Substantive Representation: Provisions for Women (Dependent Variable)

Women’s substantive representation is often measured by the inclusion of women-specific policy, which is seen as a clear indication of advocacy. For example, electoral politics researchers measure women’s substantive representation by women’s promotion of reproductive health, family leave, or childcare policy (see Htun and Jones Reference Htun, Jones, Craske and Molyneux2002; Norton Reference Norton1999; Swers Reference Swers1998; Thomas Reference Thomas1991; Vega and Firestone Reference Vega and Firestone1995). Similarly, WPS policy often defines provisions for women as pre-designated issues such as condemning sexual violence (see UN Security Council 2008). However, women-specific issues are often integrated into deliberations that are not exclusively about women. Examples of integrated women-specific policy include ceasefire commitments including gender-based violence as a recognized violation (see Burundi 2000), constitutional reforms with parliamentary quotas for women (see Central African Republic 2013), and economic and social development plans indicating funding requirements for women’s civil society groups (see Iraq 2005) (see Supplementary Appendix A.1 for a qualitative summary of provisions). We may fail to see these inclusions as provisions for women if considering only explicit references to pre-designated policy issues.

To overcome this weakness, I define provisions for women as explicit reference to women and/or girls or reference to gender-based violence. I leverage PA-X data by Bell et al. (Reference Bell, Badanjak, Beaujouan, Forster, Epple, Jamar and McNicholl2021) to apply a binary text-as-data approach that considers any reference to women, girls, widows, mothers, gender-based, or sexual-based violence as provisions for women. However, there are limitations to a binary text-as-data approach.

A binary approach restricts insight into the strength of provisions and obscures variation between provisions; agreements with a singular reference to women are weighted equal to agreements with comprehensive implementation plans for multiple provisions for women. Second, limiting the definition of “provisions for women” to policies that explicitly reference women, girls, and forms of gender-based violence omits gender-neutral policies that serve to disproportionally advance women’s equality, such as access to education (Rihani Reference Rihani2006) or property rights (Brulé Reference Brulé2020). This oversight is troubling given that women delegates may strategically advocate for women-specific issues without using gendered language to overcome barriers rooted in sexism.

Despite these limitations, I maintain a binary text-as-data definition for substantive representation and opt to address critiques through additional analysis. This is because the individual issues most relevant to women in Guatemala in 1996 (e.g., investment in social and economic growth) differ from the issues most important to women in Darfur in 2006 (e.g., ceasefire and security arrangements). While case-study analysis enables researchers to capture women’s advocacy for specific issues in a specific country during a specific negotiation, I would be remiss to identify any issue as “women-specific” in all 116 cases, spanning 54 countries and three decades. A binary text-as-data measurement of substantive representation offers a more equitable approach that encompasses what local women consider to be “women-specific,” rather than assigning specific issues as gendered based on my own biases. This approach aligns with Htun and Weldon’s call to “disaggregate women’s rights” to better assess the variation of women’s substantive representation over time and space (Htun and Weldon Reference Htun and Weldon2018, 3). In sum, any measurable conceptualization of provisions for women has consequences. A binary text-as-data approach casts the largest possible net for substantive representation, which is advantageous for a transnational study.

Peace processes are comprised of multiple agreements that can span decades. Consequently, provisions for women can be included or excluded along the way, complicating how researchers quantify representation. I use comprehensive peace agreements as my unit of analysis because they can be seen as the final step in negotiations, which offers a consistent snapshot in time across diverse peace processes and eliminates redundancy stemming from a single conflict with multiple agreements. As defined by PA-X, comprehensive agreements concern parties that are “engaged in discussion and agreeing to substantive issues to resolve the conflict and appear to be set out as a comprehensive attempt to resolve the conflict” (Bell et al. Reference Bell, Badanjak, Beaujouan, Forster, Epple, Jamar and McNicholl2021, 7). Further, comprehensive agreements are the last opportunity for delegates to include provisions of interest. It cannot be deemed strategic for women delegates to withhold advocacy for provisions for women during the comprehensive stage because the formal negotiation period ends with the signing of comprehensive peace agreements. In other words, comprehensive agreements close the window for addressing gender inequality, which is often widest during peace negotiations (Anderson Reference Anderson2016). Within the specific scope of peace negotiations, comprehensive peace agreements offer the broadest lens to analyze women’s substantive representation during peace processes.

