Can COVID-19 Lockdown Reduce Crimes Against Women? A District- Level Analysis from India

Abstract

In response to controlling the COVID-19 pandemic, the Indian government implemented a nationwide lockdown on 24 March 2020. We study the effect of lockdown on crimes against women. Using district-level panel data from 457 districts in India for five months (before, during and post-lockdown), we examine the interaction effect of COVID-19 containment zones and lockdown on crimes against women. Results suggest a differential impact of the lockdown on crime across different containment zones. Compared to the most COVID-19 affected zone, the less affected zones show a larger fall in crimes against women due to the imposition of a lockdown.

JEL Codes: C33, J16, J78

Keywords

COVID-19 lockdown crimes against women zero-inflated negative binomial district-level India

Introduction

In the face of the pandemic’s challenges, the Indian government took large steps to prevent the community transmission of the COVID-19 virus. The first phase of the national lockdown in India started on 24 March 2020. To further curb the spread of the virus, the lockdown was extended until 17 May 2020. The central government formed red, orange and green containment zones based on the total number of COVID-19 cases and the doubling rate of viral infections in each Indian district (Government of India, 2020).¹ This classification allows us to differentiate between high-, medium- and low-risk containment zones (Government of India, 2020).

On the one hand, stringent and organised measures helped reduce the spread of the virus and save human lives. On the other hand, the lockdown adversely affected various sectors of the economy, including the socio-economic lives of individuals. In the Indian context, little research has been done to examine the higher risk of gender-based criminal victimisation during the lockdown period. The crime opportunity theory (Cohen & Marcus, 1979) explains the association of crime incidences during periods of stringent lockdown. According to the theory, crime occurrences require the simultaneous intersection of probable offenders and vulnerable victims in space and time. Periods of lockdown reduced the opportunity for the potential perpetrators to target suitable victims and, hence, constrained the victimisation rate. Poblete-Cazenave (2020) examined the impact of lockdown on various criminal activities only for the state of Bihar, India. Our study attempts to understand the impact of the COVID-19 lockdown and other unprecedented institutional measures on crimes and violence against women using district–month level data in India.

There have been few studies on the subject (Peterman & O’Donnell, 2020). Our article contributes significantly to the existing literature as it explores the effect of lockdown on crime against women and explores variation in gender-based crime incidences across different COVID-19 zones. Moreover, we provide insights into the crimes committed against women inside the home versus outside and the variation during the lockdown period. We also provide a brief analysis of the comparison of gender-based crimes and gender-neutral crimes and the changes in the pre-post pandemic period. None of the papers have provided a comprehensive overview of crime across different categories in pre-, during and post-pandemic lockdown to the best of our knowledge.

Crime against women is a pressing social problem in developed and developing countries. According to the World Health Organization’s estimates, 40% of women in Africa and Southeast Asia face physical or sexual violence (World Health Organization, 2013). Nearly two in five currently married women aged 15 to 49 experience domestic or spousal violence, with slapping as the most common form of physical violence (Kishor & Gupta, 2009). The social and economic consequences of violence against women are huge, particularly in low-income countries that already face widespread poverty and inequality (Luca et al., 2015). In addition, living with the continued threat of violence has an adverse impact on the health and well-being of women (Garcia-Moreno et al., 2006; Kishor & Johnson, 2006). Violence against women is increasing globally, but, at the same time, there are recent concerns over the rising levels of social stresses coupled with the COVID-19 crisis (United Nations, 2020). Although the pandemic has affected both men and women due to widespread job losses, the fact is that women bear the risk of victimisation as they are stranded at home with their abusers (Deshpande, 2020a).

This article investigates the impact of the COVID-19 lockdown and other stringent institutional policies on crimes against women in India. We consider both intimate and non-intimate violence against women because every form of violence adversely affects their well-being and has negative social impacts. The victimisation data in India from 1 January to 3 May 2020 allows us to capture the distribution of crime in all COVID-19 affected districts, classified as red, orange and green zones. We use district-level data from the National Commission for Women (NCW), an apex organisation at the national level. NCW compiles gender-based crime data based on complaints about various offenses against women. The district-level crime data aggregated across all crime categories are available at NCW. To complement our district-level empirical analysis, we provide a graphical analysis of crime data disaggregated by the nature of the crime. NCW has its limitations; it provides disaggregated crime data only at the all-India level, and this disaggregation by crime categories is not available at the district level.

In this context, we study the aggregated crimes against women, including domestic violence, harassment, dowry,² torture, desertion, polygamy, rape, refusal to register First Information Report (FIR), deprivation, gender discrimination and sexual harassment at the workplace.³ Further, using data on different types of crime, we study crimes against women inside and outside the household in the pre- and during COVID lockdown periods. We expect a reduction in aggregate crime against women during this lockdown period due to strict policing measures in enforcing lockdowns. During institutional restrictions, fewer people went outside their homes, implying that women were less exposed to crimes. However, lockdowns may increase crimes inside the household due to the ongoing economic crisis, which leads to psychological stress and women being confined within the household with their perpetrators. With these mechanisms in mind, our article mainly focuses on three questions: First, has the institutional lockdown reduced overall crimes against women in India? Second, do we see a differential effect of the lockdown on crimes against women in various COVID-19-affected districts, classified by zones? Third, has the incidence of gender-based crime shifted from outside to inside the household due to the lockdown?

Results suggest that the stringent shutdown of the economy has resulted in lesser overall crimes against women across Indian districts. However, we see a zone-level variation in the incidence of crimes against women. Although crimes against women outside the household have decreased, we find that crimes inside the household have also fallen; however, the proportionate fall in inside crimes is lesser than the outside crimes. Therefore, we can argue that compared to inside crimes, the fall is more substantial in outside crimes, and inside crimes did not fall largely even in the presence of institutional restrictions. In the case of inside crime, we find a rise in domestic violence and cybercrimes due to the lockdown. Similar trends in domestic violence have been observed across different geographies due to COVID-19 restrictions (Boserup et al., 2020). Therefore, we find that the stringency of the lockdown has increased crimes against women inside the household in specific crime categories such as domestic violence and cybercrime. Therefore, we can argue that compared to inside crimes, the fall is more substantial in outside crimes, and inside crimes did not fall largely even in the presence of institutional restrictions. We find a reduction in other categories of inside crimes, which could also be possibly due to reporting issues during the lockdown. In what follows, we study the association between crime rates and the stringency with which the lockdown is imposed in respective zones.

Mechanisms Underlying the Prevalence of Crimes Against Women

We analyse the prevalence of crimes against women in pre-, during and post-lockdown periods at the district level in India. We extend our analysis for the trends in reported crimes for the different containment zones during the lockdown. The Indian government imposed the first phase of the lockdown from 24 March to 14 April to curb the spread of coronavirus across the country. We find that confirmed cases due to coronavirus doubled every six days from 24 March to 14 April.

Moreover, the Government of India adopted additional measures to prevent the spread of the coronavirus. One of the measures included categorising districts into red, orange and green zones based on the incidence of COVID-19 cases reported, the doubling rate of the incidence reported, the extent of testing, surveillance feedback and speed of recovery.⁴ In these three zones, an additional restriction was imposed based on the severity levels to minimise the spread of the virus. The government classified the type of activities that were allowed in each zone. For example, in red zones, private and public transport, barbershops, educational institutes, movie theatres and restaurants and bars, and religious and social gatherings were banned; all standalone shops and e-commerce for essential services were permitted to operate during the lockdown period of the analysis. Similarly, four-wheelers, including taxis and e-commerce for both essential and nonessential items, were allowed in orange zones, but bus services remained closed. In green zones, all services were approved except for the services, which were prohibited throughout the nation. Based on the date of declaration (30 April 2020), the law enforcement institution formed the three zones to break the chain of virus spread. The red containment zones were classified as severely affected districts with the most stringent restrictions. Orange zones were classified as relatively less affected. The green zones were defined as the least affected districts, with no reported cases in the last 21 days.

