Investigating non-invariance of psychological intimate partner violence measures in the demographic and health surveys: roles of survey design and national context in 19 countries.

BACKGROUND

One third of adult women report lifetime psychological intimate partner violence (IPV). Controlling behavior is a common dimension of psychological IPV; however, evidence is mixed on its cross-national and cross-time measurement invariance, limiting its use to monitor Sustainable Development Goal (SDG) 5.2.1, to eliminate all forms of violence against women. We explored easier-to-modify survey-design features and harder-to-modify individual-level and national-level characteristics that may account for non-invariance of these controlling-behavior items.

METHODS

We analyzed data on five controlling behaviors administered to 373,167 ever-partnered women 15-49 years in 19 low- or middle-income countries in which at least two national Demographic and Health Surveys were administered during 2005-2019. We performed multiple-group confirmatory factor analysis (MGCFA) to test for exact forms of invariance and alignment optimization (AO) to test for approximate invariance across 7-9 survey-design groups, defined by the number of preceding questionnaire modules (to proxy respondent burden) and weeks of interviewer training (to proxy interviewer skills). Adjustment for covariates in the MGCFA assessed whether individual- and national-level characteristics could account for any observed non-invariance across survey-design groups.

RESULTS

In MGCFA without covariates, configural invariance of the controlling-behavior items was observed across survey-design groups. Exact invariance, partial invariance (with 20% of parameter estimates freed), and approximate invariance were not observed across groups. In adjusted MGCFA, neither woman-level covariates (schooling, attitudes about IPV against women) nor national-level covariates (women's mean schooling, mean attitudes about IPV against women, gender-related legal environment) alone or combined accounted for the non-invariance of controlling-behavior items across survey-design groups.

CONCLUSIONS

Comparing estimates for controlling behavior across country, time, and survey design variations warrants caution. Standardizing questionnaire length and interviewer training may improve the invariance of these items. Other characteristics, like ethnicity and language, may account for the non-invariance of controlling-behavior items across survey-design groups and should be tested. Current controlling-behavior items should be refined to enhance their comparability, and new controlling-behavior items should be identified and tested to improve the item set's content validity. Given current evidence of the high prevalence and health impacts of psychological IPV against women, advancing this research agenda is needed to monitor SDG 5.2.1.