Re-Examining the Origins of Policy Related to Violence Against Women The Role of Women’s Descriptive Representation in Achieving Substantive Representation

Re-Examining the Origins of Policy Related to Violence Against Women:

The Role of Women’s Descriptive Representation in

Achieving Substantive Representation

CHAPTER 1

INTRODUCTION

It is estimated that 35 percent of women worldwide have

experienced either physical and/or sexual intimate partner violence

or sexual violence by a non-partner (not including sexual

harassment) at some point in their lives. (UN Women 2019)

Violence against women is an imminent problem devastating women’s physical and

mental health. According to UN Women, 35 percent of women worldwide have experienced

physical or sexual violence by an intimate partner or a non-partner. Violence against women

is not only problematic in developing countries but also developed countries. In South Korea,

an employee of a bar near Gangnam station in Seoul stabbed a woman who entered the

karaoke room on May 17, 2016. This crime is classified as a crime targeting women because

the criminal spent about half an hour sending six men to the bathroom and commit the crime

when a woman entered the bathroom. The criminal stated that “I committed the crime

because I was ignored by women, and I don’t know the victim.”

Why do some countries adopt laws and policies combating violence against women,

while others do not? Even though the fundamental way to eradicate and control violence

against women, and to help abused women is unquestionably adopting and implementing

laws and public policies addressing violence against women, studies about violence against

women have been largely conducted in other disciplines such as sociology, psychology, and

criminology rather than in political science. Even though some exceptional studies have dealt

with the above question (Htun & Weldon 2012; Weldon 2002, 2004, 2006a), there are still

holes needed to be filled. This dissertation tries to answer the question with the focus on the

role of women’s descriptive representation in parliaments to lead countries to promote laws

and public policies concerning violence against women.

Figure 1-1: Average Women in Parliaments (%) over the World

Source: The Inter-Parliamentary Union (IPU) database.

The level of women’s parliamentary descriptive representation has increased

continuously after the acquisition of passive suffrage, even if several reasons behind female

underrepresentation in political offices such as voters’ bias against female candidates (Anzia

& Berry 2011; Baskaran & Hessami 2018; Beaman 2007; Fréchette, Maniquet, & Morelli

2008), party leader’s bias against women (Casas-Arce & Saiz 2015; Esteve-Volart & Bagues

2012), and the unwillingness of women to be in competitive environments (Eisenkopf,

Hessami, Fischbacher, & Ursprung 2015; Niederle & Vesterlund 2007) still exist. The world

average percentage of female legislators was less than 1% before 1970, exceeding 10% in

2000 and reaching 22% in 2019. This optimistic upward trend is not confined to a specific

region but is occurring worldwide.

With the increased women’s parliamentary descriptive representation, the differences

of female representatives can make in politics have been examined both theoretically and

empirically. Theories behind the differences between female and male representatives are

largely based on the politics of presence (Mansbridge 1999a; Phillips 1995; Pitkin 1967) and

ethics of care (Chorodow 1978; Gilligan 1982; Ruddick 1980). Based on those theories, the

distinction between them are examined related to attitudes (Flammang 1997; Frankovic 1982;

Thomas 1991; Thomas & Wilcox 2014), priorities (Diaz 2005; Reingold 2003; Swers 1998;

Wängnerud 2000), and policy promotion (Berkman and O'connor 1993; Bratton and Ray

2002;

Caiazza 2004; Li-Ju 2008; Crowley 2004; Thomas 1991; Tolbert and Steuernagel 2001;

Cowell-Meyers and Langbein 2009). Compared to studies about the gendered-differences in

attitude and priority, studies about the difference in policy promotion female representatives

can make provide mixed empirical evidence because of the lack of a well-developed

timeseries cross-national analysis dealing with policy promotion “from the start”

(Courtemanche & Green 2017; Wängnerud 2009).

Figure 1-2: Average Women in Parliaments (%) by Geopolitical Regions

Source: The Inter-Parliamentary Union (IPU) database.

This study examines the influence of women’s parliamentary descriptive

representation on the adoption and implementation of laws and public policies addressing

against violence against women “from the start” of such laws and policies. By conducting

large-N cross-sectional time-series statistical analysis with 162 countries from 1908 to 2016,

and by examining the case of South Korea, I test the existing theories, including “the politics

of presence”, “the ethics of care” and “the Critical Mass”, about the influence of women’s

descriptive representation well-developed over decades and develop mechanisms behind the

influence of women’s descriptive representation.

This dissertation shows that higher women’s descriptive representation increases the

probability of adopting the first laws about violence against women in Chapter 4. In addition,

the conducive impacts of women’s descriptive representation in legislative branches on the

implementation of public policies concerning violence against women are also supported by

the empirical evidence from Chapter 5. Moreover, in Chapter 6, based on the case of the

Korean National Assembly, this dissertation demonstrates that female lawmakers are more

likely to cosponsor bills addressing violence against women compared to their male

counterparts and shows that male legislators who have sponsored any bills with female

counterparts tend to cosponsor such bills. In addition, it is revealed that female

representatives involved in centerleft political parties have a higher probability of sponsoring

violence against women bills rather than female colleges from center-right political parties.

Lastly, I find that a higher percentage of female legislators in the total number of sponsors of

violence against women bills tends to increase the success rate of such bills. Those empirical

findings are consistently robust in other modeling strategies and model specifications,

indicating that those relationships are not the artificial findings from an empirical strategy.

1.1. Violence Against Women

Violence against women, defined as “any act of gender-based violence that results in,

or is likely to result in, physical, sexual or psychological harm or suffering to women,

including threats of such acts, coercion or arbitrary deprivation of liberty, whether occurring

in public or in private life” (UN Woman), has received great attention based on not only

domestic but also international perspectives (Devries et al. 2013; Edwards 2011; Fitzpatrick

1993; Htun & Weldon 2012; Montoya 2013; Russell & Deven 1976; Watts & Zimmerman

2002). Even though definitions of violence against women can vary widely, violence against

women typically includes physical assault, sexual violence and harassment, and domestic

violence (Montoya 2013, 17). Moreover, violence against women sometimes also includes

emotional assault (Crowell & Burgess 1996). UN Women categorizes eight forms of violence

as violence against women: domestic and intimate partner violence, trafficking, sexual

violence, sexual harassment, Female Genital Mutilation and Cutting (FGM/C), Child, Early,

and Forced Marriage (CEFM), stalking, femicide or feminicide, and other forms of violence

against women.

Violence against women, which is not a new phenomenon, is an exacting problem

regardless of the level of developments, regime types, regions, religions, cultures, or histories

(Krug, Dahlberg, Mercy, Zwi, & Lozano 2002). Violence against women is one of the major

problems affecting women’s integrity and every socioeconomic group at every life stage

(True 2012). UNWOMEN (2018) provides a series of facts and figures about the severity of

violence against women in the world. According to its’ analysis, 35 percent of women in the

world experience sexual violence including trafficking, stalking, early childhood marriage,

forced marriage forced sex, female genital mutilation, etc. Also, Comprehensive research

from UNODC (2018) shows that 58 percent of women intentionally murdered in 2017 were

killed by intimate partners or family members.

In addition, World Health Organization (WHO) provides a summary of violence

against women including intimate partner violence and non-partner sexual violence around

the world by analyzing seemingly all available data.1 According to its summary, the estimates

1 WHO uses the following data: IVAWS, WHO multi-country study on women’s health and

domestic violence against women, DHS, GENACIS, CDC RHS, and crime surveys across

the globe (World Health Organization 2013).

of the percentage of women who experience physical and/or sexual intimate partner violence

among ever-partnered women are above 20% for all regions and high-income countries

(23.2%). The region with the highest estimate is South-East Asia (37.7%) followed by

Eastern Mediterranean (37%), Africa (36.6%), and Americas (29.8%). Even though the

estimates in Europe (25.4%) and Western Pacific (24.6%) are relatively low, the estimated

percentage is not negligible. The estimated proportion of women reporting intimate partner

violence and/or nonpartner sexual violence by each region and high-income countries is

averagely 10% higher. Violence against women is also widely spread in politics. Elected

women, female candidates, journalists, judges, academics, activists, and practitioners have

also suffered from violence against women at nearly all stages of the political process (Krook

2020b).

Violence against women imposes profoundly negative impacts on women’s life. Abused

women who experience violence against women have suffered from severe physical and

mental health consequences (Kilpatrick et al. 2003; Schnurr & Green 2004). Eby, Campbell,

Sullivan, and Davidson (1995) revealed that abused women tend to be suffered from physical

health symptoms including low energy, sleep problems, headaches, muscle tension/soreness,

constant fatigue, etc., gynecological symptoms such as Pelvic pain and Bladder infection, and

potential exposure to sexually transmitted diseases (STDs) or Human Immunodeficiency

Virus (HIV) infection.

Mentally, violence against women affects women’s mental health by influencing

dignity, self-identity, and self-esteem (Babu & Kar 2009; Bonomi et al. 2006; Coker et al.

2002;

Ellsberg, Jansen, Heise, Watts, & Garcia-Moreno 2008). Trevillion, Oram, Feder, and Howard

(2012) show that the probabilities of experiencing depressive disorders and anxiety disorders,

and post-traumatic stress disorder (PTSD) for abused women are three times, four times, and

seven times higher. Oram, Khalifeh, and Howard (2017) also demonstrate that women who

suffered from domestic abuse or sexual violence tend to have an increased risk of mental

disorders.

Despite the severity of violence against women, public policies combating violence against

women are insufficient (Krook & Sanín 2016). Some countries do not have criminal laws that

are the very basic tools to penalize violence against women and others have laws without

specific words for indicating which behaviors should be considered as violence.

To explain the variation among countries and the reason behind the governmental

provision of public policies combating violence against women, I focus on the role of

women’s descriptive representation. The influence of women’s descriptive representation on

the substantive representation of women has been studied by scholars (Atkinson & Windett

2019; Mansbridge 1999b; Phillips 1995; Pitkin 1967; Swers 2002). Even though there are

strong theoretical backgrounds of the influence of women’s descriptive representation from

the politics of presence and the ethics of care, the empirical results about the influence of

women’s descriptive representation are at best mixed. For example, a study from Htun and

Weldon

(2012), which is might be the most relevant study to the topic of this dissertation, shows that

the influence of women’s descriptive representation is insignificant or marginal compared to

that of feminist autonomous movements on the policy changes combating violence against

women.

However, there are still some limitations that should be overcome through better research

design. First, the time span of the previous literature does not include the period when the

level of women’s descriptive representation was low enough. Second, years, when there was

the adoption of laws and policies about violence against women, are also excluded from the

previous empirical research. Moreover, even though there are various types of public policies

combating violence against women, no empirical studies test the influence of women’s

descriptive representation while controlling potential heterogeneities across different types of

public policies.

1.2. Research Design

To investigate the influence of women’s descriptive representation on the provision

of public policies about violence against women properly, I employ the triangulation research

strategy with a large-N cross-national time-series statistical analysis about 162 countries from

1908 to 2016, and statistical analysis with the data from the Korean National Assembly from

1948 (the start of 1st Congress) to 2020 (the end of 20th Congress). Widely employed in

comparative public policy research (Wolf 2010), triangulation refers to “the use of multiple

theories, data sources, methods or investigators within the study of a single phenomenon.”

(Heale and Forbes 2013, 98). To be specific, triangulation includes employing both qualitative

and quantitative approaches at the same time to answer a specific research question and

testing hypotheses with different data (Tarrow 1995). By relying on different data including

large-N cross-national time-series data and the Korean National Assembly data, this

dissertation examines the relationship between women’s descriptive representation and

substantive representation in terms of laws and policies concerning violence against women.

First, I conduct a large-N cross-national time-series statistical analysis. A large-N

cross-national time-series analysis has relatively less internal validity compared to properly

designed experiments because randomization cannot be used, but more internal validity and

external validity compared to comparative methods and case study approach generally. In the

statistical method, instead of randomization, conditional expectation function (CEF), also

called partial correlation (Lijphart 1971), is used to see whether there are statistically

significant relationships between dependent variables and independent variables by holding

other variables at constant values. In other words, CEF enables us to conduct quasi-

experiments (Collier 1993). If our sample is representative of the population that we are

interested in, the results from statistical methods can be generalized to the population.

Moreover, combined with other methods, the statistical method can be used as the start line

for further in-depth case studies (Lieberman 2005) or can be used for testing hypotheses

generated from the case and comparative methods. In addition, the statistical method is a

useful tool to check generalizability in the absence of randomization. It can efficiently test the

external validity of theories and give robustness to the existing theories or reject theories

(Zuckerman 2009).

Despite the strengths of a large-N cross-national time-series statistical analysis, the large-N

approach is not enough to unveil the relationship between women’s descriptive and

substantive representation related to legal reforms combating violence against women. For

instance, even though I can examine the influence of the percentage of women in legislative

branches on the adoption and implementation of violence against women laws and public

policies at the aggregated-level, it is hard to study how gender affects legislative behaviors

related to the issue of violence against women. For instance, are female legislators more

likely to sponsor bills concerning violence against women? Do male legislators with previous

experience of sponsoring bills with female legislators tend to sponsor violence against women

bills? In addition, violence against women bills sponsored by female legislators from different

political parties are more likely to be passed? Given that collecting information about all

individual legislators and proposed bills over the globe is nearly impossible, answering those

essential research questions with large-N time-series cross-national data is difficult.

Therefore, I conduct a statistical analysis with the data on the Korean National Assembly

which provides more information about individual legislators and their legislative behaviors.

The combination of a large-N cross-national time-series statistical analysis with the data on

162 countries from 1908 to 2016, and statistical analysis with the Korean National Assembly

data enables me to examine the generalizability of the link between women descriptive and

substantive representation, and to unveil the mechanism of how the entrance of women in

legislative branches affect legal reforms concerning violence against women.

1.3. Organization of the Dissertation

The rest of this dissertation is divided into six chapters. In Chapter Two, which

follows, I thoroughly review previous literature on women’s descriptive representation and

violence against women. To be specific, previous literature about the relationship between

women’s parliamentary descriptive representation and countries’ efforts to combat violence

against women will be presented. In addition, theories related to the influence of women’s

descriptive representation in parliaments on social and political phenomena will also be

presented. Lastly, I point out the necessity to conduct mixed analysis of large-N and case

study which can support the generalizability of the findings from previous literature and re-

examine the theories behind the relationship between women’s descriptive representation and

substantive representation.

In Chapter 3, theories and hypotheses tested in this dissertation will be presented. First,

theories and hypotheses tested with the time-series cross-national data will be introduced

based on the long-lasting theories including the politics of presence, the critical mass, and the

ethics of care theories. I hypothesize that the improved women’s descriptive representation in

legislative branches leads countries to adopt and implement criminal laws and public policies

addressing violence against women. In addition, the conditional relationship between

women’s descriptive representation and the proliferation of global norms about violence

against women will be theorized. Furthermore, the hypotheses tested with the data on the

Korean National Assembly will be presented. To be specific, I hypothesize the relationship

between gender and bill sponsorship on violence against women bills, the conditional

relationship between gender and partisanship of individual legislators, and the influence of

previous cosponsorship experiences with female legislators on the legislative behaviors of

male counterparts. Moreover, related to bill success, the relationship between the percentage

of female cosponsors of the total number of cosponsors and bill success will be theorized.

Lastly, I theorize that violence against women bills sponsored by female legislators from

different political parties are more likely to be passed.

In Chapter 4, I, relying on time-series cross-national data on 162 countries from 1908

when Japan first adopted laws combating violence against women to 2016, examine the

influence of women’s descriptive representation in legislative branches on the adoption of

laws concerning violence against women. By estimating discrete event history models based

on logistic regression, I demonstrate that a higher percentage of women in parliaments leads

countries to adopt violence against women laws. In terms of predicted probabilities, this

relationship is not only statistically significant, but also substantially meaningful. After I

control the potential heterogeneous impacts of women’s descriptive representation on the

adoption of violence against women laws according to the types of violence, the relationship

is consistently robust.

In Chapter 5, I examine the relationship between women’s descriptive representation

and the implementation of public policies categorized into 20 types of policies. Based on

discrete pooled event history analysis (PEHA) developed by Kreitzer and Boehmke (2016), I

find that a higher percentage of women in legislative branches increases the log-odds of

implementing public policies concerning violence against women. The random effect of

women’s descriptive representation is not statistically significant, meaning that the positive

influence of women’s descriptive representation does not depend on the types of public

policies. More interestingly, by examining countries implementing violence against women

laws and public policies despite no female representative in legislative branches, I show that

the laws and public policies adopted and implemented without supports from female

legislators in Asian, MENA, and Sub-Saharan countries have non-negligible limitations in

terms of details of clauses and sentence. On the contrary, in Western European countries such

as Denmark and France, the adoption and implementation of successful violence against

women laws and public policies are rooted in the active feminist movements.

In Chapter 6, I further examine the mechanisms behind the link behind the statistically and

substantively significant relationship between women’s descriptive representation in

parliaments and the adoption and implementation of laws and public policies addressing

violence against women with the case of the Korean National Assembly. In this chapter, by

analyzing 85,860 bills proposed from 1948 (the start of 1st Congress) to 2020 (the end of 20th

Congress) based on the Bill Information System developed by the Korean National Assembly,

I demonstrate that female legislators are more likely to sponsor bills addressing violence

against women compared to their male counterparts, that the previous experience of

cosponsorship with female legislators leads male legislators to promote bills combating

violence against women, and that female legislators involved in the center left political parties

tend to focus on sponsoring violence against women bills rather than those from center-right

political parties. Moreover, I show that bills sponsored by a higher percentage of female

legislators have a higher probability of being passed.

Finally, Chapter 7 summarizes the empirical results from the two time-series

crossnational analysis chapters and the statistical analysis with the data on the Korean

National Assembly. In addition, this chapter discusses the contributions of my dissertation to

both the literature on Women and Politics and the literature on violence against women.

CHAPTER 2

LITERATURE REVIEW

If the representation be so formed as to give one or more of the natural

classes of men in the society an undue ascendancy over the others, it is

imperfect; the former will gradually become masters, and the latter

slaves. (Storing 1981, 75)

The saying from (Storing 1981, 75) emphasizes the fact that representatives usually do

not represent all public regardless of gender, class, race, etc., but represent some of the public.

Given this fact, in Chapter 2, I review the literature about women’s descriptive representation

in parliaments and the differences female legislators or representatives can make in policy

outcomes. After the review of previous literature, I introduce theories and hypotheses that will

be tested throughout this dissertation.

