Gender, Justice and Deliberation: Why Women Don’t Inﬂuence Peacemaking

Scholars have pinpointed that women’s underrepresentation in peacemaking results in gendered outcomes that do not address women’s needs and interests. Despite recent increased representation at the negotiating table, women still have a limited inﬂuence on peacemaking outcomes. We propose that differences in female and male speeches reﬂected in the gendered patterns in discourse during peacemaking explain how women’s inﬂuence is curtailed. We examine women’s speaking behavior in transitional justice debates in the post-conﬂict Balkans. Applying multimethod quantitative text analysis to over half a million words in multiple languages, we analyze structural and thematic speech patterns. We ﬁnd that men’s domination of turn-taking and the absence of topics reﬂecting women’s needs and interests lead to a gendered outcome. The sequences of men talking after men are longer than those of women talking after women, which restricts women’s deliberative space and opportunities to develop and sustain arguments that reﬂect their concerns. We ﬁnd no evidence that women’s limited inﬂuence is driven by lower deliberative quality of their speeches. This study of gendered dynamics at the microlevel of discourse identiﬁes a novel dimension of male domination during peacemaking. does not translate into policies that promote gender equality. The article proceeds with a presentation of data and research design, followed by the results and analysis of the gendered nature of discourse. The conclusion reﬂects on the contribution of this study to scholarship and policy.


T O M PASKHALIS New York University, United States
Scholars have pinpointed that women's underrepresentation in peacemaking results in gendered outcomes that do not address women's needs and interests. Despite recent increased representation at the negotiating table, women still have a limited influence on peacemaking outcomes. We propose that differences in female and male speeches reflected in the gendered patterns in discourse during peacemaking explain how women's influence is curtailed. We examine women's speaking behavior in transitional justice debates in the post-conflict Balkans. Applying multimethod quantitative text analysis to over half a million words in multiple languages, we analyze structural and thematic speech patterns. We find that men's domination of turn-taking and the absence of topics reflecting women's needs and interests lead to a gendered outcome. The sequences of men talking after men are longer than those of women talking after women, which restricts women's deliberative space and opportunities to develop and sustain arguments that reflect their concerns. We find no evidence that women's limited influence is driven by lower deliberative quality of their speeches. This study of gendered dynamics at the microlevel of discourse identifies a novel dimension of male domination during peacemaking.

