## wpiea2024239-print-pdf

## Source details

**Canonical URL:** [wpiea2024239-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024239-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024239-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024239-print-pdf.pdf.json)

---

### Executive Summary — Introduction and Scope
- Focus: intimate partner violence (IPV) in India; India’s female population represents more than one sixth of the global female population.
- Data: micro-level evidence from up to 235 thousand respondents from three waves of the National Family Health Survey (2005-06 (NFHS-3, 2007), 2015-16 (NFHS-4, 2017) and 2019-21 (NFHS-5, 2022)) and macro-level economic and institutional information from Indian states.
- Framework: socio-ecological model to disentangle individual, family, community and societal (labeled “socio-economic”) factors correlated with individual risk of IPV.

### Key Facts on Incidence and Costs
- Lifetime IPV incidence in India: 31.8 percent in 2019-21, down only by 1.5 percentage points compared to 2015.
- Physical violence: 29.8 percent in 2015-16 to 28.2 percent in 2019-21.
- Sexual violence: 7 percent in 2015-16 to 6.1 percent in 2019-21.
- Emotional violence: increased by +0.2 ppt to 14 percent in 2019-21.
- State variation: rates around 10 percent in Goa, Himachal Pradesh, and Nagaland, and up to 50 percent in Karnataka, Bihar, and Manipur.
- Health and human capital associations:
  - Domestic violence increases women’s risk of anemia and being underweight.
  - Correlates with children’s risk of being stunted; children of mothers who experience IPV face higher infant and under-five mortality risks.
- Macroeconomic costs and labor context:
  - A one-percentage point increase in the share of women experiencing violence could reduce economic activity (as measured by nightlights) by 8-9 percentage points.
  - Each incidence of IPV implies at least five paid days of work lost.
  - A study for Nagpur found the average household loses the equivalent of almost two weeks’ income per occurrence.
  - If labor force participation gaps in India narrowed to the average of countries in the Asia and Pacific region, this could yield GDP gains of more than 25 percent (estimate based on Ostry and others (2018)).
  - Female labor force participation context: For every 10 men, about 4 women work.

### Drivers of IPV — Stylized Findings (Socio-ecological model)
- Individual-level patterns:
  - For most types of employment, a woman being employed and earning more than her partner translates into a higher risk of IPV.
  - Employment can improve bargaining power, but empirical evidence in India shows higher rates of violence for working women and for those more educated or better employed than their husband.
- Conditional effects and interactions:
  - Impact of women’s employment on IPV depends on household and state characteristics.
  - When combined with household-level empowerment and agency, lower state acceptance of IPV, higher male education, and normalization of reporting, women’s formal and paid employment can reduce individual IPV risk.
- State-level patterns:
  - Higher share of women in leadership positions at the state-level is associated with a lower risk of violence for individuals in that state.
  - Better reporting infrastructure for victims and higher charge-sheeting rates at the state-level translate into a lower risk of IPV for individuals and lower incidences of IPV at the state-level.
- Institutional and normative factors:
  - Reporting infrastructure and charge-sheeting reduce individual risk.
  - Social norms—acceptance of wife beating and normalization of reporting—shape incidence and help-seeking behavior.

### Empirical Approach and Evidence
- Methodology:
  - Pooled analyses using NFHS waves combined with state-level economic and institutional indicators; no causal claims are made.
  - Weighted logistic regressions (pooled logit) with NFHS two-stage sampling probability weights; time and state fixed effects included.
- Main empirical summaries:
  - Micro-level: employment and earning more than partner generally associated with higher IPV risk absent mitigating household or state factors.
  - Interactions: formal paid employment can lower IPV risk conditional on household empowerment, male education, lower state-level acceptance of IPV, and established reporting norms.
  - Macro-level: higher female leadership shares and stronger enforcement/reporting systems at the state-level associate with lower IPV incidence.

### Empirical Magnitudes and Interaction Effects (exact reported average marginal effects)
- Household empowerment (decision-making):
  - If respondent is not involved in household decision-making, female employment is associated with a probability of IPV that is 1.5 percentage points higher.
  - If respondent is involved in household decision-making, having a job is associated with a probability of IPV that is lower by approximately 3.5 percentage points.
- Partner education:
  - If partner has at least secondary or higher education, having a job is associated with a lower risk of IPV.
  - If partner has no education, having a job is associated with a risk of IPV that is higher by around 4.5 percentage points.
- Normalization of violence (state-level justification rates):
  - At state justification levels between 10 and 20 percent, economic empowerment is associated with a likelihood of IPV that is lower by approximately 4.9 to 6 percentage points.
  - In states with justification levels between 60 and 70 percent, having a job is associated with a likelihood of IPV that is higher by approximately 3.5 to 6 percentage points.
- Social acceptance of reporting:
  - If state-level reporting to friends/family is around 35 percent, having a job is associated with a likelihood of IPV that is lower by approximately 6.9 percentage points.
  - If speaking out is less widespread (female employment less than 14 percent), female employment is associated with the risk of IPV being up to 4.2 percent higher.

### Government Initiatives (legal protections, services, and empowerment schemes)
- Protective laws and amendments:
  - Protection of Women Against Domestic Violence Act (2005).
  - Prohibition of Child Marriage Act (2006).
  - Criminal Law Amendment Act (2013) and 2018 amendment (increased penalties for offenses of rape; mandates completion of investigation and trials within 2 months each).
- Services and reporting initiatives:
  - Women helplines, One Stop Centres (integrated emergency, medical, psycho-social and legal services), Central Victim Compensation Fund (one-time grant in 2016).
  - One Stop Centre scheme and scheme to universalize Women Helplines administered from 2015.
  - Mahila Police Volunteers (MPV) scheme implemented in 2016.
- Empowerment, health, education, entrepreneurship, leadership schemes (selected and as stated):
  - Stand-Up India Scheme (2016), Pradhan Mantri Ujjwala Yojana (2016), POSHAN Abhiyaan (2018), Vigyan Jyoti (2020), Mahila Shakti Kendra (2017), Integrated Women Empowerment Programme (2022), and others.
  - Political leadership reservations: Seventy-third Amendment (1992) one-third reservation for women in Panchayats; One Hundred and Sixth Amendment (2023) reserves one-third of all seats for women in the Lok Sabha, state legislative assemblies, and the Legislative Assembly of the National Capital Territory of Delhi.

