Measuring Attitudes About Intimate Partner Violence Against Women: The ATT-IPV Scale

Abstract

In lower-income settings, women more often than men justify intimate partner violence (IPV). Yet, the role of measurement invariance across gender is unstudied. We developed the ATT-IPV scale to measure attitudes about physical violence against wives in 1,055 married men and women ages 18–50 in My Hao district, Vietnam. Across 10 items about transgressions of the wife, women more often than men agreed that a man had good reason to hit his wife (3 % to 92 %; 0 % to 67 %). In random split-half samples, one-factor exploratory factor analysis (EFA) (N₁ = 527) and confirmatory factor analysis (CFA) (N₂ = 528) models for nine items with sufficient variability had significant loadings (0.575–0.883; 0.502–0.897) and good fit (RMSEA = 0.068, 0.048; CFI = 0.951, 0.978, TLI = 0.935, 0.970). Three items had significant uniform differential item functioning (DIF) by gender, and adjustment for DIF revealed that measurement noninvariance was partially masking men’s lower propensity than women to justify IPV. A CFA model for the six items without DIF had excellent fit (RMSEA = 0.019, CFI = 0.994, TLI = 0.991) and an attitudinal gender gap similar to the DIF-adjusted nine-item model, suggesting that the six-item scale reliably measures attitudes about IPV across gender. Researchers should validate the scale in urban Vietnam and elsewhere and decompose DIF-adjusted gender attitudinal gaps.

Introduction

Intimate partner violence (IPV) refers to psychological, physical, or sexual harm or the threat of such harm by a former or current dating partner or spouse (Saltzman et al. 2002). In response to high estimates of IPV against women, feminists have advocated for legal change (Weldon 2002). By 2006, 109 countries had or were drafting laws prohibiting violence against women (United Nations General Assembly 2006), which offer new cultural meanings about gender, power, and violence. Still, in lower-income settings, women more often than men justify IPV against women (Uthman et al. 2010). These gender gaps are hard to interpret because of variability in the attitudinal measures used and limited research on measurement invariance across gender. We present the ATT-IPV (Attitudes about IPV) scale to measure the attitudes of women and men about physical IPV against a wife. Items from two national surveys in Vietnam were adapted and administered in a probability sample of 1,055 married men and women ages 18–50 in My Hao district. The initial scale included agree/disagree items about whether a man had good reason to hit his wife for any of 10 behaviors meant to reflect varying degrees of gender transgression (Kishor and Johnson 2004). This article describes the scale’s development, its measurement properties overall and by gender, and promise for use in surveys, program evaluations, and attitudinal studies of women and men.

Background

Global Prevalence of IPV Against Women, Consequences, and Legal Response

Worldwide, 15 % to 71 % of women report lifetime physical IPV (Garcia-Moreno et al. 2006) and often report co-occurring psychological and sexual IPV (Garcia-Moreno et al. 2006). Such exposures have adverse effects on behavioral, nutritional, physical, and psychological outcomes in women and children (e.g., Nguyen et al. 2009; Yount et al. 2011a).

The high levels and adverse effects of violence against women, including IPV, have spurred global dialog and legislation (Weldon 2002; Yount et al. 2014). In the 1990s, several international conferences urged states to outlaw IPV and to see its prevention as a public health priority (Garcia-Moreno and Stöckl 2009). Remarkably, although no democratic state (of 36) in 1974 had taken any action on violence against women, 31 of 36 had taken some action by 1994, most often to address IPV (Weldon 2002). By 2006, 109 of 211 states had or were drafting laws against IPV (United Nations General Assembly 2006).

Measuring Attitudes About IPV Against Women Globally

Legislation on IPV signals its political salience (Weldon 2002) and offers new meanings about gender and power (Merry 1995; Yount et al. 2014). Naming IPV in the penal code redefines it as a criminal act rather than normal “discipline.” Thus, where laws prioritize gender equality, men and women should be less tolerant of IPV against women. In a U.S. city after passage of the Violence Crime Control and Law Enforcement Act (1994), public justification of IPV declined and criminalization of specific acts increased (Johnson and Sigler 2000).

In lower-income countries, attitudinal data on IPV against women began to accumulate in the 1990s, when the Demographic and Health Surveys (DHS) began to ask whether wife beating was justified for behaviors reflecting a continuum of gender transgressions (Kishor and Johnson 2004). Since 1995, 88 DHS in 53 lower-income countries have asked these questions of adult women, and rates of any justification of wife beating range from 4 % to 90 % (Yount et al. 2011b).

