Women’s Employment and Intimate Partner Violence: Understanding the Role of Individual and Community Structural Drivers in Low- and Middle-Income Countries

Abstract

Empirical findings on the relationship between women’s employment and intimate partner violence (IPV) in low- and middle-income countries (LMICs) are mixed. These varied findings may arise because research thus far has given insufficient attention to how individual attributes and community context shape the pathways between women’s employment and IPV. Using publicly available Demographic and Health Survey (DHS) data from 20 LMIC settings (n = 168,995), we investigate (1) how women’s employment is associated with past-year IPV and (2) if associations differ by household- or community-level structural drivers of IPV: women’s attitudes toward IPV, women’s participation in household decision-making, and relative wealth. We fit mixed-effects logistic regression models exploring the total, individual, community, and contextual effects of women’s employment on past-year IPV; effect measure modification by structural drivers; and cross-level interactions between community-level structural drivers and individual employment. Our analyses reveal positive associations between total (odds ratio [OR] = 1.31; 95% CI [1.27, 1.35]), individual (OR = 1.23; 95% CI [1.19, 1.27]), community (OR = 1.06; 95% CI [1.06, 1.07]), and contextual effects (OR = 1.04; 95% CI [1.03, 1.05]) of women’s employment for IPV. Only individual wealth demonstrated statistically significant effect measure modification for the relationship between individual employment and past-year IPV (ratio of OR = 0.95; 95% CI [0.92, 0.99]). These findings suggest interventions that focus only on increasing women’s employment may be associated with harmful increases in the occurrence of IPV, even when these interventions enable a large proportion of women in a community to be employed. Structural interventions that change norms of women’s autonomy or attitudes toward IPV at the household or community levels may be insufficient to ameliorate these negative effects, whereas interventions that increase household wealth partly may buffer these effects.

Intimate partner violence (IPV), defined as “behavior within an intimate relationship that causes physical, sexual or psychological harm, including acts of physical aggression, sexual coercion, psychological abuse and controlling behaviors” (World Health Organization, 2017), is a critical health and human rights concern. Although affecting persons of all genders, women experience a disproportionate burden of harm from male use of violence (Caldwell et al., 2012). More than one in four ever-partnered women of reproductive age (15–49 years old) experience physical and/or sexual IPV globally (Sardinha et al., 2022), and this burden is higher in many low- and middle-income countries (LMICs, i.e., countries, such as Afghanistan and Peru, with gross national per capita incomes of 13,205 USD or less) (World Bank, 2022). Reflective of its importance for physical, mental, reproductive, and intergenerational health (Campbell, 2002), there is a burgeoning evidence base for prevention and intervention strategies that aim to transform the gender inequalities that underlie IPV (Jewkes et al., 2015; Stewart et al., 2021).

Among these prevention and intervention strategies, researchers and program implementers increasingly are examining access to economic resources as a potential leverage point for empowering women and reducing IPV (Peterman et al., 2017). Whereas research on cash transfers is growing rapidly (Baranov et al., 2021), women’s participation in employment is receiving relatively less empirical attention, despite its frequent use in IPV interventions (e.g., microfinance interventions). Extant epidemiological and intervention research points to inconsistent associations between women’s employment and IPV (Abramsky et al., 2011; Gichuru et al., 2019; Heise & Kotsadam, 2015; O’Malley & Burke, 2017; Orton et al., 2016). These inconsistent findings may reflect methodological challenges, such as self-selection into employment (Vyas & Heise, 2014) and evaluations that do not consider contextual factors. Research has thus far given insufficient attention to how individual attributes and community context shape the meaning and consequences of women’s employment for IPV.

