Household Decision Making with Violence

Transcription

1 Implications for Transfer Programs Trinity College Dublin ASSA, 2018

2 Prevalence PREVALENCE of Intimate Partner Violence 1 in 3 women throughout the world will experience physical and/or s violence by a partner or sexual violence by a non-pa 25.4% WHO European Region 29.8% WHO Region of the Americas KEY: 24.6% 23.2% Western Pacific Region 37.0% High income WHO Eastern Mediterranean Region 36.6% 37.7% South-East Asia Region WHO African Region Region of the America African Region Eastern Mediterranean Identification and European Region South-East Asia Region Western Pacific Region High income countries Map showing prevalence of intimate partner violence by WHO region HEALTH exposed partner violence 1 out ofimpact: every Women 3 women hasto intimate been physically or are sexually abused by an intimate partner. Mental Health Sexual and Reproductive Health Death and Injury

3 Transfers to Women and Violence Governments of developing countries have social assistance programs that give transfers to women. The implicit assumption is that transfers allow women to achieve better outcomes for themselves and for their children. Transfers to women can reduce violence by making women less economically dependent on their partners; alleviating poverty stress. But can also increase violence by threatening men s dominant position; increasing the resources men can appropriate through violence.

4 Research Questions How does intimate partner violence respond to transfers to women? Does such response vary when the transfer is in-kind or in-cash? How to deliver transfers to women to minimize violence?

5 This paper 1. Propose a model of household decision making in which the husband can use violence to solve spousal disagreement; violence reduces female labor productivity. 2. Estimate the model using data from Food, Cash, or Voucher, a randomized controlled trial giving in-kind or cash transfers to poor households in Ecuador. 3. Make out-of-sample predictions and simulate a policy giving in-kind or cash transfers to women in poor households, at the national level.

6 Main Findings 1. In-kind transfers have an additional margin in the reduction of violence, relative to cash transfers. 2. Delivering the transfers in-kind is cost-effective. 3. Introducing in-kind transfers at the national level can reduce violence.

7 Contributions Theoretical Contribution Depending on the level of disagreement, any transfer is potentially extra-marginal. In-kind and cash transfers have different effects on violence. Empirical Contribution Complement the results of a reduced-form impact evaluation. Make out-of-sample predictions relevant at the national level. Provide a market value for the cost of violence in an easily interpretable scale that can be used for cost-benefit analysis.

8 Related Literature Collective model of the household with endogenous weights: Chiappori (1988); Basu (2006); Iyigun & Walsh (2007); Attanasio & Lechene (2014). decision making with instrumental violence: Tauchen, Witte & Long (1991); Bloch & Rao (2002); Bowlus & Seitz (2006); Eswaran & Malhotra (2011); Anderberg & Rainer (2013). In-kind vs cash transfers: Cunha (2014); Cunha, De Giorgi & Jayachandran (2015). Effect of cash transfers on violence: Angelucci (2008); Bobonis, Gonzalez-Brenes & Castro (2013); Hidrobo & Fernald (2013); Hidrobo, Peterman & Heise (2016).

9

10 Problem of the max q,l f,d,v µ(v, ω f )u f (Q, q) + (1 µ(v, ω f )) u m (Q, q) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c

11 Problem of the max q,l f,d,v µ(v, ω f )u f (Q, q) + (1 µ(v, ω f )) u m (Q, q) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c Q is a home produced public good, q is a market acquired public good, d is a market input of home production, f has a relative preference for Q,

12 Problem of the max q,l f,d,v µ(v, ω f )u f (Q, q) + (1 µ(v, ω f )) u m (Q, q) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c τ k is a non-marketable in-kind transfer, τ c is a cash transfer,

13 Problem of the max q,l f,d,v µ(v, ω f )u f (Q, q) + (1 µ(v, ω f )) u m (Q, q) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c v is violence, ω f = w f +τ k +τ c w m marriage, is the potential female income outside the µ(v, ω f ) is increasing in ω f and decreasing in v, Γ(v) is decreasing in v. The goal is to recover Γ(v) and µ(v, ω f ).