2.1. Previous Literature on the Female Legislators and Differences They Make

Various previous studies handle the questions such as does women’s descriptive

representation lead to substantive representation (Pitkin 1967) and do female legislators can

make differences in a polity such as attitudes, interests, priorities, and differences of policy

outcomes (Bratton & Haynie 1999; Darcy 1994; Norris 1996; Thomas 1994). Representatives

sharing common experiences and social identity including gender, race, or class are more

likely to represent their own groups' interests rather than other representatives do

(Mansbridge 1999b; Phillips 1995; Sapiro 1981; Swers 2002). Gender differences between

female and male legislators have been studied with different focuses. I review the gender

differences among legislators depending on genders around attitudes, priorities, and policy

promotion.

Differences in attitudes between female and male legislators are widely studied by scholars

and there is a consent that gender influences the attitudes which individual legislators have

(Esaiasson & Holmberg 2017; Lovenduski & Norris 2003; McAllister & Studlar 1992; Narud

& Valen 2000; Norris & Lovenduski 1995), despite some debates about the degree of the

influence.

As Wängnerud (2009) summarized, the direction of the gender gap in political

attitudes between female and male legislators is relatively clear. A series of studies show that

female legislators tend to be more liberal or politically left compared to their male colleagues

(Flammang 1997; Frankovic 1982; Thomas 1991; Thomas & Wilcox 2014). By collecting

data from a mail survey of more than 2,500 legislators across states, Poggione (2004)

demonstrates that female legislators show more support on the welfare policy liberalism

which index is composed of 17 dichotomous items about legislators’ support on the principles

of welfare reform and initiatives. Also, even though McCarty, Poole, and Rosenthal (1997)

argue that the pro-liberal attitude of female legislators can be totally explained by district

factors, S. Welch

(1985) and Burrell (1996), based on conservative coalition support scores from the 93rd and

96th U.S. Congresses, demonstrate that congresswomen tend to be more ideologically liberal

rather than their male counterparts. Moreover, Swers (2002) creates a database about the

number of bills related to women’s issues and unveils that female legislators pay more

attention to women’s issues regardless of their partisanships by comparing male and female

members of House of Representatives in the United States during 103rd and 104th Congresses.

Differences in attitudes are also observed in the attitudes toward women’s issues or women’s

interests including social policy, affirmative actions, or pornography (Burrell 1996; Gehlen

1977; Leader 1977; Wängnerud 2009). Berkman and O'connor (1993) show that female

representatives are more likely to secure committee assignments allowing them to block

prolife legislation such as abortion policy. In addition, with the voting records of all

representatives in the 103rd Congress on women’s issues, Swers (1998) reveals that female

legislators are likely to vote for women’s issues while controlling other major factors of

congressional voting such as individual characteristics, ideology, and constituency factors.

I should note that some previous literature, especially the U.S. context, argues that

there has been a growing divide between Democratic and Republican women and similarity

among male and female legislators in the same political parties, and emphasize the crucial

role of a political party (Frederick 2009; Osborn 2012). Based on Aldrich (2011), Thomsen

(2015) demonstrates that both self-selection and party recruitment make the influence of

partisanship in determining the ideology of individual legislators.

While attitudes are the matter of “on the issues” which are already politicized or publicized,

priorities are related to what kinds of issues will be politicized or put on the agenda

(Wängnerud 2009). As the differences between female and male representatives on attitudes

toward issues have been studied broadly, the different priorities of issues from female and

male representatives are also widely studied by scholars (Diaz 2005; Reingold 2003; Swers

1998; Wängnerud 2000). The differences in priorities between female and male

representatives are largely based on the distinct psychological development of women and

men (Chorodow 1978; Jones 1997). As Childs and Krook (2006) said “There is some, albeit

inconclusive, evidence that increases in the percentage of female deputies also lead to the

advancement of policy issues that are important women, such as childcare, equal pay, and

protection against women”, there has been a consensus that female representatives tend to

prioritize women’s issues, even though empirical results can be different based on how

researchers determine and measure women’s issues.

In American Politics, Thomas and Welch (1991), based on the 1988 survey of members of

the lower houses of the state legislatures in twelve states, show that female state legislators’

list of priority bills tend to include more legislations about children and family than the lists

of men; that female representatives are likely to introduce more welfare bills than male

counterparts; that women tend to be on health and welfare committees; and that women are

likely to list women’s issues as accomplishments. They support the gender influence on the

issue priorities by saying that “Increased numbers of women legislators mean a larger

potential support network for women who push issues of importance to them”.

More recently, Atkinson and Windett (2019) analyze all bills introduced in the U.S. House of

Representatives from 1963 to 2009. They classify issues of bills into 19 topic areas and

demonstrate that female representative tends to devote more attention on the issues including

“civil rights, and liberties”, “education”, “health”, “law, crime, and family issues,” and “social

welfare compared to their male counterparts. Moreover, some studies also show that female

legislators tend to focus on social and family issues compared to male legislators (Diamond

1977; Kirkpatrick 1974; Werner 1968), and the influence of potential support network for

women is the pre-requisite of the difference between male and female priorities

(Flammang 1985; Saint-Germain 1989).

The differences in priorities between women and men also exist in other developed

and developing countries. Schwindt‐Bayer (2006) reveals that female legislators tend to place

a higher priority on women’s issues and children/family concerns by examining all bills

initiated during two congresses in Costa Rica and Colombia, 1994-98 and 1998-2002, and

two years in Argentina, 1995 and 1999. Taylor-Robinson and Heath (2003)’s case study on

Honduran Congress from 1990-93 also finds that female legislators are likely to emphasize

women’s rights issues, even though they show that there is no difference between female and

male legislator’s priorities on children’s and family issues. In addition, Wängnerud (2006)

analyzes Swedish parliaments over 20 years and finds that 75% of women in parliaments

focused on the issues of social policy, family policy, care for the elderly, and health care when

they conduct election campaigns. Compared to women, only 44% of men in parliaments

address those issues in 1985. More recently Tam (2017) finds that female legislators in an

undemocratic or semi-democratic legislature focus on women’s interests rather than male

legislators do by conducting a case study of Hong Kong.

Likewise, the differences between female and male representatives are widely studied and

empirically tested. However, the policy outcome or policy promotion that women legislators

can make has been relatively less examined. Wängnerud (2009) emphasized this relative

marginalization as “Although there are studies on policy promotion, the closer one gets to

outcomes in citizens’ everyday lives, the fewer empirical findings there are to report.” Given

the importance of politics on citizens’ everyday lives, the evaluation on the link between

women’s descriptive representation and substantive representation including policy changes

should be conducted to fully understand the influence of women’s descriptive representation

and the differences that women representatives can make. As Bratton and Ray (2002)

emphasize the importance of evaluating the connection between women’s descriptive

representation and policy outcomes by mentioning that “The link between descriptive

representation and substantive representation is most meaningful when descriptive

representation affects not only a representative’s attitudes and behavior but also the decisions

made by the overall legislative body”, studying the association between women’s descriptive

representation and substantive representation is essential.

Even though the relative deficiency of studies about policy outcomes that women can make,

there are also previous studies about this topic (Berkman & O'connor 1993; Bratton &

Ray 2002; Caiazza 2004; Cowell-Meyers & Langbein 2009; Crowley 2004; Li-Ju 2008;

Thomas 1991; Tolbert & Steuernagel 2001). Even though the previous studies on this issue

are concentrated on American Politics, there are also some studies in developing and other

developed countries.

Related to the United States, Saint-Germain (1989) conducts a longitudinal analysis about

public policy output in the Arizona State Legislature between 1969 and 1986 and reveals that

women legislators tend to initiate more bills about women’s interests and feminist issues than

male counterparts, which in turn produces more public policies about women’s interests.

More recently, Cowell-Meyers and Langbein (2009) investigate the association between

women’s descriptive representation and substantive representation by using logistic and

leastsquares estimations. Based on the examination of 34 policies including violence against

women, child policies, welfare policies, etc. which are identified by the Institute for Women’s

Policy Research (IWPR) as the interests of women in the U.S. States, they find evidence that

a higher level of women’s descriptive presentation in state legislatures tends to lead state-

level policies favoring women in the 1990s on 5 of the 34 policies such as favorable policies

and child support.

In addition, Eckhouse (2019) reveals that descriptive representation increases criminal justice

by conducting a case study in the United States.

In addition to the direct influence of women’s descriptive representation on substantive

representation, Caiazza (2004) shows the indirect effect of women’s descriptive

representation on policy outcomes. She suggests a two-staged relationship from party and

gender attitudes to women-friendly policies including employment protections of women,

legislation protecting sexual minorities, and reproductive rights, through women’s

representation. Using data about state policies in 2000, she revealed that the favorable

attitudes toward women representatives and Democratic dominance tend to increase the

proportion of female state legislators and in turn lead to a higher level of women-friendly

policies.

Studies about the link between women’s descriptive representation and substantive

representation are not confined only to the U.S., but also conducted with a focus on other

countries in various regions in the world. Bratton and Ray (2002) investigate the link between

women’s descriptive representation and substantive representation measured by the provision

of child care with the focus on the Norwegian municipalities in 1975, 1979, 1983, 1987, and

1991. They show that the proportion of women in the council leads the provision of child care

by using Ordinary Least Square (OLS) regressions. In terms of the adoption of policies

related to family leave, Kittilson (2008) demonstrates that the presence of women in

parliaments significantly influences the adoption and scope of both maternity and childcare

leave policies by analyzing family leave policies in 19 democratic countries from 1970 to

2000. More importantly, she argues that the women’s descriptive representation in legislative

branches has played a more important role in adopting such policies rather than the ideology

of the party in power. A more recent study from Wiener (2021) relying on an experimental

approach provides evidence for the link between women’s descriptive and substantive

representation.

Some evidence comes from studies on non-democratic countries. Bauer and Burnet

(2013) conduct a case study about two different sub-Saharan African countries: Botswana

which is a democratic country and Rwanda which is an authoritarian single-party dominant

country. They argue that women’s descriptive representatives are helpful to achieve

substantive representation of women even in undemocratic countries. Nistotskaya and

Stensöta (2018) also demonstrate that the positive dynamics between women’s political

representation and womenfriendly policy outcomes in a large electoral autocracy, Russia.

Some of the previous cross-national studies also provide evidence for the influence of

women’s descriptive representation on substantive representation. Bolzendahl and Brooks

(2007) analyze the influence of women’s political power on welfare state effort within 12

capitalist democracies. With Prais-Winsten regression with panel-corrected standard errors,

they demonstrate that higher women’s descriptive representation tends to increase the levels

of social expenditure while controlling a series of variables such as year, unemployment,

GDP per capita, and constitutional structure. Moreover, Schwindt-Bayer and Mishler (2005)’s

seminal work shows that time-series cross-national study on the examination of the link is

promising. The main question they try to answer in their article is whether women’s

descriptive representation increases legislatures’ responsibilities to women’s issues or not. As

main dependent variables, they use a series of policies including weeks of maternity leave,

women’s social and political equality, and marital equality in law. Using data about 31

democracies, they provide evidence for the positive influence of women’s descriptive

representation on substantive representation.

However, findings on the association between women’s descriptive representation and

substantive representation are mixed at best. There are some studies which demonstrate that

the influence of women’s descriptive representation in parliaments on policy outcomes and

policy promotions is insignificant, at best very marginal.

Tolbert and Steuernagel (2001) build an index of women’s health mandates based on the

nine specific policies such as breast cancer screening and direct access to ob-gyns without a

referral from primary care physicians and test whether women’s descriptive representation in

state legislatures leads the adoption of women’s health mandates or not. They show that the

presence of women in the state legislature in the 1990s and of women in leadership positions

does not influence policy adoptions about women’s health policies. Also, Thomas (1994)

distinguishes between legislative procedures such as political speeches, co-work with

colleagues, and the interaction with lobbyists, and legislative products including voting

records, issue attitudes, and policy priorities. By empirically comparing the difference

between female and male legislators, she demonstrates that the gender gap in legislative

procedures is closing but that in legislative products is not.

2.2. Previous Literature on Violence Against Women Laws and Policies

To reiterate, violence against women and girls is a global phenomenon which is not limited

to certain geopolitical regions or to specific socio-economic classes (Alesina, Brioschi,

& La Ferrara 2021; García-Moreno et al. 2015; Krook 2018; Montoya 2013). In the era of the

COVID-19 pandemic, the problem of violence against women is getting worse (Roesch,

Amin, Gupta, & García-Moreno 2020). Even though a global consensus on that violence

against women is one of the most pressing global concerns has been made and the reparations

for the victims of violence had been addressed by the Convention on the Elimination of All

Forms of Discrimination against Women (CEDAW) over decades (Rubio-Marin & Estrada-

Tanck 2020), the problem of violence against women has not been completely mitigated.

Given that violence against women disproportionately imposes profound negative both

physical and psychological problems on women and girls (Reingold 2003; Swers 2002),

violence against women has been widely regarded as an issue of women’s rights and integrity

(Faxon, Furlong, & Sabe Phyu 2015; Swers 1998; Wiener 2021). However, previous

literature on the influence of women’s descriptive representation on the adoption and

implementation of laws and policies addressing violence against women also has provided

mixed results. Some studies demonstrate that a higher percentage of women in parliaments

tends to lead countries to adopt and implement public policies combating violence against

women (e.g., Woo (2021a) and Woo (2021b)), while others show that the role of women’s

descriptive representation in legislative branches is marginal or even not statistically

significant (e.g., Weldon and Htun (2013), Htun and Weldon (2012), and Weldon (2002)).

Htun and Weldon (2012)’s crucial cross-national study on the association between female

autonomous movement and progressive policy change focusing on violence against women

also does not support the influence of women’s descriptive representation on policy

outcomes. They build an index to measure the level of progressive policies in randomly

selected 70 countries in 1975, 1985, 1995, and 2005. The index of progressive policy, ranging

from 0 to 10, is based on whether a country has services to victims of violence against

women, legal reform, policy or program targeting vulnerable populations of women, training

professionals, prevention programs, and administrative reforms. Through OLS regression,

they reveal that strong autonomous domestic feminist movement, women’s policy machinery,

and global and regional norms tend to lead countries to redress violence against women. On

the contrary, they argue that women’s descriptive representation in legislatures measured by

the proportion of women is insignificant and has only negligible impacts.

In their subsequent study with the same dataset, Weldon and Htun (2013)

reemphasize the important role of feminist activism rather than the number of women

legislators. Previous to Htun and Weldon (2012), Weldon (2002) shows that the interaction

between the feminist autonomous movements and women’s policy machinery has a strong

significant impact on the policy action against violence against women in 36 democratic

countries. In addition, some studies on individual countries have also provided empirical

supports for the null effects of the women’s presence in legislative branches on the

comprehensiveness of violence against women legislations (Beer 2017).

Even though Htun and Weldon (2012)’s research is groundbreaking in the sense that it

provides an alternative factor of policy changes about violence against women, there are still

rooms to be improved. First, an analysis based on cross-sectional data only shows the

correlations among variables, not causalities. Moreover, Htun and Weldon (2012) do not

exactly test when countries adopt policies about violence against women, but they only test

how the policies change. In other words, Htun and Weldon (2012) evaluate the influence of

women’s descriptive representation on the “aggregated” index of progressive policies toward

violence against women. However, the influence of female legislators can be different among

the components of the index. For example, the influence can be relatively strong related to the

adoption laws, but less strong to the training professionals or administrative reforms.

Second, the four cross-sectional years from 1975 to 2005 (1975, 1985, 1995, and

2005) of Htun and Weldon’s (2012) study are not broad enough to test the influences of

independent variables on policy adoptions. For example, the influence of the percentage of

women in parliament can be biased downward with the limited time span because the global

average of women in parliament (%) has significantly increased after 1970. The limited time

span is also problematic because Japan firstly adopted laws, which is one of the components

of Htun and Weldon’s index, about violence against women in 1908. To fully understand the

influence of women’s descriptive representation, we need to go back to when the first

adoption happened, as Beckwith (2007) advises that researchers must follow what happens

“from the start” when women are few, up to the point when more women present in

parliaments.

At the same time, studies supporting the explanatory power of women’s descriptive

representation in parliaments on the adoption and implementation of laws and public policies

concerning violence against women have similar limitations. Even though the two previous

studies on the adoption and implementation of violence against women policies from Woo

(2021a) and Woo (2021b) and demonstrate the strong link between women’s descriptive and

substantive representation in terms of violence against women policies, those studies still do

not examine what happens “from the start.” For instance, Woo (2021b) find that an increase in

the percentage of women in parliaments leads countries to adopt laws addressing violence

against women based on a time-series cross-national analysis on 49 Asian countries from

1950 to 2016, which geographical focus of the study prevents us from generalizing the

finding. In addition, although Woo (2021a) examines the link between women’s descriptive

representation and the adoption of policies combating violence against women, the time span

of the countryyear observations is limited to four decades from 1975 to 2016. Given that

Japan first adopted violence against women law in 1908, the time span is quite limited to fully

evaluate the impact of women’s descriptive representation.

To sum up the previous research, the findings from the previous literature about the

link between women’s descriptive representation and substantive representation remain

ambiguous (Courtemanche & Green 2017) and there is a lack of good cross-country

comparative research (Wängnerud 2009). Given that careful examination about the

association between women’s descriptive representation and substantive representation is

essential, I believe that a well-developed time-series cross-national analysis “from the start” is

the next step to broaden the understandings of the influence of women’s descriptive

representation by overcoming the difficulties of conducting longitudinal analysis on this topic

(Wängnerud 2009).

Thus, this dissertation aims to empirically test the link based on unprecedentedly

large data about the policy outcomes combating violence against women in 162 countries

from 1908 when Japan firstly adopted a domestic law punishing violence against women to

2016. Also, I conduct statistical analysis with the data about legislative behaviors from the

first Korean National Assembly (1948-1950) to 20th (2016-2020). In the case of South Korea,

I will focus on the tests of mechanisms behind the potential link between women’s descriptive

representation and policy outcomes about violence against women.

