Can Legislation Reduce Domestic Violence in Developing Countries?

Abstract

This study investigates the extent to which the legislation targeting domestic violence may influence both women’s victimization by their partners and marital dissolution in 54 developing countries. We find that the legislation is effective in reducing domestic violence against women, evidenced by the decreases in the composite indices of emotional abuse, less severe violence, more severe violence, and sexual violence by 13.6, 14.4, 19.6, and 11.5%, relatively, relative to the sample averages. The legislation also makes women less likely to be divorced. Our heterogeneity analysis reveals that the disadvantaged population, i.e., rural women, poorly educated women, women having poorly educated spouses, and women from relatively poorer households, might receive less protection from the domestic violence law. Our findings call for more reforms in the legislative systems, so that domestic violence victims can be better protected.

1. Introduction

According to the United Nations, domestic violence, also called intimate violence or domestic abuse, refers to behavior intended to achieve or retain power and control over an intimate partner in any relationship [1]. Abuse can be emotional, physical, or sexual actions or threats of actions that might influence the victim. Approximately 30% of women worldwide have been subject to domestic violence, and around 38% of all murders of women are committed by intimate partners [2]. Domestic violence is regarded as a major public health problem and a violation of women’s human rights. Domestic violence can lead to permanent injuries, psychological distress, and even suicide, for women [3,4]. The consequences for children include a wide range of behavioral and emotional disturbances [5,6].

This paper evaluates the extent to which the legislation targeting domestic violence might influence women’s victimization by their partners and marital dissolution in 54 developing countries. A country is considered to have legislation addressing domestic violence if there is a legal code that includes criminal sanctions or provides for protection orders for domestic violence in that country. The contribution of our study is three-fold. Firstly, we comprehensively investigate multiple dimensions of domestic violence such as emotional, physical, and sexual abuse, instead of looking at just one dimension. We also look at marital dissolution, to evaluate the impacts of domestic violence legislation. Secondly, we conduct a rigorous heterogeneity analysis, to see which group of the population is most affected by the policy. Finally, rather than concentrating on a single country, our context study includes 54 developing countries between 1990 and 2019, thus providing meaningful implications for many governments.

To investigate the impacts of interest, we employed the Demographic and Health Surveys (DHS) and the Women, Business, and the Law 2021 file released by the World Bank [7]. The DHS provided us with detailed information on women and partners’ characteristics. Importantly, the DHS contained various questions on women’s experience of domestic violence, which allowed us to construct composite indices of emotional abuse, physical violence in the less severe form, physical violence in the more severe form, and sexual violence. Our empirical model hinges upon the variation in the experience of domestic violence and divorce of women living in the same area but with some exposed to the law at the time of marriage and others not. The empirical model rests on the assumption that the timing of the legislation is unrelated to within-neighborhood unobserved factors that could potentially affect domestic violence.

Our study reached the following findings. Firstly, if the domestic violence law is effective at the time of marriage, women are less likely to be victimized by their partners. Specifically, the legislation decreases the composite indices of emotional abuse, less severe violence, more severe violence, and sexual abuse, by 0.03, 0.04, 0.02, and 0.01 points, respectively. These estimates correspond to the declines by 13.6, 14.4, 19.6, and 11.5% relative to the sample averages. Secondly, we also detected the lower probability of marital dissolution where women are 1.3 percentage points less likely to be divorced, representing a 15.8% reduction in the proportion of divorced women. Thirdly, analyzing the heterogeneous impacts of the legislation, we found that the disadvantaged population, i.e., rural women, poorly educated women, women having poorly educated spouses, and women from relatively poorer households, might receive less protection from the law.

Our findings emphasize the meaningful role of the domestic violence legislation in protecting women. To the extent that domestic violence imposes dreadful short-term and long-term consequences on women and children [4,6], our study suggests that legislation targeting domestic violence should be implemented and enforced. The findings also call for more reforms in the legislative systems so that domestic violence victims can be better protected. Such policies are important in order to ensure an environment free of violence for women and children. Furthermore, policymakers should also pay more attention to disadvantaged groups such as rural women, poorly educated women, women having poorly educated partners, and women from relatively poor households, since they tend to be more vulnerable.

The paper proceeds as follows: Section 2 reviews related literature, Section 3 describes the data and study sample, Section 4 presents the findings, and Section 5 provides the discussion and conclusion.

2. Literature Review

Our study is related to the literature on women’s empowerment and gender equality. Specifically, Bilan et al. [8] point out that discrimination against women is still prevalent in enterprises. However, the authors also document the fact that countries are making progress in empowering women, evidenced by the increase in the number of girls being enrolled in school, and the number of women serving in leadership positions. Moreover, it is reported that women’s empowerment is negatively related to poverty [9]. Empowering women through education could be effective at reducing gender inequality at birth by lessening the preference for sons over daughters [10,11]. Among the various factors that could influence women’s empowerment, such as education, income, and assets, Street et al. [12] show that being a business mentor to other women raises their sense of empowerment. Furthermore, training in microfinance is positively linked to various empowerment indicators such as intra-household decision making, as well as political and community participation [13].