Figure 1(a) depicts crime incidences reported on average at the district level in red, orange and green zones. These average numbers are influenced by the presence of outlier districts (districts with a very high or low incidence of crimes). In absolute terms, we see that the fall in incidences of crime against women in the red zone is larger than the green zone and comparable to the orange zone, as shown in Figure 1(b). However, as shown in Figure 1(b), a proportionate change in the monthly average reported incidence of crimes suggests that the percentage fall in the incidence of crime is the largest in the orange zone followed by the green and red zones. We observe that the cases reported have declined by 47% from the pre- to during COVID period in the red zone. The orange and green zones have registered a decrease in criminal cases, 59% and 56%, respectively, during the same period. These numbers are consistent with the empirical findings, where we exploit the variation across districts to study the change in crime numbers in pre- and during lockdown rather than averaging the crime numbers across districts.

Figure 1(a).

Zone-Wise District Average of Crimes Against Women, JanuaryÐMay 2020

Figure 1(b).

Source: The authors.

We also briefly study the socio-economic characteristics of the districts classified on the basis of the following parameters: health and nutrition, education, agriculture, water resources, financial inclusion and skill development, and basic infrastructure (NITI Aayog, Government of India, 2019). In our sample, we find that 28%, 44% and 28% of COVID-affected developed districts are in the red, orange and green zones, respectively (see Table A.1). As can be seen, the red zone consists of districts with the largest population density and a higher percentage of the population living in urban areas. Zhang et al. (2020) have suggested that the virus spreads relatively faster in regions that are densely populated. One plausible reason for the greater numbers of coronavirus cases in the red containment zone is the relatively higher density of population in the districts classified as red than in the orange and green zones. In addition, the difference in average crime rates between the three zones may be explained by the differences in social, economic and cultural variables, such as attitudes towards women, the socio-economic status of women and the strength of patriarchal norms. In India, the sex ratio⁵ is an indicator of women’s socio-economic status (Barua et al., 2017). The sex ratio in the red zone, on average, is 940 women for 1,000 men. The sex ratio is 954 and 955 in the orange and green zones, respectively. Thus, the orange and green zones with better sex ratios are seen to have lower incidences of crimes against women.

India is historically a patriarchal country with strong gender biases and persistent discrimination against women. Women have consistently been relegated to a lower socio-economic status and face oppression owing to low literacy levels, weak social and institutional support, poor workforce participation and other contributing factors (Amaral & Bhalotra, 2017; Barua et al., 2017). During economic and social crises, women are more likely to face oppression and discrimination. This may have greater effects on crime rates, specifically for women and in general (Agarwal Goel, 2021). Therefore, the prevalence of crime is also associated with multiple crises that result from a forced lockdown. Massive economic disruption, unemployment, and an increased threat of poverty (Mahmud & Riley, 2021) are some underlying socio-economic problems that are expected to trigger violence against women during the lockdown period (Anderberg et al., 2016). Although men and women are affected by the pandemic and the associated economic collapse, cases of rising crimes against women are reported during times of unemployment (Deshpande, 2020b).

A more disaggregated crime data analysis at the all-India level in Figure 2(a) compares crime incidents during pre-lockdown months (January to March) with crime cases reported during the lockdown month (April) and the post-lockdown month (May). Given the limitations of our data, the disaggregated descriptive analysis is obtained for all of India. The graphical analysis in Figure 2(b) depicts the change in monthly average for crimes that took place inside homes (classified as inside crimes) and outside homes (classified as outside crimes) in pre-, and during and post-lockdown periods.⁶

We find that both inside and outside crimes declined more during the lockdown month than during the pre-lockdown months. However, the proportionate fall in outside crimes is much larger (56%) than inside crimes (27%) due to the imposition of restrictions. In Figure 2(a), we find that only two specific categories of inside crimes, domestic violence and cybercrimes against women increased during the lockdown period. Due to the institutional restrictions to constrain the spread of the virus, women and other household members spend much more time together at home, making them more vulnerable to crimes, such as domestic violence. It can be expected that due to increased stress and economic uncertainty, the lockdown provided an opportunity for the perpetrator to instigate domestic violence. In the household production framework, men seek to extract household work from their wives at lower prices by inducing intimate partner violence (IPV) (Grossbard, 2020), which may explain the fall in women’s economic value and a consequent rise in domestic violence at times of economic uncertainty and lower employment opportunities. Therefore, we find that the stringency of the lockdown has increased crimes against women inside the household in these two categories. However, we find a reduction during the lockdown in other categories of inside crimes such as ‘right to exercise choice in marriage’. This may be due to reporting issues. Crimes such as ‘cybercrimes’ do not have any social stigma attached to them, and ‘domestic violence’ may be associated with violence and severe injuries which are likely to be reported. Other forms of inside crimes may go unreported in events of high family pressure and non-violent forms of perpetration.

Figure 2(a).

Disaggregated Average Crimes Against Women in Pre-, During and Post-lockdown Months

The increased digitalisation of financial transactions may explain cybercrime’s rise. Sociolegal research shows that online crimes target women because the existing laws in the Indian judicial system are not adequate to ensure safety in cyberspace (Halder & Jainshankar, 2016). All other forms of crime have registered a decline due to the lockdown. In aggregate, we find that the overall crimes against women have declined during the lockdown. Using descriptive graphical analysis, we show that the incidence of gender-based crime has shifted from outside to inside the household due to the lockdown. In the post-lockdown month of May, there is a substantial increase in both inside and outside crimes. The numbers are found to be even greater than the pre-lockdown months. This substantiates the opportunity and motivation theory of crime. The uplift of lockdown provided greater opportunity to the perpetrators, and loss in employment and fall in social security benefits due to the pandemic acted as the motivation to higher crime rates.

Figure 2(b).

Source: The authors.

It is also important to get an overview of gender-based versus gender-neutral crimes and study the behaviour of two crime categories before and after the pandemic.

Annual data on different crime categories under the Indian Penal Code (IPC) is available with National Crime Records Bureau (NCRB). It is not possible to establish a causal relationship between the pre- and post-lockdown period since the data is not available across months, and the lockdown was levied only for the month of April. Figure 3 compares gender-based crimes with gender-neutral crimes between 2019 (pre-pandemic) and 2020 (pandemic). Data suggests that the incidence rate of gender-neutral crimes is higher than women-specific crimes.

Figure 3.

Source: The authors.

Data suggests that both gender-neutral and gender-based crimes have increased during 2020 compared to 2019. This could be explained by the loss in employment, weakening of the social security system, migration and rise in psychological stress due to the pandemic. However, gender-neutral crimes recorded a rise of approximately 47% on average as compared to 27% in the case of crime against women. Besides Assam, Jammu & Kashmir, West Bengal, Karnataka and most states have witnessed a fall in overall crime against women between 2019 and 2020. These states are high in women empowerment and literacy, an important determinant of incidence reporting. On the other hand, gender-neutral crimes have risen for all the states barring Punjab and Delhi. Data suggests that the pandemic caused a larger rise in crime in general compared to women-specific crimes. Thus, policies to address employment issues and social security measures are pertinent to curb the rise in crime incidences, both gender-specific and gender-neutral, at times of pandemic.

