Violence against women, including intimate partner violence (IPV), is a grave violation of human rights and is highly prevalent worldwide. The highest rates have been reported in resource-limited settings, although wide variations exist.1,2 Personal history of IPV has been associated with numerous adverse sexual and reproductive health consequences for women, including low self-efficacy for condom use,3 inconsistent condom use,4,5 unplanned pregnancies,6 and a range of other perinatal outcomes.7 The specific mechanisms underlying these associations are not well-understood, but previous work suggests at least two hypotheses. First, at the individual level, IPV may lead to, or may be a marker for, direct controlling behaviors perpetrated by male partners to promote pregnancy.8 Second, even when their male partners do not directly engage in coercive behaviors, women may fear violent reprisals if they live in communities where IPV is highly prevalent or where IPV is culturally sanctioned. This latter hypothesis suggests that context may exert an effect independent of direct victimization at the individual level, ie, a “culture of terror”9 may indirectly compromise women's sexual decision-making. Studies that explicitly model the pattern of violence against other women in the community would help to disentangle these two hypotheses, but few studies have attempted to do so.10
There is some evidence that women desire greater reproductive control but are unable to attain it. In diverse settings across sub-Saharan Africa, a desire to achieve optimal birth spacing is the primary reason underlying demand for family planning services;11 however, women's preferences for longer birth intervals are not met through existing family planning programs.12 With the exception of one study of married women living in the Philippines, which showed that greater autonomy in household decision-making was associated with longer birth-to-conception intervals,13 it is unclear whether IPV is associated with shortened birth intervals. The public health implications of these two gaps in the literature are important, given that shortened birth intervals have been associated with worsened maternal and child health, including uterine rupture,14,15 uteroplacental bleeding disorders,16,17 and a wide range of adverse perinatal outcomes.18,19 Therefore, we undertook this study to estimate the association between IPV and birth interval length and to estimate the comparative influence of personal history of IPV compared with community prevalence of IPV.
MATERIALS AND METHODS
The data for this analysis were drawn from the Demographic and Health Surveys, which are collected for their use in policy formation, program planning, and monitoring and evaluation. The Demographic and Health Surveys are publicly available population-level surveys implemented by host country governments with funding and technical assistance from ICF Macro and the United States Agency for International Development. Data on fertility have been collected through the Demographic and Health Surveys since 1984, and the quality is generally regarded to be among the highest for data on births and infant and child deaths in resource-limited settings20 and comparable in accuracy to prospectively gathered data.21 Rigorous training and close supervision of field staff ensure high response rates at the individual and household levels.22 Each survey used a multistage stratified design with probabilistic sampling (with each household having an equal probability of selection) and was designed to be nationally representative of all women of reproductive age (ie, 15–49 years). The primary sampling unit was the smallest clustering unit of analysis. In the Demographic and Health Surveys, the primary sampling unit typically represents a village or cluster of villages in rural areas and a ward or residential neighborhood in urban areas; for ease of exposition in the succeeding discussion, we use the term village to refer to this level of analysis. The data collection procedures for the Demographic and Health Surveys were approved by the ICF Macro Institutional Review Board as well as by the relevant Ethical Review Boards in the host country for each survey. All participants provided verbal informed consent. Additional details on staff training, pretesting, and other survey procedures are detailed in the Demographic and Health Surveys final reports for each country, which are publicly available from the ICF Macro web site (http://www.measuredhs.com).
Our analysis was reviewed by the Harvard School of Public Health Office on Human Research Administration and was considered exempt from full review, because it was based on anonymous public-use data with no identifiable information on participants. We selected recent surveys administered in continental sub-Saharan Africa in which a randomly selected subset of ever-married women of reproductive age were administered a standardized questionnaire module containing questions about IPV. Eleven Demographic and Health Surveys conducted between 2004 and 2008 in the following countries met these criteria: Cameroon (2004), Democratic Republic of the Congo (2007), Ghana (2008), Kenya (2008), Liberia (2007), Malawi (2004), Nigeria (2008), Rwanda (2005), Uganda (2006), Zambia (2007), and Zimbabwe (2005–2006). To create the variables derived from the village level, we used data from all women who had been administered the IPV module. For the regression analyses, we included data from only multiparous women who had been administered the IPV module. Therefore, the birth data included information only from second and higher-order births.
