REPRODUCTIVE HEALTH AND DOMESTIC VIOLENCE: ARE THE POOREST WOMEN UNIQUELY DISADVANTAGED?*

Domestic violence is a pervasive form of violence against women. Worldwide, 10% to over 50% of women report having experienced domestic violence (Heise, Ellsberg, and Gottemoeller 1999; Kishor and Johnson 2004). For the individual who experiences domestic violence, the consequences can range from

aches and pains to severe injury, chronic mental and physical illness, and even death (Campbell 2002; Heise et al. 1999). Of particular relevance to the study of demography is that domestic violence can result in an increased risk of adverse reproductive health outcomes (Campbell 2002; Gazmararian, Adams, and Pamuk 1996). However, the validity of this relationship needs to be carefully examined because adverse reproductive health outcomes are more likely in poverty-stricken settings (Gazmararian et al. 1996; Pebley, Goldman, and Rodriguez 1996), and such settings have also been associated with an increased risk of domestic violence (Moore 1999; Plichta and Abraham 1996; Stark and Flitcraft 1996). Accordingly, we explore the relationship between poverty and domestic violence and the relative roles they play as obstacles to women's reproductive health as measured by three different reproductive health outcomes. The results suggest that women who are both poor and have experienced violence are no more disadvantaged in terms of their reproductive health than are women who are wealthier and have also experienced violence; in fact, the health outcomes for wealthier women who have been abused are often less favorable than those of poorer abused women.

BACKGROUND: VIOLENCE, POVERTY, AND REPRODUCTIVE HEALTH

A common assumption in the literature on domestic violence is that women who are poor are more likely to experience violence than women who are not poor (Ellsberg et al. 1999; Heise 1998; Jewkes 2002). The argument is that families living in impoverished conditions are subject to higher levels of stress than families not living in poverty, and as a result, poor families are more prone to family violence than families that are not impoverished (DeKeseredy and Schwartz 2002; Martin et al. 1999; Raphael 2001). However, empirical findings have been mixed, with some analyses indicating that poverty is a strong predictor of violence (e.g., Ellsberg et al. 1999); others finding poverty to be an insignificant factor after controlling for variables such as education and residence (e.g., Diop-Sidib? 2001); and still others finding that poverty has an effect on the likelihood of domestic violence in some geographic locations but not others (e.g., Johnson 2003; Martin et al. 1999).

The varying effects of household economic status in the domestic violence literature may be due in part to the large variability in defining household wealth, income, or socioeconomic status. Some researchers have used income (e.g., Kirn and Cho 1992; Rodgers 1994), others have used information on assets (e.g., Diop-Sidib? 2001; Johnson 2003; Martin et al. 1999), and still others have used a composite measure of socioeconomic status (e.g., Ellsberg et al. 1999; Hoffman, Demo, and Edwards 1994), with definitions and data being subject to different sets of assumptions and caveats. What is needed for a comparative assessment of the role that poverty plays in mediating the association between violence and reproductive health disadvantage is an indicator of economic status that is similarly defined and measured across countries and is free of confounding variables that have an independent association with domestic violence and/or reproductive health.

A growing literature links violence to adverse reproductive health outcomes. Evidence from health facilities suggests that abused women's reproductive health is compromised through much higher rates of gynecological problems, HIV and sexually transmitted infections (STIs), miscarriages, abortions, low birth weight, and unwanted pregnancy (Campbell 2002). However, there is little research on whether women who are both poor and have experienced violence have more disadvantaged reproductive health outcomes than women who are not poor and have experienced violence. Most research has been restricted to special populations (e.g., populations in housing projects or in shelters) based on data from the developed world. With only a few notable exceptions, almost no research has examined the relative effect on women's reproductive health of both being poor and experiencing domestic violence in developing countries.