Data and Empirical Strategy

We use the following data sets to examine the impact of the lockdown and the stringency with which the restrictions were imposed on violence against women. District-level monthly data for crimes against women are taken from the National Commission of Women (NCW).⁷ The Complaints & Investigation Cell of the commission records oral, written or suo moto complaints under Section 10 of the NCW Act. During the national lockdown, complaints were filed through social media, emails and online registrations. We use data from January to April 2020 for the empirical analysis. We have extended our data to 31 May 2020 for a robustness analysis. A record of crime events that took place in similar months from the previous year, 2019, controls for unobserved factors in the district, such as people’s attitude about and tolerance of crime, cultural beliefs and attitudes towards gender parity, and women’s status. We collect district-level women’s police station⁸ data from ‘Data on Police Organizations’, 2016 report, Bureau of Police Research and Development, New Delhi.

In our data set, we consider April as the lockdown month. Between January and April 2020, nearly 50% of crimes against women were recorded in red zone districts and 34% and 16% of crimes were reported in orange and green zones, respectively. The red zone comprises 25% of the districts in India, and the orange and green zones consist of 43% and 32% of the districts, respectively. To evaluate the effect of the lockdown and additional administrative measures on crimes against women, we use district-level monthly data for 457 districts in India. The empirical model in the district-level analysis is as follows:

\begin{matrix} C r i m e_{m d} = β_{0}^{} + β^{}_{1} l o c k d o w n_{m d} + β_{2} O Z_{m d} + β_{3} G Z_{m d} + β_{4 (} l o c k d o w n^{*} O Z_{) m d} \\ + β_{} 5 (l o c k d o w n * G Z_{) m d} + β_{} 6 X ´_{m d} + θ_{d} + λ_{m} + ε_{m d}_{,} \end{matrix}

(1)

where m denotes months (from 1 January 2020 to 31 May 2020), d denotes districts (1 … 457), Crime_md is the number of crimes against women reported in month m in district d; lockdown_md dummy is equal to 1 in all districts for the month of April, otherwise zero; OZ_md is the orange zone dummy equal to 1 if the district belongs to the orange zone, otherwise zero; GZ_md is the green zone dummy equal to 1 if the district belongs to the green zone, otherwise zero; the red zone is the reference dummy variable⁹; (lockdown * OZ)_md is the interaction term between the lockdown dummy and the orange zone dummy; (lockdown * GZ)_md is the interaction term between the lockdown dummy and the green zone dummy, X´_md denotes the vector of control variables (crimes against women in 2019, women police station dummy [districts with women police station is equal to 1, otherwise zero] and no liquor ban dummy [states with no ban on liquor is equal to 1, otherwise zero]); θ_d control for unobserved district fixed effects, which are correlated with the explanatory and outcome variables in our model; λ_m denotes month fixed effects; and ε_md is the error term.

In Equation (1), the outcome variable is ‘the number of crimes against women’ reported in various districts of India in the respective months. The outcome is a count variable, and its variance is greater than the mean (mean = 2.36 and variance = 36.95), resulting in overdispersion. Moreover, reported incidences of crimes against women are zero for 40% of the month-district observations. Given the nature of the dependent variable, we estimate zero-inflated negative binomial (ZINB) and zero-inflated Poisson (ZIP) models.¹⁰ The ZINB estimation method allows overdispersion of the outcome variable and divides the total number of observations into zero and nonzero cases of crimes against women. Therefore, it is a combination of the logit (or Probit) model that accounts for the excessive number of zeros, and the negative binomial specification estimates the remaining counts. Akaike’s information criterion (AIC) and Bayesian information criterion (BIC) values are lower in the case of ZINB than ZIP estimates. Therefore, we conclude that the ZINB estimation method is appropriate for our data set.

We control unobserved time-invariant district fixed effects that are correlated with our dependent and other explanatory variables of interest, such as lockdown dummy, COVID-19 containment zone dummies and the interaction variables. For example, crimes against women in India are influenced by culture, religion, social norms, legal rights and geographical vulnerability. These unobserved factors do not change over time and can be considered as omitted variables in our model. Therefore, we eliminate the problem of omitted variables bias by including district dummies in the regression analyses. Moreover, Ravallion (2007) argues that inadequate control for unobserved time-invariant characteristics generates inconsistent and biased estimations.

In another model, we replace the district dummies with district-level observed indicators that might affect crimes against women. Sex ratio and literate population data are taken from the census of India, a decennial data collected by the Office of the Registrar General and Census Commissioner, India (ORGI). We calculate the annual population growth rates between 2001 and 2011. The growth figures are used to estimate the population (total, men and women, literate) in 2020. Urbanisation (percentage of urban population to total population) is included to control for the fact that the virus spreads faster in densely populated urban regions. We also include the Gini coefficient of inequality for urban areas because inequality increases criminal activities (Kelly, 2000).

The sex ratio is an important determinant of crimes against women, especially in countries like India, with a history of patriarchy, sex-selective abortions and the lower socio-economic status of women. An improvement in the sex ratio can be considered as an indicator of a better fallback position for women: higher bargaining power for women and stricter laws preventing crime against women (Amaral & Bhalotra, 2017; Barua et al., 2017; Cameron et al., 2019). Dreze and Khera (2000) find that criminal violence is strongly correlated with the sex ratio, implying that Indian districts with higher women-men ratios have lower criminal violence rates.

There is mixed evidence available on the effects of education and schooling on crime rates. On the one hand, a few studies suggest that schooling alters criminal behaviour and causes a decline in incarceration and arrests for crimes of a violent nature (Groot & van den Brink, 2010; Lochner & Moretti, 2004; Machin et al., 2011). These studies suggest that education reduces the likelihood of crime, and therefore, literacy acts as a constraining factor. On the other hand, a few studies show increased reporting and apprehension with enhanced literacy rates and representation, specifically for women (Iyer et al., 2012; Soares, 2004). Education increases the likelihood of reporting crimes. If the reporting effect of literacy is higher than the constraining effects, we would see a positive association between literacy rate and reported incidences of crime.

In addition to unobserved time-invariant characteristics, we control unobserved time-variant factors in our models that are common for all districts. For example, during the lockdown for COVID-19, the central government of India provided direct cash transfers and cooking gas cylinders free of cost to poor households (The Times of India, 2020). These unobserved time-variant factors influence the outcome variable and other variables of interest included in the regression models. Therefore, we include month dummies to control the unobserved time-variant effects in our models.

The fixed effects (FE) model addresses the omitted variable bias problem resulting from unobserved and constant factors. However, the FE model cannot control for other types of bias. First, the FE model produces biased results in the presence of unobserved time-varying socio-economic variables such as the district-level sex ratio, urbanisation, urban inequality, literacy and population density.¹¹ These socio-economic variables influence the outcome variable and are also correlated with the explanatory variables that are included in the regression models (Wooldridge, 2013, p. 512). Second, the FE model does not solve the endogeneity problem resulting from omitted variable bias in our models. Thus, in Table 2, we control the aforementioned socio-economic variables instead of district-fixed effects to address possible omitted variable bias in estimation.¹² We also estimate FE Negative Binomial (FE NB) and FE Poisson models, and the estimated results are consistent with ZINB and ZIP estimates. The FE Poisson model is used because it completely controls time-invariant unobserved district fixed effects (Wooldridge, 2002, pp. 674–676).

We present overall and zone-level summary statistics of all variables in Table A.2. The summary statistics reveal that, on average, three crimes against women were reported between 1 January 2020 and 31 May 2020. This implies that over the four months, on average, three crimes took place across all 457 districts in India. Crimes against women, on average, were the highest in the districts classified as red zone (6), followed by the orange zone districts (2) and the green zone districts (1). The green zone witnessed the lowest crime incidences on average among the zones during the same period. Concerning socio-economic variables, we also find the red zone has the largest population density, higher urbanisation (percentage of people living in urban regions), relatively higher inequality and a lower sex ratio.