Comparability of Attitudinal Questions About IPV Against Women

Variations in women’s reported attitudes about IPV against women arise in part from the broader context of gender equity/inequity in familial, social, economic, and political institutions (Yodanis 2004; Yount et al. 2011b). Such variation also may arise from noncomparability of the scale across surveys. Of 67 DHS conducted in 48 countries between 1995 and 2007, 40 DHS included the following attitudinal items about IPV against women (Yount et al. 2011b):

Variations in this question across the 67 DHS pertained to wording: in the preamble (21 %), to depict the husband’s aggression (19 %),¹ to evaluate the husband’s aggression (28 %),² and to depict the wife’s transgressions (36 %). In 24 % of these DHS, other items (transgressions of the wife) were added to the five in the preceding extract. All else being equal, using an alternate preamble and adding items multiplied the expected prevalence of “justifying” IPV against a wife by 1.49 and 1.32, respectively; using an alternate wording for the husband’s aggression and for both his aggression and its evaluation multiplied the expected prevalence of “justifying” such treatment by 0.66 and 0.44, respectively (Yount et al. 2011b). Noncomparability in these aspects of the scale accounted for 8 % of the unexplained variance in women’s attitudes about IPV against women (Yount et al. 2011b).

In wealthier settings, experimental variations in the ordering, wording, and response options of attitudinal questions have altered the distribution of participants’ responses (e.g., Holleman 1999; Schuman and Presser 1996; Tourangeau et al. 2003). In an experiment of the 2007 Bangladesh DHS attitudinal questions on wife beating, most women respondents in six villages initially gave indeterminate responses to all items (Yount et al. 2012), and depicting the wife’s transgressions as unintended yielded lower rates of justifying IPV than depicting them as willful (1 % to 8 % vs. 38 % to 68 %, respectively) (Yount et al. 2012).

Variants of the most common DHS attitudinal question also have been added to national surveys in Vietnam, which may confound estimates of attitudinal change over time and preclude clear estimates for attitudes in a given year. For example, the 2006 and 2010–2011 Multiple Indicator Cluster Surveys (MICS)—multipurpose surveys supported by the United Nations Children’s Fund to monitor the situation of women and children—added the following questions:

In 2006, 63.8 % of women ages 15–49 responded that beating a wife is acceptable for at least one listed transgression (General Statistics Office GSO 2006). In 2010–2011, 35.8 % of same-aged women found good reason for a husband to hit or beat his wife (General Statistics Office GSO 2011). Less favorable attitudes about IPV against women in 2010–2011 may reflect real attitudinal change after passing the 2007 law on domestic violence (National Assembly, Government of the Socialist Republic of Vietnam (NAGSRV) 2007) or response effects to alterations of the question (e.g., preamble, wording to depict the husband’s aggression, wording to judge his aggression). In 2010, the United Nations–Government of Viet Nam Joint Programme on Gender Equality undertook the National Study on Domestic Violence Against Women in Viet Nam (NSDVAWV), the first such study in Vietnam. In contrast to estimates from the 2010–2011 MICS, 41.3 % of women in the NSDVAWV found enough reason for a husband to mistreat/beat his wife in at least one of six situations (General Statistics Office GSO 2010):

Such variations in this attitudinal question expose a need for standard measures to capture men’s and women’s attitudes about IPV for comparison across countries and over time.

Measurement Invariance to Attitudinal Questions About IPV Against Women

Using such a scale to compare attitudes across gender also assumes that its measurement properties are met for women and men (Yount et al. 2014). From the perspective of social learning theory, attitudes develop from gendered life course experiences as well as from the gendered influences of family, peers, social norms, and institutions (Bandura 1973; Wekerle and Wolfe 1999). Thus, women and men should reflect back many of these norms in attitudinal surveys, such that observed gender differences in attitudes reflect true variation. Yet, observed differences also may reflect nonequivalent measurement properties across gender.

Statistical item bias, or differential item functioning (DIF), is one conceptualization of the nonequivalence of measurement scales (Yount et al. 2014) and refers to the distinct measurement properties of a scale item for different groups, after accounting for overall differences between the groups on the construct being measured (Holland and Wainer 1993). DIF exists if individuals from two groups with equivalent levels of an underlying construct have different probabilities of selecting specific response categories for an item intended to measure the construct (Mellenbergh 1989). In our case, women and men may understand attitudinal items differently or be variously motivated to select certain response categories, such that women and men who share a similar view respond differently to the same attitudinal item (Yount et al. 2014).

In wealthier countries, the nonequivalence of measurement across gender has been observed with nonattitudinal scales (e.g., Cauffman and MacIntosh 2006; Fletcher and Hattie 2005; Gelin and Zumbo 2003) and attitudinal scales about teen dating violence (Edelen et al. 2009). In lower-income countries, researchers have reported gender gaps in attitudes about IPV against women without assessing the scale for DIF by gender (e.g., Uthman et al. 2010). Qualitative research in lower-income settings has shown gender-discrepant interpretations of and responses to attitudinal questions about IPV against women (Yount et al. 2012). Although some men justified their own perpetration, other men seemingly gave the socially desirable response of “not justified,” and women often justified such violence because they saw it as inevitable, irrespective of their own views. Ignoring imbalances in DIF across scale items results in attitudinal scores with measurement bias (Reise et al. 1993), confounding interpretations of observed group differences. Identifying the sources and extent of measurement noninvariance can improve measurement by removing bias and clarifying how groups interpret or respond to items differently (Edelen et al. 2009; Yount et al. 2014).

