How does empowering men affect intimate partner violence? Evidence from Bangladesh

Abstract

This paper investigates the relationship between women's and men's empowerment and intimate partner violence (IPV). To do so, it uses a sample of 4548 households with a husband and a wife who are present in two or three rounds of the Bangladesh Integrated Household Survey (BIHS). Measures of empowerment are constructed separately and symmetrically for husbands and wives. Conditional fixed-effects logistic regressions are employed to examine the statistical association between measures of empowerment and intimate partner violence. Findings indicate that IPV is widespread in rural Bangladesh, as 67.6% and 22.4% of households report experiencing verbal and physical IPV, respectively, in at least one round. Husbands tend to be more empowered than their wives in most, but not all, dimensions. Yet, even in dimensions of empowerment dominated by men on average, the percentage of households in which individual wives are more empowered than their husbands is notable. In regressions, some measures of men's empowerment (e.g., ownership of assets) are negatively associated with verbal and any IPV. Some measures of women's empowerment (e.g., community influence) are associated with verbal, physical, and any IPV, but the signs are mixed. All in all, the findings suggest that men's empowerment may be a determinant of intimate partner violence in rural Bangladesh, and they also underscore the need to extend theories of empowerment.

Highlights

• •Little is known on the link between men's empowerment and intimate partner violence.
• •This paper uses longitudinal data from Bangladesh to investigate the question.
• •Measures of men's empowerment are negatively associated with verbal/any IPV.
• •The findings suggest that men's empowerment may be a determinant of IPV.
• •They also underscore the need to extend theories of empowerment.

1. Introduction

The prevalence of intimate partner violence (IPV) varies considerably across the world, and it remains a critical public health issue (Coll et al., 2020). Motivated by the necessity to address this global problem, quantitative research investigates the determinants of IPV in developing countries. Studies highlight sociodemographic factors like women's education (Ahinkorah et al., 2018; Oluwagbemiga et al., 2023; Rahman et al., 2020a), women's age at marriage (Murshid, 2017; Terrazas-Carrillo & McWhirter, 2015; Tsegaw et al., 2022), and exposure to violence during childhood (Abramsky et al., 2011; Murshid & Murshid, 2018). Studies also highlight contextual factors like patriarchal norms (Heise & Kotsadam, 2016; Murshid & Critelli, 2020), religious fractionalization (Kaya & Cook, 2010), and community inequality (Munguia & Martinez-Zarzoso, 2022).

Additionally, measures of women's economic status and empowerment are linked to the likelihood of experiencing IPV. Women's employment and earnings are positively associated with IPV in many studies (Bhalotra et al., 2021; Bourey et al., 2023; Heath, 2014; Krishnan et al., 2010; Stöckl et al., 2021; Terrazas-Carrillo & McWhirter, 2015), though not in all of them (Chin, 2012, Vyas et al., 2015; Raj et al., 2018). Results are mixed with respect to women's receipt of economic transfers (Angelucci, 2008; Canedo & Morse, 2021; Leite et al., 2022) and women's participation in microcredit programs (Green et al., 2015; Ranganathan et al., 2021; Sato et al., 2022; Yount et al., 2021). Most studies find that women's decision-making power, freedom of movement, and asset ownership are negatively associated with intimate partner violence (Bourey et al., 2013; Chowdhury et al., 2018; Raj et al., 2018; Ranganathan et al., 2021; Sabarwal et al., 2014; Svec & Andic, 2018; Treves-Kagan et al., 2022). Yet, some studies do not find these dimensions of empowerment are protective (e.g., Forty, 2022; Murshid & Critelli, 2020; Tenkorang, 2018).

Scholars and policymakers have long recognized men's complex role in their perpetuation of gender violence as well as their participation in programs to reduce it (Jewkes et al., 2015). For example, studies have examined men's alcohol consumption and unemployment (e.g., Angelucci, 2008; Bhalotra et al., 2021; Krishnan et al., 2010; Oluwagbemiga et al., 2023). However, the literature has devoted little attention to exploring whether empowering men would reduce, or even raise, violence against intimate partners.

The women's empowerment literature understands empowerment as the process of gaining control over decisions and resources that determine one's ability to make critical life choices (Kabeer, 1999). This conceptualization is a useful starting point for defining men's empowerment. Accordingly, quantitative papers have operationalized measures of empowerment symmetrically for men and women. Several studies look at the relationship between development outcomes and either the empowerment gap between spouses (Kumar et al., 2021; Seymour, 2017; Sraboni et al., 2014) or spousal empowerment measured separately for husbands and wives (Crookston et al., 2021; Malapit et al., 2019; Ntakyo & Van Den Berg, 2022; Osiewalska, 2018). One study looks at the relationship between husbands' and wives' empowerment (Akter & Francis-Tan, 2021).

But nearly all research on empowerment and intimate partner violence focuses on women's empowerment. A few studies, nevertheless, do explore the role of men's empowerment, even if indirectly. Men's unemployment (Bhalotra et al., 2021; Krishnan et al., 2010), lower relative status (Goodman et al., 2017; Luke et al., 2007), and lower life satisfaction (Yount et al., 2016) appear to raise the likelihood of perpetuating IPV. This evidence may suggest that men's empowerment is inversely associated with IPV. There are further reasons to believe this may be the case.

Current theories maintain that men who subscribe to ideologies of male superiority use violence against women as a tool to maintain gender inequalities (Fulu et al., 2013; Jewkes et al., 2015). Thus, if men's empowerment brings economic security, education, and/or egalitarian beliefs, intimate partner violence may decrease. Other perspectives are consistent with the conjecture that men's empowerment may be inversely related to IPV. It may be that violence is an outlet for frustration regarding one's low socioeconomic status, or that violence is a substitute or compensation for one's poor self-actualization, particularly if male violence is socially acceptable.