We attempt to fill this research void by using data from the Demographic and Health Surveys (DHS) from three countries to examine the relationship among each of three different reproductive health outcomes, spousal violence, and a household poverty-wealth indicator. DHS surveys are conducted in the developing world by using face-to-face interviews, and they typically collect nationally representative data on demographic and health indicators for women aged 15-49, as well as on the characteristics of the respondents' households. Reproductive health disadvantage is measured in terms of three indicators: (1) having a non-live birth, (2) having an STI or symptoms of an STI, and (3) having an unwanted birth. The first two measures directly assess reproductive health outcomes, and all three measures are indicators of women's ability to fulfill their reproductive health needs. Research suggests that all three outcomes are likely to have significant associations with women's experience of violence.

Having a Non-live Birth and Violence

Abusive spousal behavior can pose direct risks to the viability of a pregnancy, including physical trauma to the abdomen due to violence or intentional abortion to preclude an unwanted birth. Jejeebhoy (1998) found that women in India who had ever been beaten by their husbands were almost twice as likely to experience a fetal loss, even after controlling for a variety of social, economic, and demographic factors. Indirect risks to the viability of the fetus that have been associated with domestic violence include increased levels of stress and delays in seeking antenatal care (Newberger et al. 1992; Taggart and Mattson 1995).

STIs and Violence

The literature indicates that women who have experienced spousal violence are more likely to suffer a range of gynecological problems, including STIs (cf. Augenbraun, Wilson, and Allister 2001 ; Plichta and Abraham 1996; Schei 1991). Maman et al. (2002) found that HIV-positive women were more than twice as likely to report current experience of violence and to report a greater frequency of violent events than HIV-negative women. Several pathways are likely to lead to higher rates of STIs in abused women. The experience or even the threat of domestic violence tends to limit women's ability to control when and whether to have sexual relations or to negotiate condom use. Wives of abusive men may be at higher risk of STIs because abusive men appear to be more likely to indulge in other high-risk behaviors that are positively associated with STIs, such as alcohol abuse, promiscuity, and polygamy (Abrahams et al. 2004). Finally, domestic violence may also be an outcome of the disclosure of an STI to a partner (Zierler, Witbeck, and Mayer 1996).

Unintended Pregnancies and Violence

Unintended pregnancy, in part a consequence of non-use or inconsistent use of contraception, is positively associated with domestic violence (Campbell et al. 1995; Gazmararian et al. 1995; Gazmararian et al. 2000). Women in abusive relationships are much less likely to use condoms than nonabused women and are more likely to experience further abuse if they attempt to discuss condom use with their partners (Wingood and DiClemente 1997). Using data on women in the United States, Goodwin et al. (2000) found that women with unintended pregnancies had 2.5 times the risk of experiencing physical abuse compared with women whose pregnancies were intended. In addition, they found that although women on Medicaid were significantly more likely to experience abuse, the association between unwanted pregnancy and abuse was stronger for women who were not on Medicaid than for women who were.

In the sections that follow, we examine the association of each of these three reproductive health outcomes with the experience of spousal violence and poverty-wealth status for women in Cambodia, the Dominican Republic, and Haiti. The analysis also explores whether women who are poor and have experienced violence are the ones most likely to have experienced reproductive health disadvantage. The use of similarly defined variables across all countries increases the validity of cross-country comparisons and the robustness of conclusions.

DATA AND MEASUREMENT

Data

This study uses data from the DHS conducted in Cambodia and Haiti in 2000 and in the Dominican Republic in 2002. Although these countries were selected because they have an identical set of questions on domestic violence, it is notable that the countries are culturally distinct, reflecting Asian, Afro-Caribbean, and Hispanic-Caribbean backgrounds. The degree to which the findings of this analysis are consistent across the economic and cultural diversity represented by these countries is expected to contribute further to a generalized understanding of the interaction of poverty and violence and the association of that interaction with women's reproductive health.