Here, we describe the development of the ATT-IPV scale to measure women’s and men’s attitudes about physical IPV against a wife. Vietnam was a germane study site, having a high reported lifetime prevalence of physical IPV against women (31.5 %) (General Statistics Office GSO 2010), recent legislation to mitigate such violence (National and Government of the Socialist Republic of Vietnam NAGSRV 2007), variation in prior scales used to measure attitudes about physical IPV against wives (General Statistics Office GSO 2006, 2010, 2011), and evidence that women more often than men justify such violence (Gender and Community Development Network (GENCOMNET) 2011).

Method

Vietnam Setting

Rapid economic and legal change in Vietnam has created tensions in gender norms and relations (Yount et al. 2014). In 1986, the state initiated a policy to maintain socialism under an expanded market economy, with mixed results for the gendered structuring of work (Khuat et al. 2010). The labor migration of some women prompted their husbands to assume unpaid family work (Lan and Yeoh 2011), although most women still do most of this work (Knodel et al. 2005; Thanh et al. 2012; World Bank 2011). Women’s high rates of market work in 1990 (81.3 %) continued in 2010 (to 78.1 %) (World Bank 2013), but women more often than men are unskilled workers (42.9 % vs. 36.2 %) and vulnerably employed (69.0 % vs. 54.4 %) (World Bank 2011). The occupational structure still is gender segregated (Thanh et al. 2012; World Bank 2011).

Vietnam also has adopted a legal framework founded on the principle of gender equality and has created institutions to support women’s progress. In the 1980s, the government outlawed all physical violence against women and children, but enforcement was uneven (Rydstrøm 2003). In 1995, the state endorsed the National Plan of Action for the Advancement of Vietnamese Women, and the National Committee for the Advancement of Women in Vietnam (established in 1993) strengthened its structures and extended its governmental and national networks (Khiet 2000). In 2007, the National Assembly adopted a new law on domestic violence, which specified acts of IPV and strategies for prevention and intervention (National and Government of the Socialist Republic of Vietnam NAGSRV 2007).

Still, inequitable gender norms and relations persist (Nguyen et al. 2011; Thanh et al. 2012; Werner 2009). Confucian principles of the patrilineal hierarchy still define men as inside lineage and women as outside lineage (Horton and Rydstrøm 2011; Khuat et al. 2010). Thus, men remain the symbolic and practical pillars of the house, maintaining authority over major patrilineal rituals and decisions (Khuat et al. 2010; Phan 2008; Thanh et al. 2012; Werner 2009). Women should adjust to changing social situations (Khuat et al. 2010; Nguyen 2012; Nguyen et al. 2011; Phan 2008), and perceived failure to do so may incite blame (Nguyen 2012; Thanh et al. 2012) and “warranted” punishment (Khuat et al. 2010; Nguyen 2012; Phan 2008). Taoist ideas about men’s and women’s bodies associate men with heat (nong) and women with coolness (lanh), which may legitimize men’s violence and pressure women to endure it (Horton and Rydstrøm 2011). The Women’s Union, which promotes women’s progress in public life, urges respect for patriarchal hierarchy and familial harmony (Schuler et al. 2006), which may extend to tolerating IPV.

Tensions in gender norms more generally are reflected in national surveys. In 2001, a majority of men (58.0 %) and women (54.6 %) agreed that men make better political leaders (World Values Survey Association 2013), and substantial minorities of men (22.1 %) and especially women (26.2 %) agreed that university education was more important for boys than girls. Moreover, high percentages of men (83.8 %) and women (88.2 %) agreed that being a housewife was as fulfilling as working for pay. Despite these more customary views, the vast majority of men (96.5 %) and women (97.6 %) agreed that the husband and wife should contribute to family income. These figures, and those from a 2003 survey of three marriage cohorts (1963–1971, 1977–1985, 1992–2000) in the Red River Delta and Hanoi (Knodel et al. 2005), suggest a norm of femininity in Vietnam in which women should be mothers, housewives, and family earners, while men should retain public (and private) power in decision-making.

Thus, on the one hand, marked social and legal change in Vietnam may lower public tolerance for IPV against women. On the other hand, persistent familial inequities are reflected in ongoing IPV against women (Hoang et al. 2002). About one-third of women in rural Northern Vietnam (32.7 %) and nationally (34.4 %) report lifetime exposure to physical or sexual IPV (General Statistics Office GSO 2010; Nguyen et al. 2008), exceeding levels elsewhere in Southeast Asia (Yount and Carrera 2006). Where inequitable gender norms and practices persist, women still may blame themselves for IPV (Horton and Rydstrøm 2011; Nguyen 2012; Thanh et al. 2012;). Indeed, 41.3 % of women nationally justify IPV against women (General Statistics Office GSO 2010). With ideas about gender and IPV in flux, comparable and psychometrically sound measures are needed to capture women’s and men’s attitudes about IPV against wives.

