Under what conditions does bureaucracy matter in the making of global public policies?

This study investigates how configurations of bureau-cratic autonomy, policy complexity and political contes-tation allow international public administrations (IPAs) to influence policymaking within international organizations. A fuzzy-set Qualitative Comparative Analysis of 17 policy decisions in four organizations (FAO, WHO, ILO, UNESCO) shows that all IPAs studied can be influential in favorable contexts. When policies are both contested and complex, even IPAs lacking autonomy can influence policy. If either complexity or contestation is absent, however, it is the variant of autonomy of will that helps the IPA exploit procedural strategies of influence. Low autonomy of will, among other factors, explains why IPAs cannot exert influence. Conversely, the variant of autonomy of action appears largely irrelevant. The study provides new insights into the role of bureaucracy beyond the state, exemplifying how research of bureaucratic influence can yield more systematic results in various empirical settings.

basis of international bureaucratic autonomy is put center stage, as the essence of bureaucratic organization resides there (Caughey et al., 2009;Trondal, 2011).This approach allows to identify and compare IPA influence across systematically selected cases of real decision-making.We go beyond individual-level survey data measuring perceived influence (e.g., Nicholson-Crotty & Miller, 2012).Instead, we use a combination of in-depth case studies and a standardized coding scheme for assessing administrative goal attainment (Dür, 2008) to systematically compare actual bureaucratic behavior in 17 concrete cases of intra-organizational decision-making in four IOs: the Food and Agriculture Organization of the United Nations (FAO), the International Labour Organization (ILO), the World Health Organization (WHO), and the United Nations Educational, Scientific and Cultural Organization (UNESCO).Using fuzzy-set Qualitative Comparative Analysis (fsQCA), the article clarifies under which conditions policy influence by international secretariats does or does not take place.Our main finding is that all IPAs studied can be influential-but only if the policy context is favorable.If a decision is not simultaneously complex and contested, autonomy of will becomes important because it enables the secretariat to actively structure the process and develop independent policy solutions.Surprisingly, autonomy of action appears irrelevant for enabling the IPA to exert influence when not accompanied by autonomy of will.
While our study provides an approach to how research on bureaucratic influence can achieve more systematic and comparative results, we cannot generalize our empirical results beyond the scope conditions outlined below.Hence, future research should broaden the range of IOs studied to establish especially whether and how IPAs with strong autonomy of both will and action influence policymaking.In sum, while the external validity of our findings remains necessarily restricted, the approach developed is of high relevance for the study of influence beyond the transnational context studied here.
The next section indicates three major gaps in the debate on the role of IPAs in IO policymaking and how this study addresses them.Based on these considerations, we develop our analytical framework to explain international bureaucratic influence.Section 3 details the logic of the case selection, how the data was collected, and the use of fsQCA.The final section discusses the implications of our findings for future research about policy influence of international (and national) bureaucracies.

| WHEN DO IPAS INFLUENCE POLICY, AND WHAT ROLE DOES STRUCTURAL BUREAUCRATIC AUTONOMY HAVE?
In recent years, International Relations scholars have increasingly studied IOs as autonomous actors and used principal-agent approaches to open the black box of their organization (see e.g., Hawkins et al., 2006).For instance, research on IO delegation (Hooghe & Marks, 2015) and performance (Lall, 2017;Tallberg et al., 2016) have fostered our understanding of how intra-organizational factors (including international bureaucracies) affect IO policymaking.Yet, despite the growth of research on IPAs, also within other disciplines (Bauer et al., 2017;Eckhard & Ege, 2016;Littoz-Monnet, 2021;Moloney & Stone, 2020;Weinlich, 2014), three limitations remain with regard to the comparative study of international bureaucracies' influence on the making of global public policies.First, scholarship is still highly divided on how to systematically observe, measure, and compare such bureaucratic influence across cases.Second, the relationship between bureaucratic autonomy and policy influence remains unclear despite the core role political science grants to it.Studies do stress structural bureaucratic autonomy as an important intra-organizational factor (Cortell & Peterson, 2022;Haftel & Thompson, 2006;Manulak, 2016;Urpelainen, 2012).Yet, empirically grounding the impact of structural bureaucratic autonomy in actual instances of IO policymaking proved difficult.Third, while scholars have posited potential conditions for bureaucratic influence (Biermann & Siebenhüner, 2009a;Cox & Jacobson, 1973;Johnson & Urpelainen, 2014;Weinlich, 2014;Xu & Weller, 2004), we do not know the exact configuration of these conditions that trigger influence.Each of these gaps are central to our research question.