In all three countries, all women aged 15-49 in the sample households were eligible for the DHS. A subsample of DHS households was selected for the domestic violence questions. In keeping with the ethical guidelines provided by the World Health Organization (WHO 2001) on the conduct of domestic violence research, special training was provided for this component of the DHS. In addition, two mechanisms were used to protect the security of women: within each household, only one randomly selected eligible woman received the module, and the module was not implemented if privacy could not be obtained. Weights were constructed to make the data on violence nationally representative.1 The analysis sample is restricted to women who have ever been in a union, unless otherwise specified. The resulting sample sizes are 2,403 for Cambodia, 6,807 for the Dominican Republic, and 2,347 for Haiti.

Multivariate logistic regression techniques are used to explore the linkages among domestic violence, poverty, and reproductive health outcomes. This approach makes it feasible to examine, for each of the three reproductive health outcomes defined below, the separate and interactive association with being poor and/or experiencing violence net of all other variables that are also known to be associated with the specific reproductive health outcome.

Dependent Variables

Three aspects of women's reproductive health are examined:

Had a pregnancy that did not end in a live birth. All women were asked whether they had ever had a pregnancy that did not end in a live birth (i.e., ended in a miscarriage or abortion or was a stillbirth). In defining this variable, no attempt is made to differentiate among the three types of pregnancy terminations because respondents are not always able to correctly differentiate among them and because all three types of terminations are expected to be positively associated with domestic violence. In the analysis, women who have ever had a pregnancy that did not end in a live birth are coded as 1; all others are coded as 0. Only women who had ever had a birth or had reported a terminated pregnancy are in the subsample of women for this analysis.

Had an STI or symptoms of an STI in the 12-month period preceding the survey. This variable is based on all ever-married women in the sample and derives from selfreports. In all three countries, women were asked a similar combination of questions to determine whether they had had an STI in the past year. This combination included a direct question ("In the past 12 months, have you had a sexually-transmitted disease?") as well as one or more questions on possible symptoms (e.g., "In the past 12 months, have you had a genital sore or ulcer?"). If women said "yes" to one or more of these questions, they are counted as having had an STI/STI symptom in the past 12 months and are coded 1 for the analysis. All other women are coded 0.

Had an unwanted birth in the five years preceding the survey. Women with a live birth in the five years preceding the survey and women who were pregnant at the time of the survey were asked whether, at the time they became pregnant with their last birth/ current pregnancy, they had wanted a child then, later, or not at all. Women who did not want another child at all at the time they first became pregnant with their last live-born child or current pregnancy are coded 1, and others are coded 0. Only women who had a birth in the five years preceding the survey or who were currently pregnant are in the subsample for this analysis.2

Key Explanatory Variables

The two explanatory variables central to the discussion in this paper are domestic violence and the poverty-wealth status of the household to which the respondent belongs. Accordingly, the definitions of these variables are discussed in some detail.

Domestic violence. Domestic violence is defined here as violence experienced by women at the hands of their current or earlier spouse(s).3 The violence indicator is derived from responses given by ever-married women to three different sets of questions on violence. The first set is based on a modified and shortened version of the conflict tactics scale (CTS) used by Straus (1990) and asks each respondent whether her husband (current if married or last if formerly married) ever did any of the following to her: (1) "Push you, shake you, or throw something at you?"; (2) "Slap you or twist your arm?"; (3) "Punch you with his fist or with something that could hurt you?"; (4) "Kick you or drag you?"; (5) "Try to strangle you or burn you?"; (6) "Threaten you with a knife, gun, or other type of weapon?"; (7) "Attack you with a knife, gun, or other type of weapon?"; (8) "Physically force you to have sexual intercourse even when you did not want to?"; and (9) "Force you to perform types of other sexual acts you did not want to?" Women could answer "yes" or "no" to each item, and when the answer was "yes," women were asked about the frequency of the act in the 12 months preceding the survey.

Although it is the most commonly used quantitative measure of domestic violence, the original CTS has been criticized on several grounds (cf. DeKeseredy and Schwartz 1998). The modified CTS in use by the DHS accounts for the two major criticisms by including questions on sexual violence and by not assuming that violence takes place only in circumstances characterized by conflict. Further, the literature on the CTS has emphasized that physical trauma is but one component of the damage sustained by an abused spouse (for a brief review, see Gordon 2000). Accordingly, the violence measure used here does not weight violence that has resulted in injury more than violence with no reported injuries.

