Designing international organizations for debate? A factor analysis

International organizations (IOs) constitute key arenas in which states discuss common issues. Such debates are central prerequisites for taking qualitatively good decisions. Yet researchers have not examined how IOs foster discussion through their institutional provisions. We conduct a factor analysis of institutional rules of 114 IOs which reveals that two ideal types how IOs seek to induce discussion exist: The first type creates room for debate in the negotiation stage of the policy cycle. In contrast, the second type gives member states a strong say in the agenda-setting, thereby facilitating debate. Why do IOs opt for either strategy? A limited policy scope, heterogeneity among actors, and diplomatic socialization increase the probability that IOs place emphasis on debate during negotiations, while a high number of members is the main reason for IOs to promote debate during agenda-setting. These choices reflect the strive of IOs to balance extensive debate with speedy decision-making.


Introduction
International organizations (IOs) are key elements in international politics. They cover a multitude of different policy fields and constitute the arenas in which national delegates exchange their positions and legal, factual, political or normative arguments in order to develop and eventually pass policy outcomes, for instance in form of resolutions, directives, programs, declarations or decisions.
Like other political systems, IOs are institutionally designed to pursue two competing aims: creating opportunities for debate in their decision-making bodies, while limiting these opportunities to allow for speedy decision-making (Aaken 2004;Golub 2007). They do so by formulating explicit rules and provisions on how debates should be conducted. Yet, this comes with strings attached: Usually, rules that seek to induce debate are likely to reduce the speed of decision-making and vice-versa (Rada 2000;Panke 2006a, b;Golub 2007). For instance, the World Health Organization (WHO)'s internal reforms were geared toward speeding up discussion between delegates in order to allow for greater efficiency of decision-making (Eckl 2017). On the one hand, extensive debates bring about the risk of reducing the speed of decision-making (Rada 2000;Golub 2007), thereby making IOs with extensive debates less efficient than IOs that limit the time for discussion considerably. On the other hand, debates can be conducive to high-quality negotiation outcomes. IOs that provide room for extensive debates increase the chances of producing negotiation outcomes that carry a high level of high problem-solving potential and strongly reflect the collective perceptions on what is good, appropriate and just (Johnstone 2003;Panke 2010;Reinhard, et al. 2014).
In their treaties and rules of procedures, IOs specify, whether debates should take place or not, who can speak and for how long or how often, or who has access to debate in the agenda-setting, the negotiation and decision-taking stages of the policy cycle. When examining these provisions, it is striking that most IOs follow a common logic, regulating similar elements which seem to be important. For instance, many IOs define whether and how the member states can participate in agenda-setting or how often national delegates can make proposals during debates. Yet, IOs also show considerable differences with regard to how their institutional rules structure debate. While some organizations, like the Organisation for Economic Co-operation and Development (OECD) or the United Nations General Assembly (UNGA), have adopted extensive and detailed rules of procedures, usually set out in separate documents, other IOs such as the Black Sea Economic Cooperation (BSEC), and the Wassenaar Arrangement (WA) only include few provisions in their treaties.
While much of IO scholarship initially examined the creation, evolution and change, as well as the operation and effectiveness of individual organizations, in recent years, IO research became more strongly comparative in nature (Panke et al. 2019(Panke et al. , 2020e.g., Hurd 2017;Hooghe et al. 2019). In addition, a focus on institutional design of IOs became increasingly prominent in International Relations research as well (Goodin 1995;Pierson 2000). Thus, in the last decade, scholarship has put strong emphasis on studying the institutional design of IOs and on doing so in a comparative manner, focusing on larger samples of IOs. With the legalization approach, scholars have developed a set of mainly institutional criteria, which allow to map out and compare how strongly different IOs entail different types of binding or non-binding obligations as well as different (authoritative) forms of compliance monitoring and dispute resolution (e.g., Abbott et al. 2000). The rational design of IO approach has examined which institutional rules (e.g., membership, scope, centralization, control, and flexibility) match different types of challenges, which IOs face, for instance with respect to the number and heterogeneity of their member states (e.g., Koremenos et al. 2001;Rathbun 2011). The authority of IOs also plays a role in research, which comparatively examines the nature of IOs alongside pooling and delegation dimensions and explains why IOs vary concerning the authority they have been equipped with (Hooghe et al. 2017). The institutional setup of IOs in general, as well as with respect to specific features such as the access of transnational actors to IO decision-making (Tallberg et al. 2014), can have important implications for their operation and longevity as well as their legitimacy, which has recently been studied in a comparative manner as well (c.f. Gray 2018; Eilstrup-Sangiovanni 2020).
