Acceptance of Intimate Partner Violence in Rural Malawi: An Empirical Analysis on the Impacts of Lineage and AIDS Conversation Networks

Abstract

Acceptance of intimate partner violence (IPV) is both a barrier to its reduction and a strong predictor of prevalence. This study examines whether lineage and AIDS conversation networks impact acceptance of IPV among men and women in rural Malawi. The sample consists of 4,422 male and female respondents from the 2001, 2004, and 2006 waves of the Malawi Longitudinal Study of Families and Health (MLSFH). The longitudinal nature of the data enables us to use not only random effects models in the empirical analysis, but also fixed effects models, which help alleviate omitted variable bias. Our findings suggest that in the patrilineal Northern Region of Malawi, AIDS conversation networks may be conducive toward reducing acceptance of IPV. However, AIDS conversation networks may reinforce violent norms in the Central Region of Malawi as men with larger AIDS conversation networks are more accepting of IPV. These findings imply there may be an opportunity to incorporate network-based policies to reduce IPV, and lineage may be an important consideration in forming policies to alleviate IPV. In addition, we find a relatively strong age-acceptance gradient, with younger, never-married individuals being less accepting of IPV than older, married individuals. This finding may help inform intervention targeting as older, married women in Malawi may be at a higher risk of IPV.

Keywords

Intimate partner violence Malawi lineage AIDS conversation networks domestic violence

Introduction

Intimate partner violence (IPV) is an international problem that transcends geographic, economic, and cultural divides. Despite a plethora of international and national commitments, declarations, and plans to combat IPV, it is still common in many societies. The World Health Organization (2013) estimates that worldwide, 30% of ever-partnered women have experienced IPV. There are significant economic costs associated with IPV as it drains resources from individuals, the private sector, and the public sector (Brown, Finkelstein, & Mercy, 2008; Max, Rice, Finkelstein, Bardwell, & Leadbetter, 2004). In addition to physical harm, victims of IPV face increased risk of depression (Beydoun et al., 2012; Kim et al., 2013), are more likely to contemplate or attempt suicide (Devries et al., 2011), and have poorer reproductive health (Sarkar, 2008).

An important barrier to reducing IPV is its social acceptance in many parts of the world. A 2002 study in Malawi found that 30% of men and 39% of women agreed that women sometimes deserve to be beaten (Andersson, Ho-Foster, Mitchell, Scheepers, & Goldstein, 2007). Acceptance of IPV is also a strong predictor of prevalence as women who are accepting of IPV are more likely to be victims (Alio et al., 2011; Boyle et al., 2009; Faramazi et al., 2005), and men who are accepting of IPV are more likely to be aggressors (Abrahams, Jewkes, Hoffman, & Laubsher, 2004).

Given these findings, there has been increased interest in examining acceptance of IPV as a means to better understand and reduce incidence of IPV. A study of 17 countries in Sub-Saharan Africa found acceptance of IPV is decreasing with education and decreasing with economic status at both the individual and community level, and counterintuitively is higher for woman than men (Uthman et al., 2009). However, empirical studies have found that determinants sometimes differ across contexts. For example, education reduces acceptance of IPV in Kenya, but increases acceptance in Zambia (Lawoko, 2008). A recent study on IPV in Malawi concluded that more research is needed to “examine cultural attitudes including acceptability and justifying attitudes towards IPV in Malawi, ultimately providing the background needed for effective, targeted interventions” (Bazargan-Hejazi, Medeiros, Mohammadi, Lin, & Koustav, 2013, p. 47).

This study contributes to the literature on acceptance of IPV by exploring whether AIDS conversation networks, one proxy for social networks, impact acceptance of IPV in rural Malawi. Economic household bargaining models imply that social networks, via increased support and opportunities outside of marriage, will lower levels of IPV. Social networks may empower women by providing them with a variety of opportunities, emotional or financial support, accommodation, legal advice, employment, envisioning relationships without violence, and other assistance (Sen, 1999). Alternatively, social networks may reinforce violent norms (Fagan, 1989) and reduce the negative impact of violence, making it easier for victims to stay in violent relationships (Katerndahl, Burge, Ferrer, Becho & Wood, 2013).

This study also examines the importance of lineage in accepting IPV. The three regions of Malawi are characterized by lineage differences that impact married and divorced life. Given these differences, certain factors, including AIDS conversation networks, may be more important in one region than another. Therefore, we run separate regressions for each region (the North, the Center, and the South) to allow for more flexibility by not forcing variables to have the same impact in all regions and, thus, giving us the opportunity to discover more accurate results.

While previous studies have often examined acceptance of IPV using cross-sectional data from the Demographic and Health Surveys (DHS), we use data from the Malawi Longitudinal Study of Families and Health (MLSFH), a unique panel dataset that allows us to use random effects and fixed effects models. Fixed effects models allow us to control for unobserved heterogeneity among respondents that otherwise may be difficult to capture. For example, personality traits may be important in determining acceptance of IPV. These traits are difficult to observe and measure in cross-sectional data, but time-invariant traits can be controlled for in fixed effects models.

Regional Heterogeneity, Lineage, and IPV in Malawi

Malawi is comprised of three heterogeneous regions, the North, the Center, and the South. As this article is looking at attitudes toward leaving violent marriages, it is important to consider the marriage practices in each region as well as women’s and men’s perceived rights upon dissolution of marriage. The North is predominately Protestant and patrilineal with patrilocal residence after marriage and inheritance traced through sons. In patrilineal systems, land belongs to men, and women obtain any inheritance rights through their husbands. For women, land ownership is temporal and linked to the existence of the marriage (Chiweza, 2008). The Center is both Protestant and Catholic and characterized by both matrilineal and patrilineal households where residence may be matrilocal or patrilocal. The South is largely Muslim and follows a matrilineal system of inheritance where residence after marriage is primarily matrilocal. In matrilineal systems, land belongs to the women’s family and children inherit the property of their maternal uncles. Upon dissolution of a marriage, the wife and children are undisturbed in terms of residence and land use (Chiweza, 2008). In both matrilineal and patrilineal societies, women in Malawi are considered inferior to men (White, 2007).

