Why International Organizations Commit to Liberal Norms

Recent decades have witnessed the emergence and spread of a broad range of liberal norms in global governance, among them sustainable development, gender equality, and human security. While existing scholarship tells us a lot about the trajectories of particular norms, we know much less about the broader patterns and sources of commitments to liberal norms by international organizations (IOs). This article offers the first comparative large-N analysis of such commitments, building on a unique dataset on IO policy decisions over the time period 1980–2015. Distinguishing between deep norm commitment and shallow norm recognition, the analysis produces several novel findings. We establish that IOs’ deeper commitments to liberal norms primarily are driven by internal conditions: democratic memberships and institutional designs more conducive to norm entrepreneurship. In contrast, legitimacy standards in the external environment of IOs, often invoked in existing research, mainly account for shallower recognition or “talk” of norms.


Introduction
Recent decades have witnessed the emergence and spread of a broad range of liberal norms in global governance. From hardly being recognized as global policy principles in the 1970s, norms such as sustainable development, gender equality, human security, good governance, and deregulation have been integrated into the policy portfolios of many international organizations (IOs). This development has inspired an impressive literature on norms in global governance, examining, for instance, the importance of norm entrepreneurship (Finnemore 1993;Keck and Sikkink 1998;Price 1998;Risse, Ropp, and Sikkink 1999;True and Mintrom 2001;Joachim 2003), the evolution and contestation of norms (Finnemore and Sikkink 1998;Sandholtz 2007;Kelley 2008;Krook and True 2012;Towns 2012;Wiener 2018;Deitelhoff and Zimmermann 2019), and the processes of norm effectiveness (Acharya 2004;Simmons 2009;Sikkink 2011;Risse, Ropp, and Sikkink 2013;Zimmermann 2017).
Existing scholarship tells us a lot about the trajectories of particular norms. But we know less about the broader patterns and sources of commitments to norms in global governance, especially as they are espoused by IOs. How have norms spread across IOs and over time? Are some norms adopted by IOs more quickly or broadly than others? Are some IOs generally leaders or laggards in the uptake of new norms? And, perhaps most importantly, what explains variation in norm commitments by IOs?
Commitments to norms by IOs are consequential for states and societies. Norms focus attention on particular problems, prescribe or proscribe behaviors, and may generate distributional effects. When integrated into the policies of an IO, norms guide subsequent decisions by the organization, affect its allocation of resources, and place demands on its member states (Barnett and Finnemore 2004;Martin and Simmons 2012). Such policy commitments by an IO may in turn strengthen norms as prescriptive principles, contributing to further norm cascades (Finnemore and Sikkink 1998;Risse, Ropp, and Sikkink 1999). Empirical research documents effects of norm commitments by IOs across a variety of policy fields, including election monitoring (Kelley 2012), gender equality (Simmons 2009), global development (Park and Vetterlein 2010), human rights (Risse, Ropp, and Sikkink 2013), international security (Gheciu 2005), and sustainable development (Biermann and Siebenhüner 2009). Whether, how, and why norms spread among IOs therefore matters for real-world concerns in international politics.
A common explanation of liberal norm commitment by IOs privileges a diffusion of global scripts. It suggests that IOs are driven by concerns of organizational legitimacy and look to their external environment when deciding whether to embrace a new norm. The more prominent a norm is within an IO's organizational field and wider society, the more likely it is to be adopted. The result is a global diffusion of liberal policies and institutions broadly considered appropriate. This explanation suggests IOs may demonstrate homogeneity in norm commitment, despite different functional tasks, memberships, and geographical origins. This explanation informs recent accounts of the spread of accountability mechanisms, civil society instruments, good governance policies, human rights provisions, and parliamentary institutions across IOs (Grigorescu 2010;Risse, Ropp, and Sikkink 2013;Börzel and van Hüllen 2015;Risse 2016;Rocabert et al. 2019;Sommerer and Tallberg 2019).
However, this account struggles to explain important patterns of variation in IOs' commitments to norms. IOs differ in whether they have policies dedicated to a norm. For example, the United Nations (UN), African Union (AU), and Commonwealth, among others, have full-fledged policies on gender equality, while others like the Southern African Development Community (SADC) and Caribbean Community (CARICOM) do not. Similarly, some IOs are quicker than others to adopt norms. The European Union (EU) adopted a policy on sustainable development already in 1992, while the Organization for Islamic Cooperation (OIC) did not do so until 2012.
We advance an alternative explanation of IOs' commitments to liberal norms that account for variation across IOs by privileging actors and institutions in the internal machinery of IO policy-making. Starting from the premise that entrepreneurs aim to convince IOs of new norms, it theorizes the conditions under which organizations are more likely to be receptive to such overtures. Specifically, it highlights two conditions as decisive for IOs' openness to liberal norm entrepreneurship and their resulting commitments. First, IOs whose memberships are more democratic offer more hospitable environments for state, supranational, and transnational entrepreneurs promoting new liberal norms. Second, IOs whose institutional designs are characterized by more pooling of decision-making authority, more delegation to supranational bodies, and more access for transnational actors (TNAs) are more conducive to norm entrepreneurship. This account suggests that liberal norm commitment by IOs varies in foreseeable ways and identifies which conditions are particularly favorable to commitment.
We test these explanations through the first comparative large-N analysis of norm commitment by IOs. Building on a unique dataset on IO policy decisions, we map and explain the spread of eight liberal norms across eighteen multi-issue IOs from 1980 to 2015. To identify norm commitment, we collected and analyzed all policy decisions taken by the main intergovernmental decision-making bodies of these IOs over the full time period. To differentiate between an IO's commitment to a norm and shallower "talk" about a norm, we use two alternative measures. We identify commitment as the first codification of a norm through a full policy devoted to its implementation, and recognition as the first rhetorical reference to a norm in IO policy. This research design breaks with previous scholarship on norms, which seldom is comparative in orientation and so far has not differentiated systematically between commitment and talk. 1 The findings lend strong support to our argument. The analysis shows that norm commitment by IOs is driven by internal conditions conducive to norm entrepreneurship: the democratic density of memberships and the degree to which institutional designs provide for pooling, delegation, and TNA access. As a result, norm commitment is concentrated to the liberal core of IOs with largely or entirely democratic memberships and institutional rules that facilitate norm 1 For two exceptions, see Park and Vetterlein (2010), who study the adoption of nine policy norms by the International Monetary Fund and the World Bank, and Börzel and van Hüllen (2015), who study the adoption of good governance policies across twelve regional IOs. entrepreneurship. In contrast, the findings offer more limited support for global scripts about appropriate action as an explanation of norm commitment by IOs. This does not mean that this account is all wrong. If we focus on shallow recognition of norms by IOs, we find such talk or lip service is best explained by a norm's prominence in an IO's external environment. Taken together, these findings suggest an important corrective to existing knowledge: while global scripts help us to understand why norms spread across IOs at a rhetorical level, accounting for deeper norm commitment requires an appreciation of the role of democracies and institutional rules that facilitate norm entrepreneurship.
Our findings have several broader implications. First, they demonstrate how careful empirical measurement may allow us to identify the multiple logics of action that drive IOs, as well as the scope conditions under which they hold. Second, our findings show how a large-N approach to norms can complement prevailing qualitative work by identifying general sources of variation in the spread of norms. Third, they raise critical questions about the performance of IOs, showing that organizations frequently stop at rhetorical recognition, indicating less effectiveness of norms than is often assumed based on public discourse. Finally, our findings suggest why liberal norms may face retrenchment in global governance: while they were never fully supported outside democratic IOs, liberal norms are now weakened from within as well, as authoritarian tendencies gain ground even within the democratic core.
The article proceeds in four steps. First, we introduce our dataset and illustrate how IO commitment to norms varies. Second, we outline two theoretical explanations for when IOs might be inclined to make norm commitments. Third, we assess the explanatory power of these accounts based on multivariate analysis. Finally, we summarize the findings and discuss the broader implications for research on global norms, IOs, and the liberal international order.

