Evidence shows that transfers targeted to women can reduce intimate partner violence (IPV), but little evidence exists on the effects of transfers targeted to men. Whereas impact pathways for transfers to women are posited to include increased women's empowerment and reduced household poverty-related stress and conflict, the former is unlikely to occur from transfers to men. Transfers to men are common in Africa, where household structure tends to differ from Latin America, the context for much of the existing evidence on IPV. We use data from a randomly-assigned national cash transfer program in Mali, targeted to (mostly male) household heads, to analyze impacts on IPV. Given high prevalence of polygamy in the sample, we disaggregate impacts by whether the household head is polygamous. We find that, on average, there are no reductions in IPV, but in the nearly 40% of households that are polygamous, there are significant reductions in physical and emotional IPV, as well as controlling behaviors. Analysis of mechanisms reveals significant reductions in the household head's stress and reductions in reported disputes only in polygamous households, confirming that poverty-related stress may drive the results. We discuss potential explanations for why patterns differ by polygamy status.
1. Cash transfers, polygamy, and intimate
partner violence: Experimental evidence
from Mali
Rachel Heath, Melissa Hidrobo, Shalini Roy
PHND Brown Bag
December 13, 2017
2. Motivation
Intimate partner violence (IPV) is the most pervasive form of violence globally: 1
in 3 women affected in her lifetime
o In Mali, nearly 4 in 10 women have experienced physical violence in their
lifetime (DHS 2012/2013)
A growing body of literature shows that cash transfer programs in developing
countries can reduce IPV (Angelucci 2008; Bobonis et al 2013; Hidrobo et al 2016; Haushofer and Shapiro
2013; Perova and Vakis 2013; Roy et al 2017)
o Find decreases in physical violence ranging from 5-11 percentage points
although some subgroups found at greater risk for increases in violence
o Nearly all target women and are in Latin America
Policy relevant given that ~1 billion transfer beneficiaries in developing world
(DFID 2011)
3. Motivation
Knowledge gaps remain
1. What happens to IPV when transfers are targeted to men?
o Many cash transfer programs around the world target household heads,
who are usually male
2. Do existing findings generalize globally?
o Polygamy is common in Mali, as in much of West Africa, and different
wives may be treated differently and have different bargaining power
3. What are the pathways through which transfers affect IPV?
o Evidence on pathways is scarce
4. Motivation
Pathways (not exhaustive and not exclusive) (Buller et al 2017)
1. Improved economic security and emotional wellbeing;
o At their core, CTs are economic safety nets designed to reduce poverty
o Recent studies on the psychology of poverty show links between poverty and mental health
(Lund et al 2010; Haushofer and Fehr 2014)
2. Reduced intra-household conflict;
o Increased access to cash decreases conflict by reducing arguments over tight budgets and
daily money needed to run the household
o However, could also increase conflict if used on temptation goods
3. Increased women’s empowerment
o Bargaining models predict that transfers (targeted to women) improve a woman’s bargaining
position within the household by increasing her threat point
o However, an increase in women’s resources may put a woman at increased risk of IPV if men
feel threatened and use violence to reassert authority or extract resources
5. Source: AM Buller, A Peterman, M Ranganathan, A Bleile, M Hidrobo & L Heise (2017). A Mixed Methods Review of
Cash Transfers & Intimate Partner Violence in LMICs (working paper)
6. Motivation
How pathways play out likely depend on household structure
oIf polygamous households are less cooperative and have more conflict
(Barr et al 2017), this could affect the impact of CTs on IPV.
oWithin polygamous household which wife is more affected may depend
on her bargaining power
How pathways play out likely depend on targeting
oLast pathway (increased women’s empowerment), is likely not relevant
if women are not targeted, unless husband redistributes transfers to
wife.