Prevalence of IPV varies by region in Malawi; 17.9% of women living in the North report experiencing physical violence in the past year, compared with 13% in the Center and 14.5% in the South (NSO & ICF Macro, 2011). These statistics may be underestimating IPV as Malawian culture discourages women from openly talking about IPV (Chakwana, 2005). Acceptance of IPV also varies by region. The 2010 Malawi DHS survey consistently finds acceptance of IPV is highest in the North and lowest in the South.

Conceptual Framework

Economist have analyzed IPV in the context of household bargaining models (Farmer & Tiefenthaler, 1997; Tauchen, Witte, & Long, 1991). In these models, two partners, often a husband and wife, bargain over resources. A key determinant of bargaining power is reservation utility, the utility each partner would experience if the marriage (or union) were to dissolve. Assuming individuals want to maximize their utility, if one’s utility within marriage is less than their reservation utility, he or she will initiate divorce or separation. Divorce is common in Malawi. In the North, 40% of first marriages end in divorce. Comparable figures for the Center and the South are 45% and 65%, respectively (Reniers, 2003). A higher reservation utility leads to more bargaining power. Many factors have the potential to impact reservation utility, including income, wealth, opportunities outside of marriage, and social networks. The implications from these models may be that social networks provide women with more resources and opportunities outside of marriage, thus raising their reservation utility and bargaining power. Consequently, one may expect women with stronger social networks to be less accepting of IPV.

However, the potential impacts of social networks on acceptance of IPV may be more complex than the story above. Fagan (1989) finds that perpetrators of IPV are less likely to stop violence if they are engaged with social networks that support and condone violence. Katerndahl, Burge, Ferrer, Becho, and Wood (2013) argue that if victims of IPV seek out other victims, these social networks may actually reinforce violence if it is seen as “normal.” Even if social network members see violence as egregious, they may not know how to help. Wilcox (2000) argues that if single motherhood and single woman status are less culturally desirable, the social support for leaving a relationship may be less than positive. Therefore, the effect (if any) of social networks on acceptance of IPV is an empirical question and likely context specific. However, a significant finding in either direction may suggest the need for network-based policies and programs in addition to individual approaches to combat IPV.

Data and Method

Data

The data used in this article are from the MLSFH, a long-standing survey initially developed for the purpose of exploring the role of social networks in changing demographic attitudes and behaviors. This project is rare in that it is one of only a few studies that explicitly examines social networks; it is also one of only a few longitudinal studies in Malawi. The MLSFH survey is not nationally representative; it was designed to take place in three rural districts, one in each region of the country: Rumphi district in the North, Mchinji district in the Center, and Balaka district in the South.

The MLSFH survey was designed to interview 500 randomly selected ever-married women in each of the three districts and their husbands. The initial survey was conducted in 1998 and follow-up surveys on the same respondents and any new spouses were conducted in 2001, 2004, 2006, 2008, and 2010. In addition, in 2004, 998 adolescents were added to provide a better representation of the young and unmarried rural population. Survey questions do differ from across waves in part to reflect changes in health concerns. However, in every wave, respondents were asked questions about their background, marital history, family planning, and gender-related questions. Our variables of interest, which include questions on AIDS conversation networks, are available in the 2001, 2004, and 2006 waves. Thus, this article focuses on these 3 years. The analysis includes the adolescent sample to allow for more variation in age and marital status.

Our final dataset has information on 4,422 individuals. Despite a smaller sample size, sample characteristics from the MLSFH closely match characteristics from the rural population of the Malawi DHS data. The MLSFH cohorts were selected to represent the rural population of Malawi, in which residents live in conditions similar to those in rural areas of other countries with high HIV prevalence (Kohler et al., 2015). Thus, findings using the MLSFH may be applicable to other rural areas with similar characteristics, but may not be generalizable to urban areas. In our dataset, 61% of individuals were interviewed in at least two waves. Part of the reason for the unbalanced panel is the adolescent sample was not added until 2004. The main reasons for attrition were migration and mortality.

Both male and female respondents were asked, “In your opinion, is it proper for a wife to leave her husband if he beats her frequently?” Respondents replied either yes or no. The response to this question is the focus of this article and is used to measure acceptance of IPV. We create a binary dependent variable in which 1 indicates that the respondent believes it is proper for a wife to leave her husband if he beats her frequently and 0 indicates otherwise. In 2006, 86% of women and 81% of men responded that it was proper for a woman to leave a violent marriage.

The MLSFH contained a series of questions on AIDS conversation networks. The responses to these questions may provide us with some information on whom (if anyone) respondents are speaking to about personal, important health issues. These types of name generator questions are a common method to gather information in ego network research and often ask respondents a series of questions about whom they discuss important matters with (Burt, 1984) or whom they chat with or visit (Campbell & Lee, 1991).

Specifically, the MLSFH asked respondents, “how many people have you chatted with about AIDS?” Interviewers then collected information from the respondent on up to four of these network members, including information on their sex, relationship with respondent, closeness to respondent, and residence. Characteristics of these AIDS conversation network members are displayed in Table 1. In 2006, women provided additional information on more than 6,500 individuals with whom they had chatted about AIDS. Individuals most frequently discuss AIDS with members of the same sex. For example, in 2006, 95% of the AIDS conversation network members identified by women respondents were women. Relatives comprise between 23% and 45% of AIDS conversation network members. Nonrelatives, which comprise the majority of AIDS conversation networks, include friends, colleagues, and health workers, among others. With regard to the closeness of the relationship between the respondent and AIDS conversation network members, respondents most frequently identify members as friends, with 18% to 28% of respondents being identified as confidants. Women are more likely to have AIDS conversation network members in their village and the percentage of AIDS conversation network members residing in the respondents’ village ranges from 50% to 69%.