| How to systematically observe and compare bureaucratic influence of IPAs?
IPA influence is "the sum of all effects observable for, and attributable to, an international bureaucracy" (Weinlich, 2014, pp. 60-61).Yet, influence is an elusive concept which is hard to pin down (Ege et al., 2019(Ege et al., , 2021)).In order to empirically study bureaucratic influence, one needs to define the situation or process in which influence is thought to occur.One also needs to specify the object of influence, that is, the specific subject matter that changed assumingly as influence was exerted.In the literature, there is no consensus regarding these epistemological problems.For instance, Biermann et al. (2009; see also Nay, 2012;Widerberg & van Laerhoven, 2014) study cognitive influence as the capacity of the bureaucracy to gather, shape and disseminate information and knowledge.While the generation and dissemination of information may help secretariats to become influential, defining it already as a type of influence renders it difficult to identify instances of non-influence.If one conflates the characteristics of a bureaucracy (such as their capacity to disseminate information) with the observable and attributable effect of their intervention, one cannot establish an empirical relationship between bureaucratic characteristics and actual influence.Even if IPAs disseminate information and member states use this information, this does not (yet) unequivocally constitute IPA influence.Therefore, we focus our analytical perspective on the sum of observable effects related to the most important potential impact of bureaucracies, namely, on policy content.This helps us put forward a more precise estimation of IPA influence.
Any approach to study IPA influence empirically faces difficult tradeoffs.IOs exert both service and program functions (Cox & Jacobson, 1973).When putting policy content center stage, the program-related, norm-setting (and often regulatory) activities of IO policymaking are crucial.According to our understanding, bureaucratic influence is the particular aspect of an IO decision that can be attributed to the presence and specific behavior of an IPA.Therefore, we focus on "relative change" of policy content and follow the counter-factual reasoning that if the administration had not been there, or had it not acted the way it did, then the decision under scrutiny would have been different (Biermann et al., 2009, p. 45).Furthermore, successful IPA influence requires the existence of explicit administrative preferences in favor of a particular aspect or policy option.The IPA's efforts must be discernible and attributable on the part of the administration to justify or defend this particular aspect or option.Going beyond a structural potential for action, these preferences need to manifest themselves in concrete entrepreneurial behavior (Nay, 2011).A congruence between observable IPA preferences and the final policy output (Ege et al., 2019, p. 13) that can be traced back to specific bureaucratic influence strategies is required (see Table 4).

| How does structural bureaucratic autonomy impact IPA influence?
(Intra-)bureaucratic characteristics are seen as the main source for organizational "autonomization" (Weber, 1972).The very factors that enable the bureaucracy to fulfill its tasks efficiently are also important sources of autonomous power and influence, allowing the administration to develop a life of its own and escape the control of its political principals (Eisenstadt, 1958, p. 103).In this context, it is important to remember that autonomy and influence are two different concepts that must be kept analytically distinct (see Weinlich, 2014, p. 57)."Even if an actor has autonomy […], it may still have very little influence" (Haftel & Thompson, 2006, p. 256).Yet, despite recent research on formal and informal administrative patterns (see e.g., Bauer et al., 2017;Knill et al., 2019;Trondal et al., 2012), systematic empirical analyses linking bureaucratic autonomy to IO policymaking and IPA influence are scarce (Cortell & Peterson, 2022;Ege et al., 2019; but see Barnett & Finnemore, 2004;Manulak, 2016;Urpelainen, 2012).
Conceptualizations of autonomy (see Verhoest et al., 2004) concur that structural features, especially those related to executive characteristics, administrative resources as well as organizational competences, determine whether an IPA can act independently of its members, and therefore capitalize on its organizational autonomy (Brown, 2010;Haftel & Thompson, 2006;Hooghe & Marks, 2015;Trondal, 2011).Focusing on intra-organizational features (instead of others, like, perceived expertize or reputation), Bauer and Ege (2016) conceptualize bureaucratic autonomy as being rooted in structural factors of, and formal relationships within IPAs.Bureaucratic autonomy thus consists of both an IPA's ability to develop autonomous preferences (autonomy of will), and the capacity to transform these preferences into action (autonomy of action).While structural features are used to operationalize both autonomy of will and autonomy of action, the conceptualization itself is broader and relies on four dimensions.
In order to develop autonomy of will, an administration first requires "administrative cohesion" to overcome obstacles to collective action and to interact with political actors as an integrated organizational entity (see Mayntz, 1978, p. 68).Cohesion enables the bureaucracy to develop a "single set of corporate goals," allowing its members to work toward the same cause (Caughey et al., 2009, p. 3).Cohesion does not refer to policy-specific units that prepare concrete proposals but describes the general ability of the administration to act as one entity.Moreover, an autonomous will entails "administrative differentiation," that is, the bureaucratic capacity to develop preferences that can potentially differ from those of the political principals (Bauer & Ege, 2017).Establishing autonomy of action crucially depends on statutory powers (i.e., formal secretarial competences vis-à-vis political principals throughout the policy cycle) and independent financial and personnel resources (Brown, 2010;Hooghe & Marks, 2015).Figure 1 summarizes the conceptualization of bureaucratic autonomy.
This distinction between autonomy of will and autonomy of action is employed as an empirical yardstick for measuring international secretariats' differential potential to influence IO decision-making, and thus provides the basis for subsequently examining the relevance of different degrees of autonomy for the occurrence and non-occurrence of influence (see Section 3.4).