CHAPTER 3

THEORIES AND HYPOTHESES

How does women’s descriptive representation lead countries to adopt and implement

laws and public policies combating violence against women? Pitkin (1967)’s explanation of

women’s descriptive and substantive representation gives a clue to the conducive influence of

women’s descriptive representation on the adoption and implementation of laws and public

policies related to violence against women. She defines political representation as an activity

that makes voters “present” in policy-making processes, descriptive representation as a body

mirroring the characteristics including gender and race of the represented groups, and

substantive representation as a body reflecting the interests of the community. Put simply,

substantive representation is “acting for” representation and descriptive representation is

“standing for” representation (Celis & Childs 2008). Pitkin (1967) argues that an increase in

“descriptive representation” tends to lead to greater “substantive representation” (Bailer,

Breunig, Giger, & Wüst 2021).

In a similar vein, the politics of presence theory developed by Phillips (1995) asserts

that the gender of representatives is an influential factor in representative democracy

(Mansbridge 1999b; Phillips 1998; Tamerius, Duerst-Lahti, & Kelly 2010; Williams 2000;

Young 2002). Based on the assumption that female legislators and female citizens have

shared life experiences such as child-rearing or exposure to violence, Phillips (1995, pp. 67–

68) underlines that the entrance of women into politics is a prerequisite for the politicization

of women’s issues. Some studies interviewing legislators and legislative staffers provide

empirical evidence for the politics of presence theory (Burden 2007; Grose 2011; Lawless

2015; Swers 2016; Wangnerud & Sundell 2012).

In addition to the politics of presence theory, the critical mass theory has also been

widely used as a theoretical framework to explain the association between women’s

descriptive and substantive representation (Celis & Childs 2008; Cowell-Meyers & Langbein

2009; Espírito-Santo, Freire, & Serra-Silva 2020; Homola 2021). The critical mass theory

posits that a certain proportion of female legislators in legislative bodies is needed for women

to make their voices audible in the policy-making process (Dahlerup 1988, 2006; Kanter

1977). According to Kanter (1977, pp. 966–972), in skewed groups where a large

preponderance of one dominant group exists, the dominant tends to control the minority (also

termed “tokens”). Due to their minority status, the tokens with more visibility in a society

tend to suffer from stereotypes, isolation from the society, and confirmation to dominant

models, which, in turn, reinforce the skewed structure (Kanter, 1977, pp. 971–2). In this case,

it is hard for female representatives to make their voice for women’s interests and rights

heard.

On the other hand, when the number of “tokens” increases to a certain point, this

prevents them from being isolated by the dominant group. The possible allies and coalitions

among female representatives under this condition make it possible for them to break the

vicious circle, to escape female stereotypes and role entrapment, and to pursue their interests

(Kanter 1977, p. 966). Therefore, once the proportion of female representatives reaches a

certain point, female legislators politicize women’s issues including violence against women

more efficiently and can make their male counterparts take bills related to women’s issues

more seriously (Celis 2007; Dahlerup, Campbell, & Childs 2014).

Theories developed by Gilligan (1982) and Mansbridge (1999b) also support the

positive influence of women’s descriptive representation on substantive representation.

Gilligan (1982) argues that men and women have been taught differently regarding views on

morality. The main thrust of her theory is that women, unlike men, tend to be taught empathy

and compassion as the core of morality from childhood. This difference in the “ethics of care”

between men and women makes women pay more attention to issues such as welfare policies

and the avoidance of using military forces. (Chorodow 1978; Ruddick 1980). In addition,

Mansbridge (1999), using a case study of the U.S., theorizes that female representatives can

appropriately reflect the needs of female citizens based on their shared experience.

Mansbridge (1999) argues that because of their common female identity, female legislators

are more likely than male legislators to understand the problems female citizens experience,

formulate legal solutions to those problems, and incorporate the solutions into the legislative

processes. She also demonstrates that female legislators tend to focus more on women’s

issues including education, health, and welfare than male legislators do, because female

representatives facilitate conversations with female citizens and politicize the problems

affecting them.

Even though the above-mentioned theories rely on different explanations, all aim to

explain how women’s descriptive representation can promote their substantive representation.

It should be noted that a wide array of studies with data about individual representatives has

highlighted heterogeneities among women representatives according to their partisanships

(Barnes & Cassese 2017; Osborn 2012; Schilling & Osborn 2020), individual characteristics

(Costa, Greenlee, Nteta, Rhodes, & Sharrow 2019; Deckman & Swers 2019), or

intersectionality related to race and class (Barnes 2016; Smooth 2011). Therefore, taking on

board the caveat that female representatives are not a homogenous group, this article based on

unprecedentedly large data on 162 countries from 1908 to 2016 tests whether higher levels of

women’s descriptive representation at the aggregate level leads countries to adopt and

implement laws and public policies combating VAW.

Hypothesis 1: Higher levels of women’s descriptive representation lead countries to

adopt laws addressing violence against women, while all other things being equal

(ceteris paribus).

Hypothesis 2: Higher levels of women’s descriptive representation lead countries to

implement public policies addressing violence against women, while all other things

being equal (ceteris paribus).

In addition to the independent impacts of women’s descriptive representation on the adoption

and implementation of laws and public policies concerning violence against women, this

dissertation sheds light on the conditional relationship between the presence of women in

legislative branches and the proliferation of global norms2 about violence against women. The

rich literature on global norms has emphasized the pivotal role of global norms on normative

and behavioral changes in states (Checkel 1998; Finnemore 1996). The strong linkage

between domestic and transnational actors generated from the proliferation of global norms

about violence against women can raise moral consciousness, support domestic movements

toward

human rights, and challenge norm-violating governments’ behaviors (Hovi, Huseby, & Sprinz

2005; Risse, Ropp, & Sikkink 1999).

I expect that the conducive influence of the proliferation of global norms about violence

against women on the adoption and implementation of violence against women laws and

public policies is magnified by a higher women’s descriptive representation in legislative

branches. This expectation is based on the previous findings that female legislators are more

responsive to women’s issues regardless of their partisanships compared to their male

counterparts (Swers 2002) and that differences in attitudes are also observed in the attitudes

toward women’s interests in terms of social policy area (Poggione 2004; Schwindt‐Bayer

2006). Given that national assemblies or legislatures play a crucial role in legal reforms which

are one of the ways to show countries’ commitment to the proliferated global norms about

violence against women, the higher percentage of women in decision-making positions may

amplify the impacts of the global norms on the adoption and implementation of laws and

public policies concerning violence against women.

2 Following Finnemore and & Sikkink (1998, 891), a norm in this dissertation is defined as “a

standard of appropriate behavior for actors with a given identity.”

Hypothesis 3: The impacts of the proliferation of global norms about violence

against women on the adoption of laws addressing violence against women are

magnified by the improved women’s descriptive representation in legislative

branches, while all other things being equal (ceteris paribus).

Hypothesis 4: The impacts of the proliferation of global norms about violence against

women on the implementation of public policies addressing violence against women are

magnified by the improved women’s descriptive representation in legislative branches,

while all other things being equal (ceteris paribus).

With the data on the Korean National Assembly from 1948 (the start of 1st Congress) to 2020

(the end of 20th Congress), I further examine why a higher percentage of women in

parliaments leads countries to adopt and implement laws and public policies about violence

against women. The focus on the Korean case is based on several reasons.

First, relying on the Korean case enables me to take analytical advantages. The Bill

Information System developed by Korean National Assembly provides bill sponsorship

record with detailed information about each proposed bill from the 1st Congress (1948-1950)

to the 20th Congress (2016-2020), which makes it possible to fully understand the influence of

women’s descriptive representation in legislative branches on the adoption and

implementation of laws and public policies concerning violence against women from the

start. Second, given that Tsebelis (2002) argues that executive dominance over legislative

bodies on agenda-setting in legislative branches has been observed frequently in

parliamentary democracies unlike presidential democracies, South Korea classified as a

presidential democracy is a valuable case where the Korean National Assembly has played a

crucial role in shaping legislative agenda on violence against women. Thrid, South Korea is a

country in which both women’s descriptive and substantive representation in terms of

violence against women laws and policies have progressed dramatically in a short period,

providing enough variations to be examined. In addition, examining the link between

women’s descriptive and substantive representation in terms of legal reforms combating

violence against women with the Korean case will broaden our understanding of the influence

of women’s descriptive representation in policy outcomes in the Asian context.

With the Korean case, I examine whether the gender of individual legislators is a

driving factor in sponsoring bills about violence against women. Even though previous

literature on the legislative behaviors demonstrates that female legislators are more likely to

support bills related to women’s issues rather than their male counterparts are, studies on this

association in Asian context especially with a focus on violence against women bills

relatively have not been developed. Based on the politics of presence and the critical mass

theories, I expect to observe a strong relationship between gender and legislative behaviors

related to bills about violence against women.

Hypothesis 5: Female legislators are more likely to sponsor bills addressing violence

against women compared to male legislators are, while all other things being equal

(ceteris paribus).

In addition to the direct influence of women’s descriptive representation, this

dissertation goes one step further based on those previous literature focusing on the influence

of women’s parliamentary descriptive representation and the difference between male and

female legislators (Dovi 2002; Mansbridge 1999a; Phillips 1995; Rosenthal 1995; Swers

2002; Wängnerud 2009; Williams 2000) and tests the influence of legislative networks with

female legislators on the legislative behaviors of male legislators. I relying on the data about

bills proposed in the Korean National Assembly test the indirect influence of female

representatives on the adoption and implementation of laws and public policies addressing

violence against women by changing the legislative behaviors of their male counterparts. I

argue that the previous experience of sponsoring bills with female legislators makes male

legislators focus on the issue of violence against women.

I build theories based on the rational approach and informational theories about

legislators’ policy-making behaviors. Proposing bills require legislators to invest their limited

resources including time and budgets. Assuming legislators are risk-averse, the limitation of

accessing information about specific issues will prevent legislators from engaging in

proposing bills about the issues, because the conditions of relative uncertainty legislators

cannot sure about the results (Krehbiel 1992).

Based on the previous findings, female legislators have more easy access to information

about women’s issues and they already have more information compared to male counterparts

because of women’s expertise and shared experience (Dovi 2002; Mansbridge 1999a; Phillips

1995; Rosenthal 1995; Swers 2002; Wängnerud 2009; Williams 2000). Thus, male legislators

with less information and interest about women’s issues relatively put less effort into

proposing bills about women’s issues. Male legislators can get access to the information by

conversing with other female representatives through legislative networks. Thus, male

legislators with legislative networks with female representatives have less informatic

limitations on proposing bills about women’s issues, because they can get information from

female legislators. The decreased uncertainty to propose bills about women’s issues will lead

male legislators to propose more bills about violence against women. Therefore, I expect that

male legislators with experience in sponsoring bills with female legislators are more likely to

sponsor bills addressing violence against women.

Hypothesis 6: Male legislators with experience of sponsoring bills with female

legislators are more likely to sponsor bills addressing violence against women, while

all other things being equal (ceteris paribus).

In addition to the independent and indirect impacts of the presence of women’s descriptive

representation in legislative branches, the interactions between gender and partisanship of

individual legislators will be examined. There have been a number of studies unveiling

heterogeneities among female legislators according to their partisanships rather than assuming

female legislators are homogeneous (Cooperman 2014; Kreitzer 2015; Osborn 2012; Osborn

& Kreitzer 2014; Reingold, Kreitzer, Osborn, & Swers 2021; Swers 2002). Recently,

Reingold et al. (2021) demonstrate that conservative Republican women tend to lead

antiabortion policy and their leadership is conditioned by policy contexts. To be specific, they

argue that such leadership is most likely to emerge under the conditions of competitive

partisan environments and when women-centered issue frames are used.

In the South Korean context, political parties and party systems have been developed

differently from those in other advanced democratic countries. Even after its democratization

in 1987, a comparable and stable party system did not emerge in South Korea over decades (J.

Kim 2020). The South Korean party system is highly structured both by elite networks and

political dynasties based on strong regional connections (H. Lee & Singer 2020). There have

been 87 political parties involving legislators from the 1st Congress (1948-1950) to the 20th

Congress (2016-2020). Despite a high number of political parties, there are actually two

political parties: “Center-Right (conservative)” and “Center-Left (liberal)” with different

names or labels. The high number of historically existing political parties is rooted in the

attempts of parties to distance themselves from the demeaning legacies of individual

presidents (Lim & Alsford 2021). In addition, South Korea’s heavy reliance on personalistic

party cues, centralization of power, and strong political attention in the public also contribute

to the immense party relabeling (M.-s. Kim 2015).

As mentioned earlier, even though several smaller leftist political parties have existed

and emerged during two recent decades, the two dominant wings can be categorized as

“CenterRight (conservative)” and “Center-Left (liberal)”. The two parties’ relationship to the

former authoritarian regime acts as the most salient programmatic cleavage within these

camps (Kang 2008). Like left-leaning political parties in young presidential democracies, the

center-left political parties that existed in South Korea are identified with political cleavages

over class conflicts and minority protection including the protection of women’s rights and

integrity (Shim 2019). Thus, as is common in other countries (Erickson 1997; Reynolds

2013), I expect that female legislators with liberal party affiliation are more likely to sponsor

bills concerning violence against women.

Hypothesis 7: Female lawmakers’ legislative behaviors in terms of sponsoring bills

addressing violence against women are not homogeneous across their partisanships.

In terms of the destinations of proposed bills, I examine whether the bills sponsored by

female legislators and whether the bills sponsored by female legislators from different

political parties have higher probabilities of being passed or not. Based on the recent study

from Holman, Mahoney, and Hurler (2021) arguing that one of the ways by which female

legislators overcome more electoral vulnerability and gendered performance expectation is

bill success, I expect that bills sponsored by female legislators are more likely to be passed.

On the contrary, some studies argue that female legislators are marginalized in legislative

branches and prevented from passing their bills, due to the restricted access to important

political resources compared to their male counterparts (Kerevel & Atkeson 2013; Miguel

2012).

Therefore, I will test the counterargument will be also tested at the same time.

Hypothesis 8: Bills sponsored by a higher percentage of female legislators are more

likely to be passed, while all other things being equal (ceteris paribus).

Hypothesis 9: Bills sponsosred by a higher percentage of female legislators are less likely

to be passed, while all other things being equal (ceteris paribus).

In addition, I will examine whether the bills supported by female legislators from

multiple political parties are more likely to be passed compared to other bills only supported

by female legislators from a certain political party. It is based on the rationale that female

legislators tend to reach across the aisle (Osborn 2012) and they often overperform their male

counterparts to overcome legislative marginalization and higher electoral standards imposed

by voters for women (Holman et al. 2021; Lazarus & Steigerwalt 2018; Thomsen & Sanders

2020). Given that previous studies on gender and partisanship are largely conducted with a

focus on the U.S. context, the hypothesis test on the relationship between the success of bills

and the composition of cosponsors according to the partisanships of female cosponsors will

also broaden our understanding of such relationship in the Asian context.

Hypothesis 10: Bills sponsored by female legislators from different political parties are

more likely to be passed, while all other things being equal (ceteris paribus).

In the following two empirical chapters based on the time-series cross-national data on 162

countries from 1908 to 2016, the relationship between women’s descriptive representation in

parliaments and the adoption and implementation of laws and public policies addressing

violence against women will be examined. In the last empirical chapter on the South Korean

case study, hypotheses related to bill sponsorship and the success of bills about violence

against women will be tested.

Given the fact that even the basic link between women’s descriptive and substantive

representation in terms of violence against women laws and public policies still has not been

empirically examined thoroughly with the data from the start of such laws and policies, this

dissertation only focuses on testing theories and hypotheses about how the gender

composition of legislative branches and gender of individual legislators affects legal reforms

and legislative behaviors on the issue of violence against women respectively.

However, I should note that previous literature on violence against women has

examined crucial roles of domestic feminist autonomous movements and transnational

feminist mobilization, and transnational advocacy and transnational actors such as the

European Union.

For instance, as mentioned in the previous section, Htun and Weldon (2012) with cross-

national policy data on 70 in four years demonstrate that strong feminist autonomous

movements tend to lead countries to promote legal reforms concerning violence against

women. They argue that feminist movements can raise the issue of violence against women

from the marginal issue to the central, and make it possible to challenge male privilege in

sexual matters (see also, Htun and Weldon 2018). In addition, Medie (2013), by examining

the Liberian women’s movement, argues that women’s NGOs with political and material

support from international supports can make an impression on the implementation process of

laws criminalizing rape and other forms of violence against women (see also, Medie (2020)

and Anyidoho, Crawford, and Medie (2020)).

Moreover, transnational advocacy and actors, and networks have also played crucial

roles to lead countries to pay attention to the issue of violence against women (Montoya 2008,

2013). For instance, by examining data on the Daphne program which is the primary

mechanism for the European Union to handle the problem of violence against women,

Montoya (2008) based on a network analysis shows how transnational cooperation provides

venues for domestic organizations to improve their strength to address violence against

women. More recently, Montoya (2013, 90) emphasizes that the role of World Conferences

on Women (especially the conference in Beijing in 1995) in raising the issue of violence

against women among European countries, and that the important roles of European Union

actors to facilitate transnational cooperation against the issue of violence against women.

Even though this dissertation does not extensively focus on those explanations in the large-N

cross-national time-series analyses, I take into account the influences of those factors on the

adoption and implementation of violence against women laws and public policies by

including relevant explanatory variables.

CHAPTER 4

ADOPTION OF LAWS ADDRESSING

VIOLENCE AGAINST WOMEN

Given that criminal law is the very basic tool to control violence, protect individual

rights, and maintain social justice (Harel 2014), unveiling the impacts of the presence of

women in legislative branches on the adoption of violence against women laws should be the

first step to examine the influence of women’s descriptive representation on legal reforms

combating violence against women. Even though interpretations of laws sometimes depend

on social norms (Macnamara & Sagarin 1977), the absence of laws prevents us from

discriminating whether a certain behavior is penalizable or not.

Does a higher percentage of women in legislative branches lead countries to adopt

laws addressing violence against women? To answer this academic question and test the

generalizability of long-lasting theories arguing that the improved women’s descriptive

representation tends to promote policy changes for women’s rights, this chapter conducts a

time-series cross-national analysis on 162 countries from 1908 when Japan first adopted laws

combating sexual violence to 2016. The data coverage is determined by data availabilities of

different time-series cross-national datasets and databases.

The rest of this chapter proceeds with the following orders. First, dependent,

independent, and control variables employed in this article to test the association between

women’s descriptive representation in legislative branches and adoption of laws addressing

violence against women will be introduced with coding rules and primary data sources. Then,

modeling strategies and empirical results will be followed. In addition, empirical results from

alternative model specifications will be presented to evaluate whether the results from the

main model are sensitive to model specifications or not. Finally, I conclude.