Discussion of Empirical Results

In this section, we discuss the empirical estimates of the association between reported incidences of crimes against women and the period during which India was under a complete lockdown to constrain the spread of the COVID-19 virus. The results also discuss the differential impact of the lockdown in the red, orange and green zones. We compute the marginal effects at the mean (MEM) and incidence rate ratio (IRR)¹³ derived from ZINB and ZIP estimates in columns 1 through 4 in Table 1. Columns 5 through 8 show the MEM and IRR from the FE NB and FE Poisson models.

Table 1.

Lockdown, COVID-19 Zones and Crime Against Women

Variables	Zero-Inflated Negative Binomial (ZINB) Estimates		Zero-Inflated Poisson (ZIP) Estimates		FE NB Estimate		FE Poisson Estimate
	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)
	C1	C2	C3	C4	C5	C6	C7	C8
Lockdown dummy (Lockdown in April month is equal to 1, otherwise zero)	Ð0.327** (0.149)	0.718** (0.108)	Ð0.445*** (0.161)	0.635*** (0.105)	Ð0.329*** (0.146)	0.671*** (0.100)	Ð0.526*** (0.158)	0.582*** (0.095)
COVID-19 orange zone dummy (Orange zone is equal to 1, otherwise zero)	Ð0.479*** (0.034)	0.616*** (0.020)	Ð0.479*** (0.031)	0.613*** (0.019)	Ð0.489*** (0.032)	0.609*** (0.019)	Ð0.484*** (0.029)	0.608*** (0.018)
COVID-19 green zone dummy (Green zone is equal to 1, otherwise zero)	Ð0.190*** (0.034)	0.825*** (0.028)	Ð0.202*** (0.035)	0.813*** (0.029)	Ð0.203*** (0.032)	0.813*** (0.026)	Ð0.217*** (0.032)	0.800*** (0.026)
Lockdown dummy * COVID-19 orange zone dummy	Ð0.352* (0.189)	0.700* (0.134)	Ð0.326* (0.194)	0.716* (0.142)	Ð0.331* (0.184)	0.714* (0.134)	Ð0.321* (0.189)	0.719* (0.140)
Lockdown dummy * COVID-19 green zone dummy	Ð0.379* (0.209)	0.681* (0.144)	Ð0.328 (0.221)	0.715 (0.161)	Ð0.322 (0.203)	0.721 (0.149)	Ð0.280 (0.211)	0.750 (0.162)
Women police station dummy (Districts have women police station is equal to 1, otherwise zero)	Ð0.538*** (0.008)	0.580*** (0.005)	Ð0.525*** (0.008)	0.585*** (0.004)	Ð0.552*** (0.005)	0.571*** (0.002)	Ð0.539*** (0.004)	0.575*** (0.001)
Crime against women in 2019	0.028*** (0.004)	1.029*** (0.004)	0.022*** (0.003)	1.023*** (0.004)	0.029*** (0.004)	1.030*** (0.005)	0.022*** (0.004)	1.023*** (0.004)
No liquor ban dummy (States have no ban on liquor is equal to 1, otherwise zero)	0.124*** (0.006)	1.133*** (0.007)	0.139*** (0.004)	1.152*** (0.003)	0.113*** (0.005)	1.121*** (0.005)	0.132*** (0.001)	1.146*** (0.001)
No. of districts	463	463	463	463	463	463	463	463
No. of months (January 2020 to April 2020)	4	4	4	4	4	4	4	4
Observations	1,852	1,852	1,852	1,852	1,852	1,852	1,852	1,852
District FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Month FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Source: The authors.

Notes: Clustered standard errors at district-level are reported in parentheses. *** p < .01, ** p < .05 and * p < .1 denote 1%, 5% and 10% levels of significance, respectively. The dependent variable is the number of crimes against women reported from January 2020 to April 2020. The COVID-19 red zone is the base line dummy variable. FE = Fixed effect; NB = negative binomial.

We find that the reported number of crimes against women significantly declined during the lockdown on average. The MEM in column 1 suggest that the probability of crimes is 32.7% less at the mean during the lockdown than during the non-lockdown months (January to March 2020). The IRR for the ZINB estimate shows that the overall crime rate during the lockdown is 0.72 units lower than the crime rate in the non-lockdown period. We argue that the lockdown reduces the exposure of women outside the boundaries of the household and reduces the risk of victimisation. Our results related to lockdown are consistent with the stay-at-home order for the United States (USA), where fewer people outside their homes for a shorter period implies that individuals are less exposed to crimes during institutional restrictions due to COVID-19 (McDonald & Balkin, 2020).

We also exploit the variation in the stringency of restrictions imposed in different districts of India. As discussed in the section titled, ‘Mechanisms Underlying the Prevalence of Crimes Against Women’, the red zone is the most severely affected by the COVID-19 virus in terms of the total number of cases and doubling rate and, therefore, has the most stringent restrictions. Compared to the red zone, the orange zone includes districts with fewer restrictions and the green zone districts have minimum institutional restrictions.

The MEM from the ZINB estimation shows that the probabilities of crime in the orange and green zones are 47.9% and 19%, respectively, lower than in the red zone at the sample means. Moreover, crime reduction due to the COVID-19 lockdown is significantly higher in the orange and green zones. Compared to the red zone, the probability of crime in the orange zone is 35.2% less, and 37.9% less in the green zone, at the sample mean, during the lockdown. Therefore, the interaction effects suggest that the orange and green zones have witnessed lower crime rates than the red zone. We find directionally similar results for the ZIP model.

The results suggest that government restrictions on the mobility of individuals are associated with a lower rate of crimes against women,¹⁴ and the lockdown effects are stronger in the orange and green zones. The findings are in line with the numbers in Figure 1(b) that depict a relatively larger decline in reported crime cases in the orange and green zones during the lockdown period. It may be argued that the decrease in crimes may be due to a decrease in reporting due to restrictions imposed during the lockdown. We do not completely deny the presence of under-reporting bias, which has always been an issue of concern because of social stigma and societal pressure associated with women’s reporting of crimes. However, we perform different specifications such as FE Negative Binomial and FE Poisson in the following section to establish the robustness of our estimates.¹⁵

Our model controls for the presence of women’s police station in the district. The presence of women’s police stations in the literature is associated with increased reporting of crimes against women (Amaral et al., 2018; Iyer et al., 2012). Our results are consistent with the literature, and the MEM from ZINB estimation suggests that the probability of crimes against women declines by 53.8% points due to the presence of women’s police stations. We also control for the crime incidents that happened in 2019, and estimated results suggest that the incidence of crime in 2019 for similar months is associated with a higher rate of crime reporting in the current year (2020). A plausible reason is that the districts with historically higher intolerance of crimes against women are likely to report higher crime rates in the future. For instance, Thurman et al. (2003) explore community norms to address violence against women as a community problem in Native American communities in the western USA. They argue that historical issues are responsible for creating culturally appropriate strategies to prevent violence against women.

It is also important to note that the states of India that did not implement bans on liquor sales¹⁶ have experienced a 12.4% higher probability of crimes against women than the states with the complete prohibition of alcohol. The estimates obtained from the ZIP model in columns 3 and 4 in Table 1 are directionally consistent with our findings from the ZINB model and provide robustness checks for the estimates. The existing literature suggests that liquor consumption can lead to violent behaviour and abuse, especially towards women and children (Lucas et al., 2015). Most liquor shops were indeed closed during lockdowns, which temporarily banned the buying and selling of alcohol. However, social and cultural norms around the use of alcohol can influence crimes against women. For instance, societies with a higher tolerance for alcohol consumption show a stronger association between alcohol use and violence than societies with moderate use of alcohol (Rossow, 2001).

