sikkink 1998). State interests are seen to "emerge from and are endogenous to interaction with [international] structures" (Checkel 1998, 326). In contrast, rationalism maintains that interests are formulated within national boundaries and then bargained in international interactions (e.g., Stone 2011;Moravcsik 2013). Exchanges at the international level can modify the cost-benefit analysis of states, while leaving their properties unaltered (Mearsheimer 1994). From this perspective, preferences are exogenous to interstate processes.
The origins and evolution of state preferences are central to so many debates in IR that it represents "perhaps the most fundamental issue" in the discipline (Johnston 2005(Johnston , 1040. As Wendt (1992, 423) has argued, the issue is empirical, and resolving the debate requires researchers to examine the causal relationship between interactions and states' identities and interests. Within this research agenda, some have investigated the effects of processes developed within international organizations on the behavior and attributes of states (e.g., Checkel 2005;Lewis 2005;Bearce and Bondanella 2007;Greenhill 2010;Taninchev 2015). IGOs are seen to shape state preferences through exchanges and "conditions that are unique to social groups qua social groups, namely, socialization processes" (Johnston 2001, 487). This literature, however, has struggled to test, let alone validate, the transformative capacity of international institutions on state preferences.
Socialization, for instance, is often used to denote processes of social influence and peer pressure that operate within IGOs and prompt pro-norm behavior (Johnston 2014). In such situations, states alter their behavior because of the distribution of social rewards and punishments such as social liking, public recognition, naming and shaming, or shunning. These processes, however, do not require a transformation of preferences. The change in states' behavior may be brought about by a consequentialist choice, whereby the social environment of an IGO can increase the costs of noncompliance or the benefits of group conformity (Zurn and Checkel 2005). Such shifts in behavior do not tell us whether states' underlying preferences have changed. Hence, they could represent "public conformity without private acceptance" (Johnston 2001, 499). Our objective is to examine whether interstate contacts within IGOs trigger a redefinition of members' preferences along common lines, without material or social incentives. This "purest type of socialization" (Johnston 2001, 494) can help disentangle the exogenous/endogenous nature of state preferences. To do this, we further specify what we mean by socialization, distinguishing preference convergence due to socialization from convergence resulting from other processes such as coercion or material incentives linked to IGO membership.

Specifying Socialization
We understand socialization as a set of processes whereby actors acquire new beliefs and interests "through regular and sustained interactions within broader social contexts and structures" (Bearce and Bondanella 2007, 706). We examine the introduction of new members into the values and practices of a given community and consider what happens to older members, and to the broader community, once novices become a "conventional" part of the IGO (Taninchev 2015, 135).
IGOs can act both as a site and promoter of socialization and membership offers various venues and mechanisms for states to interact and be socialized (Checkel 2005;Zurn and Checkel 2005). IGOs can facilitate the diffusion of values at the societal level through seminars with stakeholders, media campaigns, or engagement with civil society (Gheciu 2005, 23-32). In more structured IGOs, such as the EU, the diffusion of norms also occurs through legislation (e.g., Börzel and Risse 2000). Nonetheless, in line with most studies, we posit that government officials are at the forefront of socialization processes (e.g., Schimmelfennig and Sedelmeier 2005;Bearce and Bondanella 2007;Fink 2013). Two types of officials are involved in these processes: national bureaucrats who interact regularly in IGOs (Smith 2004;Cao 2009) and high-level representatives, whereby socializing exchanges occur at the political level (Greenhill 2010;Taninchev 2015). Through interactions in IGOs, both national bureaucrats and political leaders exchange ideas, obtain new information, acquire the rules of the group, and develop sentiments of trust (March and Olsen 1998). IGOs can promote preference convergence through these processes.
The convergence of state interests, however, may be due to issues that have little to do with socializing interactions. The policy convergence literature identifies several factors that foster convergence, some of which are linked to ideational influences or economic pressures. They include international, national, and policy-specific causal processes (Heichel, Pape, and Sommerer 2005). Holzinger and Knill (2005) examine this literature and list five main factors: independent problem-solving (i.e., convergence derives from similar but independent responses to common problems), regulatory competition (i.e., increasing economic integration generating competitive pressures to converge), transnational communication (this largely corresponds to socialization processes), international harmonization (i.e., convergence is the result of compliance with international legal obligations), and imposition (i.e., convergence is coerced by external actors who exploit unequal economic and/or political power). 1 For the purposes of this paper, the first two factors are less relevant-as we evaluate whether IGOs promote convergence of foreign policy preferences. Nonetheless, countries may join IGOs because they have already shifted their preferences. A crucial issue that we need to consider is that convergence with the IGO may be due to legal requirements or the sanctioning power of the IGO (the fourth and fifth factors). IGOs control resources that can be used as incentives or punishments to ensure conformity with their rules.
As Fink (2013, 633) states, any socialization-based explanation "has to rule out (or control for) coercion by IOs as an alternative explanation." Imposition and binding obligations are particularly strong and visible during the accession period. IGOs offer various benefits to members-including security, trade opportunities, international legitimacy, regulatory capacity, and lower transaction costs (see Simmons and Martin 2002). A rationalist approach would emphasize the strong incentives states have to alter their behavior in the short term to gain these benefits. Furthermore, conditionality is most effective during this accession period, as the IGO requires applicant states to align with its norms to join. The adjustment of state preferences during this phase is strategic, shallow, and ephemeral-and does not reveal a transformation of preferences through socialization. According to this rationalist perspective, we should see preferences converge with the IGO most notably during accession (e.g., Schimmelfennig and Sedelmeier 2005). We refer to this as the incentivization thesis (IT).
In contrast, the socialization thesis (ST) assumes that new preferences, acquired through a state's participation in the IGO, are internalized by the state. As Alderson (2001, 417) explains, internalization is the hallmark of socialization-whereby states incorporate norms emerging elsewhere in the international system. Internalization implies that these preferences become part of the actors' properties and are not simply short-term behavioral adjustments. This ideational change should persist over time and be reproduced without active corroboration. The new norms become institutionalized within state structures.
Internalization may be triggered by various mechanisms. Students of socialization usually emphasize mechanisms of non-reflective role-playing and persuasion (Checkel 2005). The convergence literature refers to similar processesoften grouped together as transnational communicationsuch as learning, emulation, or joint problem-solving (Holzinger and Knill 2005;Cao 2009;Fink 2013). This paper focuses on the macro-effects of socialization. As Bearce and Bondanella (2007, 705) explain, if the macro-effect "could not be demonstrated, it would arguably make little sense to debate its underlying micro foundations." As to the ST, we evaluate whether convergence occurs postmembership, is sustained throughout the IGO affiliation, and internalized by the new member states-but we remain open as to which mechanism fosters internalization. These mechanisms share an important feature-they operate mainly through communication, information exchange, and horizontal state-to-state cooperation (Simmons and Elkins 2004;Holzinger and Knill 2005;Fink 2013). Of course, some incentives to align with the IGO's preferences also exist post-accession. However, as a rationalist approach emphasizes, these incentives are greatly reduced once the main benefits that come with membership are secured and would have to compete with the temptation to revert to preaccession preferences.
The ST also considers the content of the IGO's norms internalized by states. State socialization is not just about absorbing the IGO's procedures and working practices; it involves the incorporation of norms that define the mandate, values, and interpretive schemes of the IGO. Socialization is, in this sense, "normative," reflecting the norms promoted by the IGO (Alderson 2001). The challenge in examining whether IGOs elicit this internalized convergence is separating the endogenizing capacity of IGOs from factors related to strategic cost-benefit considerations. We do this by adopting a novel approach to measuring state preferences based on the application of text analysis to UNGD statements and explicitly incorporating the pre-membership accession period of IGOs into our analysis. We then evaluate the extent of post-membership convergence and consider the normative content of alignment.