In addition to the modified CTS, women were asked whether they had experienced violence at the hands of anyone other than their current or last husband, using the question, "From the time you were 15 years old, has anyone other than your (current/last) husband hit, slapped, kicked, or done anything else to hurt you physically?" Women who responded "yes" to this question were asked about the person(s) who had done this. A similar question was used to measure violence during pregnancy.

In this article, a woman is counted as having experienced domestic violence if she responded "yes" to either one or more of the modified CTS questions, if she reported that a previous husband was violent since she was 15, or if she reported violence during a pregnancy perpetrated by a current or past husband. Based on this definition, 17% of evermarried women in Cambodia, 22% in the Dominican Republic, and 29% in Haiti have ever experienced domestic violence.4 A preliminary analysis of these data indicates that few factors consistently distinguish women who have experienced spousal violence from those who have not. These factors include having a husband who abuses alcohol and a family history of violence, but do not include characteristics such as educational attainment or employment status (Kishor and Johnson 2005).

The poverty-wealth measure. In addition to measuring violence, this article rests critically on the ability to estimate the poverty-wealth status of households in a uniform and comparable manner. The wealth index used here was recently developed and tested in a large number of countries with regard to inequities in household income, the use of health services, and health and other outcomes (Filmer and Pritchett 2001; Gwatkin et al. 2000). It is an indicator of wealth that is best interpreted as an indicator of a household's permanent income status.

This wealth index is constructed using DHS data on household assets and dwelling characteristics (including country-specific assets) and principle components analysis. The asset data include household ownership of a number of consumer items, such as a television, a radio, a bicycle, or a car; the dwelling characteristics data include type of drinking water, sanitation facilities, roofing, and flooring. Each asset/characteristic is assigned a weight (factor score) that is generated through principle components analysis, and the resulting asset scores are standardized in relation to a standard normal distribution with a mean of O and a standard deviation of 1 (Rutstein and Johnson 2004). Each household is then assigned a score for each asset/characteristic, and the scores are summed by household. The sample is then divided into population quintiles; each quintile is designated a rank, from 1 (poorest) to 5 (wealthiest), and individuals are ranked according to the total score of the household in which they live.5

Control Variables

Several variables are used as control variables in the multivariate analyses (see Appendix Table A for the distributions by country). Variables common to the analysis of all three reproductive health outcomes are as follows: age of the respondent measured in years, her number of years of education, whether she is exposed to any media (television, radio, or newspapers/magazines) at least once a week, whether she is currently employed, whether she lives in a nuclear or nonnuclear family, and whether she lives in a rural or urban area. Several of these variables are considered to be relevant because of their linkages to women's empowerment, access to knowledge, and other resources. Education and media exposure have been shown to be a source of empowerment for women, facilitating their ability to gather and assimilate information, to manipulate aspects of their circumstances within a modern world, and to interact effectively with modern institutions (Caldwell 1986; Kishor 2000). Women who are employed are often assumed to be more empowered economically and, by extension, vis-?-vis their male partners, and thus may be less likely to experience domestic violence and/or have greater control over their reproductive health. However, the association of employment with women's empowerment and with the likelihood of experiencing violence does not appear to be uniform across settings (Johnson 2003; Malhotra and Mather 1997). Age is included because several anthropological and empirical studies undertaken in disparate cultures (e.g., Fernandez 1997: India; Johnson 2003: Nicaragua and Haiti; McCluskey 2001: Belize) have found that age is negatively associated with the experience of violence. Some of these studies suggest that as women get older, their social status increases, and they become less vulnerable to acts of domestic violence. Age is also likely to have independent effects on the different reproductive outcomes being examined. For example, the older a woman is, the longer her exposure to the likelihood of getting an STI.