4.1. Data

4.1.1. Dependent Variable: Adoption of Laws Addressing Violence Against Women

To examine the impacts of women’s descriptive representation, the adoption of laws

addressing violence against women is employed as a dependent variable. To build this

dependent variable, I rely on the Global Database on Violence against Women developed by

the UN Women.

The Global Database provides detailed information about laws addressing violence

against women in the UN member states from 1908 to 2016. In addition to the information

about when a country adopts laws about violence against women, a brief summary and

primary source of each violence against women law are presented. Furthermore, the Global

Database categorizes such law by various forms of violence against women: 1) Feminicide, 2)

Forced Sterilization, 3) Stalking, 4) Trafficking, 5) Violence Against Women and Girls

(without clauses targeting certain forms of violence, 6) Domestic and Intimate Partner

Violence, 7) Sexual

Violence, 8) New Forms of Violence Against Women, 9) Sexual Harassment, 10) Female

Genital Mutilation, and 11) Child, Early and Forced Marriage. I assign 1 to a country-year

observation when a country adopted laws targeting any forms of violence against women in a

given year. Otherwise, 0 is assigned.

Figure 4-1: Maps for Countries with/without Violence Against Women Laws

(A) 1980

(B) 2016

Note: Figure 4-1 is visualized based on the information from the Global Database on

Violence against Women.

Figure 4-1 illustrates countries with and without laws about violence against women

in 1980 and 2016. As you can see, only 17 countries had laws about violence against women

in 1980. The number of countries with violence against women laws has increased

dramatically since 1980 regardless of geopolitical regions, and it reaches 164 in 2016. It

demonstrates that there is a non-negligible variation in the dependent variable.

4.1.2. Independent Variable: Percentage of Women in Parliaments

To examine the influence of women’s descriptive representation in legislative

branches on the adoption of laws addressing violence against women, I employ the

percentage of women in parliaments. This decision is based on the fact that the percentage of

women in parliaments has been widely used to measure women’s descriptive representation

(Coffé & Reiser 2021;

Htun & Weldon 2012; Smith, Reingold, & Owens 2012; Taflaga & Beauregard 2020). The

Inter-Parliamentary Union (IPU) database is used to construct this variable. The IPU database

provides data about the percentage of women in parliaments over decades with information

about when the first female legislator was elected in each country. To fill the missing

countryyear observations, I assign 0 to country-year observations before a country had its first

female legislator in legislative branches. Moreover, if there are missing values between

election terms, the percentage of women in parliaments of the previous election year is used

to fill in the missing values.

4.1.3. Control Variables

To evaluate the independent influence of women’s descriptive representation in parliaments

and to prevent the beta coefficient estimate of the independent variable from being biased due

to the omission of relevant variables, I include a series of control variables.

The Level of Economic Development

First, the level of economic development is controlled. Some previous literature on

the modernization theory has argued that economic development tends to lead countries to put

more emphasis on women’s rights and integrity (Doepke, Tertilt, & Voena 2012; Inglehart,

Norris, & Ronald 2003; Inglehart & Welzel 2005; Wilensky 1974). To control the positive

influence of economic development, this dissertation includes GDP per capita (real GDP per

capita in 2011$) as one of the control variables, using the Maddison Project Database (Bolt,

Inklaar, Jong, & Zamden 2018). This database provides historical information on comparative

economic performances of 169 countries up to the present.

In addition to the economic development, the level of democracy also has been

regarded as an important determinant of legal reforms to promote women’s rights and the

empowerment of women (Haggard & Kaufman 2008; Orenstein 2008; Ottaway & Ottaway

2004). With the focus on free and fair elections and political competition (Donno, Fox, &

Kaasik 2021), women’s rights are indispensable bundled norms from democracy based on

democratic equality and inclusion (Inglehart et al. 2003; Tremblay 2007).

The Level of Democracy

To parcel out the conducive impacts of democracy on the adoption of violence

against women laws, I rely on the Polyarchy Index developed by the Varieties of Democracy

(V-Dem) dataset (Coppedge et al. 2018). Given that the selection of measures for the level of

democracy can lead to totally different empirical results and inferences (Högström 2013;

Teorell, Coppedge, Skaaning, & Lindberg 2016), selecting appropriate measurements of the

level of democracy should be based on concrete reasons. In this dissertation, the Polyarchy

Index is employed instead of other time-series cross-national datasets regarding democracy

such as the Freedom House Index and the Polity Score. The Polyarchy Index covers a wide

range of country-year observations. For instance, while the Polyarchy Index covers from 1789

to the present, the Freedom House Index covers only from 1972. Furthermore, both the

Freedom House Index and the Polity Score have been criticized because of their

methodological performances. The Freedom House Index has been questioned by scholars

due to its bias and the lack of consistency, and the Polity Score is likely to overestimate the

level of democracy in countries (Knutsen et al. 2019). Therefore, the Polyarchy Index is

employed rather than the

Freedom House Index and the Polity Score.

Women’s Participation in Civil Society Organizations and Women’s Property Rights

Two women-related factors, women’s participation in civil society organizations

(CSO) and women’s property rights, are the two important variables requiring attention. Those

two factors have been emphasized as crucial factors positively influencing women’s rights and

integrity, and legal reforms for women. Some previous studies have demonstrated that the

improved women’s property rights are likely to increase women’s political efficacy and to lead

governments to take legal actions related to women’s issues (Mishra & Sam 2016; O'Hara &

Clement 2018; Pradhan, Meinzen-Dick, & Theis 2019), while others have argued that

women’s active participation in civil society organizations or feminist autonomous movements

make countries adopt progressive policies combating violence against women (García-Moreno

et al.

2015; Htun & Weldon 2012; Weldon & Htun 2013). Especially, Htun and Weldon (2012)

based on cross-national data in four years (1975, 1985, 1995, and 2005) emphasized the

relative importance of feminist autonomous movements compared to other women-related

variables.

To control the potential impacts of the two women-related factors, the indices for

women’s participation in civil society organizations and women’s property rights from the

VDem dataset are used. It should be noted that there are no available and reliable time-series

cross-national data to measure the feminist autonomous movement. Given that the

Spearman’s rho between the women’s participation in civil society organization index from

the V-Dem and the feminist autonomous movement measure from Htun and Weldon (2012) is

0.638 and statistically significant at the level of p < 0.001, Women’s CSO participation index

is used as a proxy measure for feminist autonomous movement.

The Ratification of the CEDAW

Furthermore, a binary variable indicating whether a country is a member state of the

Committee on the Elimination of Discrimination Against Women (CEDAW) is included as a

control variable. Adopted by the United Nations General Assembly on December 18, 1979,

the CEDAW has played an important role in dispersing women’s rights agenda including

violence against women over the world (True 2016). The CEDAW is the only international

human rights treaty aiming to handle the issue of sex and gender-based discrimination and to

improve women’s rights (Hellum 2021; Hellum & Aasen 2013). Even though the problem of

violence against women was not indicated as a priority for action when the CEDAW was

presented at Copenhagen in 1980 (Keck & Sikkink 1998; Weldon 2006a), the CEDAW has

expressed its concerns over the problem of violence against women through general

recommendations

(Qureshi 2020; Zada 2021). Especially based on the General Recommendation No.12 and

No.19, the CEDAW urges member states to submit periodic reports for legal measures taken

to combat violence against women (McQuigg 2007). Along with some pessimistic

quantitative literature arguing the null effects of the ratification of the CEDAW on promoting

women’s rights, some studies have demonstrated that the CEDAW has non-negligible impacts

on improving women’s rights and integrity (Gray, Kittilson, & Sandholtz 2006; Hill Jr 2010).

Thus, with an expectation that the ratification of the CEDAW increases the

probabilities of adopting violence against women laws, I control the membership of the

CEDAW. Based on the membership information of CEDAW provided by the United Nations

Human Rights Office of the High Commissioner, 1 is assigned to a country-year observation

after the country ratified CEDAW. Otherwise, 0 is assigned.

The Proliferation of Global Norms about Women’s Rights

In addition, the effect of the proliferation of global norms about women’s rights is also

controlled for. Individual countries follow international human rights norms to avoid the

pressure of being accused of violating those norms, which can lead to the imposition of

economic sanctions. Moreover, a strong linkage between domestic and transnational actors

generated from the proliferation of global norms related to VAW can also make countries

adopt and implement laws addressing VAW (J. Fitzpatrick 2012; Risse et al. 1999). The

percentage of countries ratifying the CEDAW treaty, the most prominent international treaty

related to women’s rights (Englehart & Miller 2014; Och 2018), is used to control for the

anticipated positive influence of the proliferation of global norms about women’s rights.

Legislature Corrupt Activities

Given that legislative rules directly affect policy adoptions and who benefits from legal

reforms (Cox 2006; Cox & McCubbins 2007), legislative corrupt activities and their impacts

on legislative effectiveness have been continuously discussed (Herzfeld & Weiss 2003;

McMann, Seim, Teorell, & Lindberg 2020; Warren 2006; Yadav 2012). Among comparative

political scientists, it has been regarded that legislative corrupt activities tend to decrease

institutional efficacy (Fish, Michel, & Lindberg 2015; Karklins 2002). For instance, Karklins

(2002, 32) mentions that “By distorting the purpose of public institutions, corruption

undermines good governance and democracy, both in the minds of the citizenry that no longer

trusts state and political institutions”.

Therefore, to control the anticipated negative impacts of legislature corrupt activities

on the adoption of violence against women laws, I use the legislature corrupt activities index

developed by the V-Dem. It measures legislature corrupt activities based on the behaviors of

1) “accepting bribes”, 2) “helping to obtain government contracts for firms that the legislator

own”, 3) “doing favors for firms in exchange for the opportunity of employment after leaving

the legislature”, and 4) “stealing money from the state or campaign donations for personal

use.”

Even though other cross-national measures such as the Transparency International’s

Corruption Perceptions Index (CPI) and the index from the Freedom House’s “Nations in

Transit” project are available, it is revealed that these indicators are prone to measurement

error (Fish et al. 2015). In addition, the ordinary nature of these variables does not allow us to

capture the magnitude of corruption (Svensson 2005).

Reserved Seats Gender Quotas

After the introduction of gender quotas in Latin American and African countries, many

countries have implemented legislative gender quotas to promote women’s substantive

representation (Hillman 2018; Tan 2015). With the increase in the number of countries with

legislative gender quotas, previous studies have broadened our understanding of the

conducive impacts of gender quotas on various political phenomena including the improved

legislative diversity (Barnes & Holman 2020), women’s political efficacy, and engagement

(Franceschet, Krook, & Piscopo 2012; Kerevel & Atkeson 2017; Zetterberg 2009), and even

international reputation (Bush & Zetterberg 2020).

Especially, the association between the introduction of legislative gender quotas and

women’s substantive representation has been widely examined by literature on gender quotas

(Clayton & Zetterberg 2018; Franceschet & Piscopo 2008; Krook 2020a; Woo 2021b). The

findings on the impacts of gender quotas are mixed at best. Some scholars demonstrate that

female legislators are more likely to propose bills related to women’s rights after the

introduction of legislative gender quotas (Barnes 2012; Franceschet & Piscopo 2008), while

others find that the adoption of legislative gender quotas does not change legislative

behaviors of female legislators (Clayton, Josefsson, & Wang 2017; Kerevel & Atkeson 2013),

or sometimes negatively influence women’s substantive representation in terms of policy

outcomes (Htun, Lacalle, & Micozzi 2013).

Given this academic controversy about the impacts of legislative gender quotas, I

control the influence of reserved seat gender quotas to examine the independent influence of

women’s descriptive representation. The binary variable named Reserved Seats Gender

Quotas is built based on the information about individual countries’ legislative gender quotas

from the Gender Quotas Database from the International Institute for Democracy and

Electoral Assistance (IDEA). I assign 1 to the binary variable if a country has a reserved seat

gender quota.

The Ruling Party’s Political Ideology

The ruling party’s political ideology in terms of left and right scale is also included as

one of the control variables. This decision is based on the conflicting arguments and findings

of whether the political ideology of a ruling party has a meaningful influence on the

promotion of legal reforms for women’s rights and integrity. Some studies have supported the

important role of a ruling party’s political ideology in determining the probabilities of taking

legal reforms promoting women’s rights (Blofield, Ewig, & Piscopo 2017; Hieda 2013;

Mazur 2002). On the contrary, other scholars have empirically demonstrated that the political

ideology on the leftright scale is not associated with the adoption of feminist policies

including violence against women laws (Beer 2017; Weldon & Htun 2013). In this

dissertation, rather than disregarding this controversial variable and taking the risk of omitted

variable bias, I control the ruling party’s political ideology based on the inclusion of a binary

variable indicating whether a ruling party is left-oriented or not. To construct this variable, the

classification from the Quality of Government (QOG) database (Teorell, Charron, Samanni,

Holmberg, & Rothstein, 2010) is employed.

International and Civil Conflicts

The presence of international and civil conflicts in a country is controlled to parcel

out the negative impacts of those conflicts in legal reforms addressing the problem of

violence against women. It is based on the rationale that when a country is involved in

international or civil conflicts which intimidate national security, the country would be

expected to pay less attention to the problems of violence against women and legal reforms

promoting women’s rights. I assign 1 to a country-year observation when a country

experiences internal or international armed conflicts.

Government System

The potential impacts of different government systems of countries are also

considered. Some scholars argue that parliamentary systems are more efficient to provide

public goods and promote women’s rights than the presidential system (Bohlken 2018; Persson

& Tabellini 2005). Moreover, even though legislative branches play a considerable role in

policy-making and agenda-setting in some presidential and semi-presidential systems

(Schwindt‐Bayer 2006), some might argue that the existence of an additional veto player in

presidential systems and the presidential agenda setting can influence the adoption of laws

concerning violence against women. Thus, to parcel out the influence of government systems,

I categorize countries’ government systems into 1) Presidential, 2) Semi-Presidential, and 3)

Parliamentary by following the classification from the QOG database

Legal System

Legal systems of countries are controlled to isolate the influence of women’s descriptive

representation on the adoption of violence against women laws from potential impacts of

different legal systems. According to legal systems, the procedures of adopting new laws and

legal reforms are different, and the different procedures or characteristics of various legal

systems might affect the tendency of legal reforms. Some previous studies have demonstrated

that the probabilities of adopting new policies or laws are determined by legal systems. For

instance, Zehri and Chouaibi (2013) find that countries with the English Common Law legal

system are more likely to adopt the International Account Standards (IAS) and the

International Financial Reporting Standards compared to other legal systems.

Thus, to control the potential impacts of legal systems on the adoption of violence

against women laws, each country’s legal system is controlled based on the classification of

the legal system from the QOG database. The QOG database divides legal systems of

countries into five categories: 1) English Common Law, 2) French Commercial Code, 3)

Socialist/Communist Laws, 4) German Commercial Code, and 5) Scandinavian Commercial

Code.

Geopolitical Regions

Furthermore, to control unobserved and unmeasured factors related to geopolitical

regions such as cultural aspects related to violence against women, I control each country’s

geopolitical regions. In addition, the regional fixed variables allow us to control region-

specific characteristics such as the presence of regional organizations and agreements

combating violence against women. Following the classification of geopolitical regions from

the QOG database, I assign countries into the eight geopolitical regions: 1) Western Europe

and North America (including Australia and New Zealand), 2) Latin America (including

Cuba, Haiti, and the Dominican Republic), 3) the Middle East and North Africa (including

Israel, Turkey, and Cyprus), 4) Sub-Saharan Africa, 5) Eastern Europe and post Soviet Union

(including Central Asia), 6) East Asia (including Japan and Mongolia), 7) South-East Asia,

and 8) South Asia.

Regional Diffusion

Lastly, the percentage of countries adopting laws concerning violence against women is

included to take into account the regional diffusion mechanism. It has been demonstrated that

policy diffusion is likely to occur among countries in the same geo-political region. The

regular contacts in international regional organizations including both political and economic

organizations make countries learn or emulate other countries’ policies (Berry & Berry 1999;

Boushey 2010). In addition, feminist NGOs also can take ideas for laws concerning violence

against women from neighboring countries and lead governments to take legal reforms

combating violence against women (Htun & Weldon 2012). Thus, I include the percentage of

countries with violence against women laws in each geo-political region defined previously to

isolate the influence of women’s descriptive representation in legislative branches from the

regional diffusion mechanism.

Given that socio-economic and political factors are not translated into legal reforms in

a short period, independent and control variables are lagged by one year. Descriptive statistics

of dependent, independent, and all control variables are presented in Table 4-1a in Appendix.

4.2. Empirical Analysis

In this Empirical Analysis section, the selection of modeling strategy and empirical

results will be presented. In addition, a robustness check will be followed.

4.2.1. Modeling Strategy

Given that the dependent variable is the adoption of laws addressing violence against

women, this dissertation relies on the discrete event history model to examine the influence of

women’s descriptive representation in parliaments on the probability of adopting violence

against women laws. Event history analysis is the standard approach to analyze factors behind

policy diffusion or policy innovation (Kreitzer & Boehmke 2016). A great number of previous

studies have applied event history analysis to examine driving factors of the adoption of

various policies and laws such as stay-at-home orders (Shay 2020), social distancing policies

(Adolph,

Amano, Bang-Jensen, Fullman, & Wilkerson 2021), public communication reforms

(EckerEhrhardt 2018), and smoking bans (Pacheco 2012).

Among various types of event history models such as the Cox proportional hazard

model and other parametric models, a discrete event history model based on logistic

regression is employed because of the fact that the available data on the adoption of violence

against women laws follows the discrete-time process (observations year by year). Discrete

event history models estimate the probability of an event occurrence at a given time t and the

estimation can be expressed as below (Box-Steffensmeier & Jones 1997).

(t) = P(E=t | Tt; , `X)

After the individual unit experiences an event, the individual will be dropped from the

dataset. If we are interested in repeated events, we do not need to drop the observation with

the experience of an event. The above formula can be transformed as below.

ln ( (t) ) =  + `X + .

1−(t)

Compared to Cox proportional hazard models, one of the advantages of a discrete event

history model is that the discrete event history model is not sensitive to the existence of tied

events since it is not based on partial likelihood. In other words, the order of events does not

matter for estimating coefficients of independent variables. Moreover, the interpretation of

coefficients can be easily translated into predicted probabilities. The unit of analysis is a

country-year and, given that Japan firstly adopted a law penalizing sexual violence in 1908,

the risk set is composed of country-year observations from 1908. Consistent with event

history analysis, if a country adopts a law combating violence against women, it is dropped

from the

risk set.