Yet, despite the richness of the comparative IO literature, no one has shed light on how and to which extent IOs seek to induce discussion between the delegates or delimit them in order to speedup decision-making. 1 This is surprising given that both features can have important implications for the effectiveness and legitimacy of global governance.
In order to contribute to these strands of research, we study whether IOs are designed to induce discussion between delegates, although this reduces the speed of arriving at decisions (Eckl 2017). Specifically, we address the question if there are distinct patterns in the institutional deliberative design of IOs in the first part of the paper. Put differently we examine which institutional design elements tend to cluster together and whether there are types of IOs with respect to how they seek to foster discussion between delegates. To shed light on these elements, we apply a factor analysis examining the composition of the institutional designs of IOs. This reveals that there are two types. First, there is negotiation-stark type of IOs, which fosters discussion between delegates by including many provisions to this effect in the negotiation stage. Second, there is an agenda-setting-stark type of IOs, which attempts to induce debate between delegates in the agenda-setting stage. This indicates that IOs tend to balance both aims: debate and good outcomes on the one hand and speedy decision-making on the other hand. Yet, IOs tend to place emphasis on different aims in different stages of the policy cycle. While the first type of IOs is likely to be subject to a slower speed of negotiations, the second type of IOs might require more time in the agenda-setting stage.
In the second part of the paper, we shift the focus to a second question and seek to explain why IOs differ in their institutional setups: Which variables induce specific design choices? Are the driving forces behind the two types of IOs the same? The results show that apart from IO size, different forces are at play. The most important variables inducing debate in the negotiation-stark-type of IOs are a limited policy scope, heterogeneity among the actors and diplomatic socialization. Driving factors behind the agenda-stark type are IO size as well as a more limited policy scope.

Sample selection and data collection
This section discusses the sample selection and data collection necessary to conduct the factor analysis in order to uncover whether there are different types of IO institutional designs to induce debate between delegates.
IOs are institutionalized forms of cooperation between three or more states and their numbers have increased after WWII and again after the end of the Cold War. Thus, we need to be selective concerning which IOs we examine. The basis for the selection of IOs is the Correlates of War database (Pevehouse et al. 2004) as well as the Yearbook of International Organizations. 2 Together they cover the universe of IOs, while each database alone omits some IOs. 3 To select IOs, we further specify the above definition by three criteria. First, an IO must still be in operation in 2016 rather than just existing on paper, which we captured by the existence of an updated homepage. Second, IOs must be composed of member states and in exceptional cases regional organizations, whereas organizations also entailing firms, private persons, etc., are excluded. Third, we only include IOs whose purpose is to create or reinforce international norms and rules, while all IOs that have no authority to engage in such standard-setting activities are excluded (e.g., commercial purpose organizations such as banks, advisory bodies like think tanks, scientific study groups). On this basis, we selected a subsample of 114 IOs (c.f. Table 4) which is approximately representative with respect to the COW dataset and the IO Yearbook, as it includes organizations that vary with regard to their size, age, policy field, as well as regional vs. global membership. Furthermore, we ensured a proportional representation of world regions concerning the regional IOs.
IO institutional design structures the interactions between state actors in the main legislative arena. 4 For the coding of institutional design elements of IOs that seek to induce discussion between delegates, we used two types of sources: primary law (e.g., founding treaties and treaty changes, protocols, annexes) as well as the rules of procedure (e.g., terms of references, procedures). Similar to states, policy cycles of IO usually entail five stages: agenda-setting, negotiations, decision-taking, implementation and enforcement (Panke 2013, Howlett et al. 2009). Since we are interested in how IOs organize debates between delegates, we only focus on the three stages of the policy cycle in which such formal debates can take place within the IO, namely agenda setting, negotiation, and decision-taking. In the agenda-setting stage, IO members need to decide which issues to put on the IO negotiation agenda. This phase starts with the drafting of agenda items and ends with a finalized agenda. The subsequent negotiation stage begins once the actors have a meeting agenda and ends after substantive text changes of the outcome document/IO decision are finalized. In the decisiontaking stage, actors formally pass the IO outcome. This encompasses pre-voting statements as well as explanations of votes.