Introduction
Peace is much more than the cessation of violence. Scrutiny of the quality of peace has revealed that peace often fails women (Wallensteen 2015, 45). The end of a conflict provides an opportunity to lay the foundations for gender-just peace, which transforms unequal gender relations providing for women's political, social, and economic agency (Björkdahl and Mannergren Selimovic 2013;Lake 2018). However, post-conflict peacemaking can also introduce norms, structures, and power relations that disadvantage women. Some of these are an extension of gendered conflict dynamics, stemming from the different experience of violence by men and women (Melander 2016); others entail a reversal of women's wartime gains in political and social agency (Tripp 2015;Berry 2018;Kreft 2019;Østby, Leiby, and Nordås 2019). An imperative to make peace work for women has motivated scholars and practitioners to tackle gender inequality during peacemaking. Women's participation in peace processes matters; it is associated with longer and better peace (Demeritt, Nichols, and Kelly 2014;Krause, Krause, and Bränfors 2018). Transitional justice is integral to peace (Sharp 2013). Peace that works for women advances women's representation and rights, including the right to justice for wartime sexual and gender-based violence (SGBV) (Chinkin and Kaldor 2013). 1 Women's presence in peace efforts is critical for bringing about gender-just peace, because women's presence provides women "communicative advantages" (Mansbridge 1999, 642). Women can insert their perspectives into the peacemaking process, which paves the way for recognition of their needs (Brown and Aoláin 2015, 147). Criticism of women's marginalization in peacemaking has resulted in international and national efforts to include women in peace processes (Adjei 2019). However, women's increased presence in these processes has had only a limited impact on their outcomes. Scholars have shown how provisions of peace agreements and mandates of transitional justice instruments overlook or marginalize women's needs, interests, and entitlements (Borer 2009;Bell and O'Rourke 2010;Haynes, Ní Aoláin, and Cahn 2011;Sandole and Staroste 2015). The failure to bring about gender-just peace in contexts where women are represented in the peacemaking process poses a significant puzzle (Castillo Diaz and Tordjman 2012;Coomaraswamy 2015;Paffenholz et al. 2016).
We address this puzzle of women's representation without influence by examining the black box between women's representation indicated by 'bodies at the table' and outcomes that do not reflect women's concerns. It is important to understand women's contributions to debates during peacemaking. A gendered pattern of speaking behavior, which reveals differences in speech by men and women, has a "cumulative effect on power and influence" (Kathlene 1994, 573). We study how discourse in a transitional justice process is gendered since recognition of women's justice needs is part of "gender-sensitive and gender-responsive" perspectives on peace (Coomaraswamy 2015;Davies and True 2019). Gender is a political, social, and cultural construction, which should not be conflated with the sex identity of women and men (Carver 1996). The concept of gender makes visible how behaviors, norms, and discourses associated with the female and male sexes in institutional and informal processes are implicated in the production and reproduc- tion of inequality and oppression (Sandole and Staroste 2015, 119-20;Cameron 1998;Krook and Mackay 2011, 4). Our analysis identifies how gender-based differences in discourse during peacemaking account for a gendered outcome that does not reflect women's needs and interests.
To explain the lack of women's influence despite their broadly equal representation, we propose and test three mechanisms operating in discourse during peacemaking: deliberation, emboldening, and decentering. The first mechanism concerns how women's quality of deliberation, focused on justification of arguments, compares with that of men's. It is premised on the feminist critique of democratic deliberation and institutions that male domination is perpetuated by prescribed modes of communication (Acker 1990;Sanders 1997;Mackay, Kenny, and Chappell 2010). Foregrounding the structure of discourse and focusing on turn-taking sequences, the second mechanism probes whether women are emboldened to put forth their views in deliberative enclaves (Sunstein 2007). It assumes that when women speak after each other in succession in mixed-sex debates, a deliberative space is created that is conducive to the articulation of their needs. Decentering is the third mechanism. Considering that women often decenter, i.e., they avoid discussing directly violence that they have suffered (Kashyap 2009;Theidon 2013;Porter 2016), we explore whether thematic differences in speeches by men and women can explain a gendered outcome that marginalizes women's interests and concerns.
We find that, in conditions where women are broadly represented equally, gendered outcomes that disadvantage women result from men's domination of turn-taking and the absence of topics reflecting women's concerns and interests; in mixed-sex debates, the sequences of men talking after men are longer than those of women talking after women, which restricts women's deliberative space and opportunities to develop and sustain arguments that reflect their needs and interests. We find no evidence that women's limited influence is driven by lower deliberative quality of their speeches. This research shows that a microlevel of discourse during peacemaking is a domain of male domination that has been overlooked by scholars and practitioners puzzled by the elusive influence of women who are present at the peacemaking table.
We use a case of a civil society-led transitional justice process in the post-conflict Balkans, known by its acronym RECOM, 2 to scrutinize gender differences in discourse. From 2010 to 2011, the multiethnic initiative organized debates dedicated to designing the mandate of the regional fact-finding commission, which had emerged as a preferred transitional justice approach in previous rounds of consultations. These debates produced the commission's draft Statute. 3 Defining the commission's mandate, the document failed to respond to women's concerns, needs, and interests, exposing the gender dimension as a weakness of the RECOM's process (Bonora 2019). The draft Statute did not provide for women's equal inclusion in different facets of the commission's operation, nor did it envisage appropriate procedures to facilitate recognition of women victims of SGBV (although SGBV was listed among abuses to be investigated). To advance gender justice, transitional justice instruments also need to include appropriate gender-responsive procedures (Swaine 2018, 231-32), for example for staging women's testimonies about their experience of violence. Why did seemingly vocal contributions of women present in RECOM's debates not lead to the commission's mandate that responds to women's needs, concerns, and interests? To answer this question, we interrogated whether the patterns in discourse are gendered by applying quantitative content analysis, which involves human coding and computer-assisted text analysis of a corpus of over half a million words comprised of the transcripts of RECOM's debates.
Our evidence drawn from the study of women's speaking behavior that reveals how women's influence is curtailed in a mixed-sex deliberative setting advances research about peace more broadly. First, it demonstrates theoretical benefits of the empirical study of processes that can help ensure gender-just "quality peace" (Waylen 2014;Wallensteen 2015), which has lagged behind the study of peacemaking outcomes, such as peace agreements and their effects. Second, we sound a note of caution about crude measurement of women's participation in peacemaking in the existing scholarship and practice (Paffenholz et al. 2016); it captures women's physical presence, describes their roles as signatories or negotiators and specifies whether they take up senior roles (Coomaraswamy 2015, 45), but neglects more refined measures such as how often women take the floor, how many arguments they make, and how long they speak relative to men. Third, we expose the untapped potential of quantitative analysis of discourse for the study of peace, which contributes to insights gained through qualitative study of discourse and its effects (Jennings 2019). Quantifying and understanding the gendered patterns of discourse during peacemaking can help us devise practical interventions that advance peace that works for women.
In the next part of this article, we review scholarly debates about gendered peace and justice, and outline a critique of existing approaches to women's representation and limited influence in peace processes. We then present mechanisms in discourse that can explain why women's representation in peacemaking does not translate into policies that promote gender equality. The article proceeds with a presentation of data and research design, followed by the results and analysis of the gendered nature of discourse. The conclusion reflects on the contribution of this study to scholarship and policy.