Our study also fits into the literature that explores various factors affecting women’s experience of domestic violence. In particular, Le and Nguyen [14] document the fact that highly educated women from developing countries are less likely to be exposed to intimate violence, especially psychological abuse. Other studies report that an increase in women’s income can also make women less susceptible to domestic violence. For instance, in the context of India, Bhattacharyya [15] find that women’s engagement in paid work is associated with a reduction in domestic violence. In another setting, Bobonis [16] show that Mexican women receiving cash transfers from the Mexican Oportunidades program tend to face less physical violence. Moreover, it is documented that women’s ownership of land can improve women’s position within an intimate relationship, making it less likely for their partners to act out aggression against them [15,17,18]. Nevertheless, other events such as armed conflict and natural disasters are adverse shocks that can worsen women’s experience of intimate violence [19,20,21].

Closely related to our paper are the studies that examine the impacts of government policies that specifically target intimate relationships. Dugan [22] examines policies on domestic violence in different states within the United States. The author finds that women residing in states with aggressive legislation have a higher probability of reporting domestic violence offenders to the police, and a lower incidence of domestic violence. Also in the context of the U.S., Stevenson and Wolfers [23] concentrate on divorce policy. The authors detect the decline in domestic violence, female suicide, and spousal murder in states that make the divorce process easier. In a different setting, Beleche [24] focuses on the Penal Code Reform in Mexico, which clarifies what constitutes domestic violence and what punishments the perpetrator faces. Beleche [24] finds that criminalizing domestic violence in a state’s criminal code leads to a decrease in female suicide rates, which can be attributed to the decline in domestic violence.

In this paper, we aim to address two main research questions. The first question is, what are the impacts of the legislation targeting domestic violence both on women’s experience of domestic abuse by their partners and on their likelihood of divorce? The second question is whether the impacts vary across different groups of women from various socio-economic backgrounds.

By investigating the impacts of legislation specifically addressing domestic violence on women’s experiences of intimate violence and divorce in 54 developing countries, our study complements the literature in three ways. Firstly we comprehensively examine multiple dimensions of domestic violence captured by the composite indices of emotional, physical, and sexual violence, along with specific incidences behind each dimension. We then look at marital dissolution, to evaluate the impacts of domestic violence legislation. Secondly, we rigorously conduct a heterogeneity analysis, to see which groups of the population the legislation tends to have a larger effect on. Finally, instead of concentrating on a single country, our study context includes 54 developing countries between 1990 and 2019. Therefore, our results can have meaningful implications for many governments.

3. Data

3.1. Data on Domestic Violence

Data on domestic violence are drawn from the Demographic and Health Surveys (DHS). Targeting women of reproductive ages (15–49) in developing countries, the DHS gathers information on their demographic backgrounds, fertility, and birth history, among others. In addition, there are specific questions on women’s experience of domestic violence in the past 12 months, which can be categorized into four groups. Specifically, the first group consists of three questions on the incidences of emotional abuse, which ask if the woman was humiliated, threatened with harm, or insulted by her partner in the past 12 months. The second group comprises four questions on women’s experiences of physical violence in the less severe form which ask if the woman was pushed/shook or had something thrown at her, slapped, punched with fists or something harmful, or had her armed twisted or hair pulled by her partner in the last 12 months. The third group is composed of three questions on the incidences of physical violence in the more severe form which ask whether the woman was kicked or dragged, strangled or burnt, or threatened with a knife or gun by her partner in the last 12 months. The fourth group includes two questions on women’s experiences of sexual abuse which ask whether the woman was forced into unwanted sex or forced into other sexual acts by her partner in the last 12 months.

For each question in each group, we created a one-zero indicator that equals one if the woman was subject to each incident of violence in the last 12 months, and zero otherwise. We generated four composite indices corresponding to four groups by averaging the item indicators within each group. Specifically, the first index, Emotional Abuse, is calculated as the average of the three indicators Humiliated, Threatened, and Insulted. The second index, Less Severe Violence, is computed by taking the average of the four indicators Pushed/Shook, Slapped, Punched/Hit, Arm Twisted/Hair Pulled. The third index, More Severe Violence, is the average of the three indicators Kicked/Dragged, Strangled/Burnt, and Threatened with Knife/Gun. The fourth index, Sexual Abuse, is calculated as the average of the two indicators Unwanted Sex and Other Sexual Acts.