Liberal Norm Commitments by IOs, 1980-2015
We describe the central patterns in liberal norm commitments by IOs based on a novel dataset covering eight liberal norms and eighteen IOs over the period 1980-2015. The norms included are democracy promotion, gender equality, good governance, sustainable development, deregulation, debt relief, human security, and responsibility to protect (R2P). These principles qualify as norms because they articulate shared standards of appropriate behavior (Finnemore and Sikkink 1998). This does not mean that they were not once contested or are now taken for granted. Indeed, as we will show, commitments to these norms vary significantly among states and IOs.
We have selected these eight norms based on four considerations. First, they are all grounded in liberal political ideals, stressing individual freedom, equal rights, economic openness, social justice, and democratic governance (Moravcsik 1997, 525-27). Since liberalism is a multivalent and contested concept (Gaus 2018), we have chosen a broad and inclusive approach. While these norms all appreciate liberty as a core political value, each norm may be more strongly associated with a distinct strand of liberalism. For instance, the norms of gender equality, human security, and R2P build on individual rights as emphasized in classical liberalism. Deregulation builds on ideas of limited government interference in the market associated with economic liberalism. Debt relief and sustainable development accord with conceptions of social justice inherent to social or new liberalism. Democracy promotion and good governance emphasize principles common to most strands of political liberalism. These norms contrast with illiberal international norms, such as non-interference, which privileges state sovereignty at the expense of individual and humanitarian values.
Second, these eight norms cover a wide range of issues in global governance, from development and environmental protection to security and human rights. They also address the broad domains where norms have been identified in previous research (see Table A1 in the online appendix). This variety allows us to go beyond norm-specific accounts and assess general explanations of norm commitment by IOs. Third, these norms have all emerged and spread over the past four decades, and several have reached a status of prominence in international political discourse (see Figure  A1 in the online appendix). Concentrating on norms that have emerged during the same time period facilitates comparative analysis. Fourth, while their substantive content differs, these eight norms are comparable with regard to their nature and scope. All are regulative self-binding norms, which prescribe or proscribe certain behavior of IOs and their member states, rather than other binding norms directed at third parties, such as the norm of corporate social responsibility.
Our sample of IOs (Table A2 in the online appendix) is guided by three considerations. 2 First, all eighteen IOs have a multi-issue orientation, ensuring that the eight norms are potentially relevant to their activities. 3 Several of the norms we study make little sense for specialized single-issue IOs. Second, the sample has a wide geographic scope. It includes six global IOs and twelve regional IOs, three from each of the four major world regions (Europe, Africa, Americas, and Asia-Pacific). This stratified selection mirrors the distribution in the full population of fifty-four multi-issue IOs included in the correlates of war-intergovernmental organization (COW-IGO) dataset and corresponds to one-third of this population (Table A3 in the online appendix). Third and crucially, this sample ensures variation on the explanatory factors we examine, including the democratic composition of IO memberships and the institutional design of IOs.
In accord with earlier research, we look at formal policy decisions to identify norm commitment (Finnemore and Sikkink 1998;Park and Vetterlein 2010;Börzel and Stapel 2015). To identify norm commitment, we collected and analyzed all policy decisions taken by each IO's main intergovernmental decision-making body between 1980 and 2015. 4 The policy decisions, such as resolutions, declarations, and decisions, were gathered from each IO's electronic and physical archives, producing a corpus of about 45,000 documents. We focus on the policy decisions of the main intergovernmental decision-making body because they represent the collective will of the IO and constitute the strongest possible form of political commitment an IO can make. In comparison, policies developed in other parts of an IO's machinery are less optimal indicators. IO bureaucracies produce reports or guidelines promoting norms, but the documents do not necessarily reflect the collective will of the IO membership. Similarly, lower level IO organs may develop policy documents invoking norms, but these do not entail the same level of political commitment as top-level decisions. In other words, it is precisely because they are taken by the full membership at the pinnacle of the organization that policy decisions by the main intergovernmental decision-making body represent a good approximation of norm commitment by IOs.
We develop two measures to capture the degree to which an IO embraces a norm, which we subsequently use in searches of the corpus. 5 The principal measure-norm commitment-represents the first full, comprehensive policy devoted to implementing or promoting a norm. As these policy decisions emerge from negotiations in which an IO's membership agrees to codify a norm, norm commitment is probably closest to what most people think of as norm adoption. It reflects the first time an IO makes an assurance that it will integrate the norm in its operations by mobilizing resources and developing a framework to follow through on the decision. In our coding, full policies are identified through documents in which the norm represents the main content of the policy decision (see Table A6 in the online appendix).
The second measure-norm recognition-represents the IO's first rhetorical reference to a norm in its policy output. First references suggest that an IO's main decision-making body recognizes a norm and regards it as sufficiently important to merit a discussion and formal recording at the highest political level. However, they do not amount to a firm commitment to a norm and thus represent only a shallow endorsement of a norm, not a dominant frame for multilateral negotiations. When such first references are not followed by deeper commitments, they may be accurately described as rhetoric, talk or lip service.
One important difference between commitment and recognition refers to norm robustness (Deitelhoff and Zimmermann 2019). We assume that the robustness of norms in discourse and practice is higher in the case of commitment compared to recognition (Legro 1997, 35). Just like structural factors that shield norms against challenge (Wiener 2018), the legal character of a full policy makes a norm less prone to weakening by contestation (Deitelhoff and Zimmermann 2019, 10).
The two measures can be illustrated by comparing the adoption by the UN and the Shanghai Cooperation Organization (SCO) of the norm of sustainable development. The UN General Assembly (UNGA) made a first reference to "sustainable development" in Resolution 35/73 in 1980. The main topic of this resolution is a different one, and thus it counts as norm recognition. When the UNGA adopted Resolution 47/190 in 1992, endorsing the Rio Declaration on Environment and Development, the decision urged member states to ensure "the means of implementation [of] Agenda 21, stressing in particular the importance of financial resources and mechanisms." The 1992 commitment to Agenda 21 paved the way for subsequent deepening of UN policy on sustainable development, including the 2015 Sustainable Development Goals. Since the 1992 resolution represents the first UN decision setting out a collective pledge to implement sustainable development, incorporates extensive and specific provisions, and provides a framework for subsequent policy, it counts as norm commitment. In contrast, when the Summit of the SCO in 1998 adopted a 5 The main search term was the exact wording of the norm. We also searched for widely used acronyms ("R2P") and reformulations ("promotion of democracy"; "relief of debt"). For deregulation, we included the terms "liberalization" and "privatization." communiqué that set out the goal of achieving "durable peace and sustainable development of the region," the document only made a cursory reference to the norm and did not pave the way for successive and more far-reaching commitments. We therefore code it as norm recognition. Using this two-fold measure, we map how the eighteen IOs embrace our eight norms through policy decisions between 1980 and 2015. 6 Figure 1 suggests several observations. First, norm commitment (as indicated by the black markers) varies across norms and IOs. The majority of IOs have committed to sustainable development (fifteen IOs), gender equality (eleven IOs), and good governance (nine IOs), while commitments by IOs are rarer for some norms, specifically human security, R2P, and debt relief.
Second, some IOs commit to norms earlier than others. In the case of gender equality, for instance, we saw a first wave of commitments in the early 1980s, followed by a second wave in the early 2000s, and then a third after 2010. And while the EU and the Organization for Economic Cooperation and Development (OECD) already committed 6 In some cases, the first reference and full policy are found in the same document or year.
to the norm of deregulation at the beginning of our observation period, IOs like the Council of Europe (COE), the UN, and the Asia-Pacific Economic Cooperation (APEC) followed about fifteen years later.
Third, some IOs commit to a larger number of norms than others. The UN stands out with full policies on all eight norms, followed by the EU and the COE with five commitments each (see also Figure A2 in the online appendix). Three IOs have committed to only one norm-Association of Southeast Asian Nations (ASEAN), SADC, and CARICOM-while the Arab Maghreb Union (AMU) and SCO have committed to none.
Finally, not all IOs commit to norms that they recognize. Norm recognition is relatively high; six of the eight norms are recognized by at least half of the sample of IOs. Two norms, debt relief and human security, are predominantly endorsed at this shallower level suggesting that their international presence is a function of norm talk not matched by deeper commitments. Some IOs seem more inclined to follow up recognition with a full policy. For example, the UN committed to seven norms that it had previously recognized. The time span between these two levels varies considerably, from one year (e.g., good governance in the COE) to twenty plus years (e.g., sustainable development in the AU). Also, some IOs distinguish themselves by rarely going beyond norm recognition; notably, SADC has recognized five norms but only committed to one (sustainable development).