### Policy Implications and Recommendations
- Core takeaway: complexity of IPV requires a multipronged approach combining protection with economic empowerment.
- Policy priorities:
  - Strengthen reporting infrastructure and enforcement to increase reporting and charge-sheeting rates.
  - Promote women’s economic empowerment via formal, paid employment combined with household-level empowerment programs so employment translates into reduced IPV risk.
  - Increase female representation in leadership to change social perceptions and expand outside employment opportunities for women.
  - Address social norms: reduce acceptance of IPV through public campaigns and education; normalize reporting and help-seeking.
  - Coordinate interventions across health, legal, education, and economic policy areas to capture interplay of micro and socio-economic drivers.
- Specific supply- and demand-side labor market recommendations:
  - Improve education for both men and women and create more paid jobs for women in the formal sector by addressing services, infrastructure, skills, safety (supply-side) and promoting non-agricultural sectors and addressing occupational segregation (demand-side).
  - Increase gender diversity within formal employment, especially at senior levels, and implement policies to improve retention of women in the formal labor force.
- Institutional and awareness actions:
  - Strengthen institutions, reporting infrastructure, availability of services for IPV victims; increase awareness of available services.
  - Build more infrastructure such as police stations, domestic violence help desks, social services, and hospitals.
  - Tailor awareness-raising campaigns on the impact of IPV on victims, families, society, and the economy.
- Overarching recommendation:
  - A multipronged approach is required that simultaneously empowers women economically while protecting individuals; government action should be supported by communities and stakeholders (family counselors, law enforcement, courts, NGOs, private sector).

### Prevalence and Reporting — Survey Coverage and Patterns
- NFHS sample sizes for ever-married women (age 15-49): 83,703 (2005-06), 79,729 (2015-16), and 72,320 (2019-2021).
- Acceptance of wife-beating (female acceptance, series displayed): 39.8, 33.3, 31.8 (2005, 2015, 2019).
- Latest female acceptance: 41 percent (figure label as provided).
- Reporting and help-seeking aggregates (figure labels preserved): 47.3, 44.5, 40.7 and 41.8, 31.7, 33.5.
- Only one in 100 sought help from official sources.
- State examples:
  - Manipur: IPV incidence 40 percent; only 3 percent sought help from any source.
  - Punjab and Goa: IPV rates around 10-13 percent; more than 30 percent sought help.

### Annex I — Data Description (variables, datasets, and measurement)
- Main dependent variable: dummy = 1 if respondent has ever experienced any form of domestic violence from her husband or partner (physical, emotional, sexual).
- NFHS waves for violence indicators: 2005, 2015, 2019.
- PLFS/NSS used for state-level participation and leadership (employer) indicators: 2005-06, 2017-18, 2021-22.
- CAW Charge-sheeting rate data: NCRB Crime in India (years 2006, 2016, 2022).
- Per Capita Net State Domestic Product (Current Prices, in Rupees): National Statistics Office years 2005, 2015, 2019.
- Selected variable codings preserved exactly as in source (examples):
  - Highest educational level: 0: No education; 1: Primary education; 2: Secondary education; 3: Higher education.
  - Occupation codes: 0: not in work force/no occupation; 1: professional/ technical/ managerial; 2: clerical; 3: sales; 5: agricultural; 7: services/household and domestic; 8: manual - skilled and unskilled.
  - Economic Empowerment dummy = 1 if respondent works in professional/ technical/ managerial (code 1), clerical (code 2), or sales (code 3) role, and is paid in cash only (code 1).
  - Respondent earns more than her partner: dummy = 1 if both earn and respondent earns more, or partner not bringing in money and respondent is earning.
  - Justifies wife-beating dummy = 1 if respondent responds “yes” to at least one justification question.
  - Help-seeking dummies and state-level weighted averages (multiplied by 100 to get percentage).
- Measurement issues:
  - Domestic violence data subject to underreporting; NFHS uses anonymity and safe, specific questioning to reduce under-reporting.

### Annex IV — Logistic Regression Results (overview and selected coefficients)
- Models and sample:
  - Four logistic regression models for outcomes: (1) Intimate Partner Violence (all); (2) Physical Violence; (3) Sexual Violence; (4) Emotional Violence.
  - Observations: 124,158 (for each model).
  - Wald (81): 7481 (model 1), 7620 (model 2), 2629 (model 3), 3314 (model 4).
  - Prob > Wald: 0.000 (all models).
  - Pseudo R2: 0.123 (model 1), 0.127 (model 2), 0.103 (model 3), 0.0784 (model 4).
- Selected individual-level coefficients (coefficient (z-statistic)):
  - Current age: 0.108*** (10.282) overall; age^2: -0.001*** (-8.808) overall.
  - Highest education =2, Secondary: -0.105*** (-3.497) overall; =3, Higher: -0.551*** (-9.361) overall.
  - Occupation =3, Agricultural or Manual: 0.298*** (11.759) overall.
  - Respondent's father beat her mother: 1.206*** (46.862) overall.
  - Respondent drinks alcohol: 0.489*** (7.160) overall.
  - Respondent earns more than her partner: 0.138*** (2.825) overall.
  - Respondent involved in household decision-making: -0.386*** (-18.165) overall.
  - Number of children aged 5 and under: 0.081*** (5.944) overall.
  - Rural residence: -0.091*** (-3.210) overall.
- Selected partner-level and community/institutional coefficients:
  - Partner secondary education: -0.102*** (-3.392) overall.
  - Partner drinks alcohol: 0.298*** (12.992) overall.
  - Share of people that justify beating: 0.029*** (14.234) overall.
  - Share of women that sought help from official sources: -0.187*** (-7.767) overall.
  - CAW Charge-sheeting rate: -0.004*** (-3.166) overall.
  - Share of women who are employers: -0.216*** (-3.465) overall.
- Wealth effects (relative to poorest):
  - Middle: -0.160*** (-4.731) overall.
  - Richer: -0.290*** (-7.263) overall.
  - Richest: -0.560*** (-10.992) overall.
- Time effects:
  - Year = 2015: -0.097* (-1.943) overall.
  - Year = 2019: 0.010 (0.172) overall.
- State fixed effects: coefficients reported for each state (selected examples include Himachal Pradesh, Punjab, Haryana, Uttar Pradesh, Bihar, Tripura, Maharashtra, Kerala, Telangana, Lakshadweep) with statistical significance indicated in the table.