This paper uses longitudinal data to investigate the statistical association between empowerment and IPV in Bangladesh. It leverages the 2011, 2015, and 2018 rounds of the Bangladesh Integrated Household Survey (BIHS), a longitudinal survey of persons residing in rural areas of the country. Measures of empowerment, inspired by the Women's Empowerment in Agriculture Index, are constructed separately for wives and husbands. The paper adopts a logistic regression model with household-level fixed effects to examine the sample of households reporting IPV in some, but not all, survey rounds. The advantage of the method is that it is able to control for time-invariant, observed and unobserved characteristics.

This paper contributes to the literature on intimate partner violence in developing countries. It is the first to estimate the statistical association between measures of men's empowerment and intimate partner violence using longitudinal methods intended to reduce confounding. Thus, it joins a rising trend in the literature to consider the causes and consequences of men's empowerment. The paper not only adds to descriptive knowledge about the correlates of IPV in a context where gender inequality is relatively high, but it also raises underexplored questions regarding the theory of women's and men's empowerment.

2. Previous research on IPV in Bangladesh

Bangladesh is a crucial setting in which to study empowerment and IPV. Gender inequality has been decreasing in recent decades due in part to policies and programs meant to address gender differences in rights and resources (Shamsuddin, 2015). While some men are accepting of changing gender norms, others feel resentment as they view women's rise as a threat to their identity (Karim et al., 2018; Schuler, Lenzi, et al., 2018). Traditional gender norms remain deeply entrenched, especially in rural areas (Schuler, Lenzi, et al., 2018).

The earliest quantitative studies focus on the role of microcredit programs. Based on a 1992 survey of married women in six villages in Bangladesh, Schuler et al. (1996) find that participation in microcredit programs is negatively associated with IPV. However, their qualitative evidence (Schuler et al., 1998) paints a complex picture that mirrors conflicting results from other early studies (Bates et al., 2004; Koenig et al., 2003; Naved & Persson, 2005). Koenig et al. (2003) report that at the community level, women's participation in savings/credit groups is negatively associated with IPV, but at the individual level, it is positively associated. Later studies aim to understand why the findings are mixed. Some make use of propensity score matching and report no statistically significant association between microcredit and IPV (Bajracharya & Amin, 2013; Yount et al., 2021). Others make use of stratified analyses and find a significant association only for certain subgroups of women (Dalal et al., 2013; De & Christian, 2020; Murshid et al., 2016). For example, Dalal et al. (2013) discover participation in microcredit programs and IPV are positively associated only for women who are more educated and for those who are from wealthier households.

Beyond microcredit, research investigates economic factors. Two factors emerge repeatedly in the literature: woman's employment and household economic status. A number of studies report that woman's employment is positively associated with IPV (Bates et al., 2004; De & Christian, 2020; Heath, 2014; Naved & Persson, 2005; Rahman et al., 2013). Naved and Persson (2005) find that wife's income earning is positively associated with IPV in rural but not urban areas, while Heath (2014) finds that only wives who have a lower age at marriage or have a lower education face a higher risk of IPV when participating in the labor force. Studies also report that household economic status (often measured by assets, landholdings, and/or access to household services) is negatively associated with IPV (Bates et al., 2004; Haque et al., 2022; Koenig et al., 2003; Murshid et al., 2016; Rahman et al., 2011; Schuler, Field, & Bernholc, 2018). For example, using a 2016–2017 survey of households in two rural districts, Haque et al. (2022) find that household economic status (measured by assets, access to electricity, and income) is inversely associated with IPV, and household food insecurity is positively associated.

Researchers have extensively explored the contribution of education and demographic factors. Of the wide range of demographic characteristics, a handful are clear and consistent correlates of IPV. Both wife's and husband's level of education is negatively associated with the prevalence of IPV (Bates et al., 2004; Koenig et al., 2003; Murshid, 2017; Murshid et al., 2016; Naved & Persson, 2005; Rahman et al., 2013; Schuler et al., 2018). Based on a 2014 survey conducted in the same villages studied by Schuler et al. (1996), Schuler et al. (2018) find that a wife's likelihood of reporting IPV decreases as her education increases. Additionally, studies discover that wife's age at marriage is negatively associated with IPV (De & Christian, 2020; Haque et al., 2022; Murshid, 2017; Rahman et al., 2013), and that husband's childhood exposure to maternal marital violence is positively associated with his perpetration of IPV (Murshid, 2017; Murshid & Murshid, 2018; Naved & Persson, 2005). Using the 2015 Bangladesh Violence Against Women Survey, Rahman et al. (2020a) take an intersectional approach to understanding patterns of physical IPV. They find that married women in younger age and lower education locations and married women in higher education and poorer household locations had an elevated risk of IPV. Complementing these results, Rahman et al. (2020b) examine interactions between community-level factors. They find that married women living in communities with higher women's education and labor force participation were more likely to report IPV.