Gendered Peace and Justice: (Re-)Assessing Women's Representation in Peace Processes
Inaugurating the Women, Peace and Security (WPS) Agenda in 2000, the UN Security Council Resolution (UN-SCR) 1325 prompted critical rethinking by scholars and practitioners about how to bring about gender-just peace (Kirby and Shepherd 2016, 252). Accounting for "gendered peace" (Pankhurst 2008), scholars have pinpointed systematic underrepresentation of women in peace processes, despite a slow but steady trend of their greater inclusion following UNSCR 1325 (Coomaraswamy 2015, 45). Only 9 percent of negotiators in thirty-one major peace processes between 1992 and 2011 were women (Castillo Diaz and Tordjman 2012). Unsurprisingly, the outcomes of those processes were gendered, in that 16 percent of 585 major peace agreements in 102 peace processes from 1990 to 2010 had references to women and their concerns (Bell and O'Rourke 2010; Ellerby 2016). Our understanding of women's influence in processes that define mandates of transitional justice instruments is even more limited, although their proceedings and effects, for example those of the International Criminal Tribunal for the former Yugoslavia (ICTY), are also gendered (King, Meernik, and Kelly 2017;Gallagher, Prakash, and Li 2020). 4 Analyzing the absence of women in peace processes and its consequences is important (McLeod 2019), but we also need to better understand the limited influence of (the increasing number of) women who are present in these processes. Looking beyond women's proportional representation, scholars have queried the kind of representation of women in peace processes. The inclusion of elite women in peace processes who are linked to elite men or clan leaders has resulted in a "veneer of female legitimacy" and underrepresentation of non-elite women's concerns (Ní Aoláin 2016, 155). Others stress that women often remain silent during meetings, negotiations, and other events in peacemaking processes (Ellerby 2016). Structural constraints provided an alternative explanation. Bell and O'Rourke (2010, 978) contend that incorporating women's concerns would make it harder to reach an agreement or might destabilize existing agreements in the context of power sharing. Normative considerations also play a role. Local men in many conflict and post-conflict settings oppose women's emancipation and gender equality, which are often perceived to be externally imposed (Anderlini 2007;Khodary 2016).
The WPS Agenda has promoted institutionalization of gender equality norms in peace processes (Adjei 2019). However, peacemaking that now includes more women still produces outcomes that do not adequately reflect women's needs and concerns. The study of gender-just peace can benefit from engaging with the scholarship on political representation and communication that examines the gap between women's descriptive and substantial representation. Descriptive representation refers to women's presence in political processes (e.g., national legislatures), while substantive representation captures their influence on policy (Pitkin 1967). Mendelberg, Karpowitz, and Oliphant (2014, 19) observe that "even high descriptive representation does not consistently erase [women's] low substantive representation" in deliberative settings. Women's limited impact on policy outcomes is evident in both Western and non-Western countries, such as Rwanda (Devlin and Elgie 2008). At the same time, women's greater representation may cause a backlash. Kathlene (1994) confirmed Yoder's "intrusiveness thesis" (Yoder 1991), which holds that men react to women's increased presence in the legislative setting by themselves becoming more vocal. 5 However, underrepresentation can also motivate greater participation by women in political debates by incentivizing women to increase their visibility (Pearson and Dancey 2011b, 910).
In contrast to scholars of peacemaking, scholars of political representation and communication have refined measures of women's participation in public debates. Beyond counting women, they consider the proportion of women's speaking turns to men's, as well as the duration of their speeches (Karpowitz, Mendelberg, and Shaker 2012, 21). As Kathlene (1994, 564) points out, "men and women may take an equal number of turns, but men may talk longer than women in any given turn." Men talk more than women in mixed-sex groups and in different conversational contexts (Leaper and Ayres 2007;Parthasarathy, Rao, and Palaniswamy 2019). The shorter length of women's speeches can offset any benefits to descriptive representation-even if 4 Women were underrepresented even in trial panels involving SGBV charges (Sharratt 2011).
there is gender equality in terms of the number of speaking turns. Male dominance in communication holds true even when family issues, which might be expected to stimulate women's participation, are discussed in legislative settings (Kathlene 1994, 569). Furthermore, during a single speaking turn, speakers may present arguments about one or more policy points. Men's dominance in terms of the number of policy points they address may amplify their impact on policy outcomes; alternatively, women's dominance over policy points may compensate for the fewer speaking turns they have. These insights from the fields of political representation and communication reveal a need for a more robust assessment of women's representation in peacemaking beyond the binaries: women's presence versus women's absence or women's silence. When addressing the question of why women's representation does not translate into influence in peacemaking, assessment of women's representation needs to consider both women's proportional presence and the number of speaking turns, the duration of their turns, and the number of policy points they make. If women's presence-thus reassessed-is (broadly) at parity with men's and if it fails to translate into influence on policy, we can turn to the analysis of speaking behavior to find out what hinders translation of descriptive representation into substantive representation of women's concerns in peacemaking.

Presence without Influence in Peacemaking: Mechanisms
We contend that understanding speaking behavior is integral to understanding processes that result in gendered peacemaking outcomes. As Kathlene (1994, 573) points out, discourse analysis of political discussions can explain the gap between women's representation and women's influence on policy outcomes. We propose three mechanisms operating in public discourse that can account for gendered outcomes in peace and justice processes: deliberation, emboldening, and decentering.

Deliberation
Deliberation spotlights the content of speakers' contributions. Focused on how female and male speakers substantiate their views when they take the floor, scholars of democratic deliberation have studied the quality of speakers' arguments. A reason-giving requirement is at the core of democratic deliberation (Thompson 2008): speakers provide reasons for their positions and respond to reasons offered by others in an exercise of deliberative reciprocity (Gutmann and Thompson 1996). Deliberation also entails respect for interlocutors and openness to hearing their views. Steiner et al. (2005, 22) point out that respect requires empathy: "[t]he capacity and the willingness to put oneself in the shoes of others and to consider a situation from their perspective." Such other-regarding communication embodies the principle of reflexivity. Deliberators reflect on their positions, weighing them in the light of counterarguments (Bächtiger and Steiner 2005, 156). Deliberative communication plays an important role in the transition from conflict to peace. Deliberative virtues can help overcome mistrust and polarization in divided societies (Dryzek 2005; O'Flynn 2006; Steiner 2012; Caluwaerts and Deschouwer 2014). They can also promote justice-seeking by considering the perspectives of not only the ethnic Other, but also those of women. Recognition of women's concerns depends on their being an equal deliberative partner to men. Women's communication styles, including deliberation, can be understood from the prism of the difference/dominance debate (Cameron 1998, 14-15). The former centers on suitability of deliberation as a mode of communication in terms of women's ways of speaking, while the latter highlights structural underpinnings of male dominance of communication styles.
Difference democrats have pointed out that "[s]ome citizens are better than others at articulating their arguments in rational, reasonable terms" (Sanders 1997, 349). Recognition that the requirement for dispassionate argument in deliberation particularly disadvantages women has led to calls for valuing diverse models of communication in a democratic discussion, such as greeting, rhetoric, narratives, and storytelling (Sanders 1997;Young 2001). Directing attention to the gendered nature of institutions, scholars have posited that communication and language are implicated in the process of control (Mackay, Kenny, and Chappell 2010, 579-83). Male dominance in these gendered social structures is secured by legitimizing certain rules, norms, and behaviors. The absence of emotionality is prescribed in institutions (Acker 1990, 151), which restricts the range of permissible forms of articulation of needs and interests and impacts women adversely.
Addressing the question of "gendered deliberation" , emerging empirical research has not produced compelling evidence that the quality of deliberation in national parliaments differs between men and women (Bächtiger and Hangartner 2010). 6 Nonetheless, a deliberative perspective points to a possibility that women's substantive marginalization in peacemaking may be driven by different overall quality of argumentation between men and women.