*International Monetary Fund — content derived from wpiea2024239-print-pdf.*

### Executive Summary ......................................................................................................

### Executive Summary

### Introduction and Scope
- Focus: intimate partner violence (IPV) in India, using India as a case study because the country’s female population represents more than one sixth of the global female population and data are comprehensive and consistent at the national and state levels.
- Data: micro-level evidence from up to 235 thousand respondents from three waves of the National Family Health Survey (2005-06 (NFHS-3, 2007), 2015-16 (NFHS-4, 2017) and 2019-21 (NFHS-5, 2022)) and macro-level economic and institutional information from Indian states.
- Framework: socio-ecological model to disentangle individual, family, community and societal (labeled “socio-economic”) factors correlated with individual risk of IPV.

### Key Facts on Incidence and Costs
- Lifetime IPV incidence in India: 31.8 percent in 2019-21, down only by 1.5 percentage points compared to 2015.
- Physical violence: 29.8 percent in 2015-16 to 28.2 percent in 2019-21.
- Sexual violence: 7 percent in 2015-16 to 6.1 percent in 2019-21.
- Emotional violence: increased by +0.2 ppt to 14 percent in 2019-21.
- Large state variation: rates around 10 percent in Goa, Himachal Pradesh, and Nagaland, and up to 50 percent in Karnataka, Bihar, and Manipur.
- Health and human capital: domestic violence increases women’s risk of anemia and being underweight, correlates with children’s risk of being stunted; children of mothers who experience IPV face higher infant and under-five mortality risks.
- Macroeconomic costs:
  - A one-percentage point increase in the share of women experiencing violence could reduce economic activity (as measured by nightlights) by 8-9 percentage points.
  - Each incidence of IPV implies at least five paid days of work lost.
  - A study for Nagpur found the average household loses the equivalent of almost two weeks’ income per occurrence.
  - If labor force participation gaps in India narrowed to the average of countries in the Asia and Pacific region, this could yield GDP gains of more than 25 percent (estimate based on Ostry and others (2018)).
- Female labor force participation context: For every 10 men, about 4 women work.

### Drivers of IPV — Stylized Findings
- Individual-level patterns:
  - For most types of employment, a woman being employed and earning more than her partner translates into a higher risk of IPV.
  - Employment can improve bargaining power, but empirical evidence in India shows higher rates of violence for working women and for those more educated or better employed than their husband.
- Conditional effects and interactions:
  - The impact of women’s employment on IPV depends on household and state characteristics.
  - When combined with specific interventions—household-level empowerment and agency, lower acceptance of IPV in the state, higher male education, and normalization of reporting—women’s formal and paid employment can reduce individual IPV risk.
- State-level patterns:
  - A higher share of women in leadership positions at the state-level is associated with a lower risk of violence for individuals in that state.
  - Better reporting infrastructure for victims and higher charge-sheeting rates at the state-level translate into a lower risk of IPV for individuals and lower incidences of IPV at the state-level.
- Institutional and normative factors:
  - Strong institutions and enforcement matter: reporting infrastructure and charge-sheeting reduce individual risk.
  - Social norms—acceptance of wife beating and normalization of reporting—shape both incidence and help-seeking behavior.

### Empirical Approach and Evidence
- Methodology: pooled analyses using NFHS waves combined with state-level economic and institutional indicators to analyze the interplay of microeconomic and socio-economic factors; no causal claims are made.
- Main empirical results (summary):
  - Micro-level: employment and earning more than partner generally associated with higher IPV risk absent mitigating household or state factors.
  - Interactions: formal paid employment can lower IPV risk conditional on household empowerment, male education, lower state-level acceptance of IPV, and established reporting norms.
  - Macro-level: higher female leadership shares and stronger enforcement/reporting systems at the state-level associate with lower IPV incidence.

### Government Initiatives (Highlights)
- Protective measures (examples):
  - The Protection of Women Against Domestic Violence Act (2005).
  - The prohibition of Child Marriage Act (2006).
  - Criminal Law Amendment Act (2013) and 2018 amendment (increased penalties for offenses of rape; mandates completion of investigation and trials within 2 months each).
  - Women helplines, one-stop centers (integrated emergency, medical, psycho-social and legal services), and the Central Victim Compensation Fund (one-time grant in 2016).
- Empowerment schemes (examples):
  - Stand-Up India Scheme (2016), Pradhan Mantra Ujjwala Yojana (2016), POSHAN Abhiyaan (2018), Vigyan Jyoti (2020).

### Policy Implications and Recommendations
- Core takeaway: the complexity of IPV requires a multipronged approach that combines protection with economic empowerment.
- Policy priorities (inferred from findings):
  - Strengthen reporting infrastructure and enforcement (to increase reporting and charge-sheeting rates).
  - Promote women’s economic empowerment via formal, paid employment combined with household-level empowerment programs to ensure employment translates into reduced IPV risk.
  - Increase female representation in leadership to change social perceptions and expand outside employment opportunities for women.
  - Address social norms: reduce acceptance of IPV through public campaigns and education; normalize reporting and help-seeking.
  - Coordinate interventions across health, legal, education, and economic policy areas to capture the interplay of micro and socio-economic drivers.

### Conclusion
- IPV in India remains high with slow declines and wide state heterogeneity, imposing significant human, social, and economic costs.
- Empirical evidence points to interdependent roles of individual, household, community and state-level socio-economic factors: empowering women economically can reduce IPV risk when combined with household agency, lower normative acceptance of violence, higher male education, and strong reporting/enforcement institutions.
- A multipronged strategy that simultaneously empowers women and strengthens protections and institutions is needed to reduce IPV and unlock India’s development potential.