| The conditions under which IPAs are able to exert influence
IPAs can be powerful actors who use their central position within the organization, their informational advantage, and their expert authority vis-à-vis member states to influence IO actions and decisions (Eckhard & Ege, 2016).While IPAs have been shown to influence policymaking in particular cases (Margulis, 2018), the conditions under which this influence occurs are controversial, and so is their relative explanatory power (Widerberg & van Laerhoven, 2014, p. 304).Accordingly, our framework (summarized in Table 1) rests on three sets of explanatory factors that enable influence to occur (see e.g., Eckhard & Ege, 2016;Liese & Weinlich, 2006, p. 515).Bureaucratic-organizational factors (in the form of autonomy of will and action) capture the characteristics of the IPA itself (Bauer & Ege, 2016;Biermann & Siebenhüner, 2009b;Trondal, 2011).Functional factors (such as programmatic complexity) relate to the underlying policy problem (Cox & Jacobson, 1973;Xu & Weller, 2004), whereas power-related factors (such as contestation) concern the policy preferences of the most important IO members (Copelovitch, 2010;Urpelainen, 2012).

| High bureaucratic autonomy
Characteristics of the administration itself are crucial when explaining bureaucratic influence.The secretariat's autonomy from member states is an important resource that-under specific conditions-can be used to influence policy (Weinlich, 2014, p. 62).This autonomy provides the administration with room for maneuver beyond its formally delegated discretion (Carpenter, 2001;Cortell & Peterson, 2022, p. 403).Autonomy of will is expected to allow the IPA to become influential early in the process.Being internally cohesive and equipped with the capacity of administrative differentiation, this condition allows it to actively shape the policy discourse and offer particularly innovative solutions to the problem at hand.Autonomy of will (even in the absence of formally delegated authority in form of powers and resources) permits the secretariat to take the lead in policy conceptualization and the search for suitable policy F I G U R E 1 The dimensions of structural bureaucratic autonomy of international public administrations.Source: Bauer and Ege (2017).solutions by applying strategies related to expertize and framing.Organizing conferences, inviting speakers, collecting information and publishing their own research are just some of the activities by which the IPA can influence the course of events and the context of the final decisions.Moreover, a tendency to recruit the executive head of the IPA from within (instead of relying on experienced diplomats from a member state) is also conducive to such behavior because such a leader would be more likely to engage in this type of autonomous behavior.Autonomy of action, by contrast, which is rooted in the powers and resources of an IPA, allows an administration to translate its preferences into action during the actual decision-making process.Autonomy of action reflects the formal authority delegated to the IPA by the IO members.Thus, even if the ability to develop independent preferences is weak, autonomous action capacities may still open up avenues of influence.In these cases, influence mechanisms might be particularly related to the strategic use of procedural knowledge and the intricate involvement of the IPA in policy-making.One major way to influence policy negations is by actively setting the agenda.This is especially successful if the executive head possesses an explicit right of initiative.But even if the IPA is not granted such a formal power, there might be other means such as the cooperation with like-minded members during the negotiation process that would allow it to still influence the final decision (Ege et al., 2021, pp. 741-742).In sum, the broader expectation is that an international bureaucracy needs both, the capacity to develop autonomous preferences (autonomy of will) and the ability to translate these preferences into action (autonomy of action; see Bauer & Ege, 2016;Caughey et al., 2009;Maggetti & Verhoest, 2014, p. 239), to exert bureaucratic influence.

Condition contributes to the non-occurrence of influence when…
To be sure, while at the level of individual policy participants, explicit policy preferences (or an autonomous will) are a precondition for influence (and are thus part of our conceptualization of influence as described below), the structural-organization level, at which our autonomy concept is observed, can only capture a potential that still requires concrete administrative agency.Thus, we do not expect autonomous IPAs to be always or automatically influential.Conversely, one can conceive of situations in which even IPAs with low autonomy of will and action may make a difference.For instance, despite lacking structural autonomy potential at the IPA level, there may still be individual bureaucrats who in certain situations may act as policy entrepreneurs, thereby influencing the run of events.Even though these situations are rare, our configurational analysis (see Section 3.1) allows us to identify and discuss the reasons and context factors for such a low-autonomy-high-influence scenario.

| High (programmatic) complexity
Policy problems require different capacities to address them (Biermann & Siebenhüner, 2009b;Thomann et al., 2019).Especially at the international level, policy problems are often programmatically complex and the policy related support that IO decision-makers receive from their constituents (e.g., governments, permanent missions, and social partner associations), is not always sufficient to adequately address a particular problem.A policy can be considered programmatically complex when it is demanding or difficult to address because it requires context-specific knowledge or technical expertize.When policy problems are programmatically complex, the IPA is often the only actor in the international system that can provide well-informed policy solutions and expertize (Busch & Liese, 2017;Johnson & Urpelainen, 2014).This need for a particular kind of expertize creates opportunities for IPAs to eventually influence the underlying decision.Thus, we expect high programmatic complexity to be a condition that invites high bureaucratic influence.