14 Optimality Conditions Technology of Home Production For any input z {d, 1 l f }, Q z q z = µ(v, ω f ) uf q + (1 µ(v, ω f )) um q µ(v, ω f ) uf Q + (1 µ(v, ω f )) um Q Ratio between the marginal productivity and the marginal cost of the input z = Ratio between the household marginal willingness to pay for home good and the market good

15 Optimality Conditions Technology of Home Production Under the separability assumption, Q (1 l f ) Q d = q (1 l f ) q d Relative marginal productivity of female labor and the market input = Relative marginal cost of female labor and the market input

16 Optimality Conditions Violence µ(v, ω f ) uf m = v + [µ(v, ω f ) uf Q + (1 µ(v, ω f )) um Q [µ(v, ω f ) uf q + (1 µ(v, ω f )) um q ] Q v ] q v Marginal benefit of violence = Marginal cost of violence

17 Bargaining u m Cash In-Kind u m no transfer u f no transfer u f The weighted sum of the utilities is a short-cut for a bargaining problem.

18 Bargaining u m Cash In-Kind u m no transfer µ u f no transfer u f In the absence of violence, the allocation is Pareto-efficient.

19 Bargaining u m Cash In-Kind u m no transfer Violence u f no transfer u f Yet the male can use violence to increase his say in the household decisions.

20 Bargaining u m Cash In-Kind u m no transfer Violence Violence u f no transfer u f But violence comes at the cost of destroying the overall resources available.

21 In-kind vs Cash Transfers u m cash u m in-kind u m no transfer No Transfer Cash In-Kind u f no transfer uf in-kind, cash Consider a transfer that is infra-marginal for the female, but extra-marginal for the male.

22 In-kind vs Cash Transfers u m cash u m in-kind u m no transfer No Transfer Cash In-Kind u f no transfer uf in-kind, cash The utility gains the husband can appropriate are lower when the transfers are in-kind.

23 Cash Transfers and Violence Violence ucash,v=1 m ucash,v=0 m Violence There is a trade-off between male s say in the household decisions and the overall resources available.

24 In-kind Transfers and Violence u m In-kind,v=0,1 Violence Violence There is a trade-off between male s say in the household decisions and the overall resources available.

25 In-kind vs Cash Transfers and Violence Violence u m cash,v=1 uin-kind,v=0,1 m ucash,v=0 m Violence Violence In-kind transfers make violence less productive as an appropriation device.

26

27 Ecuador 35% of women have been physically abused by an intimate partner, yet 90% of the victims are still married to the perpetrator. The main social assistance program, Bono de Desarrollo Humano gives transfers to women and covers at least 40% of the population.

28 "Food, Cash or Voucher" Type of program Randomized control trial Year 2011 Objective Improving nutrition Target population Poor households Payee Women Duration 6 months Treatment Modality Time of observation 40 dollars monthly transfer (10% of monthly income) In-Kind (Food or Voucher) or In-Cash Baseline and follow-up Female labor time allocation and wages demographics and food expenses Intimate partner violence Impact Evaluation Hidrobo, Peterman & Heise (2016). Main Result The program reduces violence by 6 to 7 percentage points. The effects do no differ across treatment arms.

29 Effects of Food, Cash, or Voucher v ij1 = c + β in-kind T in-kind i + β cash T cash i + v ij0 + φ j + e ij Full Sample Not Working Female Working Female Violence at baseline: 16% 16% 17% Any transfer * * (0.033) (0.035) (0.050) In-kind * ** (0.035) (0.037) (0.052) Cash (0.037) (0.038) (0.057) p-value: In-Kind vs. Cash Clusters N 1,230 1,

30 Heterogeneous Effects Use the to understand heterogeneity. Women not working at baseline The technology of home production requires d > τ k. The transfer is infra-marginal for both agents. Lower reduction of violence, as violence is less costly. Women working at baseline The transfer can be extra-marginal for one of the agents. In-kind transfers resolve part of the spousal disagreement. Higher reduction of violence, as violence is more costly.