Explaining Norm Commitment by IOs: Theories and Hypotheses
The previous section showed that liberal norms have spread across IOs over time, while also highlighting significant variation in commitment. In this section, we outline two alternative accounts of this development: an explanation centered on the diffusion of global scripts and our privileged explanation centered on conditions internal to an IO that make it conducive to norm entrepreneurship.

External Conditions: Diffusion of Global Liberal Scripts
A common explanation of norm commitment by IOs privileges the diffusion of global liberal scripts, understood as templates of appropriate liberal standards. This explanation is grounded in the sociological institutionalist notion that organizations adopt features considered legitimate by their external environment. Accordingly, institutions and policies do not spread across organizations because of their functional virtues, but because the wider environment supports and legitimizes them (Meyer and Rowan 1977). The external environment consists of peers in the organizational field as well as the broader society in which organizations are embedded (DiMaggio and Powell 1983;Meyer et al. 1997). The environment matters, not because it presents competition, but because it conveys standards of appropriateness, and conformance to these standards gives organizations legitimacy, resources, stability, and sustainability. This logic is usually referred to as emulation in the literature on diffusion and has been used to explain patterns of isomorphism in various contexts (see Simmons et al. 2006;Gilardi 2012). In IR, this approach has become increasingly influential as scholars have recognized its potential to explain organizational similarities in world politics (Finnemore 1996). Important contributions draw on this logic to explain the spread of macro-institutional arrangements in global governance, such as regionalism (Risse 2016), multilateralism (Ruggie 1992), and transnational governance (Dingwerth and Pattberg 2009). In research on IOs, this perspective accounts for why these organizations sometimes demonstrate significant homogeneity, despite different functional tasks, memberships, and geographical origins. A range of studies suggest that IOs look to their external environment when deciding whether to adopt a particular design or policy, giving rise to homogeneity across organizations. While the terminology used to describe these processes varies, the literature increasingly refers to a diffusion, emulation, or mimicking of global liberal scripts. In differentiating these claims from our alternative account, it is important to recognize both the logic at work-diffusion of externally legitimated models-and the resulting pattern of adoption-a spread beyond the liberal core of IOs with democratic memberships.
A number of contributions highlight how liberal democratic designs have spread across IOs, reflecting templates considered appropriate among peers and in society. Several studies show how a growing number of IOs have established international parliamentary institutions to legitimize themselves (Rittberger and Schroder 2016; Rocabert et al. 2019). This global trend extends to IOs whose member-ships are authoritarian, pointing to the strong impact of externally legitimated models. Other studies submit that IOs' increasing tendency to open up to civil society actors reflects the diffusion of a new global participatory practice, spreading to IOs outside the liberal core, such as the AU and OIC (Steffek et al. 2008;Sommerer and Tallberg 2019). A similar argument has been advanced to account for the spread of accountability mechanisms, transparency policies, and dispute settlement mechanisms across IOs (Grigorescu 2010;Alter 2012). One telling illustration of this explanatory logic finds that prior adoption by peers in an organization's environment trumps all IO-specific factors in explaining accountability mechanisms, including democratic memberships (Grigorescu 2010).
Likewise, several contributions explain the spread of liberal policies among IOs by invoking the notion of global scripts. According to one influential study, human rights originated in the West, but subsequently gained universal prescriptive status: "[T]he norm-guided logic of appropriateness now requires both governments and non-state actors in world society to at least pay lip service to the idea that there are such things as fundamental human rights" (Risse and Ropp 2013, 9). Another study traces the spread of policies on democracy, human rights, the rule of law, and anti-corruption across regional IOs over recent decades, submitting that "these standards for legitimate governance institutions form part of a global script for "good governance," which enjoys universal acceptance" (Börzel and Stapel 2015, 24). Similarly, emulation is the favored explanation for the global spread of regional integration-often modeled on the EU and extended to non-Western IOs such as the Andean Community CAN and ASEAN (Jetschke and Murray 2012;Risse 2016;Lenz and Burilkov 2017).
This logic of global liberal scripts yields two hypotheses, focused on the main sources of legitimate standards in the external environment. The first highlights the importance of organizational fields, suggesting that IOs are more likely to commit to a liberal norm if peer IOs already have done so. The second emphasizes the importance of the broader societal environment in which IOs are embedded, stipulating that IOs are more likely to commit to a liberal norm that is prominent in wider society.

H1:
The greater the prominence of a liberal norm among peers in an IO's organizational field, the more likely the IO is to commit to this norm. H2: The greater the prominence of a liberal norm in wider society, the more likely the IO is to commit to this norm.

Internal Conditions for Norm Entrepreneurship: Democratic Density and Institutional Design
The descriptive patterns identified earlier raise doubts about the extent to which the spread of global liberal scripts can explain norm commitment by IOs. Importantly, commitment to liberal norms appears to be more selective than a logic of global scripts would suggest. While IOs' recognition of norms is reminiscent of an isomorphic pattern, 7 IOs' deeper commitment to norms appears to follow a different logic. In the following, we therefore advance an alternative account that privileges actors and institutions in the internal machinery of policy-making. This account 7 For several norms-deregulation, gender equality, good governance, and sustainable development-the pattern of recognition over time resembles the typical S-curve of diffusion. While this is not readily visible in Figure 1, because the figure orders IOs according to norm commitment (not recognition), it shows more clearly in Figure A1 in the online appendix.