The literature is increasingly examining the connections between women's empowerment and IPV. Most studies consolidate multiple variables to construct a single index of empowerment. Dimensions of empowerment vary across studies, but they tend to focus on decision-making authority (e.g., input on decisions about large household purchases) and autonomy of movement (e.g., can visit family or relatives without permission from husband). In short, the results are mixed. One study does not find a significant relationship between autonomy and IPV (De & Christian, 2020). Three others report that women's empowerment is positively associated with IPV (Koenig et al., 2003; Murshid et al., 2016; Rahman et al., 2011). For example, Murshid et al. (2016) examine the 2007 Bangladesh Demographic and Health Survey and find that the higher is women's autonomy (regarding access to health services), the higher is their likelihood of experiencing IPV. They do not find a significant association with woman's decision-making power. Yet, three studies report that empowerment is negatively associated with IPV (Haque et al., 2022; Rahman et al., 2013; Schuler et al., 2018). For example, based on the 2007 Bangladesh Demographic and Health Survey, Rahman et al. (2013) find that married women with higher overall autonomy are less likely to experience IPV. Exploring the components of autonomy, they report that IPV is negatively associated with economic decision-making power, negatively associated with attitudes toward IPV, positively associated with freedom of movement, and unassociated with health and family planning autonomy.

In summary, there is considerable quantitative research on IPV in Bangladesh. Much of it leverages cross-sectional surveys and centers on predictors including microcredit participation, economic status, demographic factors, and women's empowerment. While the evidence on some correlates of IPV (e.g., education) is unambiguous, the evidence on other correlates (e.g., women's empowerment) is ambiguous. Moreover, the IPV literature has paid inadequate attention to the role of men's empowerment. This paper's aim is to examine the association between women's and men's empowerment and IPV in Bangladesh.

3. Methods

3.1. Data

This paper uses the Bangladesh Integrated Household Survey (BIHS). The International Food Policy Research Institute (IFPRI) designed the survey to study development issues in Bangladesh as well as to evaluate the U.S. government's Feed the Future initiative (Ahmed, 2013; IFPRI 2016; IFPRI 2020). Round 1 was conducted during 2011–2012 in 64 districts and 279 villages of rural Bangladesh. Subsequent rounds, i.e., round 2 conducted during 2015 and round 3 conducted during 2018–2019, were administered to the same households that participated in previous rounds and to new households that split from the initial ones. Of those households interviewed in round 1, about 92% were re-interviewed in round 2 and about 72% were re-interviewed in round 3. The BIHS is representative of rural Bangladesh and the seven administrative divisions of the country.

Interviews were done in person by two interviewers, one male and one female. The survey questionnaire covers a large set of themes from employment and agricultural production to food consumption and nutrition. Notably, it includes a module on domestic violence, which is administered to wives, and a module on empowerment, which is administered to both husbands and wives. The estimation sample consists of households containing both a husband and a wife with non-missing variables in at least two survey rounds. Altogether, the sample size is 11,882 household-rounds, composed of 2786 households that appear in three rounds and 1762 households that appear in two rounds. Fig. S1 in the Supplemental File provides additional details about the determination of the sample size.

3.2. Dependent variables

Separately from other household members, female survey enumerators interviewed married women about their experiences with domestic violence in the past year. Respondents were asked (a) whether their husband threatened them with divorce, (b) whether their husband threatened them with taking another wife, (c) whether their husband or another household member verbally abused them, and (d) whether their husband or another household member physically abused them.

Only in round 3, respondents were asked separate questions about the abuse committed by their husband, other adult male household members, and adult female household members. Information from this round shows that husbands committed the vast majority of abuse. For those women who reported any verbal abuse in round 3, 97.4% reported abuse by their husbands specifically. For those who reported any physical abuse in round 3, 100% reported abuse by their husbands specifically. Thus, the questions about domestic violence are essentially questions about intimate partner violence.

Three measures of intimate partner violence serve as dependent variables in the analysis:

Verbal violence. This variable is equal to one if a respondent said that in the past year their husband threatened them with divorce or with taking another wife or their husband or another household member verbally abused them; it is equal to zero otherwise.

Physical violence. This variable is equal to one if a respondent said that in the past year their husband or another household member physically abused them; it is equal to zero otherwise.

Any violence. This variable is equal to one if a respondent said that in the past year they experienced verbal or physical violence; it is equal to zero otherwise.

3.3. Independent variables

The set of empowerment variables was inspired by the framework used by the Women's Empowerment in Agriculture Index or WEAI (Alkire et al., 2013). The WEAI framework identifies five domains or dimensions of empowerment: production, resources, income, leadership, and time. One or more indicators, which map to questions in the BIHS, belong to each domain. An overall empowerment score is calculated by applying specific aggregation and cut-off rules to dichotomize each indicator (whether or not “inadequate”) and combining the dichotomized indicators with numerical weights. This paper's objective is not to estimate an overall empowerment score for individuals in the sample, so its approach departs from the WEAI in that the indicators are not dichotomized or combined. Thus, the measures of empowerment are not binary but continuous (except for one), which allows for greater variation across households, and they are not combined but separate, which allows for correlations across measures.

Seven measures of empowerment are independent variables in the analysis.1 They are constructed identically but separately for husbands and wives. Note the measures are not zero-sum. For each household, the relationship between the husband's measure and the wife's measure is not constrained artificially. The variables are defined as follows:

Input in productive decisions. This variable is equal to a person's average input in making decisions across six income-generating activities (food crop farming, cash crop farming, livestock raising, non-farm economic activities, wage and salary employment, and fishing). For each activity, a person is assigned a value of 0 if no input or input into very few decisions, 1 if input into some decisions, and 2 if input into most or all decisions. If one spouse reports participation in an activity while the other spouse reports no participation, the one who reports no participation is assigned a value of 0. If both spouses report no participation in an activity, the activity is not included in their averages.