Emboldening
Scholars have highlighted gendered differences in the use of language between women and men "in terms of both what they say and how they say it" (Krook 2010, 233). However, besides the quality of deliberation, a gendered pattern of speech also includes a structural dimension of public discourse: who takes the floor, when do they do it, and to what effect? Gender-specific features of discourse thus result from conversational interactions (Hannah and Murachver 2007, 275). Who follows whom may also matter: what if men are more likely to speak in succession than women?
Research in political science, social psychology, and communication has shown that women are interrupted more frequently by men than men are by women in legislative and nonlegislative settings, while specifying conditions under which interruptions occur (Winsky Mattei 1998;Mendelberg, Karpowitz, and Oliphant 2014) and whether they are hostile or not (Kathlene 1994). A gendered pattern of interruptions produces gendered consequences. Women are less successful than men at taking and holding the floor (Grob, Meyers, and Schuh 1997, 293), and their influence in the group is undermined (including women's perception of their own efficacy) (Mendelberg, Karpowitz, and Oliphant 2014, 29). The gendered pattern of interruptions underscores the importance of sustaining speaking opportunities in public debates. However, women do not necessarily have to be interrupted by men in order to be marginalized in mixed-sex debates. A gendered pattern of dominance may be sustained at the level of speaking sequences throughout the debate. A speech by a previous woman participant may embolden another woman to contribute.
Whether a woman speaker is more likely to be followed by another woman or a man indicates whether women are speaking in succession, thereby creating a deliberative space or an enclave conducive to women asserting their perspectives. Recognizing that women's perspectives are often marginalized in public fora, Sunstein (2007, 277) has argued that "a special advantage of enclave deliberation is that it promotes the development of positions that would otherwise be invisible, silenced, or squelched in general debate." Deliberative enclaves can "protect" (Mansbridge 1999, 63) the discourse of the disadvantaged and marginalized, ensuring greater equity and quality of deliberation (Karpowitz, Mendelberg, and Shaker 2012, 605). For these scholars, enclaves refer to opportunities when people from marginalized groups deliberate together, away from the advantaged and privileged. Alternatively, the structure of speaking turns can also restrict the deliberative space. It can pave the way for dominance in debates if speaking sequences are gendered and if men speak in longer sequences than women in mixed-sex debates. If this is the case, such a pattern across the whole debate can cumulatively limit women's substantive contributions and their influence on the outcome of the debate.

Decentering
Both scholars of political representation and transitional justice have found evidence that themes of contributions by men and women differ. Therefore, topics women address in public debates can also be an indicator of their influence or of the lack of it. Women and men talk about different issues, and these thematically gendered patterns persist in a range of settings: in face-to-face and virtual communication, e.g., in national parliaments, on a campaign trail, or on social media (La Cour Dabelko and Herrnson 1997;Carroll 2008;Krook 2010).
Besides diversifying policy and legislative agendas (Greene and O'Brien 2016), women speaking about women's issues enhances women's representation and influence (Bratton and Haynie 1999;Herrnson, Lay, and Stokes 2003;Pearson and Dancey 2011a). This includes legislation on gender-based violence, as illustrated by the toughening of sentences in the Egyptian parliament for performing female genital mutilation (Abdelgawad and Hassan 2019). However, in post-conflict settings, during proceedings in truth commissions and war crimes trials, women decenter, i.e., they are reluctant to talk about their own experience of conflict-related violence. Instead, when they talk about violence, women center their narrative on others: their husbands, partners, and children (Kashyap 2009;Crosby and Lykesy 2011;Theidon 2013;Yarwood 2013). This stands in contrast to women's public advocacy on women's issues, including SGBV. Studies of women's advocacy reveal the efficacy of frames, opportunity structures, and network dynamics (Berry 2018), but this is of limited analytic utility for understanding how women's influence is limited in mixed-sex, face-to-face public debates.
Investigating the topics women and men address can also indicate to what extent peacemaking outcomes, such as mandates of transitional justice instruments, are responsive to women's concerns. Considering that conflicts impact women differently than men-regardless of whether women talk about violence they suffered in debates addressing the criminal legacy, or they decenter-captures an important aspect of a likely broader pattern of thematic differences between women and men. This gendered thematic pattern can lead to a gendered outcome.
In sum, the biggest challenge for researchers of gender and language is to establish "why and where differences exist" (Hannah and Murachver 2007, 275). These differences matter; a gendered speech pattern reflects differences in power, status, and authority, which in turn determine speakers' influence on policy. Despite the growing scholarship focusing on women's speaking behavior, gender-based thematic and structural differences in speech patterns have not been studied together with the deliberative quality of women's contributions. The mechanisms we propose and test to account for why women's representation in a transitional justice process fails to produce a gender-responsive transitional justice instrument incorporates novel measures of the gendered nature of discourse: the deliberative quality of speeches and the sequential structure of turn-taking, alongside the thematic content of their speeches.

Research Design
To study women's representation without influence and to test the proposed mechanisms to explain gendered outcomes, we focus on the RECOM process in the post-conflict Balkans.

The Background
The RECOM grassroots civil society initiative advocates the creation of a regional fact-finding commission that would compile a list of all victims of the wars surrounding the dissolution of the former Yugoslavia through the 1990s and into the early 2000s. 7 With their narrow focus on perpetrators inherent in the pursuit of retributive justice, externally led efforts to promote justice in the Balkans through the operation of the ICTY from 1993 to 2017 had left a shared sense of elusive justice among victims of the Balkan wars on all sides (Orentlicher 2018). As a victimcentered transitional justice process, RECOM has provided a local restorative approach to the legacies of the Balkan wars. RECOM organized consultations with a wide range of stakeholders in 134 one-or two-day-long debates from 2006 to 2011. Like other human rights initiatives in the poor, post-conflict region, RECOM's activities were supported by foreign donations. Nonetheless, the agenda-setting for RECOM's consultations remained in the hands of local actors, who launched and drove this bottom-up transitional justice process (Rangelov and Teitel 2014). 8

