COVID-19 & Violence against Women and Children: What Have We Learned So Far?

Six months into the COVID-19 crisis, thousands of news stories have been published warning of the increased risks of violence against women and children (VAW/C). Research from previous health, economic, and political crises supports this dynamic, predicting increases in multiple risk factors for diverse forms of violence. Yet most press coverage relies on month-to-month statistics from highly volatile single sources from high-income countries like helplines, hospitalizations, and police records. 

In this note, we review rigorous studies that have analyzed how COVID-19 and related policies are impacting rates of VAW/C and highlight more reliable methods, while acknowledging limitations of underlying data sources. We propose recommendations for how to both broaden and deepen our collective understanding of how COVID-19 is impacting these forms of violence, and what can be done in response.

How should we interpret media-reported rates?

In some countries, reported rates of violence against women have both increased and decreased during COVID-19, depending on the type of violence and source of reporting. For example, in April in South Africa, calls to gender-based violence centers were reportedly increasing, while a group working closely with the National Prosecuting authority, police, and the Department of Social Development reported in May that rape and sexual assault cases were down by 50 percent. These conflicting reports are not unique, but they raise many questions around how to interpret and draw conclusions from different sources of data.  

While it is easy to produce such statistics, it is nearly impossible to identify meaningful changes based on simple month-to-month reporting from administrative data. This is due to numerous factors, including seasonality (i.e., certain events—including crimes—may be more likely to occur during certain times of the year), general time trends (i.e., VAW/C may be already increasing or decreasing at different rates in different locations), and noisiness of the data (accuracy of administrative data may be compromised, due to a variety of factors, including incomplete or poor reporting systems). Further, VAW/C data specifically suffers from widespread underreporting due to stigma, shame, and fear of retaliation, which may further compound biases in reported data. This makes it difficult, if not impossible, to understand underlying trends from month-to-month reports from specific service data.

Thankfully, researchers are working on producing more reliable analysis.

What does the emerging evidence say?

A handful of recent papers have begun to account for sources of bias, by moving beyond simple month-to-month comparisons in reported statistics. In total, we find 17 analyses looking at the relationship between COVID-19 and VAW/C. Methodologies vary, but new evidence broadly falls into three categories: 

1. time series or difference-in-differences analysis of administrative data, 
2. surveys of service providers and 
3. primary data collection of violence and/or related factors. 

While many types of time series modeling exist, the main benefit is the ability to account for time trends, including things like seasonalities and other patterns over time. Primary data collection also has many advantages, but cross-sectional data represents only a snap shot in time and thus may not be able to credibly disentangle whether changes are a due to COVID-19 or are product of previous conditions. We summarize some of the key findings from these papers, while acknowledging they have many additional interesting details we are not able to fully explore here.

While aiming to be inclusive, we do not include papers that do not attempt to correct for any biases, or do not clearly explain their underlying methodology, indicators, or assumptions. We note that many of the papers we rely upon are still in pre-print and working paper form, and are thus preliminary. In addition, we primarily find papers on VAW/C inside the home, yet acknowledge that experience of violence is certainly broader. Finally, VAW/C is not the only type of violence we may be concerned about during COVID-19, as inter-group and political violence, for example, may also be impacted. With these caveats, we round up existing studies, highlight limitations, and re-cap what we still do not know.

The majority of papers rely on reported crime or service call data, and analyze primarily domestic violence and assault measures (only one analyzes child maltreatment). 

Is reported VAW/C increasing? Results are mixed. Among 12 studies, three report decreases in VAW/C measures, three report increases, three report mixed findings, and three report no changes. Among the 12 studies, only two use primary data (Beland et al. 2020 in Canada and in Bangladesh Rashid et al. 2020). In addition, only two are in low- and middle-income country settings (LMICs, aforementioned Bangladesh study and Silverio-Murillo & de la Miyar, 2020 in Mexico). Only one, again from Bangladesh, uses qualitative data. Two papers go beyond reports of COVID-19 “effects” and explore the relationships with complementary programming (financial assistance, Beland et al. 2020) or alcohol prohibition policies (Silverio-Murillo & de la Miyar, 2020), neither finding evidence of moderating or interaction effects on violence-related measures.

Group A: Papers that measure impacts of COVID-19 or associated response measures on VAW/C

Table notes: Studies are ordered by direction of impact and alphabetical. ARIMA = auto regressive integrated moving average; BSTS = Bayesian structural time series; SARIMA = seasonal auto-regressive integrated moving average models
* Exact question asked is “How concerned are you about each of the following impacts of COVID-19?” option: “Violence in the Home.” With approximately 13 percent of respondents answering they were very or somewhat concerned.

Papers that measure impacts of COVID-19 or associated measures on VAW/C 

What do these estimates mean?