draws from earlier research on democratic density and institutional design to identify conditions internal to an IO that shape the success of liberal norm entrepreneurship. It integrates theories of how memberships and institutions matter into a coherent account to explain IO commitments to liberal norms. It suggests that IOs are more conducive to liberal norm entrepreneurship when their memberships are more democratic and their institutional designs involve more delegation, pooling, and access, making them more likely to commit to liberal norms. We start from the premise that IOs in general face efforts by entrepreneurs to convince them of the appropriateness of new liberal norms. While simplifying, this assumption is consistent with a rich literature focused on states, supranational bureaucracies, and TNAs as norm entrepreneurs. Work on state entrepreneurs shows how governments seek to use their standing in IOs to place new norms on the agenda, build support among the likeminded, shame opponents, and push for adoption (Waltz 2001;Kelley 2008;Park and Vetterlein 2010;Sikkink 2011). Studies on supranational entrepreneurs highlight that "many IO staff have as their stated purpose to shape state action by establishing best practices and by articulating and transmitting norms that define what constitutes acceptable and legitimate state behavior" (Barnett and Finnemore 2004, 33; see also Hawkins et al. 2006;Biermann and Siebenhüner 2009;Johnson 2014). Accounts of transnational entrepreneurship demonstrate how NGOs, interest groups, and epistemic communities employ their information, credibility, and moral authority to persuade IOs to adopt norms (Keck and Sikkink 1998;Price 1998;Joachim 2003;Betsill and Corell 2008;Tallberg et al. 2018).
Assuming IOs face such demands to integrate liberal norms, a first condition shaping the likelihood of commitment is the democratic density of memberships (Pevehouse 2005, 46). We expect that entrepreneurs will be more successful in convincing IOs to embrace liberal norms when their memberships are more democratic. For democracies, liberal ideals come naturally, making it easier for state, supranational, and transnational norm entrepreneurs to find allies in the membership and build support for commitment.
This argument builds on the assumption that states' preferences over international outcomes are partly rooted in domestic political conditions. This logic is sometimes referred to as "liberal constructivism" (Risse-Kappen 1996) or "ideational liberalism" (Moravcsik 1997) as it derives the preferences that states promote internationally from their domestic commitments to particular political ideals. A fundamental feature of such domestic conditions is the distinction between democratic and autocratic rule, leading democracies and autocracies to promote systematically different ideals abroad.
Democracy as a political system rests on liberal political ideals, some of which pertain to the organization of the polity and have become constitutive of democracy (e.g., free and fair elections, individual rights, rule of law), and some of which pertain to the organization of society and have become particularly common in democracies (e.g., equality, economic openness, social justice). These domestic political ideals make democracies more likely to favor liberal international outcomes than autocracies (e.g., Mansfield et al. 2000;Pevehouse 2005;Simmons 2009;Tallberg, Sommerer, and Squatrito 2016).
When IO memberships are democratically dense, commitment to liberal norms becomes more likely. Norm entrepreneurs are then more likely to encounter an audience receptive to their advocacy efforts. Since democratic states already adhere to liberal principles domestically, accepting these ideals internationally is not a radical step. Liberal norm entrepreneurs will therefore find more like-minded states with whom to partner and face fewer hurdles as they seek to convince the membership of commitment. Conversely, efforts to promote liberal norms can face greater resistance in IOs with less democratic memberships. Norm entrepreneurs are more likely to encounter an audience that perceives liberal ideals as foreign, potentially threatening to domestic control, and synonymous with demands for adjustment. Gaining deep acceptance for new liberal norms in this inhospitable environment will be more difficult.
The second condition shaping the likelihood of IOs committing to liberal norms are the institutional rules shaping norm entrepreneurship. Institutional rules specify who has standing in policy-making, what authority these actors enjoy, and how decisions are made. In our account, institutional rules matter by enabling and constraining norm entrepreneurs as these actors seek to influence IO policy. We focus theoretically on design features that we expect to be important in this respect, leaving aside other dimensions of institutional design. 8 We focus on one central institutional condition for each type of entrepreneur: pooling, delegation, and TNA access.
First, state entrepreneurs are more likely to be successful when institutional rules provide for a higher degree of pooling in interstate decision-making. Pooling refers to majoritarian decision-making and has implications for the capacity of state entrepreneurs to secure the required level of support for norms (Hooghe et al. 2017). The lower the institutional threshold, the easier it is for a state entrepreneur to push through a commitment to a norm, all else equal. When IO decisions require unanimous support, all member states have to be convinced, making the mission of a state entrepreneur more difficult. By contrast, when IO decisions only require the support of a simple or qualified majority, it becomes easier for a state entrepreneur to build a winning coalition (Scharpf 1997). 9 Second, supranational entrepreneurs are more likely to be successful when institutional rules delegate more power to international bureaucracies. IOs vary extensively in the extent to which they delegate agenda-setting, implementation, and enforcement powers to international bureaucracies (Hooghe et al. 2017). This variation in delegated authority influences the ability of supranational actors to exert influence over outcomes (Barnett and Finnemore 2004;Hawkins et al. 2006). Delegation comes with greater opportunities for IO staff to shape agendas and policies. It is not a coincidence that some of the most prominent examples of supranational entrepreneurship involve IOs such as the EU, IMF, and World Bank, all of which have highly empowered international bureaucracies.
Third, transnational entrepreneurs are more likely to be successful when institutional rules provide for non-state access to IO policy-making. The openness of IOs to TNAs has expanded considerably in recent decades, yet continues to vary across organizations (Tallberg et al. 2014). While some IOs grant TNAs considerable access to policy-making, others remain closed. Such variation is likely to affect TNAs' abilities to convince IOs to adopt new norms. Institutional access is frequently identified as a central determinant of TNA influence in IOs and multilateral negotiations (Betsill and Corell 2008;Tallberg et al. 2018). Rather than having to rely exclusively on strategies of outside lobbying, such as mobilizing public opinion, TNAs with access can employ a broader range of resources and strategies, including inside lobbying and information exchange. In sum, our argument leads to the following expectations:

H3:
The higher the democratic density of an IO's membership, the more likely the IO is to commit to a liberal norm.

H4:
The more an IO's institutional rules provide for pooling, delegation, and access, the more likely the IO is to commit to a liberal norm.

Empirical Analysis
We begin by assessing the distribution of norm commitment across the key features of our privileged argumentdemocratic density and institutional rules. Figure 2 plots the commitment rates for three categories of IOs, sorted by the democratic density of the membership. We note that IOs with the highest share of democratic members tend to adopt norms sooner. The medium category of IOs exhibits slower but more expansive commitments, mainly because the UN falls in this group, while the IOs with the lowest share of democratic states trail both groups in scope and speed. We observe a similar pattern when plotting the average adoption rates by institutional rules (Figure 3). IOs with higher levels of pooling, delegation, and TNA access tend to adopt more norms, and sooner, than IOs with lower levels. Overall, these patterns offer preliminary support for our argument. 10 Even though our analysis does not cover illiberal norms, these patterns suggest that memberships and institutions may help to explain IOs' treatment of norms such as non-intervention. The IOs with lower democratic density demonstrate efforts by state entrepreneurs to consolidate non-intervention as a norm, while resisting norms that conflict with it, notably, R2P. 11 For example, ASEAN members have remained committed to non-intervention and resisted pressure to adopt R2P as a guiding principle for the organization (Bellamy and Drummond 2011). Instead, ASEAN, 10 We note that democratic density is correlated with both delegation (r = 0.32) and TNA access (r = 0.42). While this means that democratic density to some extent predicts delegation and TNA access, our sample contains IOs with considerable diversity regarding these variables. For example, the Nordic Council scores high on democratic density (0.85) but low on delegation (0.09), whereas the AU scores low on democratic density (0.25) but above average on delegation (0.25) and TNA access (0.31).
11 Non-intervention is not a new norm, and most IOs give it some recognition.
For example, non-intervention is codified in the UN Charter.
which scores low on democratic density and medium on institutional rules, has sought to adjust the principle of nonintervention to allow for superficial integration of R2P. 12 This example suggests that conditions for illiberal norm entrepreneurship may be more favorable in IOs with less democratic memberships and permissive institutional rules.