Measuring State Preferences: UN General Debate Speeches
In examining whether IGO membership leads to preference convergence through socialization, a key challenge is measuring state preferences (Wendt 1994;Johnston 2014). The difficulty arises because measures of state preference are derived from observable state behavior, but this behavior may result from factors other than preferences. It may, for example, reflect strategic actions due to the international context. As Johnston (2001, 491) notes, "what may appear to be a change in preferences may, instead, be a change in strategies." Studies that systematically examine state preferences typically use countries' votes in the UN General Assembly (UNGA) (see Voeten 2013). The benefit of using UNGA votes to measure preferences is that they provide information about states' positions on a range of issues and can be compared across countries over time. There are, however, significant shortcomings of using UNGA votes to measure preferences-particularly for testing whether preferences change via socialization. The main limitation is that UNGA voting is significantly influenced by various external constraints and coordination mechanisms. For example, states have used foreign aid to influence countries' votes (see Dreher, Nunnenkamp, and Thiele 2008). The impact of strategic voting blocs in the UNGA is also widely recognized (e.g., Voeten 2000). International and regional organizations, including the EU, often have extensive coordination processes that shape members' voting behavior (Smith 2006b;Degrand-Guillaud 2009).
This suggests that UNGA voting similarity among IGO members is more likely to reflect the efficacy of powerful states or IGOs' coordination efforts on politicized resolutions than preference convergence (Jin andHosli 2013, 1275). This is problematic because coordination incorporates factors only partially, if at all, related to socialization. Successful coordination is closely linked to social incentives, peer pressure, and other mechanisms of social influence previously discussed (Johnston 2001). Material incentives such as issue linkages, side payments, and aid flows are used to elicit group conformity. In such cases, coordination may induce high levels of voting cohesion; however, such coordination only alters state behavior and does not affect preferences.
To address this issue, we propose an alternative approach to measuring state preferences based on countries' annual UNGD statements. The General Debate takes place every September, marking the start of each new UNGA session. It consists of heads of government, and other high-level representatives, delivering addresses to the UNGA on behalf of their state. Scholars of diplomacy have long recognized that leaders' public statements convey important information about foreign policy priorities (see Jönsson and Hall 2003). Indeed, a primary purpose of the UNGD is to enable states to put on public record their position on different issues (Baturo, Dasandi, and Mikhaylov 2017). All UN member states deliver an annual statement, which means they can be compared across countries and over time. Hence, it is a unique forum, seen "as a barometer of international opinion on important issues, even those not on the agenda for that particular session" (Smith 2006a, 155). 2 A key distinction between UNGD speeches and UNGA voting is that the latter is directly connected to the adoption of UN resolutions, whereas UNGD statements have no such institutional link to formal decision-making. Subsequently, governments face fewer external constraints when delivering UNGD statements compared to UNGA voting (Baturo, Dasandi, and Mikhaylov 2017). Significantly, governments produce UNGD speeches without consulting-let alone coordinating-with other states. Social processes that generate pressure for group conformity with votes have less influence on the content of countries' UNGD statements. Similarly, material incentives or threats are less likely to impact these speeches. The limited impact of external factors and the absence of coordination among countries make UNGD statements especially well suited to examining the effect of socialization on preference convergence.
Leaders certainly consider the ramifications of their UNGD addresses. Therefore, some posturing and strategic signaling can occur in these speeches. We argue, however, that generally the influence of external factors is minimal on UNGD statements-and certainly much lower than with UNGA votes. Some might argue that the lack of constraints and consequences means these speeches are "cheap talk" that do not accurately represent states' preferences. However, analyses of UNGD statements suggest this is not the case (Smith 2006a;Baturo, Dasandi, and Mikhaylov 2017;Kentikelenis and Voeten 2020). The absence of external constraints, instead, enables leaders to consider a wide range of issues and provide normative justifications for these positions. Furthermore, the lack of immediate consequences means states can discuss contentious issues without causing damaging repercussions (Smith 2006a). Hence, the absence of external constraints allows representatives to present a more complete picture of their state's foreign policy preferences. Indeed, a recent study comparing UNGD speeches with the positions held by state officials in different global governance fora finds that UNGD statements "convey meaningful information on underlying state preferences" (Kentikelenis and Voeten 2020, 26).