Study Site

The study site was 12 communes and one district town of My Hao district in Hung Yen province, 30 km from Hanoi and home to 97,733 residents, almost all of Kinh ethnicity. My Hao is typical of the Red River Delta region, with a history of rice cultivation and customary local products. Livelihoods, however, have changed markedly in the last 10 years. Although agriculture remains popular, new industrial investments in the early 2000s have spurred the construction of factories for domestic and international companies, creating jobs for thousands of residents, including women. Moreover, the three highways traversing the district connect it with the provincial towns and other provinces in the North. These changes make My Hao an ideal place to study gender issues, including IPV against women, because the district reflects changes that are challenging Vietnam more generally.

Following patrilocal patterns of residence, men more often than women live in their commune of birth (95 % vs. 58 %) and in the same commune as their natal family (95 % vs. 60 %) (Yount et al. 2014). Joint household residence also is common (40 %) (Yount et al. 2014). Relatively few residents (7 %) live in poor households, which are below the official poverty line for rural areas (VND 400,000 per person per month) and eligible for certain state benefits (World Bank 2012; Yount et al. 2014). Schooling attainments are high for women and men (9.5 grades), and almost all women (97 %) and men (98 %) work for money (Yount et al. 2014). Residents often engage in multiple types of work, including farming (64 %), work in local factories (30 %), and self-employment in small enterprises (23 %) (Yount et al. 2014). In addition to their market work, women perform most of the domestic labor. Relatively more women (50 %) than men (23 %) attend a group or organization at least once per year (Yount et al. 2014). Mass unions, such as the Women’s Union and Youth Union, are located in the communes, and reconciliation groups are present to resolve conflicts. Local People’s Committees govern the communes, and the Communist Party ensures the residents’ ideological position. More than one-fourth of women (29 %) report lifetime exposure to physical IPV, and a similar percentage of men (28 %) report ever perpetrating physical IPV (Yount et al. 2014). Nearly equivalent numbers of women and men report witnessing IPV as a child (26 % and 27 %, respectively), but men (72 %) more often than women (50 %) report being hit or beaten as a child by a parent or other adult relative (Higgins et al. 2013; Yount et al. 2013).

Sample Survey

The sample was drawn from a household census of the study site (Yount et al. 2014). Married women and men ages 18–50 were eligible.⁶ Villages were paired by the size of the eligible married population, starting with the two largest villages, and were randomly assigned within pairs to the women’s and men’s samples to mitigate the risk of breached confidentiality and escalated violence against women who disclose IPV (World Health Organization (WHO) 2001). The smallest village was omitted because it was not paired and had too few married persons (n=36) given the sample design. Twenty village pairs were selected with probability proportional to the size of the married population in each pair relative to the total married population in all 74 villages. In selected villages located in 12 communes, 27 households with at least one eligible participant were randomly selected. In households with multiple eligible participants, one was selected at random to ensure privacy. With an expected response rate of 93.0 % and aiming for 500 interviews each with men and women (1,000 total), we selected 540 men and 540 women for interviews. Of these, 1,069 persons were located, and 1,055 were interviewed (533 women, 522 men), yielding a high final response rate (98.7 %) that was similar for women (99.3 %) and men (98.1 %) and across study villages (92.6 % to 100 %).

The questionnaire included, in sequence, modules on sociodemographic and economic background; a module on attitudes about physical IPV against women and women’s recourse-seeking after IPV; and a module on perpetration and exposure to IPV, exposure to violence in childhood, and knowledge of laws about IPV and women’s rights. The Institutional Review Boards of Emory University and the Center for Creative Initiatives in Health and Population approved the study. After obtaining verbal informed consent, all interviews were held in separate rooms at the commune health station to ensure privacy and confidentiality (World Health Organization WHO 2001). Survey participants and interviewers were gender-matched to enhance disclosure (World Health Organization WHO 2001).