Table 1.

Characteristics of AIDS Conversation Network Members.

Sex	Women (%)			Men (%)
Year	2001	2004	2006	2001	2004	2006
Sex of network member
Male	7	8	5	95	93	95
Female	93	92	95	5	7	5
Network member’s relationship to respondent
Male relative	5	4	3	31	20	24
Female relative	40	29	35	3	3	2
Other (nonrelative)	55	67	62	66	77	74
Closeness of relationship between respondent and network member
Confidant	18	28	24	19	26	23
Friend	65	57	69	65	61	70
Acquaintance	14	14	6	14	12	6
Met once or twice	3	1	1	4	1	1
Residence of network member
Same village as respondent	60	62	69	50	51	54
Different village from respondent	40	38	31	50	49	46
No. of network members	3,705	4,287	6,539	2,430	3,041	5,404

We create two AIDS conversation network variables for inclusion in the regression analysis. The first, AIDS conversation network size, is the number of people the respondent reports chatting with about AIDS. This variable may give us an indication of how many people the respondent is talking to about present health and gender-related issues; it may also give us an indication of the size of the respondent’s social network. Second, we create a binary variable in which 1 indicates the respondent has identified at least one AIDS conversation network member as a confidant and 0 indicates otherwise. This variable can provide an indication if the respondent has a person who is close to them, whom they can confide in and perhaps count on for support. Having such a person may be important in determining if one is accepting of IPV.

In addition to the AIDS conversation network variables, we include a number of control variables, including age, number of living children, marital status, education, and wealth groups (poorest, quite poor, average wealth, quite wealthy, and wealthiest). Wealth is a subjective measure according to the interviewer who was asked to characterize the relative wealth of each household compared with others in the village. Wealth is used because the MLSFH does not include consistent questions about income; for example, there is no income information for 2004. In addition, in measuring living standards, wealth is often preferable to income as it is not as sensitive to temporary shocks (Filmer & Pritchett, 2001). The subjective measure of wealth is used because no information on asset ownership was included in the 2006 questionnaire. Finally, time fixed effects are also included in all regressions. Summary statistics by year and sex are shown in Table 2.

Table 2.

Summary Statistics by Year and Sex.

Variable	Mean
	Women			Men
	2001	2004	2006	2001	2004	2006
Not accepting of IPV	72%	78%	86%	65%	69%	81%
Age	34.8	32.5	33.7	40.9	33.1	36.5
No. of children	4.1	6.6	3.7	4.7	10.8	3.9
Marital status
Married	90.3%	75.5%	82.0%	97.2%	62.6%	79.1%
Divorced/separated	6.2%	5.4%	7.1%	1.9%	0.0%	1.4%
Widowed	3.4%	4.2%	4.2%	0.9%	0.0%	0.8%
Never married	0.0%	14.8%	6.5%	0.0%	37.4%	18.6%
Education
No education	33.6%	24.5%	29.5%	17.5%	10.6%	16.5%
Primary education	59.8%	66.2%	61.4%	68.6%	68.1%	63.3%
Secondary education	6.6%	9.2%	9.1%	13.9%	21.3%	20.3%
Wealth
Poorest	11.5%	16.1%	5.8%	7.1%	13.4%	5.5%
Quite poor	32.4%	36.2%	25.6%	24.6%	36.0%	19.5%
Average wealth	44.6%	37.3%	54.7%	53.3%	40.5%	55.4%
Quite wealthy	9.4%	8.7%	11.7%	12.8%	7.9%	17.1%
Wealthiest	2.1%	1.7%	2.1%	2.3%	2.2%	2.5%
AIDS conversation networks
AIDS conversation network size	5.7	4.2	11.6	7.3	6.1	13.7
At least one network member is a confidant	32.8%	56.2%	83.6%	35.9%	63.0%	81.7%
No. of observations	1,508	1,650	1,737	949	971	1,422

Note. By design, sample is fairly evenly distributed between regions (not shown). IPV = intimate partner violence.

In 2006, women and men reported an average of 11.6 and 13.7 AIDS conversation members, respectively. This is significantly higher than in previous years, suggesting increased discussion of HIV/AIDS. In 2006, approximately 84% of women and 82% of men reported that at least one of these conversation members is a confidant.

Econometric Methods

The baseline estimation for the analysis may be represented by the following formula:

Prob (Y) = f (\begin{matrix} AIDS conversation network size, confidant member, \\ age group, marital status, wealth, education, \\ No . of living children, time fixed effects . \end{matrix})

where Y is a binary variable equal to 1 if the respondent believes it is proper for a woman to leave her husband if he beats her frequently.

The determinants of acceptance of IPV are first estimated using a random effects logit model. A logit model is preferable to ordinary least squares (OLS) because of the binary nature of our dependent variable. The random effects model allows us to take advantage of the panel nature of our data, whereas a simple pooled logit would not account for the fact that there are repeated observations over time. The random effects logit model introduces a random effect term (α_i) in the error structure to account for unobserved time-invariant individual specific effects. The error term takes the form, e_i = α_i + µ_it, where α_i is assumed to capture differences across individuals caused by unobserved factors; α_i are independent and identically distributed, N(0, $σ_{α}^{2}$ ; and µ_it is an idiosyncratic error term with variance, $σ_{μ}^{2}$ , equal to 1. The model is represented by the following equation:

P r (y_{i t} = 1 | x_{i t} {, α}_{i}) = Φ (x_{i t} b {+ α}_{i}),

where y_it is acceptance of IPV; x is a vector of explanatory variables including sociodemographic characteristics, AIDS conversation network variables, and time fixed effects; b are the coefficients to be estimated; and Φ (x_itb + α_i) is the cumulative normal distribution function.

