VoxEU Column COVID-19 Gender

Assessing the magnitude of the domestic violence problem during the COVID-19 pandemic

Concerns that domestic violence would intensify during the COVID-19 lockdowns were not borne out by early research. But that research relied on police records. This column suggests that studies based solely on police-recorded incidents of domestic violence provide a less accurate picture than sources less susceptible to changes in reporting behaviour. Based on internet search activity, the authors find a 40% increase in domestic violence incidents in London during the lockdowns – seven to eight times larger than estimates relying on police data alone.

At the beginning of the ongoing COVID-19 pandemic, many countries implemented strict self-isolation and stay-at-home orders. Although these measures effectively curbed the spread of the virus, they brought with them the risk of increasing domestic violence. Indeed, many women’s support groups and domestic abuse helplines around the world reported a sharp increase in DV help-seeking of anywhere between 25% and 80% (e.g. Human Rights Watch 2020).

Yet, a set of recent empirical studies that use police-recorded calls-for-service or crime data to estimate impacts of COVID-19 lockdowns on the incidence of domestic violence have reported relatively modest effects, generally far below media-reported increases in calls and contacts to helplines and charities. One major concern with this evidence is the following: it is highly plausible that COVID-19 and the associated lockdowns have not only affected the incidence of domestic violence, but also victims’ reporting behaviour to the police. Indeed, lockdowns left victims of  domestic violence trapped at home with their violent partners, limiting their ability to safely report to the police (Campbell 2020, Kofman and Garfin 2020). Therefore, any empirical study that narrowly focuses on police-recorded DV incidents alone runs the risk of underestimating the domestic violence problem during COVID-19 and similar crises.

Measuring domestic violence using internet search activity

To overcome this measurement problem, in a recent paper (Anderberg at al. 2021) we propose a model-based algorithm for measuring temporal variation in incidence of domestic violence based on internet search data. Our approach uses five years of pre-2020 data to regress daily internet search activity for 35 terms related to seeking help for domestic violence on daily incidents of domestic violence recorded by the London Metropolitan Police (both observed). Since both reflect the same underlying (unobserved) temporal variation in domestic violence incidence, this leads to a positive correlation that is stronger for the most relevant/least noisy internet search terms. This, in turn, allows us to use estimated signal-to-noise ratios as weights to create a composite domestic violence index based on internet search activity. Our model establishes two conditions under which this measure yields less biased estimates of the domestic violence problem during COVID-19 than those based on police-recorded crime data: lockdowns have made seeking help generally more difficult for victims and have limited help-seeking through the police relatively more than through the internet.

The scale of the domestic violence problem during COVID-19

Our study of domestic violence during COVID-19 yields several interesting findings. First, after using pre-2020 data (from 1 April 2015 to 31 December 2019) to train our algorithm, we use pre-lockdown 2020 data (from 1 January to 15 March) to validate that our internet search-based domestic violence index has predictive power for such crimes recorded by the London Metropolitan Police. This is illustrated in Figure 1(a), which shows that our search-based domestic violence index exhibits positive correlation and simple co-movement with recorded domestic violence crimes. As an additional verification of the validity of our approach, we show that higher temperatures, which are known to increase the risk of domestic violence (Butke and Sheridan 2010), are not only significant predictors of police-recorded crimes but are also highly correlated with our search-based index.

Figure 1 Time series for testing period and 2020 until end of first lockdown

(a) Testing period (Jan 2020 – Mar 2020)  



(b) Testing period + first lockdown



Notes: The figure shows the residuals of the normalised daily counts of domestic violence crimes recorded by the London MPS and of the search-based domestic violence index after removing year, month, and day-of-the-week fixed effects from each series. The initial normalisation rescaled both variables to have a mean of 100 over the algorithm training period of 1 April 2015 to 31 December 2019. The residualised series are shown in Panel (a) for the testing period 1 January to 15 March 2020. Panel (b) extends the series plotted in panel (a) to include the lockdown period.