Analysis

Data and Item Construction

The data for analysis come from agree/disagree responses about whether a husband has good reason to hit his wife for any of 10 behaviors meant to capture a continuum of gender-transgressions, such as she does not complete the housework to his satisfaction and he finds out that she has been unfaithful (Table 1).⁷ The wording of four items (1, 2, 4, and 5) of the 10 came from the 2010 NSDVAWV. The wording of four other items (6, 7, 8, and 9) came from the 2010–2011 MICS in Vietnam. The wording of item 3 came from both surveys, and one item (10) was added based on qualitative research described elsewhere (Yount et al. 2014). To standardize the question, we relied on insights from formative qualitative work and prior experience of Vietnamese team members working on the NSCVAWV. Namely, the preamble in the 2010–2011 MICS version of the question was removed (Sometimes a husband . . .), and the question wording from the 2010–2011 MICS was retained (In your opinion, does a man have a good reason to hit his wife in the following situations). Relative frequencies of all 10 items were estimated overall and by gender in random split-half samples (described later) to assess their completeness and distributions. One item with a low relative frequency (item 7; see Table 1) was not analyzed further. Because of the binary response options for each item, tetrachoric correlations were estimated in each random split-half sample to assess the level of bivariate association between any two items (Bandalos and Finney 2010). These correlation matrices were the basis for exploratory and confirmatory factor analyses.

Factor Analyses and Structural Model of DIF by Gender

Exploratory factor analysis (EFA) is a preferred method used to identify the factor structure for a set of items with minimal prior psychometric testing (Bandalos and Finney 2010; Yount et al. 2014). In EFA, items are not constrained to load on specific factors, so the researcher may identify the factor structure for the set of items. For sufficiently large samples, confirmatory factor analysis (CFA) can be estimated with an independent subsample to test the factor structure determined by the EFA (Bandalos and Finney 2010). Our total sample size exceeded guidelines for random split-half-sample analyses (Bandalos and Finney 2010), so we performed an EFA and subsequent CFA with random subsamples. Excluding one participant from the EFA subsample with missing data for all 10 attitudinal items yielded final subsamples of N₁ = 527 for the EFA and N₂ = 528 for the CFA. These subsamples were similar on all observed attributes.⁸

We used EFA to assess the scale’s unidimensionality, factor loadings for the nine remaining attitudinal items, and fit indices for a one-factor model (root mean square error of approximation (RMSEA), comparative fit index (CFI), and Tucker-Lewis index (TLI)) (Muthén and Muthén 1998–2012). We planned to retain items that were strongly and positively related to the underlying attitudinal construct and to remove items sequentially that were weakly related to the construct (a loading <0.3). Because of strongly positive loadings, adequate fit indices (RMSEA close to 0.06 or less; CFI close to 0.95 or greater; TLI close to 0.95 or greater) (Brown 2006; Harrington 2008), and theoretical interpretation, we retained all nine items.

Using the other random split-half sample (N₂ = 528), we performed CFA to test the factor structure and measurement invariance of the final one-factor, nine-item EFA model. After assessing the fit indices and factor loadings of the estimated CFA model, we tested the scale for uniform DIF by gender. We used a single-group multiple indicator multiple cause (MIMIC) structural equation model,⁹ adding to the CFA model a covariate for gender (with women as the reference group) to test for the invariance of indicator intercepts and factor means. We then examined modification indices (or estimated improvements in model fit) for allowing the direct effects of gender on the attitudinal items to be estimated freely. We added a single direct effect of gender on the item with the largest modification index and retained this effect if it was statistically significant (p ≤ .05) and if its inclusion improved model fit (p ≤ .05, chi-square test for difference). These iterations proceeded until adding direct effects of gender on single attitudinal items no longer improved model fit. We identified significant direct effects of gender on three items (1, 2, 10; see Table 1). To minimize DIF by gender in the attitudinal scale, we removed these three items; reestimated a CFA with the remaining six items, first without and then with an indirect effect of gender; and assessed the fit statistics for a final six-item one-factor model. All models were estimated with Mplus7 (Muthén and Muthén 1998–2012) using an estimation approach (mean and variance-adjusted weighted least squares (WLSMV)) suitable for models with binary data and accounting for the complex sample design.

Using single-group MIMIC models allows the researcher to identify DIF, to adjust for it in a final model, or to remove the items that exhibit DIF, and to generate DIF-adjusted or DIF-free factor scores. Yet, the effect of such strategies at the score level may be too modest to warrant adjustment for DIF with this more complex scoring approach (Edelen et al. 2009; Yount et al. 2014). To assess whether observed DIF altered inferences about gender differences in attitudes about IPV against women, we computed five scores for the ATT-IPV scale and evaluated Wilcoxon-Mann Whitney tests comparing men and women on each score (Edelen et al. 2009; Yount et al. 2014). Attitudinal scores were based on (1) the sum of all nine 0/1 (no/yes) items in the original measure (summative 9); (2) factor scores from the nine-item one-factor model that ignored identified DIF (factor 9, no DIF); (3) factor scores from the nine-item one-factor model adjusted for identified DIF (factor 9, DIF); (4) factor scores from the six-item one-factor CFA model in which items with identified DIF were removed (factor 6); and (5) the sum of the six 0/1 (no/yes) items that did not exhibit DIF (summative 6). Any differences in the inferences for gender gaps in attitudes based on the nine-item summative scale and nine-item factor score without DIF would reflect the influence of factor analysis alone. Any differences in the inferences based on the nine-item DIF-unadjusted and DIF-adjusted factor scores would reflect the influence of adjusting for DIF. Finally, we expected that the factor mean difference in attitudes by gender estimated from the six-item one-factor model would be similar to that estimated from the nine-item, DIF-adjusted one-factor model.