Multiple theories have been used to explain potential relationships between women’s employment and IPV; these describe both negative and positive relationships (Vyas & Watts, 2009). Among theories hypothesizing a negative relationship between women’s employment and IPV, household bargaining models posit that the resources and power accompanying women’s employment will decrease IPV risk by decreasing the cost of relationship exit and increasing the cost of violence, allowing the woman to “bargain” for better outcomes (Strenio, 2020). Social capital theorists hypothesize that women’s employment can increase social capital by exposing women to non-kin-based networks (Sanyal, 2009), which increases women’s value at home and enhances their agency through exposure to ideas and resources outside of the home (Yount et al., 2021). By contrast, resource theory describes that women’s employment will increase IPV risk when women’s employment precipitates an imbalance in key resources that favors women (Allen & Straus, 1975; Goode, 1971), usurping traditional social scripts of resource allocation by gender (i.e., that men maintain power through resource ownership). In the absence of material sources of power, men may use violence to reestablish control and reaffirm masculine norms (Basile et al., 2013). Similarly, women’s employment may challenge normative beliefs that women should not work outside the home. Men may respond to this challenge by enacting social scripts that dictate their use of violence (Huesmann, 1988, 2018). Current evidence provides support for all these different associations in different populations, settings, and times (see, e.g., Lenze & Klasen, 2017; Tandrayen-Ragoobur, 2020).

Broader theories of IPV suggest positive and negative associations may be present in empirical research because complex interactions across the social ecology determine the symbolic meaning and practical consequences of women’s employment (Vyas et al., 2015). According to the ecological framework (Bronfenbrenner, 1977), a woman’s characteristics interact with the characteristics of her male partner, relationship, community, and macrosocial context to increase risk for IPV (Heise, 1998, 2011). This suggests that the effect of women’s employment, including whether it confers bargaining power and social capital or threatens male control, varies by the attitudes held by a woman’s male partner, relationship dynamics, and community norms and values that establish social scripts and expected behaviors for women and men, as well as broader contextual factors such as laws and policies that govern women’s access to employment and social capital.

Our analysis applies the ecological framework (Heise, 1998, 2011) to investigate the potential for individual and community processes and norms to explain associations between women’s employment and experience of past-year IPV in 20 LMIC settings. In addition to exploring how community-level women’s employment norms affect past-year IPV, we draw on three structural drivers of IPV: normalization and acceptability of violence in relationships, gender inequality favoring men, and poverty (Gibbs et al., 2020). Gibbs et al. (2020) describe that these factors individually and synergistically drive IPV, while influencing individual risk factors. When mapped onto the ecological framework, these drivers can manifest within individual attitudes, community norms, and/or macrosystem factors. Using publicly available data, we address the following research questions:

1. How are individual, community, and contextual variation (i.e., the effect of community variation holding individual variation constant) in women’s employment associated with past-year IPV?

2. Do associations between women’s employment and past-year IPV differ by household- or community-level women’s attitudes toward IPV (normalization and acceptability of violence in social relationships), women’s participation in household decision-making (gender inequality favoring men), and relative wealth (poverty)?

Answering these questions can help to inform the content and form of economic interventions targeting IPV in diverse settings.

Method

Data

These analyses used publicly available data from the Demographic and Health Surveys (DHS). The DHS are nationally representative surveys on population demographics, health, nutrition, and HIV that have been implemented in over 90 LMICs. For this study, we included the most recent survey from all countries that had harmonized data available in IPUMS-Demographic and Health Surveys (IPUMS-DHS) and for which all independent and dependent variables were available. Twenty countries were used, the majority of which are in Africa or Southeast Asia. Although there is significant variation and mixed results by country, past-year IPV in these regions tends to be prevalent, ranging 16.6% to 44.5% in Africa and 13.7% to 24.9% in Southeast Asia. Each country survey follows a two-stage cluster sampling procedure. In the first stage of the women’s survey, enumeration areas, which represent the primary sampling units (PSUs), are drawn from census files. In the second stage, women aged 15 to 49 years from 20 to 30 households per PSU are sampled (ICF International, 2012). To protect respondent safety, domestic violence questions typically are administered privately to one randomly selected ever-partnered woman aged 15 to 49 years in every second sampled household (ICF International, 2012).