We analyze general framework features that might be conducive to debate between delegates throughout the three stages. A design element is regarded conducive to promoting discussion if it provides an opportunity for the exchange of positions, claims, and demands between IO member states (c.f. Table 5).
There are four institutional rules that might foster discussion in the agendasetting stage. These include the possibility to conduct exceptional meetings (AS_ except_meetings), the explicit participation of IO member states in the setting of the negotiation agenda (AS_state_participation), the possibility that an agenda can be changed later on (AS_change_agenda), and the possibility that states discuss the agenda at the beginning of a meeting (AS_discuss_agenda).
In the subsequent negotiation stage, 10 different institutional design features could prompt debate between the state actors. These consist of the competency of the chair to accord the right to speak to delegates (N_chair_speakers), the possibility to flexibly change the order of speakers during ongoing discussion (N_order_speakers), the right of reply of delegates (N_chair_right_of_reply), the opportunity of delegates to make proposals during debate (N_make_proposals), the rule that proposals can be made even without secondments by other actors (N_no_secondment_proposals), the ability of delegates to engage with additional exceptional proposals (N_timing_exceptions_proposals), the right of the actors to reintroduce formerly withdrawn proposals (N_reintroduce_proposals), the ability of delegates to reconsider formerly rejected proposals (N_reconsider_proposals), the rule requiring a discussion before closing the debate on one agenda item (N_ discuss_close_debate), as well as a rule that ensures that delegates engage in a discussion before finally closing the meeting (N_discuss_close_meeting).
In the decision-taking stage, three institutional design elements can possibly induce pre-and post-voting debates. These are the requirement of a quorum for the ability of the delegates to pass final decisions (D_quorum), the rule that decisions can only be taken by consensus (D_consensus), and the rule that all states carry equal weight when it comes to voting (D_one_state_one_vote).
There are five institutional design elements that might impact debates across the stages. These include the opportunity of delegates to flexibly decide whether or not the meeting takes place behind closed doors (FRAME_transparency), the organization of translations by the IO (FRAME_translation), the rule allowing state delegates to bring additional advisers and experts to the meetings (FRAME_advisors), the access of external actors to the meetings (FRAME_ access_externals) and granting these external actors speaking rights in the meetings (FRAME_speaking_externals).
For coding purposes, each of these 22 elements was formulated as a question, which was answered by checking the primary law and rules of procedure of every IO in turn (c.f. Table 5). 5 Whenever the institutional rule fostering discussion was present, it was coded with a 1, whenever it was absent it was coded with a 0.
For each IO, we measured the presence or absence of an institutional design element geared toward fostering discussion between delegates for the year 2016. We opted for a snapshot of one year rather than covering a long period of time for two reasons. First, most of IO primary law hardly varies over time with respect to the provisions on debate of IOs. Second, while rules of procedure can be changed over time, usually only the latest version can be systematically accessed on IO websites or via IO secretariats. Thus, a time-series analysis of IO institutional designs could only be based on primary law and would neglect rules of procedure, although the latter are usually more detailed than the former when it comes to fostering or preventing discussion between delegates.

Factor analysis: the identification of IO institutional design types
In order to examine the relations between the different design elements and to identify common patterns in the institutional rules of IOs, we conduct an exploratory factor analysis. Factor analysis is a statistical technique that attempts to reduce the complexity of a set of observed variables by identifying a number of underlying factors. Factor analysis assumes that a number of un-observed variables (the so-called factors) exist that account for correlations among the observed variables (Leschke 2000). Hence, we use factor analysis to identify common patterns in the data. Our data is categorical in nature, as each of the 22 institutional design elements is either coded as 1 (seeking to foster discussion) or 0 (being absent). Thus, we opt for a tetrachoric factor analysis, a technique which is suitable for binary data (Heinen 1996).