Descriptive Representation: Women’s Involvement (Independent Variable)

The WPS sector defines women’s involvement in peace processes along three separate tracks: Track 1, Track 2, and Track 1.5. Track 1 refers to official, formalized peace diplomacy between high-level officials representing warring parties. Track 1 involvement is often understood to be the “primary peacemaking tool of a state’s foreign policy” (Mapendere Reference Mapendere2000, 67). This is distinguishable from Track 2, which includes unofficial dialogues, often involving civil society actors or public protests. Track 2 is a widely used trajectory for women to shape agreement outcomes. For example, during the 2003 Accra Peace talks, Liberian women-led protestors staged an effective sit-in, barricading meeting room doors to physically demand a settlement. Although a singular woman was involved in Liberia’s 2003 peace negotiation as a Track 1 observer to the African Union (Adwoa Coleman), women had an irrefutable influence over negotiation outcomes via Track 2 involvement.

As the name suggests, Track 1.5 is a hybrid between Track 1 and 2 involvement. Track 1.5 is defined as informal talks where both formal and informal delegates act in their “personal capacity” (Chang et al. Reference Chang, Alam, Warren, Bhatia and Tukington2015, 14; see also Bercovitch and Jackson Reference Bercovitch and Jackson2009; Mapendere Reference Mapendere2000). WPS promotes Track 1.5 involvement through parallel women-specific committees as a means of facilitating women’s engagement pending women’s prohibition from official Track 1 negotiations (e.g., the Syrian Women’s Advisory Board) (UN Women 2018). However, the power of women’s committee members is context-dependent since recommendations made by women’s committees are at the discretion of formal Track 1 delegates. Parallel but separate women’s committees have the potential to enhance influence by offering women an opportunity to coordinate and advance interests as a unified bloc (Paffenholz et al. Reference Paffenholz, Nick, Dixon, Shluchter and True2016). Alternatively, women’s committees may simply pay lip service to international stakeholders and women’s proposals are ignored or dismissed by Track 1 delegates.

I limit my analysis to Track 1 involvement because it is the most constrained environment within which women operate and is therefore the greatest test for the theory that power influences women’s ability to advocate for provisions for women. Focusing on power distribution, I find just as much variation within Track 1 as between Tracks. Women may be in Track 1 positions of power, acting as signatories or negotiators, or they may be present as silent observers with limited to no opportunities to speak. Despite the difference in power between signatories and silent observers, most research considers women to be involved in peace negotiations if a singular woman fills either role. Consequently, women’s involvement can only be defined by what it is not—an absence of women.

True and Riveros-Morales (Reference True and Riveros-Morales2019) weigh silent observers and lead negotiators equally. Paffenholz et al. (Reference Paffenholz, Nick, Dixon, Shluchter and True2016) distinguish types of involvement but do not highlight differences in direct involvement based on positions held by women delegates. Krause, Krause, and Bränfors (Reference Krause, Krause and Bränfors2018) exclusively account for local signatories, omitting alternative forms of involvement altogether. This difference in conceptualizing “involvement” likely accounts for contradicting findings between studies, but it also points to a larger methodological issue within the WPS sector. The difference between lead negotiators and silent observers is not a difference in degree, but a difference in kind. Measuring involvement through a power lens allows more explicit research on women’s role in formal peace negotiations and facilitates a better understanding of how women influence final agreement outcomes.

