Beliefs about Relationship Violence Against Women Scale: First and Second Order Confirmatory Factor Analysis and Gender Invariance in Ghana

Background: Standardized instruments for examining gender-based violence beliefs have grown rapidly (Flood, 2008) and recent advancements suggest that exposure to patriarchal structures and ideologies remains vital to comprehending the underlying beliefs (Hunnicutt, 2009). Regrettably, validation of instruments beyond the primary society in which they were developed remains sparse despite the relevance of cross-cultural validations to objective comparisons across societies (Arafat, Chowdhury, Qusar, & Hafez, 2016; Byrne & Watkins, 2003; Milfont & Fischer, 2010). Among instruments for examining attitudes toward gender-based violence, beliefs about relationship violence against women scale (BEREVIWOS) advances knowledge with its capture of psychological manipulation and control in intimate relationships (Fakunmoju et al., 2016, 2017). The present analyses examine first and second order factor structure and gender invariance of the instrument in Ghana. Specifically, the analyses determine whether the same items measure the constructs, whether the constructs have the same meanings for men and women, and whether the latent means can be compared across gender.

Methods: Respondents (N = 322) included males (n = 162, 50.3%) and females (n = 160, 49.7%) with an average age of 31.24 years (SD = 10.57). BEREVIWOS comprise 12-item, three sub-constructs, namely physical violence (e.g., because a man is the head of his household, he has a right to hit his woman), psychological manipulation and control (e.g., a man should do everything within his power to make his woman obey him) and sexual violence (e.g., a man has the right to force his woman for sex) (Fakunmoju, 2016). Response choices range from strongly disagree=1 to strongly agree=5. Cronbach’s alpha is as follows: “Physical violence” .76, “Psychological manipulation and control” .89, and “Sexual violence” .83. CFA using maximum likelihood estimation method (Stata-15) and measurement invariance across gender was examined. For the CFA, RMSEA, CFI, were examined to determine acceptable model fit (Kline, 2006). Non-significant change (Δ) in chi-square, RMSEA values falling within the confidence interval, and the ΔCFI being smaller than -0.01 were used to determine measurement invariance across gender (Cheung & Rensvold, 2002; Timmons, 2010).

Results: Acceptable fit for single-group first order (RMSEA .063; CFI .966, TLI .953) and second-order (RMSEA .082; CFI .940, TLI .920) CFA with three error-correlations were identified. Multi-group CFA also demonstrated reasonable configural (RMSEA .073, CFI .953, TLI .935) and metric (RMSEA .067, CFI .956, TLI .945) invariance for gender (Δχ² 3.54, df = 9, ns; ΔCFI -0.003). Although RMSEA values fell within the confidence interval for scalar invariance, the change in chi-square was significant and the change in CFI was greater than -0.01, suggesting inadequate capture of scalar invariance.

Conclusions and implications: The hypothesized model of BEREVIWOS was valid for Ghana. Configural invariance suggests that the same items measured the constructs and metric invariance suggests the constructs have the same meaning for respondents across gender. The non-invariance of intercepts (i.e., scalar non-invariance) across gender suggests that future studies may consider personal, socio-cultural, and religious factors that may precipitate systematic responses to BEREVIWOS items in order to offer valid explanations of gender differences in latent means.