This blog by originally appeared on website on 13 November 2020.
There has been a global surge in domestic violence since the onset of Covid-19. This column provides insights into what may be driving this rise, drawing on evidence from Brazil. Job loss leads to increases in domestic violence, irrespective of whether it is the perpetrator or victim whose job is lost. Both income stress and an increase in time spent together seem to contribute to this. Unemployment benefits have mitigation potential if they can be supplemented by policies designed to encourage a return to work.
Social distancing in the wake of the global pandemic has led to millions of families being 鈥榣ocked down鈥� together in their homes and to widespread job and income losses. This has coincided with a substantial global surge in domestic violence, which the United Nations has described as a 鈥渟hadow pandemic鈥� (UN Women 2020).
Identifying the behavioural causes is key for devising potential solutions. It seems plausible that both job loss and mobility restrictions have fuelled domestic violence, but there is limited evidence. Our research analyses administrative data from Brazil that pre-date the pandemic, but which demonstrate the relevance of these causes (Bhalotra et al. 2020). Our study also provides the first evidence on the mitigating potential of unemployment benefits.
Mechanisms that potentially link job loss to domestic violence
The literature on domestic violence suggests alternative mechanisms by which job loss of either partner may influence domestic violence:
- Job loss can alter the balance of bargaining power within the couple by modifying 鈥榦utside options鈥� (Anderberg et a.l 2016, Aizer 2010). It can challenge gender stereotypes by changing the relative earnings of the man, and domestic violence can emerge as a manifestation of 鈥榤ale backlash鈥� (Macmillan and Gartner 1999).
- Job loss of women can blunt domestic violence motivated by instrumental control (Bloch and Rao 2002, Anderberg and Rainer 2011, Carr and Packham 2020).
- Job loss will tend to lead to the couple spending more time together, increasing 鈥榚xposure鈥� or opportunities for violence (Dugan et al. 2003).
Job loss constitutes a significant shock to household income. This can trigger renegotiation of the allocation of a shrunken pie, with stress lowering the bar for conflict. Previous work has demonstrated an association of job loss with psychological stress and substance abuse (Black et al. 2015, Schaller and Stevens 2015) consistent with job loss tightening liquidity constraints, generating uncertainty about future income and/or creating a sense of failure. Stress can, in turn, lead to domestic violence. Card and Dahl (2011) highlight the relevance of emotional cues for violence, and several studies show that substance abuse is a proximate determinant (Lee Luca et al. 2019)
Data for Brazil
We gained access to court registers for Brazil that contain every domestic violence case between 2009 and 2017. There were two million cases, representing 11% of all criminal justice cases. We linked the plaintiff and defendant in these cases to administrative data containing longitudinal employment records. The data contain about 100 million workers, 60 million employment spells and 10 million layoffs per year.
Identification of causal relationships
To identify causal impacts of unwarranted individual job loss, we track individuals who lose their jobs in mass layoffs. For each individual suffering job loss, we generate matched controls defined as individuals of the same birth cohort, earnings category, tenure, industrial sector and state, who do not lose their jobs in the same year.
This facilitates identification of dynamic heterogeneous treatment effects (De Chaisemartin and D鈥橦aultfoeuille 2019). It also purges region-industry level shocks, which we test by additionally controlling for municipality-industry-year fixed effects. We conduct a number of robustness checks, on selection, anticipation, judicial lags, missing data and, importantly, the concern that job loss changes reporting behaviour. Our analysis of unemployment insurance exploits a discontinuity in eligibility criteria to achieve identification.
Job loss of men and women leads to higher domestic violence
Male job loss results in a roughly 30% increase in the chances that a man perpetrates violence. Female job loss delivers a percentage increase in the chances that a woman is victimised that is almost twice as large.
These effects persist through the three years in our frame. They hold across the distribution of age, education and income. The same pattern of results emerges when we use data identifying couples. We find the same pattern when we measure domestic violence by women鈥檚 use of public shelters rather than as judicial cases.
The results are unlikely to be driven by reporting bias
A common problem with analysing data on reported acts of violence is that it can be difficult to disentangle changes in the occurrence of violence from changes in reporting behaviour. Women may be more likely to report violence when the man loses his job, and less likely to report it when they lose their jobs 鈥� for example, because they are financially reliant on their partners.
Consider the direction of the likely bias. If job loss influences reporting in the hypothesised direction and reporting drives our results, then we expect to see an increase in reported violence following male job loss, but a decrease following female job loss. Since we find an increase in reported violence (and use of public shelters) following female job loss, it seems unlikely that this result is driven by reporting. But it remains possible that the results for male job loss are.