Measuring Power: Positions Held by Women Delegates

I use positions held by Track 1 delegates as a proxy for power.Footnote ³ I classify all Track 1 peace negotiation delegates mentioned in comprehensive agreements by the following positions: signatory, negotiator, mediator, advisor, observer, logistical support, women’s committee member, or unknown. These eight categories encompass the vast majority of positions involved in comprehensive peace negotiations from 1990 to 2021 and expand on current peace negotiation data (Bigio et al. Reference Bigio, Vogelstein, Bro and Connell2021).

Signatories have the greatest capacity to alter agreement outcomes by demanding amendments before signing or simply refusing to sign documents.Footnote ⁴ While it is feasible that parties may replace a signatory who refuses to sign an agreement with someone willing, this would not impact research findings since a replaced signatory will no longer be coded as a signatory. Women may perceive their role as a signatory to be more precarious given a male majority. Consequently, women signatories may be less likely to voice disagreement or request amendments for fear of being removed from the process or hindering women’s future involvement. Further, signatories may be a “symbolic role,” allowing women to sign an agreement without allowing them to participate in the negotiation (UN Women 2015, 15). Despite social and patriarchal pressures that may alter women delegates’ willingness to dissent, any signatory has the technical ability to alter the outcome. Even in a symbolic role, women signatories have the capacity to ask for an amendment before signing. It is for this reason that Krause, Krause, and Bränfors argue that any signatory can be “assumed to have directly participated in negotiations, with voice and influence” (Reference Krause, Krause and Bränfors2018, 987). Further, these critiques are relevant to women in any delegate position. Therefore, women signatories have the greatest relative power to women in other positions.

Negotiators and mediators have the capacity to influence agenda items and push for specific provisions or amendments. While negotiators represent a specific interest group and strive to obtain stakeholders’ desired outcomes, mediators are third-party members (rarely belonging to the conflict parties) that strive for consensus (Buchanan et al. Reference Buchanan, Cooper, Griggers, Low, Manchanda, Peters and Prentice2012; O’Reilly, Súilleabháin, and Paffenholz Reference O’Reilly, Súilleabháin and Paffenholz2015; Paffenholz et al. Reference Paffenholz, Nick, Dixon, Shluchter and True2016). Therefore, negotiators have greater relative power and ability to advocate for women-specific provisions than mediators. Advisors are in a position to guide signatories, negotiations, or mediators. While advisers may have the greatest technical knowledge, they have relatively less autonomy and power than signatories, negotiators, or mediators.

Observers are limited to overseeing the peace process. Observers serve an important role by enhancing the legitimacy of negotiations, increasing local or international buy-in to peace, and may sway outcomes by acting as a physical reminder of certain interest groups or informing grassroots protestors (Paffenholz Reference Paffenholz2014). However, observers are generally deemed “powerless” because they are not in a position to negotiate and are rarely able to make demands (Paffenholz Reference Paffenholz2014, 79). This is also true for those in positions of logistical support. While logistical support staff are necessary for peace negotiations to function, they have little power over agreement outcomes and are rarely documented in final agreements.

Lastly, including women’s committee members (a Track 1.5 form of involvement) as a distinct category is controversial given the Track 1 scope of analysis. In developing PADD, I include all individuals listed in comprehensive peace agreements. Only two cases within the 116-case dataset documented the presence of parallel women’s committee members (Somalia, 2007 and Nigeria, 2016). While a sample size of two is not sufficient to quantitatively determine women’s committee members’ relative power, I distinguish women’s committee members as a distinct entity within PADD because it signifies an emerging WPS trend (UN Women 2018).

PADD Composition

This research uses Peace Agreements (PA-X) Database and Dataset (Bell et al. Reference Bell, Badanjak, Beaujouan, Forster, Epple, Jamar and McNicholl2021) coding to determine which agreements are comprehensive. I extract all comprehensive peace agreements finalized between 1990 and April 2021 for 116 observations across 54 countries (see Supplementary Appendix A.2 for a regional distribution). I do not limit my analysis to peace negotiations that involve the United Nations because this restricts samples to negotiations between parties that invited the International Organization to mediate or observe. The UN Women Report, “Women’s Participation in Peace Negotiations,” focuses on 31 peace processes (Diaz and Tordjman Reference Diaz and Tordjman2012), while PADD captures data on 74 comprehensive agreements finalized during the same study period (1992–2011).