31

32 1. Impose functional forms. 2. Use the optimality conditions and identification restrictions to recover Γ(v) and µ(v, ω f ). 3. Use the recovered parameters and functional forms to simulate the model. 4. Simulate the effect of a policy giving in-kind or cash transfers to women in poor households, at the national level.

33 Functional Forms Productivity cost of violence Γ(v) = e γ(v) (0, 1] and e γ(0) = 1 Female relative weight Technology of home production Utility of the female µ(v, ω f ) = µ(v) Q = e γ(v) (d + τ k ) θ (1 l f ) 1 θ u f (Q, q) = α f i log(q) + log(q) for every household i Utility of the male u m (Q, q) = log(q) + log(q) for all households

34 Problem of the ( ) max µ(v) αi f log(q) + log(q) q,l f,d,v st. Q = e γ(v) (d + τ k ) θ (1 l f ) 1 θ q + d = e γ(v) l f w f + w m + τ c Observable from the data v, l f, 1 l f, e γ(v) l f w f, w m, d, τ k, τ c To be Identified θ, e γ(v), α f i, µ(v) + (1 µ(v)) (log(q) + log(q))

35 Optimality Conditions Technology of home production Relative marginal productivity of female labor = and the market input Relative marginal cost of female labor and the market input Violence Marginal benefit of violence = Marginal cost of violence

36 Technology of home production 1. Optimality condition: Q (1 l f ) Q d 2. Replace the functional forms: 1 θ θ = d + τ k (1 l f ) 3. Apply logs: ( ) dit + τ k,it log (1 l f,it ) w f,it q (1 l f ). q d = e γ(v) w f. ( ) θ = log + γ(v it ) + ɛ it. 1 θ 4. Estimate θ and e γ(v) through OLS: ( ) dit + τ k,it log = β 0 + β 1 v it + β 2 vit ɛ it. (1 l f,it ) ω f,it

37 Violence 1. Optimality condition: µ(v, ω f ) v [ uf m = + µ(v, ω f ) uf Q + (1 µ(v, ω f )) um Q [ µ(v, ω f ) uf q + (1 µ(v, ω f )) um q ] Q v ] q v. 2. Replace the functional forms and assume µ(v) = Γ(v) δ e κ : (1 + ρ) = ( αi f 1 ) ε µ(v) log(q) µ v 1 ρ is the ratio between female labor income and q. ε µ v and ε Γ v are the elasticity of the female relative weight and the productivity cost. δ is a new parameter to be identified e κ is a constant that captures female s weight in the absence of violence. ε Γ v δ

38 Violence 3. Apply logs: log (1 + ρ it ) + (log(q it ) + 1) log ( α f i 1 ) + κ + δγ(v it ) + log (δ 1) + ɛ it, }{{} η it 4. Estimate the α i f through a household FE-OLS: ( log (1 + ρ it ) + log( Q ) it ) + 1 = a i + η it. 5. Use the residuals to estimate δ through a NLLS, and recover µ(v): ˆη it κ = δ γ(v it ) + log (δ 1) + ɛ it

39

40 Trade-Off of the Perpetrator Productivity cost of violence On average, violence destroys 4% of female labor productivity with a market value of 10 dollars a month. Q it = e 0.85v 2 it (dit + t k,it ) 0.86 (1 l f,it ) 0.14 Effect of violence on weights On average, violence reduces the female say in the household decision making by 12%. µ(v) = 1 2 e3.05 γ(v) Trade-off It is is if, perpetrators were willing to sacrifice one day of female labor income every month (10 dollars) to reduce their partners say by 12%.

41 Predicted Prevalence of Violence Prevalence of Violence No Transfer % Food, Cash, or Voucher 8.23% Cash transfers (only) 9.86 % In-kind transfers (only) 7.41 % 17 out of every 100 women beneficiary of Food, Cash, or Voucher are victims of intimate partner violence. A cash transfer reduces violence for 7 of these 17 women. An in-kind transfer reduces violence for 10 of these 17 women.

42 Cost-benefit Hidrobo, Hoddinott, Peterman, Margolies & Moreira (2014) suggest that the monthly costs of providing a transfer for Food, Cash, and Voucher are: Food dollars Cash 2.99 dollars Voucher 3.27 dollars The 8.5 dollars cost difference of delivering the transfers in-kind instead of in-cash are offset by the 10 dollars monthly reduction of income per victim of violence.