| High contestation
Contestation of problems and their solution creates room to maneuver for the IPA, especially opportunities for collaboration with like-minded members, and is therefore considered an important condition for IPA influence (Dijkstra, 2017).For instance, a contested decision can help break up the collective principal into multiple principals, thus allowing the IPA to cooperate with certain decision-makers to ensure an agenda item is formally discussed.If, by contrast, all relevant members are united and had already agreed on a particular solution, there is no room for the IPA to influence the output.Classically, this condition is of particular relevance for decisions based on majority rule (Lyne et al., 2006, p. 45).But even under the consensus requirements that usually coin UN negotiations, it is possible for the IPA to act as mediator and help overcome the divergent preferences of members by offering its own solution to the imminent problem.However, contested decisions are also found to be highly politicized by member states and hence potentially associated with low bureaucratic influence (see Copelovitch, 2010;Louis & Maertens, 2021).Given these ambiguous findings, we follow a conservative approach and refrain from formulating directional expectations about bureaucratic influence for politically contested decisions in order to avoid unfounded counterfactual assumptions during the analysis.
It should be noted that the assessment of the explanatory relevance of autonomy for bureaucratic influence constitutes the main research interest of this article.Bureaucratic autonomy is the angle chosen to look at the influence equation.Therefore, autonomy of will and action receive particular attention when we discuss the conceptualization and the empirical results.For the empirical analysis, however, all conditions are equally relevant.Furthermore, we acknowledge that the literature associates more factors with influence than can be accounted for in the present analysis.Thus, we also coded issue salience and preference heterogeneity of IO members (Eckhard & Ege, 2016, pp. 968-969).Yet, we decided to stick with the more parsimonious model because issue salience is usually a feature of individual states and thus difficult to observe across all IO members.It also overlaps both theoretically and empirically with issue contestation, which combines the notion of high perceived importance (i.e., salience) with the existence of heterogenous preferences of IO members.To make sure we did not exclude other relevant factors, we conducted additional checks for potentially omitted conditions by identifying the best-matching pairs of most typical and most deviant cases for qualitative follow-up analysis (Oana et al., 2021) and present the main findings in Section 4. Owing to space limitations, the details about the case studies can be found in Supporting Information S1. 1

| RESEARCH DESIGN
We first introduce the method of fsQCA and discuss the case selection and data collection.We then detail the measurement and calibration of the outcome "international bureaucratic influence" and the four explanatory conditions.

| Analytical method
Given the intricacy of administrative action, we seek to capture how bureaucratic autonomy of will and action combine with issue complexity and contestation, leading to either the occurrence or non-occurrence of IPA influence.FsQCA (Ragin, 2014) is a set-theoretic method that models the interplay of several conditions, and whether such combinations of factors are necessary or sufficient for an outcome to occur, in our case for bureaucratic influence.FsQCA thus allows one to assess whether bureaucratic influence has more than one equally plausible explanation (equifinality), and whether the absence of bureaucratic influence has a distinct explanation (asymmetry)-characteristics that are important in our empirical cases.As our analytic interests do not correspond with a "ceteris paribus" approach, fsQCA is better suited than regression methods to analyze the case-specific data at hand.Simultaneously, fsQCA offers a more formalized comparative approach than a simple comparative case study design, for which our N is also too high (Oana et al., 2021).
Technically, fsQCA conceptualizes the phenomena of interest as "sets" in which cases can have "membership" or not, as well as different "degrees of set membership" between a range from 0 to 1. Values above 0.5 indicate "partial" or "full" membership in a set, and values below 0.5 "partial" or "full" non-membership.For instance, some cases may experience the occurrence of IPA influence (INFL), and others may not (∼INFL where the tilde sign is read as "not").Sometimes the IPA's preferences are only partially (fuzzy set membership 0.66), and sometimes fully reflected in the policy output (fuzzy set membership 1; see Table 4).We conceptualize explanatory factors as conditions (high programmatic complexity, high contestation, and high bureaucratic autonomy), and the explanandum as "outcome" (high bureaucratic influence).
We first assess necessary conditions for the outcome.To assess sufficient conditions in a second step, all logically possible combinations of conditions are displayed in the rows of a "truth table" (see Table 5).Each case displays one of these configurations.We then assess whether all, or the majority of, the cases in a given truth table row also display the outcome-in which case, the configuration is sufficient for the outcome.During the process of logical minimization, the QCA package implemented in the R software then derives the shortest possible logical description of those configurations that are sufficient for the outcome (Duşa, 2019).The logic is straightforward: If, say, one consistently observes the occurrence of bureaucratic influence under conditions of high autonomy of will, high autonomy of action, and high complexity; but also under the conditions of high autonomy of will, low autonomy of action, and high complexity, then whether autonomy of action is high or low does not matter for the outcome to occur.Thus, the result is a shorter expression: that high autonomy of will together with high complexity is sufficient for the occurrence of bureaucratic influence.
All QCA results must pass three criteria using various parameters that all range from 0 to 1: that of empirical consistency, empirical coverage, and substantive importance.Employing Enhanced Standard Analysis (Oana et al., 2021), we make theoretically informed counterfactual assumptions based on Table 1.The Supporting Information S1 displays all materials and indications needed for replication.

