Challenges to measuring violence against women
Department of Gender, Women and Health
Session March 10 at 9 AM
See aldo: Slide presentation
This paper draws heavily on Chapter 6 of "Researching Violence Against Women: Practical Guidelines for Researchers and Advocates" by Ellsberg M and Heise L, a WHO/PATH publication forthcoming in 2005 and on the work of the Gender and Women's Health Department (GWH) in WHO on the measurement of violence against women (VAW).
It is only in the last decade that violence against women was put on the international agenda, largely through the tireless efforts of women's organizations worldwide. In 1993, the International Conference on Human Rights recognized that women's rights were human rights and that violence against women was an abuse of women's human rights. That same year, the UN General Assembly approved a Declaration for the Elimination of Violence against Women (VAW) which clearly identified violence against women as being based on gender inequality and called on governments to recognize it and respond. In 1995 the Fourth World Conference on Women in Beijing dedicated a chapter of its platform of action to violence and in 1996 (1) , the World Health Assembly of the World Health Organization (WHO) passed a Resolution declaring the prevention of violence, including violence against women and children, as a public health problem requiring urgent action.
WHO's own work on violence against women started in
1995. Hardly any data was available on the magnitude of the problem or its
consequences 10 years ago (2).
WHO therefore decided to implement as one of its first endeavours in this
area a multi-country study to look into the prevalence of violence, particularly
intimate partner violence, and its consequences on women's health and lives.
The study also looks at strategies that women use to deal with violence
and at risk and protective factors which contribute to violence. This study,
the WHO multi-country study on women's health and domestic violence (WHO
VAW study for short), provides for the first time population based data
on the prevalence of different forms of violence against women and on the
association of intimate partner violence with a range of health outcomes
intimate partners and ex partners. Violence is now firmly on the international agenda, and countries, governments, NGOs, national statistics offices, women activists, policy makers and service providers worldwide have recognized the need for reliable data on violence in order to create awareness of the problem, convince policy makers of the magnitude of the problem and related issues, as a basis for the development of interventions and to monitor the impact of programs.
It should be stated up front that the methodology that WHO has developed is for a specialized survey, which provides extensive information on the problem, as described above, but is resource intensive, needing important investment in human resources, time and money. However, a short instrument to measure prevalence and key contextual issues has also been developed by WHO from the larger questionnaire. Currently we are working to develop an international consensus around what is a minimum set of questions and a minimum set of conditions that need to be in place to measure violence in a way that provides reliable information.
This paper will set the stage by outlining some of the main challenges for measuring violence against women. These challenges are on: enhancing comparability, disclosure and safety. This paper will use some experiences and lessons learnt by the WHO study as well as other VAW studies to illustrate some of the issues around the measurement of violence (5).
2. HOW TO DEFINE AND MEASURE VIOLENCE
2.1. What do we mean by "violence against women"?
There are many forms of violence against women, yet no universally agreed definition. The United Nations Declaration on the Elimination of Violence against Women of 1993, provides a broad conceptual framework, as follows: Violence against women is….."any act of gender-based violence that results in, or is likely to result in physical, sexual, or psychological harm or suffering to women, including threats of such acts, coercion or arbitrary deprivation of liberty, whether occurring in public or private life" (United Nations, 1993).
Violence against women includes: partner abuse, sexual abuse of girls, rape including marital rape, dowry related violence, female genital mutilation, trafficking in women, forced prostitution, sexual harassment at the workplace and violence condoned or carried out by the state (i.e. rape in war).
Domestic violence or intimate partner violence is a pattern of assaultive and coercive behaviors including physical, sexual and psychological attacks, as well as economic coercion used by adults or adolescents against their current or former intimate partners. Examples of physical abuse include slapping, shaking, beating with fist or object,
strangulation, burning, kicking and threats with knife or gun. Sexual abuse includes coerced sex through threats or intimidation or through physical force, forcing unwanted sexual acts, forcing sex in front of others and forcing sex with others.