43

44 Scaling-up the Program Use the cross-sectional national representative data. Concentrate among the households beneficiaries of Bono de Desarrollo Humano. Assume the technology of home production (θ), the productivity cost of violence (e γ(v) ), and the effect of violence on weights (µ(v)) are the same for all poor households. The disagreement in the household (αi f ) is not observable. 1. Use Food, Cash, or Voucher. 2. Regress α f i on household observable characteristics. 3. Use these coefficients to predict α f i at the national level. 4. Use θ, e γ(v), µ(v) from Food, Cash, or Voucher and the distribution of α i f and the empirical distribution of w f to at the national level simulate the model.

45 National Level Encuesta Nacional sobre Relaciones Familiares y Violencia de Género contra las Mujeres Representative National Year 2011 demographics, wages, and violence Bono de Desarrollo Humano Social assistance program Target population Poor households Treatment 50 dollars monthly cash transfer (2011) Payee Women Prevalence of violence 37%

46 Increasing the Size of the Transfers The differential effect of in-kind and cash transfers is not linear in the size of the transfer.

47 Depending on the level of disagreement, any transfer if potentially extra-marginal. Not all forms of empowerment are equally relevant for all women. Even abstracting from the human right dimension, intimate partner violence imposes productivity cost. The fact that a woman is no longer abused represents an economic gain of 10 dollars a month.

48 Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level

49 Definition of violence

50 Index of Violence Count of the different forms of violence that reported by the respondent. Physical violence Sexual violence Range v [ 0 9, 9 ] 9 Average v = 2 9 punch, kick, strangle, attack with weapon, threaten with a weapon, push, or slap forced sex, non approved sex acts Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

51 Extensions of the model

52 Partially Marketable In-kind Transfer Problem of the max q f,q m,l f,d,v µ(v, ω f )u f (Q, q f, v) + (1 µ(v, ω f )) u m (Q, q m, v) st. Q = F (d + φτ k, Γ(v) (1 l f )) q f + q m + d = Γ(v)l f w f + w m + τ c + (1 φ)τ k Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher A share φ > 0 of the in-kind transfer τ k is non-marketable. Equivalent to τ k = φτ k and τ c = τ c + φτ k, with τ k τ c. Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

53 Private Goods and Direct (dis)utility from Violence Problem of the max q f,q m,l f,d,v µ(v, ω f )u f (Q, q f, v) + (1 µ(v, ω f )) u m (Q, q m, v) st. Q = F (d + τ k, Γ(v) (1 l f )) q f + q m + d = Γ(v)l f w f + w m + τ c Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

54 Private Goods and Direct (dis)utility from Violence Optimality Conditions Q [µ(v, ω f ) uf d Q 1 l f Q + (1 µ(v, ω f )) um Q [µ(v, ω f ) uf Q + (1 µ(v, ω f )) um Q ] = 1 ] [µ(v, ω f ) uf 2 q f + (1 µ(v, ω f )) um q q m d ] = 1 ] [µ(v, ω f ) uf 2 q f + (1 µ(v, ω f )) um q m µ(v, ω f ) uf q f = (1 µ(v, ω f )) um q m Definition of violence Extensions of the model q Alternative (1 l f ) interpretation of the model Food, Cash, or Voucher µ(v, ω f ) uf m v + (1 µ(v, ω f )) um v = [µ(v, ω f ) uf Q + (1 µ(v, ω f )) um Q ] Q v [µ(v, ω f ) uf q f q f v + (1 µ(v, ω f )) um q m q m v + µ(v, ω f ) uf v ] Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

55 Alternative interpretation of the model

56 Alternative Interpretation max q,l f,d,v max q,l f,d,v u m (Q, q) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c u f (Q, q) u f (Q, q, v) [ ] u m (Q, q) + λ u f (Q, q) u f (Q, q, v) st. Q = F (d + τ k, Γ(v) (1 l f )) q + d = Γ(v)l f w f + w m + τ c Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