In a first step, we created the matrix, which displays the tetrachoric correlation coefficients between the different design elements. It shows that some items have negative values and therefore need to be excluded so that the factor analysis can be carried out with valid results (Bartholomew et al. 2011). Thus, six of the initial items were omitted (exceptional meetings in the agenda-setting stage, quorum, consensus and one state one vote rule in the decision-making stage, as well as the ability of delegates to bring additional advisors and experts to the meetings, the access of external actors and their right to take the floor). The subsequent tetrachoric correlation matrix does not entail negative values (see Table 6, Appendix) and allows to run the factor analysis. We used orthogonal varimax rotation to extract interpretable results. This suggests that a selection of two factors is promising, as the addition of further factors would only add little additional explanatory power (see Fig. 1). Taken together, the two factors carry 67.72% of the variation.
The analysis reveals that there are two factors, but out of the 16 institutional design variables one does not belong to any of the two types (bringing advisers to IOs), so that there are 15 different institutional features to differentiate IO ideal typical institutional designs that seek to foster discussion between delegates (see further Table 1). A first group of IOs attempts to induce discussion by including many design elements in the actual negotiation stage, complemented by two additional supportive features (such as IOs provide translations, opting for closed door sessions). Thus, we can refer to these types of IOs as 'negotiation-stark IOs'. The factor constituting this group of IOs alone accounts for 52.12 percent of the total variation. Negotiationstark IOs are prone to extensive debate between delegates in the negotiation phase of the policy cycle. Consequently, these IOs can be expected to be slower in this specific phase.

Fig. 1 Plot of eigenvalues
A second group of IOs places emphasis on agenda-setting rules in order to foster discussion between delegates early on. The corresponding factor carries a lower explanatory power, as it resembles for 15.57 percent of observed variation. Since this group of IOs places the strongest emphasis on agenda-setting rules, we refer to this type of IO intuitional design as 'agenda-stark IOs'. Agenda-stark IOs have institutional designs that seek to induce discussion in the early stage of the policy cycle in particular and are expected to be slower in this stage of the policy cycle.
Each type of IO only seeks to induce deliberation between member state delegates in one stage of the policy cycle, rather than in several all stages at once. In this sense, negotiation-stark as well as agenda-stark IOs both strike a balance between deliberation and speedy decision-making, but do so in different manners. Figure 2 plots the two factors against each other. This reveals that there are actually two very distinct clusters. The three items related to the agenda-setting stage are situated at the upper-left quarter of the graph, while the items concerning the negotiation stage cluster in the lower-right quarter. The two framework condition items are placed in between, but are slightly closer to the other factor 1 items.
In the next step, we take a look at how different IOs score with regard to the two factors (see Fig. 3). In other words, which IOs resemble the negotiation-stark type that is prone to discussions in the negotiation stage, and which resemble the agenda-setting-stark type that seeks to foster debates at the start of the policy cycle? Some IOs can be grouped into one of the two clusters very easily. For instance, the Latin American Integration Association (ALADI) or Intergovernmental Oceanographic Commission (IOC) is situated very closely to the position of the agenda-setting cluster in the upper left-half of Fig. 2. Other IOs like the Human Rights Council (HRC) or the International Coffee Organization (ICO) are approximately at the same position as the negotiation cluster in Fig. 2. These IOs can be seen as typical examples of the two types. Figure 3 shows that also many IOs exist which include elements of both types, as for example the Food and Agriculture Organization of the United Nations (FAO) or World Intellectual Property Organization (WIPO). In addition, Fig. 3 illustrates that the extent to which IO design seeks to induce debates and where these IO institutional design rules place debates are not mutually exclusive. While there are some IOs that are geared toward fostering debates during agenda-setting rather than negotiations (the IOs on the far left side of Fig. 3), and some IOs that place great emphasis on negotiations but not on debates in agenda-setting (at the bottom of Fig. 3), many IOs combine both (lower left to top right of Fig. 3). IOs  Fig. 3, such as UNFCCC, have institutional designs with a greater number of rules fostering debates than the group of IOs that are placed in the left bottom corner of the figure (e.g., the World Customs Organization (WCO)), leading to the expectation that delegates engage in more debates in these IOs in practice.

Accounting for the choices in IO Institutional design
In the literature, factor analysis is often applied to create indices that are used as independent variables in subsequent analysis (e.g., Logan and Mattes 2012). We, however, are also interested in why IOs choose different institutional designs and, therefore, turn the two types into dependent variables. Thus, in the remainder of this paper, we examine two research questions: Which variables induce specific design choices? Are the driving forces behind the two types of IOs the same?