Psychological abuse involves isolation from others, excessive jealousy, control over her activities, verbal aggression, intimidation through destruction of property, harassment or stalking, threats of violence and constant belittling and humiliation. Withholding funds, spending family funds, making most financial decisions, not contributing financially to the family and controlling the partner's access to health care, employment or other, are all types of psychological abuse. Using children to control an adult woman can be done through physical and sexual abuse of children, hostage taking of children, custody battles and using children to monitor the partner.
To operationalize violence in quantitative studies, particularly when the aim is to estimate the prevalence of violence in different settings, researchers need to develop clear definitions that can be compared across settings.
2.2. What do we mean by “prevalence of violence against women”?
The prevalence of violence against women is defined as the proportion of women who are abused in a given study population. Researchers face two major challenges in obtaining accurate prevalence data: how to define “abuse” and how to determine the study population. A further complication is that surveys do not measure the actual number of women who have been abused, but rather, the number of women who are willing to disclose abuse. As with all self-reported information, it is always possible that results are biased by either over-reporting or under-reporting.
However, in reality, researchers around the world have found no evidence that abuse is over-reported (6). To be identified as a victim of abuse in most societies is so shameful that few women report abuse when it has not actually occurred. Women are far more likely to deny or minimize experiences of violence due to shame, fear of reprisals, feelings of self-blame, or loyalty to the abuser.
2.3. The Study Population: How to choose whom to interview?
Populations used for research on violence vary greatly across studies. Many studies include all women within a specific age range (frequently 15-49 or over 18), while other studies classify women according to marital status, and interview only women who have been married at some point in their lives, or women who are currently married. The rationale behind such underlying differences in the range of women included rests in the way researchers define the population at risk of abuse. Table 1 gives examples of approaches used in past studies of partner violence.
Table 1. Variation among study populations from recent population-based surveys on partner violence
Sometimes, researchers decide not to interview women below a certain age (usually 18 years) because of specific legal requirements regarding the participation of minors as research informants. In some countries, marriage determines 'adulthood' so that this decision depends on the average age of marriage. Because many prevalence studies are embedded in larger studies that focus on women’s reproductive health—such as the Demographic and Health Surveys (DHS)—only women of reproductive age (15-49 years old) are included, thereby excluding the experiences of older women (7).
Some studies further refine the study population according to their relationship status. For example, studies on partner violence often include only women who are currently married because experience shows that these women are at greatest risk of current partner abuse. In some cases, researchers exclude women in common-law relationships, or those who have been married for less than one or two years (See Table 1). The choice of study population will often be influenced by public norms about male-female relationships or national policies on families, but researchers should be alerted to the fact that any woman who could have ever had an intimate partner, whether they have been married or not, may be exposed to the risk of partner violence. While we recommend using the broadest criteria possible to define the study population of ever-partnered women (the at risk population), it should be recognized that this will be different in different contexts (e.g. in Bangladesh is was not feasible to ask unmarried women about their partners, but then, an unmarried woman in Bangladesh cohabiting with a partner would in most instances would have identified herself as being married and in this way still be included in the study population).
Restricting the study population may bias results for the following reasons:
An example from Nicaragua shows how prevalence estimates for intimate partner violence can vary greatly according to how the study population is defined, and whether the figures include only recent experiences or lifetime experiences of violence. In a study of 488 women aged 15-49 in León, Nicaragua, researchers compared the prevalence of current violence (in the past 12 months) between different subgroups of the sample and found large differences (10). (Table 2)
Table 2. Prevalence of current and lifetime violence, Nicaragua
2.4. How to operationalize various types of violence in a survey?
be said up front that due to the complex and sensitive nature of the of
partner violence researchers should realize that quantitative research will
not give a complete picture, and complementary qualitative approaches are
needed to interpret the data on women’s experiences violence (as also took
place in the countries in the WHO VAW Study). However, since this paper
focuses on measurement of prevalence of VAW qualitative approaches, important
as they are, will not be addressed here.
The way in which violence is defined has an enormous impact on the final results. Thus, it is crucial to establish from the beginning how violence will be defined and who will be considered a “case of abuse,” to borrow a term from epidemiology.