We examined birth interval as the primary outcome variable, defining it as the length of time (in months) between two live births. There were two primary explanatory variables of interest, both related to IPV. In the Demographic and Health Surveys, lifetime exposure to IPV was measured with a modified Conflict Tactics Scale,23 which inquired about whether the partner or husband had ever engaged in different acts of physical and sexual violence, ranging from being pushed or slapped to being burned or forced to have sexual intercourse. Acts of violence were classified into one of two mutually exclusive categories: acts of physical violence (including being pushed, slapped, punched, kicked, choked, burned, attacked with a weapon, or threatened with a weapon) and acts of sexual violence (including being forced into sexual intercourse or other sexual acts). Using these definitions, personal history of IPV was measured at the individual level. To measure the community prevalence of IPV, we created a derived variable at the village level by calculating the proportion of women in the village who reported a personal history of IPV.
All analyses were conducted using Stata 12.0. We used multilevel linear regression to model the variation in birth intervals (see the Appendix, available online at http://links.lww.com/AOG/A293) while adjusting for the following additional explanatory variables measured at the level of the individual: age; residence in an urban (compared with rural) setting; domestic partnership status (currently married or not formally married but cohabiting with a partner in a consensual union compared with never married, divorced, or widowed); religion (Christian compared with Muslim compared with other); household headship (respondent is the head of the household compared with other family member is head of the household); professional occupation status (employed outside the home and in a professional, technical, managerial, clerical, or sales capacity compared with employed in the home or outside the home in any other capacity); educational attainment; within-country quintiles of household asset wealth;24 and contraception use (modern type of contraception, such as intrauterine device, diaphragm, condom, or spermicide, compared with traditional type, such as abstinence, lactational amenorrhea, or withdrawal, compared with none). Three explanatory variables were specific to the level of births: birth order, cumulative number of boys at the time of the birth, and whether the previous sibling died. We estimated two regression models. Model 1 focused on physical violence as the primary exposure of interest while adjusting for the explanatory variables listed. Model 2 focused on sexual violence as the primary exposure of interest while adjusting for the explanatory variables listed. Our modeling approach allowed us to partition the variation in birth intervals to the four different levels of variance (births, individuals, villages, and countries) and to calculate intraclass correlations, ie, the proportion of variance in birth intervals that could be attributed to countries, villages, and women. Finally, we also estimated three-level multilevel regression models to produce estimated associations specific to each country.
We implemented two additional analyses to determine whether our findings were robust to alternative categorizations of the variables used. Many studies relating birth intervals to maternal or child health outcomes have adopted varying cut-offs to define a “short” birth interval, eg, less than 18 months.14,15,18,19 To more closely relate our findings to these other studies, we estimated the associations between IPV and short birth intervals using logistic regression models (while accounting for clustering in a fashion similar to the primary specification), with the dependent variables defined according to three different cut-offs: less than 18 months, less than 24 months, and less than 36 months. Second, we disaggregated physical violence into two categories using a classification system common to epidemiological studies in this body of literature25: “severe” acts of violence, including being pushed or slapped, and “more severe” acts, including being punched, kicked, choked, burned, attacked with a weapon, or threatened with a weapon. We then fit the multilevel regression models to the data using these variables as the primary exposures. Although victimized persons may experience all forms of IPV as uniformly traumatic, we would expect to observe a gradient in the effects of IPV to further support the plausibility of our findings.
The response rates for the country surveys ranged from 85.6% (Zimbabwe) to 97.9% (Rwanda and Lesotho; Appendix Table 1, http://links.lww.com/AOG/A293). The number of villages for the country surveys ranged from 298 to 886 (Appendix Table 2, http://links.lww.com/AOG/A293). Of the 122,091 women who completed interviews, our analytic sample included data from 46,697 women (38.3%) who reported 168,457 previous births and who were also administered the module on IPV. Summary statistics for the analytic sample are displayed in Table 1. Among the women in the analytic sample, 11,730 (25.1%) reported a personal history of physical violence and 4,935 (10.6%) reported a personal history of sexual violence.