Survey data were collected between 2010 and 2019 in the 20 included countries (Table 1). Although domestic violence questions are asked of ever-partnered women, we restricted the analytic sample to currently married or partnered women who were administered domestic violence questions to ensure comparability across countries. Of 380,908 women interviewed across the 20 countries, 287,757 women (76%) were currently married or partnered of whom 218,045 (76%) were selected and interviewed with the domestic violence module. Complete data for all IPV questions were available for 182,862 (84%) women. Variables for which women had missing data included current employment (missing = 153 women), attitudes toward IPV (missing = 5,995), decision-making (missing = 4,326), and wealth (missing = 3,956). We performed complete case analysis such that our final analytic sample included 168,995 (76% of the sample interviewed with the domestic violence module) women with data available for all variables of interest.

Table 1.

Countries Represented in the Pooled Sample.

Region	Countries	n (%)
Central Africa		11,036 (6.5)
	Angola Democratic Republic of Congo
Eastern Africa		41,662 (24.7)
	Burundi Ethiopia Kenya Malawi Mozambique Rwanda Tanzania Uganda Zimbabwe
Southern Africa		1,096 (0.6)
	Namibia
Western Africa		25,138 (14.9)
	Mali Nigeria Senegal
Eastern Mediterranean		23,099 (13.7)
	Afghanistan Egypt
Southeast Asia		66,964 (39.6)
	India Myanmar Nepal
		168,995 (100)

Measures

Past-Year IPV Outcome.

A modified version of the Revised Conflict Tactics Scale was used to capture IPV experience. This instrument has been used extensively in LMICs, has good internal consistency across subscales (psychological/emotional: Cronbach’s α = .79, physical: α = .86, sexual: α = .87), and has been identified as having good construct validity in high-income countries (Straus et al., 1996). Women answered questions about three types of violence: psychological/emotional (ever having been humiliated, insulted, or threatened by her husband/partner), physical (ever having been slapped, pushed, or shaken; having something thrown at her, her hair pulled, or her arm twisted; or being kicked, dragged, beat up, choked, burned, or threatened/attacked with a weapon by her husband/partner), and sexual (ever having been physically forced to have sex or perform another sexual act when it was not wanted by her husband/partner). For each violent behavior they reported, women were asked about the timing (whether in the past 12 months) and frequency (not at all, sometimes, often) of their experience. This was used to create a composite past-year IPV variable, which captured any of these experiences of violence (sometimes or often) within the 12 months preceding the survey. This timeframe was selected to reduce temporal bias.

Individual, Community, and Contextual Variables.

Demographic characteristics included women’s age in 5-year age groups, parity (i.e., the number of times a woman has given birth), sex of the head of household, and residential urbanicity (i.e., urban or rural residence). The primary individual-level covariate of interest, women’s employment, captured if the respondent is currently working. The precise wording of this question varied across surveys, but most surveys included an introductory statement to distinguish between household labor and labor outside of the home, followed by a single question about whether the respondent is currently employed.

Consistent with prior literature (e.g., Speizer, 2010), we created a dichotomous variable indicating any attitudes justifying IPV from a set of questions asking whether the respondent thinks a husband is justified in hitting or beating his wife for: arguing with him, burning the food, going out without telling him, refusing to have sex, and/or neglecting the children. Women reporting yes to any of these attitudes were coded as “1,” whereas women who indicated that none of these situations were justified were coded as “0.” We similarly created a dichotomous variable indicating participation in all major household decisions from a set of questions asking whether the respondent has final say on four tasks: making large household purchases, on visits to family or relatives, on spending husband’s earnings, and on woman’s health care. Women reporting independent or joint decision-making with their partner or another family member for all tasks were coded as “1,” whereas women reporting no participation in one or more of the tasks were coded as “0.” The measure of relative wealth was already created in the database by DHS through principal components analysis of selected assets (e.g., bicycles), construction materials for housing, water access, and sanitation facilities for each household.

We created community-level variables by combining the results from individual respondents within a PSU (i.e., community). PSUs are often census enumeration areas but may be electoral zones or villages (ICF International, 2012). For attitudes toward IPV, we calculated the percentage of respondents in the community endorsing any attitudes justifying IPV. For participation in decision-making, we calculated the percentage of respondents in the community reporting participation in all decisions. For average wealth, we calculated the mean of individual responses in the community. We scaled the community-level variables by multiplying each prevalence estimate by 10; therefore, each unit increase represents a 10% increase in community prevalence of attitudes justifying IPV, participation in decision-making, and wealth.