Ownership of assets. This variable is equal to a person's average ownership across fourteen items of productive capital (e.g., land, livestock, farm equipment, house, consumer durables, cell phone). For each item, a person is assigned a value of 0 if he or she says the household owns the item but does not list him or herself as sole or joint owner, and 1 if a person lists him or herself as sole or joint owner. If one spouse says the household owns the item while the other spouse says the household does not, a zero value is assigned to the spouse who reports no household ownership. If both spouses report the household does not own the item, the item is not included in their averages.

Input in credit decisions. This variable is equal to a person's average input in making decisions to borrow across five sources of credit (non-governmental organization, informal lender, formal lender, friends/relatives, ROSCA or credit group). For each source, a person is assigned a value of 0 if he or she says the household took a loan but does not list him or herself as sole or joint decision-maker, and 1 if a person lists him or herself as sole or joint decision-maker. If one spouse says the household took a loan while the other spouse says the household did not, a zero value is assigned to the spouse who reports no loan. If both spouses report the household did not take a loan from the source, the source is not included in their averages.2

Community influence. This variable is equal to a person's response to a question about their perceived influence on their local community. Respondents are shown a picture of a ladder with nine steps. They are asked to place themselves on the ladder between steps 1 and 9, where 1 indicates no influence on the community and 9 indicates highest influence on the community.

Group membership. This variable is equal to 1 if a person is an active member of at least one community group (e.g., agricultural producer's group, mutual help group, local government, religious group) and is equal to 0 if a person is not. Since most persons who belong to a group belong to only one, it makes more sense for this variable to capture the extensive margin rather than the intensive margin.

Speaking in public. This variable is equal to a person's average level of comfort with speaking up in public when the community needs to make a decision on an important issue (deciding on infrastructure to be built in the community, ensuring proper payment of wages for public works, and protesting the misbehavior of authorities or elected officials). For each issue, a person is assigned a value of 0 if he or she is not comfortable speaking up in public, and 1 if he or she is comfortable speaking up in public.

Satisfaction with leisure. This variable is equal to a person's response to a question about their satisfaction level with their available time for leisure activities like watching TV, listening to radio, seeing movies, or doing sports. Respondents are asked to select a response from 1 to 10, where 1 indicates not satisfied, 5 indicates neither satisfied nor dissatisfied, and 10 indicates very satisfied.

In the primary analysis, the set of husband's empowerment variables and the set of wife's empowerment variables are included as independent variables in regression models. In the secondary analysis, a single index of husband's empowerment and a single index of wife's empowerment are the independent variables. Separately for husbands and wives, each of the empowerment variables is standardized so that its mean is zero and standard deviation is one. Then, the standardized variables are averaged to obtain a single index. Note these indexes are not equivalent to WEAI empowerment scores, since not all WEAI indicators are used to form the indexes and the standardized measures are weighted equally.

3.4. Covariates

Time-varying covariates are constructed for the analysis. These variables reflect the spouses’ capabilities (employment, income), their asset endowments (non-land assets and land holdings), and their household size. Note that current employment equals 1 if a person worked for pay in the past week receiving remuneration from salary, wage, self-employment, farming, or animal husbandry and equals 0 otherwise. Log income is the logarithm of average monthly income from all sources including salary, wage, self-employment, farming, and animal husbandry; if a person does not have any income, they are assigned a value of zero. The value of household assets and the amount of land owned are each placed into quintiles calculated by survey round. Table S1 in the Supplemental File displays summary statistics for covariates used in the analysis.

3.5. Statistical models

Stata version 18 was used to perform the statistical analyses. Conditional fixed-effects logistic regressions are employed to examine the statistical relationship between empowerment and intimate partner violence within households (Greene, 2012). The main advantage of the model is that fixed effects control for all time-invariant factors at the individual and household levels, whether observed or unobserved by researchers. Note that only time-varying covariates may be included in fixed effects models; time-invariant characteristics like education, birth year, and spousal age gap are not necessary (and not possible) to include. Note also that sample size decreases in fixed effects models because only households that exhibit changes in IPV across time are used to estimate the coefficients.

That is, the following model is implemented for household h in survey round r:

${IPV}_{h,r}={\beta }_1{HE}_{h,r}+{\beta }_2{WE}_{h,r}+{\beta }_3{X}_{h,r}+{\delta }_r+{\theta }_h+\epsilon$, where IPV is intimate partner violence, HE is husband's empowerment, WE is wife's empowerment, X is a vector of time-varying covariates, $\delta$ is a survey round fixed effect, and $\theta$ is a household fixed effect (Greene, 2012). Survey round fixed effects account for time-varying, nation-level variables associated with IPV.

The main tables report adjusted odds ratios and 95% confidence intervals for empowerment variables, while Table S2 reports estimates for covariates. Stata version 18 was used to produce models, regressing each type of IPV on empowerment variables, with and without individual and household covariates. The tables also report two diagnostic tests: the Hausman and Likelihood ratio (LR) tests. The former is a specification test that compares models with random and fixed effects to determine the necessity to include fixed effects, while the latter is a goodness of fit test that compares nested models to determine the predictive value of additional independent variables.

Three robustness exercises are conducted. Table S3 makes use of two alternative measures of empowerment. For input in productive decisions and input in credit decisions, if a spouse reports no participation in an activity or reports the household did not take a loan from a source, the activity/source is not included in his or her average (regardless of what the other spouse reports). Table S4 makes use of alternative measures of IPV. For round 3, the measures of verbal, physical, and any violence pertain specifically to violence committed by husbands, a refinement which the data allows. Technically, this creates an inconsistency between the definition of round 3 variables and the definition of round 1 and 2 variables, but practically, the difference is trivial because husbands committed the vast majority of abuse. Table S5 stratifies the models according to the number of survey rounds that households appear in the estimation sample. Taken together, the results from the robustness exercises are comparable to the results from the primary analysis.