Case Selection
Case selection refers to the event that the theory tries to explain as well as to the selection of countries, both of which require attention (Gerring and Cojocaru 2016, 408). With its lack of provisions that reflect women's concerns and ensure women's equal involvement in the commission's work, RECOM's draft Statute is a typical case of peacemaking with a gendered outcome. 9 Further, as a transitional justice process in the Balkans, RECOM is a response to the criminal legacy typical of civil wars fought along identity lines where sexual violence is a part of the overall repertoire of violence (Wood 2014, 461). 10 A typical case contributes to theory development by producing arguments that can explain some, but not all, cases (Toshkov 2016, 292). Its value is in "contingent generalizations that apply to the subclass of cases" similar to those that are studied (George and Bennett 2005, 32). Our insights about the patterns of discourse at the intersection of identity, gender, and wartime victimization can shed light on gendered peacemaking outcomes after other intrastate conflicts (Allansson, Melander, and Themnér 2017), and, specifically, those fought along the ethnic cleavage. Because religion does not frame peacebuilding efforts in the Balkans, 11 the findings are of limited applicability to understanding women's influence in the aftermath of intrastate conflicts in contexts where religious norms (which are not limited to a single religious group) shape women's participation in peacemaking. 12

Data
The textual data analyzed in this study is generated from transcripts of twenty consultations about the commission's draft Statute organized by the RECOM. 13 These transcripts are publicly available and comprise over 500,000 words. 14 In the text corpus, the gender and the role (discussant or moderator) of each speaker are coded. The order of speaking was determined by moderators, who responded to participants' requests to take the floor. Participants themselves were drawn from broad sections of civil society in the region. Because of the consultative nature of the RECOM's process, the organizers' priority was to make debates diverse and inclusive along different identity axes (Bonora 2019, 145): men and women, people from all ethnic groups involved in the Balkan conflicts, from different constituencies (victims, veterans, human rights activists, and professionals, such as lawyers, journalists, and teachers), and different generations. Their aim was to ensure a wide representation of different experiences of conflict and views on their appropriate redress by the regional commission, which would be codified in the draft Statute. Holding consultations both in rural and urban locations in all countries of the former Yugoslavia was an additional strategy to ensure the diversity of views.

Preprocessing
Preprocessing the text data involved a number of steps. Each speaker's speaking turn (i.e., speech) was tagged and, within each speaking turn, codes were assigned to speech acts, i.e., arguments (demands in the DQI terms) about any given issue under discussion, e.g., location of the commission's seat or selection of commissioners. This strategy enables us to capture the gendered nature of discourse by analyzing not only who spoke and for how long, but also how many arguments they made (about the articles of the draft Statute) and how well substantiated those arguments were. Lastly, 10 For an overview of violence during the Bosnian war, see Berry (2018, 116-29  as the original sessions were held in multiple languages (Albanian, Bosnian, Croatian, Macedonian, Montenegrin, Serbian, and Slovenian), the corpus was manually translated into Serbian. 15 We then applied a set of natural language processing tools developed for Balkan languages.

Methods
We combine quantitative content analysis, which relies on interpretative coding of text segments, to measure the quality of deliberation with a Discourse Quality Index (DQI), as well as computer-assisted quantitative text analysis to quantify the word frequencies in utterances at the speaker level. The combination of these two methods allows us to conduct a granular analysis of the content of speakers' contributions and of the frequency of participants' speeches and their speaking sequences, which could not be achieved by conducting only a computer-assisted text analysis of gendered speech patterns. 16 We fit Bayesian multilevel and structural topic models to estimate the effects of gender on speech behavior: the quality of deliberation, turn-taking, and thematic content. In what follows, we first reassess women's representation in a transitional justice process, with measures that have previously not been used by scholars of peacemaking.

Verifying the Puzzle of Women's Representation and Influence in Peacemaking
We have argued that a more robust measure of women's representation in peacemaking is needed before we can claim that their representation results in gendered outcomes that marginalize their interests and concerns. Consequently, we distinguish three levels of women's representation: (1) physical presence, (2) representation in taking the floor, and (3) participation in argumentation. The first level captures the turnout rate of women and men in the debates about the draft Statute; the second level reflects the proportion of participants who actually spoke during the debate as opposed to those who remained silent; and the third level accounts for those speakers who made an argument about the provision(s) of the draft Statute when they spoke. Table 1 presents average participation levels across twenty debates at different levels for discussants of both sexes. The number of participants varies between eighteen and seventy. 17 Of all the participants, 38 percent were female discussants, of whom 37 percent made at least one utterance (i.e., spoke as opposed to remaining silent), and of those who spoke, 74 percent made a statement pertaining to at least one of the articles of the draft Statute. While it is impossible 15 Details on the linguistic aspects of preprocessing carried out by Denisa Kostovicova with some research assistance are provided in the online appendix. 16   to entirely rule out underrepresentation of women driving a gendered outcome, other indicators of women's representation in speech, which have been overlooked in the existing literature, need to be considered. We proceed to measure the length of participation: instead of time it took for each individual to deliver a speech (Karpowitz, Mendelberg, and Shaker 2012), we use the number of words that each speech contained. To assess the association between a speaker's sex and speech length, measured by the logarithm of the number of word units (tokens), we use a random-intercept Bayesian multilevel model with consultations as level-2 units. Given the number of consultations included in our analysis, Bayesian estimation allows us to avoid potential problems with biased coefficient estimates and confidence interval coverage that can be encountered when the number of groups is small (Stegmueller 2013). Table 2 shows the estimated coefficients for speaker's sex and consultation-level predictors (ethnic diversity, level of debates, participating community, and whether translation was required, e.g., because of the ethnic composition of a debate). In addition to categorical descriptors of the consultations, we include the proportion of women present as another independent variable. Although the coefficient for female speakers is slightly negative (β female = −0.076), meaning that women's utterances, on average, tend to be shorter in terms of the number of words spoken, this relationship is not statistically significant.
Having checked women's representation at several levels, in addition to their physical presence, our analysis provides comprehensive assessment of their participation in the RECOM's debates; women made a vocal contribution both in terms of taking the floor and presenting arguments. Although it is somewhat lower than men's, women's (under)representation does not provide a convincing explanation for the absence of provisions reflecting women's needs and interests in the draft Statute of the regional fact-finding commission. We therefore examine empirically three mechanisms: deliberation, emboldening, and decentering, to assess whether there are any gendered patterns in discourse that could lead to a gendered outcome.