Finally, we used one question on the marital status of women to construct an indicator (Being Divorced), taking the value of one if the woman was divorced, and zero otherwise. In total, we have one divorce indicator, four domestic violence composite indices, and 12 underlying items. In addition, the DHS also provided us with details on women’s backgrounds as well as their partners’ backgrounds such as age, educational levels, etc.

3.2. Data on Domestic Violence Legislation

Data on legislation on domestic violence are retrieved from the Women, Business, and the Law 2021 file released by the World Bank [2]. This document provides information on laws and regulations that affect women’s economic opportunities in 190 countries. One section of the document records whether domestic violence legislation is present in each country. The existence of domestic violence legislation means that the country has legislation specifically targeting domestic violence that includes criminal sanctions and protection orders for domestic violence. Domestic violence legislation is considered non-existent if there is no such legislation, if the legislation does not include sanctions or protection orders, or the legislation only protects a specific category of women or family members. Moreover, if there is only one provision in the criminal code that raises the punishment for ordinary offenses committed between spouses or within the family, domestic violence legislation is still regarded as non-existent [2]. The information on the legislation in each country in our sample is provided in Appendix B. To capture whether or not the woman is protected by domestic violence legislation, we constructed our main explanatory variable DVL (Domestic Violence Legislation) that takes the value of one if the domestic violence legislation as defined in World Bank [2] is effective in the woman’s country of residence on the date of her marriage, and zero otherwise.

3.3. Estimation Sample

Our sample consists of approximately 818,000 women from 54 low- and middle-income countries.

Table A1. Country List.

Order	ISO3	Country Name	Order	ISO3	Country Name
1	AFG	Afghanistan	28	LBR	Liberia
2	AGO	Angola	29	MDA	Moldova
3	ARM	Armenia	30	MDV	Maldives
4	AZE	Azerbaijan	31	MLI	Mali
5	BDI	Burundi	32	MMR	Myanmar
6	BEN	Benin	33	MOZ	Mozambique
7	BFA	Burkina Faso	34	MWI	Malawi
8	BGD	Bangladesh	35	NAM	Namibia
9	CIV	Cote d’Ivoire	36	NGA	Nigeria
10	CMR	Cameroon	37	NPL	Nepal
11	COD	Dem. Rep. Congo	38	PAK	Pakistan
12	COL	Colombia	39	PER	Peru
13	COM	Comoros	40	PHL	Philippines
14	DOM	Dominican Rep.	41	RWA	Rwanda
15	EGY	Egypt	42	SEN	Senegal
16	ETH	Ethiopia	43	SLE	Sierra Leone
17	GAB	Gabon	44	STP	Sao Tome & Principe
18	GHA	Ghana	45	TCD	Chad
19	GMB	Gambia	46	TGO	Togo
20	GTM	Guatemala	47	TJK	Tajikistan
21	HND	Honduras	48	TLS	Timor-Leste
22	HTI	Haiti	49	TZA	Tanzania
23	IND	India	50	UGA	Uganda
24	JOR	Jordan	51	UKR	Ukraine
25	KEN	Kenya	52	ZAF	South Africa
26	KGZ	Kyrgyzstan	53	ZMB	Zambia
27	KHM	Cambodia	54	ZWE	Zimbabwe

in Appendix A displays the list of countries in our study. The mean and standard deviation of outcome variables are provided in

Table 1. Descriptive Statistics for Dependent Variables.

	Mean	Standard Deviation	Sample Size
	(1)	(2)	(3)
Emotional Abuse	0.221	0.415	743,872
Humiliated	0.149	0.356	693,312
Threatened	0.082	0.275	685,680
Insulted	0.173	0.378	614,396
Less Severe Violence	0.278	0.448	818,525
Pushed/Shook	0.177	0.382	804,177
Slapped	0.235	0.424	782,402
Punched/Hit	0.103	0.304	782,283
Arm Twisted/Hair Pulled	0.087	0.282	586,237
More Severe Violence	0.102	0.303	800,584
Kicked/Dragged	0.096	0.294	781,385
Strangled/Burnt	0.029	0.167	786,301
Threatened with Knife/Gun	0.027	0.161	767,573
Sexual Abuse	0.087	0.281	782,703
Unwanted Sex	0.081	0.273	782,650
Other Sexual Acts	0.034	0.181	653,987
Being Divorced	0.082	0.274	815,751

. The index Emotional Abuse has the mean value of 0.221. Approximately 14.9, 8.2, and 17.3% of our sampled women were humiliated, threatened with harm, and insulted by their partners in the last 12 months, respectively. The average value of the index Less Severe Violence is 0.28. Approximately 17.7, 23.5, 10.3, and 8.7% of women in our sample were pushed/shook or had something thrown at them, slapped, punched with fists or something harmful, and had their armed twisted or hair pulled by their partners in the last 12 months, respectively. The index More Severe Violence has the mean value of 0.10. The proportions of women being kicked or dragged, strangled or burnt, and being threatened with a knife or gun by their partner in the last 12 months are 9.6, 2.9, and 2.7%, respectively. The mean of the Sexual Abuse index lies approximately at 0.09. The fractions of women being forced into unwanted sex and other sexual acts are 8.1 and 3.4%, respectively. Around 8.2% of women in our sample are divorced.