The median birth interval pooled across countries was 29 months (interquartile range, 23–40 months). The country-level median birth interval was longest among women in Zimbabwe (36 months) and shortest among women in Uganda (27 months). Consistent with this, a null regression model with no explanatory variables revealed statistically significant variation in birth intervals at all levels. Birth interval variation attributable to the geographic environment (ie, the combined effect of village and country) was comparable to birth interval variation between women: 4.7% of the total variation in birth intervals resulted from between-individual differences, 2.9% resulted from between-village differences, and 2.7% resulted from between-country differences. After multivariable adjustment (Table 2), 4.3% of the total variation was attributable to individuals, 2.0% was attributable to villages, and 2.1% was attributable to countries. In short, accounting for birth-level and individual-level explanatory variables did not substantively alter the apportioning of the total variation in birth intervals to the different levels of analysis. Compared with the regression models containing only individual sociodemographic variables and birth-level variables, addition of personal history of physical or sexual violence explained an additional 0.1–0.3% of the variation in birth intervals, whereas addition of community prevalence of physical or sexual violence explained an additional 1.9–2.1% of the variation.
Shorter birth intervals were associated with personal history of physical violence (b=−0.91, 95% confidence interval [CI] −1.16 to −0.65) as well as the community prevalence of physical violence against women (b=−2.38, 95% CI −3.42 to −1.35) when these variables were considered separately in unadjusted regression models. These associations persisted after multivariable adjustment. Personal history of sexual violence also was associated with shorter birth intervals (b=−0.76, 95% CI −1.22 to −0.30), as was the community prevalence of sexual violence against women (b=−1.74, 95% CI −3.18 to −0.29). The magnitudes of these estimates were comparable with those estimated for physical violence. Country-specific analyses revealed between-country differences in the statistical significance and magnitude of the associations between IPV and birth intervals (Table 3). The strongest associations with personal history of IPV were observed for Ghana and Cameroon, and the strongest associations with community prevalence of IPV were observed for Kenya and Nigeria.
A number of other important patterns emerged from the multivariable regression models. Consistent with previous research,13 measures of socioeconomic empowerment were associated with longer birth intervals, including household headship, professional occupation status, and greater educational attainment and household asset wealth. The magnitudes of these associations were comparable to the estimated effect of IPV: women in professional occupations had a predicted mean birth interval of 36.5 months (compared with 35.7 months among women who were unemployed or in nonprofessional occupations), and women with a higher education had a predicted mean birth interval of 38.1 months (compared with 35.8 months among women with only a primary education).
In the sensitivity analyses, using logistic multilevel regression modeling, we found that both personal history of IPV and community prevalence of IPV were associated with increased odds of shortened birth intervals (Appendix Table 3, http://links.lww.com/AOG/A293). When a cut-off of less than 18 months was used, both the community prevalence of physical violence (adjusted odds ratio 1.47, 95% CI 1.25–1.73) and the community prevalence of sexual violence (adjusted odds ratio 1.28, 95% CI 1.01–1.62) were associated with increased odds of shortened birth intervals. When we disaggregated physical violence by levels of severity, a dose–response gradient was apparent: the estimated effects of more severe acts of violence (eg, being choked, burned, attacked with a weapon) were larger in both magnitude and statistical significance at the levels of both the individual and the community (Appendix Table 4, http://links.lww.com/AOG/A293).
Using multilevel modeling to analyze data on 168,457 births and 46,697 reproductive-age multiparous women from 11 countries in sub-Saharan Africa, we investigated the association between IPV and birth intervals. Our study yielded four key findings. First, we found a statistically significant association between IPV and shortened interbirth intervals. Second, we found evidence of these associations at the levels of both the individual and the community. Third, at both the level of the individual and the community, the estimated effects of more severe acts of physical violence on birth intervals were larger in both magnitude and statistical significance compared with severe acts. Fourth, these estimated associations were comparable in magnitude with known correlates of interbirth intervals that traditionally have been the focus of policy and programming, such as women's education and employment status.
The association between personal history of IPV and shortened birth intervals is consistent with recently published findings on reproductive coercion.8 What is new about our study is that we found evidence of adverse effects of IPV not only at the individual level but also at the level of the community: using multilevel modeling, we found that an increased community prevalence of IPV was associated with shortened birth intervals independent of a woman's personal history of IPV. Our interpretation of these data is that a greater prevalence of IPV at the village level may contribute to a pervasive sense of “everyday violence”26 and acquiescence of unequal power relations between men and women,27 which, in turn, undermine women's self-efficacy for achieving their preferred patterns of birth spacing. These findings are consistent with previous work in which community prevalence of IPV has been associated with inconsistent condom use5 and unintended pregnancy.28
The independent associations between interbirth intervals and all of the explanatory variables investigated in our study—including IPV—were small in magnitude, with predicted interbirth interval differences of approximately 1–2 months. However, even small population shifts in key health parameters may be consequential.29 At the level of the individual, increasing a specific woman's subsequent interbirth interval by only 1–2 months may not result in substantive changes in health outcomes for her or her children. At the population level, however, a 1-month to 2-month upward shift in the entire population distribution of interbirth intervals could correspond to large differences in health outcomes.29,30 Policy-oriented recommendations in this area traditionally have focused on improving women's education and economic status.31–33 In our analysis, the estimated effects of education and occupational status were comparable to the estimated effects of IPV, suggesting that additional attention in this area is warranted.