In addition, women's marital status (where marriage includes both formal and informal unions) is controlled with a simple dichotomous variable-currently married/currently not married-in all the analyses except the analysis of STIs/STI symptoms. For the analysis of STIs/STI symptoms, a higher number of sexual partners is an important risk factor. To control for this, the dichotomous marital status variable is replaced by one that groups women into three categories: those who are currently married to their first husband, those who are currently married but to a second or higher-order husband, and those not currently married. (Note that the term husband includes partners in an informal union, as explained in footnote 3.) In this sample of only ever-married women, the distinction between being currently married and formerly married is important, both for the continued exposure to domestic violence and for the evaluation of the risk of each of the health outcomes.

High fertility has been associated with domestic violence both as a potential causal factor (e.g., Martin et al. 1999, who found little support for this contention) and as an outcome of violence (Campbell et al. 1995). Parity is also clearly related to women's reproductive health. Thus, controls for fertility are included in all of the analyses. While the analysis of STIs/STI symptoms controls for the number of children ever born, the analyses of having a non-live birth and having an unwanted birth each control separately for a woman's number of living children and number of children who have died.

Finally, the analysis of STIs/STI symptoms has one additional control, namely a categorical variable that measures whether the woman's husband is frequently, occasionally, or never drunk. This control is included because excessive drinking is known to be positively correlated with risky sexual behavior and is thus a risk factor for STIs.

RESULTS

There is a substantially significant association between domestic violence and the wealth index (Table 1) in each country (p < .05 based on Pearson's chi-square test), but the direction and strength of the association varies between countries. In Cambodia, women in the poorest quintile are more likely than women in the other quintiles to have ever experienced violence. Even in Cambodia, however, the likelihood of spousal violence declines with wealth only from the first (the poorest 20% of the population) to the third quintile and does not vary at all between the top three quintiles. In the Dominican Republic, by contrast, only women in the wealthiest quintile have a significantly lower likelihood of experiencing spousal violence; a similar proportion of women in all the quintiles from the first to the fourth report experiencing spousal violence. Finally, in Haiti, spousal violence and the wealth index have an inverted U-shaped association, with women in the third quintile being the most likely to have experienced spousal violence.

Table 2 shows how each of the three dependent variables-namely, ever experiencing a pregnancy that ended in a non-live birth, having had an STI/STI symptom in the year preceding the survey, and having had an unwanted birth in the five years preceding the survey-varies by women's experience of spousal violence. All these outcomes are more likely to occur if women have ever experienced violence, with the single exception of the "wantedness" of the most recent birth in Haiti.

In Table 3, we present a set of two models for each reproductive health outcome within each country. The top panel of the table contains the odds ratios for additive models, in which we regressed the outcome on spousal violence, the wealth quintiles (with the first quintile as the reference category), and a set of control variables. The first panel shows that the bivariate findings reported in Table 3 hold up when wealth and other factors are held constant-that is, spousal violence is positively and significantly associated with the probability of poor reproductive health outcomes, with the exception of unwanted births in Haiti. The table also shows that wealth has no strong or consistent association with these reproductive health outcomes.

Despite the lack of a significant association between wealth status and women's reproductive health disadvantage, wealth status may have an interactive effect on the observed positive association of experiencing violence and reproductive health disadvantage. Women who are poor and have experienced violence may be particularly disadvantaged with regard to their reproductive health outcomes compared with other women who have experienced violence. Accordingly, an interactive model in which wealth and spousal violence were combined into a single 10-category variable (i.e., spousal violence first quintile, spousal violence second quintile . . . no spousal violence fifth quintile) was run. This analysis (tables not shown) showed that, with the exception of non-live births in Haiti, there were no significant differences (p < .05) in any of the reproductive health outcomes by wealth among women who have never experienced spousal violence. Accordingly, those who did not experience spousal violence were collapsed into one category. We present the results of the interactive model with this six-category variable in the lower panel of Table 3 (the odds ratios for the control variables for this model are shown in Appendix Table B). A comparison of the model chi-square statistic indicates that the additive model fits the data as well as the interactive model for all outcomes in the Dominican Republic and Haiti and for all outcomes in Cambodia except having had an STI/STI symptom in the last 12 months. In fact, in Cambodia, neither the additive model nor the interactive model fit the data well for the STI outcome.