One potential problem in estimating the effects of AIDS conversation networks on acceptance of IPV is omitted variable bias. This problem may arise if there is an unobserved variable that is correlated with an independent variable and the dependent variable. For example, a woman who is self-confident may be more likely to chat with others about AIDS and less accepting of IPV. To deal with omitted variable bias, we run a fixed effects logit model. The fixed effects logit estimation allows us to control for unobserved characteristics of the respondents by essentially adding an individual specific dummy variable that is fixed over time. Thus, the advantage of a fixed effects logit model is that we are able to control for unobserved heterogeneity among respondents that impact their AIDS conversation networks and acceptance of IPV. As the fixed effects logit model controls for all fixed characteristics over the survey waves, these estimates focus on the relationship between changes in the respondents’ attitudes toward IPV and changes in their AIDS conversation networks. The fixed effects logit examines only within respondent variation, and consequently, respondents who do not change their attitude toward IPV over the survey waves will not contribute to the estimates. Thus, the sample size for the fixed effects logit is smaller than for the random effects logit, which can reduce power in statistical analyses. We discuss results for both models; the fixed effects helps to control for omitted variable bias and understand changes in attitudes, while the random effects allows us to examine the attitudes of all respondents and look at both within and between variation among the respondents.

There are significant regional differences in acceptance of IPV in Malawi. Furthermore, there are lineage differences between regions. Women in matrilineal societies are undisturbed in terms of residence and land use when their marriage dissolves. Therefore, women in matrilineal societies may be less accepting of IPV as they do not have to sacrifice as much by ending the marriage. Women in patrilineal societies lose all rights to residence and land upon divorce. Consequently, women in patrilineal societies may face a greater cost of divorce and, therefore, be more accepting of IPV. In addition, women in matrilineal societies are often living in close proximity to their family; whereas, women in patrilineal societies often move after marriage to their husband’s village. Living close to one’s family may provide women with a strong support system and protection against harm such as IPV. AIDS conversation networks may be more important to women in patrilineal societies where this close support system is not available. Although the dynamics of lineage and IPV are relatively unexplored, several studies have found differences in patterns of IPV between matrilineal and patrilineal groups (Asiedu, 2016; Sedziafa, Tenkorang, & Owusu, 2018). Therefore, separate regressions are run for each region to allow for different potential effects of the AIDS conversation network variables and other variables as well.

Results

Results for the Northern Region

Table 3 displays the results for the Northern Region. Columns 1 and 2 show odds ratios for the random effects logit estimation and the fixed effects logit estimation, respectively, for women. In both models, AIDS conversation network size is statistically significant and indicates women who have chatted with more people about AIDS are less accepting of IPV. The fixed effects model finds that speaking to one additional person about AIDS increases the odds of not accepting violence by 4.5%. The results from the fixed effects model also indicate that the odds of not accepting violence are 1.5 times greater for women with at least one confidant. These findings support the predictions of the household bargaining models that social networks may lower acceptance of IPV.

Table 3.

Determinants of IPV in the Northern Region.

Sex	(1)	(2)	(3)	(4)
Sex	Women		Men
Model	RE Logit	FE Logit	RE Logit	FE Logit
AIDS conversation network
Network size	1.020**	1.045***	1.006	1.004
	(0.010)	(0.017)	(0.008)	(0.013)
	[.050]	[.006]	[.453]	[.521]
Any confidant	1.255	1.556**	1.393**	1.106
	(0.188)	(0.342)	(0.214)	(0.251)
	[.128]	[.043]	[.031]	[.537]
Age (30-39 is base group)
13-19	3.064***	2.583	1.699	3.586
	(1.185)	(2.828)	(0.704)	(4.277)
	[.004]	[.414]	[.201]	[.287]
20-29	1.463**	0.659	1.393	0.713
	(0.273)	(0.307)	(0.349)	(0.391)
	[.041]	[.382]	[.186]	[.592]
40-49	0.975	0.600	0.905	0.873
	(0.192)	(0.243)	(0.917)	(0.463)
	[.900]	[.196]	[.647]	[.740]
50-59	1.037	0.564	0.998	0.603
	(0.285)	(0.406)	(0.257)	(0.510)
	[.894]	[.416]	[.995]	[.494]
60+	1.269		0.597*	0.922
	(0.623)		(0.177)	(1.119)
	[.628]		[.083]	[.867]
Marital status (married is base group)
Divorced/separated	2.586**	2.313	0.967
	(0.990)	(2.0397)	(0.769)
	[.013]	[.345]	[.967]
Widowed	1.026	0.493
	(0.340)	(0.419)
	[.937]	[.407]
Never married	1.787	1.722	1.071	0.796
	(0.817)	(1.910)	(0.365)	(0.816)
	[.204]	[.673]	[.841]	[.962]
Education (primary education is base group)
No education	0.673	0.419	0.534	0.963
	(0.267)	(0.297)	(0.229)	(1.267)
	[.318]	[.224]	[.145]	[.948]
Secondary education	0.739	0.614	1.223	0.803
	(0.148)	(0.366)	(0.207)	(0.367)
	[.132]	[.429]	[.234]	[.667]
Wealth (average wealth is base group)
Poorest	0.928	0.640	1.458	1.034
	(0.359)	(0.326)	(0.507)	(0.549)
	[.848]	[.384]	[.278]	[.941]
Quite poor	1.002	0.791	0.853	0.532**
	(0.164)	(0.191)	(0.148)	(0.144)
	[.991]	[.331]	[.359]	[.030]
Quite wealthy	0.988	0.847	0.989	1.261
	(0.225)	(0.319)	(0.226)	(0.512)
	[.959]	[.672]	[.963]	[.408]
Wealthiest	0.721	0.376	0.522	0.813
	(0.323)	(0.299)	(0.216)	(0.491)
	[.465]	[.212]	[.116]	[.844]
No. of children	0.993	1.022	1.024	0.986
	(0.021)	(0.037)	(0.016)	(0.025)
	[.739]	[.424]	[.120]	[.600]
Year fixed effects	Yes	Yes	Yes	Yes
No. of observations	1,587	561	1,206	489