4. Results

4.1. Descriptive statistics

Before presenting the main results, it is useful to present descriptive statistics on the households in the sample, reported prevalence of IPV, and empowerment variables. Table 1 provides background characteristics for the 4548 households in the estimation sample. Almost 88% of them are Muslim. On average, husbands are 8 years older than their wives. The schooling gap between spouses is close to zero, in part because many respondents (48% of husbands and 41% of wives) do not have any education at all.

Table 1.

Background characteristics of households.

	N	Mean or %	95% CI
Muslim %	4548	87.77	[86.82, 88.73]
Age at baseline (husband)	4548	42.46	[42.08, 42.83]
Age at baseline (wife)	4548	34.23	[33.90, 34.56]
Age gap between spouses	4548	8.22	[8.09, 8.36]
No education (husband) %	4548	47.67	[46.22, 49.12]
No education (wife) %	4548	41.07	[39.64, 42.50]
Years of schooling (husband)	4548	3.43	[3.31, 3.55]
Years of schooling (wife)	4548	3.52	[3.42, 3.63]
Schooling gap between spouses	4548	−0.09	[-0.19, 0.01]

NOTE. Data source is Bangladesh Integrated Household Survey 2011–2012 (round 1), 2015 (round 2), and 2018–2019 (round 3).

Table 2 displays the reported prevalence of IPV in the past year by survey round. Violence is widespread in rural Bangladesh. 68.0% of households report any type of violence in at least one round. Verbal violence is more common than physical violence. 67.6% and 22.4% of households report verbal and physical IPV, respectively, in at least one round. As the table shows, the prevalence of verbal violence is increasing with time. For households appearing in three rounds, 7.4% report verbal violence in only round 1, 13.6% report in only round 2, and 17.4% in only round 3. However, the prevalence of physical violence does not exhibit a clear trend. For households appearing in three rounds, the prevalence of physical violence seems to peak in round 2.

Table 2.

Reported prevalence of verbal, physical, and any IPV in the past year by survey round.

	Verbal		Physical		Any
	%	N	%	N	%	N
For households in two or three rounds
Violence reported in no rounds	32.45	(1476)	77.59	(3529)	31.97	(1454)
At least one round	67.55	(3072)	22.41	(1019)	68.03	(3094)
Total	100%	(4548)	100%	(4548)	100%	(4548)
For households in three rounds
Violence reported in no rounds	25.41	(708)	76.53	(2132)	25.09	(699)
Only round 1	7.43	(207)	6.10	(170)	7.57	(211)
Only round 2	13.64	(380)	7.82	(218)	13.50	(376)
Only round 3	17.41	(485)	4.77	(133)	17.12	(477)
Rounds 1 and 2	5.10	(142)	1.69	(47)	5.35	(149)
Rounds 1 and 3	7.72	(215)	0.75	(21)	7.72	(215)
Rounds 2 and 3	14.54	(405)	1.65	(46)	14.54	(405)
All rounds	8.76	(244)	0.68	(19)	9.12	(254)
Total	100%	(2786)	100%	(2786)	100%	(2786)
For households in two rounds
Violence reported in no rounds	43.59	(768)	79.28	(1397)	42.85	(755)
Only earlier round	12.37	(218)	8.06	(142)	12.66	(223)
Only later round	27.41	(483)	9.42	(166)	27.19	(479)
Both rounds	16.63	(293)	3.23	(57)	17.31	(305)
Total	100%	(1762)	100%	(1762)	100%	(1762)

NOTE. Data source is Bangladesh Integrated Household Survey 2011–2012 (round 1), 2015 (round 2), and 2018–2019 (round 3).

Table 3 displays summary statistics for the empowerment variables.3 On average, men dominate in most dimensions of empowerment. For five measures, the percentage of households where husbands are more empowered than their wives is greater than the percentage of households where wives are more empowered than husbands. For example, husbands report higher empowerment in asset ownership than their wives in 86.1% of households. Yet, for two measures, the percentage of households where wives are more empowered than husbands is greater than the percentage of households where husbands are more empowered than wives. For example, wives report higher satisfaction with leisure than their husbands in 45.6% of households. Even in the dimensions in which men tend to have relatively more power on average, the percentage of households where individual wives are more empowered than their husbands is nontrivial. For example, wives are more comfortable speaking in public in 17.3% of households.

Table 3.

Summary statistics for empowerment variables.

		Husband relative to wife (%)			Husband		Wife
	N	Higher	Equal	Lower	Mean	SD	Mean	SD	Min	Max
Input in productive decisions	11,882	72.02	14.12	13.85	1.61	0.52	0.87	0.62	0	2
Ownership of assets	11,882	86.06	7.54	6.40	0.76	0.19	0.37	0.22	0	1
Input in credit decisions	11,882	31.06	56.18	12.77	0.82	0.32	0.67	0.43	0	1
Community influence	11,882	43.01	23.98	33.01	3.63	1.79	3.37	1.79	1	9
Group membership	11,882	9.45	63.55	27.00	0.20	0.40	0.38	0.49	0	1
Speaking in public	11,882	41.52	41.22	17.25	0.60	0.44	0.40	0.46	0	1
Satisfaction with leisure	11,882	36.90	17.48	45.62	5.59	2.41	5.89	2.72	1	10
Empowerment index	11,882				0.00	0.45	0.00	0.48	−2	1.5

NOTE. Data source is Bangladesh Integrated Household Survey 2011–2012 (round 1), 2015 (round 2), and 2018–2019 (round 3).