| Case selection and data collection
We applied a two-step selection of policy cases.Starting at the organizational level for which autonomy data for 20 IPAs is available (Bauer & Ege, 2017), we tried to hold constant the organizational context factors that may also affect IPA influence, enabling us to focus on our four explanatory conditions.Thus, we selected only specialized agencies of the United Nations system from the issue area of social regulation, with a degree of organizational maturity (older than 50 years), and with equality-based voting rights for each member.We consider these factors essential for the study of IPA influence because they determine how the secretariat interacts with IO members.Consequently, our findings are restricted in their external validity and are directly applicable only to UN organizations that mainly regulate global social policy and have established institutional structures where all members are formally treated equally.By contrast, to study the autonomy-influence nexus we selected IOs that are characterized by different configurations of structural autonomy, stemming from differences in administrative cohesion, differentiation, powers and resources.Against this background, we conduct the empirical analysis of IPA influence in UNESCO, ILO, FAO and WHO.
Table 2 shows that both UNESCO and ILO have low autonomy of will and action.We expect this type of organization to be relatively passive during policymaking, rather providing technical assistance or monitoring tasks.Yet, because both UNESCO and ILO have some degree of administrative differentiation, they serve as important control cases that may still be able to influence policies under the right contextual conditions.The FAO, with high autonomy of will but lacking autonomy of action, is expected to be especially influential in the early policymaking stages of problem definition and policy initiation.While the FAO is the only IPA in our sample with high autonomy of will, it can be considered a typical "ideational bureaucracy" (see Ege, 2017).The WHO complements our sample by featuring the opposite configuration-with no autonomy of will, but a strong capacity for autonomous action.We expect that our findings will be indicative of other IPAs with similar autonomy configurations.
The next step was to identify concrete and meaningful policy cases-coping with the challenges in measuring the influence concept discussed in Section 2.1.We used an online survey to ask IPA staff members, member states and stakeholders (permanent representations and transnational actors [TNA] involved in the respective issue area) what they consider the most important policies adopted in their respective work environment in recent years (see Supporting Information S1 for a distribution of respondents). 2The involvement of the IPA in these important policies was not addressed to avoid selection bias.We first grouped answers according to the IO's area of activity, and then summarized and counted similar answers.Third, we conducted follow-up research on the most frequently mentioned policies to dismiss cases that could not be linked to a concrete decision. 3We excluded decisions taken more than 10 years ago, since they would be too difficult to reconstruct in follow-up interviews.
Initially, we focused on six policies for each organization.To study these cases in more detail, we conducted 69 semi-structured interviews with IPA staff members, member states and some transnational actors as the major source of coding for all conditions (except autonomy) and the outcome (see Supporting Information S1 for a full list of interviews).We also used organizational documents (decision documents such as a resolution texts or reports) to complement and Note: 1: feature is present; 0: feature is absent.The sub-concepts are aggregated by using a logical AND: for example, if both COH AND DIFF are present, autonomy of WILL is present.
T A B L E 2 Bureaucratic autonomy configurations of the selected four IPAs cross-validate our information from the interview.Since the interviews did not provide a sufficient data basis for all cases, we selected the following 17 for further analysis (see Table 3). 4 While the selection of policy cases was based on the results of the online survey, the final coding of influence, complexity and contestation was done using interview material and documents.The interviews were fully transcribed and (together with the relevant decisional documents and reports) entered into MAXQDA.The data was first coded by trained research assistants and later validated and finalized by the authors.The coding process included the identification of the different observable implications of the outcome and conditions for each of the 17 policies as discussed below.We used frequencies and qualitative assessments of the relevance of the code segments to aggregate the different sources to statements that best reflected the characteristics of the policy case.Finally, the two autonomy conditions were coded at the organizational level based on the structural characteristics described below.

| Measurement and calibration of influence
IPA influence is observed through three essential elements, as summarized in Table 4: (1) the existence of explicit administrative preferences in favor of a particular policy option; (2) efforts on the part of the IPA to justify or defend this option (entrepreneurialism); and (3) congruence between IPA preferences and the final policy output.If all of these three elements are present, bureaucratic influence is considered successful.Adding analytical depth to capture the "degree of preference attainment" (Dür, 2008, p. 566), we distinguish between partial preference attainment and full preference attainment in order to differentiate "partial" from "full occurrence of influence."This allows for a more nuanced conceptualization of different degrees of influence.As we apply relatively hard requirements for determining influence, if anything, we (consciously) risk underestimating rather than overestimating bureaucratic influence.