UNGD Statements and National Missions to the United Nations
To provide further support for this argument, we conducted interviews with representatives from the EU member states' national missions to the United Nations and officials from the EU delegation to the UN. 3 These interviews provide evidence of significant coordination among EU member states on UNGA voting and the absence of such coordination with UNGD statements. Furthermore, they shed greater light on how UNGD statements are viewed by national delegates. The interviews also explain how UNGD statements are produced, highlighting the role of national bureaucrats and high-level political representatives in the process-which we present in the Supplementary Information. 4 The representatives interviewed all described the intense coordination among EU member states to ensure voting 2 The voluntary fifteen-minute time limit on UNGD speeches ensures that representatives typically focus on the most important issues for their country. 3 We conducted interviews with ten representatives from the national missions and two representatives from the EU delegation between June 2015 and October 2017. 4 As discussed, national bureaucrats and high-level political representatives are also at the forefront of IGO socialization processes, suggesting UNGD statements are well suited for examining socialization. alignment on UNGA resolutions. As one official explained, this coordination occurs "on almost every issue, every time." 5 EU coordination mechanisms on UNGA voting are considered highly successful. As an Austrian representative described, "it is rare that we don't find an agreement." 6 Typically, once a draft of the UN resolution is received, it is debated among member states. Following this debate, representatives report back to their capitals, before meeting again to agree on their stance. These coordination meetings occur every morning during UNGA sessions, across different committees. As one interviewee described, in these meetings, "the peer pressure [to find common positions] is very real." 7 A German official, referring to a UN resolution about the Georgia crisis on which Cyprus had wanted to abstain, explained: In these cases, we lobby these countries to vote the same way as other EU member states. Both the EU delegation and individual member states do the lobbying, and we tell the representatives from these countries, "well, you are standing in the way of consensus, which is not good" … we keep the pressure up. 8 In contrast, the representatives all confirmed the absence of coordination among EU members in preparing and delivering annual UNGD statements. As a Finnish official explained, "each country does it its own way … this is a national speech; there is no coordination with other countries." 9 A German representative similarly stated, "coordination for the General Debate does not happen as governments do not want other states looking over their speeches," adding that "these speeches are the most sovereign thing that a country does as a member of the UN." 10 A similar statement was made by another official: Speeches at the General Debate are interesting because they flesh out national policies-what states think. It is the one place where states can speak their mind … it reflects the issues that countries consider to be most important. In this way, the speeches in the General Debate are unique… These speeches are one of the least coordinated parts of the [EU's] UN activities. 11 The interviews provide broader support for the importance of the UNGD statements in capturing state preferences. A representative from the Danish mission, for example, explained: The September [UNGD] speech is not cheap talk. It is in fact grounded in real policy priorities of the nation. It says what we are, what we want to represent in international affairs … It is one of the very few opportunities to speak with your own voice. 12 National officials agree that UNGD statements offer an opportunity to present "the country's vision of the world, the priorities for the year to come and how you think the challenges of the world should be addressed." 13  it officially sums up the position of Italy on the most important things in international politics in that particular year … the speech is a summary of a country's foreign policy in a given year." 14 The relevance of UNGD statements, together with the lack of coordination with other actors, makes these speeches particularly well suited to empirically test the transformative capacity of IGOs.

Membership Process in IGOs: The Case of the EU
There is a further issue that needs to be addressed to confirm that any preference convergence that occurs with IGO membership is due to socialization. Foreign policy alignment with an IGO could occur because of the material benefits, imposition, or legal commitments associated with membership rather than socialization. Membership changes the IGO's coercive powers and the incentive structure of the new member state. Material benefits are particularly high and detectable in the accession period. Hence, we explicitly incorporate convergence during the accession process into our analysis. We do this by focusing on the EU. The EU has a well-established process of enlargement and the procedure for becoming a member is clear and detailed. For example, we know that Slovakia submitted its membership application in June 1995; the European Council granted it candidate status in December 1997; negotiations started in February 2000; and Slovakia joined the EU in May 2004. Crucially, the socializing capacity of the EU and the incentives for candidate countries vary significantly during the accession process compared to the post-accession period. During the accession period, the incentives to align with EU policies are greatest, whereas socializing contacts with EU institutions are relatively infrequent (Schimmelfennig 2005). Research on EU enlargement finds that the assignment of rewards and punishments has effectively promoted policy convergence in Central and Eastern Europe (CEE) during the accession process. The credibility of membership incentives and of EU sanctioning powers convinced these countries' policymakers to align with EU policies, rules, and standards (Schimmelfennig and Sedelmeier 2005). As Epstein and Sedelmeier (2008, 803) explain, the preaccession period is "a most likely case for rationalist … approaches," while convergence due to socialization is limited in this phase.
Conversely, once a country joins, contacts with EU institutions become intense at all levels of government. This sustained cooperation creates an institutional environment rife with opportunities for national officials to share experiences and learn the perspectives of other states (Lewis 2005). We expect this socialization to lead to foreign policy convergence. After accession, the incentives to align certainly do not disappear: EU policies generate material interests that can produce converging pressures. However, these interests significantly diminish compared to the accession phase and are unlikely to offset the temptation to return to pre-accession preferences. The literature on EU enlargement argues that if the incentives-based explanation is correct, we should see compliance with EU norms cease or be overturned by the new members (Schimmelfennig and Sedelmeier 2005;Dimitrova 2010). Absent EU conditionality, new member states now face strong pressures and incentives-from political elites and other domestic actors-to revert to pre-accession policies and positions. A rationalist approach anticipates post-accession noncompliance precisely in areas such as foreign policy where EU 14 Interview, Italian Mission to the UN, September 11, 2017. conditionality is weak to begin with, and EU postmembership sanctioning power is limited and mostly based on non-coercion (Epstein and Sedelmeier 2008). If policy convergence was a purely strategic move, we would expect some divergence post-membership (Schimmelfennig 2005, 857).
If IGO membership does lead to preference convergence, then according to the ST and IT, it occurs at different stages of the membership process. Furthermore, the ST and the IT differ on whether the resulting change in positions continues after membership. This enables us to derive hypotheses related to the ST and the IT. Based on the ST, we would expect: H1: State preference convergence with the EU position is largest when a state gains EU membership.