4.2. Association between empowerment and IPV

Table 4 displays the results from conditional fixed-effects logistic regressions of intimate partner violence on the full set of empowerment variables. Patterns are roughly the same with and without covariates. The LR tests suggest that the introduction of covariates adds some, but not much, explanatory power.

Table 4.

Conditional fixed-effects logistic regressions of verbal, physical, and any IPV in the past year on husband's and wife's empowerment (odds ratios with 95% confidence intervals in brackets).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	Verbal	Verbal	Physical	Physical	Any	Any
Input in productive decisions (husband)	1.001	0.990	0.989	0.979	1.001	0.988
	[0.897,1.117]	[0.884,1.107]	[0.838,1.168]	[0.826,1.160]	[0.898,1.116]	[0.884,1.104]
Ownership of assets (husband)	0.644*	0.651*	0.969	0.931	0.661*	0.665*
	[0.480,0.864]	[0.484,0.876]	[0.606,1.550]	[0.579,1.495]	[0.493,0.886]	[0.495,0.893]
Input in credit decisions (husband)	0.860	0.850	0.809	0.797	0.841	0.832*
	[0.723,1.023]	[0.714,1.012]	[0.613,1.068]	[0.602,1.056]	[0.707,1.000]	[0.699,0.990]
Community influence (husband)	0.912*	0.911*	0.988	0.993	0.915*	0.914*
	[0.878,0.947]	[0.877,0.946]	[0.928,1.052]	[0.932,1.058]	[0.881,0.950]	[0.880,0.949]
Group membership (husband)	0.954	0.953	0.908	0.908	0.949	0.947
	[0.830,1.098]	[0.828,1.097]	[0.724,1.138]	[0.723,1.141]	[0.826,1.091]	[0.823,1.090]
Speaking in public (husband)	0.986	0.978	0.978	0.978	0.985	0.976
	[0.864,1.125]	[0.857,1.116]	[0.794,1.207]	[0.791,1.209]	[0.863,1.124]	[0.855,1.114]
Satisfaction with leisure (husband)	0.940*	0.939*	0.996	0.996	0.944*	0.942*
	[0.919,0.962]	[0.917,0.961]	[0.960,1.033]	[0.960,1.034]	[0.922,0.966]	[0.920,0.965]
Input in productive decisions (wife)	0.975	0.961	0.923	0.933	0.979	0.964
	[0.886,1.074]	[0.872,1.060]	[0.793,1.075]	[0.798,1.090]	[0.889,1.077]	[0.875,1.063]
Ownership of assets (wife)	1.194	1.165	1.911*	1.958*	1.227	1.201
	[0.907,1.574]	[0.881,1.540]	[1.237,2.951]	[1.258,3.048]	[0.932,1.615]	[0.909,1.586]
Input in credit decisions (wife)	1.011	1.021	0.736*	0.723*	1.013	1.025
	[0.877,1.165]	[0.885,1.178]	[0.588,0.922]	[0.576,0.908]	[0.879,1.168]	[0.889,1.182]
Community influence (wife)	0.954*	0.954*	0.945	0.940*	0.954*	0.953*
	[0.920,0.990]	[0.920,0.989]	[0.889,1.004]	[0.884,0.999]	[0.920,0.989]	[0.919,0.988]
Group membership (wife)	1.306*	1.289*	1.430*	1.447*	1.308*	1.287*
	[1.150,1.482]	[1.135,1.465]	[1.164,1.755]	[1.176,1.781]	[1.153,1.484]	[1.133,1.462]
Speaking in public (wife)	1.222*	1.224*	1.078	1.081	1.189*	1.193*
	[1.069,1.396]	[1.070,1.400]	[0.871,1.335]	[0.872,1.340]	[1.041,1.358]	[1.043,1.364]
Satisfaction with leisure (wife)	0.962*	0.962*	0.939*	0.939*	0.957*	0.957*
	[0.942,0.983]	[0.942,0.983]	[0.908,0.972]	[0.908,0.972]	[0.937,0.978]	[0.937,0.978]
Household fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Individual and household covariates	No	Yes	No	Yes	No	Yes
N	6904	6904	2521	2521	6903	6903
Hausman test chi2 (p-value)	65 (0.0000)	80 (0.0000)	77 (0.0000)	91 (0.0000)	58 (0.0000)	71 (0.0001)
LR test chi2 (p-value)	Model 2 vs. 1:	24 (0.0585)	4 vs. 3:	12 (0.6827)	6 vs. 5:	25 (0.0440)

NOTE. Coefficients are expressed as odds ratios. Asterisks indicate statistical significance at the 5% level. 95% confidence intervals are in brackets below point estimates. All models control for survey round and household has loan. Covariates include husband's current employment, wife's current employment, log of husband's income and log income squared, log of wife's income and log income squared, value of household assets, amount of land owned by household, and household size. Sample sizes are less than 11,882 because the estimation of fixed effects models involves only households with any variation in IPV over time. Data source is Bangladesh Integrated Household Survey 2011–2012 (round 1), 2015 (round 2), and 2018–2019 (round 3).

Three measures of husband's empowerment and four measures of wife's empowerment are significantly associated with the prevalence of verbal IPV in Model 2. For men, a one-unit increase in ownership of assets, community influence, and satisfaction with leisure is associated with a 35%, 9%, and 6% decrease, respectively, in the adjusted odds of reporting verbal IPV. For women, a one-unit increase in community influence and satisfaction with leisure is associated with a 5% and 4% decrease, while a one-unit increase in group membership and speaking in public is associated with a 29% and 22% increase, respectively.