PA-X lists the signatories of agreements but does not code for gender. This information is not available from other datasets including the UN Peacemaker Peace Agreements Database, the UCDP dataset, or the Transitional Justice Institute Peace Agreement Database. While Krause, Krause, and Bränfors (Reference Krause, Krause and Bränfors2018) develop a dataset that codes for gender, it is limited to local signatories between 1989 and 2011 and therefore does not account for the full range of positions held by Track 1 delegates. I turn to PA-X and the UN Peacemaker Peace Agreements Database to access original and translated peace agreement texts. I code for all individuals mentioned in the agreement.Footnote ⁵ I separate given (first and middle) names from family (last) names using local naming practices. I determine the gender of given names using Gender API software which uses country-specific naming data to associate a name with a gender and provides a confidence score. For example, the name “Ali,” identified as male, has a 61% confidence score in the United States given the unisex application of the name but has a 98% confidence score in Jordan. I manually code for all names with a confidence score of 75% or below, as well as any discrepancies between the gender associated with first and middle names. I additionally use a randomized number generator to manually code for 30 names with a confidence score above 75%. This randomized check did not find any false coding.

In the event manual coding is inconclusive when finding a specific delegate, I turn to social media profiles, coding the delegate as a woman or not depending on social media profiles associated with the given name in question. In the event a name remains androgynous after looking at Facebook, Twitter, LinkedIn, employee profiles, and news reports, I rely on the Gender API software findings, even if the confidence score is below 75%. In the event a manual check and the Gender API software are inconclusive, the individual is coded as “NA.” This methodology may misgender an individual who does not follow traditional naming structure, such as a Jordanian woman named Ali. However, this coding error would be equally distributed across the 2,299 individual delegates involved in the 116 comprehensive peace negotiations, minimizing the impact of measurement error.

A more thoroughgoing shortcoming of this research is that it mistakenly views gender as binary, which does not represent the author’s view of gender and fails to account for the plurality and diversity of gender identities and expressions across socio-cultural and historical contexts. Despite non-binary pronouns only recently gaining greater traction in Western popular and academic discourses, a rich array of nuanced gender concepts and a spectrum of gender identities have long been recognized in a wide variety of indigenous contexts. While coding gender along a binary—a delegate is labeled a woman (1), or not (0)—risks misgendering delegates, this is also likely to be equally distributed across the dataset and is thereby unlikely to skew findings. Regardless, this theoretical limitation signals the need for greater work in this area to attend to the complex interplay between gender identity and gender performance.

The position held by individual delegates is recorded from original and translated peace agreements published by PA-X and UN databases and is categorized by position, for example, signatory; negotiator; mediator; or observer. In the event an individual holds more than one position, I defer to their highest position of power. For example, if an individual is listed as a signatory and observer, they are coded as a signatory.Footnote ⁶ Although this increases observations for positions of greater power relative to positions of lesser power, I do so because I am interested in how power influences the relationship between descriptive and substantive representation. Therefore, while an individual may be formally recognized as a signatory and observer, they are more likely to advocate for provisions for women through their signatory role. Further, signatories represent a range of different political actors including members of armed groups, government representatives, and delegates from civil society groups, subsequently minimizing issues of endogeneity (Krause, Krause, and Bränfors Reference Krause, Krause and Bränfors2018).