*International Monetary Fund.*

### 1. National Level 2. State Level (2005 vs. 2019)

### 1. National Level 2. State Level (2005 vs. 2019)

### Prevalence and Reporting of Intimate Partner Violence (IPV)
- Data coverage: three NFHS waves covering 83,703 (2005-06), 79,729 (2015-16), and 72,320 (2019-2021) ever-married women (age 15-49).
- Acceptance of wife-beating:
  - The share of respondents who justify wife beating is substantial, with higher acceptance rates among women.
  - Female acceptance (latest survey) is 41 percent.
  - Female acceptance has decreased over time in the series displayed as 39.8, 33.3, 31.8 (2005, 2015, 2019) (figure labels preserved from source).
  - Male acceptance rose by +2 ppts between 2015 and 2019.
  - State variation in 2019-21: Andhra Pradesh, 80 percent of women justified wife-beating for at least one reason; Tamil Nadu and Telangana, around 75 percent; Uttar Pradesh, around 36 percent; Himachal Pradesh and Delhi, below 15 percent.
- Reporting and help-seeking:
  - High incidence of IPV contrasts with low reporting rates and low help-seeking.
  - Reporting of IPV to anyone is generally low and declined between 2005 and 2019.
  - Aggregate help-seeking shares shown in figures: 47.3, 44.5, 40.7 and 41.8, 31.7, 33.5 (figure labels preserved from source).
  - Only a small share sought help from unofficial sources; even fewer sought help from official sources—only one in 100 did.
  - State examples:
    - Manipur: IPV incidence 40 percent; only 3 percent of victims sought help from any source.
    - Punjab and Goa: IPV rates around 10-13 percent; more than 30 percent of victims sought help.

### Drivers of IPV: The Ecological Model and Stylized Facts
- Conceptual framework:
  - Uses the socio-ecological model (Bronfenbrenner 1977; Bronfenbrenner 1986; Heise 1998) with concentric systems: microsystem (individual), mesosystem (intimate partner interactions), exosystem (household/community), and macrosystem (state-level socio-economic, legal, political context).
- Individual and partner-level associations:
  - Established correlates include lower educational levels, family history of violence, and alcohol consumption.
  - Strong negative relationship between female and male education and IPV (coefficients reported in Figures 6-10 account for time fixed effects).
  - Family history of violence is associated with higher rates of IPV.
- Community-level associations:
  - Rigid and patriarchal community attitudes that devalue women increase IPV risk.
  - Higher acceptance toward wife-beating at the state level is related to higher incidence of IPV (Figure 8).
- State-level (socio-economic) factors:
  - Reporting infrastructure, state policies and laws, societal perceptions, and enforcement matter for IPV incidence.
  - States with higher shares of victims who sought help from any source have lower average IPV incidence (Figure 9).
  - Seeking help from official sources is a stronger deterrent than seeking help from unofficial sources.
  - Higher charge-sheeting rates for crimes against women are associated with lower IPV incidence (Figure 10).
  - No clear correlation observed yet between IPV incidence and conviction rates, possibly because conviction rates are very low and take a long time to materialize.

### Economic Empowerment and IPV: Mixed Evidence
- Literature background:
  - Two strands: economic empowerment may reduce IPV by increasing women’s bargaining power; alternatively, women’s employment or earning more than partner may increase IPV in some contexts.
  - Prior findings cited: Bhattacharyya and others (2011), Panda and Agarwal (2005) (protective effects); Dalal (2011), Rani and Bonu (2009), Burton and others (2000) (higher IPV for working or relatively advantaged women in some samples).
- Findings at state and individual levels (Figure 7 summary):
  - State-level:
    - A higher ratio of women to men working as regular salaried employees is significantly associated with lower IPV (Figure 7.1).
    - A higher share of female employers correlates with lower incidences of violence (Figure 7.2).
    - Women’s inclusion in household decision making relates negatively with IPV (Figure 7.3).
    - Higher rates of female financial inclusion at the state level are associated with lower incidences of violence (Figure 7.4).
  - Individual-level:
    - Controlling for other factors, women who are not working experience less violence, on average, relative to employed women.
    - Working respondents face a significantly higher risk of IPV, with the strongest associations for work in agriculture or manual labor.
    - Compared to being not working, working in professional, clerical, managerial, technical, or sales roles is associated with higher risk of sexual and emotional violence.
    - IPV is significantly higher if the respondent earns more than the husband.
    - Other empowerment dimensions—respondent’s involvement in household decision making and female-headed households—are associated with lower risk of IPV (decision-making lowers all types of IPV; female-headed households lower physical IPV).

### Empirical Methodology
- Outcome and sample:
  - Dependent variable: domestic violence dummy (ever experienced specific type of violence).
  - Pooled analysis using three NFHS waves; results reported for IPV and individual types (physical, sexual, emotional).
- Regression approach:
  - Weighted logistic regressions (pooled logit) with probability weights based on NFHS two-stage sampling to obtain nationally representative results.
  - Regression includes individual-level covariates, partner characteristics, household/community variables, state-level socio-economic indicators, religion and caste dummies, and time and state fixed effects.
  - Time fixed effects: year fixed effects for 2015-16 and 2019-21, taking 2005-06 as the base.
  - Macro or state fixed effects represented by σ_s; year fixed effects by θ_t.
- Variables included (as specified in the pooled logit specification):
  - Individual: age, age^2, education dummies ("no education", "primary education", "secondary education", "higher education"), employment type dummies (professional/technical/managerial/clerical/sales; services household/domestic; agriculture/manual labor), formal sector paid employment dummy, familial history of violence, alcohol consumption.
  - Partner/relationship: partner education, partner employment type, partner alcohol consumption, age gap, dummy if respondent earns more than partner, respondent participation in household decision-making, respondent access to finance (bank account).
  - Household/community: number of children aged 5 and under, household members, female head of household, rural indicator, wealth quintile dummies ("poorest", "poorer", "middle", "richer", "richest"), religion categories (Hindu, Muslim, Christian, Sikh, Buddhist / Neo-Buddhist, Jain, other/none), caste categories (SC/ST, OBC, no caste or tribe, other).
  - State-level socio-economic factors: state averages for share who justify IPV, share of victims who sought help from unofficial sources, share who sought help from official sources, state-level charge-sheeting rate for crimes against women, share of women who are employers.
- Outcome transformation:
  - The dependent variable Y is the log-odds ratio of experiencing domestic violence:
    - Y = ln(Pr(domestic violence) / (1 − Pr(domestic violence)))

### Baseline Regression Findings (stylized)
- Individual-level:
  - Non-linear relationship between respondent age and IPV risk: risk peaks in the respondent’s 30s and decreases with age thereafter.
  - Respondent education is consistently associated with lower IPV risk:
    - Physical and sexual IPV risk significantly decreases at least at secondary education.
    - Emotional violence significantly lower only for respondents with higher education.
  - Family history of violence associated with higher IPV rates.
  - Working respondents and respondents who earn more than their partners are associated with higher IPV risk in the micro-level regressions.
  - Employment type matters: agriculture/manual labor associated with the highest increased risk across types of IPV.
  - Women’s involvement in household decision making is associated with lower risk of all types of IPV.
  - Female-headed households show lower risk of physical IPV.