The following are examples of criteria that have been used in studies of intimate partner violence:
For quantitative measurement of violence, the danger in relying exclusively on women’s own definitions of abuse (referred to in social science as an “emic” approach), is that these may vary greatly from one woman to the next, and between cultures, so that it may not be possible to draw meaningful conclusions from the results.
One advantage of using externally derived definitions (an “etic” approach) is that this enables the researcher to make comparisons across different groups of women. The method most commonly used is to ask women whether they have experienced a series of behaviorally specific acts of physical, sexual, or emotional violence, such as hitting, slapping, kicking, or forced sex.
Although intimate partner violence researchers initially focused primarily on physical violence, today emotional and sexual abuse are usually also considered. Many studies also include other kinds of abusive or controlling behaviors, such as limiting decision-making power or mobility, or forms of economic violence.
While it may be useful to identify women who have experienced abuse overall, it is important to present separately the prevalence of each type of violence. Aggregating emotional, sexual, and physical abuse in a single domestic violence figure is likely to lead to confusion, because definitions of emotionally abusive acts vary across cultures, which makes it difficult to define cross-culturally. Furthermore, combining these categories may lead some to question the credibility of the prevalence rates, as emotional abuse is often considered to be less severe than the other types of violence or is more likely to be seen as a 'normal' part of relationships.
The list of abusive behaviors need not be exhaustive. Their purpose is not to describe every possible act that a woman may have experienced. Rather, the aim is to maximize disclosure and to allow for general characterizations regarding the most common types and severity of violence. There is broad agreement now on the types of acts that capture physical violence, less so for sexual violence and emotional abuse. Box 1 show how physical and sexual partner violence was operationalized in the WHO VAW Study (11).
Childhood sexual abuse. Definitional and recall issues complicate research into the prevalence of childhood sexual abuse. Definitions of child sexual abuse in the North American literature have varied along several dimensions; whether non-contact abuse (e.g. exhibitionism) is included together with sexual touching; the maximum age of the victim, the minimum age of the perpetrator, a minimum age difference between victim and perpetrator (generally five years), and whether only incidents experienced as unpleasant or abusive are considered (12). The upper limit on childhood, for example, varies in studies from age 12 to age 18. Some definitions require that the incident be experienced by the victim as abusive; others define all sexual contact between a child and someone significantly older as inherently abusive. Research has shown that differences in definition can greatly affect estimates of prevalence (12).
The challenges of deriving appropriate definitions for child sexual abuse are even more difficult when contemplating cross-cultural research, in which even definitions of childhood can be points of contention and debate. It should be noted that information on child sexual abuse usually comes from asking adults about their experiences during childhood which also introduces important recall issues.
The WHO VAW study chose age 15 as a cutoff point between childhood and adulthood and asked respondents whether before the age of 15 anyone had ever touched them sexually or forced them to do something sexual that they did not want.
Just as it is crucial to be able to distinguish between different types of abuse, researchers also need specific information about the number of perpetrators and their relationship to the victim. Some researchers suggest that providing specific cues about context may be more effective than a single general question in helping a woman remember violent events (13). Therefore, interviewers may be trained to probe about specific situations in which women might have experienced violence, such as the workplace, school, or violence by family members.
Therefore, interviewers may be trained to probe about specific situations in which women might have experienced violence, such as the workplace, school, or violence by family members.
2.6. Time frame and frequency
To fully understand patterns of abuse, researchers need a time frame by which the abuse can be measured. To determine how many women have experienced partner abuse, for each of the acts, it is generally sufficient to obtain information pertaining to the last year, and to lifetime experiences of abuse and whether it happened once, a few times or many times.
When the aim of the study is to examine experiences of physical or sexual abuse in childhood, it is important to ask the respondent's age when the abuse started, as well as the age of the perpetrator. Did it happen one time? A few or many times? How many years did the abuse last?
3. ENHANCING DISCLOSURE OF VIOLENCE
Experience from international research nonetheless indicates that some methods are more effective than others in encouraging women to talk about violence. The following series of issues may affect women’s willingness to discuss experiences of violence.