Controlling for Women in Conflict

To account for alternative hypotheses, I control for numerous variables that may explain the relationship between women’s descriptive and substantive representation in comprehensive peace negotiations (see Supplementary Appendix A.5 for a full list of controls and data sources). I control for women’s involvement in conflict as both combatants and as survivor-victims of wartime sexual violence. Women’s exclusion from peace negotiations is often linked to women’s military absence (Aharoni Reference Aharoni2011). Israel’s deputy foreign minister defended women’s exclusion in the Oslo Process (1993–2000) by saying “… women don’t reach high military ranks. And since the military plays such a vital role not only in war but also in the peace process, women have less to say in this process” (Aharoni Reference Aharoni2011, 403). Similarly, Kampwirth (Reference Kampwirth2002) finds Central American women gained greater prestige after participating in combat. Women’s participation as combatants may increase the likelihood that women are involved in peace processes (Brannon and Best Reference Brannon and Best2022). Thomas (Reference Thomas2024) finds gendered provisions in peace processes are more likely when women serve as rebel combatants. To account for this alternative explanation, I control for women’s involvement as armed combatants using the Women in Armed Rebellion Dataset (WARD) (Wood and Thomas Reference Wood and Thomas2017).Footnote ⁷

The international community widely advocates for women’s inclusion in peace negotiations based on conflict-related sexual violence (see UN Security Council 2008; 2019a). This corresponds to the finding that women are more likely to be included in peace negotiations by eliciting sympathy (Brannon and Best Reference Brannon and Best2022). High levels of conflict-related sexual violence may provoke a strong response from stakeholders and increase women’s involvement in peace negotiations as a result. I control for rates of sexual violence using the Sexual Violence in Armed Conflict (SVAC) Dataset (Cohen, Nordås, and Nagel Reference Cohen, Nordås and Nagel2021). I calculate the average rate of documented sexual violence for all conflict years preceding the signing of the comprehensive peace agreement.

Controlling for International Involvement

International commitment to women’s involvement in peace and security formalized with the ratification of UNSCR 1325 in 2000 and has been reiterated in nine other related resolutions between 2008 and 2019. Given the international community’s focus on WPS, gender provisions are more likely when the UN acts as a signatory to a peace agreement (Bell and O’Rourke Reference Bell and O’Rourke2010). I account for international involvement through the following proxies: the number of UN Security Council Resolutions issued about the conflict up to the date the peace agreement is signed; if the UN is a signatory to the agreement; if any other international actor, state representative, or international organization personnel, such as NATO or the African Union, is a signatory to the agreement; and if the country has a 1325 National Action Plan at the time of signing the comprehensive agreement. National Action Plans (NAPs) are country-specific strategies to “secure the human rights of women and girls in conflict settings” (WILPF 2022).

Controlling for Gender Equality

Countries with higher levels of gender equality may be more likely to include women in peace negotiations and may be more likely to include provisions for women in final agreements. This is because women’s rights are more likely to be codified through laws and norms and women are more likely to access high-level institutional opportunities in places with greater gender equality. Further, women with higher economic, social, and cultural capital, are more likely to gain power at the household level, which filters to the community and national level (Domingo et al. Reference Domingo, Holmes, O’Neil, Jones, Bird, Larson and Presler-Marshall2015). Therefore, are women delegates influencing peace agreement outcomes to contain provisions for women, or are provisions for women and women’s involvement outcomes stemming from societal norms?

To answer this question, I control for gender equality using the Gender Development Index (GDI) issued by the United Nations Development Programme because it has the most robust data relative to other gender indexes. I exclude the Gender Inequality Index (GII) and the Global Gender Gap Index from regression analysis due to collinearity.Footnote ⁸ These three indexes use different variables and proxies to calculate gender equality at the country level (see Supplementary Appendix A.5). I independently control for the percentage of secondary pupils who are girls and rates of teen pregnancy using World Bank data.

Overcoming Missingness: Multiple Imputation

Nine agreements (7.8% of cases) had missing delegate data, that is, the individuals involved could not be determined through publicly accessible peace agreements. This rate of missingness is not disruptive to analysis. However, control data vary widely across the 116 cases due to conflict. For example, press coverage contains no missing values; girls’ access to secondary education is missing in 9% of cases; the Gender Development Index omits 40% of cases; and the public perception of women is missing in 72% of cases. To overcome missingness, I implement multiple imputations.