We derive our main results using ZINB and ZIP Poisson models that deal with the presence of a high proportion of ‘zeros’ for the crime variable because crime is not reported in these districts. We, however, run robustness tests for our estimates with FE NB and FE Poisson models. As we see in the columns 5 and 6 of Table 1, the estimates confirm the findings from the ZINB and ZIP models. The estimates are not only directionally consistent but are also highly significant. Moreover, we witness a marginal difference in coefficients, suggesting that the results are robust to different estimation strategies.

In Table 2, we replace district fixed effects with district-specific socio-economic factors such as sex ratio, literate population, population density, urbanisation and inequality measured by urban Gini to address possible omitted variable bias in estimation. We control state-fixed effects and cluster the standard errors at the state level in all regression models. Although the lockdown period, on average, lowered the incidence of crimes, the coefficient is statistically insignificant. The plausible reason is that although the overall occurrence of crime against women declined during the lockdown period, inside crime rates were proportionately higher during the same period (see Figure 2(b)). The institutional restrictions on mobility during the lockdown reduced female victimisation and crime by minimising exposure. Therefore, women were kept safe from perpetrators outside their households and from crimes such as rape, kidnapping, sexual assault at work, and so on.

Controlling for district-level socio-economic factors and state fixed effects, we find that the orange and green containment zones are estimated to have a lower incidence of crimes against women than the red zone. The interaction effects of the lockdown period with the orange and green zones remain negative and significant. The results are mostly directionally significant, with minor variations in estimates compared to the results presented in Table 1. A brief discussion on the state-level controls shows that a higher sex ratio is associated with lower crime rates.

Table 2.

Lockdown, COVID-19 Zones, Sex-Ratio and Crime Against Women

Variables	Zero-Inflated NB (ZINB) Estimates			Zero-Inflated Poisson (ZIP) Estimates		FE NB Estimate		FE Poisson Estimate
	Marginal Effect at Mean (MEM	Incidence Rate Ratio (IRR)		Marginal Effect at Mean (MEM	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incident Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incident Rate Ratio (IRR)
	C1	C2		C3	C4	C5	C6	C7	C8
Lockdown dummy (Lockdown in April month is equal to 1, otherwise zero)	Ð0.220 (0.453)	0.849 (0.280)		Ð0.638 (0.579)	0.644 (0.255)	Ð0.192 (0.389)	0.859 (0.262)	Ð0.704 (0.552)	0.588 (0.240)
Covid_19 orange zone dummy (Orange zone is equal to 1, otherwise zero)	Ð0.147 (0.204)	0.896 (0.138)		Ð0.232* (0.119)	0.852* (0.072)	Ð0.164 (0.184)	0.878 (0.127)	Ð0.276* (0.146)	0.812* (0.091)
COVID-19 green zone dummy (Green zone is equal to 1, otherwise zero)	Ð0.233 (0.165)	0.841 (0.107)		Ð0.497*** (0.104)	0.710*** (0.047)	Ð0.259* (0.140)	0.815* (0.089)	Ð0.507*** (0.108)	0.681*** (0.055)
Lockdown dummy * COVID-19 orange zone dummy	Ð0.563* (0.327)	0.659* (0.155)		Ð0.615** (0.310)	0.654** (0.136)	Ð0.507* (0.289)	0.670* (0.154)	Ð0.437* (0.248)	0.719* (0.136)
Lockdown dummy * COVID-19 green zone dummy	Ð0.582** (0.295)	0.650** (0.140)		Ð0.501 (0.319)	0.708 (0.153)	Ð0.549** (0.271)	0.648** (0.140)	Ð0.392 (0.254)	0.744 (0.144)
Women police station dummy (Districts has women police station is equal to 1, otherwise zero)	Ð0.104 (0.096)	0.925 (0.068)		0.041 (0.191)	1.028 (0.134)	Ð0.109 (0.088)	0.917 (0.064)	Ð0.039 (0.178)	0.970 (0.130)
Crime against women in 2019	0.072*** (0.020)	1.055*** (0.018)		0.125*** (0.024)	1.032*** (0.005)	0.072*** (0.022)	1.058*** (0.018)	0.042*** (0.007)	1.032*** (0.005)
No liquor ban dummy (States have no ban in liquor is equal to 1, otherwise zero)	1.287*** (0.108)		2.590*** (0.241)	1.686*** (0.314)	3.187*** (0.722)	1.254*** (0.100)	2.685*** (0.236)	1.545*** (0.281)	3.204*** (0.758)
ln (Sex ratio)	Ð6.08*** (1.105)	0.011*** (0.008)		Ð8.229*** (2.468)	0.003*** (0.005)	Ð5.660*** (0.982)	0.011*** (0.009)	Ð7.998*** (1.978)	0.002*** (0.003)
ln (Literate population)	0.181* (0.095)	1.143* (0.085)		Ð0.018 (0.199)	0.987 (0.135)	0.201** (0.089)	1.172** (0.084)	0.051 (0.195)	1.039 (0.153)
ln (Population density)	0.167** (0.067)	1.132** (0.058)		0.117 (0.079)	1.084 (0.062)	0.131** (0.066)	1.109** (0.058)	0.083 (0.074)	1.046 (0.060)
Urbanisation	0.016*** (0.003)	1.012*** (0.002)		0.013*** (0.003)	1.009*** (0.002)	0.016*** (0.003)	1.012*** (0.002)	0.015*** (0.002)	1.011*** (0.002)
Urban Gini	1.210 (0.817)	2.448 (1.509)		0.445 (0.681)	1.358 (0.641)	1.097 (0.815)	2.373 (1.517)	0.368 (0.689)	1.319 (0.688)
No. of districts	457	457		457	457	457	457	457	457
No. of months (January 2020 to April 2020)	4	4		4	4	4	4	4	4
Observations	1,828	1,828		1,828	1,828	1,828	1,828	1,828	1,828
No. of states	22	22		22	22	22	22	22	22
State FE	Yes	Yes		Yes	Yes	Yes	Yes	Yes	Yes
Month FE	Yes	Yes		Yes	Yes	Yes	Yes	Yes	Yes

Source: The authors.

Notes: Clustered standard errors at state-level are reported in parentheses. *** p < .01, ** p <.05 and * p < .1 denote 1%, 5% and 10% levels of significance, respectively. The dependent variable is the number of crimes against women is reported from January 2020 to April 2020. The COVID-19 red zone is the base line dummy. We control district-wise sex ratio, literate population, population density, urbanisation and urban and Lorenz ratio (2004Ð2005) instead of district fixed effect in all models.

The high female–male ratio indicates a better socio-economic representation of women, greater intra-household bargaining, higher reporting and enhanced human capital investment in children, particularly girls (Afridi et al., 2016; Iyer et al., 2012; Sudha & Ranjan 2003). The existence of such gender-neutral norms constrains criminal engagements in societies. The literate percentage in districts is seen to be correlated with higher crime rates on average, although the estimate is statistically insignificant. Literacy increases the reporting of crime and may thus explain the positive association between the two. An increase in population density at the mean raises the probability of female victimisation by 0.167 additional crimes, as estimated by the ZINB model. Greater urbanisation is also found to be associated with greater crimes against women.

The literature argues that urbanisation and high population density offer economies of scale in the form of specialisation of markets and labour but also cause huge diseconomies with the occurrence of crimes. Rural–urban migration, large inequalities in income levels, relatively larger gains from burglary are some determinants of higher crime rates in urban areas (Gibbons, 2004; Pressman & Carol, 1971; Shichor et al., 1979). Our findings suggest that higher inequality in urban income is positively associated with crime, although insignificantly. The literature reports similar findings where the greater disparity in income distribution is attributed to higher economic aspirations and social stress, which encourage criminal engagements (Brush, 2007; Choe, 2008).