Measuring the Quality of Deliberation
The first potential mechanism to account for a gendered outcome is the quality of deliberation. To test this empirically, the measurement instrument-the DQI for the Study of Transitional Justice (DQITJ)-was constructed by adopting and adapting a technique for analyzing the quality of deliberation in a post-conflict context (Steenbergen et al. 2003;Steiner 2012). The DQI is a set of analytical constructs that jointly measure the quality of deliberation. Its construction responded to a need to supplement theorizing about deliberation with "empirical investigations of real-life deliberations" (Steiner et al. 2005, 43), originally applied to parliamentary debates. The dimensions of the DQI are underpinned by Habermas' notion of "communicative action" (Habermas 1984), which stipulates: "individuals give and criticize reasons for holding or rejecting particular validity claims, so that universally valid norms can be discovered through reason" (Steenbergen et al. 2003, 25). The DQITJ contains eight components: the first code captures the presence of (1) interruptions. 18 Habermas' level of justification of demands is denoted by (2) justification rationality, and content of justification is considered individually with reference to the (3) common good of a community, (4) specific subgroup, such as victims or young generations, or (5) abstract principles, such as peace, while respect is subdivided into two types: (6) respect toward participants and their arguments and (7) respect toward groups, and, lastly, (8) storytelling captures whether participants use stories alongside rational arguments.
The principal unit of analysis in the DQI coding strategy is a speech act, defined as "the public discourse by a particular individual delivered at a particular point in a debate" (Steenbergen et al. 2003, 27). The relevance of a speaker's utterance for coding is determined by whether it contains a demand, i.e., "a proposal on what decision should or should not be made" (Steenbergen et al. 2003, 27). A position on whether the commission's seat should be in Sarajevo illustrates a demand. The RECOM's text corpus consists of 1,211 speech acts uttered by discussants over twenty debates (excluding speech acts by moderators in line with the practice followed in the analysis of parliamentary debates, as well as in experimental studies). All speech acts were identified and manually coded according to the DQITJ codebook. 19 Each of the 1,211 speech acts was coded twice, independently, along all dimensions of deliberation. 20 Aggregation of multiple components of the DQI has received only limited attention in the literature. It is not uncommon to calculate a simple additive index by summing up the ordinal codes assigned to each speech act (Hangartner et al. 2007) or to consider each of them separately . Other researchers have applied principal component analysis to combine the items of DQI (Caluwaerts and Deschouwer 2014). Here we largely follow the approach of Gerber et al. (2018) and estimate a two-parameter Bayesian item-response theory (IRT) model to calculate an aggregate measure of the quality of deliberation. This modeling strategy is appealing for several reasons. Substantively, it assumes that the quality of deliberation is a unidimensional latent construct that is manifested through multiple items (DQI components), each of which has a difficulty and discrimination parameter. In the context of deliberation, the former can be viewed as how big of a political, social, or psychological challenge each of the items presents to a discussant. For example, while it could be a relatively easy task not to interrupt other participants, in the context of a transitional justice process, substantiating one's arguments with references to the shared abstract principles can be far more challenging.
At the same time, some objectives, such as delivering an argument that overcomes an ethnic interest, can 18 In the original work (Steenbergen et al. 2003), this component is labeled as participation, but here we use interruption to avoid confusion with a general term participation referring to a speaker's overall contribution to debates. 19 The codebook was developed by Denisa Kostovicova. 20 Coding was done by Denisa Kostovicova and a trained coder. The intercoder reliability statistics are available in the online appendix. All of them indicate an acceptable level of inter-coder agreement. differentiate better between a lower and higher quality speech act. This idea is captured by the second parameter of the model: discrimination. Apart from theoretical appeal, IRT offers a number of statistical advantages. First, it allows one to model the observed variables, derived from handcoded speech acts, as categorical variables, 21 without making assumptions that they are measured on an interval scale and possess an additivity property that allows them to be meaningfully summed up. Second, other types of modeling strategies such as factor analysis would require another implausible assumption of normally distributed error terms. 22 Figure 1 shows the response functions of each of the DQI components. Avoiding interruptions and showing respect toward other participants appear to be the easiest deliberation criteria to satisfy. These results are also a consequence of very few identified interruptions 23 and instances of open disrespect that we observe in the data. This is a noteworthy observation, given how demanding it is for people from different sides of an armed conflict to discuss the criminal legacy together. All components related to the content of justification as well as storytelling are the hardest principles of deliberation to meet in practice. The steep curves for justification rationality and respect toward other groups show that these two components can best discriminate between those speech acts the quality of which falls just below or just above their respective difficulty.
To estimate the association between gender and the quality of deliberation, we fit multilevel models that include both demand-level (model 2) and consultation-level (model 3) explanatory variables. Table 3 shows that speech acts delivered by women tend to have slightly higher quality of deliberation (β female = 0.031), although this relationship is not significant. The empirical assessment of the quality of deliberation in a civil society context does not support arguments that deliberation as a mode of communication disadvantages women. As such, it is consistent with the findings from parliamentary debates (Bächtiger and Hangartner 2010). In the case of a transitional justice process, these results suggest that the deliberative quality of women's speeches does not account for the content of the adopted draft Statute that does not respond to women's needs, interests, and concerns.