Multiple imputation replaces missing values with estimated values derived from the distribution of, and relationship among, observed variables (Li, Stuart, and Allison Reference Li, Stuart and Allison2015; Rubin Reference Rubin and van Buuren2018). Specifically, I leverage predictive mean matching imputation (pmm). Drawing on available data, the model calculates values closest to the predicted value for the missing entry. This method predicts the distribution of missing data to align with the distribution of observed data (see Supplementary Appendix A.8 for examples). I create five imputed datasets to ensure sufficiency (Allison Reference Allison2000). In addition to excluding my independent and dependent variable from imputation, I exclude variables with overwhelming proportions of missing data,Footnote ⁹ non-control variables that have perfect collinearity,Footnote ¹⁰ and variables that are irrelevant to prediction.Footnote ¹¹

Linking Descriptive and Substantive Representation

I find a significant positive correlation between women’s involvement in peace negotiations and the inclusion of provisions for women in comprehensive peace agreements, supporting the claim that women’s descriptive representation (women in the room) increases substantive representation (provisions for women). A total of 72% of agreements that involve women delegates include provisions for women, while only 48% of agreements that do not involve women delegates include provisions for women (Table 1).

Table 1. Descriptive Statistics

A positive relationship holds when measuring women’s involvement as a binary and when accounting for the percentage of women delegates in a negotiation (see Supplementary Appendix A.11, model 1 for binary and Table 2 for percentage). The percentage of women delegates captures variation in the total number of delegates (both men and women) involved in comprehensive peace agreements. For example, Nigeria’s 2016 agreement involved a total of 235 delegates while Nepal’s 2006 agreement involved 2. On average, women comprise just 3.9% of delegates to negotiations. Despite the small percentage of women delegates, their influence is significant (Table 2). Models 1–5 (Table 2) indicate that a 1% increase in woman delegates nearly doubles the likelihood an agreement will contain provisions for women. Low statistical power may explain insignificant findings in model 6. Note that agreements referencing women, girls, widows, mothers, gender-based, or sexual-based violence do not necessarily equate to beneficial policy or legal protections for women. A binary measurement of “provisions for women” means women’s inclusion may increase the likelihood women are simply referenced within the agreement. While references are important (Cohn Reference Cohn, Rai and Waylen2008; Reid Reference Reid2021), they are not a silver bullet for gender equality.

Table 2. Women Delegates and Agreements Containing Provisions for Women (OLS Regression)

Note: Analysis omits agreements that negatively reference women (Adadda Agreement 2007). See Supplementary Appendix A.12 for analysis including the omitted case. See Supplementary Appendix A.13 for lagged Gender Development Index variables for 1 and 5 years before the year the agreement was signed. ***p < 0.001; **p < 0.01; *p < 0.05.

Robustness Checks for a Text-as-Data Approach

Measuring provisions for women (substantive representation) using a binary text-as-data approach omits provisions for women that do not explicitly reference women, girls, widows, mothers, gender-based, or sexual-based violence. To address this, I leverage PA-X data (Bell et al. Reference Bell, Badanjak, Beaujouan, Forster, Epple, Jamar and McNicholl2021) to assess the relationship between women delegates and provisions for narrow interests understood to disproportionally benefit women in patriarchal societies (Table 3).

Table 3. Relationship between Women’s Inclusion in Peace Negotiations and Provisions from Which Women Would Disproportionally Benefit

Note: “Provisions for children” are defined as “comprehensive mechanisms or commitment to children or youth” (Bell et al. Reference Bell, Badanjak, Beaujouan, Forster, Epple, Jamar and McNicholl2021, 14). See Supplementary Appendix A.14a–e for GLM regression tables. **p-value at the 0.01 level.