3.1 How are women asked about violence?
The methods used to ask women about violence may influence how comfortable they are disclosing abuse. Studies in industrialized countries have found that for the purpose of identifying intimate partner abuse, either face-to-face interviews or interviews by telephone give better results than self-administered questionnaires (14;15).
On the other hand, anonymous techniques frequently encourage greater disclosure of childhood sexual abuse. The WHO Multi-country Study tested the use of different methods for eliciting disclosure of child sexual abuse. Women were asked during a face-to-face interview whether they had ever been touched sexually or made to do something sexual against their will before the age of 15. At the end of the interview, women were asked to mark on a separate piece of paper whether they had been sexually abused as a child by placing a check next to either a happy or sad face, regardless of what they had chosen to reveal during the face-to-face interview (Figure 1). Women were assured that as their name was not on the paper, that their answer would never be traced back to them. To further preserve the anonymity of the respondents these papers were placed together in a large plastic bag. In most countries, more women disclosed violence using this method than they did in the direct question. This card method worked well in all settings but one - Bangladesh. Women especially in rural settings would get confused and would often call the husband for assistance or permission as they were not used to putting things down on paper.
To explore patterns of disclosing childhood sexual abuse further, more recently the WHO VAW study developed a way to link the anonymous question to the questionnaire, by having the woman put the marked paper in a blank envelope that she would sealed and that would be attached to the questionnaire.. Results for Tanzania capital were as follows: by interview 4%, anonymous 11%,and using both methods combined 12.2%. While we hypothesized that the anonymous reports would be "closer to the true prevalence" and thus include at least all those who in the interview had disclosed childhood sexual abuse plus a certain percentage that had not disclosed during the interview, we noted that among those who reported childhood sexual abuse during the interview there were women who would not disclose anonymously and vice versa. Other studies have also found this and this is explained by the fact that women may have different reasons for disclosing verbally and not anonymously (for example because of fear of putting something on paper, as in Bangladesh) and vice versa.
3.2 Who is asking?
As in all research on sensitive topics, disclosure rates are affected by the skill of the interviewer, and her or his ability to establish rapport with the informant. Women are more likely to be willing to share intimate and potentially painful or embarrassing aspects of their lives when they perceive the interviewer as empathetic, non-judgmental, and genuinely interested in their situation. It is generally believed that female interviewers are more successful in eliciting personal information from women, although this has rarely been tested. In some settings, difficulties have been encountered when using young or unmarried women as interviewers, or when using interviewers who lack experience discussing sensitive issues. This highlights the importance of using carefully selected and appropriately trained female interviewers.
3.3 How many times should you ask a woman about violence?
Numerous studies have shown the importance of giving women more than one opportunity to disclose violence during an interview. Women may not feel comfortable talking about something so intimate the first time it is mentioned, or they may not recall incidents that took place long ago. This is why studies that include only one or two questions on violence usually result in substantial under-reporting of abuse. Researchers have found that many women initially deny having experienced violence, but over the course of the interview, overcome their reluctance to talk. For this reason, it is also wise to avoid using “gateway” or "filter" questions, where women who reply negatively to the first violence question are not asked the more specific questions in the survey.
Example from the field: A Demographic and Health Survey carried out in Nicaragua used two sets of questions to identify partner abuse. One question asked in general, “Have you ever been physically beaten or mistreated by anyone?” Women who responded affirmatively were questioned about the perpetrator. The next set of questions referred to specific acts such as pushing, slapping, choking, beating, and forced sex. For each act, women were asked whether their partner had carried out the act within the last year, or at any time during their marriage. While 14 percent of women reported partner abuse using the first set of more general questions, 29 percent of women reported acts of physical or sexual partner abuse in the more specific set of questions (16).
3.4 The context of the interview
The overall framework of the survey and the items immediately preceding questions on abuse can also affect how women interpret and respond to violence-related questions. For example, embedding questions on physical assault immediately following items on relationships will cue respondents to the issue of partner abuse, whereas asking a similarly worded question after items on crime victimization will tend to cue respondents towards assaults perpetrated by strangers.