Rational choice approaches on international cooperation expect that states are the main actors and engage with one another in IOs as arenas to further their respective interests and preferences (Keohane 1984). The higher the number of actors, the less likely it is that winsets naturally overlap, and the more difficult it is to develop negotiation outcomes that reach the IOs' respective thresholds for passing decisions (Snidal 2002). With an increasing number of IO member states, debate is more time-intensive, as there is a higher number of actors that are likely to speak up at the negotiation table (Hertz and Leuffen 2011). On this basis, it can be expected that smaller IOs are likely to be equipped with more institutional provisions fostering debates, than larger IOs. Accordingly, the higher the number of actors, the fewer institutional design provisions conducive to debate are adopted by IOs (Hypothesis 1).
The composition of membership might also be a feature with impact on IO institutional design choices as heterogeneity can influence the speed of decisionmaking (Golub 2008): The more heterogeneous IO member states are, the greater is the demand for debate to increase the chances that the members can develop a negotiation outcome that they can pass based on the voting requirements. By contrast, IOs with homogeneous membership can more easily afford to cut the time for debate short without risking non-decisions. Thus, the second hypothesis states: The less homogeneous the IO member states are, the more design features conducive to debate an IO is equipped with (Hypothesis 2).
Another important aspect might be whether an IO engages in policy-making only or in operational activities in addition (Hurd 2011). For IOs with operational focus, speedy negotiations and decisions are important to effectively address the problems or issues on the ground. In IOs without operational activities, the speed of negotiations is less crucial, since the process of passing decisions is usually followed by member state domestic activities (ratification, transposition or implementation) that are time-intensive in themselves before an IO decision can have an impact on the ground (Cox et al. 1973). Accordingly, operational-activity IOs should have fewer design elements conducive to discussion than policy-making IOs (Hypothesis 3).
Unlike rational choice, constructivist approaches consider exchanges of positions, claims, and demands that are complemented by reasons as an integral element of interaction (Wendt 1999). Actors pursue national positions, but can change their causal, legal, technical, scientific or normative ideas in the wake of the better argument, which can lead to endogenous position changes, and outcomes well beyond a lowest common denominator compromise (Johnstone 2003). Such processes of deliberation can be induced through the professional training that diplomats receive prior to being posted to represent their country in an IO. Diplomats are socialized into presenting reason-giving exchanges in which they communicate technical, scientific, legal, normative ideas underlying the national positions that they put forward in the international realm (Alderson 2001). In contrast, governmental actors and ministers are socialized in a context in which they aim for re-election and therefore have to make public marks and take a stance with their respective national positions vis-a-vis their counterparts. The latter needs less time and room for actual debate than the former. Thus, IOs in which diplomats constitute the actors in the main legislative body (rather than government members) should be equipped with more institutional design features fostering discussion (Hypothesis 4).
Apart from the question of whether an IO should foster debate between delegates in its institutional design in general, it is also interesting to see why IOs pursue different approaches to do so. The factor analysis revealed that there are two types of how IOs foster discussion: either by including a lot of institutional design elements regulating the actual negotiation stage or by allowing member states to play an important role in the agenda-setting stage. One factor which might account for this difference is the policy scope of an IO. IOs can be task-specific or general purpose in nature as they differ with respect to the range of policy mandates they are equipped with (Jupille et al. 2013). IOs with a broad policy scope might be less inclined to strongly foster debate between delegates in the negotiation stage, as extensive debates reduce the speed of decision-making, especially if there are many items on the agenda. By contrast, IOs which cover a lot of different policy fields are more likely to allow their member states to put issues on the agenda and engage in discussion to this effect. Thus, the fifth hypothesis expects that if the policy scope of an IO is broad, it is more likely that this IO adopts institutional design features conducive to discussion in the agenda-setting stage, rather than in the negotiation stage (Hypothesis 5).

Analysis and discussion
This section examines which of the hypotheses are plausible in accounting for differences in IO institutional design choices.
In a first step, we operationalize the independent variables as follows 6 : The first independent variable, IO size, measures the number of member states of an IO in 2016 (Hypothesis 1). We determine the number of member states with full membership rights in 2016, based on information from the official IO websites.