H2: State preference convergence with the EU position is limited during the EU accession period.
In contrast, if the IT better explains how IGO membership produces preference convergence, then our expectation would be: H3: State preference convergence with the EU position is largest during the EU accession period.

H4: State preference convergence with the EU position is reversed when a state gains EU membership.
We also focus on the EU because it is the most institutionalized organization in international politics, and therefore, as Checkel (2005) explains, the most likely case for deep transformation of state preferences. This is important, as demonstrating the potential endogenous transformation of state interests via international institutions remains a hard case for IR (Johnston 2001). We are specifically interested in the transformation of foreign policy preferences, and the EU has the most established foreign affairs partnership among IGOs. This partnership, which began in 1970 under the European Political Cooperation, has become increasingly institutionalized (Smith 2004) and was upgraded to the Common Foreign and Security Policy (CFSP) in 1993. EU foreign policy remains a largely intergovernmental field-that is, governed by treaties and consensual decisionmaking. In fostering convergence, it can thus hardly rely on the hard powers and binding legislation that the EU deploys in areas related to its single market. Nonetheless, EU foreign policy is a thick institutional regime with formal and informal rules that have created automatic reflexes of coordination among diplomats-an environment that constructivists have long considered apposite for socialization (Wendt 1994). Significantly, the cumulative number of meetings among ministers and diplomats is higher in the CFSP than in any other EU policy area (Chelotti 2016). These meetings can foster socialization, as they facilitate information exchanges, learning, perspective-taking, and the redefinition of policies and interests. They occur at every level of the EU Council structure: from thirty to forty working parties in the field of foreign policy to more senior bodies, such as the Committee for Civilian Aspects of Crisis Management, the EU Military Committee, and the ambassadorial-level representatives of the Political Security Committee, and monthly meetings between Foreign Affairs ministers.
Another important reason for focusing on the EU is that EU foreign policy has a highly distinctive character, with its own scope, principles, and responsibilities. Indeed, for all the waves of enlargement the EU has undergone, EU policymakers had strong reservations about applicants' preferences conflicting with relevant features of EU foreign policy (see Elvert and Kaiser 2004;Ferreira-Pereira 2006). Prior to applying and joining, applicant countries' preferences were not aligned to EU foreign policy on key issues such as the degree of their pro-Atlanticism (e.g., United Kingdom and Denmark in 1973; many CEE countries in the 2004/2007 enlargement); neutrality (Ireland in 1973; Austria, Finland, and Sweden in the 1995 enlargement); 15 and "Third World" sympathies (Greece, Portugal, and Spain in the 1981/1986 enlargement). 16 The EU has developed its foreign policy along ideational and humanitarian dimensions (Sjursen 2006;Aggestam 2008). Its relations with the rest of the world have been informed by a series of values and commitments to multilateral frameworks. The notion that best captures these attributes is normative power Europe (NPE). Manners (2002) argues that due to its historical origins, hybrid polity, and politico-legal system, the EU can shape the "normal" in world politics and is inclined to act normatively in its foreign policy. Manners identifies nine norms of EU international identity: peace, liberty, democracy, rule of law, human rights, social solidarity, anti-discrimination, sustainable development, and good governance. The EU not only gives a particular European interpretation to these norms, which separates it from liberal powers such as the United States, but also translates these values into specific principles guiding EU foreign policy (Manners and Lucarelli 2006).
Irrespective of the virtues, limits, or contradictions of NPE, EU foreign policy appears normatively motivated. As Forsberg (2011Forsberg ( , 1183 states, if "normative power is best seen as an ideal type … the EU approximates [it] more closely than other" actors. Areas of application of NPE include issues such as poverty and development, the death penalty, gender equality, and human rights (see, e.g., Manners and Lucarelli 2006;Scheipers and Sicurelli 2007). Therefore, we expect new member states to develop these normative values once they come in contact with the EU. If organizations can socialize agents, they should be able to induce them not just to the organization's procedural rules, but also to its constitutive rules, that is, the norms that define the identity and values of an institution (Alderson 2001). The manifestation of this normative socialization in new EU member states (or its absence) enables us to further disentangle the nature of EU foreign policy convergence and better address the ST-IT debate. If during the accession period and, even more clearly, post-membership, we see evidence of internalized normative values-and not only of new material EU-related interests-this would provide a strong indication of deep socialization.

Data and Methodology
To test whether IGO membership leads to foreign policy convergence through a socialization process, we conduct a regression analysis of the effects of EU membership on states' foreign policy positions. We discuss the data and methodology in this section.