The issue of context is particularly relevant in large-scale surveys in which abuse-related questions are integrated into questionnaires designed for other purposes. When questions on violence immediately follow lengthy discussions on other topics such as family planning, nutrition, or childhood illness, a woman may be less likely to disclose experiences of violence. In this case, it is particularly important to give the respondent a chance to "switch gears." An introductory paragraph can make it clear that the interview is shifting to a completely different subject.
Questions on violence may themselves be framed in a variety of ways that convey different messages to the women being interviewed. For example, one well-known instrument, the Conflict Tactics Scale, presents the use of violence as a way of resolving conflict and includes a series of questions about non-violent ways to resolve conflicts, such as “discussing the issue calmly” or “sulking” (17;18)
This lead-in could be problematic in a country where physical assault is not understood as a way to resolve conflict, but rather as a form of punishment or discipline. Other instruments describe acts that the partner “does when he is mad.”
Even the timing of the violence questions within the overall interview may affect how women respond, particularly when the questions are part of a much larger study. Asking about violence too early may not provide interviewers enough time to build rapport with the informant. On the other hand, if questions are placed at the end of a long interview, both the interviewers and the respondents may be tired or anxious to finish the interview and therefore less likely to probe into experiences of violence.
One important way to ensure confidentiality is to interview only one woman per household. When the study is exclusively focused on violence, this is relatively easy to achieve. The situation is more complex when other members of the household are to be interviewed on other topics, such as reproductive health. When incorporating a module on violence in the Nicaraguan DHS, researchers selected only one woman per household to answer the violence portion of the survey, presented as a section on “Household Relations.” Interviewers were instructed to inform her that she was selected by chance to be asked these questions, and that no one else in the household would know she was being asked.
By the same token, both ethical and methodological principles suggest that it is better to avoid interviewing women and men from the same household on violence. It is true that questioning both partners of a couple would enable comparisons between wives and husbands, as well as direct information about live events preceding abusive behavior by men. However, this method may place a woman at risk if her abusive husband suspects that she has been talking about his behavior. In anticipation of his reaction, she may be reluctant to disclose violence. Therefore, we recommend that surveys not include women and men from the same household. If it is necessary to interview husbands for some other reason, do not ask about violence, and let the woman know that she is the only one being asked about abuse.
4. ETHICAL AND SAFETY ISSUES
Research on VAW raises important ethical challenges. WHO has developed “Putting Women First: Ethical and Safety Recommendations for Research on Domestic Violence Against Women,” which were used for the WHO Study. WHO considers that adhering to these guidelines is are essential for doing research in this field and for ensuring the quality of the data (19).These guidelines lay out some of the key principles that should guide research on domestic violence, such as ensuring absolute privacy when doing the interview and maintaining absolute confidentiality of information provided by respondents. It also identifies specific actions needed to promote each of the principles identified. The principles are summarized in Box 2. These guidelines are now being used widely for research on vaw and in other fields.
5. SELECTION AND TRAINING OF INTERVIEWERS
International research indicates that women’s willingness to disclose violence is influenced by a variety of characteristics such as the sex, age, marital status, attitudes, and interpersonal skills of interviewers. The WHO VAW study used female interviewers and supervisors, and their careful selection and appropriate training was of paramount importance (20). A standardized three-week training for interviewers was developed. The training included sensitizing interviewers to gender issues and to the dynamics of VAW, as well as familiarizing them with proper interview techniques, the questionnaire, and the field procedures they would be using. Interviewers were trained to respond sensitively to women who disclosed violence, but not to assume the role of a counselor or to raise respondents’ expectations unrealistically about what the study could do for them.
Evidence of the Value of Training
Additional evidence of the value of the WHO VAW Study training approach comes from Serbia and Montenegro, where the survey was conducted, between March and May 2003. With technical support provided by WHO, thirteen previously inexperienced interviewers were fully trained over 2.5 weeks. However, because of pressure to finish the fieldwork, six weeks after the start of the fieldwork, an additional group of twenty-one professional interviewers from a survey company were recruited to assist with the interviews. This new batch of experienced interviewers received less than a day’s training, which included orientation on gender and violence issues and a brief review of the questionnaire and field procedures.