*International Monetary Fund*

### 1. Intimate Partner Characteristics

### 1. Intimate Partner Characteristics

### Key empirical findings on correlates of IPV
- Note: *p<.10; **p<.05; ***p<.01. Results based on regression that also includes covariates in Tables 1 and 3, as well as a dummies to capture religion, cast, and state and time fixed effects.
- Partner’s education and employment status:
  - IPV is lower if the respondent’s partner has attained at least secondary education, or, in the case of sexual violence, when he has attained higher level of education.
  - IPV is lower when the intimate partner is employed, though with some variation in significance depending on the type of employment.
- Alcohol consumption:
  - The IPV risk is consistently higher if the intimate partner, or the respondent herself drinks alcohol.
- Household structure and dynamics:
  - Incidence of IPV significantly increases with the number of children aged 5 and under in the household.
  - Women living in larger households are less likely to experience IPV.
  - The higher the household’s wealth, the lower the risk of each type of violence.
- State-level socio-economic and institutional factors:
  - Women living in states where a larger share of the population justifies wife beating for at least one reason are at a significantly higher risk of IPV (consistent for all types of violence).
  - Higher reporting rates of violence to unofficial sources (friends and family) at the state level are significantly associated with a lower risk of sexual violence at the individual level.
  - A higher share of victims who report to official sources (police, medical professionals, social services) in a state is associated with a significantly lower IPV risk for the individual.
  - A higher charge-sheeting rate within the state is associated with lower physical, sexual, and overall violence but a somewhat higher level of emotional violence.
  - Greater female leadership: women living in states where the share of women who are employers is higher are less likely to be at risk of physical (and overall) IPV.

### When does formal paid employment affect IPV risk?
- Definition used: female paid formal employment = dummy equal to 1 if respondent is working in a professional, technical, managerial, clerical, or sales position and getting paid in cash only (an approximation since NFHS does not use formal/informal categories for occupation type).
- Pooled logit regression specification:
  - 푌 = 훽1 푿1 + 훽2 푿2 + 훽3 푿3 + 훽4 푿4 + 훽5 (푓표푟푚푎푙 ∗ 풁) + 휃 + 휖, where 푌 is the log-odds ratio of ever experiencing a specific type of violence; 푿1, 푿2, 푿3, 푿4 are individual, relationship, household, and societal factors; 푓표푟푚푎푙 is the female formal paid employment dummy; 푓표푟푚푎푙∗풁 are interactions with four factors.
- Interaction factors tested:
  - Individual/family level: (1) respondent’s involvement in decision-making within the household; (2) respondent’s partner’s highest level of education (primary, secondary and higher).
  - State level: (1) level of justification of IPV within the state of residence; (2) share of IPV (physical or sexual) victims who sought help from unofficial sources.
- Statistical significance notes:
  - For all interaction terms included except for the interaction with husband’s highest education level, the coefficient of the interaction term was statistically significant at the 95 percent significance level.
  - For the interaction with husband’s highest education level, the coefficient of interaction was significant at the 90 percent significance level for secondary education.
- Average marginal effects (exact reported values):
  - Empowerment and agency within the household:
    - If a respondent is not involved in household decision-making, female employment is associated with a probability of IPV that is 1.5 percentage points higher.
    - If the respondent is involved in household decision-making, having a job is associated with a probability of IPV that is lower by approximately 3.5 percentage points.
  - Education of the partner:
    - If the intimate partner has at least secondary or higher education, having a job is associated with a lower risk of IPV.
    - If he has no education, having a job is associated with a risk of IPV that is higher by around 4.5 percentage points.
  - Normalization of violence at the societal level:
    - At relatively lower levels of justification of IPV within the state of residence (between 10 and 20 percent), economic empowerment is associated with a likelihood of IPV that is lower by approximately 4.9 to 6 percentage points.
    - In states with relatively higher levels of justification of IPV (between 60 and 70 percent), having a job goes hand in hand with a likelihood of IPV that is higher by approximately 3.5 to 6 percentage points.
  - Social acceptance of reporting IPV:
    - If reporting (physical or sexual) IPV to friends and family is more widespread within the state of residence (around 35 percent of victims report IPV), having a job is associated with a likelihood of IPV that is lower by approximately 6.9 percentage points.
    - If speaking out against IPV is less widespread, female employment (less than 14 percent) the risk of IPV is up to 4.2 percent higher.
- Note: Bars in the referenced figures indicate 95 percent confidence intervals.

### Government initiatives and institutional context
- Legal protections and specialized services:
  - Dowry Prohibition Act enacted in 1961.
  - Crimes Against Women (CAW) Cell established at the central level in the Delhi police in 1983.
  - Protection of Women Against Domestic Violence Act (PWDVA 2005) introduced first legal definition of domestic violence and provided civil remedies (secure accommodation, protection orders, monetary relief, compensation, temporary separation, relief from other suits and legal proceedings).
- Recent government schemes to support victims and reporting:
  - One Stop Centre (OSC) scheme and scheme to universalize Women Helplines (WHL) administered from 2015.
  - Mahila Police Volunteers (MPV) scheme implemented in 2016.
  - OSCs provide specialized services; WHL offers 24-hour emergency and non-emergency responses linking victims to police, OSCs, medical professionals, and legal services.
- Schemes for empowerment, health, education, entrepreneurship, and leadership (selected programs and years as stated):
  - National Commission for Women Act statutory board (1992).
  - National Policy for the Empowerment of Women (2001).
  - National Plan of Action for the Girl Child (1991-2000).
  - Beti Bachao Beti Padhao Scheme (2015).
  - Nai Roshni (Scheme for Leadership Development of Minority Women, launched in 2012-13).
  - Mahila Shakti Kendra (2017) implemented in 115 districts.
  - Integrated Women Empowerment Programme (2022).
  - Health and nutrition: Pradhan Mantri Matru Vandana Yojana (PMMVY 2017), Janani Suraksha Yojana (2005), POSHAN Abhiyaan (2018), Pradhan Mantri Ujjwala Yojana (PMUY 2016).
  - Education, skills, and entrepreneurship: STEP (1986-87), TREAD (2014), Skill Upgradation and Mahila Coir Yojana (2017), Women Entrepreneurship Platform (WEP 2017), Stand-Up India Scheme (2016).
  - Women in STEM and fellowships: WISE-KIRAN (2002-03), Overseas Fellowship Scheme (2017-18), Vigyan Jyoti (2020).
  - Political leadership reservations: Seventy-third Amendment (1992) one-third reservation for women in Panchayats; One Hundred and Sixth Amendment (2023) reserves one-third of all seats for women in the Lok Sabha, state legislative assemblies, and the Legislative Assembly of the National Capital Territory of Delhi.