The summary statistics of independent variables are displayed in

Table 2. Summary Statistics for Independent Variables.

	Mean	SD	N
	(1)	(2)	(3)
DVL	0.330	0.470	818,722
Woman Age	32.392	8.508	818,722
Woman Education	6.294	5.010	818,414
Partner Age	38.013	10.263	705,352
Partner Education	7.262	5.057	756,717
Age Difference	5.986	6.598	705,352
Number of Children	2.824	1.985	818,722
Male Household Head	0.796	0.403	818,720
Living in Urban Areas	0.435	0.496	818,722
Wealth Quintile	2.880	1.392	781,152

. Approximately 33% of our sampled women were exposed to the legislation on domestic violence at the time of marriage. The mean ages of women and their partners are 32.4 and 38 years old, respectively. The average age difference is roughly six years. On average, women complete approximately 6.3 years of education, and their partners complete around 7.3 years of education. The mean number of children is 2.8. The proportion of households headed by a male is 80%. Around 43.5% of households live in urban areas. The average wealth quintile is 2.9.

4. Empirical Methodology

We quantify the effects of domestic violence legislation through the regression equation given by

$$Y_{ijs}={\beta }_0+{\beta }_{1}DV{L}_{ijs}+{\delta }_j+{\lambda }_s+X^{\prime }{}_{ijs}\mathsf{\Omega } +{ϵ}_{ijs}$$

(1)

The subscripts correspond to woman i, residential cluster j, year of birth t, and year of survey s. The variable $Y_{ijs}$ represents the outcomes of interest (four indices of domestic violence, the underlying items, and the divorce indicator). The variable $DV{L}_{ijs}$ (Domestic Violence Legislation) is an indicator taking a value of one if domestic violence legislation is effective at the marriage date. In addition, we denote by ${\lambda }_s$ and ${\delta }_j$ survey year fixed effects and residential cluster fixed effect, respectively.

$X^{\prime }{}_{ijs}$ is a covariate of woman characteristics, including woman age, squared woman age, woman education, partner age, squared partner age, partner education, age difference between woman and partner, the number of children, whether the household is headed by a male, whether the household lives in a rural area, household wealth quintile, fixed effects for woman’s and partner’s birth year, and woman and partner country-specific birth cohort trends. Finally, ${ϵ}_{ijs}$ stands for the error term. Robust standard errors are clustered at the residential cluster.

Our coefficient of interest is ${\beta }_1$, which summarizes the effectiveness of domestic violence legislation in mitigating women’s experience of domestic violence and divorce. Our identification hinges upon the variation in the experience of domestic violence and divorce of women living in the same area with some exposed to the law at the time of marriage and others not. The empirical model rests on the assumption that the timing of the legislation is unrelated to within-neighborhood unobserved factors that could potentially affect domestic violence.

5. Results

5.1. Main Results

Table 3. The Effectiveness of Domestic Violence Law—Main Results.

	(1)	(2)	(3)	(4)
Panel A: Emotional Abuse
DVL	−0.056 ***	−0.055 ***	−0.027 ***	−0.026 ***
	(0.001)	(0.001)	(0.002)	(0.002)
Observations	743,872	743,872	612,632	610,877
Panel B: Less Severe Violence
DVL	−0.039 ***	−0.061 ***	−0.045 ***	−0.038 ***
	(0.001)	(0.001)	(0.002)	(0.002)
Observations	818,525	818,525	657,853	655,895
Panel C: More Severe Violence
DVL	−0.021 ***	−0.044 ***	−0.023 ***	−0.021 ***
	(0.001)	(0.001)	(0.001)	(0.001)
Observations	800,584	800,584	657,739	655,781
Panel D: Sexual Abuse
DVL	−0.029 ***	−0.027 ***	−0.014 ***	−0.013 ***
	(0.001)	(0.001)	(0.001)	(0.001)
Observations	782,703	782,703	642,655	640,705
Panel E: Being Divorced
DVL	−0.008 ***	−0.013 ***	−0.010 ***	−0.013 ***
	(0.001)	(0.001)	(0.001)	(0.001)
Observations	815,751	815,751	777,884	776,353
Residential Cluster FE	.	.	.	Y
Individual Characteristics	.	.	Y	Y
Survey Characteristics	.	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refers to survey year and country fixed effects. Individual Characteristics includes woman age, squared woman age, woman education, partner age, squared partner age, partner education, age difference between woman and partner, the number of children, whether the household is headed by a male, whether the household lives in a rural areas household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