To isolate the influence of women’s descriptive representation on the adoption of violence

against women laws from duration dependency, I include time polynomial functions.

Compared to time fixed effects and splines, time polynomial functions allows us to avoid

inefficiency and perfect prediction problems that can be generated from time fixed effects,

and to avoid arbitrary selection of knots, and we can model any hazard shapes recovered by

parametric models and cox models by using time polynomial functions (Carter & Signorino

2010). Even though time variables will be highly correlated and introduce multicollinearity in

an estimated model, Carter and Signorino (2010) show that multicollinearity is not a problem

for a large data set and time cubic functions outperform time splines and time fixed effects

based on the results from Monte Carlo simulation. To be specific, they test biasness and

efficiency of estimates from models with time fixed effects, time splines, and time cubic

functions. Under the existence of constant, decreasing, increasing, and non-monotonic

duration dependency with 2,000 samples and 10,000 Monte Carlo iterations, they find that the

estimates from the models with time cubic functions recover duration dependency more

precisely than the other two models with time fixed effects and time splines. Moreover, they

argue that when the number of observations is large enough the multicollinearity problem

does not a big problem. Time stands for the years after the first adoption of violence against

women law by Japan in 1908.

4.2.2. Empirical Results

Figure 4-2 presents the point estimates of variables with 95% confidence intervals based on

the estimations from the discrete even history model based on logistic regression. The

variance inflation factor (VIF) test was conducted to check whether there is a

multicollinearity problem in the model. The mean VIF is 2.48 and individual VIFs are less

than 4 except time polynomials, indicating there is no problematic correlation among

independent and control variables.3 To deal with the potential heterogeneity problem in error

terms, I estimate robust cluster standard errors by countries instead of normal standard errors.

Consistent with Kanter’s Critical Mass theory and the politics of presence theory, the key

independent variable, Women in Parliaments (%), is statistically significant at the level of p <

0.05 with a positive direction as expected. It means that a higher percentage of women in

3 Even though there are no golden standards about the threshold of VIF to assure that there is

legislative branches tends to lead countries to adopt laws concerning violence against women.

To be specific, the estimated coefficient of Women in Parliaments (%) is 0.024, meaning that

one unit increase in the percentage of women in parliaments increases the log-odds of the

adoption of laws addressing violence against women by 0.024. In the line with previous

studies demonstrating the positive association between women’s descriptive and substantive

representation, this empirical result supports the arguments that the improved women’s

descriptive representation is likely to be followed by legal reforms for women’s rights and

integrity.

Turning to control variables, the level of economic development and democracy

measured by GDP per capita and the Polyarchy Index (V-Dem), and CEDAW (%) have

statistically significant relationships with the probability of adopting violence against women

laws at the level of p < 0.05. One unit increase in those three variables increases the log-odds

of introducing laws about violence against women by 0.079, 1.453, and 0.033 respectively.

These results imply that previous findings on the conducive roles of economic development,

democracy, and the proliferation of global norms for women’s rights in promoting women’s

rights can be applied to explaining variations of the adoption of violence against women laws

among countries. Furthermore, the Regional Diffusion (%) is statistically significant at the

level

no multicollinearity problem, VIFs less than 4 have been regarded as an indication of no

consequential collinearity. Please refer Mela and Kopalle (2002), James, Witten, Hastie, and

Tibshirani (2013), Akinwande, Dikko, and Samson (2015), and Zaman et al. (2015) for more

information about multicollinearity and VIF test.

of p < 0.01, indicating that a higher percentage of countries with violence against women

laws in the same geo-political region leads countries to adopt such laws.

Figure 4-2: Estimations from Discrete Event History Model

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 4-2a in

Appendix for the detailed information about the result. In this figure, Women in Parliaments

(%), CEDAW (%), and Regional Diffusion (%) are presented as their proportion for

visualization sake.

Also, there are interesting findings in terms of the three categorical variables:

Government System, Legal System, and Geopolitical Regions. Related to Government System,

countries with a semi-presidential system are less likely to adopt violence against women

laws compared to countries with a presidential system. In terms of Legal System,

Scandinavian Commercial Code is statistically significant at the level of p <0.01 with a

negative sign, indicating that countries having Scandinavian Commercial Code, compared to

countries with English Common Law, tend to have a lower probability of adopting laws

addressing violence against women. Also, it is demonstrated that Latin American countries

have a lower probability of adopting violence against women laws compared to countries

located in Western Europe and North America.

To evaluate whether the influences of statistically significant variables are also substantially

meaningful or not, predicted probabilities of adopting laws concerning violence against

women are estimated by the different values of the variables.

According to the upper-left panel of Figure 4-3, the predicted probability of adopting laws

about violence against women is 1.492% when there are no female legislators. It increases to

1.855% if 10% of legislative seats are filled with female representatives. When the percentage

of women in parliaments reaches 20% and 30%, the predicted probabilities of introducing

violence against women laws are 2.297% and 2.829% respectively. Despite a smaller number of

country-year observations, the predicted probabilities are 4.216% and 5.095% when the

percentage of women in parliaments increases to 50% and 60%.

Figure 4-3: Predicted Probabilities of Adopting Violence Against Women Laws

Note: Predicted Probabilities are presented with 95% confidence intervals with the kernel or

normal density of each variable. Predicted Probabilities are estimated while all other variables

are held at their observed values.

The level of economic development and democracy have also substantial impacts on the

probability of introducing laws addressing violence against women. In terms of GDP per

capita, the predicted probabilities are less than 5% if a country has a GDP per capita lower

than

5. The predicted probabilities exceed 5% when GDP per capita is higher than 20. When the

Polyarchy Index increases from 0 to 0.5, the predicted probabilities increase from 1.043% to

1.916%. At the extreme, the predicted probability reaches over 3%. In addition, the impacts

of CEDAW (%) and Regional Diffusion (%) on the predicted probabilities of adopting

violence against women laws are also substantially significant as presented in Figure 4-3.

Figure 4-4 presents the predicted probabilities of adopting violence against women laws with

95% confidence intervals. As you can see, countries with a semi-presidential government

system and Scandinavian Commercial Code have substantially less predicted probabilities of

adopting laws about violence against women. To be specific, the predicted probability in

countries with a semi-presidential system is 1.285%. Considering the predicted probabilities

in countries with presidential and parliamentary systems are 2.261% and 1.803%, the

negative impact of a semi-presidential system on the predicted probability is substantially

meaningful. In terms of Legal System, the predicted probabilities of adopting such laws from

countries having an English Common Law system and countries with a Scandinavian

Commercial Code are 1.791% and 0.483%. Also, the predicted probability of introducing

violence against women laws in Latin American countries is 0.890%. Given that the predicted

probability in Western European and North American countries is 2.265%, Latin American

countries are substantially less likely to adopt laws addressing violence against women.

Figure 4-4: Predicted Probabilities based on the Categorical Variables

Figure 4-4—continued

Note: Predicted Probabilities are presented with 95% confidence intervals with the normal

density of each variable. Predicted Probabilities are estimated while all other variables are

held at their observed values.

To sum up the empirical results from the discrete event history model, it has been

revealed that women’s descriptive representation in legislative branches measured by the

percentage of women in parliaments is not only statistically significant, but also substantially

meaningful.

The empirical analysis is replicated with the interaction term between Women in Parliaments

(%) and CEDAW (%) to examine whether the percentage of women in legislative branches

magnified the positive influence of the proliferation of global norms about women’s rights

and integrity. Figure 4-5 demonstrates that the interaction term is not statistically significant

at any level of p-value, indicating that the influence of CEDAW (%) is not conditioned on the

women’s descriptive representation in parliaments and vice versa.

Figure 4-5: Estimations with the Interaction Term

between Women in Parliaments (%) and CEDAW (%)

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 4-3a in

Appendix for the detailed information about the result.

In terms of control variables, the empirical results and inferences are similar to those of

Figure 4-2. As in Figure 4-2, GDP per capita, the Polyarchy Index (V-Dem), Regional

Diffusion (%) have statistically significant relationships with the log-odds of the adoption of

laws addressing violence against women at the level of p < 0.05. Also, Semi-Presidential,

Scandinavian Commercial Code, and Latin America are statistically significant at the level of

p < 0.05 with negative signs. Even if not presented, the influences of the statistically

significant explanatory variables are substantially meaningful.

Figure 4-6: Average Marginal Effects of CEDAW (%) and Women in Parliaments (%)

Note: Average marginal effects are presented with 95% confidence intervals with kernel density

of Women in Parliaments (%) and CEDAW (%). Predicted Probabilities are estimated while all

other variables are held at their observed values.

Given that “[t]he analyst cannot even infer whether X has a meaningful conditional effect on

Y from the magnitude and significant of the coefficient on the interaction term”

(Brambor, Clark, & Golder 2006, 74), I graph estimated average marginal effects of CEDAW

(%) and Women in Parliaments (%) based on the values of the other variable respectively.

Even in terms of average marginal effects, it turns out that the influence of Women in

Parliaments (%) is not conditioned on CEDAW (%) and vice versa. Although the point

estimate of the average marginal effect of CEDAW (%) increases with the increase in Women

in Parliaments (%), the estimates are not statistically different from each other. In the right

panel, the estimated average marginal effects are not statistically different from 0.

4.2.3. Robustness Check

Considering that the importance of a robustness check with alternative model

specifications has been widely emphasized by scholars over decades (Esarey & Danneman

2015; Neumayer & Plümper 2017), I conduct a robustness check to evaluate whether the

empirical findings in the discrete logistic event history models are sensitive to model

specifications or not.

Some previous studies have demonstrated that not all forms of violence against

women are identical (Bouché & Wittmer 2015; Keefer 2006; Vives-Cases, Ruiz-Cantero,

EscribàAgüir, & Miralles 2011; S. A. Welch 2016). Especially related to the influence of

women’s descriptive representation on the adoption of violence against women laws, O'Brien

(2010) argues that trafficking of women and girls should be regarded as one of the national

security issues and that Female Genital Mutilation is a religious issue. Therefore, the impacts

of women’s descriptive representation are relatively less conspicuous on the adoption of laws

about trafficking and Female Genital Mutilation.

Thus, to control the potential heterogeneous impacts of women’s descriptive

representation in parliaments on the adoption of violence against women laws according to

the various forms of violence against women, I conduct a pooled event history analysis

(PEHA) developed by Kreitzer and Boehmke (2016). In the pooled event history model, the

number of observations is multiplied by the eleven forms of violence against women, which

increases the degree of freedom in the model and, in turn, allows me to get more precise

estimates about coefficients of independent and control variables (Boehmke 2009). Moreover,

estimating the pooled event history model is more parsimonious than several separate event

history models by the eleven forms of violence against women and enables me to catch the

potential commonalities of the eleven forms of violence against women. For instance, the

adoption of laws addressing sexual harassment and sexual violence can be a function of

women’s descriptive representation in parliaments.

The unit of analysis in the pooled event history model is “country-forms of violence

against women-year”. Countries are not dropped out of the risk set until they adopt all laws

concerning the eleven forms of violence against women. In other words, countries remain in

the risk set if there are unadopted laws related to any forms of violence against women4. As in

previous estimated models, Time and Time Squared variables are included to model potential

linear and nonlinear time trends. In the pooled event history model, Time stands for years

after the first adoption of laws according to each form of violence against women (e.g., see

Table 46a in Appendix for the sample dataset).

According to Kreitzer and Boehmke (2016, 125), there are three potential approaches

to examine or to control heterogeneity across different forms of violence against women that

each law aims to penalize: 1) including fixed effects, 2) adjusting standard errors by

clustering the errors within the eleven forms of violence against women, 3) estimating

multilevel models

4 Please see Boehmke (2009, 235) for more detailed explanation on the risk set in the pooled

event history model.

with random effects. In this dissertation, I estimate pooled event history models based on

multilevel models with random effects of forms of violence against women and random

coefficients of Women in Parliaments (%).

This decision is based on the following rationale. Even though the inclusion of fixed

effects by forms of violence against women imposes fewer assumptions on the distribution of

the heterogeneity and the form-specific errors are not needed to be independent of the

explanatory variables, fixed effects model specification might introduce multicollinearity

problems (Kreitzer and Boehmke 2016, 126). Especially, given that some explanatory

variables such as GDP per capita and Polyarchy Index (V-Dem) are changing slowly, the

inclusion of binary indicators by the eleven forms of violence against women brings a high

level of multicollinearity (Cheah 2009). Moreover, although estimating clustered standard

errors have been widely employed in the application of the pooled event history analysis

(Boehmke 2009; Makse & Volden 2011), it does not enable me to examine and to control the

heterogenous impacts of women’s descriptive representation on the adoption of laws

concerning violence against women by forms of such violence.

Model 1 and Model 2 are estimated without and with the interaction term between

Women in Parliaments (%) and CEDAW (%). The empirical results from Figure 4-2 and

Figure

4-5 are consistent with the results from Figure 4-7. The independent variable, Women in

Parliaments (%), in Model 1, is statistically significant at the level of p < 0.05, while the

interaction term in Model 2 is not statistically significant at the level of p < 0.05. These

results demonstrate that the previous empirical results about the conducive impact of

women’s descriptive representation in parliaments are not sensitive to model specifications.

Even though the substantive influence of the independent variable is not presented, Women in

Parliaments (%) are substantially meaningful in terms of predicted probabilities. Even in a

repeated failure context (Table 4-5a in Appendix) with the consideration of the number of

adopted laws by 11 forms of violence against women, Women in Parliaments (%) is

statistically significant at the level of p < 0.05, while the interaction term is not statistically

significant at any level of pvalues.5

Related to control variables, GDP per capita is not anymore a statistically significant

determinant of probabilities of adopting violence against women laws, while the level of

democracy and Regional Diffusion (%) is statistically significant at the level of p < 0.01.

Interestingly, the adoption of reserved seat gender quotas is statistically significant in both

Model 1 and Model in Figure 4-7. Even though Reserved Seats Gender Quotas is not

statistically significant in Figure 4-2, the results from Figure 4-7 provide partial supports for

the argument that the adoption of reserved seat gender quotas tends to improve women’s

substantive and symbolic representation by increasing women’s political participation (Ciftci

& Bernick 2015) and awareness of gender rights (Htun & Jones 2002), while contradicts the

studies arguing the null or negative effects of gender quotas on the politicization of women’s

issues (Franceschet et al. 2012; Josefsson 2014).

5 I should note that countries remain in the analysis if they have not adopted laws addressing

11 forms of violence against women. For more information about the comparison between

pooled event history analyses in the first event context and in the repeated events contxt, see

Boehmke (2009).

Figure 4-7: Estimations from Pooled Event History Models

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 4-4a in

Appendix for the detailed information about the result.

Furthermore, compared to countries with a presidential system, countries with

semipresidential and parliamentary systems are less likely to adopt violence against women

laws. In addition, unlike the results from Figure 4-2 and Figure 4-5, French Commercial Code

turns out to be statistically significant variables decreasing the log-odds of introducing

violence against women laws. The random effect according to the eleven forms of violence

against women is not statistically significant at any level of p-value, meaning that the

influence of women’s descriptive representation in parliaments on the adoption of laws

addressing violence against women is not heterogenous according to the various forms of

violence against women.

4.3. Conclusion

In Chapter 4, it has been demonstrated that a higher women’s descriptive

representation in legislative branches tends to lead countries to adopt laws concerning

violence against women. With the unprecedentedly large time-series cross-national data on

162 countries from 1908 to 2016, I find that the relationship between the percentage of

women in parliaments and the adoption of violence against women laws is not only

statistically significant in the discrete event history models, but also substantially meaningful

in terms of predicted probabilities. However, unlike my expectation, the conducive impacts of

the proliferation of global norms about women’s rights and integrity measured with the

percentage of countries ratifying the CEDAW are not conditioned by a higher percentage of

women in parliaments, and vice versa. Those findings are consistently robust in other model

specifications such as the PEHA and models with different combinations of explanatory

variables.

CHAPTER 5

IMPLEMENTATION OF PUBLIC POLICIES CONCERNING

VIOLENCE AGAINST WOMEN

Does a higher women’s descriptive representation in parliaments also lead countries

to promote public policies dealing with the problem of violence against women? In Chapter 5,

the relationship between women’s descriptive representation in legislative branches and the

implementation of public policies concerning violence against women will be examined with

the time-series cross-national data on 162 countries from 1908 to 2016.

Even though the adoption of domestic statutory laws penalizing violence against

women is the most basic and the first step to eliminate violence against women, the

implementations of other public policies such as preventive measures, police training,

shelters, and social and medical assistance for the abused women have also been essential

parts to overcome the problem of violence against women (Avdeyeva 2007; Htun & Weldon

2012; Weldon 2006b; Weldon & Htun 2013; Woo 2021a, 2021b). Given that introducing

multisectoral public policies is essential to women’s and girls’ safety, protection, and

recovery, examining the influence of women’s descriptive representation in parliaments as a

driving factor behind the implementation of public policies about violence against women, in

addition to the adoption of violence against women laws, is a crucial task.

The rest of Chapter 5 is organized as follows. First, I will introduce a new dependent

variable to examine the impacts of women’s descriptive representation on the implementation

of public policies addressing violence against women. Next, a modeling strategy and

empirical results will be presented. Then, I will further examine why some countries without

female representatives in legislative branches adopted public policies and laws combating

violence against women. Also, the potential limitations of such policies and laws will be

reviewed.

Finally, based on the results from the time-series cross-national analyses in Chapter 4 and

Chapter 5, I discuss empirical findings regarding the influence of women’s descriptive

representation on women’s substantive representation in terms of the adoption of violence

against women laws and public policies.

5.1. Data

To examine the potential conducive impacts of women’s descriptive representation in

legislative branches on the implementation of public policies addressing violence against

women, I use a binary dependent variable indicating whether a country implements public

policies combating violence against women.

To build the new dependent variable, I again rely on the Global Database on Violence against

Women developed by the UN Women. The database provides information about when each

UN member state implemented public policies addressing violence against women according

to 20 types of public policies: 1) national action plan to address violence against women, 2)

budget allocation, 3) regulation, 4) referral service, 5) integrated service, 6) onestop center, 7)

shelter, 8) psychosocial counseling, 9) hotline, 10) longer term service, 11) health protocols

and 12) health training, 13) justice protocols and 14) justice training, 15) police protocols and

16) police training, 17) prevention campaign, 18) prevention community, 19) formal and 20)

non-formal education.