Hypothesis 2 argues that IOs with less homogeneous membership should allow for more discussion. We measure the variable by a proxy, capturing whether an IO has a global or regional membership. Our assumption is that the heterogeneity of member states regarding regime type, economic power and social economic performance is greater in global IOs than in regional ones, as member states in the latter often share a common history, often face similar socio-economic conditions and are often exposed to similar contextual challenges and opportunities. We collect the information from IO treaties and code all IOs in which state membership is based on geographical criteria with 0, whereas all IOs open to countries worldwide are coded with 1.
The operationalization of hypothesis 3 starts from the empirical observation that all IOs in our sample engage per definition in some form of standard-setting (i.e. passing policies, norms, rules). In addition, some of these IOs also have an operational function (i.e. electoral observation, peace keeping, monitoring nuclear nonproliferation). The independent variable of hypothesis 3 captures if IOs are also engaged in operational activities. We check the websites of the IOs to find information on the activities conducted by IOs. IOs which run own operational projects and activities are coded as 1, while IOs which only focus on standard-setting, regulation and coordination among the member states are coded as 0.
The fourth hypothesis expects that professional diplomats are more socialized to open discussion than government members. In order to measure the diplomatic level of national delegates attending IO negotiations, we use a binary indicator: IOs, in which diplomats convene in the main legislative body are coded with 0. IOs, in which heads of state or government or ministers discuss policies, in the main legislative body are coded with 1 (this also applied too IOs in which the heads of state and governments are accompanied by diplomats). The information is collected based on the primary law of IOs.
Hypothesis 5 states that IOs that cover many different policy fields should allow for more debate between delegates in the agenda-setting but not in the negotiation stage. To assess the IO policy scope, we code the primary-law based competencies of IOs in eight different fields, namely economic/finance/labor cooperation, security/ disarmament cooperation, health/safety issues, environment/nature, science/technology/transport, culture, human rights and other issues. Subsequently, we count the number of fields in which an IO holds competencies, leading to a score from 1 to 8.
The dependent variables are the two factors as identified in the first part of the paper: the negotiation-stark and the agenda-stark type. Both dependent variables are continuous in nature. As neither the DVs themselves, nor-more importantlythe residuals-are normally distributed, we use OLS regression with Huber-White robust standard errors.
The results of the analysis are depicted in Table 2. While the first three columns show the effect of the different independent variables on the likelihood of an IO to belong to the negotiation-stark type, the last three columns, respectively, do the same for the agenda-stark type. Taking a look at the results, we see that some but not all theoretical expectations find empirical support.  Concerning hypothesis 1 on IO size, we face a counterintuitive finding: in contrast to our expectations, a higher number of member states does not decrease, but increase the likelihood of an IO to include many design elements fostering discussion in the negotiation stage (Model 2) as well as in the agenda-setting stage (Model 5). This might reflect that the need for the opportunity to engage in debates is more pronounced the larger IOs are. More member states bring a higher number of different positions to the negotiation table, which have greater chances to become reconciled when the IO provides more opportunities to engage in debate. In other words, larger IOs might exhibit more institutional features fostering debates in the agendasetting as well as in the negotiation-stage than smaller IOs, in order to avoid gridlock and IO non-decisions that might, in the longer run, turn IOs into zombies (Gray 2018).
Hypothesis 2 is plausible for the negotiation-stark type. The regression analysis shows that a homogeneous membership, the independent variable of Hypothesis 2, makes an IO less likely to belong to the negotiation-stark type (Model 1). As expected, heterogonous IOs call for more debate to allow the diversity of positions to be articulated during the negotiations. For IOs resembling the agenda-stark type, the coefficient is also negative but not significant. This suggests that heterogonous IOs place stronger emphasis on providing room for voicing the high number of different positions once the agenda is already set.
The regression table further illustrates that IOs with an operational focus have a considerably higher chance of belonging to the negotiation-stark type (Models 1, 2 and 3), while the results for the agenda-stark type (Models 4, 5 and 6) are inconclusive due to the lack of significance. These findings are not in line with hypothesis 3, which expected that IOs with operational focus crucially depend on swift decisionmaking and therefore opt for institutional designs that limit opportunities for debate in the agenda-setting as well as the negotiation stage.