displays our main results on the impacts of domestic violence law on women’s experiences of domestic violence and divorce. Each panel presents the results for a different outcome variable. Coefficients from the most parsimonious specification that only includes the Domestic Violence Legislation indicator as the regressor are displayed in Column 1. In Column 2, we control for survey characteristics (survey year and country fixed effects). In Column 3, we further account for individual characteristics (woman age, squared woman age, woman education, partner age, squared partner age, partner education, age difference between woman and partner, the number of children, whether the household is headed by a male, whether the household lives in a rural area, household wealth quintile, fixed effects for woman’s and partner’s birth year, woman and partner country-specific birth cohort trends). Column 4 is our most extensive specification, and incorporates survey characteristics, individual characteristics, and residential cluster fixed effects.

According to our most parsimonious specifications, exposure to the domestic violence law at the date of marriage decreases the composite indices of emotional abuse, less severe violence, more severe violence, and sexual abuse by 0.06, 0.04, 0.02, and 0.03 points. Nevertheless, these estimates only capture the correlation between the law and domestic violence because key factors potentially affecting both the implementation of the legislation and women’s victimization are not controlled for. For instance, women in a more progressive country might be more likely to be exposed to the law and experience less domestic abuse at the same time. Column 1 also shows that the law is negatively correlated with the probability of marital dissolution by 0.8 percentage points.

In Column 2, we introduced survey year and country fixed effects to account for time and country specific factors that are jointly correlated with the implementation of the domestic violence law and women’s experience of domestic abuse. We still detected negative and statistically significant effects of the law on women’s experience of domestic violence and divorce. However, the specification in Column 2 does not account for women’s and spouses’ characteristics that might confound the impacts of interest. An example may be the possibility that women getting married later tend to be exposed to the law and face less domestic abuse simultaneously. Therefore, in Column 3, we further control for women’s and spouses’ characteristics. The estimates are negative and statistically distinct from zero, although the magnitude slightly decreases.

Finally, we incorporated residential fixed effects into our most extensive specification to account for location-specific factors affecting both the implementation of the legislation and domestic violence. For example, the enforcement of the legislation might be relatively weak in some poor regions of a country, and women in such regions were more likely to experience domestic violence. Evident from Column 4, we found that the domestic violence law was effective in reducing domestic abuse against women and the likelihood of women being divorced. Specifically, the implementation of the domestic violence legislation is associated with the decreases in the emotional abuse, less severe violence, more severe violence, and sexual abuse indices by 0.03, 0.04, 0.02, and 0.01 points, respectively. These estimates correspond to 13.6, 14.4, 19.6, and 11.5% decreases relative to the sample averages. As women are less likely to experience aggression from partners, the probability of divorce declines. Specifically, women are also 1.3 percentage points less likely to be divorced, representing a 15.8% reduction in the proportion of divorced women.

Next, to explore the source of the decrease in women’s experience of domestic abuse, we estimated separately how the domestic violence law affected each of the violent incidents captured by the individual items that constitute the composite indices (

Table 4. The Effectiveness of Domestic Violence Law by Items.

	(1)	(2)	(3)	(4)
Panel A: Emotional Abuse
	Humiliated	Threatened	Insulted
DVL	−0.019 ***	−0.010 ***	−0.019 ***
	(0.002)	(0.001)	(0.002)
Observations	591,837	591,769	505,107
Panel B: Less Severe Violence
	Pushed/ Shook	Slapped	Punched/ Hit	Arm Twisted/ Hair Pulled
DVL	−0.024 ***	−0.038 ***	−0.018 ***	−0.011 ***
	(0.002)	(0.002)	(0.001)	(0.001)
Observations	644,220	642,467	642,374	506,521
Panel C: More Severe Violence
	Kicked/ Dragged	Strangled/ Burnt	Threatened with Knife/Gun
DVL	−0.020 ***	−0.006 ***	−0.005 ***
	(0.001)	(0.001)	(0.001)
Observations	640,208	644,172	628,219
Panel D: Sexual Abuse
	Unwanted Sex	Other Sexual Acts
DVL	−0.013 ***	−0.004 ***
	(0.001)	(0.001)
Observations	640,663	562,524
Residential Cluster FE	Y	Y	Y	Y
Individual Characteristics	Y	Y	Y	Y
Survey Characteristics	Y	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refers to survey year and country fixed effects. Individual Characteristics includes woman age, squared woman age, women education, partner age, squared partner age, partner education, age difference between women and partner, the number of children, whether the household is headed by a male, whether the household lives in rural areas, household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

). Starting with emotional abuse, we found that the implementation of the domestic violence law decreased the probability of women being humiliated, threatened with harm, and insulted by their partners in the past 12 months by 1.9, 1.0, and 1.9 percentage points, respectively. Domestic violence law is also associated with the reduction of less severe violence in all forms, particularly 2.4, 3.8, 1.8, and 1.1 percentage point reductions in the incidences of women being pushed/shook, slapped, punched/hit, and having arm twisted/hair pulled by their partners in the past 12 months, respectively.