Multivariate Analysis: Modeling Specification and Measurements
To test our hypotheses regarding IO norm commitment, we employ pooled event history analysis (PEHA), conventionally used in diffusion research (Shipan and Volden 2006;Kreitzer and Boehmke 2016). Given our interest in the broader dynamics of IO norm commitment, rather than in the determinants of particular norms, PEHA allows us to study the effects of variables across multiple norms and diverse IOs in a single model. Following Kreitzer and Boehmke (2016), we implement PEHA by estimating parameters for average effects of covariates on norm commitment using multilevel logit models, capturing temporal effects with a simple polynomial. The unit of analysis is IO-norm-year, with IO-years nested within norm groupings. Following conventions in event history analysis, our dichotomous dependent variables are coded as 1 in the year of norm commitment (or recognition), 0 in all preceding years, and missing for years after. We exclude observations prior to norm emergence, as determined by Google Ngram data. 13 To account for heterogeneity inherent to the data, we include random effects for IO, norm, and year. IO random effects allow us to adjust for unobserved, time-invariant IO-specific factors, such as organizational culture. Norm random effects allow us to account for unobserved particularities of individual norms that affect IOs similarly, such as sovereignty costs. And year random effects allow us to control for temporally specific changes that affect all IOs, such as global crises. Taken together, these adjustments allow us to identify systematic determinants of norm spread in a diverse sample of IOs and norms.
To test the effect of norm commitment by other organizations in an IO's organizational field (Hypothesis 1), we construct the variable norm prominence in field. This variable is based on spatial lags of prior commitments to each norm 12 Indicative of ASEAN's resistance, this minor adjustment to R2P was undertaken by a subsidiary body-not the ASEAN Summit-which is why it does not register as norm recognition in our data. 13 This correction minimizes the problem of zero-inflation for human security and R2P, which were more or less unknown before 1990.
among IOs in specific reference groups. We construct spatial lags for two such reference groups: (1) other IOs in the same region, assuming that organizations look to other IOs in the same geographic community and (2) all other IOs in the sample, assuming that multi-issue organizations are inspired by their global peers (Sommerer and Tallberg 2019). The variables are the cumulated rate of prior norm commitment, lagged by one year. 14 For instance, the spatial lag for norm prominence in field (region) in 2010 would indicate that commitment to the gender equality norm was 14 The spatial lags are calculated using the spmon command (Neumayer and Plümper 2010).
50 percent among American IOs in 2009 and 33 percent among African IOs. We assess Hypothesis 2 through the variable norm prominence in society, which captures norm references in global public discourse. We use the Google Ngram tool and data from the Google Books corpus of scientific and nonscientific English-speaking publications (Michel et al. 2011). We construct a 2-gram for each norm and extract time series data from 1985 to 2008. 15 We operationalize the democratic density of an IO's membership (Hypothesis 3) based on the mean liberal democracy index (V-Dem) of an IO's member states in the year of observation, calculated based on IO membership data from the COW-IGO dataset (Coppedge et al. 2017;Pevehouse et al. 2019). Next to electoral quality, the index factors in the protection of civil liberties, rule of law, the independence of the judiciary, and limitations on the exercise of executive power (Coppedge et al. 2017). IOs with more liberal democratic memberships will have higher values on this variable.
We assess the influence of institutional rules (Hypothesis 4) with three different indicators, reflecting our interest in pooling, delegation, and access. We measure the effect of majoritarian decision rules through aggregated pooling scores taken from the Measuring International Authority dataset (MIA) (Hooghe et al. 2017). It aggregates the voting rules for agenda-setting and final decisions across the state-dominated bodies of an IO, weighted by bindingness and ratification. 16 The variable delegation measures the allocation of authoritative competences to non-state bodies in an IO's decision-making process (MIA). It aggregates political delegation in agenda-setting, decision-making, and dispute settlement across six decision areas: accession, suspension, constitutional reform, budgetary allocation, financial compliance, and policy-making.
Finally, the variable TNA access integrates data on the depth and range of formal access to an IO's bodies (Tallberg et al. 2014). These two dimensions are constitutive of all participatory arrangements by defining what rights are granted to whom. The depth of access captures the level of involvement offered to TNAs through institutional rules, and the range of access captures the breadth of TNAs entitled to participate.
Additionally, we control for several possible confounders. First, it may be easier to reach agreement among fewer member states (Axelrod and Keohane 1985). Membership size equals the IO membership count in a given year (COW-IGO data).
Second, political heterogeneity may make decisionmaking on new norms more cumbersome. We adjust for this possibility using polity heterogeneity, measured as the standard deviation of the democratic density of the observed IO in a given year.
Third, commitment to liberal norms might be more likely in the presence of a major democratic power (Park 2014). The variable democratic major power is coded 1 for IOs that include a major or regional democratic power, but lack major or regional non-democratic powers that could veto liberal norms. 17 Fourth, IOs may be more likely to commit to norms if they have resources to implement new policies. The categorical variable IO resources reflects the scope of staff and budget of an IO, coded 1-5 as described in Table A4 in the online appendix.
Finally, we include three variables in supplemental tests. Global IO is coded as 1 for IOs with members from at least three continents. Norm relevance is coded as 1 for IOs with a mandate that corresponds to the norm. IO productivity is the count of key decisions made by the main interstate body in the year of observation. Table 1 reports the results of the pooled multilevel models on IO norm commitment. We specify models so as to 16 For a full description of the coding of this variable see Hooghe et al. 2017, 213-17. 17 Major powers operationalized based on COW data; we add regional powers for the period after 1989 (Cline et al. 2011).