Differing patterns emerge in the three countries. In all three countries, the odds of having a non-live birth among those who have not experienced spousal violence are significantly lower (OR = 0.6-0.7) than those for women who have experienced spousal violence and are poor. As we noted earlier, in models not shown, we found that there are no differences in any country in the probability of a non-live birth by wealth among women who have never experienced spousal violence.

In Cambodia, the association between spousal violence and a non-live birth is attenuated by wealth, as shown by the finding that the wealthiest women (those in the fifth quintile) who have experienced violence are less than half as likely as the poorest women who have experienced violence to experience a non-live birth. In fact, there are no differences between the wealthiest Cambodian women who experienced spousal violence and those who did not experience spousal violence.

In the Dominican Republic and Haiti, by comparison, there is no consistent finding regarding the moderating effect of wealth on the association between spousal violence and a non-live birth. In the Dominican Republic, women who have experienced spousal violence and are in the second quintile appear particularly disadvantaged, while in Haiti, being in the third quintile appears protective among those who have experienced spousal violence.

The second vertical panel of Table 3 presents the results with regard to the risk of STIs. It shows that women who have not experienced violence have significantly lower odds (OR = 0.4-0.6) of having had an STI or STI symptom in the past 12 months. The elevated risk for STIs among women who have experienced violence does not differ by wealth in either Cambodia or the Dominican Republic. In Haiti, however, women who are in the wealthiest quintile and have experienced violence are more than twice as likely as women in the reference category to have an STI/STI symptom. Thus, in Haiti, the risk of an STI is highest for women who have experienced violence and are the wealthiest, not, as expected, women who are the poorest.

Finally, in Cambodia, women who are poor and have experienced violence have higher odds of having an unwanted birth than virtually all other categories of women, as we hypothesized (Table 3, last vertical panel). In the Dominican Republic, wealth does not moderate the higher risk of having an unwanted birth among women who have experienced spousal violence. In Haiti, women who have experienced violence and belong to the middle three wealth quintiles are two to three times as likely to report their last child as unwanted as the poorest women who have experienced violence; the richest women, whether they have experienced violence or not, are no different in their likelihood of reporting an unwanted birth than the poorest women who have experienced violence.

SUMMARY AND CONCLUSIONS

We used DHS data from Cambodia, the Dominican Republic, and Haiti to investigate whether women who experience both poverty and violence are unique in their disadvantage in terms of selected reproductive health outcomes. We defined violence in terms of ever having experienced spousal violence and used a robust and validated wealth index to categorize the households women live in by economic status. The key population of interest consisted of women who have ever experienced spousal violence and belong to the lowest wealth quintile. Our bivariate analyses did not reveal a consistent or monotonically negative association between increasing household wealth and women's experience of domestic violence.

In our multivariate models, we found, first, that the wealth quintile did not have a strong or consistent main association with poor reproductive outcomes in and of itself. Of course, wealth may still moderate the effect of spousal violence, even when it has no main effect. The results show, however, that although the association between spousal violence and poor reproductive health outcomes does vary by wealth, it does so mainly for women who have experienced spousal violence and is rarely according to expectations. For most outcomes, the poorest women who have experienced violence were either no more disadvantaged than women who were not poor and had experienced violence, or were actually somewhat better off than women who were wealthier and had experienced violence. Only in Cambodia does wealth play the moderating effect we hypothesized; but even here, the moderating effect of wealth is neither monotonie nor evident for all outcomes.

Thus, it is the experience of violence per se that disadvantages women in terms of reproductive health. In fact, it appears that wealthier women who have experienced violence are sometimes worse off than those who are poor. When the experience of violence adversely impacts a particular health outcome, it does so whether the woman is poor or not. For these health outcomes at least, the negative effect of having experienced violence extends across the poverty-wealth barrier and is not limited to women living at the crossroads of poverty and violence.