Outcome Variables
The outcome variables are derived from annual UNGD addresses taken from the UN General Debate Corpus (Baturo,15 The European Commission indicated that neutrality was a major obstacle for the accession of Austria (see European Commission 1992a), Finland (see European Commission 1992b), and Sweden (see European Commission 1992c)-mainly for its incongruity with EU's sanctions regime and defense policy. 16 Before joining the European Community, these countries voted very differently from EU members in the UNGA (Luif 2003).
Dasandi, and Mikhaylov 2017). To assess whether EU membership leads to preference convergence, we examine whether countries' foreign policy positions move closer to the EU position using measures derived from UNGD statements. We estimate foreign policy positions using the Wordscores approach, which is a method of extracting policy positions from texts using computerized content analysis (Laver, Benoit, and Garry 2003). The approach treats words as data and assumes that the relative frequency with which actors use specific words reflects underlying political positions. The technique derives policy positions by comparing patterns of word use in a set of reference texts to those in other virgin texts to estimate the positions of the latter on the policy dimension (Klemmensen, Hobolt, and Hansen 2007). This requires the policy dimension of interest to be determined a priori, whereby the reference texts are selected because they represent established positions on a specific policy dimension.
We use Worescores to derive countries' positions on two foreign policy dimensions: (1) an EU-Russia dimension, in which the EU and Russian UNGD statements are reference texts, and (2) an EU-US dimension, in which the EU and US statements are reference texts. For both dimensions, the EU reference text is selected as the positive anchor for the dimension, with the Russian and US reference texts selected as the negative anchor for the EU-Russia the EU-US dimensions, respectively. The virgin texts in our analysis are the UNGD statements of all UN members. Therefore, the more similar a country's statement is to the EU statement, the higher its score, and the more similar a country's statement is to the Russia/US statements, the lower its score.
To measure the EU position, we use the UNGD statement of the country holding the EU Council Presidency as a reference text for 1971-2010 and of the President of the European Council from 2011 to 2014. Prior to 2011, the member state holding the rotating EU Council Presidency in a particular semester delivered the UNGD statement on behalf of the EU. Since 2011, the President of the European Council has delivered a separate EU address. 17 The statements and resolutions of the rotating Presidency and the President of the European Council are widely used as measures of EU foreign policy positions (see Smith 2006b). The annual Russian UNGD statement provides the other reference text for the EU-Russia dimension and the US statement for the EU-US dimension. We derive the positions of all UN member states on the two dimensions for 1971-2014 using the raw scores generated by the Wordscores algorithm implemented in the quanteda package (Benoit et al. 2018). 18 These two dimensions represent key axes in international politics with real-world relevance. The main division during the Cold War was between the communist East bloc, led by the Soviet Union, and the liberal West, led by the United States and supported by the EU. Figure 1 presents ridgeline plots of annual Wordscores positions of EU member states for both EU-Russia and EU-US dimensions. This consists of density plots showing the distribution of EU member states' positions on the two dimensions for each year (shown by the 17 The EU statement is drafted by the rotating Council Presidency/cabinet of the European Council president, with comments then provided by member states. For more information on how the EU statement is produced, see Supplementary Information. 18 Using quanteda v1.2.0, we performed standard preprocessing. To avoid any potential concept drift, we ran the Wordscores models separately for each year. Wordscores estimation is done with the classic linear posterior weighted wordclass differences and a smoothing parameter for class affinities 0.5 (Jeffreys prior). For the empirical analysis, we use "raw" untransformed scores, which are multiplied by 100 for presentational purposes in the regression tables. main curves in the center). The figure also indicates the position of the reference texts for the two dimensions, which are the smaller curves on either side of the main distribution curves-with the EU reference text as the positive anchor and the Russia/US reference text as the negative anchor. Hence, the scores on the x-axis indicate the proximity of EU member states to the EU position and the Russia/US position. The graphs are broadly in line with our expectations. During the Cold War, EU members were further from the Russian position, moving closer at the end of the Cold War. After 1989, there was a rapprochement in relations, whereby the EU-Russia dimension largely collapsed. This can be seen in figure 1 with the positions of the EU, Russia, and EU members all moving closer together in the 1990s. However, efforts to integrate Russia into the West ended in the early 2000s and were swiftly followed by Russia's reemergence as a more assertive world power (Trenin 2006). The graph shows EU members moved away from the Russian position during this period. In contrast, the EU has traditionally aligned with the United States, and hence the EU-US dimension represents a "narrower" foreign policy dimension (Steffenson 2005). This allows us to assess whether membership leads countries to shift toward the EU position rather than simply becoming more liberal. The "narrowness" of the EU-US dimension is shown in figure 1 with EU members' positions close to 0 for much of the period up to the mid-1990s. However, in the late 1990s, there is a sharp move away from the United States toward the EU position. This is consistent with the argument that during this period, European countries have sought to counterbalance US power in the United Nations (Voeten 2004). As the EU is noted for its distinct normative foreign policy preferences, including the EU as a reference text for both Wordscores dimensions allows us to directly test whether states adopt these distinct preferences.
We shed light on the broader substantive interpretation of the two dimensions through an analysis of the words with the highest scores on the two dimensions and a topic model analysis. In the former, we examine the words in the UNGD statements that differentiate those countries closest to the EU position from those states closest to Russian or US positions. Figure 2 presents the top 30 words for both ends of the EU-Russia dimension from the Wordscores estimations for the entire Cold War and post-Cold War periods for the years in our analysis. It shows the words most strongly associated with the EU position (left) and the words most closely linked to the Russia position (right) for the two periods. Figure 3 does the same for the EU-US dimension. We use the UNGD texts to interpret these individual words. 19 The analysis of high-scoring words suggests that EU statements cover a much wider range of foreign policy themes, including "softer" and normative issues such as development, human rights, and international cooperation. In contrast, Russia and the United States focus on traditional security issues. Keywords associated with the Soviet position during the Cold War emphasize the Superpower rivalry (e.g., "socialist," "NATO," and "hegemon"). The EU also has a more global perspective. Russia, particularly post-Cold War, has a more regional focus, while the United States has a strong Middle East emphasis. The EU also focuses on more specific human rights and international legal issues than the United States. Furthermore, keywords associated with the US position are more confrontational and emotive (e.g., "tyranny," "hate," and "murder" in figure 3). The EU position, in contrast, emphasizes international cooperation and multilateral initiatives (e.g., "Lomé" Convention, "CSCE," 20 and Convention on "Cluster Munitions").
We also conduct a topic model analysis to identify the main themes in UNGD statements, and the extent states discuss these different themes (presented in the Supplementary Information). The topics with the highest usage by the United States are international security, conflict and terrorism, and Middle East peace; the topics most discussed by Russia are disarmament and colonialism and independence; and the topics with highest engagement by EU member states are Africa peace and security, economic development and the United Nations, Africa region, Latin America region, and sustainable development and climate change. While the United States and Russia have a focus on traditional security themes, the EU engages more with topics, such as development, climate change, and individual rights, and has a more outward foreign policy agenda. Therefore, the topic model analysis and the analysis of words with the highest score support the perspec-20 CSCE' stands for the Conference on Security and Cooperation in Europe, created in the 1970s to reduce East-West tensions. tive that the EU has a more value-oriented foreign policy approach.