Given that various types of public policies addressing violence against women have

their important roles to protect the abused women of violence against women and to address

violence against women (García-Moreno et al. 2015; Htun & Weldon 2012; Woo 2021a,

2021b), I do not arbitrary select certain types of public policies addressing violence against

women. Thus, when a country implements any type of violence against women public

policies in a given year, I assign 1 to the country-year observation. Otherwise, I assign 0.

Figure 5-1 demonstrates the maps for countries with a number of public policies addressing

violence against women in 1980 and 2016. As presented, the number of countries with any

type of public policy was much smaller in 1980 than in 2016. To be specific, France was the

only country with five different violence against women public policies: integrated service,

health training, justice training, police training, and prevention campaign. In addition,

Australia, Gabon, India, and Thailand have two such policies.

Figure 5-1: Maps for Countries with the Number of Public Policies

Concerning Violence Against Women

1980

2016

Note: Figure 5-1 is visualized based on the information from the Global Database on

Violence against Women.

With the entrance of women in politics, many countries have introduced various

types of public policies handling the problem of violence against women. In 2016, countries

with a relatively higher percentage of women in parliaments such as Canada (26%), Australia

(26.7%), Argentina (34.6%), Denmark (37.4%), Norway (39.6%), Spain (41.1%), Iceland

(41.3%), Ecuador (41.6%), and Mexico (42.4%) have implemented more than 10 types of

public policies addressing violence against women. On the contrary, countries with a lower

percentage of women in parliaments (%) including Oman (1.2%), Iran (3.1%), Haiti (4.2%),

Republic of the Congo (7.4%), Egypt (2%), Japan (9.5%), Brazil (9.9%), the United States

(19.4%) have introduced less than 10 types of such policies. The two-sample t-test

demonstrates that the difference of means in the percentage of women in parliaments between

countries with 10-20 public policies and countries with less than 10 policies is statistically

significant at the level of p < 0.056.

5.2. Empirical Analysis

5.2.1. Modeling Strategy

Some might argue that the influence of women’s descriptive representation may differ across

various types of public policies addressing violence against women. For instance, providing

formal or non-formal educations about the problem of violence against women and

developing budgets to combat violence against women can be followed by different functions

6 In 2016, the mean Women in Parliaments (%) of countries adopting less than 10 public

policies about violence against women is 19.731, while that of that of countries implementing

10 to 20 policies is 24.313.

of women’s descriptive representation in parliaments. The influence of women’s descriptive

representation on the adoption of budgets addressing violence against women might be

smaller than on the implementation of other public policies, because the implementation of

budgets

specific to the problem of violence against women is more likely to be affected by the

economic development of a country.

Therefore, to examine the influence of women’s descriptive representation in

legislative branches on the implementation of various types of public policies addressing

violence against women with the consideration of the potential heterogenous impacts of the

percentage of women in parliaments, I employ a discrete pooled event history analysis

(PEHA) developed by Kreitzer and Boehmke (2016) as in the robustness check in Chapter 4.

To reiterate, the application of the pooled event history analysis increases the degree of

freedom in the model, allowing me to get precise coefficient estimates of explanatory

variables (Boehmke 2009). Also, it enables me to examine the potential commonalities of the

20 types of public policies and this approach is more parsimonious than estimating separate

discrete event history models by 20 types of policies.

In the pooled event history models estimated in this chapter, the number of country-

year observations is multiplied by the 20 types of public policies addressing violence against

women. Thus, the unit of analysis is “country (𝑖)-policy (𝑘)-year(𝑡)” and the dependent

variable is whether a country 𝑖 implements a policy 𝑘 in year 𝑡.7 Countries are included in the

risk set if they do not implement all 20 public policies concerning violence against women.

7 I should note that the explanation and notation on the pooled event history model is largely

based on Kreitzer and Boehmke (2016).

To be specific, countries are included in the risk set from the years when the first adoption of

each policy occurred in any country to the years when a country adopts all 20 public policies.

Time and Time Squared are employed to control potential duration dependency. Time stands

for years

after the first implementation of public policies according to each type of policies addressing

violence against women (e.g., see Table 5-4a in Appendix for the sample dataset).

As pooled event history models estimated in Chapter 4, I estimate multilevel models

with random effects for policies rather than including fixed effects or adjusting standard

errors by clustering the errors within the twenty types of public policies concerning violence

against women. To reiterate, the inclusion of fixed effects of policies might introduce

multicollinearity problems given that some explanatory variables are changing slowly over

the years, and adjusting standard errors will not enable me to examine or to control the

potential heterogeneous impacts of violence against women according to the twenty types of

violence against women public policies (Cheah 2009; Kreitzer & Boehmke 2016). The

application of multilevel models with the policies as the second-level group and the random

effect of the percentage of women in parliaments allows me to model the potential

heterogeneous effects of women’s descriptive representation in legislative branches by

different public policies addressing violence against women.

Below is the formula for the pooled multilevel event history model estimated in this chapter.

Again, 𝑖 represents a country, 𝑘 indicates types of public policies, and 𝑡 denotes a year.

0 𝑖𝑓 𝑌𝑖𝑘𝑡

𝑌𝑖𝑘𝑡 = {1 𝑖𝑓 𝑌𝑖𝑘𝑡 ,

𝑌𝑖𝑘𝑡 𝑘 × 𝑊𝑜𝑚𝑒𝑛 𝑖𝑛 𝑃𝑎𝑟𝑙𝑖𝑎𝑚𝑒𝑛𝑡 (%)𝑖𝑡−1 + 𝛽2 × 𝐺𝐷𝑃 𝑝𝑒𝑟

𝑐𝑎𝑝𝑖𝑡𝑎𝑖𝑡−1 +

𝛽3 × 𝑃𝑜𝑙𝑦𝑎𝑟𝑐ℎ𝑦 𝐼𝑛𝑑𝑒𝑥𝑖𝑡 𝛽𝑛 × 𝑅𝑒𝑔𝑖𝑜𝑛𝑎𝑙 𝐷𝑖𝑓𝑓𝑢𝑠𝑖𝑜𝑛

(%)𝑖𝑘𝑡−1 +𝑖𝑘𝑡,

𝐵0𝑘 = 𝐵00 + 0𝑘,

𝐵1𝑘 = 𝐵10 + 1𝑘.

𝑌𝑖𝑘𝑡 is the binary variable indicating whether a country adopts a policy 𝑘 in a year 𝑡, and

𝑌𝑖𝑘𝑡∗ is the latent variable. 𝐵0𝑘 and 𝐵1𝑘 are not fixed but they vary according to types of

violence against women public policies based on the random effect for policies (0𝑘) and the

random coefficient of women’s descriptive representation (1𝑘) by types of such policies. It is

assumed that 0𝑘, 1𝑘, and 𝑖𝑘𝑡 are not correlated with the explanatory variables, and that 0𝑘

and 1𝑘 are independent of 𝑖𝑘𝑡. While other explanatory variables are measured at the

country-time level, Regional Diffusion (%) is measured at the country-policy-year level. To

be specific, Regional Diffusion (%) in this chapter is the percentage of countries with a given

policy in the same geopolitical region according to the types of violence against women

policies.

5.2.2. Empirical Results

Figure 5-2: Estimations from Pooled Event History Models

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 5-1a in

Appendix for the detailed information about the result. In this figure, Women in Parliaments

(%)is presented as its proportion for visualization sake.

Figure 5-2 shows the estimations from the pooled event history model (Model 1) with a

random intercept, and the model (Model 2) with a random coefficient for Women in

Parliaments (%). The VIF tests were conducted to check the presence of multicollinearity

problems in the two models. The mean VIF and the individual VIFs of explanatory variables

excluding time functions are less than 4, indicating no consequential collinearity among

independent and control variables. Even though the random coefficient for Women in

Parliaments (%) is not statistically significant at any level of p-value (see Table 5-1a in

Appendix), the Akaike Information Criterion (AIC) and the Bayesian Information Criterion

(BIC) are smaller in Model 2.

Across the two models, the estimated coefficients and robust standard errors are similar

despite some small differences. In both models, the key independent variable, Women in

Parliaments (%), is statistically significant at the level of p < 0.05 in a positive direction. The

estimated coefficients of the independent variable are 0.009 and 0.008, meaning that 1%

increase in Women in Parliaments (%) increases the log-odds of adopting public policies

addressing violence against women by 0.009 and 0.008. This result provides empirical

evidence for the positive impact of women’s descriptive representation in parliaments is not

limited to the adoption of violence against women laws, but leads countries to introduce

public policies regarding violence against women.

Related to control variables, like the empirical results concerning the adoption of violence

against women laws, GDP per capita, the Polyarchy Index (V-Dem), and CEDAW (%) have

statistically significant relationships at the level of p < 0.01 with the probability of

introducing public policies combating violence against women in the two models. Consistent

with previous studies, it emphasizes the important roles of the level of economic development

and democracy, and the global norms on women’s rights on the adoption of public policies

addressing violence against women. Moreover, Regional Diffusion (%) is also statistically

significant at the level of p < 0.01. At the same time, it is revealed that countries with a

semipresidential government system and Scandinavian Commercial Code are less likely to

adopt public policies concerning violence against women as well as violence against women

laws.

Unlike the results related to the adoption of violence against women laws, some other control

variables turned out to have statistically significant impacts on the dependent variable. The

three variables including Women’s CSO Participation, Women’s Property Rights, and

Ratification of CEDAW are statistically significant at the level of p < 0.01 with positive

directions, while the increase in the Legislature Corrupt Activities makes countries less likely

to adopt public policies concerning violence against women. One of the potential reasons why

those variables are not statistically significant in the models presented in Chapter 4, but

significant in the models in the two models in Figure 5-2 is that the implementation of public

policies, compared to the adoption of laws is more likely to be affected by factors less

relevant to decision-making processes.

In terms of the two women-related factors, the empirical results are in the line with

previous studies arguing that women’s active participation in civil society organizations and

the improved women’s property rights tend to lead countries to pay more attention to

women’s issues such as violence against women (Dahlum & Wig 2020; Datta 2006). In

addition, given the statistically significant correlation between Women’s CSO Participation

and the feminist autonomous movement index from Htun and Weldon (2012), this result also

provides indirect support for the crucial role of the feminist autonomous movements. The

results regarding the positive and negative influences of the ratification of the CEDAW and

the corrupt activities in legislative branches are also consistent with my expectations that the

monitoring body of the committee of CEDAW urges member states to take actions on policies

addressing violence against women, and that the corruption in legislature impede legislative

branches from performing effectively.

Lastly, the results from both Model 1 and Model 2 demonstrate that Sub-Saharan

African countries are less likely to implement public policies addressing violence against

women, while East Asian and South Asian countries are more likely to introduce such public

policies compared to Western European and North American countries. These findings also

make sense given that Sub-Saharan African countries are culturally more lenient to the

problem of violence against women. For instance, there is a relatively pervasive culture of

violence against women and “unfettered social and cultural acceptance of wife-beating”

intensifying such practices and creating public policies to mitigate violence against women

(Rani, Bonu, & Diop-Sidibe 2004, 117).

As in Sub-Saharan African countries, East and South Asian countries have the

practices of female genital mutilation (Muteshi, Miller, & Belizán 2016), sex-selective

abortion, and systematic neglect of girl children (Abrejo, Shaikh, & Rizvi 2009; Hesketh, Lu,

& Xing 2011), intimate partner violence (Jejeebhoy, Santhya, & Acharya 2014), and other

harmful practices related to the problem of violence against women (Hester 2004; Piper

2003). Especially in South Asia, there have been rigid cultures, discriminatory attitudes, and

inequitable practices and violent behaviors towards women at all stages of life (Niaz 2013).

However, unlike Sub-Saharan African countries, among East and South Asian

countries the state and women’s organizations have actively put their efforts on the

implementation of extensive constitutional and statutory safeguards addressing violence

against women. For instance, women’s movements in South Korea have taken the lead in

advocating policy changes addressing violence against women and brought the issue of

violence against women onto the public agenda in Korea (Jung 2013; Postmus & Ah Hahn

2007). Even though those activities from women’s organizations and governmental responses

aiming the problem of violence against women frequently face resistance at the community

level both in East and South Asian countries under the strong patriarchal moral code and

culture (Manderson, Bennett, & Bennett 2003; Mehta 2004; Najmah 2021), they have

engendered policy implementations handling violence against women.

Figure 5-3: Predicted Probabilities of Adopting Violence Against Women Public Policies

Figure 5-3—continued

Note: Predicted Probabilities are presented with 95% confidence intervals with the kernel or

normal density of each independent and control variable. Predicted Probabilities are estimated

while all other variables are held at their observed values.

To examine the substantive effects of statistically significant variables in Model 2 in Figure

5-2, I estimate predicted probabilities according to the different values of the significant

variables. Even if the changes in the predicted probabilities presented in Figure 5-2 are

relatively smaller than the changes related to the adoption of violence against women laws,

the small probability changes are still not negligible considering that the implementation of

public policies about violence against women is quite a rare event.

Based on the changes in the percentage of women in legislative branches, the predicted

probability of adopting public policies addressing violence against women increases steadily.

At 0% of Women in Parliaments (%), the predicted probability is 0.513% and it increases to

0.560% when the percentage reaches 10%. If the percentage of women in parliaments

increases to 30%, the predicted probability reaches 0.612%. At the extreme value, the

predicted probabilities are over 0.793% if the percentage of women in parliaments exceeds

50%.

Other control variables also have a similar degree of substantive effects on the adoption of

public policies concerning violence against women. For instance, when Women’s CSO

participation and Women’s Property Rights increase from -3 to 4 and -5 to 5, the predicted

probabilities increase from 0.298% to 0.924% and from 0.214% to 1.081% respectively. Also,

countries ratifying the CEDAW have around 0.605% predicted probability, while non-

member states have lower than 0.398% predicted probability. In addition, when a country has

the lowest Legislative Corrupt Activities, the predicted probability of adopting public policies

about violence against women is 1.019%. However, if Legislative Corrupt Activities reaches

over 0, the predicted probabilities decrease to 0.596%.

Figure 5-4: Predicted Probabilities based on the Categorical Variables

Figure 5-4—continued

Note: Predicted Probabilities are presented with 95% confidence intervals with the normal

density of each variable. Predicted Probabilities are estimated while all other variables are

held at their observed values.

Figure 5-4 presents the estimated predicted probabilities according to the different values of

the three categorical variables: Government System, Legal System, and Geopolitical Regions.

Put it simply, the predicted probabilities demonstrate that the statistically significant

categorical variables in Model 2 in Figure 5-2 are also substantially meaningful.

First, related to Government System, countries having a semi-presidential system have

0.476% predicted probability of adopting public policies addressing violence against women,

while countries with a presidential system and parliamentary system have 0.646% and

0.607% predicted probabilities. Regarding Legal System, the predicted probability in

countries with English Common Law system is 0.661%, and that in countries with

Scandinavian Commercial Code is 0.482%. Lastly, Sub-Saharan African countries tend to

adopt public policies combating violence against women with 0.368% predicted probability,

while East and South Asian countries have 0.880% and 1.117% predicted probabilities

respectively.

Figure 5-5: Estimations with the Interaction Term

Between Women in Parliaments (%) and CEDAW (%)

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 5-2a in

Appendix for the detailed information about the result.

Next, to evaluate the conditional relationship between Women in Parliaments (%) and

CEDAW (%) on the implementation of public policies addressing violence against women, I

estimate a pooled event history model with a random effect of Women in Parliaments (%) by

different types of public policies. As in Chapter 4 about the adoption of laws concerning

violence against women, the interaction term between the percentage of women in

parliaments and the proliferation of global norms about women’s rights measured by CEDAW

(%) is not statistically significant at any level of p-value.

In terms of control variables, as the results from Model 1 and Model in Figure 5-2, GDP per

capita, the Polyarchy Index (V-Dem), Women’s CSO Participation, Women’s Property Rights,

and Ratification of CEDAW, Legislature Corrupt Activities, and Regional Diffusion (%) have

statistically significant impacts on the log-odds of adopting such public policies. In addition,

the semi-presidential government system is negatively associated with the adoption of

violence against women public policies, even after the inclusion of the interaction term.

SubSaharan African countries are less likely to adopt such policies, while South Asian

countries are more likely to adopt them.

Figure 5-6: Average Marginal Effects of CEDAW (%) by Women in Parliaments (%)

Note: Average Marginal Effects are presented with 95% confidence intervals with kernel

density of Women in Parliaments (%) and CEDAW (%). Average Marginal Effects are

estimated while all other variables are held at their observed values.

As in Chapter 4, I graph estimated average marginal effects of Women in Parliaments (%)

and CEDAW (%) in Figure 5-6. It shows that those two variables are not conditioned by each

other. First, even though the point estimates of the average marginal effects of CEDAW (%)

increase with the increase in the percentage of women in parliaments, each point estimate is

not statistically different from each other. In addition, the point estimates of the average

marginal effects of Women in Parliaments (%) are not statistically different from 0 until

CEDAW (%) reaches around 81%. Although the average marginal effects increase when

CEDAW (%) increases from 81% to 100%, the changes of the average marginal effects of

Women in Parliaments (%) are very marginal at best.

As a whole, the empirical results support the link between women’s descriptive

representation in legislative branches and women’s substantive representation in public policy

outcomes concerning violence against women. This relationship is not only statistically

significant in the pooled event history models with the random effects of Women in

Parliaments (%) by types of public policies, but also substantially meaningful with regard to

the predicted probability of implementing public policies about violence against women.

However, the conditional relationship between women’s descriptive representation and the

proliferation of global norms on women’s rights is not statistically and substantively

significant.

The empirical results on independent variables from Figure 5-2 and Figure 5-5 are

consistently in pooled event history analyses in repeated events context (Table 5-3a in

Appendix) with the consideration of the number of public policies by 20 types of public

policies. To be specific, the influence of women’s descriptive representation in parliaments on

the implementation of public policies is statistically significant, but the interaction term

between Women in Parliaments (%) and CEDAW (%) is not.