The previously inexperienced but carefully selected and trained interviewers
obtained a significantly higher response rate (93% vs. 86%; p<.0001) and
a significantly higher disclosure rate (26% vs. 21%; p<.05) for physical
and/or sexual partner violence than the ‘professional’ interviewers. Respondent
satisfaction at the end of the interview was significantly higher for women,
both with (46% vs. 29%; p<.01) and without violence (46% vs. 38%; p<.05),
interviewed by the trained interviewers.20 These findings highlight the
degree to which interviewer selection and training affect levels of participation,
disclosure and satisfaction with the interview, and illustrate that it is
not advisable to assume less training is needed when using professional
6. SPECIALIZED SURVEYS VS. USE OF A SHORT MODULE
Two distinct research trends are emerging as more international data on violence against women become available. First, large-scale surveys primarily designed for other purposes increasingly solicit information on violence. For example, several Demographic and Health Surveys (DHS) and Reproductive Health Surveys conducted by they Centers for Disease Control (CDC) have included questions on violence in national surveys (7;21). Although many of these surveys use one or two aggregate "gateway" questions to measure any type of violence, such as, "Have you ever been beaten by anyone since you were 15/were married? By whom?", some of the more recent DHS studies include a module on domestic violence with questions that go into more detail..
The second trend is represented by studies that are primarily designed to gather detailed information on women’s experiences of violence. Many of these studies, such as the prevalence studies in Nicaragua, South Africa, and the WHO Multi-country Study, have relatively smaller sample sizes and cover a limited geographical region, while others such as the National Surveys on Violence against Women in Canada (1993), (22) the United States (1997), (23) Sweden, (24) and Finland (1997)(25) collect national data. These specialized studies tend to gather much more information about different types of violence and perpetrators, as well as information on circumstances and women's responses to violence. They also tend to devote more attention to the interaction between interviewers and respondents and issues of safety.
There are potential advantages to including violence questions in national surveys designed primarily for other purposes. For example,
There are also drawbacks to this strategy: In general, prevalence estimates have been higher in the more focused studies than in the national surveys designed primarily for other purposes (26). One explanation suggested is that because the focused studies are more likely to use methods for enhancing disclosure, such as repeated questioning, they are able to produce more accurate prevalence estimates (16). They are also more likely to adhere to ethical and safety standards which contribute to enhancing disclosure and data quality.
Thus, one tradeoff of using multi-purpose surveys to produce prevalence estimates on violence is that violence can be significantly under-reported. Such under-reporting can dilute associations between potential risk factors and health outcomes, leading to results that are falsely negative. Underestimating the dimensions of violence could also prevent violence intervention programs from receiving the priority they deserve in the allocation of resources. Finally, because many of these studies have not systematically addressed safety concerns, women who participate in them may face increased risk of retaliation or other harm.
In many instances, where official data is lacking, and where resources are not available for a specialized survey, to create awareness and to influence policy a short module in a survey designed for other purposes could be sufficient initially. However, in order to ensure the validity of the results the utmost care should be given to enhancing disclosure. Furthermore, in this case WHO strongly recommends that ethical and safety guidelines are adhered to as closely as possible (see Box 3).
1. Beijing Declaration and Platform for Action, adopted by the 4th World
Conference on Women, Beijing 4-15 Sept. 1995, UN Doc. ACONF.177/20 Rev.
1 (96.IV.13), para. 129a.
14. Johnson J, Sacco V. Researching violence against women: Statistics Canada's
national survey. Canadian Journal of Criminology 1995;37:281-304.
17. Straus MA, Gelles RJ. Measuring intrafamily conflict and violence: The
Conflict Tactics (CT) Scale. Journal of Marriage and the Family 1979;41:75-88.
19. World Health Organization Putting women first: Ethical and Safety Recommendations
for Research on Domestic Violence against Women WHO/FCH/GWH/01.1(Geneva:
World Health Organization, 2001)
Edited by Aldo Campana,