Note. Standard errors in parenthesis, p values in square brackets. Columns display odds ratios. IPV = intimate partner violence; RE = random effects; FE = fixed effects.

p < .10. **p < .05. ***p < .01.

The results from the random effects model indicate that younger women in the North are less accepting of IPV than older women in the North. Teens are 3 times more likely to believe it is acceptable for a woman to leave a violent marriage than women in their 30s (the base group). Divorced is also statistically significant, indicating that the odds of not accepting IPV are 2.5 times greater for divorced women than married women.

Columns 3 and 4 show the odds ratios for the random effects logit and fixed effects logit for men in the North. In the random effects model, the variable for identifying at least one confidant is statistically significant at the 5% level and indicates that men who identify having at least one confidant in their AIDS conversation network are 1.5 times more likely to believe it is proper for a woman to leave a violent marriage. This finding is consistent with Stets (1991) who argues that social isolation increases IPV. Our results also seem to suggest that younger men are less accepting of violence than older men. The chances of a man who is 60 plus years in age believing it is appropriate for a woman to leave a violent marriage are approximately 60% of the odds of a man in his 30s believing it is acceptable for a woman to leave a violent marriage.

Results for the Central Region

The results for the Central Region are in Table 4. Columns 1 and 2 display odds ratios for the women’s random effects logit and the fixed effects logit, respectively. The AIDS conversation network variables are not statistically significant, indicating that AIDS conversation networks do not impact women’s acceptance of IPV in the Central Region of Malawi. This finding differs from the North Region and may potentially be due to the fact that the Central Region is characterized by both matrilineal and patrilineal societies. Women in matrilineal societies may have an existing support system as they often reside in their home villages. Therefore, AIDS conversation networks may not do as good a job of representing women’s social support systems. They also may not be as important as other support systems, such as family, that are local.

Table 4.

Determinants of IPV in the Central Region.

Sex	(1)	(2)	(3)	(4)
Sex	Women		Men
Model	RE Logit	FE Logit	RE Logit	FE Logit
AIDS conversation network
Network size	0.994	0.989	0.986*	0.995
	(0.008)	(0.012)	(0.007)	(0.014)
	[.469]	[.366]	[.054]	[.701]
Any confidant	1.187	1.152	1.225	1.053
	(0.182)	(0.258)	(0.218)	(0.377)
	[.265]	[.526]	[.254]	[.885]
Age (30-39 is base group)
13-19	0.838	1.407	0.902	1.717
	(0.261)	(0.963)	(0.392)	(2.644)
	[.571]	[.618]	[.813]	[.726]
20-29	0.872	0.852	1.257	2.571*
	(0.167)	(0.372)	(0.290)	(1.478)
	[.476]	[.713]	[.321]	[.100]
40-49	0.844	0.770	1.134	0.779
	(0.202)	(0.377)	(0.278)	(0.604)
	[.478]	[.594]	[.606]	[.748]
50-59	0.552*	0.717	1.043	0.323
	(0.185)	(0.657)	(0.310)	(0.424)
	[.076]	[.717]	[.887]	[.389]
60+	0.715	0.313	1.221	0.025*
	(0.286)	(0.557)	(0.503)	(0.049)
	[.401]	[.514]	[.627]	[.060]
Marital status (married is base group)
Divorced/separated	1.295	1.059	1.030	0.595
	(0.415)	(0.582)	(0.732)	(1.062)
	[.419]	[.917]	[.967]	[.771]
Widowed	1.026	0.322	1.343
	(0.340)	(0.282)	(1.298)
	[.658]	[.196]	[.1344]
Never married	2.161*	7.951*	3.109***	2.627
	(0.959)	(10.257)	(1.191)	(2.897)
	[.082]	[.101]	[.003]	[.381]
Education (primary education is base group)
No education	0.987	0.918	1.135	2.971
	(0.171)	(0.422)	(0.276)	(2.515)
	[.938]	[.852]	[.603]	[.198]
Secondary education	0.888		0.741	0.186*
	(0.329)		(0.187)	(0.186)
	[.749]		[.235]	[.098]
Wealth (average wealth is base group)
Poorest	0.985	0.664	0.623	0.592
	(0.227)	(0.226)	(0.189)	(0.376)
	[.948]	[.228]	[.120]	[.410]
Quite poor	1.066	1.028	0.746	0.630
	(0.177)	(0.269)	(0.151)	(0.278)
	[.701]	[.917]	[.148]	[.295]
Quite wealthy	1.061	1.659	1.192	1.063
	(0.265)	(0.635)	(0.280)	(0.483)
	[.813]	[1.85]	[.454]	[.893]
Wealthiest	1.076	0.909	1.736
	(0.518)	(0.713)	(1.023)
	[.880]	[.914]	[.349]
No. of children	.974	.924*	.996	.963
	(0.018)	(0.042)	(0.018)	(0.038)
	[.158]	[.083]	[.883]	[.350]
Year fixed effects	Yes	Yes	Yes	Yes
No. of observations	1,610	516	953	236

Note. Standard errors in parenthesis, p values in square brackets. Columns display odds ratios. IPV = intimate partner violence; RE = random effects; FE = fixed effects.

p < .10. **p < .05. ***p < .01.