Table 5-1: Lists of Countries Adopting Violence Against Women Laws

without Female Legislators

Country

Japan

Australia

Denmark

Egypt

India

Iraq

Thailand

Gabon

Morocco

Senegal

Central African Republic

Year

1908

1924

1930

1937

1953

1959

1960

1963

1965

1966

Sexual Violence

Sexual Violence, Stalking Sexual

Violence

Violence Against Women

Female Genital Mutilation

Shelter

Long Term Support

Table 5-1—continued

Chad 1967 Child, Early, and Forced Marriage

Laws

Policies

Bulgaria

France

1968

1972

Child, Early, and Forced Marriage

Police training,

Prevention Campaign,

Health Training,

Justice Training,

Integrated Service

Oman

Nigeria

Bangladesh

Iceland

1974

1976

1980

1982

Violence Against Women,

Trafficking

New Forms of Violence

Justice Protocol

Shelter, Hotline

Kuwait 1983 Violence Against Women,

Domestic/Intimate Partner Violence

Shelter

Jamaica 1985

Shelter

Angola 1986 Psychosocial Counseling,

United Arab Emirates

Tajikistan

1987

1995

Sexual Violence

Justice Protocols

Psychosocial Counseling

Bosnia and Herzegovina

Qatar

Bahrain

1999

2004

2005

Violence Against Women,

Sexual Violence

Prevention Campaign,

Burma/Myanmar 2005 Trafficking

Non-Formal Education

Kyrgyzstan 2006 Violence Against Women,

Trafficking

Prevention Campaign

Saudi Arabia 2008 Shelter

Note: The information about forms of violence targeted by each law and policy is based on

the Global Database on Violence against Women

Rather than concluding only with the empirical analysis, I further examine potential reasons

why some countries with no female representatives in legislative adopt laws and policies

addressing violence against women by investigating some individual cases. Moreover, the

limitations of laws and policies introduced without the presence of women in parliaments will

be examined. Table 5-1 shows the list of countries adopting and implementing laws or any

types of public policies without female legislators in their legislative branches. Twenty-nine

countries across different geopolitical regions adopted violence against women laws and

public policies despite no presence of women in decision-making institutions.

However, the laws and public policies adopted without supports from female

representatives have their limitations in details of the clauses concerning violence against

women and have been criticized due to the light sentences on violence against women.

Continuously questioned by women’s non-governmental organizations and the committee of

the CEDAW, the Japanese Penal Code of 1908 is one of the typical cases of violence against

women laws with flaws. Two main articles related to sexual violence in the Penal Code of

1908 are below.

Article 177 (Rape): A person who, through assault or intimidation, forcibly commits

sexual intercourse with a female of not less than thirteen years of age commits the

crime of rape and shall be punished by imprisonment with work for a definite term of

not less than 3 years. The same shall apply to a person who commits sexual

intercourse with a female under thirteen years of age.

Article 178 (Quasi Forcible Indecency; Quasi Rape): (1) A person who commits an

indecent act upon a male or female by taking advantage of loss of consciousness or

inability to resist, or by causing a loss of consciousness or inability to resist, shall be

punished […].

The committee of the CEDAW has expressed concern about the Penal Code. Especially, the

fact that the crime of sexual violence is only prosecuted based on complaints from the victim.

Moreover, it has also been pointed out that the penalty for rape is lenient, and that incest and

martial rape are not explicitly defined as a form of violence against women in the Japanese

Penal Code. The Committee also has raised its concern about the low age of sexual consent

and urged Japan to raise the minimum age of sexual consent.

Those kinds of limitations are also observed in Sub-Saharan, North African, and middle east

countries. For instance, the Central African Republic adopted the “Ordinance No.66/16

Abolishing the Practice of Female Circumcision” in 1966. However, even though it shows the

governmental recognition of the fact that FGM tends to entail harmful consequences to the

physical and mental health of women and girls, it has been regarded that this ordinance

imposed any associated criminal penalty for incidents of FGM. Also, Nigeria implemented

the Legal Aid Council to provide free legal counseling services for abused women from

violence against women, but it has been revealed that the Legal Aid Council has failed to help

the victims access judicial remedies (Enemo 2020).

Among countries located in North Africa and the Middle East, Egypt, Iraq, Qatar, and

United Arab Emirates are typical examples. Egypt and Iraq early adopted laws penalizing

violence against women. They early adopted the Egyptian Penal Code and Iraqi Family Law

in 1937 and 1959 respectively, but the Committee of the CEDAW has continuously urged

those countries to modify and revise clauses concerning violence against women. The

Committee said that the Egyptian Panel Code lacks a holistic approach to the prevention and

elimination of the problem of violence against women including domestic and intimate

partner violence, martial rape, and other forms of violence against women.8 In addition, even

if the Iraqi Family Law issued in 1959 has been considered among the best laws in North

Africa and the Middle East, the problem of the child, early and forced marriage is not legally

recognized under the Iraqi Law. Similarly, the penal codes of the United Arab Emirates and of

Qatar have been criticized due to their lack of detailed definitions of the forms of violence

against women.

On the contrary, the early adopted laws and policies in European countries such as the

Danish Criminal Code and French gender equality policies related to violence against women

have been regarded as successful laws and policies combating violence against women by

scholars and the committee of the CEDAW. Among European countries, active feminist

movements have played a crucial role in the adoption of violence against women laws and

policies despite no presence of women in legislative branches. For instance, since the 1970s,

the feminist movement has been rallying public opinion against violence against women

(Corradi & Stöckl 2016). The women’s liberation movement in France in the early 1970s

strived to make changes in laws and policies, and to lead the government to take concrete

action dealing with women’s rights and termination of violence against women (Greenwald

1996). Also, in Denmark, the major women’s movement led by the Danish Women’s Society

urges the Danish government to revise the Danish constitution and to ensure women’s rights

to vote.

8 United Nations, Convention on the Elimination of All Forms of Discrimination against

Women (CEDAW). 19 January 2001. Considerations of Reports of States Parties. Egypt.

As a consequence, the active women’s movement in Denmark contributes to the provision of

equal opportunity laws during the 1920s and the adoption of clauses addressing violence

against women in the Danish Criminal Code (Griffin & Braidotti 2002).

5.3. Conclusion

In a nutshell, by employing pooled event history analysis (PEHA), I demonstrated that the

increased women’s descriptive representation in legislative branches has both statistically and

substantially significant impacts on the implementation of public policies addressing violence

against women. The conducive impacts of women’s descriptive representation on the

implementation of such public policies are not heterogeneous according to the various types

of policies concerning violence against women. As in Chapter 4, the percentage of women in

legislative branches and the percentage of countries ratifying CEDAW are not conditioned by

each other. Those relationships are consistently robust in various model specifications with

different combinations of independent and control variables, meaning that the findings are not

the results of arbitrary modeling strategies.

Moreover, by examining countries that adopted and implemented violence against women

laws and policies without supports from female legislators in legislative branches, I find that

the laws and public policies adopted and implemented without supports from female

legislators in Asian, MENA, and Sub-Saharan countries have non-negligible limitations in

terms of details of clauses and sentence. On the contrary, in Western European countries such

as Denmark and France, the adoption and implementation of successful violence against

women laws and public policies are rooted in the active feminist movements.

CHAPTER 6

THE INFLUENCE OF WOMEN’S DESCRIPTIVE REPRESENTATION:

THE CASE OF SOUTH KOREA

In Chapter 6, I relying on the South Korean case further examine how the presence of female

legislators affect the adoption and implementation of laws and public policies addressing

violence against women with the focus on the bills proposed in the Korean National

Assembly from 1948 (the start of the 1st Congress) to 2020 (the end of the 20th Congress). To

be specific, I will test the relationship between gender of individual legislators and

sponsorship on violence against women bills, and the interaction between gender and

partisanship of legislators will be studied. In addition, when do male legislators sponsor such

bills will be examined with a focus on the role of their previous experience of sponsoring bills

with female legislators. Moreover, I will examine whether bills sponsored by a higher

percentage of female legislators and by female legislators from different political parties

enjoy a higher probability of being passed or not.

To test the above-mentioned relationships and hypotheses, I rely on South Korea’s case. To

reiterate, although South Korea’s legislative politics on women’s issues especially violence

against women have not received much academic attention by the existing literature on

Women and Politics, South Korea is an ideal case to examine such relationships and

hypotheses, for several reasons. First, there are analytical advantages stemming from the bill

sponsorship record existing through the whole period of observation. The data availability

related to proposed bills from the 1st Congress enables me to fully understand the influence of

gender, and the interaction between gender and partisanship of individual legislators. Second,

considering that Tsebelis (2002) demonstrates that executive dominance over legislative

bodies on agenda-setting has been frequently observed in parliamentary democracies unlike

presidential democracies, South Korea which is a presidential democracy is a valuable case

where the Korean National Assembly plays an important role in shaping legislative agenda on

violence against women. In addition, South Korea is a country where both women’s

descriptive and substantive representation in terms of violence against women laws and

policies have progressed dramatically in a relatively short period, which provides enough

variations to be examined.

6.1. Data

6.1.1. Dependent Variables

As dependent variables, I use two binary variables: Bill Sponsorship and Bill Success.

To build Bill Sponsorship, I, first, collected all information about each bill proposed in the

Korean National Assembly from 1948 (the start of 1st Congress) to 2020 (the end of 20th

Congress) based on the Bill Information System developed by the Korean National Assembly.

The Bill Information System provides detailed information about proposed bills such as their

proposed date, lists of sponsors, contents of each bill, and results of each bill. In total, 85,860

bills have been proposed from the 1st Congress to 20th Congress. Then, I, relying on keywords

related to violence against women, classify whether each bill is a bill addressing violence

against women.

To be specific, first, I drop all bills proposed not by legislators. In the dataset, 63,374 bills

among 85,860 bills have been proposed by legislators. Next, I select bills with titles including

the following keywords: feminicide, forced sterilization, stalking, trafficking, violence against

women and girls, domestic and intimate partner violence, sexual violence, sexual harassment,

female genital mutilation, child, early and forced marriage, protection of women and children,

and victims of violence against women. Thrid, I also select bills with contents including the

above-mentioned keywords.9 Lastly, given that this approach might include bills not related to

violence against women, I review the selected bills to exclude bills not relevant to violence

against women.

As a result, 1,916 bills among 85,860 bills are categorized into the violence against

women bills. After the session of the 14th Congress when 4 bills concerning violence against

women were proposed, the number of violence against women bills proposed has increased.

The number reaches 503 in the 19th Congress, and 855 in the 20th Congress. To build Bill

Sponsorship, I assign 0 to “Bill ID – Individual Legislators” observations when legislators

sponsor violence against women bills. Otherwise, 1 is assigned to observations.

9 Classifying bills proposed in Korean National Assembly based on keywords in titles and

contents of bills is one of the common ways to categorize bills into specific topics. As

examples, please refer Shim (2019). All the keywords are used in Korean instead of English

given that the Bill Information System is in Korean.

Figure 6-1: Total Number of Proposed Bills and Violence Against Women Bills

Next, to build Bill Success, I categorize violence against women bills into Rejection and

Success. I assign 0 to “Bill ID” observations if bills are finally rejected. Otherwise, I assign 1.

Since 14th Congress, 797 bills about violence against women have been passed, while 1,119

bills are finally rejected. It should be noted that both the number of proposed violence against

women bills and the rate of rejection have increased.

Figure 6-2: Results of Proposed Violence Against Women Bills

6.1.2. Independent Variables

To examine whether women legislators pay more attention to the problem of violence

against women and focus on proposing bills concerning violence against women, I include

Gender (0 – male legislators, 1 – female legislators) as an independent variable when a

dependent variable is Bill Sponsorship. Moreover, to see whether the gender composition of

sponsors influences the probability of Bill Success, the percent of female sponsors of each

violence against women bills is employed as an independent variable when Bill Success is a

dependent variable. In addition, a binary independent variable indicating whether each

violence against women bills is sponsored by female legislators from different political

parties is included. Individual characteristics including gender and partisanship of legislators

are collected based on the information posted on the Korean National Assembly open-access

website.

In addition, I include Partisanship as an independent variable when Bill Sponsorship

is employed as a dependent variable to evaluate the conditional influence of Gender

according to individual legislators’ partisanships. In South Korea, 87 political parties had

seats in the Korean National Assembly from the 1st Congress to 20th Congress. The 87

political parties are categorized into Center Right, Center Left, and Left political parties. If

individual legislators are not involved in any political party, they are assigned to the

Independent category.

Lastly, in terms of Bill Success, I include a binary variable indicating whether a bill is

sponsored by female legislators from different political parties to see whether women can

overcome marginalization and gendered expectations of performance by sponsoring with

other female legislators or not. I assign 1 to “Bill Id” observations if a bill is sponsored by

female legislators from different political parties. Otherwise, I assign 0.

6.1.3. Control Variables

To prevent coefficient estimations from being biased due to the omitted variable bias, a series

of control variables according to the two dependent variables are controlled. In terms of Bill

Sponsorship, I first control Seniority of legislators. Previous literature on the seniority of

legislators has demonstrated that more senior lawmakers tend to have higher legislative

effectiveness (Taylor 2019; Volden & Wiseman 2014). On the contrary, some Korean case

studies have revealed that more senior members are likely to focus on the issues of unification

and national security, and to propose fewer bills (Jeong & Jang 2013; Ka 2009). Therefore,

rather than disregarding the potential impacts of the seniority of legislators, I include

Seniority as one of the control variables. It is an ordinal variable ranging from 1 to 9

according to the times of being elected.

Second, given that South Korea has a two-tiered electoral system composed of a

district tier (single-member district) and a party tier (proportional representation), Electoral

Tier of legislators is controlled. I assign 0 to “Bill ID – Individual Legislators” observations if

legislators were elected from single-member districts; otherwise, 1 is assigned. The inclusion

of Electoral Tier is based on the rationale that lawmakers elected from single-member

districts tend to focus on the specific problems and issues related to their districts rather than

more general issues such as the problems of violence against women to increase the

probability of being re-elected.

In addition, Electoral Cycle and Congress are also controlled. It is expected that

legislators are more likely to propose or sponsor bills right before the next election because

voters tend to weigh more legislators’ active participation in bill proposal or sponsorship

during such period. To control the anticipated positive impact of Electoral Cycle, 1 is

assigned to “Bill ID – Individual Legislators” observations if a bill was proposed one year

before the general elections; otherwise, 0 is assigned. Congress is a binary variable for each

congress session. The inclusion of Congress enables me to control unobserved and

unmeasured each congressspecific influence on the probability of sponsoring bills about

violence against women.

Related to the other dependent variable, Bill Success, I control Electoral Cycle, total

Number of Sponsors, Total Number of Bills per session, Center Left Party Control, and the %

of Sponsors from Dominant Party. Number of Sponsors and % of Sponsors from Dominant

Party are controlled with a natural expectation that those two variables have positive

relationships with the probability of Bill Success. Moreover, even though there is no concrete

theory about the association between the total number of bills proposed per session and the

probability of bills’ success, Total Number of Bills is controlled, given that previous literature

on bill success has controlled the total number of bills introduced (Senk 2020; Shim 2019).

Center Left Party Control is also included as one of the control variables to parcel out the

potential influence of different party controls in the Korean National Assembly. In addition, I

also control whether each bill is sponsored by legislators from different political parties or not

with the binary variable named Sponsored by Legislators from Different Parties.

I should note that observations before the 13th Congress are not included in the dataset

considering the fact that violence against women bills have been proposed since the 14th

Congress. In addition, given that 14th Congress started in 1992, and the period of military

dictatorship (by Park Jeong-Hee from 1963-1979) and the two military governments (by

Chun Doo-Hwan from 1980-1988, and by Roh Taw-Woo from 1988-1993) ended with the

1992 presidential election won by Kim Young-Sam initiating Civilian Government, it enables

me to naturally control the influence of regime types. Descriptive statistics of all variables

employed in Chapter 6 are presented in Table 6-1a and Table 6-2a according to the two

dependent variables: Bill Sponsorship and Bill Success. Again, the unit of analysis is “Bill ID

– Individual Legislators,” and “Bill ID” when the dependent variables are Bill Sponsorship

and Bill Success respectively. Across all the empirical models estimated in the next section,

the mean and individual variance inflation factors (VIFs) of independent and control variables

do not exceed

2, indicating that there is no problematic multicollinearity among explanatory variables.

6.2. Empirical Analysis

6.2.1. Modeling Strategy

Given the nature of the two binary dependent variables, Bill Sponsorship and Bill Success,

logistic regression is applied to examine the influence of independent variables on both

dependent variables. In addition, robust standard errors are estimated instead of normal

standard errors.

6.2.2. Empirical Results

Figure 6-3: Estimations from Logistic Regression on Bill Sponsorship

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 6-3a in

Appendix for the detailed information about the result.

To examine the influence of gender, partisanship, and the experience of cosponsorship

with female legislators on the probability of sponsoring bills related to violence against

women, I start by estimating two logistic regression models on Bill Sponsorship. Model 1 is

estimated with a full set of independent and control variables, while Model 2 is estimated

based on the split-sample which only includes male legislators. Detailed information about

estimated coefficients and robust standard errors can be found in Table 6-3a in Appendix.

Both Model 1 and Model 2 have similar results on each independent and control

variable, the results reveal that female legislators (Gender = 1), partisanship in the Center-

Left party, previous experience of cosponsorship with female legislators, and PR-elected

legislators have a statistically significant positive relationship with the log-odds of sponsoring

bills about violence against women at the level of p < 0.001. On the contrary, partisanship in

the Left party, seniority, electoral cycle, and Congresses after the 15th Congress have a

statistically significant relationship with the dependent variables at the level of p < 0.001 in

negative directions. In addition, non-partisanship is not statistically significant at any level of

p-value.

In Model 1, first of all, the estimated coefficient of Gender is 0.374, meaning that

women legislators have 0.374 higher log odds of sponsoring violence against women bills

than male counterparts have. It is consistent with an expectation that women legislators tend

to pay more attention to the problem of violence against women and empirical findings from

the two previous chapters that a higher percentage of women legislators in parliaments leads

countries to adopt violence against women laws and public policies.

Figure 6-3 also demonstrates that legislators involved in the Center-Left political

party are more likely to sponsor bills about violence against women, while legislators from

the Left political party are less likely to sponsor such bills. The estimated coefficients are

0.127 and 0.228, indicating that being a legislator involved in the Center-Left party or Left

party, versus involved in the Center-Right party, increases the log-odds of sponsoring

violence against women bills by 0.127 or decreases the log-odds by 0.228 respectively.