Figure 2 Comparison of estimated effect of the COVID-19 lockdown as measured using (a) police records and (b) internet search activity

(a) Police records


(b) Internet search activity



Notes: The figure plots the coefficients from two regressions estimating the effect of the London lockdown by two-week intervals. The regressions control for year, month, and day-of-the-week effects, as well as for temperature and rainfall.

Second, analysing the London lockdown, we establish that while the time series for our index and police-recorded domestic violence crimes followed each other closely over the testing period, this feature breaks down at the onset of the London lockdown. This is illustrated in Figure 1(b), showing that the increase in the search-based index after lockdown measures were implemented was substantially larger and sharper than the increase in police-recorded domestic violence crimes. However, this observation is purely descriptive, as it does not account for time and meteorological effects. Once we net out these confounding factors, we obtain our main finding, which is illustrated in Figure 2. For both domestic violence incidents recorded by Metropolitan Police (Panel a) and our search-based index (Panel b), the London lockdown had no immediate impact on domestic violence, but a significant effect emerged somewhere between three to six weeks into the lockdown. In level terms however, we find a 40% increase at peak in our search-based domestic violence index, seven to eight times larger than the increase in police-recorded crimes, and much closer to the size of the increase reported by the UK’s National Domestic Abuse Helpline. When we replicate our results for London using comparable crime and internet search data for the city of Los Angeles, California, we obtain strikingly similar results.

Third, if we assume that the increase in the search-based index accurately captures the effect of the London lockdown on domestic violence incidence, whereas the lower increase in police-recorded domestic violence crimes reflects a reduced reporting rate by victims, we are able to estimate the number of ‘missing’ police-recorded crimes over the lockdown period. The prediction we obtain indicates that the London Metropolitan Police would have recorded an additional 4,700 domestic violence crimes over the lockdown period had the rate of reporting to the police itself not been reduced by the lockdown. 

Concluding remarks

Research on domestic violence has burgeoned over the past decade (e.g. Alesina et al. 2016, Bhalotra 2020), and more recently there have been numerous investigations assessing the impact of COVID-19 and associated lockdowns (Immordino et al. 2020, Yamamura and Tsutsui 2020).  An important conclusion that can be drawn from our study is that evidence based solely on police-recorded domestic violence incidents is unlikely to provide an accurate picture of the magnitude of the problem during crises like the COVID-19 pandemic. In such assessments, using complementary data sources less prone to changes in help-seeking or reporting behaviour would allow for a better understanding of the lower and upper bounds of likely impacts on domestic violence . Our algorithm for measuring temporal variation in domestic violence incidence based on internet search activity offers one option for complementing assessments based on police records. Equally important is the use of data from domestic abuse hotlines, which to date are rarely systematically collected and made available for research.


Alesina, A, B Benedetta, and E La Ferrara (2016), “Violence against women: A cross-cultural analysis for Africa”,, 25 March 2016.

Anderberg, D, H Rainer, and F Siuda (2021), “Quantifying domestic violence in times of crisis: An internet search activity-based measure for the COVID-19 pandemic”, Journal of the Royal Statistical Society: Series A, forthcoming.

Bhalotra, S (2020), “A shadow pandemic of domestic violence: The potential role of job loss and unemployment benefits”,, 13 November 2020.

Butke, P and S C Sheridan (2010), “An analysis of the relationship between weather and aggressive crime in Cleveland, Ohio”, Weather, Climate, and Society 2(2): 127–139.

Campbell, A M (2020), “An increasing risk of family violence during the Covid-19 pandemic: Strengthening community collaborations to save lives”, Forensic Science International: Reports, 100089.

Human Rights Watch (2020), “UK failing domestic abuse victims in pandemic”, 8 June.

Immordino, G, M Berlin, F F Russo and G Spagnolo (2020), “The role of prostitution markets in domestic violence during Covid-19”,, 13 September.

Kofman, Y B and D R Garfin (2020), “Home is not always a haven: The domestic violence crisis amid the covid-19 pandemic”, Psychological Trauma: Theory, Research, Practice, and Policy. 

Yamamura, E and Y Tsutsui (2020), “COVID-19, mental health, and domestic violence: Evidence from Japan”,, 22 June.

1,575 Reads