Regarding more severe violence, if the domestic violence law is already effective at the time of marriage, women are less likely to be kicked/dragged, strangled/burnt, and threatened with knife/gun by their partners in the past 12 months by 2.0, 0.6, and 0.5 percentage points, respectively. The law further decreases women’s experience of sexual abuse, evidenced by the 1.3 and 0.4 percentage point decreases in the likelihood of being forced into unwanted sex and other sexual acts, respectively.

5.2. Heterogeneity Analysis

We continued to analyze the heterogeneous impacts of domestic violence legislation. First, we looked at the dimension of residence by examining if women living in urban areas and those living in rural areas were differentially affected. To do so, we interacted our main explanatory variable, Domestic Violence Legislation (DVL) with the urban indicator. As shown in

Table 5. The Effectiveness of Domestic Violence Law—Heterogeneity 1.

	Emotional Abuse	Less Severe Violence	More Severe Violence	Sexual Abuse	Being Divorced
	(1)	(2)	(3)	(4)	(5)
DVL	−0.022 ***	−0.037 ***	−0.021 ***	−0.011 ***	−0.007 ***
	(0.002)	(0.002)	(0.002)	(0.001)	(0.001)
DVL × Urban Areas	−0.009 ***	−0.003	−0.000	−0.005 ***	−0.013 ***
	(0.003)	(0.003)	(0.002)	(0.002)	(0.002)
Observations	610,877	655,895	655,781	640,705	776,353
Residential Cluster FE	Y	Y	Y	Y	Y
Individual Characteristics	Y	Y	Y	Y	Y
Survey Characteristics	Y	Y	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refer to survey year and country fixed effects. Individual Characteristics include woman age, squared woman age, women education, partner age, squared partner age, partner education, age difference between women and partner, the number of children, whether the household is headed by a male, whether the household lives in rural areas, household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

, the domestic violence law is more effective in reducing emotional abuse, sexual abuse, and the probability of divorce for urban women than rural women. We lack statistical evidence for the impacts on less severe violence and more severe violence, as the coefficients on the interaction terms are statistically indistinguishable from zero.

We then conducted a heterogeneity analysis along the lines of woman’s education and partners’ education. We interacted the Domestic Violence Legislation indicator with the number of educational years completed by the woman and the number of educational years completed by the partner, separately. The results are displayed in

Table 6. The Effectiveness of Domestic Violence Law—Heterogeneity 2.

	Emotional Abuse	Less Severe Violence	More Severe Violence	Sexual Abuse	Being Divorced
	(1)	(2)	(3)	(4)	(5)
Panel A: Woman Education
DVL	−0.016 ***	−0.021 ***	−0.014 ***	−0.006 ***	−0.009 ***
	(0.003)	(0.003)	(0.002)	(0.002)	(0.001)
DVL × Woman Education	−0.001 ***	−0.002 ***	−0.001 ***	−0.001 ***	−0.000 ***
	(0.000)	(0.000)	(0.000)	(0.000)	(0.000)
Observations	610,877	655,895	655,781	640,705	776,353
Panel B: Partner Education
DVL	−0.022 ***	−0.021 ***	−0.017 ***	−0.009 ***
	(0.003)	(0.003)	(0.002)	(0.002)
DVL × Partner Education	−0.001 **	−0.002 ***	−0.000 **	−0.000 ***
	(0.000)	(0.000)	(0.000)	(0.000)
Observations	610,877	655,895	655,781	640,705
Residential Cluster FE	Y	Y	Y	Y	Y
Individual Characteristics	Y	Y	Y	Y	Y
Survey Characteristics	Y	Y	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refer to survey year and country fixed effects. Individual Characteristics include woman age, squared woman age, women education, partner age, squared partner age, partner education, age difference between women and partner, the number of children, whether the household is headed by a male, whether the household lives in rural areas, household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

. The more educated the woman, the more effective the domestic violence law at reducing emotional abuse, less severe violence, more severe violence, sexual abuse, and divorce (Panel A). We also detected stronger impacts of the law for women whose partners are more educated (Panel B). In other words, education increases the effectiveness of the law in decreasing women’s experience of domestic abuse of all forms and in marital dissolution.

Finally, we tested whether or not the impacts of domestic violence law differed by wealth, by adding an interaction between our main explanatory variable with a categorical variable wealth quintile. Here, wealth is measured by the wealth quintile where the household in the lowest quintile (quintile = 1) is the poorest and the household in the highest quintile (quintile = 5) is the richest in the within-country wealth distribution. Evident from

Table 7. The Effectiveness of Domestic Violence Law—Heterogeneity 3.