57 Food, Cash, or Voucher

58 List of Pre-Approved Goods endix A. Example of poster on nutrition messaging M. Hidrobo et al. / Journal of Development Economics 107 (2014) back Fig. A.1. WFP Poster. Source: Hidrobo, Hoddinott, Peterman, Margolies & Moreira (2014) endix B. Supplementary Tables Fiedler, J.L., Villalobos, C.A., De Mattos, A.C., An activity-based cost analysis of the Honduras Community-Based, Integrated Child Care (AIN-C) programme. Health Policy Plan. 23 (6), Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level

59 Sample Flowchart Baseline (n=2357) Female respondent aged at baseline (n=2252) Married or at union at baseline (n=1488) Head of household or spouse (n=1439) Alone at time of the interviews (n=1245) Same respondent at baseline and followup (n=1230) Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

60 Descriptive Statistics All Control Treatment p-value In-Kind Cash In-kind vs. Cash Panel A. Demographics No. of household members Male head of household Panel B. Intimate Partner Violence Any type of violence Physical or sexual violence Panel C. Variables for the income a day day expenses in food Female employed Female labor income a day Female hours of work a day Female hours of household work a day Male employed Male labor income a day back Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level

61 Recovering female wages

62 Per Hour Wage from Female Labor Income The female per hour wage from the data is e γ(v) w f. To disentangle w f from e γ(v), use a Heckman Two-Step procedure among the female-working households, as if the wages of abused working females were not observed. As exclusion restrictions, use the cohabitation status of the couple and the number of children. The female wage variable used for the estimation is, { wf if v = 0 and l w f = f > 0 ŵ f if v = 1 and l f > 0 where ŵ f are the Heckman Two-Step predicted female relative wages. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

63 Heckman Log Wages Selection Age * (0.0165) (0.0273) Age, squared (0.0002) (0.0003) Female s education years ** (0.0158) (0.0199) Female with secondary education or more (0.1081) (0.1617) Female s hours of work a day *** (0.1758) (0.0491) Female s hours of work a day, squared *** (0.0119) (0.0040) Carchi * (0.0784) (0.1243) Married couple * (0.1081) No. children form 0 to (0.0910) No. children from 6 to (0.0541) Constant (0.9177) (0.5322) Lambda 0.90 Clusters 141 N 922 Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

64 Distribution of Female Relative Wages Frequency Log of wife's relative Wage Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

65 estimation

66 Identification Technology The optimality condition for the technology of home production is Q (1 l f ) Q d = q (1 l f ). q d Replacing with the Cobb-Douglas functional form for Q, 1 θ θ d + τ k (1 l f ) = e γ(v) w f. Applying logs, θ and e γ(v) are identified through ( ) ( ) dit + τ k,it θ log = log + γ(v it ) + ɛ it, (1 l f,it ) w f,it 1 θ where ɛ is a measurement error term uncorrelated with v. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

67 1. Estimate θ and e γ(v) through ( ) dit + τ k,it log = β 0 + β (1 l f,it ) ω f,it }{{} 1 v it + β 2 vit ɛ it. }{{} log( θ 1 θ ) γ(v) polynomial of v Use the estimated θ and e γ(v) to recover Q. 2. Estimate the α i f through a household FE-OLS ( log (1 + ρ it ) + log( Q ) it ) + 1 = a }{{} i +η it. log(α f 1) i 3. Use the residuals of the previous step to estimate δ through a NLLS, and recover µ(v) = e δγ(v) e κ. ˆη it κ = δ γ(v it ) + log (δ 1) + ɛ it 4. Use θ, e γ(v), µ(v), the distribution of α i f, and the empirical distribution of w f to simulate the model. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

68 Identification Violence The optimality condition for violence is µ(v, ω f ) uf m = v + [µ(v, ω f ) uf Replacing with the functional forms: µ(v) v ( ) [ 1 αi f log(q) = µ(v) αf i Q + (1 µ(v, ω f )) um Q [µ(v, ω f ) uf q + (1 µ(v, ω f )) um q Q + (1 µ(v)) 1 Q [ + µ(v) 1 q + (1 µ(v)) 1 q ] Q v ] q v ] Q e γ(v) e γ(v) v ] l f w f e γ(v) v Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