Several limitations should be considered when interpreting our findings. First, due to the questionnaire design, the practice of birth displacement potentially leads to over-estimations of birth intervals for more recent births. Although this practice may lead to biased estimates of fertility rates and trends in fertility rates,20 for it to falsely shift the association between birth intervals and IPV away from the null, Demographic and Health Surveys interviewers would need to displace births more often when interviewing high-parity women who did not report histories of IPV. Second, although there is little longitudinal evidence to support the hypothesis that women are at greater risk for victimization when they become pregnant,34 the cross-sectional design of the study does not permit us to rule out reverse causality as a potential explanation for the findings. However, even if the reverse could be shown (ie, pregnancy causing an increased probability of exposure to IPV), it would be unlikely for an individual woman's pregnancy to causally influence the community prevalence of IPV. Third, our sample contained only multiparous women who were administered the IPV module. In the Demographic and Health Surveys, the IPV module was administered only to a randomly selected subset of ever-married women of reproductive age. Our conclusions therefore do not extend to never-married or nulliparous women. Related to this limitation, no sampling weights were provided for analyses restricted to multiparous respondents to the IPV module, so our estimates could not be reweighted to be nationally representative. Nonetheless, it is the largest study of its kind to date suggesting broad generalization across diverse settings in sub-Saharan Africa. Finally, between-country heterogeneity would suggest caution in the pooled analyses. Partitioning the variation in interbirth intervals between the multiple levels of analysis suggested that 2.1% of the variation was attributable to inherent differences between countries and 2.0% was attributable to inherent differences between villages. These intraclass correlations are comparable with those observed in other multilevel analyses of neighborhood and health data.35
In summary, our analysis has shown that IPV, both at the individual and community level, is associated with shortened birth intervals among women in sub-Saharan Africa. Efforts to improve women's health globally often involve efforts to alter the structural conditions that shape women's risk for poor health, but the literature on birth spacing has been largely limited to analyses of sociobehavioral factors measured at the individual level. Additional attention should be given to violence against women in research on reproductive health.
1. Tsai AC, Leiter K, Wolfe WR, Heisler M, Shannon K, Phaladze NA, et al.. Prevalence and correlates of forced sex perpetration and victimization in Botswana and Swaziland. Am J Pub Health 2011;101:1068–74.
2. Garcia-Moreno C, Jansen HA, Ellsberg M, Heise L, Watts CH, WHO Multi-country Study on Women's Health and Domestic Violence against Women Study Team. Prevalence of intimate partner violence: findings from the WHO multi-country study on women's health and domestic violence. Lancet 2006;368:1260–9.
3. Sayles JN, Pettifor A, Wong MD, MacPhail C, Lee SJ, Hendriksen E, et al.. Factors associated with self-efficacy for condom use and sexual negotiation among South african youth. J Acquir Immune Defic Syndr 2006;43:226–33.
4. Wingood GM, DiClemente RJ. The effects of an abusive primary partner on the condom use and sexual negotiation practices of African-American women. Am J Public Health 1997;87:1016–8.
5. Tsai AC, Subramanian SV. Proximate context of gender unequal norms and women's HIV risk in sub-Saharan Africa. AIDS 2012;26:381–6.
6. Martin SL, Kilgallen B, Tsui AO, Maitra K, Singh KK, Kupper LL. Sexual behaviors and reproductive health outcomes: associations with wife abuse in India. JAMA 1999;282:1967–72.
7. Coker AL. Does physical intimate partner violence affect sexual health? A systematic review. Trauma Violence Abuse 2007;8:149–77.
8. Miller E, Decker MR, McCauley HL, Tancredi DJ, Levenson RR, Waldman J, et al.. Pregnancy coercion, intimate partner violence and unintended pregnancy. Contraception 2010;81:316–22.
9. Taussig M. Culture of terror–space of death. Roger Casement's Putumayo Report and the explanation of torture. Comp Stud Soc Hist 1984;26:467–97.