Studies suggest that crimes against women are often subject to under-reporting due to social stigma and lack of social and institutional support associated with crimes such as rape, molestation, domestic violence and crimes of a similar nature (Chattopadhyay, 2016). Most physical or sexual violence in India occurs in marriage; however, only 10% of married women report sexual violence perpetrated by their husbands (Raj & McDougal, 2014). Women have difficulty reporting violence for several reasons, such as economic dependency on the perpetrator, lack of social support, religious and cultural norms, and lack of legal information about their rights (United Nations, 2010).

NCW highlights the under-reporting of domestic violence directed towards women during the COVID-19 lockdown period. One explanation for the under-reporting of crimes can be attributed to the constant fear of violence and denial of family resources in a situation when women cannot move out to seek assistance from external sources. At the same time, judges, police, and health service workers are overwhelmed during the lockdown period and are unable to protect the victims of domestic violence (United Nations, 2020).

The cost of under-reporting is huge because it reduces the probability of detection by preventing law-enforcement officials from gathering useful information about the crime (Allen, 2007). In the absence of essential information about the particular types of crime, it is difficult to capture an accurate estimate of violence against women. The NCW has pointed out the possibility of under-reporting crimes during the lockdown period. We run some robustness checks on our data to address this potential measurement problem. We argue that a district’s development index comprised of social, economic, and institutional support measures such as accessibility to NGOs, women’s police stations and the availability of technological infrastructures such as Internet connectivity and mobile phones should address under-reporting to a great extent. We exclude the 73 most underdeveloped Indian districts as classified by (NITI Aayog, Government of India, 2019) to perform a robustness check on our main estimates with only the developed districts¹⁷ (see Table 3). The estimates are in line with the main results presented in Table 1. We do not deny the complete absence of under-reporting; however, the robustness test establishes the reliability of the statistical findings from the data.

Table 3.

Lockdown, COVID-19 Zones and Crime Against Women in Developed Districts (Marginal Effects at Mean)

Variables	Zero-Inflated		Fixed Effect
	Negative Binomial Estimate	Poisson Estimate	Negative Binomial Estimate	Poisson Estimate
	C1	C2	C3	C4
Lockdown dummy (Lockdown in April month is equal to 1, otherwise zero)	Ð0.372** (0.162)	Ð0.503*** (0.177)	Ð0.441*** (0.160)	Ð0.589*** (0.174)
COVID-19 orange zone dummy (Orange zone is equal to 1, otherwise zero)	Ð0.051 (0.039)	Ð0.064* (0.036)	Ð0.076** (0.037)	Ð0.099*** (0.031)
COVID-19 green zone dummy (Green zone is equal to 1, otherwise zero)	Ð0.194*** (0.038)	Ð0.209*** (0.039)	Ð0.208*** (0.037)	Ð0.226*** (0.036)
Lockdown dummy * COVID-19 orange zone dummy	Ð0.440** (0.211)	Ð0.404* (0.216)	Ð0.415** (0.206)	Ð0.398* (0.211)
Lockdown dummy * COVID-19 green zone dummy	Ð0.509** (0.250)	Ð0.454* (0.260)	Ð0.448* (0.244)	Ð0.402* (0.248)
Women police station dummy (Districts has women police station is equal to 1, otherwise zero)	Ð0.589*** (0.009)	Ð0.574*** (0.260)	Ð0.605*** (0.006)	Ð0.589*** (0.005)
Crime against women in 2019	0.031*** (0.005)	0.024*** (0.004)	0.033*** (0.005)	0.024*** (0.004)
No liquor ban dummy (States have no ban in liquor is equal to 1, otherwise zero)	0.598*** (0.015)	0.605*** (0.017)	0.572*** (0.006)	0.569*** (0.006)
No. of districts	390	390	390	390
No. of months (January 2020 to April 2020)	4	4	4	4
Observations	1,560	1,560	1,560	1,560
District FE	Yes	Yes	Yes	Yes
Month FE	Yes	Yes	Yes	Yes

Source: The authors.

Notes: Cluster standard errors at district-level are reported in parentheses. *** p < .01, ** p < .05 and * p < .1 denote 1%, 5% and 10% levels of significance, respectively. The dependent variable is the number of crimes against women is reported form January 2020 to April 2020. The COVID-19 red zone is the base line dummy variable. In this regression, we have excluded 73 backward districts to minimise the underreporting for the crime against women.

We have extended our data set to 31 May 2020 and re-estimated the zero-Inflated negative binomial. The estimates results are shown in Table 4 and are found to be directionally consistent. The coefficient of the lockdown dummy is negative though insignificant, which confirms that during the lockdown, crime against women has declined. Moreover, in the orange zone, crime against women witnesses a larger decline; in the green zone, crime against women falls significantly with a greater magnitude than in the red zone (see columns C1 and C3). Furthermore, the interaction coefficient between lockdown and orange zone is negative and significant. The results confirm that the probability of crime against women in the orange zone significantly declined due to lockdown, but the interaction coefficient between lockdown and green zone is negative and insignificant. This implies a limited decline in crime against women compared to red and orange zones.

Table 4.

Lockdown, COVID-19 Zones, Sex Ratio and Crime Against Women

Variables	Zero-Inflated NB (ZINB) Estimates		Zero-Inflated Poisson (ZIP) Estimates		Fixed Effect NB Estimate		Fixed Effect Poisson Estimate
	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)	Marginal Effect at Mean (MEM)	Incidence Rate Ratio (IRR)
	C1	C2	C3	C4	C5	C6	C7	C8
Lockdown dummy (Lockdown in April month is equal to 1, otherwise zero)	Ð0.386 (0.509)	0.764 (0.263)	Ð0.852 (0.645)	0.583 (0.232)	Ð0.335 (0.433)	0.778 (0.251)	Ð0.858 (0.572)	0.540 (0.216)
COVID-19 orange zone dummy (Orange zone is equal to 1, otherwise zero)	Ð0.196 (0.195)	0.872 (0.122)	Ð0.214 (0.175)	0.873 (0.009)	Ð0.207 (0.175)	0.856 (0.112)	Ð0.252 (0.187)	0.834 (0.114)
COVID-19 green zone dummy (Green zone is equal to 1, otherwise zero)	Ð0.384*** (0.159)	0.756*** (0.090)	Ð0.574*** (0.137)	0.695*** (0.061)	Ð0.392*** (0.141)	0.745** (0.078)	Ð0.572*** (0.145)	0.663*** (0.070)
Lockdown dummy * COVID-19 orange zone dummy	Ð0.504* (0.303)	0.704* (0.144)	Ð0.594* (0.303)	0.687** (0.128)	Ð0.448* (0.265)	0.714* (0.142)	Ð0.380* (0.231)	0.761* (0.126)
Lockdown dummy* COVID-19 green zone dummy	Ð0.394 (0.293)	0.760 (0.153)	Ð0.391 (0.312)	0.781 (0.151)	Ð0.377 (0.270)	0.754 (0.153)	Ð0.283 (0.250)	0.815 (0.146)
Women police station dummy (Districts has women police station is equal to 1, otherwise zero)	Ð0.035 (0.107)	0.975 (0.073)	0.092 (0.226)	1.060 (0.148)	Ð0.039 (0.097)	0.971 (0.070)	0.001 (0.178)	1.000 (0.141)
Crime against women in 2019	0.059*** (0.021)	1.042*** (0.017)	0.125*** (0.024)	1.020*** (0.003)	0.059** (0.023)	1.045** (0.018)	0.027*** (0.004)	1.019*** (0.003)
No liquor ban dummy (States have no ban in liquor is equal to 1, otherwise zero)	1.363*** (0.110)	2.577*** (0.238)	1.812*** (0.357)	3.137*** (0.764)	1.338*** (0.100)	2.720*** (0.228)	1.545*** (0.281)	3.341*** (0.831)
ln (Sex ratio)	Ð6.399*** (1.167)	0.011*** (0.009)	Ð8.638*** (2.707)	0.004*** (0.007)	Ð5.968*** (1.039)	0.011*** (0.009)	Ð8.432*** (2.110)	0.002*** (0.003)
ln (Literate population)	0.236*** (0.083)	1.178* (0.069)	0.078 (0.199)	1.051 (0.133)	0.253*** (0.077)	1.208** (0.072)	0.142 (0.194)	1.107 (0.156)
ln (Population density)	0.150** (0.076)	1.110* (0.069)	0.074 (0.085)	1.047 (0.058)	0.121* (0.074)	1.095* (0.060)	0.056 (0.074)	1.041 (0.056)
Urbanisation	0.019*** (0.003)	1.013*** (0.002)	0.020*** (0.002)	1.013*** (0.002)	0.018*** (0.003)	1.014*** (0.002)	0.021*** (0.002)	1.015*** (0.002)
Urban and Lorenz ratio	1.672** (0.770)	3.192** (1.748)	0.789 (0.661)	1.645 (0.700)	1.567** (0.761)	3.228** (1.831)	0.813 (0.674)	1.792 (0.841)
No. of districts	457	457	457	457	457	457	457	457
No. of months (1 January 2020 to 31 May 2020)	5	5	5	5	5	5	5	5
Observations	2,285	2,285	2,285	2,285	2,285	2,285	2,285	2,285
No. of states	22	22	22	22	22	22	22	22
State FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes
Month FE	Yes	Yes	Yes	Yes	Yes	Yes	Yes	Yes

