Impact of Measurement Variability on Study Inference in Partner Violence Prevention Trials in Low- and Middle-Income Countries

Validity of a common measure of intimate partner violence perpetration: Impact on study inference in trials in low- and middle-income countries

Background

In lower-and middle-income countries (LMICs), studies of interventions to reduce intimate partner violence (IPV) perpetration are expanding yet measurement equivalence of the construct has not been established. We assessed the measurement equivalence of physical and sexual IPV perpetration used in recent trials in LMICs and tested the impact of non-invariance on trial inference.

Methods

With data from three recent intervention trials among men (sample size 505-1537 across studies), we calculated tetrachoric correlations among items and used multiple-group confirmatory factor analysis to assess invariance across arms and over time. We also assessed treatment effects adjusting for covariate imbalance and using inverse probability to treatment weights to assess concordance of invariant measures with published results, where warranted.

Findings

The average correlation among items was high and increased over time with several items in two studies showing correlations ≥0.85 at endline. Increases in correlation for physical IPV were concentrated in the treatment arm in two of the studies. The increase in correlation in sexual IPV differed by arm across studies. Across all studies, a correlated two-factor solution was the best fitting model according to the EFAs and CFAs. One study demonstrated measurement invariance across arms and over time. In two of the studies, longitudinal measurement non-invariance was detected in the intervention arms. In post hoc testing, one study attained invariance with a one-factor model and study inference was concordant with published findings. The other study did not attain even partial invariance.

Conclusion

Common measures of physical and sexual IPV perpetration cannot be used for valid effect estimation without further refinement. The study highlights the need for an expanded item set, content validity assessments, further measurement invariance testing, and then consistent use of the item sets in future intervention trials to support accurate inference on the effectiveness of IPV perpetration prevention interventions.

Validity of a Common Measure of Intimate Partner Violence Perpetration: Impact on Study Inference in Trials in Low- and Middle-Income Countries

Background

In lower-and middle-income countries (LMICs), studies of interventions to reduce intimate partner violence (IPV) perpetration are expanding, yet measurement equivalence of the IPV perpetration construct that is the primary outcome in these investigations has not been established. We assessed the measurement equivalence of physical and sexual IPV perpetration item sets used in recent trials in LMICs and tested the impact of non-invariance on trial inference.

Methods

With data from three intervention trials among men (sample size 505-1537 across studies) completed in 2019, we calculated tetrachoric correlations among items and used multiple-group confirmatory factor analysis to assess invariance across arms and over time. We also assessed treatment effects adjusting for covariate imbalance and using inverse probability to treatment weights to assess concordance of invariant measures with published results, where warranted.

Findings

The average correlation among items measuring IPV perpetration was high and increased by 0.03 to 0.15 for physical IPV and 0.07 to 0.17 for sexual IPV over time with several items in two studies showing correlations ≥ 0.85 at endline. Increases in the degree of correlation for physical IPV were concentrated in the treatment arm in two of the studies. The increase in correlation in sexual IPV differed by arm across studies. Across all studies, a correlated two-factor solution was the best fitting model according to the EFAs and CFAs. One study demonstrated measurement invariance across arms and over time. In two of the studies, longitudinal measurement non-invariance was detected in the intervention arms. In post hoc testing, one study attained invariance with a one-factor model and study inference was concordant with published findings. The other study did not attain even partial invariance.

Conclusion

Common measures of physical and sexual IPV perpetration cannot be used validly for comparisons across treatment versus control groups over time without further refinement. The study highlights the need for an expanded item set, content validity assessments, further measurement invariance testing, and then consistent use of the item sets in future intervention trials to ensure valid inferences regarding the effectiveness of IPV perpetration prevention interventions within and across trials.

Collateral effects of COVID-19 stay-at-home orders on violence against women in the United States, January 2019 to December 2020

BACKGROUND

The necessary execution of non-pharmaceutical risk-mitigation (NPRM) strategies to reduce the transmission of COVID-19 has created an unprecedented natural experiment to ascertain whether pandemic-induced social-policy interventions may elevate collateral health risks. Here, we assess the effects on violence against women (VAW) of the duration of NPRM measures that were executed through jurisdictional-level orders in the United States. We expect that stay-at-home orders, by reducing mobility and disrupting non-coresident social ties, are associated with higher incident reporting of VAW.

METHODS

We used aggregate data from the Murder Accountability Project from January 2019 through December 2020, to estimate count models examining the effects of the duration of jurisdictional-level (N = 51) stay-at-home orders on femicide. Additionally, we used data from the National Incident-Based Reporting System to estimate a series of count models that examined the effects of the duration of jurisdictional-level (N = 26) stay-at-home orders on non-lethal violence against women, including five separate measures of intimate partner violence (IPV) and a measure of non-partner sexual violence.

RESULTS

Results from the count models indicated that femicide was not associated with COVID-19 mitigation strategies when adjusted for seasonal effects. However, we found certain measures of non-lethal VAW to be significantly associated in adjusted models. Specifically, reported physical and economic IPV were positively associated with stay-at-home orders while psychological IPV and non-partner sexual violence were negatively associated with stay-at-home orders. The combination measure of all forms of IPV was positively associated with the duration of stay-at-home orders, indicating a net increase in risk of IPV during lockdowns.

CONCLUSIONS

The benefits of risk-mitigation strategies to reduce the health impacts directly associated with a pandemic should be weighed against their costs with respect to women's heightened exposure to certain forms of violence and the potentially cascading impacts of such exposure on health. The effects of COVID-19 NPRM strategies on IPV risk nationally and its immediate and long-term health sequelae should be studied, with stressors like ongoing pandemic-related economic hardship and substance misuse still unfolding. Findings should inform the development of social policies to mitigate the collateral impacts of crisis-response efforts on the risk of VAW and its cascading sequelae.

Optimizing the Construction of Outcome Measures for Impact Evaluations of Intimate Partner Violence Prevention Interventions

Most impact evaluations of intimate partner violence (IPV) prevention interventions use binary measures of "any" versus "no" physical and/or sexual IPV as their primary outcome measure, missing opportunities to capture nuance. In this study, we reanalyzed secondary data from six randomized controlled trials conducted in low- and middle-income countries-Bandebereho (Rwanda), Becoming One (Uganda), Indashyikirwa (Rwanda), MAISHA CRT01, MAISHA CRT02 (Tanzania), Stepping Stones Creating Futures (South Africa), and Unite for a Better Life (Ethiopia), to assess how different conceptualizations and coding of IPV variables can influence interpretations of the impact of an intervention. We compared the standard outcome measures to new measures that reflect the severity and intensity of violence and whether interventions prevent new cases of IPV or reduce or stop ongoing violence. Results indicate that traditional binary indicators masked some of the more subtle intervention effects, and the use of the new indicators allowed for a better understanding of the impacts of the interventions. Conclusions on whether a program is perceived "to work" are highly influenced by the IPV outcomes that the investigators choose to report, and how they are measured and coded. Lack of attention to outcome choice and measurement could lead to prematurely abandoning strategies useful for violence reduction or missing essential insights into how programs may or may not affect IPV. While these results must be interpreted cautiously, given differences in intervention types, the underlying prevalence of violence, sociodemographic factors, sample sizes, and other contextual differences across the trial sites, they can help us move toward a new approach to reporting multiple outcomes that allow us to unpack the "impact" of an intervention by assessing intervention effect by the severity of violence and type of prevention, whether primary and secondary.