Results

“A Husband has Good Reason to Hit His Wife If . . .”: Frequency Distributions by Gender

Overall, the distributions of attitudes about IPV against a wife and gender gaps in these attitudes were comparable across random split-half subsamples (Table 1; p values available on request), so we discuss the distributions for the EFA sample. In general, the relative frequencies of finding good reason to hit a wife varied widely by the type of transgression. At one extreme, participants rarely agreed that a husband had good reason to hit his wife if she burns the food (2 %). Low to modest percentages agreed that a husband had good reason to hit his wife if she refuses to have sexual relations with him (8 %), asks him whether he has other girlfriends (9 %), does not complete the housework to his satisfaction (12 %), goes out without telling him (14 %), and disobeys him (17 %). Higher percentages agreed that a husband had good reason to hit his wife if she neglects the children (44 %), he finds out that she has been unfaithful (60 %), she rudely argues with him (65 %), and she argues with her parents-in-law (79 %).

Attitudes about wife hitting differed substantially by gender. Across all degrees of behavioral transgression, women expressed more favorable attitudes than men about IPV against wives. For example, at each end of the continuum, women more often agreed that a husband had good reason to hit his wife if she burns the food (3 % vs. 1 %) and if she argues with her parents-in-law (92 % vs. 67 %).

Exploratory and Confirmatory Factor Analyses

Table 2 presents the estimated factor loadings and fit statistics for the one-factor, nine-item EFA and CFA models with random-split-half samples. In both models, estimated factor loadings were positive, significant, and of similar magnitude (0.575–0.883 for EFA; 0.502–0.897 for CFA). Also for both models, fit statistics suggested adequate fits to the data (RMSEA = 0.068, CFI = 0.951, TLI = 0.935 for EFA; RMSEA = 0.048, CFI = 0.978, TLI = 0.970 for CFA).

Analysis of Differential Item Functioning by Gender

Table 2 also includes the analysis of DIF by gender (with women as the reference group). According to the baseline structural model that included only the indirect effect of gender on the latent factor, men had a −0.934 lower factor mean than did women (p ≤ .05), suggesting that men found significantly less good reason than did women for a husband to hit his wife. Estimates in the subsequent columns show effects on the attitudinal gap between women and men of taking uniform DIF by gender into account. Given the estimated factor mean difference by gender, men had significantly higher-than-expected probabilities than did women of agreeing that a husband had good reason to hit his wife if she does not complete her housework to his satisfaction (item 1), disobeys him (item 2), and argues with her parents-in-law (item 10), resulting in significantly positive path coefficients (0.669, 0.492, and 0.448, respectively). After accounting for this uniform DIF by gender, the effect of being a man on the factor mean attitudinal difference became even more negative, from −0.934 to −1.227. In other words, the factor mean difference for agreeing that a husband had good reason to hit his wife was 31.4 % lower for men than women after adjusting for men’s unexpectedly higher probabilities than women of agreeing with items 1, 2, and 10. Adjusting for uniform DIF by gender for these items significantly improved model fit based on the chi-square test for difference (at p ≤ .05). No other scale item showed significant uniform DIF by gender, so no other direct effects of gender on scale items were added.

Comparison of Gender Gaps in Derived Attitudinal Scores

Table 3 shows the results of using five different methods to estimate scores for men’s and women’s attitudes about physical IPV against a wife. For each method, we show the overall mean and median score, mean and median scores by gender, and differences in mean and median scores by gender (men – women). All mean and median scores were derived from the random split-half sample on which the MIMIC model for uniform DIF was estimated (N₂ = 528). The differences in mean factor scores (men – women) in Table 3 differ slightly from the factor mean differences in Table 2 because of indeterminacy in the computation of factor scores (Grice 2001). More notably, the results in Table 3 show that basic inferences regarding gender differences in attitudes about physical IPV against a wife do not differ across scoring methods: in all cases, men justify IPV significantly less often than do women. However, the magnitude of the gender difference in attitudes increases substantially—by about 30 %—for factor scores that take uniform DIF by gender into account (factor 9, DIF) versus those that do not (factor 9, no DIF). The magnitude of the gender gap in attitudes is similar for the scores based on the nine-item factor model that takes DIF by gender into account and six-item factor model for which items with DIF by gender are removed. Thus, the latter items may reliably capture gender gaps in attitudes about IPV in this setting.