Five measures of wife's empowerment are significantly associated with the prevalence of physical IPV in Model 4, whereas no measures of husband's empowerment are significantly associated. For women, a one-unit increase in input in credit decisions, community influence, and satisfaction with leisure is associated with a 28%, 6%, and 6% decrease in the adjusted odds of reporting physical IPV, while a one-unit increase in ownership of assets and group membership is associated with a 96% and 45% increase, respectively.

Four measures of husband's empowerment and four measures of wife's empowerment are significantly associated with the prevalence of any IPV in Model 6. For men, a one-unit increase in ownership of assets, input in credit decisions, community influence, and satisfaction with leisure is associated with a 33%, 17%, 9%, and 6% decrease, respectively, in the adjusted odds of reporting any IPV. For women, a one-unit increase in community influence and satisfaction with leisure is associated with a 5% and 4% decrease, while a one-unit increase in group membership and speaking in public is associated with a 29% and 19% increase, respectively.

Table 5 displays the results from conditional fixed-effects logistic regressions of intimate partner violence on single indexes of empowerment. The results are consistent with the results from the previous table.

Table 5.

Conditional fixed-effects logistic regressions of verbal, physical, and any IPV in the past year on husband's and wife's empowerment (odds ratios with 95% confidence intervals in brackets).

	Model 1	Model 2	Model 3	Model 4	Model 5	Model 6
	Verbal	Verbal	Physical	Physical	Any	Any
Empowerment index (husband)	0.642*	0.632*	0.860	0.854	0.646*	0.634*
	[0.561,0.735]	[0.551,0.726]	[0.693,1.068]	[0.685,1.065]	[0.565,0.740]	[0.553,0.728]
Empowerment index (wife)	1.035	1.021	0.880	0.880	1.016	1.002
	[0.907,1.182]	[0.892,1.167]	[0.712,1.088]	[0.709,1.093]	[0.890,1.160]	[0.876,1.146]
Household fixed effects	Yes	Yes	Yes	Yes	Yes	Yes
Individual and household covariates	No	Yes	No	Yes	No	Yes
N	6904	6904	2521	2521	6903	6903
Hausman test chi2 (p-value)	43 (0.0000)	68 (0.0000)	52 (0.0000)	83 (0.0000)	40 (0.0000)	61 (0.0000)
LR test chi2 (p-value)	Model 2 vs. 1:	27 (0.0313)	4 vs. 3:	10 (0.7981)	6 vs. 5:	28 (0.0198)

NOTE. Coefficients are expressed as odds ratios. Asterisks indicate statistical significance at the 5% level. 95% confidence intervals are in brackets below point estimates. All models control for survey round and household has loan. Covariates include husband's current employment, wife's current employment, log of husband's income and log income squared, log of wife's income and log income squared, value of household assets, amount of land owned by household, and household size. Sample sizes are less than 11,882 because the estimation of fixed effects models involves only households with any variation in IPV over time. Data source is Bangladesh Integrated Household Survey 2011–2012 (round 1), 2015 (round 2), and 2018–2019 (round 3).

The index of husband's empowerment is significantly associated with the prevalence of verbal and any IPV but not with the prevalence of physical IPV. Specifically, a one-unit increase in husband's empowerment is associated with a 37% decrease in the adjusted odds of reporting verbal or any IPV. Given the standard deviation of the index is 0.45, this implies a one standard deviation increase in husband's empowerment is associated with a 16% decrease in the odds. However, the index of wife's empowerment is not significantly associated with the prevalence of verbal, physical, or any IPV. This is congruent with the previous findings that some measures of wife's empowerment are positively associated with IPV while other measures are negatively associated.

5. Discussion

In summary, this paper investigates the relationship between the empowerment of husbands and wives and intimate partner violence in rural Bangladesh. The paper finds that IPV is widespread, as a majority of households report at least one type of violence in at least one round. Husbands tend to be more empowered than their wives in most, but not all, dimensions. Yet, even in dimensions of empowerment dominated by men on average, the percentage of households in which individual wives are more empowered than their husbands is notable. The regression analysis reveals that some measures of men's empowerment are negatively associated with verbal and any IPV. Moreover, some measures of women's empowerment are associated with verbal, physical, and any IPV, but the sign of the association varies by measure.

It is important to discuss the limitations of the paper. Even though household fixed effects and time-varying covariates are included in regression models, the estimates are not necessarily causal. The paper relies on the limited measures of violence against women available in the BIHS. While rounds 1 and 2 ask about domestic violence, only round 3 asks specifically about intimate partner violence. Nevertheless, information from round 3 demonstrates that almost all domestic violence is committed by husbands. 97% of women who reported any verbal violence and 100% who reported any physical violence said their husbands carried out the abuse. Additionally, underreporting and nonresponse to survey questions about violence and other sensitive themes may impact the results, despite the procedures the BIHS took to obtain accurate information. Lastly, it is crucial to recognize that the findings pertain to rural Bangladesh and are not generalizable to other countries and contexts.