Statistical Analysis

We began by estimating background characteristics and key covariates in the pooled sample and conducting univariable analyses between these covariates and women’s employment. To explore our primary research questions, we fit a series of a priori-defined mixed-effect logistic regression models with random intercepts. Mixed-effects logistic regression models were used to account for the hierarchical (i.e., nested) nature of the data. Due to multistage cluster sampling, women from the same PSU were expected to be more like one another than women from different PSUs, violating the assumption of independence that is foundational to ordinary logistic regression (Debelew & Habte, 2021). Mixed-effects logistic regression also allows for partitioning the total variance into within-group and between-group effects to understand the relative contribution of individual and community variables to the outcome.

For our first question, models were specified to investigate the total effect, within- and between-community effects, and contextual effect of women’s employment. For our second question, models were specified to investigate effect measure modification by attitudes toward IPV, women’s participation in decision-making, and relative wealth. We subsequently fit a series of mixed-effects logistic regression models with random intercepts and random slopes to test cross-level interactions (Heisig & Schaeffer, 2019) between women’s employment and prevalence of any attitudes justifying IPV, prevalence of women’s participation in all decisions, and average wealth.

Models 1 to 3 explore the total effect, within- and between-community effects, and contextual effect of community employment. For model 1, the included variable combines individual (within-group) and community (between-group) effects of employment. For model 2, we partitioned the individual and community effects by including a cluster-mean centered variable and scaled community variable. In this model, the within-group effect specifically represents the log odds of past-year IPV for women who have higher versus lower employment relative to other women in their community. The between-group effect represents the log odds of past-year IPV for a given community, scaled such that the log odds ratio (OR) reflects a 10% higher prevalence of women’s employment in the community. For model 3, we tested contextual effects, or how the community prevalence of women’s employment shapes the effect of individual employment. We included the original variable and scaled community variable. In this model, the contextual effect effectively compares two women with the same value for employment who come from communities with a 10% difference in the prevalence of women’s employment.

Models 4 to 9 investigate effect measure modification. For models 4 to 6, we included multiplicative interaction terms for individual-level and community-level attitudes justifying IPV, participation in decision-making, and wealth. Models 7 to 9 give cross-level interactions between individual-level employment and community-level attitudes toward IPV, participation in decision-making, and wealth.

We did not use sampling weights to generate nationally representative results as we did not estimate characteristics or fit regression models for individual countries. The results are representative of the pooled international sample. We used the conventional α < .05 for tests of statistical significance. All analyses were completed in Stata SE 14.2 using the meqrlogit command (StataCorp, 2015).

Results

Table 2 describes the key descriptive variables and their association with IPV. By design, age ranged from 15 to 49 years. Over two-thirds (68.7%) of women resided in rural areas; only 13% resided in woman-headed households. Slightly less than half (44.2%) of respondents were employed, agreed with any reasons justifying IPV (44.1%), and participated in all decisions (45.6%). Overall, 29% of women reported experiencing past-year physical, sexual, and/or emotional IPV.

Table 2.

Descriptive Statistics and Univariable Analyses for the Pooled Sample.