Results
alternate the two measures of norm prominence in field. Due to correlation between delegation and TNA access (r = 0.796), we estimate models with both variables included and either variable excluded. These combinations result in six models. As Table 1 indicates, neither norm prominence in field (region) (Hypothesis 1) nor norm prominence in society (Hypothesis 2) is a statistically significant determinant of norm commitment. Norm prominence in field (global) (Hypothesis 1) partly diverges from this pattern, showing positive and significant coefficients in two models (2 and 6). Taken together, these results grant weak support for global scripts as an explanation of IOs' commitments to liberal norms.
By contrast, Table 1 offers strong support for the account privileging democratic density and institutional design as explanations of norm commitment. The coefficient for democratic density (Hypothesis 3) is positive and statistically significant at the 95 percent confidence level (or higher) in all models. This finding confirms the impression from the descriptive analysis that full policies on liberal norms are concentrated in IOs with more democratic memberships.
We also find clear support for the expectation that IOs with more pooled decision-making, empowered bureaucracies, and openness to TNAs will exhibit higher rates of norm commitment (Hypothesis 4). The coefficient for pooling is positive and statistically significant in all models, suggesting that when IOs allow for majority decision-making, state entrepreneurs (or other entrepreneurs that have to build support among member states) are more likely to be successful and the IO more likely to commit to a norm. A case in point is the UNGA's adoption of the 2005 resolution on R2P, in which state entrepreneurs built cross-regional coalitions that could ensure the required level of support, leading outstanding recalcitrant states to give in (Bellamy 2017).
Similarly, the coefficients on delegation are positive and statistically significant in all models. This means that IOs whose supranational bureaucracies have been vested with higher agenda-setting, implementation, and enforcement powers present more favorable conditions to norm commitment. The EU represents a typical example of this pattern. This IO exhibits the highest degree of delegation in our sample, reflecting the far-reaching powers of the European Commission, and also has a very high rate of norm commitment. Speaking to the opposite end of the scale, the OIC features a severely constrained secretariat, leaving it with few tools to exercise norm entrepreneurship vis-à-vis the OIC Council, which likely contributes to the low commitment rate of the OIC. 18 A case where delegation to a pro-norm bureaucracy explicitly is credited with the adoption of legally binding policies is that of good governance in the AU (Leininger 2015, 63-65).
Finally, the results indicate that TNA access has predictive power. The positive and significant coefficients in Models 3 and 4 suggest that IOs that provide greater opportunities for TNAs to participate in policy-making are more likely to commit to liberal norms. The high correlation between TNA access and delegation, discussed above, should be borne in mind when interpreting the results for TNA access in Models 1 and 2. Organizations that rank high on TNA access, such as the OAS, are likely to have experienced more intense norm entrepreneurship by TNAs than IOs that remain inaccessible to TNAs, like ASEAN or SCO. A case in which TNAs are claimed to have played a major role in pushing member states to commit to a liberal norm is gender equality in the UN (Joachim 2003). Conversely, a case of TNAs with limited access failing to convince the IO of norm commitment is ASEAN in the case of human security (Allison and Taylor 2017). The control variables provide additional insight into the conditions that favor IO norm commitment. Membership size has positive and significant effects in all models, suggesting that large IOs are more likely to commit to liberal norms. This result may be driven partly by the high commitment rate of the UN. Democratic major power shows positive and statistically significant coefficients in two models (2 and 6). The negative finding for IO resources (Models 1-2, 5-6) indicates that large and well-resourced bureaucracies do not facilitate, and may even counteract, commitments to liberal norms.
Overall, we find strong evidence that democratic memberships and institutional rules that facilitate norm entrepreneurship are conducive to norm commitment. Do they also account for shallower talk about norms by IOs? We examine this question using norm recognition as the dependent variable (Table 2). Here we find stronger support for the logic of global scripts. When norm prominence in field is operationalized as the cumulative recognition by other IOs in the same region, the positive effect is significant at the 95 percent confidence level (Models 7, 9, and 11). The alternative indicator based on previous recognition among all other multi-issue IOs is positive but not statistically significant at conventional levels (p = 0.08). Norm prominence in society also has a statistically significant positive effect. The overall impression is that the global scripts account shows a strong explanatory fit with IO talk of norms. This finding is consistent with sociological institutionalist notions of ceremonial, rather than profound, adaptation to external standards of legitimacy (Meyer and Rowan 1977).
In a similar vein, Table 2 suggests democratic memberships and institutional rules are not essential to shallow talk about norms. Democratic density is not a statistically significant determinant of norm recognition by IOs (Models 7-12). These results are in line with the descriptive finding that norm recognition is common among IOs regardless of democratic density, suggesting that autocracies may see the legitimation benefits (and limited political costs) of rhetorically recognizing liberal norms. 19 Similarly, with one exception, institutional rules permissive to norm entrepreneurs do not appear to significantly contribute to Notes: * p < .05; ** p < .01. All models estimated in R version 3.5.1. Logit estimator. Errors in parenthesis. Standardized explanatory variables. Random effects for IOs, norms, and years. The lower N in models estimating norm recognition compared with models of norm commitment (table 1) is caused by the survival character of our data. Recognition typically predates commitment and thus leads to an earlier exit point. norm recognition. Pooling has positive coefficients (Models 7-11), but is not statistically significant. The results for TNA access on norm recognition are also not significant. Delegation, however, is positively correlated with norm recognition and is statistically significant. This suggests that independent supranational bureaucracies facilitate both norm recognition and commitment by IOs. Democratic major power has a positive, statistically significant effect in two models (8 and 12), suggesting a powerful democratic state may boost the likelihood of IO norm recognition.
Membership size also has a positive effect on norm recognition in all models and is statistically significant in Models 8 and 12. To contrast the difference between fully committing to a norm and only recognizing it, we calculate predicted probabilities. Figure 4 exhibits the annual probability of norm commitment and norm recognition as a function of democratic density and norm prominence (region), as these move from one standard deviation below the mean to one standard deviation above. 20 The figure suggests three 20 Predicted probabilities for all main variables with significant effects are presented in the online appendix. points. First, there is a general difference in probability between norm commitment and norm recognition, the first being considerably less common. Second, for commitment, norm prominence in field does not matter, whereas democratic density does. IOs with less democratic memberships are extremely unlikely to commit to norms in any given year. Increasing this variable from one standard deviation below the mean to one above yields about one percent higher likelihood of commitment. Aggregated over longer periods of time, such differences in annual probabilities can account for a considerable amount of the variation we observe in our sample. 21 Third, for recognition, norm prominence in field matters but democratic density does not. An increase in norm prominence from one standard deviation below the mean to one above translates into about a three percent higher probability of adoption.