We conducted further checks for male job loss. The most informative involves restricting the sample to in flagrante cases in which the perpetrator is 鈥榗aught in the act鈥�, so the case is filed without the woman having to report. Male job loss continues to be followed by a large increase in domestic violence in this restricted sample.
The shock of income loss, bargaining, and exposure seem to be relevant mechanisms
We propose the following explanation. Job loss of either partner constitutes a major shock to household income. This disturbs the equilibrium, leading the couple to renegotiate allocation of a tighter household budget, creating grounds for conflict. Additional stresses deriving from income uncertainty and a sense of unworthiness potentially aggravate this. At the same time, the couple tends to spend more time together if either is unemployed, and this increases exposure or opportunities for violence.
The manner in which any conflict is resolved depends on which partner loses their job. If the woman loses her job then, in line with the household bargaining model, she is more likely to be victimised. But if the man loses his job, then he is less likely to perpetrate violence. This can explain our finding of a larger increase in domestic violence following female compared with male job loss.
On its own the bargaining model would seem inconsistent with our findings because it predicts a decline in domestic violence following male job loss. But combined with our hypothesis that an income shock increases the potential for conflict, it helps explain the findings.
Unemployment benefits compensate income losses but can increase exposure
Unemployment is synonymous with a decline in income and an increase in time at home, and our estimates capture the joint effect of these two changes. With a view to isolating the income the role of income loss, we exploit experimental variation generated by unemployment insurance (UI) eligibility rules in Brazil. Workers who have previously received benefits in Brazil need to have at least a 16-month gap before they can claim benefits again (Gerard et al 2019). We compare the risk of domestic violence among unemployed men who are just eligible versus those who are just ineligible for UI using a regression discontinuity design. Unemployment benefits in Brazil cover, on average, about 80% of former earnings and last three to five months.
While benefits are being paid, men eligible and ineligible for UI are equally likely to commit domestic violence. Once benefits expire, eligible men are more likely to commit violence. The reason is that eligibility lengthens unemployment durations (Katz and Meyer 1990, Lalive 2008), increasing exposure. While benefits are being received, exposure effects are offset by an income effect. Overall, while we have a natural experiment that, in principle, identifies the impact of a policy designed to relax income constraints, in practice UI has behavioural effects that reintroduce the relevance of time spent at home.
In a separate exercise using all men, we confirm that income transfers mitigate by leveraging the fact that severance pay and UI increase with tenure. We find impacts of job loss on domestic violence attenuate at high tenure.
The upshot for policy
Our main findings are that job loss influences domestic violence first through generating an income shortfall, and second by increasing exposure. So, the ideal policy intervention would compensate the income shortfall and get people out of the home and back to work.
Unemployment benefits can help but need to be combined with active policies aimed at getting the unemployed back to work (e.g. through training or support with job search). Experiments conducted in Kenya (Haushofer et al. 2019) and Mali (Heath et al. 2020) show that cash transfers to men reduce violence against female partners, highlighting the potential for welfare payments to be effective if incentive effects can be avoided by design.
Conclusion
As the pandemic has led to earnings losses and to families spending more time together, our research potentially illuminates the recent surge in domestic violence.
Media coverage of the Covid-19 surge has highlighted the role of exposure. Brazilian official Luciana Azambuja recently wrote in the popular press that 鈥淸s]ocial isolation will make families spend more time together. This can generate more conflicts.鈥� In addition to forcing couples together for longer, lockdown has limited contact with social networks and this may have contributed to domestic violence (Gelles 1983, 1993, Usher et al. 2020). Etheridge and Spantig (2020) find a deterioration in mental health in the UK during the pandemic that is larger among women, and the decline in social interactions appears to be an important factor.
Lockdown has been eased in many countries. But unemployment rates look set to rise, potentially increasing exposure even in the absence of mobility restrictions. Unemployment also tightens income constraints. Our research therefore suggests that domestic violence may persist beyond lockdown.
The policy infrastructure has been primarily concerned with providing support to victims in the shape of shelters, counselling and protection orders. Interventions designed to prevent domestic violence have focused on the economic empowerment of women, though they sometimes misfire (Angelucci 2008, Heath 2014, Bhalotra et al. 2019, Tur-Prats 2019, Estefan 2019, Kotsadam and Villanger 2020, Carr and Packham 2020). Our research suggests the relevance of compensating both men and women for income losses.
References
Aizer, A (2010), 鈥淭he Gender Wage Gap and Domestic Violence鈥�, American Economic Review 100: 1847鈥�1859.
Anderberg, D, H Rainer, J Wadsworth and T Wilson (2016), 鈥淯nemployment and Domestic Violence: Theory and Evidence鈥�, The Economic Journal 126: 1947鈥�1949.
Anderberg, D and H Rainer (2011), 鈥淓conomic Abuse: A Theory of Intrahousehold Sabotage鈥�, Journal of Public Economics 97: 282-295.