### Policy conclusions and recommendations
- Main conclusions:
  - A woman in India is at a higher risk for IPV if she is approximately in her late 30s or early 40s, and if she or her partner are educated only at the primary level or not at all.
  - Family history of IPV, lower agency within the household, alcohol consumption, and no involvement in household decision-making are strong predictors of IPV.
  - Lower wealth and urban residency, everything else equal, predict higher levels of IPV.
  - Employment by itself is not a deterrent of IPV; it needs to be complemented with supporting frameworks to be associated with lower risk.
  - Work in the agriculture sector, as a manual (skilled or unskilled) worker and earning a higher income than the partner are associated with higher incidences of IPV at the individual level.
  - Institutions and enforcement matter: seeking help from official sources is a strong deterrent of IPV, and higher state charge-sheeting rates for crimes against women lower individual risk.
- Policy prescriptions emphasized:
  - Improve education for both men and women and create more paid jobs for women in the formal sector, addressing both supply-side constraints (services, infrastructure, skills, safety) and demand-side constraints (promoting non-agricultural sectors, addressing occupational segregation).
  - Increase gender diversity within formal employment, especially at senior levels in the corporate sector, and implement policies to improve retention of women in the formal labor force.
  - Strengthen institutions, reporting infrastructure, and availability of services for IPV victims; increase awareness of available services.
  - Ensure laws and supportive frameworks are effective; build more infrastructure such as police stations, domestic violence help desks, social services, and hospitals.
  - Increase awareness of the impact of IPV on victims, families, society, and the economy through tailored awareness-raising campaigns.
- Additional numeric/legal context:
  - Indian women currently enjoy just 60 percent of the legal protections that men do.
- Overarching recommendation:
  - A multipronged approach is required that simultaneously empowers women economically while protecting individuals; government action should be supported by communities and stakeholders (family counselors, law enforcement, courts, NGOs, private sector) to prevent spillovers to future generations and boost India’s macroeconomic potential.

*International Monetary Fund — Content derived from "1. Intimate Partner Characteristics" in the provided chapter PDF.*

### Annex I. Data Description

### Annex I. Data Description

### Data sources and main dependent variables
- Main dependent variable: dummy = 1 if respondent has ever experienced any form of domestic violence from her husband or partner, and 0 otherwise. Calculated by combining multiple sub-questions in the domestic violence module across three NFHS survey waves, covering physical, emotional, and sexual violence.
- Separate dummy indicators created for:
  - physical violence (severe or less severe),
  - sexual violence,
  - emotional violence.
- NFHS waves used for violence indicators: 2005, 2015, 2019.
- Periodic Labor Force Survey (PLFS) / National Sample Survey (NSS) used for state-level participation of women in leadership within the labor force:
  - Leadership dummy = 1 if individual’s principal or subsidiary activity is classified as “employer”.
  - PLFS/NSS waves used: 2005-06, 2017-18, 2021-22.
  - State-year averages of the leadership dummy for women are merged into the survey dataset.

### Independent variables (socio-ecological model)
- Independent variables divided into individual, relationship, household, and state-level factors.
- Individual- and partner-level data include age, education, employment status, alcohol consumption, etc.
- Household and contextual variables include familial, household, cultural, and socio-economic environment indicators.
- Definitions and codes for many variables are provided in Annex Table 3 (variable labels). Selected variable definitions and coding preserved exactly as in source:
  - Age: Current age of respondent in years (between 15 and 49).
  - Highest educational level (categorical): 0: No education; 1: Primary education; 2: Secondary education; 3: Higher education.
  - Occupation (categorical): 0: not in work force/no occupation; 1: professional/ technical/ managerial; 2: clerical; 3: sales; 5: agricultural; 7: services/household and domestic; 8: manual - skilled and unskilled.
  - Type of earnings (categorical): 0: not paid; 1: cash only; 2: cash and in-kind; 3: in-kind only.
  - Economic Empowerment dummy = 1 if respondent works in professional/ technical/ managerial (code 1), clerical (code 2), or sales (code 3) role, and is paid in cash only (code 1).
  - Respondent's father beat her mother: dummy = 1 if yes.
  - Respondent drinks alcohol: dummy = 1 if yes.
  - Partner's highest educational level: same categorical coding as respondent (men’s indicator).
  - Partner drinks alcohol: dummy = 1 if yes.
  - Respondent earns more than her partner: dummy = 1 if both earn and respondent earns more, or partner not bringing in money and respondent is earning.
  - Respondent has an account in a bank or other financial institution: dummy = 1 if yes.
  - Respondent involved in household decision-making (together or alone): dummy = 1 if participates in all household decision-making questions (decides alone or together).
  - Number of children aged 5 and under in household; number of household members.
  - Household head is female: dummy = 1 if yes.
  - Rural residence: dummy = 1 if de facto residence is rural.
  - Wealth Group quintiles: 1: poorest; 2: poorer; 3: middle; 4: richer; 5: richest.
  - Religion categorical codes preserved: 1: Hindu; 2: Muslim; 3: Christian; 4: Sikh; 5: Buddhist/Neo Buddhist; 6: Jain; 7: Jewish; 9: No religion; 96: Other.
  - Caste/Tribe categories (calculated): 0: No caste/tribe; 1: Schedule Caste; 2: Schedule Tribe; 3: Other Backward Class; 4: None of them.
  - Justifies wife-beating dummy = 1 if respondent (male or female) responds “yes” to at least one justification question.
  - State-level shares (calculated, multiplied by 100 to get percentage) for:
    - Share of women that justify beating for at least one reason (1998, 2005, 2015, 2019).
    - Share of men that justify beating for at least one reason (1998, 2005, 2015, 2019).
    - Share of people that justify beating for at least one reason (1998, 2005, 2015, 2019).
  - Help-seeking indicators (available only for women reporting physical or sexual violence):
    - Sought help from unofficial sources (2005, 2015, 2019) — dummy = 1 if “yes” to at least one unofficial source.
    - Sought help from official sources (2005, 2015, 2019) — dummy = 1 if “yes” to at least one official source (police, medical professionals, social services, lawyers).
    - State-level weighted averages of these help-seeking dummies (multiplied by 100 to get percentage).
  - CAW Charge-sheeting rate: state-level charge-sheeting rate for “Crimes Against Women” recorded by NCRB (Crime in India): years 2006, 2016, 2022.
  - PLFS/NSS labor market indicators (state-level averages, calculated): Share of men/women who are employers; Share of men/women attending domestic duties; Share of men/women working as regular salaried/wage employee — waves: 2005-06, 2017-18, 2021-22.
  - Women to men ratio of share working as regular salaried/wage employee (2005-06, 2017-18, 2021-22).
  - Per Capita Net State Domestic Product (Current Prices, in Rupees) from National Statistics Office: years 2005, 2015, 2019.
- Regression specification includes state fixed effects (흈s) and year fixed effects (휽t). The full regression and variable label table are provided in Annex III and Annex Table 3.