oTransfers targeted to men may even increase men’s bargaining power
relative to women’s
7. Our study
Investigate the impacts on IPV of a national cash transfer program in Mali -
Programme de Filets Sociaux (Jigisémèjiri)
oTargeted beneficiary is household head – usually male
Investigate whether impacts vary by household structure and in particular
by polygamy status
Collect and analyze detail data on impact pathways – with a special focus
on men – the main beneficiaries of the program and the perpetrator of
violence in our study
8. Jigisémèjiri
Aims:
oIn the short term: alleviate current poverty and vulnerability by
smoothing and increasing consumption of households through targeted
and regular cash transfers
oIn the medium-long term: improve human capital of children, thereby
reducing intergenerational transmission of poverty by combining cash
transfer (CT) with accompanying measures (AM) and preventive
nutrition packages (PNP)
Targeting: 62,000 poor households across 6 regions of Mali and Bamako
ogeographical- and community-based targeting
9. Jigisémèjiri
3 components
1. Cash transfers (CT): unconditional transfers of 10,000 FCFA/month (or
USD 17/month) paid every trimester
2. Accompanying Measures (AM): trainings 1-2x month on how to use the
transfers, infant and young child feeding, prenatal and postnatal care,
vaccinations, WASH, respiratory infections, income generating activities
o Open to everyone in the commune
3. Preventive Nutrition Packages (PNP): fortified food distributed every
trimester
o Targeted to pregnant women, lactating mothers (Supercereal and oil)
o Children 6-59 months (Supercereal Plus)
10. Study design
Experimental design took into account 2 requests from Gov’t of Mali
o Control group would need to be rolled into the program
o Communes in which PNP would be delivered, would be purposively selected in
consultation with a national technical committee.
Two stage randomized design
o 1st stage: 96 communes are randomly assigned to either Treatment or Control
Groups
o Treatment communes receive CT and AM.
o Control communes receive no intervention the first 1-2 years
o 2nd stage: Within a subset of selected Treatment communes, villages are randomly
assigned to either additionally receive (PNP) or to not receive PNP (Non-PNP)
11. 96 selected communes
76 Treatment communes
Random selection
20 Control communes
19 PNP-Eligible
communes
57 PNP-Ineligible
communes
Non-random selection
50% of villages with
PNP
50% of villages with
no PNP
Random selection
12.
13. Data
Data sources:
1. Baseline (Sept 2014-Feb 2015): before the start of the program
o Beneficiary households with a child 6-23 months
o N=3,080
2. Midline (Aug 2016-Nov 2016): before the control group was rolled into the program
and before the first PNP distribution
o Panel sample: random subset of baseline households, N=2533
o Repeated cross section sample: new households with a child 6-23 months,
N=1,498
3. Qualitative gender study (April-May 2017):
o Gender study focused on intrahousehold relationships
o Beneficiary households in Sikasso, N=44
14. Data
Use baseline and midline panel dataset
Panel dataset composed of 4 questionnaires
o Household survey
o Mother survey
o Decisionmaker and spouse survey
o Anthropometrics
Decisionmaker and spouse survey (only administered at midline)
o Administered on main decisionmaker of the household and his/her spouse <70 years
o If more than one spouse, mother of index child was excluded and then spouse was
randomly sampled
o Included modules on relationship dynamics (IPV, disputes, trust); Stress and anxiety;
decisionmaking; cognitive function; self efficacy
15. Outcomes
WHO Violence Against Women Instrument: questions on different acts
of violence/behaviors in the last 12 months
oEmotional violence (4 questions)
oPhysical violence (6 questions)
oControlling behaviors (7 questions)
2 types of outcome variables constructed
oAny violence in the past 12 months for each category (binary)
oNormalized indices for each category
16. Sample
Eligible Sample:
oWomen who completed the decisionmaker questionnaire (either DM or
spouse): N=2,291
oAge 49 years or less: N=1,839
oIn the household and married at baseline: 1,555
IPV sample
oEligible sample living with their spouse at midline: 1,462
oEligible sample alone at the time of the interview: 1,265
Attrition from Eligible sample to IPV sample is 19%
17. Attrition and Balance of baseline characteristics
N Mean
Control
Mean
Treatment
P-value
of diff.