In terms of the previous experience of cosponsorship with female legislators, the

estimated coefficient is 0.408 in Model 1. This result supports the argument that having

experience of cosponsorship with female legislators leads legislators to pay more attention to

women’s rights and integrity. This relationship is consistent in Model 2 where only male

legislators are included in the sample, which provides empirical evidence that the legislative

behaviors of male legislators can be influenced by working with female counterparts.

The results presented in Figure 6-3 also reveal other interesting findings related to

control variables: Seniority, Electoral Tier, Electoral Cycle, and Congress. First, Seniority in

both Model 1 and Model 2 is statistically significant with a negative sign at the level of p <

0.001. This finding is not consistent with the previous findings on the conducive impacts of

seniority on legislative effectiveness and sponsorship especially in the U.S. context. For

instance, Volden and Wiseman (2014) through their authoritative study of members’ influence

in lawmaking demonstrate that seniority is strongly and positively related to legislative

effectiveness (see also, Taylor (2019)). Unlike the previous literature on seniority in the U.S.

context, this result provides empirical supports for the findings in the Korean context. It has

been revealed that more senior members are more likely to focus on the issues of national

security and unification (Ka 2009) and generally tend to propose fewer bills (Jeong & Jang

2013).

Electoral Tier is statistically significant at the level of p < 0.05, meaning that legislators

elected through the PR electoral system are more likely to sponsor bills about violence

against women. It is consistent with my expectation that PR-elected legislators pay more

attention to general issues including women’s rights rather than district-specific bills. In

addition, Electoral Cycle has a statistically significant negative relationship with the log-odds

of sponsoring such bills in both Model 1 and Model 2. Even though this relationship is not

anticipated, it is not a nonsensical finding because legislators tend to focus on their election

campaigns to maximize their probability of being re-elected. Given that PR-elected legislators

are usually not being nominated by their political parties again, this explanation can be

applied to both PR-elected and SMD-elected legislators.

Lastly, related to Congress, it is revealed that, compared to 14th Congress which is the

baseline of the Congress variable, 16th, 17th, 18th, 19th, and 20th Congresses are negatively

statistically significant at the level of p < 0.001. Also, the estimated coefficients are getting

bigger with negative signs. It means that the log-odds of sponsoring bills addressing violence

against women getting smaller. This finding can be explained by the increasing intensity of

gender conflicts, the backlash against feminist movements in South Korea, and legislators’

risk aversion. In South Korea, gender conflict has become visible largely due to the changes

since the 1990s: the spread of individualism and gender equality, the dramatic increase in

women’s economic and social activities, and active feminist movements supporting the

reform of the patriarchal culture (J. K. Lee 2013).

Gender conflict in South Korea was escalated in 1999 when the military service

incentive for men discharged from military service was held unconstitutional by Korea’s

Constitutional Court, and the issues of affirmative actions and reverse discrimination have

been fiercely debated (Ryu & Kim 2019). Recently, misandry and misogyny have emerged

especially based on online communities (Um 2016). Under the circumstance of intensified

gender conflicts, legislators who are continuously exposed to media and sensitive to public

opinions tend to behave more carefully related to women’s issues including violence against

women. Even though violence against women bills should not be regarded as affirmative

action and reverse discrimination against men, there are no strong incentives for legislators to

focus on violence against women bills while taking the risks of being negatively spotlighted

as legislators supporting only male or female voters.

Figure 6-4: Predicted Probabilities of Sponsoring Violence Against Women Bills

Figure 6-4—continued

Note: Predicted Probabilities are presented with 95% confidence intervals with the kernel or

normal density of each independent and control variable. Predicted Probabilities are estimated

while all other variables are held at their observed values.

To examine the substantive impacts of the statistically significant variables in Figure 6-3, I

estimate predicted probabilities based on each variable. Figure 6-4 presents the estimated

predicted probabilities with 95% confidence intervals while all other variables are held at

their observed values, with kernel or normal density of independent and control variables.

Overall, the statistically significant variables are not only statistically significant, but also

substantively meaningful.

Female legislators have 4.158% of sponsoring violence against women bills, while male

legislators have 2.914%. In addition, legislators involved in the Center-Left and Left political

parties have 3.370% and 2.399% of sponsoring bills about violence against women

respectively. In terms of the previous experience of cosponsoring any bills with female

colleges, the predicted probability increases from 0.337% to 1.668% when the logged

Cosponsorship with Female Legislators increases from 0 to 4. It reaches 7.621% if the

independent variable increases to 8.

The four control variables including Seniority, Electoral Tier, Electoral Cycle, and Congress

are also substantively significant. Legislators who are first elected have 4.670%, while

legislators elected five times have 1.700%. It decreases down to 1% if legislators have been

elected more than eight times. Related to Electoral Tier, the predicted probability of

PRelected legislators to sponsor bills about violence against women is 3.377% which is

0.311% higher than the probability of legislators elected from single-member districts. In

terms of Electoral Cycle, legislators have 2.544% predicted probability of sponsoring

violence against women bills during one year before the election. During normal years, they

have 3.360% predicted probability. Lastly, a dramatic decreasing tendency is observed related

to the predicted probabilities based on the value of Congress. After the 14th and 15th

Congresses, the predicted probability decreases to 1.693% in the 16th Congress and the

probabilities decrease under 0.5% after the 17th Congress.

Figure 6-5: Estimations with the Interaction Term

between Gender and Partisanship

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 6-4a in

Appendix for the detailed information about the result.

To examine whether Gender has different impacts on the probability of sponsoring bills

addressing violence against women according to their partisanships, I estimate a logistic

regression model with the interaction term between Gender and Partisanship. Figure 6-5

shows the estimated coefficients with 95% confidence intervals.

The interaction term reveals an interesting finding. Gender x Center Left is statistically

significant at the level of p < 0.05 with an estimated coefficient 0.839. It supports my

expectation that the influence of the impacts of Gender on the probability of sponsoring bills

addressing violence against women is conditioned by their partisanships. Moreover, even in

the model with the interaction term, Cosponsorship with Female Legislators, Seniority,

Electoral Tier, Electoral Cycle, and Congress are statistically significant at the level of p <

0.001, as in the Model 1 in Figure 6-3. Even not presented, they are also substantively

significant in terms of predicted probabilities as in Figure 6-4.

Figure 6-6 illustrates the predicted probabilities of sponsoring bills according to individual

legislators’ gender and partisanship. In terms of predicted probabilities, male and female

legislators from the Center-Right political parties have 2.799% and 3.857%. Both predicted

probabilities increase to 3.102% and 4.616% respectively if legislators are involved in the

Center-Left political parties. This result is consistent with previous findings that female

legislators with liberal party affiliation are more likely to sponsor bills aiming to promote

women’s rights and integrity (Erickson 1997; Reynolds 2013; Shim 2019). More

interestingly, female legislators from Left political parties are substantially less likely to

sponsor violence against women bills. Given that Left political parties in South Korea have

more historically focused on labor issues and legislations rather than women’s issues

compared to the CenterRight or Center-Left political parties, the relatively low predicted

probability of sponsoring violence against women is not surprising.

Figure 6-6: Predicted Probabilities based on the Interaction Term

Note: Predicted Probabilities are presented with 95% confidence intervals with kernel density

of Partisanship. Predicted Probabilities are estimated while all other variables are held at their

observed values.

The findings on the sponsorship for bills addressing violence against women are

consistently robust in logistic regression models with the random effect of Bill Id (see Table

65a in Appendix) and other models with different combinations of independent and control

variables. It suggests that the empirical results in Figure 6-3 and Figure 6-5 are not the

product of arbitrary model specifications. Moreover, even though not presented, the

substantive impacts in terms of predicted probabilities are similar.

Figure 6-7: Estimations from Logistic Regression on Bill Success

Note: Point estimates of variables are presented with 95% confidence intervals, visualized

based on the dotwhisker package developed by Solt and Hu (2015). See Table 6-6a in

Appendix for the detailed information about the result. In this figure, % of Female Sponsors

is presented as its proportion for visualization sake.

Next, a logistic regression model with Bill Success as a dependent variable is estimated to

examine the relationship between % of Female Sponsors and Bill Success. Again, the mean

and individual variance inflation factors (VIFs) are less than 2, indicating there is no

consequential multicollinearity among explanatory variables. In addition, robust standard

errors are estimated instead of normal standard errors.

The key independent variable, % of Female Sponsors, is statistically significant at the level

of p < 0.05 with a positive sign. The estimated coefficient is 0.009, meaning that one percent

increase in % of Female Sponsors increases the log-odds of Bill Success by 0.009. It is

consistent with the argument that female legislators put more effort into bill success as one of

the ways to overcome their electoral vulnerability and gendered performance expectation

(Holman et al. 2021). On the contrary, Sponsored by Female Legislators from Different

Parties is not statistically significant at any level of p-value.

In terms of control variables, Electoral Cycle, Number of Sponsors, and Center Left Party

Control are statistically significant at the level of p < 0.05. The estimated coefficients of

Electoral Cycle and Center Left Party Control are -0.558 and -0.313. In addition, Number of

Sponsors is statistically significant with a positive sign as expected. While Total Number of

Bills is statistically significant at the level of p < 0.1, the other two variables, Sponsored by

Legislators from Different Parties and % of Sponsors from Dominant Party, are not

statistically significant at any level of p-value.

To see whether the impact of % of Female Sponsors on the Bill Success, I estimate predicted

probabilities with 95% confidence interval, while all other variables are held at their observed

values. The predicted probability of Bill Success is 27.298% when there is 0% of female

sponsors. It increases to 29.005% and to 30.774% when % of Female Sponsors reaches 10%

and 20% respectively. If 50% of total number of sponsors is composed of female legislators,

the predicted probability reaches 36.412%. At the extreme, the predicted probability increases

to 46.623%. Thus, Figure 6-8 reveals that the influence of % of Female Sponsors on the

probability of Bill Success is not only statistically significant, but also substantially

meaningful.

Figure 6-8: Predicted Probabilities of Bill Success

Note: Predicted Probabilities are presented with 95% confidence intervals with the kernel or

normal density of each independent and control variable. Predicted Probabilities are estimated

while all other variables are held at their observed values.

6.3. Conclusion

In Chapter 6, I examine the mechanisms behind the link between women’s descriptive

representation and the adoption and implementation of laws and public policies addressing

violence against women with a focus on the Korean National Assembly. Among 85,860 bills

proposed from 1st to 20th Congresses, 1,916 bills are about the issue of violence against

women. I, employing logistic regression, demonstrate that gender of individual legislators is a

driving factor of sponsoring bills addressing violence against women. To be specific, female

legislators have a higher probability of sponsoring violence against women bills compared to

male counterparts, which result is consistent with the empirical findings from Chapter 4 and

Chapter 5 conducting time-series cross-national analysis.

More interestingly, I find that there is a statistically and substantially significant relationship

between the previous experience of cosponsoring with female legislators and the male

legislators’ probabilities of sponsoring violence against women bills. This finding emphasizes

the importance of considering the influence of the increased women’s descriptive

representation on the legislative behaviors of male lawmakers. In addition, the empirical

result about the interaction term between Gender and Partisanship, which has been recently

spotlighted, has meaningful impacts on the sponsorship of bills addressing violence against

women. In the South Korean context, female legislators involved in Center Left political

parties are more likely to sponsor such bills rather than those from Center Right or Left

political parties.

Lastly, in terms of Bill Success, a higher percentage of female sponsors tends to

increase violence against women bills’ probabilities of being passed, while Sponsored by

Female Legislators from Different Parties is not correlated with the log-odds of Bill Success.

Also, I find that Electoral Cycle, Total Number of Bills, and Center Left Party Control are

statistically significant in a negative direction, while Number of Sponsors is statistically

significant in a positive direction.

CHAPTER 7

DISCUSSION AND CONCLUSION

Violence against women is a prevalent problem over the globe, intimidating women’s

rights and integrity. Over 35% of women have experienced violence against women during

their lifetime (UN Women 2019). Even female politicians have suffered from violence against

women and such a pattern has been confirmed at the country level (Krook & Sanín 2020).

With the widespread problem of violence against women, it has been revealed that violence

against women imposes enormous physical and emotional costs for abused women, and even

harms national securities (Córdova & Kras 2020; Krook 2017; Kuperberg 2018). Including

various forms of violence such as intimate partner violence, trafficking, and sexual violence,

violence against women is associated with numerous health problems such as depression and

posttraumatic stress disorder (Jordan, Campbell, & Follingstad 2010), and generalized fear of

crime (Ferraro 1996).

The incredibly high rates of violence against women which have been reported throughout

the world, its fatalistic impacts on individual victims, and enormous economic and social

costs have been widely discussed not only in the United Nations General Assembly but also

by other international organizations (Reeves & Meyer 2021; Woo 2021b). Especially, the four

World Conferences on Women (Mexico City – 1975, Copenhagen – 1980, Nairobi – 1985,

and Beijing 1995) played important roles to raise the issue of violence against women. Even

if not particularly focused, the first World Conference on Women recognized and emphasized

the importance of dealing with violence against women, while the second conference adopted

the resolution on “battered women and violence in family” (Qureshi 2020). In both the third

and fourth conferences, the problem of violence against women was emphasized as an

obstacle to achieving gender equality, and development and peace.

However, even though there has been a global consensus on the detrimental impacts

of violence against women, still some countries do not have laws and public policies

concerning violence against women. Why do some countries adopt and implement violence

against women laws and public policies, while others do not? By employing a mixed-method

approach composed of large-N time-series cross-national analysis on 161 countries from 1908

when Japan first adopted a violence against women law to 2016 and South Korean case study

with information about proposed bills from 1948 to 2020, this dissertation with a focus on the

role of women’s descriptive representation examines potential factors leading countries to

focus on the adoption and implementation of violence against women laws and public

policies.

Throughout the three empirical chapters, this dissertation finds several interesting

findings. First, in Chapter 4, I, estimating discrete event history models with time-series

crossnational data, provide empirical evidence supporting the strong link between women’s

descriptive representation in legislative branches and the adoption of laws addressing

violence against women. Second, in Chapter 5, the relationship between women’s descriptive

representation in legislative branches and the implementation of various types of public

policies concerning violence against women is empirically supported. Additionally, by

examining countries adopting violence against women laws and public policies, it is

demonstrated that laws and public policies adopted and implemented without the presence of

women in legislative branches tend to suffer from their limitations on details in clauses and

sentences. Exceptionally, early adopted violence against women laws and public policies of

Western European countries turn out to be successful laws and policies. Despite no presence

of women in parliaments, active feminist movements and organizations in Western European

countries played crucial roles to make Western European countries adopt such laws and

policies.

Lastly, in Chapter 6, I find that gender of individual legislators is an important

determinant of sponsoring bills concerning violence against women. Also, it is demonstrated

that the previous experience of cosponsoring bills with female legislators leads male

counterparts to pay more attention to sponsoring violence against women bills. Furthermore,

by estimating logistic regression models with an interaction term between gender and

partisanship of individual legislators, I show that legislators from center-left political parties

in South Korea tend to focus on violence against women bills compared to legislators from

the center-right and left political parties. In terms of the success of proposed bills, I find a

strong association between the percentage of female sponsors in the total number of sponsors

and the proposed bills’ probabilities of being passed. The empirical findings from the three

chapters are also consistently robust in various model specifications and alternative modeling

strategies.

This dissertation contributes both the literature on Women and Politics, and the

literature on violence against women. First, the findings on the decisive influence of the

percentage of women in parliaments on the adoption and implementation of laws and public

policies addressing violence against women from the two empirical chapters with time-series

cross-national data provide strong supports for the link between women’s descriptive

representation and women’s substantive representation in terms of policy outcomes. Given the

empirical results from previous studies on this association are mixed at best because of the

lack of well-developed time-series cross-national analysis dealing with policy promotion

“from the start” (Courtemanche & Green 2017; Wängnerud 2009), the large-N time-series

cross-national analysis on 161 countries from 1908 when Japan first adopted violence against

women law to 2016 from this dissertation emphasizes the crucial role of women’s descriptive

representation in achieving women’s substantive representation.

Moreover, considering that previous literature on Women and Politics has largely

focused on Western European countries and the U.S., the South Korean case study on 85,860

bills proposed from 1948 (the start of 1st Congress) to 2020 (the end of 20th Congress)

broaden our understanding on the influence of gender on individual lawmakers’ legislative

behaviors. The empirical finding that female legislators are more likely to promote violence

against women bills rather than their male counterparts suggests that the relationship between

women’s descriptive representation and substantive representation also exists in the Asian

context. Furthermore, the statistically and substantially significant interaction term between

gender and partisanship of legislators provides empirical supports for the argument that there

are complex interactions between gender and partisanship (Reingold et al. 2021).

The South Korean case study also contributes to the literature on Women in Politics

by demonstrating that male legislators with more experience of cosponsoring bills with

female counterparts are more likely to sponsoring bills about violence against women. This

finding raises an important point that we need to consider the influence of the entrance of

women in legislative branches can also alter the legislative behaviors of male legislators.

Furthermore, this dissertation contributes to the literature on violence against women

by unveiling a series of other determinants of the adoption and implementation of laws and

public policies concerning violence against women. For instance, aside from women’s

descriptive representation, the two empirical chapters based on the time-series cross-national

data demonstrate that the level of democracy, corrupt activities in legislative branches, legal

and government systems, and geopolitical regions have statistically significant relationships

with the adoption or implementation of violence against women laws or public policies. At

the same time, the results provide partial supports for the roles of economic development and

women’s property rights. More interestingly, the increasing negative impacts of Congress

binary indicators in the South Korean case study provide a clue for the potential backlashes

against radical feminist movements.

Besides, the findings from this project suggest some potential directions for future

research related to the relationship between women’s descriptive representation and the

adoption or implementation of laws and public policies addressing violence against women.

In Chapter 5, I demonstrate that violence against women laws and public policies adopted

without supports from female representatives in legislative branches are less developed.

Given that laws and public policies are developed, introduced, and modified continuously,

studying the influence of the improved women’s descriptive representation on the changes of

the contents in laws and policies will be promising. The application of text analysis is a

feasible option to examine this relationship. In addition, with the South Korean case study, I

provide a clue for the negative impacts of intensified gender conflicts on bill sponsorship

related to violence against women. Thus, the natural future direction of this project is to

quantify the degree of gender conflicts and directly examine the influence of gender conflicts

on legislative behaviors. In addition, the varying impacts of gender conflicts according to

gender and partisanship of individual legislators will be also promising.

Re-Examining the Origins of Policy Related to Violence Against Women The Role of Women’s Descriptive Representation in Achieving Substantive Representation

Students also viewed