Discussion

This effort is the first in a lower-income setting to test psychometrically attitudinal items about IPV against women that have been used widely in national surveys, like the DHS, since 1995 (Yount et al. 2011b). A combined EFA/CFA/DIF analysis strengthened scale development. The conduct of this study in Vietnam is timely given recent legal changes to promote gender equity and to criminalize IPV. The state also has undertaken an extensive media campaign to change norms about IPV and to curb its perpetration (Gender and Community Development Network GENCOMNET 2011; United Nations Population Fund (UNFPA) 2012). In such contexts, having a valid scale to measure women’s and men’s attitudes about IPV against women enhances the capacity to tailor interventions to the groups for whom attitudinal change is most needed. Also, using the same measurement scale over time permits an assessment of attitudinal change.

Our findings contrast with recent surveys in Vietnam, which have shown marked declines in the justification of IPV (General Statistics Office GSO 2006; 2011). In our study site in the rural North, almost all married women (95.1 %) and most married men (76.9 %) found good reason to hit a wife for at least one of 10 gender transgressions. These rates of justification are higher than those found in the 2006 and 2010–2011 MICS and the 2010 NSDVAWV surveys. These discrepancies may partly be attributed to the near doubling of items included in our survey (10 items) relative to the five items in the MICS surveys and the six items in the NSDVAWV survey, as well as to our inclusion of a completely new item (she argues with her parents-in-law). Our findings corroborate recent qualitative research, suggesting the persistence of Confucian ideas about gender hierarchy as well as Taoist ideas about “hot” masculine identity (Horton and Rydstrøm 2011; Khuat et al. 2010). Our findings also support ethnographic evidence of ongoing inequities in familial gender relations (Nguyen et al. 2011; Thanh et al. 2012; Werner 2009) and high rates of IPV against women (General Statistics Office GSO 2010; Nguyen et al. 2008).

Our findings confirm and extend research in Vietnam (Gender and Community Development Network GENCOMNET 2011) concerning gender attitudinal gaps about IPV against women. Compared with married men in My Hao district, married women more often found good reason to hit a wife for all of 10 transgressions of varying degrees. In MIMIC models including an indirect effect of gender, the factor mean was 0.934 lower for men than women, meaning that men stated systematically less often than women that a husband had good reason to hit his wife. Such gender gaps in attitudes are not unique to Vietnam but corroborate those in other lower-income settings (Uthman et al. 2010) and markedly contrast with those in wealthier settings, where men blame the victim systematically more often than do women (e.g., Bryant and Spencer 2003; Locke and Richman 1999).

Our analysis also revealed that DIF by gender for three items confounded the observed gender attitudinal gap. When we controlled for higher-than-expected probabilities for men than women of giving a positive response to these items, men had even less favorable attitudes than women about physical IPV against a wife. Indeed, controlling for DIF by gender for these items widened the gender gap in attitudes by 31.4 %. Such findings suggest the value of exploring measurement invariance for items like these before comparing the reported attitudes of women and men in Vietnam and other settings (e.g., Uthman et al. 2010).

Given these findings, we tested the factor structure for the subset of six attitudinal items that were free of DIF by gender. Analyses showed that these six items captured well a unidimensional attitudinal construct about physical IPV against women and a gender gap in attitudes of similar magnitude to the nine-item model that adjusted for DIF by gender. Thus, using the six items that were free of DIF by gender may be an economical way to measure men’s and women’s attitudes about IPV against women in multipurpose or violence surveys and in research on gender gaps in such attitudes in this context. These items should be useful elsewhere in Vietnam and in other countries; however, their validation in these other settings is recommended.

Together, our findings suggest the removal of four of the initial 10 items intended to capture attitudes about IPV against women. One item warranted removal because of a lack of variability among women and men. Only 2 % of respondents agreed that a husband had good reason to hit his wife if she burns the food (item 7). Notably, this item was included in the DHS in the 1990s for use in lower-income and/or more gender-inequitable countries, such as Egypt (EI-Zanaty et al. 1996), India (International Institute for Population Sciences (IIPS) and ORC Macro 2000), and Zimbabwe (Central Statistical Office (Zimbabwe) and Macro International Inc. 2000). In Vietnam, where food insecurity is less common and women’s economic role in the family is perhaps more visible, burning the food is less consequential for family survival and gender relations and therefore a less serious “offense” for Vietnamese women. Thus, we suggest removing this item from scales measuring attitudes about IPV against women in Vietnam but recommend its exploration in lower-income and more gender-inequitable settings.