	Prevalence Estimates	Univariable Analyses
	n= 168,995 % (n) or mean (SD)	No IPV n = 120,001 % (n)	IPV n = 48,994 % (n)	OR [95% CI]a
Respondent Age Group
15–19 years	4.83 (8,167)	5.00 (6,000)	4.42 (2,167)	—
20–24 years	16.56 (27,989)	16.21 (19,458)	17.41 (8,531)	1.27 [1.19, 1.35]
25–29 years	22.26 (37,616)	21.84 (26,206)	23.29 (11,410)	1.31 [1.23, 1.39]
30–34 years	19.81 (33,479)	19.61 (23,536)	20.29 (9,943)	1.32 [1.23, 1.40]
35–39 years	15.99 (27,017)	16.02 (19,227)	15.90 (7,790)	1.24 [1.17, 1.33]
40–44 years	11.37 (19,222)	11.71 (14,051)	10.55 (5,171)	1.12 [1.05, 1.20]
45–49 years	9.17 (15,505)	9.60 (11,523)	8.13 (3,982)	1.05 [0.98, 1.13]
Urban Residence	31.29 (52,878)	32.80 (39,363)	27.59 (13,515)	0.76 [0.73, 0.80]
Parity
0	7.50 (12,676)	8.40 (10,079)	5.30 (2,597)	—
1	15.45 (26,111)	16.38 (19,658)	13.17 (6,453)	1.35 [1.28, 1.43]
2	22.75 (38,451)	23.47 (28,163)	21.00 (10,288)	1.57 [1.48, 1.66])
3	17.57 (29,698)	17.42 (20,907)	17.94 (8,791)	1.76 [1.66, 1.86]
4	12.23 (20,672)	11.65 (13,977)	13.66 (6,695)	1.95 [1.84, 2.07]
5	8.41 (14,206)	7.95 (9,543)	9.52 (4,663)	1.92 [1.80, 2.05]
6+	16.08 (27,181)	14.73 (17,674)	19.40 (9,507)	2.00 [1.89, 2.12]
Woman Head of Household	12.97 (21,927)	13.41 (16,098)	11.90 (5,829)	0.89 [0.86, 9.26]
Agrees with any Reason Justifying IPV	44.12 (32,378)	40.23 (48,272)	59.53 (29,165)	2.00 [1.94, 2.05]
Participates in Final Say on all Decisions	45.60 (77,063)	49.14 (58,967)	36.94 (18,096)	0.57 [0.56, 0.59]
Wealth (mean, SD)	−0.01 (1.81)	0.06 (1.82)	−0.17 (1.78)	0.88 [0.87, 0.89]
Currently Employed	43.42 (73,385)	42.04 (50,452)	46.81 (22,933)	1.33 [1.29, 1.36]

Note. IPV = intimate partner violence; OR = odds ratio.

^aORs are derived from mixed-effect univariable logistic regression models.

Overall, women’s employment was associated with 1.31 times the odds of past-year IPV (95% CI [1.27, 1.35]), controlling for age, parity, urbanicity, woman head of household, and country (Table 3, model 1). As this model combines effects related to variation between individuals and communities, we partitioned the individual (within-group) and community (between-group) effects in model 2. Model 2 shows that, within a given community, the odds of past-year IPV among employed women was 123% the odds of unemployed women (95% CI [1.19, 1.27]). Model 3 gives the contextual effect. This essentially compares two women with the same value of employment living in communities with a 10% difference in women’s employment prevalence. Woman living in the community with 10% higher average women’s employment had 4% higher odds of past-year IPV (95% CI [1.03, 1.05]) than the woman in the comparison community, controlling for age, parity, urbanicity, woman head of household, and country.

Table 3.

Mixed-Effects Logistic Regression Models of Past-Year IPV, Household and Community Levels (n = 168,995).