| Measurement and calibration of the conditions
We included autonomy of will and autonomy of action as two separate conditions in the analysis because we are particularly interested in how specific combinations of these two conditions affect the outcome.We rely on the publicly available data from Bauer andEge (2016, 2017), who measure structural autonomy by using formal executive characteristics, the independence of administrative resources and organizational competences.Table 5 provides an overview of the indicators used to measure autonomy of will and action within the four dimensions.
The aggregate features used to measure the four dimensions are relatively stable over time and thus allow for a cross-case comparison of decisions made at different time points during the observational period.In this regard, autonomy of will is the more challenging part of the concept because the empirical basis of cohesion and differentiation is more difficult to observe than powers and resources.We still concur with the argument that administrative structures are also relevant for autonomy of will because they allow bureaucrats to operate jointly as unified actors that can develop their own preferences to varying degrees and thus draw on characteristics of the IPA structure and staff for operationalizing autonomy of will (Bauer & Ege, 2016, 2017).
We coded the programmatic complexity of the policy at hand.A policy is considered highly complex (coded 1) if access to context-specific knowledge or technical expertize is essential in order to address all dimensions of the underlying problem.We considered a policy to have rather high complexity (coded 0.66) if access to context-specific knowledge or technical expertize is essential to address some dimensions of the problem.Rather low complexity (coded 0.33) is attributed to a policy if context-specific knowledge or technical expertize is useful but not essential.For policy problems with a low level of complexity (coded 0) context-specific knowledge or technical expertize is useful but the usefulness is below the average of comparable problems.
To operationalize the political contestation, we first identified the most visible IO members during the decision-making process, noting who was in favor of and who opposed a particular solution.Since we are not able to talk to all relevant members, we also take debates in the (social) media and the perspective of TNAs into account.This allows us to get a broader picture of the interest constellation and to better detect potential behind-the-scenes activities that might be indicative of a contested issue.The main actors relevant to assess political contestation, however, are IO members (usually states) that are directly involved in the negotiations.Accordingly, contestation was coded 0 if there was a common understanding among IO members of how the underlying problem could and should be addressed-although individual measures might have been (partly) contested.A policy is considered highly contested (coded 1) if the political discourse surrounding a policy output is characterized by strong and possibly heterogeneous preferences among decision-makers regarding possible solutions.This may have resulted in difficult and long negotiations or consultations, and involves contestation of the goals, not only of the measures. 5