The coefficient for never married is statistically significant at the 10% level and indicates that never-married women in the Central Region are less accepting of IPV than married women. Number of children is statistically significant at the 10% level in the fixed effects regression, implying that when a woman has an additional child, her odds of accepting IPV increase by approximately 8%. This may be because women (and likely their children) will face a greater burden upon divorce or because women may be more reluctant to end a marriage when there are more children involved.

The results from the random effect and fixed effects logit estimations for men in the Central Region are shown in columns 3 and 4, respectively. In the random effects estimation, AIDS network size is statistically significant at the 10% level and indicates that men with larger AIDS conversation networks are more accepting of IPV. This result is the opposite of what we found for the Northern Region. This result implies that men in the Central Region who chat with more individuals about AIDS are less likely to think it is proper for a woman to leave a violent marriage. This finding is more consistent with Fagan (1989) who argues that sometimes social networks reinforce violent norms. Similar to the results for women in the Central Region, the random effects estimation indicates that never-married men are less accepting of IPV than married men. In the fixed effects results, we once again see an age-IPV acceptance gradient. Men in their 20s are 2.6 times more likely to believe it is proper for a woman to leave a violent marriage compared with men in their 30s.

Results for the Southern Region

Table 5 displays the results for the Southern Region. Columns 1 and 2 show odds ratios for the random effects logit estimation and the fixed effects logit estimation, respectively, for women. Similar to the results found for the Central Region, AIDS conversation networks have no impact on women’s acceptance of IPV. The Southern Region of Malawi is primarily matrilineal; thus, AIDS conversation networks may not be as important as a source of social support because women are often living in their home village in close proximity to their families. The results of the random effects model indicate that women in the poorest households are 1.6 times more likely to believe it is proper for a woman to leave a violent marriage compared with women of average wealth. It may be the case that the opportunity cost of dissolving a marriage is less for women in poorer households.

Table 5.

Determinants of IPV in the Southern Region.

Sex	(1)	(2)	(3)	(4)
Sex	Women		Men
Model	RE Logit	FE Logit	RE Logit	FE Logit
AIDS conversation network
Network size	0.981	0.984	0.987	0.990
	(0.015)	(0.019)	(0.010)	(0.018)
	[.197]	[.418]	[.196]	[.819]
Any confidant	1.134	0.869	1.084	1.080
	(0.203)	(0.223)	(0.191)	(0.331)
	[.484]	[.585]	[.649]	[.978]
Age (30-39 is base group)
13-19	1.511	1.208	0.766
	(0.595)	(1.172)	(0.360)
	[.294]	[.846]	[.571]
20-29	1.445*	1.211	1.208	2.241
	(0.323)	(0.582)	(0.327)	(1.632)
	[.100]	[.691]	[.486]	[.996]
40-49	1.126	2.536	1.162	2.989
	(0.265)	(1.553)	(0.281)	(3.387)
	[.614]	[.129]	[.534]	[.278]
50-59	1.006	1.900	0.803	0.979
	(0.313)	(1.890)	(0.195)	(1.314)
	[.984]	[.519]	[.366]	[.872]
60+	1.158	3.139	0.780	0.696
	(0.501)	(4.967)	(0.251)	(1.247)
	[.735]	[.470]	[.437]	[.980]
Marital status (married is base group)
Divorced/separated	1.496	1.499	1.814
	(0.517)	(0.780)	(1.449)
	[.244]	[.436]	[.456]
Widowed	0.715	0.688	0.800
	(0.265)	(0.433)	(0.595)
	[.365]	[.553]	[.765]
Never married	3.255*	0.891	2.814**	1.012
	(2.057)	(0.920)	(1.218)	(1.235)
	[.062]	[.911]	[.017]	[.999]
Education (primary education is base group)
No education	0.817	0.683	0.630***	0.525
	(0.153)	(0.269)	(0.113)	(0.250)
	[.279]	[.333]	[.010]	[.186]
Secondary education	0.445	0.566	0.499*	2.155
	(0.245)	(0.736)	(0.182)	(2.306)
	[.142]	[.662]	[.056]	[.332]
Wealth (average wealth is base group)
Poorest	1.644**	1.614	1.172	1.229
	(0.412)	(0.605)	(0.302)	(0.575)
	[.048]	[.201]	[.538]	[.833]
Quite poor	1.124	1.134	0.905	1.011
	(0.212)	(0.310)	(0.166)	(0.340)
	[.532]	[.645]	[.589]	[.961]
Quite wealthy	0.777	0.610	0.923	0.942
	(0.233)	(0.247)	(0.257)	(0.488)
	[.399]	[.223]	[.773]	[.864]
Wealthiest	0.636	0.329	3.668*	4.824
	(0.446)	(0.317)	(2.916)	(8.038)
	[.519]	[.249]	[.100]	[.995]
No. of children	0.992	1.042	0.985	1.009
	(0.024)	(0.040)	(0.018)	(0.038)
	[.734]	[.288]	[.400]	[.432]
Year fixed effects	Yes	Yes	Yes	Yes
No. of observations	1,698	474	1,183	345

Note. Standard errors in parenthesis, p values in brackets. Columns display odds ratios. IPV =intimate partner violence; RE = random effects; FE = fixed effects.

p < .10. **p < .05. ***p < .01.