### Measurement issues and data quality
- Domestic violence data are subject to underreporting due to fear of negative consequences, normalization of violence, or differing definitions.
- The analysis uses DHS (NFHS) survey data which:
  - Promises anonymity and does not report domestic violence cases to authorities, reducing likelihood of under-reporting due to fear of negative consequences.
  - Uses specific and direct questions asked in a safe environment, decreasing likelihood of under-reporting due to normalization or definitional differences.

### Key datasets and years (selected from Annex Table 1)
- NFHS (Individual’s/Couple’s recode): violence indicators (2005, 2015, 2019); age (1992, 1998, 2005, 2015, 2019); education, occupation, wealth, religion, caste/tribe (1992, 1998, 2005, 2015, 2019) as specified per indicator.
- PLFS / NSS (calculated): labor force shares and employer indicators (2005-06, 2017-18, 2021-22).
- NCRB Crime in India (CAW charge-sheeting rates): 2006, 2016, 2022.
- National Statistics Office (Per Capita Net State Domestic Product, Current Prices, in Rupees): 2005, 2015, 2019.

*Source: Annex I. Data Description, wpiea2024239-print-pdf.*

### Annex IV. Logistic Regression Results

### Annex IV. Logistic Regression Results

### Overview of models
- Four logistic regression models reported for outcomes:
  - (1) Intimate Partner Violence (all)
  - (2) Physical Violence
  - (3) Sexual Violence
  - (4) Emotional Violence
- Observations: 124,158 (for each model)
- Wald (81): 7481 (model 1), 7620 (model 2), 2629 (model 3), 3314 (model 4)
- Prob > Wald: 0.000 (all models)
- Pseudo R2: 0.123 (model 1), 0.127 (model 2), 0.103 (model 3), 0.0784 (model 4)
- z-statistics in parentheses; significance: *p<.10; **p<.05; ***p<.01

### Key individual-level predictors (coefficients and z-statistics)
- Current age: 0.108*** (10.282); 0.128*** (11.730); 0.052*** (2.662); 0.062*** (4.289)
- Current age (squared): -0.001*** (-8.808); -0.002*** (-10.139); -0.001*** (-2.614); -0.001*** (-3.360)
- Highest educational level:
  - =1, Primary: 0.026 (0.839); 0.011 (0.343); -0.002 (-0.042); -0.009 (-0.225)
  - =2, Secondary: -0.105*** (-3.497); -0.142*** (-4.589); -0.118** (-2.105); -0.002 (-0.044)
  - =3, Higher: -0.551*** (-9.361); -0.577*** (-9.259); -0.425*** (-3.374); -0.275*** (-3.241)
- Occupation categories (respondent):
  - =1, Professional/Technical/Managerial/Clerical/Sales: 0.199 (1.348); 0.045 (0.293); 0.589** (2.450); 0.375* (1.900)
  - =2, Services (household and domestic): 0.284*** (3.943); 0.266*** (3.543); 0.228** (2.113); 0.346*** (4.118)
  - =3, Agricultural or Manual (skilled and unskilled): 0.298*** (11.759); 0.285*** (10.953); 0.327*** (7.290); 0.258*** (7.712)
- Respondent is working in the formal sector and is paid in cash only: -0.078 (-0.495); 0.052 (0.318); -0.439* (-1.662); -0.238 (-1.129)
- Respondent's father beat her mother: 1.206*** (46.862); 1.206*** (46.681); 0.961*** (23.828); 0.952*** (30.901)
- Respondent drinks alcohol: 0.489*** (7.160); 0.379*** (5.518); 0.387*** (3.353); 0.510*** (6.450)
- Respondent earns more than her partner: 0.138*** (2.825); 0.153*** (3.073); 0.283*** (3.868); 0.318*** (5.596)
- Respondent has an account in a bank or other financial institution: 0.002 (0.099); 0.020 (0.793); -0.013 (-0.288); -0.022 (-0.680)
- Respondent involved in household decision-making (together or alone): -0.386*** (-18.165); -0.342*** (-15.603); -0.609*** (-15.606); -0.490*** (-17.070)
- Number of children aged 5 and under in household: 0.081*** (5.944); 0.091*** (6.483); 0.064*** (2.759); 0.055*** (3.064)
- Number of household members: -0.015*** (-2.654); -0.018*** (-2.946); -0.004 (-0.388); -0.021*** (-2.825)
- Household head is female: -0.071 (-1.421); -0.087* (-1.657); -0.048 (-0.583); 0.053 (0.849)
- Rural residence: -0.091*** (-3.210); -0.106*** (-3.611); 0.012 (0.227); -0.073* (-1.941)