		Adjusteda OR [95% CI]
Model 1b	Women’s employment (total effect)	1.31 [1.27, 1.35]
Model 2b	Individual women’s employment (within-community)	1.23 [1.19, 1.27]
	10% increase in community prevalence of women’s employment (between-community)	1.06 [1.06, 1.07]
Model 3b	Contextual effect	1.04 [1.03, 1.05]
Individual Attitudes Toward IPV
Model 4b	Women’s employment at community mean with no justification of IPV	1.23 [1.19, 1.27]
	Women’s employment at community mean with any justification of IPV	1.21 [1.10, 1.32]
	Interaction term (ratio of OR)	0.99 [0.90, 1.07]
Individual Decision-Making
Model 5b	Women’s employment at community mean with participation in none or some decisions	1.26 [1.21, 1.30]
	Women’s employment at community mean with participation in all decisions	1.25 [1.14, 1.38]
	Interaction term (ratio of OR)	1.00 [0.92, 1.09]
Individual Wealth
Model 6b	Women’s employment at community mean with mean wealth	1.23 [1.19, 1.27]
	Women’s employment at community mean with one unit increase in wealth	1.17 [1.11, 1.24]
	Interaction term (ratio of OR)	0.95 [0.92, 0.99]
Community Attitudes Toward IPV
Model 4b	Community women’s employment with no community attitudes justifying IPV	1.07 [1.06, 1.08]
	Community women’s employment with 10% increase in any community attitudes justifying IPV	1.07 [1.06, 1.08]
	Interaction term (ratio of OR)	1.00 [1.00, 1.00]
Community Women’s Decision-Making
Model 5b	Community women’s employment with none or some community-level participation in decisions	1.06 [1.05, 1.07]
	Community women’s employment with 10% increase in community-level participation in all decisions	1.06 [1.05, 1.07]
	Interaction term (ratio of OR)	1.00 [1.00, 1.00]
Community Wealth
Model 6b	Women’s employment at community mean with mean community wealth	1.07 [1.06, 1.08]
	Women’s employment at community mean with 10% increase in community wealth	1.07 [1.06, 1.08]
	Interaction term (ratio of OR)	1.00 [1.00, 1.00]
Cross-Level Interaction: Attitudes Toward IPV
Model 7c	Women’s employment at community mean with no community attitudes justifying IPV	1.17 [1.10, 1.25]
	Women’s employment at community mean with 10% increase in any community attitudes justifying IPV	1.18 [1.12, 1.25]
	Interaction term	1.01 [1.00, 1.02]
	Random slope	0.64 [0.56, 0.73]
Cross-Level Interaction: Women’s Decision-Making
Model 8c	Women’s employment at community mean with none or some community-level participation in decisions	1.17 [1.10, 1.25]
	Women’s employment at community mean with 10% increase in any community-level	1.19 [1.12, 1.26]
	participation in decisions
	Interaction term	1.02 [1.00, 1.03]
	Random slope	0.66 [0.58, 0.74]
Cross-Level Interaction: Wealth
Model 9c	Women’s employment at community mean with mean community wealth	1.23 [1.19, 1.27]
	Women’s employment at community mean with 10% increase in community wealth	1.22 [1.18, 1.27]
	Interaction term	1.00 [0.99, 1.00]
	Random slope	0.66 [0.58, 0.74]

Note. IPV = intimate partner violence; OR = odds ratio.

^aModels adjusted for 5-year age group (woman), parity (0–6+ births), urban/rural residence, woman head of household, and country

^bMixed-effect logistic regression model of past-year IPV on women’s employment with a random intercept for the primary sampling unit.

^cMixed-effect logistic regression model of past-year IPV on women’s employment with a random intercept for the primary sampling unit and random slope.

Models 4 to 6 explored the potential for different relationships between employment and past-year IPV by attitudes toward IPV, women’s participation in decision-making, and wealth. Model 4 found that the effect of women’s employment for past-year IPV did not differ by individual attitudes toward IPV (ratio of OR = 0.99; 95% CI [0.90, 1.07]), and the effect of community women’s employment did not differ by community attitudes toward IPV (ratio of OR = 1.00; 95% CI [1.00, 1.00]). Likewise, there was no meaningful effect measure modification by individual (ratio of OR = 1.00; 95% CI [0.92, 1.09]) or community women’s participation in decision-making (ratio of OR = 1.00; 95% CI [1.00, 1.00]). Individual wealth, however, changed the relationship between women’s employment and IPV. For women with wealth equal to their community mean, employment was associated with 1.23 times the odds of past-year IPV (95% CI [1.19, 1.27]). The odds of IPV associated with women’s employment decreased by 5% for each one unit increase in individual wealth (95% CI [0.92, 0.99]) after adjustment for age, parity, urbanicity, woman head of household, and country. By contrast, community wealth did not modify the association between community women’s employment and past-year IPV.