The results for the random effects and fixed effects logit estimations for men in the South are shown in columns 3 and 4. AIDS conversation networks do not appear to have any impact on men in the South’s acceptance of IPV. We once again find that never-married men are less accepting of IPV than married men. Never-married men are almost 3 times more likely to believe it is proper for a woman to leave a violent marriage compared with married men. In a result somewhat contrary to what was found for women, men in the wealthiest households are less accepting of IPV than men of average wealth. This finding is consistent with Vyas and Watts (2009) who in a systematic review of the published literature on IPV in low- and middle-income countries find household wealth typically reduces levels of violence.

Conclusion

Despite advances in women’s rights over the past decades, IPV remains common in many societies, including Malawi. Understanding what makes IPV unacceptable may assist in creating and targeting violence reduction programs. This article contributes to the literature on acceptance of IPV in three ways. First, we examine how AIDS conversation networks, one proxy for social networks, may impact acceptance of IPV. Second, given the diversity and lineage differences in Malawi, we run separate regressions for each region to allow for more flexibility and accuracy. Finally, we use a unique panel dataset from rural Malawi, which enables us to use not only random effects models, but also fixed effects models, which help alleviate omitted variable bias.

Our results suggest that in the patrilineal North, AIDS conversation networks may help empower women to be less accepting of IPV. In the North, women with larger AIDS conversation networks and at least one confidant are more likely to justify leaving a violent relationship than those that are socially isolated. These conversation networks may represent important sources of social support for women in the North who often have moved away from their home villages upon marriage and consequently do not always have their own family in close proximity. We do not find similar impacts of AIDS conversation networks in the other regions of Malawi, perhaps because in matrilineal areas, family support may be more important and is often in closer proximity.

We find differing impacts of AIDS conversation networks on men’s acceptance of IPV. In the North, men that identify having at least one confidant are less accepting of IPV. This finding is consistent with what we see for women and together may suggest that in the patrilineal North, connections made through AIDS conversation networks seem to be conducive toward reducing acceptance of IPV. However, in the Center region, we find that men with larger AIDS conversation networks are more accepting of IPV, perhaps suggesting that these networks may reinforce violent norms as suggested by Fagan (1989). Finally, in the South, AIDS conversation networks appear to have no impact on acceptance of IPV.

The findings of this study have several implications for developing prevention and response interventions that are context specific. In the patrilineal North, programs that promote social inclusion and discussion of HIV/AIDS and health concerns may have the added benefit of reducing acceptance of IPV. In addition, individual-based approaches to combating HIV/AIDS and IPV may have important spillover effects through social diffusion. In the Center, it may be most important to implement programs that address gender inequalities and the normative use of violence as AIDS conversation networks appear to reinforce violent norms for men. In both the North and the Center, there may be an opportunity to implement programs that simultaneously address both HIV/AIDS and IPV. For example, SASA!, a community mobilization intervention to prevent IPV and reduce HIV risk, was found to reduce acceptance of IPV, incidence of IPV, and risky sexual behaviors in Kampala, Uganda (Abramsky et al., 2014). However, these types of efforts may be less successful in the South Region of Malawi, where there does not appear to be a relationship between AIDS conversation networks and IPV.

While we do find differences across regions, one fairly consistent result is that younger, never-married men and women are more likely to believe it is proper for a woman to leave a violent marriage than older, married individuals. This finding suggests that younger generations are less accepting of IPV. However, it may be especially important for response and prevention efforts to target older, married women who may be at the most risk of IPV.

In this study, we find that AIDS conversation networks sometimes empower women and in other cases may potentially reinforce violence. Future research may try to understand the channels through which those networks work and investigate other factors that impact men’s and women’s acceptance of IPV, perhaps with a specific focus on older generations.

Footnotes

Acknowledgements

Thank you to the MLSFH study for providing data and participants at “The Frontiers of Longitudinal Research in Malawi: Informing Health and Family Policies after the Peak of the AIDS Epidemic” conference for valuable comments.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Author Biography

Allison Shwachman Kaminaga is a lecturer in the Economics Department at Bryant University. She received her PhD in economics from Clark University. Her research interests include intimate partner violence, gender and economic inequalities, social economics, and HIV/AIDS.

References

Abrahams

Jewkes

Hoffman

Laubsher

(2004). Sexual violence against intimate partners in Cape Town: Prevalence and risk factors reported by men. Bulletin of the World Health Organization, 82, 330-337.

Abramsky

Devries

Kiss

Nakuti

Kyegombe

Starmann

. . . Watts

(2014). Findings from the SASA! study: A cluster randomized controlled trial to assess the impact of a community mobilization intervention to prevent violence against women and reduce HIV risk in Kampala, Uganda. BMC Medicine, 12, Article 122. doi:10.1186/s12916-014-0122-5

Alio

A. P.

Clayton

H. B.

Garba

Mbah

A. K.

Daley

Salihu

H. M.

(2011). Spousal concordance in attitudes toward violence and reported physical abuse in African couples. Journal of Interpersonal Violence, 26, 2790-2810. doi:10.1177/0886260510390951

Andersson

Ho-Foster

Mitchell

Scheepers

Goldstein

(2007). Risk factors for domestic physical violence: National cross-sectional household surveys in eight southern African countries. BMC Women’s Health, 16, Article 11. doi:10.1186/1472-6874-7-11

Asiedu

(2016). Lineage ties and domestic violence in Ghana: Evidence from the 2008 demographic and health survey. Journal of Family Issues, 37, 2351-2367. doi:10.1177/0192513X14561523

Beydoun

H. A.

Beydoun

M. A.

Kaufman

J. S.

Zonderman

A. B.

(2012). Intimate parter violence against adult women and its association with major depressive disorder, depressive syptoms and postpartum depression: A systematic review and meta-analysis. Social Science & Medicine, 75, 959-975.