Source: The authors.

Notes: Clustered standard errors at state-level are reported in parentheses. *** p < .01, ** p < .05 and * p < .1 denote 1%, 5% and 10% levels of significance, respectively. The dependent variable is the number of crimes against women is reported from 1 January 2020 to 31 May 2020. The COVID-19 red zone is the base line dummy. We control district-wise sex ratio, literate population, population density, urbanisation and urban and Lorenz ratio (2004Ð2005) instead of district fixed effect in all models.

The presence of women police stations in districts negatively impacts crime against women and the states that do not have liquor ban witness higher incidences of crime against women. Moreover, the state with higher population density, urbanisation and urban inequality experienced higher crime against women, while a better sex ratio can cause a decline in crime against women. Our estimates in alternative specifications produced mirror results in different econometric specifications (see columns C5–C8).

Overall, our findings suggest that the imposition of mobility restrictions is associated with a decline in reported cases of average crimes against women. The lockdown effect in constraining gendered violence is stronger in districts classified under the orange and green zones than in the red zone. Statistics show that the green and orange zones, on average, have fewer reported cases of crimes against women. A brief analysis of the three zones’ socio-economic characteristics suggests that the green and orange zones have a higher sex ratio, lower population density and less urbanisation, and display lower inequality than the red zones (see the section titled, ‘Mechanisms Underlying the Prevalence of Crimes Against Women’). These factors that might explain the motivation to commit crimes might also explain the larger marginal effect of the lockdown in the orange and green zones than in the red zone, even though the restrictions were less severe. As already described, we control these factors in our empirical analysis to address any potential omitted variable bias. However, we do not study these factors in detail since they are not within the scope of this article.

Conclusions and Policy Implications

The main objective of our study is to quantify the effect of the lockdown and other socio-economic factors on aggregated crimes against women in India. We also study if the lockdown has a differential impact on gender-based crimes in the three zones. Using crime data and COVID-19 cases reported by several districts, we find that the lockdown, on average, has led to a decrease in crimes against women; however, the effect has been stronger in the orange and green containment zones than in the red zone. The lower rates of crimes against women can be attributed to early lockdown measures and strict policing activities. Besides, on the one hand, our results suggest that the COVID-19 lockdown and other socio-economic factors, such as a higher sex ratio, the presence of women’s police stations and alcohol prohibition, also contribute to reducing gender-based crimes. On the other hand, urbanisation, population density and urban inequality are positively associated with crimes against women.

It is clear that although crime activities outside the household have decreased due to institutional influence during the lockdown, domestic violence and cybercrimes against women persist (see Figures 2(a) and 2(b)). We suggest that structural changes such as confinement at home induced by the forced COVID-19 lockdown create circumstances in which women fight the virus and domestic violence. We do not argue that lockdown enforcement is an effective way to reduce gender-based violence. Of course, the institutional restrictions can lower the incidences of rape, sexual assault and other kinds of harassment that women generally face outside their household spheres. However, restricted mobility, coupled with fewer economic opportunities during a crisis period, can weaken a woman’s bargaining strength and expose her to further oppression and victimisation.

The state and NGO interventions are crucial to identify ways to prevent all forms of crimes against women during crisis periods and ensure social protection. For instance, the Odisha State Commission for Women in India created a WhatsApp helpline to address the issues related to different types of violence against women during the lockdown period.¹⁸ NCW had also set up an additional WhatsApp service to help women report cases of violence against them (Ravindran & Shah, 2020). However, women’s financial independence is limited in India, and all females do not have access to mobile phones and have lower technological literacy. This can create an impediment and result in lower crime reporting through such online services. Moreover, during the lockdown, since females are co-inhabiting with their perpetrators and being continuously monitored, it may create hindrances to incidence reporting.

In South India, Kerala started women-run community kitchens to prepare and deliver food regularly. Women’s involvement in community-level projects ensures a well-coordinated approach to control the crisis in a densely populated region. Thus, women working together in community projects promote community resilience and reduce isolation during COVID times, thereby enhancing social protection. The existence of social protection can be expected to significantly reduce the amount of crime and violence against women in India. Recently, policymakers have started prioritising a general improvement in women’s status and economic empowerment via cash transfers to women’s accounts.¹⁹ All these interventionist policies are significant steps to address the challenges women face, but, in the long run, more stringent actions and changes in society’s view of women are needed to eradicate the problem of crimes against women.

Footnotes

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship and/or publication of this article.

Funding

The authors received no financial support for the research, authorship and/or publication of this article.

Appendix

Notes

References

Agarwal Goel

(2021). Effects of female workforce participation on crime: An analysis from the districts of India. Journal of Public Affairs, 21(1), e2165.

Afridi

, Mukhopadhyay

, & Sahoo

(2016). Female labor force participation and child education in India: evidence from the National Rural Employment Guarantee Scheme. IZA Journal of Labor & Development, 5(1), 1–27.

Allen

W. D.

(2007). The reporting and underreporting of rape. Southern Economic Journal, 73, 623–641.

Amaral

, & Bhalotra

(2017). Population sex ratios and violence against women: The long-run effects of sex selection in India. SSRN 3055794. 10.2139/ssrn.3055794

Amaral

, Prakash

, & Soni

(2018). Gender, crime and punishment: Evidence from women police stations in India. NEUDC, Cornell University. Retrieved from http://barrett.Dyson.Cornell.Edu/NEUDC/paper_32.Pdf

Anderberg

, Helmut

, Jonathan

, & Tanya

(2016). Unemployment and domestic violence: Theory and evidence. The Economic Journal, 126(597), 1947–1979.

Barua

, Goel

, & Sane

(2017). The effect of age-specific sex ratios on crime: Instrumental variable estimates from India (No. 17/214). https://www.nipfp.org.in/media/medialibrary/2017/11/WP_2017_214.pdf

Boserup

, Mark

, & Adel

(2020). The financial strain placed on America’s hospitals in the wake of the COVID-19 pandemic. The American Journal of Emergency Medicine, 45, 530–531.

Brush

(2007). Does income inequality lead to more crime? A comparison of cross-sectional and time-series analyses of United States counties. Economics Letters, 96(2), 264-268.