Normalized
difference
Attrition
Has lived with spouse last 12 months 1,555 0.95 0.94 0.34 -0.06
Answered intimate partner violence 1,555 0.79 0.82 0.57 0.07
Baseline Means
Respondent's age 1,265 32.08 31.97 0.83 -0.01
Respondent is spouse of head of household 1,265 0.78 0.76 0.49 -0.05
Respondent is daughter-in-law of head of household 1,265 0.19 0.19 0.99 0.00
Respondent is literate 1,265 0.04 0.05 0.51 0.04
Husband's age 1,149 44.62 44.34 0.75 -0.03
Husband is literate 1,149 0.12 0.17 0.07 0.14
Polygamous marriage 1,265 0.39 0.37 0.66 -0.04
Household size 1,265 10.17 10.29 0.80 0.03
Number of children 0 to 5 years 1,265 2.70 2.81 0.38 0.08
Number of children 6 to 15 years 1,265 3.08 3.05 0.86 -0.02
Log value of assets 1,208 5.24 5.27 0.87 0.02
18. Monogamous households Polygamous households
N Mean
Control
Mean
Treat
P-value
of diff.
Normaliz
ed diff
N Mean
Control
Mean
Treat
P-value
of diff.
Normali
zed diff
Attrition
Has lived with spouse last 12 months 984 0.94 0.95 0.95 0.01 571 0.96 0.92 0.05 -0.17
Answered intimate partner violence 984 0.76 0.82 0.24 0.16 571 0.86 0.82 0.32 -0.12
Baseline means
Respondent's age 792 30.44 31.42 0.24 0.13 473 34.67 32.91 0.03 -0.22
Respondent is spouse of head of household 792 0.74 0.75 0.72 0.04 473 0.85 0.76 0.07 -0.21
Respondent is daughter-in-law of head of
household
792 0.23 0.19 0.32 -0.10 473 0.13 0.19 0.09 0.18
Respondent is literate 792 0.05 0.04 0.61 -0.04 473 0.02 0.05 0.04 0.19
Husband's age 719 42.48 43.02 0.65 0.05 430 47.83 46.60 0.25 -0.14
Husband is literate 719 0.11 0.16 0.07 0.15 430 0.13 0.18 0.27 0.15
Polygamous marriage 792 0.00 0.00 473 1.00 1.00
Household size 792 8.34 8.82 0.16 0.14 473 13.05 12.80 0.68 -0.05
Number of children 0 to 5 years 792 2.22 2.41 0.06 0.18 473 3.46 3.48 0.88 0.02
Number of children 6 to 15 years 792 2.24 2.51 0.04 0.16 473 4.39 3.96 0.11 -0.19
Log value of assets 754 5.12 5.04 0.68 -0.05 454 5.42 5.65 0.14 0.16
19. Estimation strategy
Take advantage of the randomization and conduct a single difference
Intent-to-Treat estimation
oControl for stratification at the region level
oControl for women characteristics and household characteristics
oCluster SE at the commune level (level of randomization)
Estimate impact on full sample and by monogamous and polygamous
households
20. Results
Table 3: Impact of treatment on IPV
Mean of
control,
non-poly
Mean of
control,
poly
Overall
effect
N Effect on
Non-polyg
N Effect on
Polyg
N Diff polyg
vs non
any physical violence 0.119 0.179 -0.025 1265 0.009 792 -0.074 473 -0.083
(0.025) (0.036) (0.027) (0.029) (0.035)** (0.036)**
index of physical violence 0.071 0.254 -0.157 1265 -0.035 792 -0.349 473 -0.314
(0.094) (0.141) (0.091)* (0.092) (0.117)*** (0.113)***
any emotional violence 0.352 0.509 -0.058 1265 -0.006 792 -0.157 473 -0.151
(0.036) (0.047) (0.034)* (0.042) (0.047)*** (0.055)***
index of emotional violence 0.005 0.381 -0.133 1265 0.021 792 -0.395 473 -0.416
(0.075) (0.124) (0.095) (0.082) (0.149)*** (0.140)***
any controlling behavior 0.528 0.679 -0.056 1265 -0.006 792 -0.152 473 -0.145
(0.038) (0.044) (0.033)* (0.046) (0.046)*** (0.065)**
index of controlling behavior 0.145 0.377 -0.237 1265 -0.139 792 -0.386 473 -0.247
(0.090) (0.111) (0.092)** (0.098) (0.134)*** (0.147)*
21. Table 5: Individual-level IPV outcomes (binary)
Mean of
control,
non-poly
Mean of
control,
poly
Overall
effect
N Effect on
Non-polyg