Boyle

M. H.

Georgiades

Cullen

Racine

(2009). Community influences on intimate partner violence in India: women’s education, attitudes towards mistreatment and standards of living. Social Science & Medicine, 69, 691-697.

Brown

Finkelstein

Mercy

(2008). Methods for estimating medical expenditures attributable to intimate partner violence. Journal of Interpersonal Violence, 23, 1747-1766.

Burt

R. S.

(1984). Network items and the general social survey. Social Networks, 6, 293-339.

10.

Campbell

K. E.

Lee

B. A.

(1991). Name generators in surveys of personal networks. Social Networks, 13, 203-221.

11.

Chakwana

C. D.

(2005). Domestic violence. In National Statistical Office [Malawi] and ORC Macro (Eds.), Malawi Demographic and Health Survey 2004 (pp. 265-280). Calverton, MD: National Statistical Office and ORC Macro.

12.

Chiweza

A. L.

(2008). Empowerment in developing countries: Women’s land ownership and inheritance in Malawi. In Banik

(Ed.), Rights and legal empowerment in eradicating poverty (pp. 201-216). Burlington, VT: Ashgate.

13.

Devries

Watts

Yoshihama

Kiss

Schraiber

L. B.

Deyessa

. . . Garcia-Morenao

(2011). Violence against women is strongly associated with suicide attempts: Evidence from the WHO multi-country study on women’s health and domestic violence against women. Social Science & Medicine, 73, 79-86.

14.

Fagan

(1989). Cessation of family violence: Deterrence and dissuasion. Crime and Justice, 11, 377-425.

15.

Faramarzi

Esmailzadeh

Mosavi

(2005). A comparison of abused and non-abused women’s definitions of domestic violence and attitudes to acceptance of male dominance. European Journal of Obstetrics & Gynecology and Reproductive Biology, 122, 225-231.

16.

Farmer

Tiefenthaler

(1997). An economic analysis of domestic violence. Review of Social Economy, 55, 337-357. doi:10.1080/00346769700000004

17.

Filmer

Pritchett

L. H.

(2001). Estimating wealth effects without expenditure data - or tears: An application to educational enrollments in states in India. Demography, 38, 115-132.

18.

Katerndahl

Burge

Ferrer

Becho Wood

R. J.

(2013). Differences in social network structure and support among women in violent relationships. Journal of Interpersonal Violence, 28, 1948-1964.

19.

Kim

Lee

(2013). Prospective study on the reciprocal relationship between intimate partner violence and depression among women in Korea. Social Science & Medicine, 99, 42-48.

20.

Kohler

H. P.

Watkins

S. C.

Behrman

J. R.

Anglewicz

Kohler

I. V.

Thornton

R.L.

. . . Kalilani-Phiri

(2015). Cohort profile: The malawi longitudinal study of families and health. International Journal of Epidemiology, 44, 394-404.

21.

Lawoko

(2008). Predictors of attitudes toward intimate partner violence: A comparative study of men in Zambia and Kenya. Journal of Interpersonal Violence, 23, 1056-1074.

22.

Max

Rice

D. P.

Finkelstein

Bardwell

R. A.

Leadbetter

(2004). The economic toll of intimate partner violence against women in the United States. Violence and Victims, 19, 259-272.

23.

National Statistical Office (NSO) and ICF Macro. (2011). Malawi Demographic and Health Survey 2010. Zomba, Malawi and Calverton, MD, USA: NSO and ICF Macro.

24.

Reniers

(2003). Divorce and remarriage in rural Malawi (Special Collection 1). Demographic Research, (Article 6), 175-206.

25.

Sarkar

(2008). The impact of intimate partner violence on women’s reproductive health and pregnancy outcome. Journal of Obstetrics and Gynecology, 28, 266-271. doi:10.1080/01443610802042415

26.

Sedziafa

A. P.

Tenkorang

E. Y.

Owusu

A. Y.

(2018). Kinship and intimate partner violence among married women in Ghana: A qualitative exploration. Journal of Interpersonal Violence, 33(14), 2197-2224.

27.

Sen

(1999). Enhancing women’s choices in responding to domestic violence in Calcutta: A comparison of employment and education. The European Journal of Development Research, 11(2), 65-86. doi:10.1080/09578819908426739

28.

Stets

J. E.

(1991). Cohabiting and marital aggression: The role of social isolation. Journal of Marriage and the Family, 53, 669-680.

29.

Tauchen

Witte

A. D.

Long

S. K.

(1991). Domestic violence: A nonrandom affair. International Economic Review, 32, 491-511.

30.

Uthman

O. A.

Moradi

Lawoko

(2009). The independent contribution of indivdiual-, neighbourhood-, and country-level socioeconomic position on attitudes towards intimate partner violence against women in sub-Saharan Africa: A multilevel model of direct and moderating effects. Social Science & Medicine, 68, 1801-1809.

31.

Vyas

Watts

(2009). How does economic empowerment affect women’s risk of intimate partner violence in low and middle income countries? A systematic review of published evidence. Journal of International Development, 21, 577-602. doi:10.1002/jid.1500

32.

White

(2007). Malawi: Country gender profile. Lilongwe, Malawi: Japan International Cooperation Agency.

33.

Wilcox

(2000). “Me mother’s bank and me nanan’s, you know, support!” Women who left domestic violence in England and issues of informal support. Women’s Studies International Forum, 23, 35-47.

34.

World Health Organization. (2013) Global and regional estimates of violence against women: Prevalence and health effects of intimate partner violence and non-partner sexual violence. Author. Retrieved from http://www.who.int/reproductivehealth/publications/violence/9789241564625/en/