The paper's results can be situated in the context of the literature on intimate partner violence in developing countries. The regression analysis finds that measures of women's empowerment are significantly associated with IPV, but that the signs of the association are mixed. That is, community influence and satisfaction with leisure are negatively associated, while group membership and speaking in public are positively associated. Even one of the covariates, wife's income, is positive and significantly associated. This pattern of results is consistent with research that discovers in certain contexts, women's empowerment can reduce the likelihood of violence, whereas in other contexts, women's empowerment can raise the likelihood (Vyas et al., 2015, Raj et al., 2018; Murshid & Critelli, 2020; Bhalotra et al., 2021; Ranganathan et al., 2021; Bourey et al., 2023; Forty, 2022; Treves-Kagan et al., 2022). The evidence suggests that women's empowerment may induce “backlash” effects in some Bangladeshi households.

The regression analysis also finds that measures of men's empowerment (including ownership of assets, input in credit decisions, community influence, and satisfaction with leisure) are significantly associated with IPV, and the signs of the association are negative. If causal, this implies that, all else equal, an increase in men's empowerment reduces the likelihood of IPV. The results lend support to the paper's conjecture that men's empowerment is inversely associated with IPV. They are consistent with previous studies that explore the role of related and indirect measures of men's empowerment like unemployment, relative status, and life satisfaction (Bhalotra et al., 2021; Goodman et al., 2017; Krishnan et al., 2010; Luke et al., 2007; Yount et al., 2016).

It is fruitful to think more about the mechanisms underlying the findings on men's empowerment as well as their implications. Two of the correlates that emerge – community influence and satisfaction with leisure – do not seem to involve wives directly. Community influence is about a husband's place in his community. It is more likely to entail comparisons with other men, not comparisons with his wife. Satisfaction with leisure is reflective of a husband's attitudes about the quantity and quality of his own employment. The other correlates that emerge – ownership of assets and input in credit decisions – do involve wives and have another common trait. Husbands who are disempowered along these dimensions are in an extreme minority of men. Most husbands have joint or sole ownership of household items; most are joint or sole decision-makers about household loans. Then, being far from male norms in these areas is associated with IPV. These insights might imply that some of the forces driving IPV in rural Bangladesh involve a man's self-esteem or status relative to other men, not necessarily an intrinsic desire to keep women in an inferior position.

Nevertheless, open questions remain regarding the specific mechanisms of effects. For example, it may be that empowering men changes men's gender ideologies to be more egalitarian, improves their relative status alleviating their need to establish dominance, or boosts their self-actualization reducing their general frustration with life. Another question that emerges is why men's empowerment is associated with verbal intimate partner violence but not physical IPV. The paper is unable to say whether this is related to differences between verbal and physical violence, the fact that physical IPV is less prevalent, or patterns of underreporting. I.e., underreporting may be more severe in households with less empowered wives and husbands.

The paper's results have policy implications for Bangladesh. One unambiguously encouraging take-away is that public policies or other programs which increase the empowerment of men may help to reduce violence against women, even if this was not the intended outcome of the policy or program. However, the results also raise challenging questions. How can policymakers ensure that initiatives to increase men's empowerment are not doing so at the expense of women's rights and resources? This is a real risk given that measures of men's empowerment are correlated with measures of women's empowerment, sometimes positively and sometimes negatively, depending on the dimension (Akter & Francis-Tan, 2021). A related issue concerns the empowerment gap between spouses. Scholars and policymakers seek the elimination of spousal empowerment gaps (Kumar et al., 2021; Seymour, 2017; Sraboni et al., 2014). In principle, this may be accomplished in multiple ways. The findings in the paper suggest how this might best occur in the context of study – by raising women's status at a faster rate than raising men's status.

Furthermore, the results have implications for theories of empowerment. The evidence in the paper demonstrates that some aspects of men's empowerment are statistical correlates of intimate partner violence in rural Bangladesh. This underscores the importance of theory to guide the definition and measurement of men's empowerment. As previously discussed, it is reasonable to begin conceptualization with accepted theories of women's empowerment. Such was this paper's starting point for its selection of empowerment indicators. Yet, at least two considerations suggest that men's empowerment should be conceptualized differently than women's empowerment.

The most salient issue is that although the actual power held by individual men in their communities and households may vary considerably, the larger societal context often favors men over women which is not only reflected by imbalances of power in institutions and social norms but is also sustained by them. The ethical, social, and economic consequences of uplifting a disempowered man in a society that privileges men may be very different from uplifting a disempowered woman in the same society. Another issue is that the concept of power as it relates to men's empowerment may be distinct from the concept of power as it relates to women's empowerment. It may be necessary to distinguish among types of power. In the context of men's empowerment, care needs to be taken to distinguish “positive” types of power that inspire and “negative” types of power that embolden toxic masculinity. Theory would also have to clarify how the process of empowerment relates to these types of power. E.g., if “true” empowerment involves positive (not negative) power, then an increase in men's empowerment cannot increase intimate partner violence.

To conclude, this paper explores the association between empowerment and intimate partner violence in rural Bangladesh. The findings suggest that, besides women's empowerment, men's empowerment may also be a determinant of IPV. At least in this specific context of study, policies and programs that enhance the empowerment of husbands may reduce violence against their wives. In light of outstanding conceptual questions, the findings also underscore the need to extend theories of empowerment. The hope is the paper, though imperfect, will spur further research on this critical issue. Societies benefit immensely when families are guided by principles of love and respect.

Ethical statement

• •Ethical standards were followed in the creation and preparation of the manuscript
• •IRB approval was given by NUS
• •The manuscript is only submitted to this journal
• •No AI was used by the author
• •The manuscript is original

CRediT authorship contribution statement

Andrew Francis-Tan: Writing – review & editing, Writing – original draft, Methodology, Investigation, Formal analysis, Conceptualization.

Appendix A. Supplementary data

The following is the Supplementary data to this article.

Data availability

Data will be made available on request.