Domestic abuse in the Covid-19 pandemic: measures designed to overcome common limitations of trend measurement

Abstract

Abstract Research on pandemic domestic abuse trends has produced inconsistent findings reflecting differences in definitions, data and method. This study analyses 43,488 domestic abuse crimes recorded by a UK police force. Metrics and analytic approaches are tailored to address key methodological issues in three key ways. First, it was hypothesised that reporting rates changed during lockdown, so natural language processing was used to interrogate untapped free-text information in police records to develop a novel indicator of change in reporting. Second, it was hypothesised that abuse would change differentially for those cohabiting (due to physical proximity) compared to non-cohabitees, which was assessed via a proxy measure. Third, the analytic approaches used were change-point analysis and anomaly detection: these are more independent than regression analysis for present purposes in gauging the timing and duration of significant change. However, the main findings were largely contrary to expectation: (1) domestic abuse did not increase during the first national lockdown in early 2020 but increased across a prolonged post-lockdown period, (2) the post-lockdown increase did not reflect change in reporting by victims, and; (3) the proportion of abuse between cohabiting partners, at around 40 percent of the total, did not increase significantly during or after the lockdown. The implications of these unanticipated findings are discussed.

Keywords

• Domestic abuse
• Intimate partner violence
• Coronavirus
• Change-point analysis
• Anomaly detection
• NLP