### Key partner-level predictors
- Age difference between partner and respondent: 0.001 (0.426); -0.002 (-0.812); -0.004 (-0.799); 0.006 (1.591)
- Partner's highest educational level:
  - =1, Primary: 0.017 (0.522); 0.023 (0.713); 0.068 (1.232); -0.014 (-0.322)
  - =2, Secondary: -0.102*** (-3.392); -0.103*** (-3.370); -0.062 (-1.215); -0.109*** (-2.779)
  - =3, Higher: -0.279*** (-5.658); -0.304*** (-5.857); -0.270*** (-2.860); -0.343*** (-4.868)
- Partner's occupation:
  - =1, Professional/Technical/Managerial/Clerical/Sales: -0.073 (-1.153); -0.067 (-0.993); -0.110 (-1.092); -0.061 (-0.747)
  - =2, Services (household and domestic): -0.185*** (-2.584); -0.132* (-1.756); -0.218* (-1.841); -0.174* (-1.907)
  - =3, Agricultural or Manual (skilled and unskilled): -0.022 (-0.378); 0.007 (0.109); -0.170* (-1.875); -0.060 (-0.811)
- Partner drinks alcohol: 0.298*** (12.992); 0.320*** (13.575); 0.316*** (7.819); 0.299*** (9.788)

### Community and institutional variables
- Share of people that justify beating for at least one reason: 0.029*** (14.234); 0.025*** (11.917); 0.030*** (8.669); 0.030*** (11.663)
- Share of women that sought help from unofficial sources: -0.002 (-0.545); 0.003 (0.860); -0.041*** (-6.781); 0.003 (0.727)
- Share of women that sought help from official sources: -0.187*** (-7.767); -0.194*** (-7.635); -0.228*** (-5.185); -0.090*** (-2.966)
- CAW Charge-sheeting rate: -0.004*** (-3.166); -0.003** (-2.340); -0.011*** (-4.900); 0.004** (2.188)
- Share of women who are employers: -0.216*** (-3.465); -0.206*** (-3.090); -0.162 (-1.279); -0.099 (-1.129)

### Wealth, religion, and caste effects
- Wealth Group:
  - =2, Poorer: 0.001 (0.018); -0.002 (-0.077); -0.024 (-0.447); -0.036 (-0.889)
  - =3, Middle: -0.160*** (-4.731); -0.164*** (-4.705); -0.138** (-2.356); -0.167*** (-3.695)
  - =4, Richer: -0.290*** (-7.263); -0.283*** (-6.863); -0.254*** (-3.579); -0.336*** (-6.258)
  - =5, Richest: -0.560*** (-10.992); -0.595*** (-11.148); -0.439*** (-4.437); -0.488*** (-6.844)
- Religion (relative categories with coefficients):
  - Hindu: 0.383*** (2.817); 0.319** (2.252); -0.180 (-0.746); 0.328 (1.563)
  - Muslim: 0.578*** (4.123); 0.517*** (3.545); -0.015 (-0.061); 0.529** (2.459)
  - Christian: 0.419*** (2.835); 0.330** (2.151); 0.017 (0.063); 0.373* (1.677)
  - Sikh: 0.311* (1.873); 0.220 (1.284); -0.405 (-1.344); 0.303 (1.204)
  - Buddhist/Neo-Buddhist: 0.581*** (3.289); 0.552*** (3.048); 0.086 (0.242); 0.636** (2.502)
  - Jain: 0.088 (0.316); 0.082 (0.277); 0.228 (0.469); -0.012 (-0.033)
- Caste or Tribe:
  - =1, SC/ST: 0.142*** (4.213); 0.167*** (4.744); -0.019 (-0.308); 0.035 (0.759)
  - =2, OBC: 0.054* (1.780); 0.074** (2.333); -0.135** (-2.438); -0.057 (-1.352)
  - =3, No Caste/Tribe: -0.018 (-0.304); -0.038 (-0.594); 0.200* (1.941); 0.005 (0.062)

### State-level fixed effects (selected coefficients)
- State = 2, Himachal Pradesh: 0.947*** (5.998); 1.076*** (6.173); 1.296*** (4.763); 0.677*** (3.325)
- State = 3, Punjab: 1.197*** (10.790); 1.395*** (11.787); 1.587*** (7.917); 0.656*** (4.485)
- State = 6, Haryana: 1.527*** (17.674); 1.764*** (18.594); 1.390*** (9.108); 0.947*** (8.431)
- State = 9, Uttar Pradesh: 1.385*** (20.358); 1.667*** (21.640); 0.815*** (6.350); 0.537*** (6.265)
- State = 10, Bihar: 1.629*** (20.667); 1.911*** (21.991); 1.163*** (8.157); 0.825*** (8.414)
- State = 16, Tripura: 1.349*** (13.110); 1.516*** (13.594); 1.555*** (8.670); 0.740*** (5.696)
- State = 26, Maharashtra: 0.884*** (10.817); 1.180*** (13.080); 0.011 (0.069); 0.315*** (3.098)
- State = 31, Kerala: 1.104*** (6.670); 1.269*** (7.080); 1.433*** (4.219); 0.030 (0.136)
- State = 35, Telangana: 0.359*** (3.341); 0.713*** (6.211); -0.448** (-2.266); -0.598*** (-4.491)
- State = 30, Lakshadweep: -0.683 (-1.559); -0.731 (-1.364); 0.397 (0.628); -1.537** (-2.156)
- (Full list of states included in table with coefficients and z-statistics for all four outcomes.)

### Time effects
- Year = 2015: -0.097* (-1.943); -0.039 (-0.756); -0.584*** (-6.350); 0.207*** (3.011)
- Year = 2019: 0.010 (0.172); 0.045 (0.751); -0.511*** (-4.793); 0.387*** (5.054)

### Model constants
- Constant: -4.419*** (-15.171); -5.209*** (-17.091); -2.667*** (-5.112); -5.238*** (-12.980)

*Annex IV. Logistic Regression Results — wpiea2024239-print-pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024239-print-pdf.pdf_