What can we take from these mixed estimates? In seeking to interpret findings, one key takeaway is that prevalence estimates from administrative data sources can tell us if reporting is changing, but may say little about actual underlying prevalence of violence and trends. This gap is because COVID-19 is likely to change both availability of services, as well as likelihood of reporting.

While there is little evidence of how large these biases may be, findings by Baron et al. 2020 in Florida demonstrate that identification of violence against children by teachers and other school personnel decreases when schools are closed. A recent similar analysis (not COVID-19 related) by Fitzpatrick and colleagues using a 14-year period shows that in the United States, showed that child maltreatment reports decrease as much as 65 percent when children are not in school. The same may be true for adult women who are not able to report in close quarters with perpetrators, or due to lost income and lockdown orders, are losing bargaining power within the home and access to their support networks outside the home.

Irrespective of COVID-19, we know from prior research that the gap between reported figures and actual underlying violence levels is large—due to the sensitive nature of VAW/C and associated risks that come with reporting. For example, in a study of population-level data from 24 LMICs, we found that on average, among women who disclosed experiencing physical and/or sexual violence from a partner or other individual, a small fraction (only 7 percent) reported to any formal source combined. Examining these formal sources individually, rates of formal reporting were even smaller: police (4 percent), social services (3 percent) or health facilities (1.5 percent). This implies that estimates from data from administrative sources could underestimate the total prevalence of violence ranging from 11- to 128-fold, depending on the region and type of reporting. Thus, it is easy to see how small changes in propensity to report could drive variation in administrative data. These dynamics are difficult to tease apart and present challenges that future studies will need to grapple with.

Analysis of service provision and public attitudes around VAW/C

Five other reports or papers speak to service providers’ experience, including their ability to provide support and public attitudes around VAW/C in light of COVID-19. They suggest a cause for optimism, showing increased attention to VAW/C has increased public support for investment in support services for survivors. However, amidst concerns of rising levels and complexity of VAW/C, they also shed light on the stark reality of providers’ limited ability to support increasing needs of survivors, due to social distancing constraints and strained institutional resources. Four surveys come from HICs, all collecting primary data, and one represents cross country data from LMICs.

Group B: Surveys of service providers’ perceptions and public attitudes towards VAW/C

Table notes: CSO = Civil society organizations; WRO = Women’s rights organization

Papers surveying service providers and public attitudes towards VAW/C

What do we still need to know?

There are still many gaps in our knowledge about how COVID-19 is changing dynamics around VAW/C. So far, initial studies only measure short-term effects, primarily from administrative data sources. In addition, they focus on violence against women inside the home, with few studies examining violence against children, or violence outside the home. In addition, fewer studies have focused on promising mitigation efforts. This shortcoming is likely due in part to the time frames, and the real challenges (both ethically and methodologically) in collecting new data on violence from women and/or from children directly during lockdowns. These efforts require honest conversations of if risks outweigh the benefits and must include full disclosure of ethical protocol by researchers, and efforts to ensure both harm mitigation. In addition, the added value of data collection should be scrutinized in relation to the actionability of results. For example, studies which simply collect cross-sectional data during lock-down may be unable to provide credible evidence of whether dynamics are changing due to COVID-19 or due to pre-existing patterns.

Promising prevention strategies can be inferred to some extent from pathways of impacts and the current knowledge around what works. For example, more than one study reviewed here suggests that continued interaction with social networks is important for accessing formal services for support. Thus, interventions which strengthen social networks without directly targeting risk factors for violence may still mitigate violence. Similarly, the evidence reviewed underscores how economic insecurity may increase risk of violence. Therefore, economic strengthening interventions addressing income or employment gaps induced by COVID-19 may be critical to reducing violence risk. Nevertheless, there are still gaps in our understanding of what works within pandemic settings, and more evidence is needed. The research agenda around political support, policy and financing aimed at preventing VAW/C and supporting survivors is also an area ripe for inquiry.

Notably, only three of the studies we review (fewer than 20 percent) use data from LMICs (Mexico, Bangladesh, and LMIC among service providers). This may be in part due to our search strategies, which excluded studies in non-English or Spanish languages. However, it also reflects a higher availability of public or open access data in HICs to study this topic. In addition, scholars in the HIC settings may have greater access to or be better positioned to respond to call for proposals for new funding and data collection efforts. For example, among eight studies recently awarded small grants to study violence and COVID-19 by the National Bureau of Economic Research, half were United States-based studies. This skewed geographic distribution must change over time. A promising example for progress lies with the Innovations for Poverty Action RECOVR hub, focused on LMIC setting research, which includes several studies focused on violence (e.g. Peru and Colombia).

We look forward to a fertile period of learning and focus on prevention of VAW/C going forward, catalyzed by the increased attention the issue has attracted in light of COVID-19. Policy relevant and actionable research in times of uncertainty can play a crucial role in understanding how to protect vulnerable populations from violence.

Thanks to David Evans for helpful comments