Angelucci, M (2008), 鈥淟ove on the Rocks: Domestic Violence and Alcohol Abuse in Rural Mexico鈥�, The B.E. Journal of Economic Analysis and Policy 8(1).
Bhalotra, S, D Britto, P Pinotti and B Sampaio (2020), 鈥淛ob Displacement, Unemployment Benefits and Domestic Violence鈥�, mimeo, July (available from the authors).
Black, S E, P J Devereux, and K G Salvanes (2015), 鈥淟osing heart? The effect of job displacement on health鈥�, ILR Review 68(4): 833-861.
Bloch, F and V Rao (2002), 鈥淭error as a Bargaining Instrument: A Case Study of Dowry Violence in Rural India鈥�, American Economic Review 92(4): 1029-1043.
Card, D and G Dahl (2011), 鈥淔amily Violence and Football: The Effect of Unexpected Emotional Cues on Violent Behavior鈥�, The Quarterly Journal of Economics 126(1): 103鈥�143.
Carr, J and A Packham (2020), 鈥淪NAP Schedules and Domestic Violence鈥�, Journal of Policy Analysis and Management.
De Chaisemartin, C, and X D鈥橦aultfoeuille (2020), 鈥淭wo-way fixed effects estimators with heterogeneous treatment effects鈥�, American Economic Review, forthcoming.
Dugan, L, D Nagin and R Rosenfield (2003), 鈥淓xposure Reduction or Retaliation? The Effects of Domestic Violence Resources on Intimate鈥怭artner Homicide鈥�, Law and Society 37(1): 169-198.
Estefan, A (2019), 鈥淔emale Labour Market Access, Divorce, and Intra-Domestic Violence: Evidence from Mexico鈥�, mimeo, UCL.
Etheridge, B and L Spantig (2020), 鈥淭he Gender Gap in Mental Well-Being During the Covid-19 Outbreak: Evidence from the UK鈥�, ISER Working Paper No 2020-08.
Gelles, R J (2007), 鈥淔amily violence鈥�, in D J Flannery, A T Vazsonyi, and I D Waldman (eds), The Cambridge handbook of violent behavior and aggression, Cambridge University Press.
Gerard, F, M Rokkanen, and C Rothe (2019), Bounds on treatment effects in regression discontinuity designs with a manipulated running variable鈥�, Technical Report, National Bureau of Economic Research.
Haushofer, J, C Ringdal, J P Shapiro, and X Y Wang (2019), 鈥淚ncome Changes and Intimate Partner Violence: Evidence from Unconditional Cash Transfers in Kenya鈥�, NBER Working Paper No. 25627.
Heath, R (2014), 鈥淲omen鈥檚 access to labor market opportunities, control of household resources, and domestic violence: Evidence from Bangladesh鈥�, World Development 57: 32鈥�46.
Heath, R, M Hidrobo and S Roy (2020), 鈥淐ash transfers, polygamy, and intimate partner violence: Experimental evidence from Mali鈥�, Journal of Development Economics, March.
Katz, L F and B D Meyer (1990), 鈥淭he impact of the potential duration of unemployment benefits on the duration of unemployment,鈥� Journal of Public Economics 41(1): 45-72.
Kotsadam, A and E Villanger (2020), 鈥淛obs and intimate partner violence 鈥� Evidence from a Field Experiment in Ethiopia鈥�, CESifo Working Paper 8108.
Lalive, R (2007) 鈥淗ow do extended benefits affect unemployment duration? A regression discontinuity approach鈥�, Journal of Econometrics 142(2): 785-806.
Lee Luca, D, E Owens and G Sharma (2019), 鈥淭he Effectiveness and Effects of Alcohol Regulation: Evidence from India鈥�, IZA Journal of Development and Migration 2019: 9-4.
Macmillan, R and R Gartner (1999), 鈥淲hen She Brings Home the Bacon: Labor-Force Participation and the Risk of Spousal Violence against Women鈥�, Journal of Marriage and Family 61(4): 947.
Schaller, J, and A H Stevens (2015), 鈥淪hort-run effects of job loss on health conditions, health insurance, and health care utilization鈥�, Journal of Health Economics 43: 190-203.
Tur-Prats, A (2019), 鈥淯nemployment and Intimate-Partner Violence: A Cultural Approach鈥�, mimeo, University of California Merced.
UN Women (2020), 鈥溾��, briefing.
Usher, K, N Bhullar, J Durkin, N Gyamfi and D Jackson (2020), 鈥淧andemic鈥恟elated behaviours and psychological outcomes; A rapid literature review to explain COVID鈥�19 behaviours鈥�, International Journal of Mental Health Nursing 20 April.