Gendered Structure of Debates: Emboldening
The focus of the previous mechanism was speech, considered in isolation. Speeches, however, do not occur in isolation; they typically constitute part of a larger in situ or ex situ conversation. It is plausible that gender becomes an important determinant not of how well substantiated one's arguments are, but of whether the arguments are voiced in the first place. As the results of the quality of deliberation indicate, there were very few direct interruptions in our corpus. Being a prominent feature of discourse that is easy to measure, interruptions are a focus of a considerable body of literature (Kathlene 1994;Winsky Mattei 1998;Mendelberg, Karpowitz, and Oliphant 2014). However, a debate can be structured in such a way that participants do not feel emboldened to speak in the first place. While a thorough analysis of this phenomenon would require looking at the underlying psychological processes, we are still able to study some observable implications of this mech-21 In our analysis we simplify the codes with more than two categories by dichotomizing them into binary variables, see the online appendix. 22 While the ensuing analysis is focused on this aggregation approach, we also demonstrate its close correspondence to other strategies in the online appendix. 23 Only about 4 percent of speech acts contain some form of interruption. anism from the transcripts of debates. Specifically, we look at the sequences in which men and women deliver a speech.
In the absence of any gendered dynamics, we would expect to find no differences in the number of speeches delivered in sequence by men and women. To test this mechanism, we fit a Poisson multilevel model with the number of speeches in a row delivered by male and female discussants as the dependent variable. Here we use hierarchical modeling to control for the percentage of female discussants, which varies at the consultation level. 24 While our approach is somewhat different from direct modeling of transition probabilities when treating speech sequences as Markov chains (Eggers and Spirling 2014), we adopt similar underlying assumptions and exclude moderators from analysis. 25 As table 4 demonstrates, contrary to our expectations, the gender of the speaker has a significant association with the number of speeches in sequence. The results show that sequences of speeches delivered by women are on average 38 percent shorter (β female = −0.48) than sequences delivered by men, while controlling for the per- 24 The number of groups is nineteen in this case rather than twenty, as one consultation contained only men, which prevents us from modeling speech sequences there. 25 As disentangling the mechanical effects of slight imbalance in participation ratios between speakers of different sexes from the genuine differences in the structure of debates is not straightforward, we provide further robustness checks and alternative modeling strategies in the online appendix. centage of female discussants and other consultation-level characteristics (model 2). Overall, the average length of women's sequences is 1.97 speeches, while that of men's is 3.22. This finding indicates that gendered dynamics of debates need not manifest itself in interruptions or other conspicuous demonstrations of power asymmetry. Rather, it can result from a subtler pattern of dominance that is sustained at the level of the sequence of speaking turns. The pattern of one woman speaking and emboldening another woman to speak, which is what we would expect if women were able to deliberate in enclaves, does not occur. This finding has substantive implications, because women's lines of argumentation cannot be given expression and be sustained in a mixed-sex setting. It can result in the absence of gender-responsive provisions in the draft Statute. In contrast, when women do speak in succession, their experience of conflict, for example as bereaved mothers of killed recruits, is articulated as a demand for the recognition of this loss in the commission's mandate. 26

Thematic Differences: Decentering
After considering the quality and sequences of female and male contributions, we shift our focus to the thematic content of the speeches by men and women. Following other scholars applying structural topic models (Terman 2017), we approach the analysis guided by theoretical expectations. The text-as-data approach (Grimmer and Stewart 2013;Benoit 2020)  civil society-led transitional justice debates. We use the text-as-data approach to augment the qualitative reading, manual coding, and statistical modeling of the structure of debates and apply structural topic models to our text corpus to estimate the differences in prevalence of different topics in male and female speeches. In order to prepare the corpus for analysis, we used a specially designed set of natural language processing tools developed as part of the Regional Language Development Initiative (ReLDI) for several Balkan languages (including the Serbian language that was used to standardize the multi-language text corpus used in this research) (Ljubešić et al. 2016), removed stopwords, and lemmatized the texts. 27 Structural topic model (Roberts et al. 2014) is an extension of classical topic models based on Latent Dirichlet Allocation, proposed by Blei, Ng, and Jordan (2003). Apart from the estimation of topic proportions for each individual document, they allow one to incorporate meta information and estimate how additional covariates affect topic prevalence. Rather than fitting a model on individual utterances, we aggregate them at the speaker level. Through an iterative procedure, we find that a ten-topic structural topic model produces the best balance between statistical fit and substantive interpretation. 28 As with other analyses presented above, we include only the speeches delivered by discussants and not those delivered by moderators. Figure 2 displays the most prevalent terms in each topic. 29 Figure 3 shows the estimated effect of gender on the proportion of different topics and indicates a gendered thematic pattern of discourse. We find that the speaker's sex significantly affects the prevalence of five topics out of ten. The topics "modality" and "evaluation" are more prevalent in women's speeches, while the topics "Kosovo," "implementation," and "outcome" are more prominent among male speakers.
The topic "evaluation" indicates that women scrutinize the proposed articles of the draft Statute from the perspective of criminal justice. They are concerned, for example, that the draft Statute might give the commission quasi-legal powers akin to those that a court exercises in criminal proceedings, 30 such as providing for a criminal sanction for nonappearance of individuals summoned to testify before the commission. We also find evidence that women center their remarks away from their interests, as illustrated by the topic "modality." They advocate that the disabled, people from all age groups, and high(ly) educated individuals be included in the commission and warn against arbitrary exclusion of people on ideological grounds, thus putting others' interests before their own. The coefficient on the topic "acknowledgement," albeit not statistically significant, points in the same direction. It demonstrates women's preoccupation with the missing male family members, while remaining silent about harms experienced by women when discussing truth about wartime violence. By contrast, men focus on practical issues involved in the operationalization of the fact-finding commission. This is captured by the topic "implementation" focused on the draft statute and its 27 More details on preprocessing are available in the online appendix. 28 The online appendix contains additional information on topic diagnostics and alternative specifications. In addition to gender, we include country, level (regional/non-regional), and type of participating community (general/professional/victims) as covariates. 29 We leave topic 10 unlabeled as it largely represents a combination of the other nine topics. 30 Italics mark words with high probability of occurring within specified topics. Speeches with highest scores on significant topics are included in the online appendix.