Explanatory Variables
The analysis considers the effect of EU membership and stages of accession on countries' foreign policy positions. Our principal explanatory variable is a binary variable that indicates whether a country is an EU member state. We also consider whether a country is an EU official applicant (i.e., the country has lodged the application) or has EU candidate status (i.e., the EU accepts the application and negotiations can start).

Other Control Variables
We include additional variables to control for other factors that may influence countries' foreign policy positions. These additional controls consist of political and economic variables, such as countries' polity scores; GDP per capita (logged); and levels of trade openness, measured by total trade as a proportion of GDP. The data for GDP per capita and trade openness are from the World Bank's World Development Indicators. We also include a dummy variable for whether a country is on the UN Security Council (UNSC) and for the post-Cold War period. In the additional tests presented in the Supplementary Information, we include other control variables such as trade with Russia and the United States to demonstrate the robustness of our findings.

Model Specification
We examine the effects of EU membership, and the pre-accession stages, on state preferences using a linear regression model with two-way fixed effects. This allows us to account for country-specific unobserved factors that are constant over time and address potential omitted-variable bias. The inclusion of year dummies allows us to address time trends that may influence positions. The time period of our analysis is 1971-2014, which includes multiple rounds of EU enlargement. We conduct the analysis on two samples: a global sample (162 countries) enabling us to better ensure that any preference convergence is the result of EU-related socialization rather than the diffusion of international norms that may occur concurrently and a restricted sample of European countries (41 countries) to further test the effect of EU membership and demonstrate the robustness of our results. Table 1 presents the results of the effects of becoming an official applicant of the EU, a candidate status country, and a full EU member state on foreign policy positions using the two Wordscores measures (EU-Russia and EU-United States) for the global sample and the European sample. It shows that EU membership has a statistically significant positive effect on both measures of state preference for the global and European samples. On both the EU-Russia dimension and the EU-US dimension, joining the EU leads countries to move closer to the EU foreign policy position. The results also suggest that there is some preference convergence during the accession period. In the two global models, we find that becoming EU official applicants has a statistically significant positive effect on similarity to the EU position. However, when we restrict our sample to European countries, this effect disappears. The results show that gaining EU candidate status has no significant effect on states' positions, with the exception of the last model. We also find that higher levels of democracy are associated with a shift toward the EU position and that trade openness has a positive effect when we consider only Europe.

Results and Analysis
We conduct additional tests to assess the robustness of these findings. One concern may be that the results are driven by changes in the Russia and US position over time rather than states moving closer to the EU position. While the use of year fixed effects accounts for time trends, we further address this by using outcome variables based on alternative text similarity measures. We employ cosine similarity and Jaccard similarity measures, and utilize a new approach to measuring text similarity, the word embeddings-based Word Mover's Distance (WMD) (Pomeroy, Dasandi, and Jankin Mikhaylov 2019). We discuss these alternative measures in the Supplementary Information, but importantly they provide measures of similarity with the EU that are unrelated to US and Russia statements. We also include additional variables in our regression models, such as countries' trade with Russia and the United States. The results, provided in the Supplementary Information, demonstrate that the effect of EU membership on preference convergence is consistent across different models. EU membership has a positive effect on similarity to the EU position when we use alternative measures of preference similarity and include additional controls. This provides strong evidence that EU membership leads to a shift toward the EU foreign policy position. In contrast, the effect of EU official applicant and EU candidate status on preference similarity is sensitive to model specification.
These findings provide stronger support for the ST than the IT. Based on the ST, we would expect convergence in state preferences to be greatest once a country becomes a full member state (Hypothesis 1), while according to the IT, convergence is greatest during the accession period (Hypothesis 3). On both the EU-Russia and EU-US dimensions, we find that EU membership has a much larger effect on similarity to the EU position than the pre-accession stages. Hence, our results provide strong support for Hypothesis 1 and suggest Hypothesis 3 can be rejected. We find weaker evidence of incentivization effects during the accession process. The results show that with the global sample, becoming an official applicant has a significant effect on preference convergence, suggesting incentivization effects may occur at the start of a country's exchanges with the EU. However, these effects disappear with the European sample and with alternative model specifications. This lack of robustness suggests that preference convergence during the accession period may be inconsistent-though we do find some evidence of partial convergence during this phase, and so Hypothesis 2 cannot be rejected. In any event, according to the IT, the largest convergence should occur during the accession period, as states seek to ensure they gain the benefits of membership-and not at the point of membership. However, the results in table 1 show that EU membership has the largest effect on preference similarity on both di-mensions. Furthermore, the effect of EU membership on preference convergence is robust and largest across the alternative model specifications, supporting the ST.
To further understand the effects of IGO membership on state preferences, we examine convergence in the years after countries gain EU membership. According to the IT, we should see divergence after a country joins an IGO (Hypothesis 4). Table 2 presents the results of the analysis of the effects of EU membership years on preferences. As our analysis specifically focuses on whether the socialization of state preferences continues after a country becomes an EU member state, we restrict the sample to EU members. 21 The results show that EU members do not move away from the EU position over time. Hence, we find no evidence of divergence-strong or weak-from the EU position in the years after membership, and so Hypothesis 4 is rejected. In fact, on the EU-Russia dimension, there is a positive effect of the length of time of EU membership, whereby member states continue to move toward the EU position and away from Russia. On the EU-US dimension, we find no significant effect of EU membership years on countries' foreign policy positions. Hence, the preference convergence that occurs when countries join the bloc appears to be an enduring shift.