10. Pallitto CC, Campbell JC, O'Campo P. Is intimate partner violence associated with unintended pregnancy? A review of the literature. Trauma Violence Abuse 2005;6:217–35.
11. Jansen WH 2nd. Existing demand for birth spacing in developing countries: perspectives from household survey data. Int J Gynecol Obstet 2005;89:S50–60.
12. Ross JA, Winfrey WL. Contraceptive use, intention to use and unmet need during the extended postpartum period. Int Fam Plan Perspect 2001;27:20–7.
13. Upadhyay UD, Hindin MJ. Do higher status and more autonomous women have longer birth intervals? Results from Cebu, Philippines. Soc Sci Med 2005;60:2641–55.
14. Bujold E, Gauthier RJ. Risk of uterine rupture associated with an interdelivery interval between 18 and 24 months. Obstet Gynecol 2010;115:1003–6.
15. Shipp TD, Zelop CM, Repke JT, Cohen A, Lieberman E. Interdelivery interval and risk of symptomatic uterine rupture. Obstet Gynecol 2001;97:175–7.
16. Conde-Agudelo A, Belizan JM. Maternal morbidity and mortality associated with interpregnancy interval: cross sectional study. BMJ 2000;321:1255–9.
17. Getahun D, Oyelese Y, Salihu HM, Ananth CV. Previous cesarean delivery and risks of placenta previa and placental abruption. Obstet Gynecol 2006;107:771–8.
18. Conde-Agudelo A, Belizan JM, Norton MH, Rosas-Bermudez A. Effect of the interpregnancy interval on perinatal outcomes in Latin America. Obstet Gynecol 2005;106:359–66.
19. Fuentes-Afflick E, Hessol NA. Interpregnancy interval and the risk of premature infants. Obstet Gynecol 2000;95:383–90.
20. Arnold F. Assessment of the quality of birth history data in the Demographic and Health Surveys. An assessment of DHS-I data quality Demographic and Health Surveys Methodological Reports, No 1. Columbia (MD): Institute for Resource Development/Macro Systems; 1990:83–11.
21. Garenne M, Van Ginneken J. Comparison of retrospective surveys with a longitudinal follow-up in Senegal: SFS, DHS and Niakhar. Eur J Popul 1994;10:203–21.
22. Vaessen M, Thiam M, Le T. The Demographic and Health Surveys. Household sample surveys in developing and transition countries Studies in Methods (Series F No 96). New York (NY): United Nations; 2005.
23. Straus MA. Measuring intrafamily conflict and violence: the conflict tactics (CT) scales. J Marriage Fam 1979;41:75–88.
24. Filmer D, Pritchett LH. Estimating wealth effects without expenditure data–or tears: an application to educational enrollments in states of India. Demography 2001;38:115–32.
25. Ellsberg M, Heise L, Pena R, Agurto S, Winkvist A. Researching domestic violence against women: methodological and ethical considerations. Stud Fam Plann 2001;32:1–16.
26. Shannon K, Kerr T, Allinott S, Chettiar J, Shoveller J, Tyndall MW. Social and structural violence and power relations in mitigating HIV risk of drug-using women in survival sex work. Soc Sci Med 2008;66:911–21.
27. Jewkes R. HIV/AIDS. Gender inequities must be addressed in HIV prevention. Science 2010;329:145–7.
28. Pallitto CC, O'Campo P. Community level effects of gender inequality on intimate partner violence and unintended pregnancy in Colombia: testing the feminist perspective. Soc Sci Med 2005;60:2205–16.
29. Rose G. Sick individuals and sick populations. Int J Epidemiol 1985;14:32–8.
30. Rose G. Strategy of prevention: lessons from cardiovascular disease. Br Med J (Clin Res Ed) 1981;282:1847–51.
31. Ainsworth M, Beegle K, Nyamete A. The impact of women's schooling on fertility and contraceptive use: a study of fourteen sub-Saharan African countries. World Bank Econ Rev 1996;10:85–122.
32. Gakidou E, Vayena E. Use of modern contraception by the poor is falling behind. PLoS Med 2007;4:e31.
33. Allen RH. The role of family planning in poverty reduction. Obstet Gynecol 2007;110:999–1002.
34. Jasinski JL. Pregnancy and domestic violence: a review of the literature. Trauma Violence Abuse 2004;5:47–64.
35. Subramanian SV, O'Malley AJ. Modeling neighborhood effects: the futility of comparing mixed and marginal approaches. Epidemiology 2010;21:475–8.