| RESULTS
Supporting Information: Table B2 S1 displays the full analysis of necessity.The most notable finding is that in all cases when IPAs did not exercise influence, they also had low levels of autonomy of will: low levels of autonomy of will are thus a necessary condition for the non-occurrence of IPA influence.However, since in only five of 17 cases the IPA experienced high levels of auton omy of will, this result, while substantively interesting, is not surprising (empirically trivial) and should be interpreted with caution.Supporting Information: Table A4 S1 shows the different combinations of conditions with the observed outcome from our cases.After performing logical minimization, Table 6 reveals that there are three sufficient paths for the occurrence of IPA influence, which reflect quite clear organizational clusters (see also Bauer et al., forthcoming).In two of them, high levels of autonomy of will play a crucial role-indicating the importance of this condition for IPA influence.Autonomy of action, by contrast, appears to be irrelevant for high IPA influence in the analyzed cases, where it never combined with high autonomy of will.As Supporting Information: Figure B1 S1 illustrates, this solution has a high consistency and explains nine of 11 cases where the IPA exerted influence.
The first combination of factors leading to the occurrence of IPA influence is when the IPA has high autonomy of will and the issue is complex.We find this scenario in four cases of FAO  B3 S1 for PRI values and justification of raw consistency threshold.Read as: * = "and", + = "or", ∼ = "not", → = "is sufficient for".
T A B L E 6 Sufficient conditions for the occurrence of IPA influence decision-making.Amongst them was the 2017 FAO Strategy on Climate Change.The secretariat was in the lead the whole time and member states only asked some questions after the strategy was presented to them.FAO senior management was strongly involved and kept control over the final draft of the strategy.The question of how to improve FAO's climate change adaptation and mitigation is a multi-faceted issue that was spread across different divisions and teams.Its high autonomy of will and the related research capacities allowed the FAO secretariat to successfully tackle such a complex issue and prepare a road map and conceptual framework through a Strategy Development Team.
The second path resulting in high IPA influence is when the IPA has high autonomy of will and the issue is contested.We observed this in two cases of FAO decision-making, for example, the 2014 FAO "Rome Declaration on Nutrition."The FAO secretariat's high autonomy of will allowed it to play a highly influential role in the policy's preparation and adoption.The initiative for another "International Conference on Nutrition" which could adopt a final Declaration originally came from the FAO secretariat, committed to bringing the global food situation back on the international agenda.By shifting the focus from food security to food systems, which included all forms of malnutrition, they were able to convince both the FAO leadership and the member states to convene the conference earlier than initially planned.This extension of the substantive focus influenced both the negotiations and the content of the Declaration.Food policy is a very controversial topic and especially countries with a strong food industry were opposed to measures to achieve healthier nutrition.From the beginning, the FAO secretariat decided to extensively consult and to provide particular technical background information in order to eliminate existing conflicts, and was willing to weaken and reframe some minor points to reach an agreement.The secretariat actively engaged in advocacy and mobilization, and used its expertize strategically to shift the focus to food systems.Bridging the divide between members was only possible because of the high autonomy of will that allowed FAO staff to steer the process, actively and entrepreneurially, towards the eventually successful negotiation and adoption of the Declaration in 2014.
Third, IPA influence occurred when an issue was both complex and contested.In contrast to the previous two configurations, autonomy of will is not part of this path.Under these particularly favorable conditions, IPAs can generally be influential, no matter their degree of autonomy of will.This scenario prevailed in two ILO and two WHO cases.The case of the 2016 ILO "Resolution concerning decent work in global supply chains" illustrates the interplay of these factors (Ege et al., 2021).Despite being structurally disadvantaged by the absence of autonomy of will and action, the International Labour Office was still able to use both expertise-based and procedural influence strategies to shape the content of the final resolution.After the collapse of the Rana Plaza factories in Bangladesh put the issue of decent work in global supply chains (GSC) prominently on the political agenda, the ILO successfully used its expertize to frame the discourse (Thomas & Turnbull, 2018).It issued a report to highlight governance gaps in GSC and point out that existing regulations were insufficient.While remaining cautious to not be perceived as taking sides, it also succeeded to include its own preferences into the draft resolution: paragraph 25 requested the ILO to review its standards and consider what is needed to achieve decent work in GSC.Applying these expertise-based and procedural strategies was only possible because of the programmatically highly complex and politically contested nature of the policy problem.Compared to the case WHO_NCD, which has the same combination of factors, we see that the ILO had to rely on its ability to maneuver prudently between the diverging interests of its constituents, while the WHO with its high degree of autonomy could act more boldly in the NCD case (Ege et al., 2021).Despite its lack of autonomy of will, the ILO administration was influential.A first reason why the resolution was adopted at all was due to certain external circumstances, namely, the willingness on part of the employer groups and especially its spokesperson to make some last-minute compromises.Second, the autonomy data already indicates that while the International Labour Office is characterized by low internal cohesion, it does still have the capacity for administrative differentiation (see Table 2).Based on this structural perspective, it is not surprising that certain individuals within the ILO administration (especially from the Bureau for Workers' Activities [ACTRAV]) were able to overcome the lack of internal cohesion, still act entrepreneurially and work towards the eventual adoption of the resolution.
As Table 7 below shows, we also detected two pathways leading to the non-occurrence of IPA influence, where autonomy of will is absent.This solution has good consistency but cannot explain two of six cases of IPA influence non-occurrence within our sample (see Supporting Information: Figure B2 S1).
In the first path we find the absence of autonomy of will and autonomy of action, and decision-making processes characterized by high complexity and low issue contestation.This path covers one UNESCO case and one ILO case.Moreover, UNESCO_HERIT where the IPA did exert influence also displays this combination-discussed below.The case of the UNESCO "Global Convention on the Recognition of Qualifications concerning Higher Education" illustrates this configuration.The secretariat's role was limited to process facilitation and providing expertize.The complexity of the issue was due to the high number of universities and the varying higher education systems of the 193 UNESCO member states.An expert group oversaw the technical and legal aspects, and the secretariat could provide expertize based on statistics and experiences concerning the situation in different countries.The member states agreed on the necessity of such a convention, therefore it was not contested.However, some of them perceived it as UNESCO's own initiative, so the secretariat was very eager to remain impartial throughout the process and it did not actively promote its preferences.Due to this passiveness and the lack of structural autonomy, the secretariat could only act carefully behind the scenes.While the secretariat was responsible for preparing the first draft, it was primarily interested in reaching a final decision rather than pushing for its preferred solution.
In the second path, we find that an IPA has no or only attempted influence on the decision-making output when the IPA has low autonomy of will and simultaneously high autonomy of action, the issue is not complex and not contested.This was the case in two WHO decision-making processes.This combination of factors played out in the 2019 WHO resolution "Community Health Workers Delivering Primary Care: Opportunities and Challenges."In the process of adopting this resolution, the secretariat acted as facilitator (supporting mainly Ethiopia and Ecuador who drafted the resolution) and gathered internal and external experts to formulate Community Health Workers (CHW) guidelines, but it did not actively promote a particular solution.The general importance of CHW had already been recognized in 1978 and the new resolution simply urged countries and partners to use WHO's CHW guidelines to optimize programs and allocate resources.This problem was not contested among members and the usefulness of the guidelines was widely recognized.The secretariat did not have to behave in an entrepreneurial manner because it was clear that in such a low-complexity, low-contestation scenario the resolution would be adopted anyway.Furthermore, despite having the necessary action capacities in terms of autonomous personnel and financial resources, as well as the executive power which may have helped the secretariat to gather qualified experts to draft the CHW guidelines, the low autonomy of will that was also clearly visible in the interviews prevented the secretariat from taking a more active stance under such arguably unfavorable conditions.
In order to provide some additional analytical diagnostics and check for omitted factors, we conducted additional post-QCA case studies (Oana et al., 2021).This helped us to under- stand what factors were at play in cases that our solution terms could not explain; and why some cases deviate from the sufficiency pattern.Supporting Information: Table B12 S1 summarizes the selection criteria and central findings.The case studies show that additional policy-specific, situational factors that stem from the particular actor constellations in the respective policy field can also affect bureaucratic influence.Trust in IPA staff helps explain why an IPA can still be influential under otherwise adverse circumstances.The general salience that a policy has for the constituents of an organization has additional explanatory power for non-influence.This general salience (or "politicization" of an issue, see Louis & Maertens, 2021) is a direct consequence of the obligations associated with a decision-be it the bindingness of an instrument or the direct practical implications for policymakers.Thus, it comes as no surprise that IPAs are found to work hard to depoliticize their work which allows them "to remain engaged in politics, even when they pretend not to" (Louis & Maertens, 2021, p. iii).Future research may want to address the role of these factors for IPA influence more systematically.