Examining the 2004 EU Enlargement
We next consider the 2004 EU enlargement in more detail to demonstrate the practical policy relevance of our results and to provide further analysis of preference convergence and the ST/IT debate. In particular, we aim to shed greater There is little to suggest that these countries aligned their preferences with the EU prior to becoming members. If, after 1989, the CEE countries oriented their political and economic relations toward the West, their foreign policies still differed significantly from EU foreign policy at the time of their application submission to the EU. In 1995, they aligned with CFSP declarations only 25.5 percent of the times they were invited to do so. The major discrepancies were related to human rights, the EU's immediate neighborhood, and the Middle East (Regelsberger 2003). As the European Commission noted at the time, substantial incongruities also existed in the areas of aid and development, 22 minority rights, and defense policy. 23 Consequently, "the early literature expected that Eastern enlargement would cause a lower cohesion of foreign policy preferences among EU member states" (Finke 2020, 192) and risk transforming EU foreign policy identity (Sjursen 2005).
There were serious concerns among EU policymakers that the accession of the new "EU10" countries would undo 22 Most CEE countries did not have a development aid budget. For Lithuania and Slovenia, the Commission concluded in July 1997 that significant changes were needed for alignment with the EU (see European Commission 1997a,b). 23 The CEE countries prioritized NATO membership over EU defense issues.
The cases of Cyprus and Malta were even more problematic. Cyprus was expected to "give up its membership of the Non-Aligned Movement … in which it continues to participate actively" (European Commission 1993a, 13). The European Commission (1993b) viewed Malta's neutrality and nonaligned status as conflicting with the CFSP. EU foreign policy cohesion precisely along the two dimensions that we have examined, namely the EU-US and EU-Russia axes. In the post-Cold War era, many CEE countries looked to the United States as the ultimate guarantor of their security. At the time the European Council was finalizing the decision to admit the new states (December 2002), the EU was in the midst of one of its biggest foreign policy crises due to the Iraq war. There was a sharp split between the more Atlanticist UK, Spain, and Italy on the one side and Germany, France, and Belgium on the other-with the latter group expressing strong criticism of US actions. The CEE countries took an unabashed pro-American position. Three of these countries-Czech Republic, Hungary, and Poland-together with five "West European" states, signed the "letter of the eight" in January 2003, which offered support for the Bush administration's stance against Saddam Hussein's regime. Soon after, the Vilnius 10 group expressed willingness to participate in the US-led coalition for the disarmament of Iraq. 24 24 The Vilnius group consisted of Albania, Bulgaria, Croatia, Estonia, Latvia, Lithuania, Macedonia, Romania, Slovakia, and Slovenia.
The then French President, Jacques Chirac, reacted angrily to the diplomatic statements of these CEE states, labeling their support for the United States as "childish" and "dangerous" (NYT 2003). He reminded these countries of the value of EU membership, stating when "you are in the family … you have more rights than when you are asking to join and knocking on the door" (NYT 2003). These divisions were also fueled by what Donald Rumsfeld called "old" versus "new" Europe, where the former referred to France and Germany, which were judged to be fusty countries, mired in international organizations, transnational rules, and postmodern visions (Kagan 2004). In contrast, he argued a "new Europe" was emerging, comprising of the CEE countries that would join NATO (and the EU) a few months later, and were full of energy, vigor, and initiatives. He argued that this new Europe was destined to leave the old Europe behind, and move closer to the position of the US administration. As Rumsfeld said in an interview, the "vast numbers of [new] countries in Europe [are] not with France and Germany … they're with the United States" (US Department of Defense 2003).

Figure 5.
Keywords in EU10's UNGD statements before and after EU membership. Note: Top 30 statistically distinct phrases (bigrams) in the UNGD statements in the ten years before (bottom) and after membership (top).
To a lesser degree, there was also possible Russian influence on EU foreign policy, especially among some CEE states. Since the mid-2000s, EU-Russia relations have deteriorated. The Baltic states and Poland have been particularly vocal regarding the threats posed by Russia under Vladimir Putin's leadership. In contrast, some states have been susceptible to Russian influence. The precise list of these countries varies, depending on shifts in domestic politics, but typically it includes Hungary, Slovakia, Cyprus, Greece, and possibly Italy (Tamkin 2017a). These countries' desire to build closer ties with Moscow has led to them being called Putin's "Trojan horses" (Orenstein and Kelemen 2017). Russian information campaigns are said to have reached and influenced politics in the entire CEE area. In the Czech Republic, for instance, President Milos Zeman openly called for sanctions on Russia to be lifted, claiming Russia to be a more important partner than France (Tamkin 2017b).
In these circumstances, we might expect that once the EU10 countries had reaped the benefits of membership in 2004, they would, to some degree, re-orientate their foreign policy toward the United States or Russia. Figure 4 shows the positions of the ten countries on the EU-Russia and EU-US Wordscores dimensions, without controlling for country-specific factors. The figure indicates that this divergence did not occur. In fact, it shows quite the opposite: on gaining membership, and in the years after, these countries moved toward the EU position. On the EU-US dimension (bottom), these countries were positioned between the EU and the United States in the 1990s. All ten countries remained equidistant between the EU and United States in the second half of the 1990s, when they had filed applications for EU membership. They shifted toward the EU position around the time they became official candidates (December 1997), and this move toward the EU gained momentum once they became full member.
Similar dynamics can be observed with the EU-Russia dimension. In the early 1990s, Europe and Russia moved considerably closer, whereby the EU-Russia foreign policy dimension de facto collapsed. The significance of this axis re-emerged in the late 1990s and increased significantly after the initial years of Putin's leadership. Figure 4 shows that the ten countries began to move closer to the EU in the late 1990s, shifting from a slightly pro-Russian position. This effect intensified considerably just before accession and then increased consistently and ostensibly from 2004 when they became full member states. Therefore, these figures suggest that the EU10's preferences have become, and remained, closely aligned with those of the EU.
The last part of our analysis delves into the normative content of the EU10's alignment with EU foreign policy. Our analysis of the EU10's UNGD statements helps to corroborate that the preference convergence that occurs once these states join the EU is due to socialization processes in the EU rather than other factors, such as integrated policies that create common material interests among members. Figure 5 shows the keywords in the CEE countries' UNGD statements in the ten years before and after accession (discussed further in the Supplementary Information). It reveals the shift from a narrower focus on security and regional issues in the ten years prior to gaining EU membership (bottom) to a broader global focus in the ten years after accession (top), which includes an emphasis on issues such as climate change, international development, and gender equality. If preference convergence was driven by the changing material interests of these states, we might expect to see keywords that emphasize strategic issues, such as those linked to EU trade and energy policy, or regional issues. Instead, in accordance with our argument, the keywords demonstrate that after 2004, there is greater emphasis on normative issues (e.g., "indigenous peoples," "gender equality," "development assistance," "mass atrocities," and "climate change"), consistent with other EU member states. In addition, the post-2004 EU10's UNGD statements reveal shifts in the ways these states perceive and discuss existing foreign policy concerns. This can be seen, for example, in how countries such as Estonia, Latvia, and Poland discuss Russia in their UNGD statements before and after they joined the EU. Prior to joining, these countries tended to emphasize historical grievances and the desire to improve bilateral relations with Russia, while after becoming members their references to Russia focus more on issues such as human rights, international law, and multilateral initiatives in line with figure 5. 25 25 For example, references to Russia in Latvia's UNGD statements in the 1990s focus on the Soviet occupation of the country and initiatives to improve bilateral relations, while in Latvia's 2004 and 2008 UNGD statements, they focus We also examine the topics discussed in the EU10 countries' UNGD statements before and after they joined the EU. To do this, we draw on the topic model analysis discussed above (and presented in the Supplementary Information) and select the five topics that feature most prominently in EU members' UNGD statements. As previously noted, these topics highlight the global and normative foreign policy agenda of the EU: Africa peace and security, economic development and the UN, Africa region, Latin America region, and sustainable development and climate change. Figure 6 presents the change in the proportion of these "EU topics" in the CEE countries' UNGD statements. There is low engagement by the EU10 in the 1990s, with an increase at the end of the decade (see top). The figure shows a significant rise in the proportion of the five topics in the EU10's statements once they join the EU. Figure 6 also shows how the EU10's engagement with these topics compares with the founding EU member states ("EU6"), 26 the United States, and Russia. The figure shows that even in the late 1990s and early 2000s, as the EU10 increase their engagement with these topics, the gap in topic proportion between the EU10 and EU6 remains more or less constant. It is only once the EU10 countries join the EU that we see the gap between the EU10 and EU6 close. Furthermore, the figure shows that EU10 and EU6 countries' engagement with these topics differs considerably from the United States and Russia. Hence, the analysis demonstrates that once the CEE countries joined the EU, they devoted a higher proportion of their UNGD statements to normative topics and engaged with these themes at very similar levels to older EU members. This higher level of engagement is sustained over time suggesting these countries have internalized these preferences. Therefore, both the keywords analysis and the topic model analysis demonstrate that the CEE countries adopted EU foreign policy preferences once they become members, providing further evidence of socialization processes leading to state preference convergence.