The remaining three of the four items warranted removal because they showed significant DIF by gender (completing the housework, disobedience, and arguing with parents-in-law). Men and women may have understood these items differently or may have had different motivations for expressing agreement/disagreement with them. First, women may have weighed various small infractions with respect to the full range of domestic chores that wives typically perform; comparatively, men may have considered only its satisfactory completion (or not) without weighing specific chores. Second, women may have seen a wife’s obedience to her husband reflected in the many small acts that she performs daily, whether or not they are discussed; men may have considered only the wife’s actions that her husband has expressly forbidden. Third, Confucian ideas still underscore the belief that a person who disrespects one’s elders could dare to do other bad things, and a man who fails to “teach” his wife to respect his parents may himself be showing disrespect (tội Bất hiếu). Therefore, compared with women, men may consider as systematically more serious the implications for the husband of a wife who argues with his parents. Finally, the Women’s Union exerts a powerful influence on women’s thoughts about their daily domestic activities and place in the family (Schuler et al. 2006), but no mass organization in Vietnam exerts a parallel influence on men. More qualitative research on these three attitudinal items might reveal alternative wordings that would reduce or eliminate DIF by gender (Yount et al. 2014), or could yield a richer understanding of the social and institutional context underlying gender gaps in responses to these items in Vietnam.

Another approach to measure gender attitudinal gaps is to assess within-couple agreement/disagreement on matched husband-wife data (e.g., Ghuman et al. 2006; Smith et al. 1998; Yount and Li 2012). In secondary analyses of couples data on IPV in Egypt (Yount and Li 2012), spousal reports of IPV showed low chance-adjusted agreement (Kappa 0.23–0.43), with wife-yes/husband-no reports being the modal form of disagreement. Likewise, spouses in Asia have differed in their perceptions of the wife’s agency and have understood the responses to these questions differently, and husbands have ascribed higher agency to their wives than wives have to themselves (Ghuman et al. 2006). In studies of spousal attitudes about IPV against women, the intent is not to measure multiple-rater reports of ostensibly objective events but to use scales with equivalent measurement properties to measure (possibly different) subjective views of an event. Studies examining couples’ perceived fairness in the housework have shown that levels of agreement between spouses on this attitude depend importantly on attributes of the couple, not just individual attributes of the husband and wife (Smith et al. 1998). In practice, ethical guidelines for research on IPV discourage within-couple or within-household designs, given the risk of lost confidentiality and escalated violence against women (World Health Organization WHO 2001). Couples studies remain a useful strategy to validate scales measuring other attitudes and behaviors.

In sum, our analysis shows that standard attitudinal items about IPV against women may show significant DIF by gender. This potential should be explored before including these items in future surveys and mitigated, if possible, with modifications to question wording. Alternatively, all 10 items that we initially considered might be administered systematically in cross-national comparative surveys, and DIF by gender could be assessed and controlled analytically before decomposing gender gaps in attitudes about IPV. Our analyses suggest that several items used in prior surveys are free of DIF by gender and thus may reliably capture gender gaps in attitudes about IPV in Vietnam, and possibly elsewhere.

The important task of future studies in Vietnam and other lower-income countries will be to understand, after controlling for DIF by gender, why women more often than men find good reason to hit a wife. Without adjusting for DIF, women in 17 sub-Saharan African countries have had twice the odds as men of justifying IPV (Uthman et al. 2010). Substantively, such differences may arise as a result of gendered social learning processes in childhood (Yount and Li 2009), gendered power processes in the couple and family (Komter 1989; Yount 2011), or structural inequities between women and men (Kandiyoti 1988; Yodanis 2004; Yount 2011), which may induce women to tolerate IPV or even to see it as legitimate or normal. Methodologically, this difference also may arise from greater “desirability bias” among men, who more often could name laws that address IPV against wives (51.8 % vs. 39.8 % for women; authors’ calculations), for example. This difference also may arise from different interpretations of the mental task being asked in the preamble, “In your opinion, does a man have a good reason to hit his wife if . . . .” Women may have understood this setup to be asking about their experience or observations of IPV against women in a context in which such violence remains salient, regardless of their personal views about the hypothetical scenarios (Yount et al. 2012). Cross-nationally, variations in the preamble to similar attitudinal items do alter the prevalence of women’s positive responses (Yount et al. 2011b). Any truth in this explanation would imply a need to experiment with different preambles to improve the consistency of interpretation across gender so that decomposition of gender gaps in attitudes can focus on the substantive explanations.

Finally, the finding of greater DIF by gender for certain attitudinal items does not lessen their practical relevance in this setting. Understanding why men and women respond differently to these items may inform communication strategies that effectively promote change toward more equitable gender norms in each group. Although such items may be problematic to include in attitudinal scales, cognitive exploration of responses to them may guide efforts to address the sociocultural underpinnings of IPV.

Acknowledgments

This work was supported by NIH research Grant 5R21HD067834-01/02 (PIs Yount and Schuler) and the Hubert Department of Global Health, Rollins School of Public Health, Emory University. The article was written while Dr. Sarah Zureick-Brown was a postdoctoral fellow in the Hubert Department of Global Health, Rollins School of Public Health, Emory University. The authors thank anonymous reviewers for their comments on a prior version of this article. The authors also thank Vu Song Ha, Quach Trang, the entire field staff, and all participants for their time, effort, and dedication to this project.

Notes

References