	Emotional Abuse	Less Severe Violence	More Severe Violence	Sexual Abuse	Being Divorced
	(1)	(2)	(3)	(4)	(5)
DVL	−0.014 ***	−0.019 ***	−0.020 ***	−0.007 ***	−0.009 ***
	(0.003)	(0.003)	(0.002)	(0.002)	(0.002)
DVL × Wealth Quintile	−0.004 ***	−0.006 ***	−0.000	−0.002 ***	−0.001 ***
	(0.001)	(0.001)	(0.001)	(0.001)	(0.001)
Observations	610,877	655,895	655,781	640,705	776,353
Residential Cluster FE	Y	Y	Y	Y	Y
Individual Characteristics	Y	Y	Y	Y	Y
Survey Characteristics	Y	Y	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refers to survey year and country fixed effects. Individual Characteristics includes woman age, squared woman age, woman education, partner age, squared partner age, partner education, age difference between woman and partner, the number of children, whether the household is headed by a male, whether the household lives in a rural area, household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

, the domestic violence law is more effective at decreasing emotional abuse, less severe violence, more severe violence, sexual abuse against women, and divorce in relatively richer households.

Collectively, women living in urban areas, being more educated, having more educated partners, and coming from relatively richer households, tend to benefit more from domestic violence legislation. In other words, the disadvantaged population, i.e., rural women, poorly educated women, women having poorly educated spouses, and women from relatively poorer households, might receive less protection from domestic violence law.

5.3. Robustness Checks

In this section, we conducted several robustness checks to test the sensitivity of our main results. Firstly, we applied the sampling weight to our regressions. We carefully noted that prior studies such as the works of Winship and Radbill [25], Gelman [26], and Solon, Haider and Wooldridge [27] mention lower efficiency and statistical power as the consequences of weighting. As shown in Panel A of

Table 8. The Effectiveness of Domestic Violence Law—Robustness.

	Emotional Abuse	Less Severe Violence	More Severe Violence	Sexual Abuse	Being Divorced
	(1)	(2)	(3)	(4)	(5)
Panel A: Weighted Regression
DVL	−0.027 ***	−0.039 ***	−0.022 ***	−0.013 ***	−0.013 ***
	(0.002)	(0.002)	(0.001)	(0.001)	(0.001)
Observations	610,119	655,137	655,023	639,947	775,571
Panel B: Narrow Window (5 Years)
DVL	−0.019 ***	−0.036 ***	−0.017 ***	−0.009 ***	−0.012 ***
	(0.003)	(0.003)	(0.002)	(0.002)	(0.001)
Observations	166,552	180,663	180,650	175,517	213,042
Panel C: Household Fixed Effects
DVL	−0.041 *	−0.067 ***	−0.060 ***	−0.040 ***	−0.027 **
	(0.023)	(0.020)	(0.013)	(0.013)	(0.013)
Observations	7437	12,352	12,352	10,946	25,599
Residential Cluster FE	Y	Y	Y	Y	Y
Individual Characteristics	Y	Y	Y	Y	Y
Survey Characteristics	Y	Y	Y	Y	Y

* p < 0.1, ** p < 0.05, *** p < 0.01. Robust standard errors are clustered at the residential cluster. Each column represents the coefficient in a separate regression. Survey Characteristics refers to survey year and country fixed effects. Individual Characteristics includes woman age, squared woman age, woman education, partner age, squared partner age, partner education, age difference between woman and partner, the number of children, whether the household is headed by a male, whether the household lives in a rural area, household wealth quintile, woman and partner birth year fixed effects, woman and partner country-specific birth cohort trends. Residential Cluster FE refers to the fixed effects at the residential cluster level.

, we still found negative impacts of the domestic violence legislation on women’s victimization by partners and marital dissolution. Estimates are statistically distinguishable from zero.

Secondly, we considered only women who married five years before and five years after the implementation of the domestic violence law. For example, since the domestic violence legislation became effective in Angola in 2013, in this exercise we only included in our sample Angolan women who married from 2008 to 2018. Narrowing the timing window can exclude other potential societal changes which might contaminate the impacts of the legislation. Doing so leaves our results virtually unchanged, despite the loss of observation (Panel B of Table 8).

Finally, we adopted the household fixed effects model. In this model, we compared the domestic violence outcomes of women who married after the implementation of the domestic violence law with the outcomes of women who married before the implementation of the law, living in the same household. Therefore, common characteristics among women from the same household that might simultaneously be correlated with marriage timing and domestic abuse could be controlled for. The estimated results are displayed in Panel C of Table 8. The household fixed effects estimates are still negative and statistically significant, although the sample size decreases considerably. In other words, our results are insensitive to the employment of the household fixed effects specification.