| DISCUSSION AND CONCLUSION
Policymaking beyond the nation state-dominated by executive-technocratic decision-making, haunted by precarious representative legitimacy linkages and with limited potential for direct participation from citizens-poses afresh the question of bureaucratic influence.As a contribution to this debate, we investigated the specific conditions under which IPAs can or cannot wield influence.We propose a parsimonious model with high explanatory power that could explain 13 of 17 cases (76%).Our analysis showed that all the IPAs studied can in general influence IO policy making.When policies are both contested and complex, even IPAs lacking autonomy stand a chance to influence organizational policy outputs.If one of these two conditions (either political contestation or issue complexity) is absent, however, autonomy-in the variant of bureaucratic autonomy of will-becomes the decisive factor.Accordingly, in scenarios with high contestation only, autonomy of will helps the IPA to actively structure the process and apply procedural strategies of influence.In scenarios with high complexity only, autonomy of will enables the IPA to make use of its ability to develop independent policy solutions and rely on expertise-based influence strategies.This result is not restricted to inward-oriented organizational decisions-such as administrative reforms-but can be observed in more substantial decisions such as soil management or shipping regulations.In sum, when the policy issue was not simultaneously complex and contested, high levels of autonomy of will lead the international secretariat to exert significant influence.Conversely, low levels of autonomy of will-together with different combinations of issue contestation and low complexity-are conducive to low levels of IPA influence.Surprisingly, autonomy of action appears irrelevant for high levels of IPA influence in the absence of autonomy of will.However, as none of the IPAs analyzed here simultaneously had both high autonomy of will and action; this finding may indicate that bureaucracies need to have a will of their own in order to pursue it through actions (Maggetti & Verhoest, 2014).
While our article indicates how research on bureaucratic influence can achieve more systematic and comparative results, QCA results are highly context-specific and allow only for modest generalization.In our analysis, the limited variation in autonomy clusters at the organizational level, and the fact that important results appear in only one organization (FAO), means our results cannot be generalized beyond the scope of the contexts in which our cases operate.Moreover, our findings might not "travel" to logically possible scenarios featuring high levels of autonomy of will or action (or both) that we did not observe empirically in our cases.In particular, we find that autonomy of action is largely irrelevant for influence if not accompanied by autonomy of will.However, we offer a comparative approach enabling future research to address whether, and how, IPAs with both strong autonomy of will and action, which we did not observe, influence policymaking.Moreover, we expect our results to apply to other IPAs that have similar characteristics as relevant scope conditions (Goertz & Mahoney, 2009).For the FAO, these could be other "ideational bureaucracies" such as the secretariat of the Organization for Security and Co-operation in Europe or the World Metrological Organization, which feature the same autonomy configurations (see Ege, 2017).In such contexts of ideational IPAs engaging in norm-setting activities that closely relate to the traditional understanding of policymaking (Cox & Jacobson, 1973), high levels of autonomy of will can plausibly be expected to contribute to high IPA influence in combination with either high issue complexity, or high issue contestation.
Overall, our article contributes to the debate about international bureaucratic influence and provides an empirical-analytical approach enabling researchers to observe, measure and systematically compare bureaucratic influence across policymaking cases and across different organizations.Its empirical focus rests on IPAs, but the disciplined investigation strategy developed here, we posit, is of general applicability.The article thus offers a method to cope with a central problem challenging scholars from various disciplines: how to come to grips with the elusiveness of bureaucratic influence during policymaking across cases and contexts.Applying the suggested approach to more constellations that encompass more types of bureaucracies and include an increasing range of conditions, scholars stand a better chance to accumulate systematic and reliable analytical insights about the conditions under which bureaucracies wield influence.Studying bureaucratic influence is part of revisiting a central question of Public Administration, how bureaucracies concretely matter under conditions of policymaking in the 21st century, and will thus help in "bringing political science back into public administration research" (Peters et al., 2022; see also Kettl, 2022).Public Administration must advance on this question, if the discipline is to keep on contributing meaningfully to the pertinent debates within the increasingly methodologically advanced and data-savvy social sciences.
further analysis, however, because of the low response rate to the survey.While this is unfortunate, it is important to remember that fully autonomous administrations are empirically rather rare.Only four of the sample of 20 already quite institutionalized IPAs display such a combination and they are active mainly in economic regulation (see Ege, 2017).Thus, given the requirement to select similar organizations, the limitations in terms of generalizability appear acceptable.
Expected conditions for the occurrence or non-occurrence of influence 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Sample of policy cases 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Measuring IPA influence 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Note: For detailed descriptions of the selected policies, see Supporting Information S1.T A B L E 3 T A B L E 4 Measuring autonomy 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License T A B L E 5 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License "Deviant case consistency in kind" displayed in bold.See Truth table in Supporting Information: Table B6 S1 for PRI values and justification of raw consistency threshold.Read as: Sufficient conditions for the non-occurrence of IPA influence 14680491, 0, Downloaded from https://onlinelibrary.wiley.com/doi/10.1111/gove.12741 by European University Institute, Wiley Online Library on [01/12/2022].See the Terms and Conditions (https://onlinelibrary.wiley.com/terms-and-conditions)on Wiley Online Library for rules of use; OA articles are governed by the applicable Creative Commons License Note:T A B L E 7