The final models (7–9) tested cross-level interactions, or how community attitudes toward IPV, decision-making, and wealth affect the relationship between individual women’s employment and past-year IPV. Confidence intervals for all interaction terms bounded the null, suggesting that only individual wealth demonstrates statistically significant effect measure modification for the relationship between individual employment and past-year IPV.

We ran a post hoc sensitivity analysis to test a cross-level interaction for community prevalence of IPV (not shown in Table 3). Individual women’s employment remained a risk factor for IPV when the community prevalence of IPV was zero (OR = 1.14; 95% CI [1.06–1.23]). For each 10% increase in the community prevalence of IPV, the odds of past-year IPV associated with employment increased by 17% (OR = 1.17; 95% CI [1.11–1.24]).

Discussion

In our analysis of national data from 20 LMIC settings, we found a significant association between women’s employment and self-reported past-year IPV that was, at least partly, explained by individual, community, and contextual factors. Women’s employment was associated with notably higher odds of past-year IPV. Contrary to our expectation, however, the between-community and contextual effects indicated that women in communities with higher prevalence of women’s employment had higher odds of past-year IPV, even when individual women’s employment status did not vary. Moreover, contrary to expectation, these associations did not differ by attitudes toward IPV or women’s participation in decision-making at the individual or community levels. Only individual wealth demonstrated statistically significant effect modification for the relationship between individual women’s employment and past-year IPV.

The primary association between individual women’s employment and increased odds of past-year IPV is consistent with a large body of literature, including prospective longitudinal research (Krishnan et al., 2010) and research using propensity scores to account for differential selection into employment (Vyas & Heise, 2014), although a body of literature suggests no or negative associations between women’s employment and IPV (e.g., Lenze & Klasen, 2017). Of the theories previously described, our findings more closely align with resource theory; violence may be used as the ultimate resource when women’s employment threatens patriarchal gender norms and other mechanisms for power and control are unavailable (Goode, 1971). Resource theory, however, is not the only theory that posits a positive relationship between women’s employment and increases in IPV, and pathways linking employment and IPV may differ between contexts in this diverse sample. Moreover, we modeled the impact of women’s employment on IPV given the use of economic interventions to promote gender equity and reduce IPV, but reverse causality is possible. Men’s use of IPV accompanied by controlling behaviors may reduce the likelihood of women’s employment, whereas men’s use of IPV without controlling behaviors may increase the likelihood of women’s employment as a means of escaping violence.

Our findings suggest that more prevalent women’s employment can heighten the risk of IPV. Although few studies are available to explain this finding, it is possible that men who live in communities with greater women’s employment experience additional threat to patriarchal gender norms. In communities with higher prevalence of women’s employment, men with employed wives might experience their wives’ employment as especially threatening because it is not anomalous but symbolic of greater empowerment for women. Men might be more likely to use violence to assert power and control—using IPV as a weapon to protect their traditional masculine norms. Alternatively, if total employment is fixed, women’s employment may decrease male employment, threatening the position of men in the household and leading to the use of violence as a mechanism to assert power and control.

The relationship between women’s employment and past-year IPV did not differ by individual or community attitudes toward IPV or women’s participation in decision-making. We anticipated that, among women who refused to justify IPV and participated in all decisions, greater equality would buffer against the harmful association between women’s employment and IPV. It is possible that the effect of these structural factors on individual and community attitudes and behaviors is less protective than their impact on macrosystem factors. In a study of cross-level effects between country-level and individual-level risks, Heise and Kotsadam (2015) found that country context modified the effect of individual employment experiences. For example, women working for cash was more harmful in countries where few women work (Heise & Kotsadam, 2015).

Relative wealth modified the association between women’s employment and past-year IPV. Among potential explanations, family stress theory suggests that increasing household wealth can improve household socioeconomic position, reduce stress from overwhelming demands, and lessen threatens to male identity stemming from poverty, thereby decreasing IPV risk associated with these stressors (Larsen, 2016; Vyas & Jansen, 2018). Alternatively, differences in household wealth may correspond to differences in women’s employment. In a cross-sectional study of employment in Tanzania, Vyas and Heise (2014) found that stable employment for women reversed harmful associations between women’s employment and IPV in urban areas, and payment in cash reversed the same association in rural areas. Women living in households with greater wealth in our sample might be more likely to have stable employment, receive cash payments, or have other unmeasured employment characteristics that confer protection. Further research is needed to differentiate among these possibilities.