According to Hypothesis 4, IOs in which diplomats are the main actors in the legislative arena should seek to induce more discussion than IOs in which high-level delegates like ministers and heads of states participate in the debate. Indeed, the results of the statistical analysis show that IOs with higher ranking national delegates are less likely to provide many opportunities for debate in the actual negotiations (Model 3). The coefficient for the agenda-setting-stark type is also negative, yet not significant (Model 6). Taken together this suggests that IO institutional design seeks to induce more extensive debate in the negotiation stage, when diplomats are the main actors, but not systematically in the agenda-setting stage.
Hypothesis 5 expects that IOs which cover many policy fields should settle for institutional designs that seek to induce debate in the agenda-setting stage, yet restrict extensive debate in the negotiation stage. The findings of our analysis support this expectation. IOs with a broad policy scope tend to be more likely to foster discussion between delegates during the agenda-setting (Models 4 and 5), but are less likely during the actual negotiations (Models 1 and 2). Thus, IOs with many policy mandates subscribe to the IO type that provide opportunities for debate in the agenda-setting stage, while IOs with narrow policy scopes are more likely to adopt institutional designs reflecting the negotiation-stark type.
Taken together, driving forces toward the negotiation-stark type, in which debate is fostered in the negotiation stage, are heterogeneity among members, diplomats as main actors, as well as a broad policy scope (see Table 3).
While these three factors were expected, the regression analysis furthermore revealed that IO size as well as an operational focus are also conducive to IOs opting for the negotiation-stark type of institutional design. With respect to the agendastark type, which provides room for debate in the agenda-setting stage, IO size and in tendency also a broad policy scope turned out to be significant factors. While IO size does not point in the expected direction as larger IOs tend to be equipped with more instead of less opportunities for debate, the effect of IO policy scope is in line with the hypothesis. IOs with broader policy scopes are more likely to be equipped with institutional designs that seek to induce debate in the agenda-setting stage.
Thus, discussion between delegates is a decisive element in different stages of IO decision-making processes. IOs seek to foster debates through a multitude of different institutional design features. However, not all IOs pursue the same techniques to achieve this goal. As the factor analysis has shown, there seem to be two preferred strategies how IOs seek to foster debate: Either they include a lot of institutional design feature conducive to debate in the negotiation stage or in the agenda-setting stage. The presence of two types, the negotiation-stark and the agenda-stark type, suggests that IOs are oriented toward high-quality outcomes as well as efficient decision-making processes and, therefore, tend to focus specifically on either fostering debate in the agenda-setting stage or in the negotiation stage of the policy cycle, while allowing for speedy processes in the others.
While the two types identified in the factor analysis provide a useful and telling heuristic to depict and compare how IOs seek to induce debate between their member states, these two strategies of inducing debate are neither mutually exclusive nor exhaustive. In reality, several IOs combine elements of both types. These are located in the upper right corner of Fig. 3 and include the United Nations Children's Fund (UNFC), the UNGA, or the UNFCCC. Besides, there are also some organizations which cannot meaningfully be aligned to either of the types, as they generally do not strongly foster discussion between delegates through their institutional designs. These organizations are located in the lower left corner of Fig. 3 and include WCO, the Organization of Islamic Cooperation (OIC) and the Association of Natural Rubber Producing Countries (ANRPC). The discussion of the findings from the regression analysis reveals that in general the hypotheses seem to be more adequate to explain why IOs include a high number of elements in the actual negotiation stage than in the agenda-setting stage. Figures 4 and 5 show the average effect of two strong dichotomous explanatory variables, heterogeneity and diplomatic level, for both types. The figures illustrate that the effect of the two variables goes in the same direction for both types: A global membership as a proxy for heterogeneity increases the chance that an IO includes design features conducive to debate in the negotiation (Factor 1) as well as in the agenda-setting phase (Factor 2) (see Fig. 4). In turn, the presence of high-level state actors (rather than diplomats) has the opposite effect (see Fig. 5).  However, the two figures also indicate that the visible distance between the averages is much larger for the negotiation-stark type of IOs (Factor 1) than for the agenda-stark type of IOs (Factor 2). The effect of both independent variables is much more accentuated for the first type. Why do our hypotheses concerning the extent of provisions for debate seem to have more explanatory power for the negotiation-stark type of IO institutional design? One potential explanation could be that the discussion-inducing institutional design elements coded in the negotiation stage are very informative and precise indicators about the extent of discussion an IO allows in this phase, and that these indicators can hardly be associated with other explanatory factors. The provisions for debate identified in the agenda-setting stage are also clearly connected to the extent of debate in IOs. However, some elements, as for instance the possibility of member states to discuss and adjust the agenda at the beginning of a meeting, might not only increase the potential for discussion within IOs, but also constitute an important provision with regard to the agenda-setting powers of participating actors. Accordingly, the decisions of IOs whether to include such a design element might not only be driven by considerations about fostering debates between actors but might also be indicative for the level of authority delegation (or the lack thereof) in an IO in this stage. IOs with less room for debate between state delegates in the agenda-setting stage tend to assign greater leeway to IO secretariats or other agents put in charge of agenda-setting (c.f. Hooghe and Marks 2015). Thus, choices for an agenda-stark type might not only reflect choices about the extent of debates that should be fostered in this IO, but might also be influenced by factors conducive toward the delegation of authority in IOs.