N Effect on
Polyg
N Diff
polyg vs
non
Tries to keep you from seeing your friends 0.074 0.089 -0.009 1262 0.001 789 -0.020 473 -0.021
(0.020) (0.027) (0.019) (0.019) (0.030) (0.032)
Tries to restrict contact with your family 0.091 0.145 -0.048 1257 -0.026 786 -0.082 471 -0.056
(0.022) (0.034) (0.021)** (0.022) (0.036)** (0.037)
Insists on knowing where you are at all times 0.360 0.400 -0.076 1259 -0.043 789 -0.127 470 -0.084
(0.036) (0.047) (0.043)* (0.055) (0.064)** (0.081)
Ignores you and treats you indifferently 0.108 0.144 -0.048 1257 -0.036 789 -0.073 468 -0.036
(0.023) (0.033) (0.025)* (0.025) (0.043)* (0.046)
Gets angry if you speak with another man 0.276 0.330 -0.056 1257 -0.025 788 -0.101 469 -0.077
(0.034) (0.045) (0.031)* (0.037) (0.047)** (0.055)
Is often suspicious that you are unfaithful 0.091 0.126 -0.040 1262 -0.036 790 -0.037 472 -0.001
(0.022) (0.032) (0.019)** (0.018)** (0.037) (0.041)
Expects you to ask his permission before seeking
health care for yourself
0.392 0.495 -0.062 1264 -0.022 792 -0.142 472 -0.120
(0.037) (0.048) (0.036)* (0.042) (0.052)*** (0.063)*
Does not trust you with money 0.165 0.259 -0.003 1254 0.025 787 -0.052 467 -0.076
(0.028) (0.042) (0.035) (0.039) (0.050) (0.054)
Insulted you or made you feel bad about yourself 0.153 0.205 -0.029 1261 -0.005 789 -0.081 472 -0.076
(0.027) (0.038) (0.024) (0.030) (0.031)*** (0.041)*
Belittled or humiliated you in front of other
people
0.063 0.143 -0.028 1259 0.007 789 -0.085 470 -0.092
(0.018) (0.033) (0.026) (0.023) (0.039)** (0.036)**
Done things to scare or intimidate you on purpose 0.131 0.205 -0.057 1262 -0.019 791 -0.119 471 -0.100
(0.025) (0.038) (0.032)* (0.028) (0.051)** (0.046)**
Threatened to hurt you or some one you care
about
0.028 0.071 -0.005 1263 0.010 791 -0.026 472 -0.036
(0.013) (0.024) (0.016) (0.015) (0.029) (0.030)
Slapped you or thrown something at you that
could hurt you
0.057 0.099 -0.030 1262 -0.003 790 -0.075 472 -0.073
(0.017) (0.028) (0.019) (0.021) (0.027)*** (0.030)**
Pushed you or shoved you or pulled your hair 0.046 0.045 -0.020 1264 -0.014 791 -0.029 473 -0.015
(0.016) (0.020) (0.013) (0.015) (0.016)* (0.016)
Hit you with his fist or with something else that
could hurt you
0.074 0.082 -0.011 1260 -0.009 790 -0.007 470 0.002
(0.020) (0.026) (0.017) (0.018) (0.033) (0.035)
Kicked you, dragged you or beat you up 0.017 0.054 -0.018 1259 0.001 787 -0.048 472 -0.050
(0.010) (0.021) (0.010)* (0.010) (0.015)*** (0.016)***
Choked or burnt you on purpose 0.034 0.054 -0.015 1252 0.001 781 -0.043 471 -0.045
(0.014) (0.021) (0.016) (0.016) (0.020)** (0.018)**
22. Pathways
We investigate the impact of Jigisemejiri on 3 theorized pathways
1. Improved economic security and emotional wellbeing;
o Emotional well-being of men, the perpetrators – stress, anxiety, and self-efficacy
2. Reduced intra-household conflict;
o “marital capital” – level of trust, quality of relationship, and disputes
3. Increased women’s empowerment
o Descriptive evidence shows that household heads were main beneficiaries (88%);
recipients of transfers (83%); and decisionmaker with respect to transfer (77%)
o Find little evidence of increased empowerment on mobility, decisionmaking, and assets
(results not shown)
23. Pathway 1: Emotional well-being of spouse
Table 8: Impact of treatment on emotional well-being (male decisionmakers)
Mean of
control,
non-poly
Mean of