69 Identification Violence After some algebra, (1 + ρ) = ( α f i 1 ) µ(v) [ log(q) εµ v ε Γ v ] 1. ρ = eγ(v) l f w f q is the ratio between female labor income and q. ε µ v = µ(v) v ε Γ v = eγ(v) v v µ(v) v e γ(v) is the elasticity of the female relative weight. is the elasticity of the productivity cost. Assume that ε µ [ v ε Γ = δ µ(v) = e γ(v)] δ e κ. v δ is a new parameter to be identified. e κ is a constant that captures female s weight in the absence of violence, µ(0) = e κ = 1 2. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

70 Identification Violence The optimality condition for violence transforms into ( ) log (1 + ρ it ) + (log(q it ) + 1) log αi f 1 where ɛ is a measurement error term. + κ + δγ(v it ) + log (δ 1) + ɛ it, }{{} η it Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

71 1. Estimate θ and γ(v) through ( ) dit + τ k,it log = β 0 + β (1 l f,it ) ω f,it }{{} 1 v it + β 2 vit ɛ it. }{{} log( θ 1 θ ) γ(v) polynomial of v Use the estimated θ and e γ(v) to recover Q. 2. Estimate the α i f through a household FE-OLS ( log (1 + ρ it ) + log( Q ) it ) + 1 = a }{{} i +η it. log(α f 1) i 3. Use the residuals of the previous step to estimate δ through a NLLS, and recover µ(v) = e δγ(v) e κ. ˆη it κ = δ γ(v it ) + log (δ 1) + ɛ it 4. Use θ, e γ(v), µ(v), the distribution of α i f, and the empirical distribution of w f to simulate the model. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

72 1. Estimate θ and γ(v) through ( ) dit + τ k,it log = β 0 + β (1 l f,it ) ω f,it }{{} 1 v it + β 2 vit ɛ it. }{{} log( θ 1 θ ) γ(v) polynomial of v Use the estimated θ and e γ(v) to recover Q. 2. Estimate the α i f through a household FE-OLS ( log (1 + ρ it ) + log( Q ) it ) + 1 = a }{{} i +η it. log(α f 1) i 3. Use the residuals of the previous step to estimate δ through a NLLS, and recover µ(v) = e δγ(v) e κ. ˆη it κ = δ γ(v it ) + log (δ 1) + ɛ it 4. Use θ, e γ(v), µ(v), the distribution of α i f, and the empirical distribution of w f to simulate the model. Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

73 Distributions of disagreement in preferences

74 Distribution of Disagreement in Preferences Frequency Preference Parameter Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

75 Scaling-up the program at the national level

76 Descriptive Statistics Food,Cash,or Voucher Bono de Desarrollo Humano No. of household members Male head of household Married couple No. children form 0 to No. children from 6 to Female age Male age Couple age difference Female education years Male education years Female more educated than male Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

77 Distribution of Female Relative Wages Density Log of wife's relative wage - INEC Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back

78 Observable Characteristics Predicting Disagreement in Preferences Observable Characteristic αi f No. of household members 0.53 (0.34) Male head of household 3.26 (2.45) Married couple 0.16 (0.83) No. children form 0 to (0.54) No. children from 6 to (0.48) Female age 0.04 (0.04) Couple age difference 0.06 (0.06) Male education years 0.02 (0.14) back Observable Characteristic αi f Female education years (0.16) Female more educated than male (1.13) Female employed 0.92 (1.14) Female labor income a day 0.21 (0.16) Female hours of work a day (0.19) Female hours of household work a day (0.11) Male employed 1.85 (3.19) Male labor income a day 0.11 (0.08) Male hours of work a day 0.08 (0.16) Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level

79 Distribution of Disagreement in Preferences National Level Density Predicted Disagremment in Preference Parameter - INEC Definition of violence Extensions of the model Alternative interpretation of the model Food, Cash, or Voucher Recovering female wages estimation Distributions of disagreement in preferences Scaling-up the program at the national level back