10.

Cameron

, Xin

, & Dandan

(2019). China’s sex ratio and crime: Behavioural change or financial necessity? The Economic Journal, 129(618), 790–820.

11.

Chattopadhyay

(2016). Gendering legal discourse: A critical feminist analysis of domestic violence adjudication in India. Law, Social Justice and Global Development Journal, 20(2), 1–17.

12.

Choe

(2008). Income inequality and crime in the United States. Economics Letters, 101(1), 31–33.

13.

Cohen

L. E.

, & Marcus

(1979). Social change and crime rate trends: A routine activity approach. American Sociological Review, 44, 588–608.

14.

Deshpande

(2020a, March 27). Protecting women is missing from pandemic management measures in India. Quartz India. https://qz.com/india/1826683/indias-approach-to-fighting-coronavirus-lacks-a-gender-lens/

15.

Deshpande

(2020b, April 16). In locked down India, women fight coronavirus and domestic violence. Quartz India. https://qz.com/india/1838351/indias-coronavirus-lockdown-leads-to-more-violence-against-women/

16.

Dreze

, & Khera

(2000). Crime, gender, and society in India: Insights from homicide data. Population and Development Review, 26(2), 335–352.

17.

Garcia-Moreno

, Jansen

H. A.

, Ellsberg

, Heise

, & Watts

C. H.

(2006). Prevalence of intimate partner violence: Findings from the WHO multi-country study on women’s health and domestic violence. The Lancet, 368(9543), 1260–1269.

18.

Gibbons

(2004). The costs of urban property crime. The Economic Journal, 114(499), F441–F463.

19.

Goel

P. A.

, & Barua

(2021). Female education, marital assortative mating, and dowry: Theory and evidence from districts of India. Journal of Demographic Economics, 1–27. https://www.cambridge.org/core/journals/journal-of-demographic-economics/article/abs/female-education-marital-assortative-mating-and-dowry-theory-and-evidence-from-districts-of-india/74A26F258A5CFE53770B282678DABE3A

20.

Government of India. (2020). Notification letter no. D, O. No. Z. 280 1 5/19/2020-EM R 30-4-2020, Department of Health and Family Welfare, Ministry of Health and Family Welfare.

21.

Groot

, & Van den Brink

H. M.

(2010). The effects of education on crime. Applied Economics, 42(3), 279–289.

22.

Grossbard

(2020). Modeling intimate partner violence with home production, marriage markets and male domination (Working paper). University of Chicago. https://static1.squarespace.com/static/59af966b49fc2bd0f750a72a/t/5e8a05b71b981f2a96bc829f/1586103736549/IPV+by+sg+march+31+w+refs+spaced.pdf

23.

Halder

, & Jaishankar

(2016). Cyber crimes against women in India. SAGE Publications.

24.

Hilbe

J. M.

, & Greene

W.H.

(2007). Count response regression models. In Rao

C. R.

, Miller

J. P.

, & Rao

D. C.

(Eds), Epidemiology and Medical Statistics, Elsevier Handbook of Statistics Series (pp. 210–252). Elsevier.

25.

Iyer

, Mani

, Mishra

, & Topalova

(2012). The power of political voice: Women’s political representation and crime in India. American Economic Journal: Applied Economics, 4(4), 165–193.

26.

Kelly

(2000). Inequality and crime. The Review of Economics and Statistics, 82(4), 530–539.

27.

Kishor

, & Gupta

(2009). Gender equality and women’s empowerment in India: National Family Health Survey (NFHS-3), India, 2005–06. International Institute for Population Sciences.

28.

Kishor

, & Johnson

(2006). Reproductive health and domestic violence: Are the poorest women uniquely disadvantaged? Demography, 43(2), 293–397.

29.

Lochner

, & Moretti

(2004). The effect of education on crime: Evidence from prison inmates, arrests, and self-reports. American Economic Review, 94(1), 155–189.

30.

Luca

D. L.

, Owens

, & Sharma

(2015). Can alcohol prohibition reduce violence against women? American Economic Review, 105(5), 625–629.

31.

Machin

, Marie

, & Vujić

(2011). The crime reducing effect of education. The Economic Journal, 121(552), 463–484.

32.

Mahmud

, & Riley

(2021). Household response to an extreme shock: Evidence on the immediate impact of the COVID-19 lockdown on economic outcomes and well-being in rural Uganda. World Development, 140, 105318.

33.

McDonald

J. F.

, & Balkin

(2020). The COVID-19 and the decline in crime. SSRN 3567500. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3567500

34.

NITI Aayog, Government of India. (2019). Transforming aspirational districts in India: A programme for New India–2022.

35.

Peterman

, & O’Donnell

(2020). COVID-19 and violence against women and children: A second research round up. https://www.cgdev.org/sites/default/files/covid-19-and-violence-against-women-and-children-second-research-round.pdf

36.

Poblete-Cazenave

(2020). The great lockdown and criminal activity: Evidence from Bihar. Ideas for India. https://www.ideasforindia.in/topics/miscellany/covid-19-lockdown-and-criminal-activity-evidence-from-bihar.html

37.

Pressman

, & Carol

(1971). Crime as a diseconomy of scale. Review of Social Economy, 29(2), 227–236.

38.

Raj

, & McDougal

(2014). Sexual violence and rape in India. Lancet, 383(9920), 865.

39.

Ravallion

(2007). Evaluating anti-poverty programs. In Schultz

& Strauss

(Eds), Handbook of development economics (Vol. 4, pp. 3787)–3846). Elsevier.

40.

Ravindran

, & Shah

(2020). Unintended consequences of lockdowns: COVID-19 and the shadow pandemic (No. w27562). National Bureau of Research.

41.

Rossow

(2001). Alcohol and homicide: A cross-cultural comparison of the relationship in 14 European countries. Addiction, 96(1s1), 77–92.

42.

Shichor

, Decker

D. L.

, & O’Brien

R. M.

(1979). Population density and criminal victimization some unexpected findings in central cities. Criminology, 17(2), 184–193.

43.

Soares

R. R.

(2004). Development, crime and punishment: accounting for the international differences in crime rates. Journal of development Economics, 73(1), 155–184.

44.

Sudha

, & Rajan

S. I.

(2003). Persistent daughter disadvantage: What do estimated sex ratios at birth and sex ratios of child mortality risk reveal? Economic and Political Weekly, 4361–4369.

45.

The Times of India. (2020, April 4). Govt launches mega cash transfer to fight Covid-19. https://timesofindia.indiatimes.com/india/govt-launches-mega-cash-transfer-to-fight-covid-19/articleshow/74975904.cms

46.

Thurman

P. J.

, Roe Bubar

J. D.

, Plested

, Edwards

, LeMaster

, Bystrom

, & Hardy

(2003). Violence against Indian women, final revised report. https://nij.ojp.gov/library/publications/violence-against-indian-women-final-revised-report

47.

United Nations. (2010). Handbook on effective police responses to violence against women.

48.

United Nations. (2020). Policy brief: The impact of COVID-19 on women.

49.

Wooldridge

J. M.

(2002). Econometric analysis of cross section and panel data. MIT Press.

50.

Wooldridge

J. M.

(2013). Introductory econometrics: A modern approach (5th ed.). South-Western Cengage Learning.

51.

World Health Organization. (2013). Global and regional estimates of violence against women: Prevalence and health effects of intimate partner violence and non-partner sexual violence.

52.

Zhang

, Litvinova

, Liang

, Wang

, Zhao

, Wu

, Merler

, Viboud

, Vespignani

, Ajelli

, & Yu

(2020). Changes in contact patterns shape the dynamics of the COVID-19 outbreak in China. Science, 368(6498), 1481–1486.