6. Discussion and Conclusions

In summary, we have presented evidence on the impacts of domestic violence legislation on women’s experience of domestic abuse and divorce. Specifically, the implementation of the legislation decreases women’s victimization by their partners as shown by the declines in the composite indices of emotional abuse, less severe violence, more severe violence, and sexual abuse by 0.03, 0.04, 0.02, and 0.01 points, respectively. These estimates correspond to the declines by 13.6, 14.4, 19.6, and 11.5% relative to the sample averages. As women tend to suffer less from intimate violence, the probability of divorce falls. More specifically, women are also 1.3 percentage points less likely to be divorced, representing a 15.8% reduction in the proportion of divorced women.

We also examined the source of the decrease in women’s experience of domestic abuse by separately estimating the effects of the domestic violence law on each of the individual items that constitute the composite indices. Regarding emotional abuse, women who married after the domestic violence law was implemented, report lower incidences of being humiliated, threatened with harm, and insulted by their partners in the last 12 months, by 1.9, 1.0, and 1.9 percentage points, respectively. Regarding less severe violence, the domestic violence law reduces the probability of women being pushed/shook, slapped, punched/hit, and having arm twisted/hair pulled by their partners in the past 12 months by 2.4, 3.8, 1.8, and 1.1 percentage points, respectively. With regard to more severe violence, the implementation of the law decreases the incidences of being kicked/dragged, strangled/burnt, and threatened with a knife/gun by partners in the last 12 months by 2.0, 0.6, and 0.5 percentage points, respectively. In the case of sexual abuse, the law is associated with the 1.3 and 0.4 percentage point decreases in the likelihood of women being forced into unwanted sex, and other sexual acts, respectively.

We also detected heterogeneity in the impacts of the legislation. Specifically, the decreases in emotional abuse, less severe violence, more severe violence, sexual abuse, and divorce are greater among women living in urban areas, those being highly educated, those having highly educated partners, and those from relatively richer households. In other words, the disadvantaged population, i.e., rural women, poorly educated women, women having poorly educated spouses, and women from relatively poorer households, might receive less protection from domestic violence law.

Our findings are consistent with the literature on women’s empowerment and gender equality [8,9]. In particular, positive treatments such as serving as a business mentor and receiving training in microfinance, improve women’s sense of empowerment [12,13]. Our results are also in line with those from studies on the impacts of government policies that specifically target intimate relationships in women. Specifically, it is documented that aggressive legislation that places heavy sanctions on intimate violence offenders in the U.S. raises the probability of reporting to the police and reduces the incidence of violence against women [22]. In the context of Mexico, criminalizing domestic violence decreased female suicide rates and women’s victimization [24]. This paper further complements prior studies on factors affecting women’s experience of domestic violence. Factors such as education, engagement in market work, and ownership of economic resources can reduce the likelihood of abuse by partners [14,15,16,18]. Nevertheless, events such as natural disasters and armed conflict might expose women to more domestic violence [19,21].

The impacts of domestic violence legislation on women could potentially be explained as follows. It is possible that the legislation changes household proclivity toward intimate violence. The consequential decline in intimate violence against women and divorce might be because would-be offenders who perceive a high cost to domestic violence might abstain from displaying aggression. Another possibility is that, as the legislation criminalizes domestic violence, women are less likely to accept violence. The implementation of the domestic violence law could increase the likelihood of victims reporting to the police, thus decreasing women’s experience of domestic violence. In brief, domestic violence against women can be reduced if countries adopt legislation that protects victims and sanctions offenders.

Our findings emphasize the meaningful role of the domestic violence legislation in protecting women. Since domestic violence or intimate violence can be regarded as serious harm within intimate relationships that could produce immense consequences on women’s health, work, economic independence, and citizenship rights [28], protecting women from intimate violence can significantly raise their quality of life. In particular, women who are protected by domestic violence legislation might be less likely to experience unemployment and earning loss [29]. They may also have better physical and psychological health, compared with those exposed to intimate violence more frequently [3,4]. Moreover, women protected by domestic violence legislation could have a lower risk of being financially controlled by partners [30]. By decreasing spousal aggression against women, the legislation might facilitate women’s access to safe housing and participation in community life, improving women’s rights to citizenship [31]. More importantly, domestic violence legislation can also protect children, since witnessing violence might interrupt the child’s socio-emotional development, leading to poor health and mental distress in the future [5,6].

Our study suggests that countries should adopt legislation targeting domestic violence, as a way to protect women and children. More reforms in the legislative systems that aggressively pursue domestic violence offenders are desirable. Such legislation needs to be fully enforced especially in rural areas, because our heterogeneity analysis shows that rural women tend to receive less protection compared with urban women. Lawmakers should also pay more attention to other disadvantaged groups such as poorly educated women, women having poorly educated partners, and women from relatively poor households, since they tend to be more vulnerable to domestic violence.