control,
poly
Overall
effect
N Effect on
Non-polyg
N Effect on
Polyg
N Diff polyg
vs non
Perceived stress scale (0-40) 15.913 16.000 -1.349 1147 -1.070 717 -1.905 430 -0.836
(0.465) (0.586) (0.787)* (0.894) (0.868)** (0.826)
Anxiety (worry) index 0.147 0.286 -0.254 1147 -0.186 717 -0.358 430 -0.172
(0.075) (0.093) (0.098)** (0.101)* (0.131)*** (0.124)
Low self efficacy index -0.007 0.054 -0.131 1147 -0.095 717 -0.215 430 -0.120
(0.080) (0.104) (0.085) (0.084) (0.135) (0.139)
24. Pathway 2: Intrahousehold conflict
Table 6: Impact of treatment on women's reported relationship quality
Mean of
control,
non-poly
Mean of
control,
poly
Overall
effect
N Effect on
Non-polyg
N Effect
on
Polyg
N Diff polyg
vs non
Any dispute 0.295 0.357 -0.021 1264 0.026 792 -0.092 472 -0.118
(0.034) (0.045) (0.043) (0.041) (0.057) (0.042)***
Dispute index 0.010 0.167 -0.017 1264 0.133 792 -0.223 472 -0.356
(0.059) (0.172) (0.093) (0.098) (0.146) (0.153)**
Relationship quality index -0.088 -0.156 0.095 1246 0.076 779 0.117 467 0.040
(0.086) (0.117) (0.080) (0.104) (0.125) (0.166)
Relationship trust index -0.056 -0.290 0.000 1265 -0.085 792 0.152 473 0.237
(0.070) (0.108) (0.087) (0.076) (0.159) (0.160)
25. Summary of results
We find large and significant negative impacts of Jigisemejiri on IPV in
polygamous households and no impact in monogamous households
Suggestive evidence that impact on IPV in polygamous households is
through improved emotional well-being of men and less disputes in
household (although latter is not significant)
Why is there such a large impact on IPV in polygamous households only?
oIs it something inherent about polygamy or are we capturing something
else that is correlated with polygamy?
26. Correlates of Polygamy and IPV
DEPENDENT VARIABLE any physical violence any emotional violence
any controlling
behavior
Polygamous 0.063* -0.015 0.159*** 0.133** 0.152** 0.150**
[0.030] [0.038] [0.053] [0.052] [0.066] [0.065]
Age -0.007** -0.011*** -0.010***
[0.003] [0.003] [0.003]
age gap (husb-wife) 0.008*** -0.007* -0.004
[0.003] [0.003] [0.003]
household members 0.016*** 0.016* 0.009
[0.003] [0.008] [0.007]
log(assets per member) -0.013 -0.011 -0.002
[0.011] [0.016] [0.015]
Observations 288 288 288 288 288 288
R-squared 0.032 0.123 0.045 0.088 0.089 0.124
27. VARIABLES any physical violence any emotional violence
any controlling
behavior
TreatCom 0.004 0.178* -0.001 -0.156 0.001 -0.092
[0.031] [0.100] [0.040] [0.141] [0.045] [0.143]
polygamous 0.082*** 0.001 0.170*** 0.136*** 0.169*** 0.159**
[0.030] [0.033] [0.050] [0.051] [0.061] [0.062]
treat X polygamous -0.081** 0.009 -0.148** -0.128** -0.142** -0.158**
[0.036] [0.040] [0.058] [0.061] [0.067] [0.070]
household members 0.016*** 0.014 0.006
[0.003] [0.008] [0.007]
treat X household members -0.018*** -0.008 0.001
[0.004] [0.009] [0.008]
treat X age 0.002 0.006* 0.003
[0.003] [0.004] [0.004]
age gap (husb-wife) 0.007*** -0.006* -0.003
[0.002] [0.003] [0.003]
treat X age gap (husb-wife) -0.008*** 0.003 0.001
[0.003] [0.004] [0.004]
log(asset per member) 0.004 -0.005 0.011
[0.009] [0.014] [0.014]
treat X log(asset per member) -0.001 -0.007 -0.009
[0.012] [0.018] [0.017]
Observations 1,265 1,265 1,265 1,265 1,265 1,265
R-squared 0.024 0.039 0.034 0.043 0.052 0.059
28. Why is there such a large impact on IPV in polygamous
households only
Are we capturing something else that is correlated with polygamy?
