Lakhasly

violence outcome data, although many studies included pre-COVID- related restrictions data from the prior year or years (i.e., pre-2020) (Bullinger et al., 2020; Campedelli et al., 2021; de la Miyar et al., 2021; Evans et al., 2021; Gerell et al., 2021; Gosangi et al., 2021; McLay, 2021; Nix & Richards, 2021; Payne & Morgan, 2020; Perez-Vincent et al., 2020; Ravindran & Shah, 2020; Rhodes et al., 2021). In addition, the domestic violence pre-post COVID-19-related restrictions data was derived from administrative/official records from police crime/incident reports, police calls for service, domestic violence hotline registries, or health records.
Fig. 2a graphically illustrates the range of the study-specific esti- mates of the percentage decrease/increase in domestic violence that occurred following the emergence of the COVID-19 pandemic and post- COVID-19-related restrictions relative to domestic violence that was documented prior to the COVID-19 pandemic and post-COVID-19- related restrictions. According to the 37 %-change estimates available from the 18 included studies,3 eight of the study estimates reported a decrease in domestic violence (range = − 0.28% to − 77.0%) compared to the 29 study estimates that reported an increase in domestic violence during the post-COVID-19 pandemic’s emergence and post-COVID-19- related restrictions (range = +0.60% to +75.0%). If we were to calcu- late an overall pre− /post- percentage difference, our results would indicate that the average of the 37 positive and negative %-changes amounts to a 7.86% increase in domestic violence. Fig. 2b reports these same results but for only the US-based studies and the result %-change in domestic violence indicates that the average of 31 positive and negative
3 The study estimates out-number the number of included studies as some of the studies provided a range (low/high) of estimates (for example, Bullinger et al., 2020) and some studies reported estimates separately for different lo- cations/jurisdictions (for example, Ashby, 2020; Nix & Richards, 2021).
%-changes equates to an 8.10% increase in domestic violence.
In the final stage of the analysis, effect sizes were generated for those included studies that reported sufficient information for an effect size to be calculated, which is not always the case when collating studies to include for meta-analyses. Fig. 3a provides a forest plot illustrating the distribution of the effect sizes with their corresponding 95% confidence intervals and related weights for the 12 studies (17 effect sizes). As can be seen, the majority of the effect sizes are positive (15 out of 17) and significant (12 out of 17) indicating that “the treatment” (i.e., the emergence of the COVID-19 pandemic and COVID-19-related re- strictions) increased domestic violence. The overall mean effect size generated from a random effects restricted maximum likelihood model was 0.66 (95% CI: 0.08–1.24; z = 2.24, p < .05), representing a medium effect.4 Fig. 3b provides the same forest plot but only for the US-based studies. Among only the US-based studies (7 studies, 12 effect sizes), the mean effect size increases to 0.87 (95% CI: 0.14–1.59). Finally,
4
3
A second overall weighted mean effect size was estimated from a random effects restricted maximum likelihood model after removing the two outlier effect sizes (Nix & Richards: Phoenix, 2021 and Gerell et al., 2021) as a sensitivity analysis. The results were similar (positive and significant) with the overall mean effect size being 0.28 (95% CI: 0.17–0.39; z = 5.04, p < .05). As such, we opted to retain these two studies in the overall mean effect size as presented in the text and main analysis.