Figure 2. Top ten words by topic.
Note: English translation from Serbian original was made after fitting STM model. Size of terms is proportional to the probability of being generated by a given topic. article(s), underpinned by their concern about legitimacy of the implementation process and feasibility of the adoption of an interstate document. Similarly, as indicated by the topic "outcome," men are preoccupied with fulfillment of the aim of the RECOM commission focused on producing facts about crimes in the context of the conflict in which they were committed. Men are also associated with the topic dedicated to documenting war crimes in Kosovo. 31 The analysis reveals a broad pattern of thematic differences between women and men, as speakers of both sexes contribute to shaping important aspects of the draft Statute. Women approach the Statute deliberations primarily concerned with preserving the distinct nature of a restorative justice process and clarifying its relationship with criminal justice systems in the region, whereas men focus on, what 31 This, however, could be an artifact of the few women present at some of the meetings that took place in Kosovo. they refer to as, "technical" and political aspects of the operationalization of the commission. 32 Further, we find decentering in women's speeches not only away from their needs rooted in their experience of conflict, but also away from their interests in equal representation. This gendered thematic pattern can be related to the gendered character of the draft Statute that does not outline procedures on how to address SGBV nor codify women's equal participation in all facets of the commission's work.

Conclusions
Peacemaking outcomes that are not responsive to women's concerns are at the heart of the reproduction of gender injustice and the persistent elusiveness of "quality peace" (Wallensteen 2015) for women. Inequalities persist even after the WPS Agenda spurred women's peace activism, women's demands for access to peacemaking, and women's articulation of their particular needs (Shepherd 2017). In the area of post-conflict transitional justice, women's advocacy has led to the codification of accountability for wartime SGBV in international law. These developments have in turn had an impact on public perceptions and policy responses to this issue in post-conflict zones (Warren et al. 2017). The burning question now is why, despite such "norm augmentation" in the post-Cold War period, do we still see "old specters of unseen hierarchies operating to the detriment of addressing harms experienced by women?" (Ní Aoláin 2014, 625).
This study of women's speaking behavior provides a new perspective on gendered dynamics of peacemaking by identifying a novel axis of male domination at the microlevel of public discourse. We know that fewer women than men are likely to be at the negotiating table, despite 32 See speeches in the online appendix. recent progress in narrowing the gap between men's and women's attendance. Nonetheless, an important part of the puzzle of women's limited substantive representation has been overlooked, given our still weak understanding of what happens when women engage in the exchange of arguments with men on the other side of the table.
Feminist scholars have noted that the "add women and stir" solution has done little to advance gender-just peace (Ní Aoláin 2016, 158). Going beyond the issue of representation, our research reveals that the patterns of men's domination during a public discussion are subtle but nonetheless consequential-even when they are not expressed in obvious forms such as interruptions. Women's relative underrepresentation in peacemaking continues to be an issue that ought to be addressed. However, the focus merely on (numerical) underrepresentation neglects the question how women's voices matter in peacemaking. As we have shown in this study of post-conflict justice debates, not only do women take the floor almost as often as men, but there is no substantial difference between men and women in terms of the quality of deliberation. While the content analysis reveals that men and women address different topics, we propose that the gendered structure of turntaking is key to women's limited influence on outcomes in peacemaking. Women's speech is restricted during debates; women speak in shorter sequences than men, which shrinks women's deliberative space to develop their argumentation, including on those issues and topics that would better reflect a whole range of women's concerns and demands for equality, even if they wish to remain silent about SGBV. 33 This lacuna in the scholarship that concerns a discursive dimension of a peacemaking process is also linked to the issue of data (Anderlini 2007) and methods used to study how peacemaking is gendered. Efforts have centered on counting "bodies at the table" or provisions of peace agreements that refer to women, which have become standard indicators of women's inequality and gendered peace that disadvantages women (Paffenholz et al. 2016). 34 Our empirical analysis of gendered speech patterns in a transitional justice process contributes to efforts to quantify gendered dynamics during peacemaking and sheds new light on constraints on women's influence.
While this research furthers the study of how conflictresolution and peacemaking are gendered (David, Rosler, and Maoz 2018), it also provides new insights for scholars of political representation and communication interested in the study of male domination in political communication. Our investigation of women's speaking behavior in civil society debates confirms the value of increasing the number of comparisons by expanding the "sites" of political representation (Krook 2010;Parthasarathy, Rao, and Palaniswamy 2019), which are usually restricted to institutional settings, such as parliaments. We elucidate women's participation in a parallel nonstate, civil society process that also marginalizes women.
Exposing men's dominance at the level of turn-taking, this study provides another possible solution toward greater substantial gender equality for women in peacemaking and in politics more generally, beyond the imperative of equal representation and access that have preoccupied scholars and practitioners. Observed at the microlevel of discourse, our findings point to the need for greater attention to 33 On silence as a site of power and agency, see Mannergren Selimovic (2020). 34 New datasets (Bell and Badanjak 2019) and definitions of inclusion (Arthur 2016) will refine explanations in this vein. the management of speaking turns during public debates. Extant research has pointed to the benefits of recognizing the marginalized and their views through enclave deliberation conceptualized and implemented as deliberation in separate groups made up of those who are disadvantaged and whose perspectives are sidelined in mainstream debates (Mansbridge 1999;Sunstein 2007;Karpowitz, Mendelberg, and Shaker 2012). This research provides a novel insight that the benefits of enclave deliberation can also potentially be gained from turn-taking sequences nested in the debates in mixed-sex groups and fora. Ultimately, we recognize that the pattern we found might be only one aspect of a number of such microlevel instances of lopsidedness that can lead to gendered outcomes. Whether this is so remains to be established by future research aided by the interdisciplinary study of the microlevel of discourse, which stands to reveal new insights into the nature and consequences of communicative interactions between women and men during post-conflict peacemaking.