oWhile higher rates of physical violence and impacts on physical
violence for polygamous households can be explained by other
correlates, this is not true for emotional violence and controlling
behaviors
Is it something inherent about polygamy?
oIn polygamous households there is less cooperation and more conflict
(Barr et al 2017) leading to higher rates of anxiety and disputes, thus
there is more room for impact from CTs
oDifferent bargaining power across wives leads to higher rates of abuse
among wives with less bargaining power, and thus there is more room
for impact from CTs
29. Differences across wives in polygamous households
Table 4: Polygamous women only
Mean of
control,
first wife
Mean of
control,
wife 2+
Effect on
first wife
N Effect on
wife 2+
N Diff, first
wife vs 2+
any physical violence 0.069 0.302 -0.011 242 -0.153 228 -0.142
(0.034) (0.064) (0.037) (0.063)** (0.077)*
index of physical violence -0.096 0.646 -0.101 242 -0.610 228 -0.509
(0.104) (0.267) (0.105) (0.207)*** (0.241)**
any emotional violence 0.448 0.585 -0.128 242 -0.200 228 -0.075
(0.066) (0.068) (0.069)* (0.067)*** (0.104)
index of emotional violence 0.146 0.656 -0.276 242 -0.520 228 -0.245
(0.140) (0.208) (0.134)** (0.204)** (0.200)
any controlling behavior 0.655 0.717 -0.129 242 -0.183 228 -0.055
(0.063) (0.062) (0.071)* (0.064)*** (0.099)
index of controlling behavior 0.210 0.580 -0.303 242 -0.484 228 -0.186
(0.140) (0.174) (0.165)* (0.171)*** (0.215)
30. Conclusion
We find limited impacts on IPV of Jigisemejiri for the full sample
But in the nearly 40% of households that are polygamous, there are
significant reductions in physical and emotional IPV, as well as controlling
behaviors, that range from 8-16 percentage points
Analysis of mechanisms reveals significant improvements in the emotional
well-being of the spouse
To our knowledge this is the first study to look at impacts of a transfer
program on IPV that is targeted to males in the context of West Africa
31. Conclusion
What is it about polygamous households?
oObservable characteristics explain associations between polygamous
households and physical violence, but not between polygamous
households and emotional violence or controlling behaviors
oInherent characteristics suggest it could be due to
oLess cooperation and more conflict in polygamous households
leading to higher levels of anxiety and disputes
oLess bargaining power among 2+ wives
oAnything else?
Talking points
These pathways do not operate in isolation
Within the same programme we can have more than one pathway operating
We do not have clarity on how the effects of these pathways might cancel each other given that some of them (2&3) might end up in either positive or negative outcomes
Not all connections in the pathway have the same amount of evidence
None of the studies included modelled the pathway in its entirety
anthropological literature suggests that co-wife relationships within polygnous households are characterized by conflict but co-wives may cooperate to achieve pragmatic goals.
In particular, circles and communes are chosen based on geographical targeting using poverty maps. In total, 18 circles and 106 communes have been selected across the six regions plus Bamako. All villages within the communes are then given a quota of households that are eligible, based on population size. Community-based targeting is used to select households within villages. The following criteria are used to select households:
Food insecurity based on the number of meals eaten and the number of months household was able to consume from last harvest
Agricultural capacity based on land, livestock, and productive assets
Household size and composition
Labor and employment
Corn-Soya Blend
updated with the currently used terminology: https://www.unicef.org/supply/files/Supercereal_Products_(CSB).pdf
My understanding of the intuition is that you don't want the index to be driven mostly by one variable, which it would do if it happened that one was inherently more variable than the others -- by standarizing, you ensuring that the information added by each variable is the same.
T his is defined as the difference in means between the treatment and control groups, divided by the square root of half the sum of the treatment and control group variances. This provides a scale-invariant measure of the size of the difference