I find an insignificant relationship between women’s involvement in peace processes and most provisions that disproportionally benefit women in contexts with high rates of gender inequality. For example, women and girls likely disproportionally benefit from access to education even if the provision does not specify girls’ access to education. However, I find women delegates across the 116 agreements are not significantly correlated to agreements containing provisions for education. This suggests high variation in the types of issues women prioritize across the 116 agreements. A text-as-data approach to measuring women’s substantive representation accounts for this variation by coding explicit references to women, girls, or gendered violence.

However, a text-as-data approach risks coding agreements as containing provisions for women if agreements reference women, girls, or gendered violence in ways that disadvantage women. To ensure provisions for women advance gender equality, I read all agreements containing provisions for women. I found a singular case (Adadda Agreement 2007), where women did not stand to benefit from provisions. The agreement stipulates that “one godobtir girl” will be given in exchange for one deceased person from the Bah Ararsame, suggesting forced marriage. Omitting this agreement from analysis does not alter the significance of models 1–5, but does result in insignificance for model 6 (Table 2).Footnote ¹⁴ Given the marginal difference in standard error and p-values between Table 2, model 6 (SD 0.95, p-value 0.059) and Supplementary Appendix A.12, model 6 (SD 0.78, p-value 0.013), insignificance does not deviate from findings in the initial analysis and likely stems from a reduced sample size (Gelman and Stern Reference Gelman and Stern2006).

More than Numbers: Requiring Seats with Power

Findings support the theory that women in positions of power are positively correlated to provisions for women—a one percentage point increase in the presence of women signatories increases the likelihood an agreement contains provisions for women by 2.1 percentage points (Table 4, model 1). A GLM model confirms this relationship (Table 4, model 1); the average marginal effect of the presence of women signatories on the likelihood agreements contain provisions for women is 31.3 percentage points. While OLS regression finds a one-unit increase in the presence of women negotiators increases the likelihood an agreement contains provisions for women by 2.4 percentage points (Table 4, model 2), the sample size (n = 18) limits the persuasiveness of this finding. There is no significant relationship between women negotiators and provisions for women using GLM regression (Table 5, model 2). There is no statistically significant relationship between women mediators or women observers and provisions for women (Table 4 or Table 5, models 3 and 4).

Table 4. Percentage Measurement of Women Delegates (OLS Regression)

Note: Policy implications are not derived from OLS models given the small number of observations for women negotiators, mediators, and observers. *p < 0.1; **p < 0.5; ***p < 0.01.

Table 5. Binary Measurement of Women Delegates (GLM Regression)

*p < 0.1; **p < 0.5; ***p < 0.01.

Using OLS regression to capture the relationship between the percentage of women in various roles and provisions for women works well for women signatories (Table 4, model 1). This is because I observe women signatories in 103 cases. However, OLS regression is less persuasive for women in other roles (Table 4, models 2–4). Only 18 agreements indicate the involvement of women negotiators (Table 4, model 2), 48 mention women mediators (Table 4, model 3), and 55 mention women observers (Table 4, model 4). This does not mean that only 18 agreements involved women negotiators, but rather 18 comprehensive agreements referenced women negotiators who exclusively acted as negotiators (i.e., were not a negotiator and signatory).Footnote ¹⁵ To increase the sample, I use a binary measurement; (1) for agreements that reference women negotiators, (0) for all others (Table 5). I do not use multiple imputations because missingness involves the independent variable.

As a robustness check, I categorize all women delegates as either “signatory” or “other” (Supplementary Appendix A.11). Using GLM regression, I find the average marginal effect of a woman delegate in any role on the likelihood agreements contain provisions for women is 23.8 percentage points while the average marginal effect of the presence of women signatories on the likelihood agreements contain provisions for women is 31.3 percentage points. Non-signatory women delegates are not significantly correlated to provisions for women (AME 14.9). I remove logistical support staff and women’s Track 1.5 committees from analysis given positions were documented in three and two cases respectively.