This study has several limitations. First, as these analyses utilized cross-sectional data, we could not control fully for endogeneity between women’s employment and past-year IPV. Our findings may overestimate associations between women’s employment and IPV where these are mutually determined. For example, men’s controlling behaviors may limit women’s economic participation. We similarly are unable to account for all known confounding factors, including men’s alcohol use and men’s employment status. We also created community-level variables by aggregating scores from individual-level variables. This generates approximations of community means that may not be reliable measures when small numbers of individuals are sampled from the community (Lüdtke et al., 2008). Additionally, we used complete case analysis, which may have introduced bias where data are not missing at random. Whereas this cannot be tested directly, we compared women with missing (not included in the sample) and complete data (represented in the sample) on their employment, as the primary exposure could be subject to nonrandom missingness. These analyses (Supplemental Appendix Table 1) suggested that there is no difference between the groups on all sociodemographic variables except whether the woman has final say on all decisions. Finally, cautious extrapolation to interventions is warranted. Although our data suggest a potentially harmful relationship between women’s employment and IPV that is mitigated by increasing relative wealth, interventions may have differing effects, especially over time (Schuler et al., 1996). The processes through which economic and other interventions initiate household, community, and broader norm change and effect IPV merit additional research.

Despite these limitations, our findings contribute to the literature by exploring how individual attributes and community context shape the meaning and consequences of women’s employment for IPV. Overall, we found that individual, community, and contextual effects of women’s employment were risk factors for IPV. Individual and community attitudes toward IPV and women’s participation in decision-making did not modify these associations, although women’s employment was less harmful with increasing household wealth. Our results highlight the need to address the public health burden of IPV for women in LMICs through multiple intervention points and through thoughtful consideration of potentially deleterious effects of individual intervention strategies. Specifically, these analyses suggest that interventions that singularly aim to increase women’s employment may have harmful effects, even when such interventions enable a large proportion of women in a community to be employed. Importantly, structural interventions that change norms of women’s autonomy or attitudes toward IPV at the household or community levels may be insufficient to ameliorate these negative effects, whereas interventions that increase household wealth partly may buffer these effects. Tackling norms around the gendered nature of power and resource allocation, that may trigger IPV for women who become employed, are an essential element of IPV interventions. It is unknown, however, whether household wealth would be protective if it accrues to men and/or women and whether there is an inflection point for household wealth after which women’s employment becomes protective (rather than a risk factor for IPV). Previous attempts to increase wealth through cash transfers have had mixed effects for IPV, with some adverse impacts despite overall beneficial findings (Baranov et al., 2021; Buller et al., 2018). Instead, employment interventions may need to be paired with structural interventions that target macrosystem factors, as country-level factors have been found to modify the effect of individual employment experiences (Heise & Kotsadam, 2015).

Biographies

Christine Bourey, MPH, MSN, is a psychiatric nurse practitioner, certified nurse midwife, and doctoral candidate at the Johns Hopkins Bloomberg School of Public Health, Department of Mental Health. Her research focuses on mental health and well-being in the context of intimate partner and collective violence.

Judy Bass, PhD, MPH, is a professor at the Johns Hopkins Bloomberg School of Public Health, Department of Mental Health and a faculty member of the Johns Hopkins Center for Humanitarian Health. Her areas of expertise include designing and evaluating methods for assessing mental health across contexts and cultures and investigating the effectiveness of innovative prevention and intervention strategies.

Rob Stephenson, MSc, PhD, is a professor and chair of the Department of Systems, Populations and Leadership at the University of Michigan School of Nursing. His work is centered primarily on sexual and reproductive health, with a specific focus on the development and testing of HIV prevention interventions and the intersection between violence and health.