Conclusion
The question of whether socialization processes within IGOs can transform state preferences is central to IR. It offers insights into wider debates on the nature of state preferences and to issues such as the impact of international institutions, state compliance, and the evolution of national identities (Johnston 2014). However, attempts to systematically examine this question have struggled to separate socialization processes from other factors that may lead to foreign policy alignment. This paper helps to address this issue. We adopt a new approach to measuring state preferences by applying text analysis to countries' UNGD statements. The absence of coordination among states in producing these speeches suggests that material or social components of external influence on them are minimal. The paper also addresses the IGO's coercive powers and the potential incentive effects of IGO membership by focusing on the EU.
Our analysis provides robust empirical evidence that IGO membership can lead to state preference convergence through a socialization process. We find that EU membership is associated with a shift toward the EU position on the EU-Russia and EU-US foreign policy dimensions derived on democracy and civil society promotion, Russia's adherence to international law in the Georgian crisis, and the need to establish an EU monitoring mission in Georgia. 26 These are Belgium, France, Germany, Italy, Luxembourg, and the Netherlands.
from UNGD statements. This effect occurs when controlling for any shifts during the accession period. We also show that this socialization has a normative and ideational dimension. By focusing specifically on 2004 enlargement, we demonstrate that on joining the EU, the CEE countries shifted their foreign policy focus from security and the near neighborhood to more global and value-oriented issues, such as sustainable development, the African region, and gender equality.
In examining the effects of IGOs on member state preferences, our analysis has focused on the EU. As such, our findings are limited to this case. We focused on the EU because our objective has been to test the "purest" form of socialization-namely the transformative potential of IGOs on preferences, while controlling for the incentives for alignment that arise through the accession process. It also has a distinctive and normative foreign policy, which has allowed us to further differentiate between convergence due to material factors and socialization. The EU is the most institutionalized IGO in world politics and hence the most likely case to observe this deep transformation of preferences. There are, however, grounds to believe that our findings can be applied beyond the EU case-and not just because some suggest growing similarities between the EU and other IGOs (Johnston 2005)-or because scholars have argued that IGOs such as the OECD and WTO have a similar capacity to promote policy convergence (Cao 2009). We have investigated preference convergence in EU foreign policy-an area largely governed by treaties, unanimity, and member states, which makes cooperation in this domain similar to less institutionalized IGOs. In fact, the EU's institutional capacity is rather weak in foreign policy, and EU institutions lack the coercive and binding powers they have in such areas as the single market, competition, or fiscal policies.
A fruitful area of future research would be to expand the analysis across different IGOs. This would help to shed light on which types and aspects of IGOs foster socialization. While studies have assessed the cumulative effects of shared IGO membership on convergence (e.g., Cao 2009;Fink 2013;Taninchev 2015), we believe that our approach can help evaluate the socialization capacity of single IGOs. An advantage of using UNGD statements to assess socializationbased preference convergence is the relative absence of external constraints and coordination mechanisms in delivering these statements. The use of speeches also allows us to better test the correspondence between the normative fabric of an IGO and preference convergence. It is possible to focus on specific issues, topics, and debates of a given IGO and observe whether member states reveal similar preferences and patterns. Taking the example of NATO or the Organization for Security and Co-operation in Europe (OSCE), this approach would be able to assess, in a more fine-grained way, whether NATO's security assessments and concepts or the OSCE's values have socialized member states' preferences in the field of security. In providing clear evidence of a socialization effect on member state preferences in the case of the EU, this paper has shown that IGOs can alter state preferences through socialization, thereby contributing to a fundamental debate in IR.

Supplementary Information
Supplementary information is available at the International Studies Quarterly data archive.