Conclusions
The investigation of 114 IOs revealed that all of them encompass institutional design elements that seek to foster debate between state delegates. Yet there is not a one-size-fits-all approach. There are two types of IO institutional designs, namely a negotiation-stark and an agenda-stark one. The negotiation-stark type seeks to foster debate through the inclusion of many design elements conducive to discussion in the negotiation stage. The agenda-stark type of IO institutional design attempts to induce debate between delegates especially in the agenda-setting stage.
IOs are confronted with two competing aims: speedy decision-making and highquality outcomes: While debate between delegates is conducive to the latter, it inhibits the former (Rada 2000;Golub 2007). Most often, IOs tend to balance both aims. While the negotiation-stark type places emphasis on fostering debate in the negotiation stage and is thereby likely to be subject to a slower speed of negotiations, the second type of IOs entails many institutional design features that seek to foster debate in the agenda-setting stage. At the same time, the first type allows for efficient agenda setting, while the second type allows for an efficient negotiation phase.
The paper has also illustrated that the variables behind the empirical variation of IO institutional design are not identical when comparing the negotiation-stark with the agenda-stark type. A high number of member states is the only explanatory variable which significantly increases the chance of both types of IOs to include design elements fostering debate in the negotiation and in the agenda-setting stage, respectively. In contrast, other variables only seem to influence the actual negotiation stage. Homogeneity of actors, high-level politicians as delegates and a focus of IOs on policy-making activities decrease the likelihood that an IO includes many provisions in its institutional rules that foster discussion between delegates in the actual negotiation stage. Finally, a broad policy mandate of IOs and the thereof resulting high workload exerts a diverse effect on both types: On the one hand, a broad mandate tends to increase the chances that the agenda-setting IO type seeks to induce discussion between delegates. On the other hand, a broad policy scope decreases the likelihood that IOs belonging to the negotiation-stark type include elements inducing debate between state actors.
Despite the existence of many different institutional design elements at different stages of the policy cycle and the multiple possibilities to combine them, institutional designs of many IOs strongly resemble each other as they follow two typical patterns, namely negotiation-stark and agenda-stark IOs. An important and interesting question is how these resemblances may be accounted for. In recent years, IR research has increasingly studied diffusion processes, examining whether IO designs are not set up independently but shaped by mutual learning, imitation and adaption (Ovodenko and Keohane 2012). Regarding the deliberative institutional design of IOs it is highly plausible to assume that similar processes are at play, but this is up to future research. Another crucial, yet so far under-investigated aspect is the nexus between the institutional provisions and IO performance. Thus, it would be highly important to study how the two types of IOs score in practice with regard to performance aspects such as problem-solving capacity and legitimacy. The coding scheme of deliberative elements in IO institutional design was developed by following a mixture of deductive and inductive practices. In a first step, we collected a list of design elements potentially inducing deliberation based on deliberation, discourse and negotiation literatures. In a second step, we used the very detailed and extensive institutional rules of the UNGA to compare our coding scheme with a real-world example and enhanced it by elements we, to this point, have not thought of. In a third step, we applied the coding scheme to a larger subsample of IOs to identify those design elements, which do not only appear in the particularly detailed institutional provisions of the UNGA, but were mentioned in other IOs as well. We arrived at the final coding scheme, by re-formulating the elements and descriptions in a manner applicable to all IOs. For instance, several items relate to the role of chairs, that are called differently in different IOs (e.g., secretary general, president, chairperson, etc.) but perform similar tasks concerning the management of debate in different stages of the policy cycle.