Robustness
We assess the robustness of the results with a series of additional models, reported in the supplementary appendix. First, while the main models include random effects for IOs, it is possible that IOs in particular geographic regions are subject to common influences that retard or accelerate norm adoption (e.g., Risse 2016). In Table A6 in the online appendix, we add four region dummies, coded as 1 for IOs headquartered in the corresponding region and 0 otherwise. Our key findings are robust to this alteration, which also suggests that norm adoption in our sample is not subject to significant and consistent geographic variation.
Second, we add controls pertaining to the heterogeneity of IOs in the sample (Table A7 in the online appendix). The great majority of the results remain unchanged when controlling for the global/regional orientation of IOs, the fit of IO mandates with norms, and the productivity of IOs. Importantly, the results for democratic density, delegation, and pooling are robust to the addition of these variables. In addition, the variables representing global scripts are insignificant in the norm commitment models, but have explanatory power with regard to norm recognition, as in the main results. Substantially, we observe that global IOs (Table A7 in the online appendix, Models 1 and 4) and IOs with a mandate offering a closer fit with the respective norm (Models 2 and 5) are not more likely to commit to norms. The productivity of an IO in issuing decisions does not matter either, although the number of observations is lower in these models (Models 3 and 6).
Third, we want to ensure that our results are robust to alternative operationalizations of democratic density. In Table A8 in the online appendix, we report results based on an operationalization of democratic density as the share of democratic states, using the Democracy-Dictatorship data (Cheibub et al. 2010). We find that this variable also predicts norm commitment but with greater variance than our preferred measure. In the same table, we also estimate the effect on norm commitment of the KOF Political Globalization Index, which measures a country's IO memberships, treaty ratifications, and bilateral investment treaties (Models 5 and 6) (Gygli et al. 2019). We calculate annual averages of IO memberships and find that political globalization is not a significant predictor of norm commitment (and even negatively predicts norm recognition).
Fourth, we assess whether our results are sensitive to sub-sample analysis. Table A9 in the online appendix presents models fitted on data excluding the three norms that have markedly lower commitments (R2P, deregulation, and human security) than the other five norms. The results differ only marginally from the full sample results, with democratic density and institutional rules predicting norm commitment while norm prominence predicts recognition. In the same table, we also present the results for a sub-sample analysis excluding the organizations with the least democratic memberships (SCO, AMU, ASEAN, and OIC).
Taken as a whole, our analysis reveals that norm commitment is best predicted by the degree of democratic density in an IO's membership and the extent to which its institutional design empowers norm entrepreneurs. In contrast, global scripts about appropriate action are mainly helpful in explaining more rhetorical recognition of norms by IOs.

Conclusion
Global governance has witnessed the spread of a range of liberal norms over recent decades. While earlier research has examined the trajectory of individual norms, we have known little about the broader patterns of norm commitment by IOs. This article is an effort to address this situation through a comparative, large-N analysis of commitments to eight liberal norms across eighteen IOs in the time period 1980-2015. Theoretically, we have examined the explanatory power of two accounts: a common explanation focused on the influence of global scripts, as IOs look to their external environment for legitimate standards, and our alternative explanation, privileging the internal conditions of IOs, with a particular focus on how democratic memberships and institutional rules facilitate norm entrepreneurship.
The findings mainly support our explanation, while also pointing to important complementarities. While global scripts account for an IO's willingness to recognize norms, they face greater problems explaining the move from talk to commitment. Instead, deep commitments to liberal norms depend on democratic memberships and permissive institutional rules, resulting in a concentration of liberal norm adoption in the democratic core of IOs. These findings suggest that a focus on global scripts would risk missing the sources and patterns of more consequential norm commitments by IOs. Only by looking at the organizational context in which a norm is translated into policy can we account for the varying nature of commitments to liberal norms in global governance.
Expanding the perspective, this article suggests four broader implications. First, it demonstrates how careful empirical measurement helps to identify the multiple logics of action that drive IOs, as well as the scope conditions under which they hold. Considering whether and how to reconcile sociological and rationalist logics of action has been a concern of both general social theorists and IR scholars (Jupille, Caporaso, andCheckel 2003, March andOlsen 2006). This article shows how empirical testing may allow us to move beyond competitive generalizations to identify the scope conditions and complementarities of institutionalist theories. While global scripts explain norm talk well, democratic density and institutional design are central determinants of more consequential commitments.
Second, our analysis shows how a large-N approach can complement qualitative work on norm entrepreneurship by identifying general sources of variation in norm spread (cf. Finnemore 1993;Keck and Sikkink 1998;Price 1998;True and Mintrom 2001;Joachim 2003;Park and Vetterlein 2010). While existing research is strong on theorizing and documenting norm entrepreneurship by states, international bureaucracies, and TNAs, it is relatively weaker on establishing the political conditions that shape norm trajectories in a comparative perspective. This article shows that important conditions affecting the likelihood of successful norm entrepreneurship relate to the composition of IO memberships and the institutional constraints encountered at IOs-factors not previously assessed in a comparative analysis.
Third, our analysis raises questions relating to the effectiveness or performance of IOs (Hafner-Burton, von Stein, and Gartzke 2008;Gutner and Thompson 2010;Lall 2017). This article addresses commitment to norms by IOs and stops short of tracing how these policies subsequently impact behavior on the ground-a process that research on norm effectiveness aptly covers (e.g., Simmons 2009;Risse, Ropp, and Sikkink 2013;Zimmermann 2017). But a focus on IOs' commitments to norms has several implications. Importantly, the fact that IOs more rarely commit to, than recognize, norms indicates that the effectiveness of norms may be lower than often assumed based on their prominence in overall discourse. IOs frequently stop at rhetorical recognition of global norms. In addition, our overview indicates gaps in the extent to which IOs respond to societal problems through the development of extensive policy committed to remedying those issues. For instance, despite extensive intrastate conflict, norms aimed at protecting civilians-R2P and human security-hardly gain traction outside the UN.
Fourth, this article carries consequences for research on the liberal international order and the retrenchment of liberal norms. This literature conventionally focuses on how the United States upholds this order and on the implications of power shifts for its long-term sustainability (Ikenberry 2011;Dunne and Flockhart 2013;Acharya 2014). This article offers several important insights. It suggests that the liberal order is more fragmented than often assumed. Liberal norms heralded as center-pieces of this order often travel little beyond their Western origins. At the same time, the liberal international order, as expressed through liberal norms, appears to rest on broader underpinnings than US power. For deep commitments to liberal norms in global governance, it is not primarily the status of a lone major power (the US) that matters, but democratic memberships as a whole and the institutional empowerment of supranational bureaucracies and TNAs. Taken together, these observations suggest why we may be observing a contemporary retrenchment of liberal norms in world politics (Cooley 2015;Diamond et al. 2016). While liberal norms were never truly supported by IO memberships outside the democratic core of global governance, which cannot be counted on for backing, a shift toward illiberal democracy in IO memberships within this core now contributes to a weakening of liberal norms from within.

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