You are on page 1of 11

World Development 138 (2021) 105149

Contents lists available at ScienceDirect

World Development
journal homepage: www.elsevier.com/locate/worlddev

Regular Research Article

Do remittances reduce women’s acceptance of domestic violence?


Evidence from Pakistan
Aniruddha Mitra a,⇑, James T. Bang b, Faisal Abbas c
a
Economics Program, Bard College, 30 Campus Road, Annandale-on-Hudson, NY 12401, USA
b
Department of Economics, St. Ambrose University, Ambrose Hall 306, 518 W. Locust St., Davenport, IA 52803, USA
c
Department of Economics, School of Social Sciences and Humanities, National University of Sciences and Technology, H-12 Sector, Islamabad 44000, Pakistan

a r t i c l e i n f o a b s t r a c t

Article history: There is an emerging interdisciplinary literature that explores the transformation of prevailing gender
norms as communities experience emigration and associated inflows of remittances. This paper focuses
on a key gender norm, namely, women’s perception of experiencing domestic violence. Specifically, we
Keywords: investigate the impact of household access to remittance income on attitudes held by women in the
Remittances household regarding the acceptability of domestic violence. Using data from a Multiple Indicators
Domestic violence Cluster Survey conducted in the Punjab province of Pakistan in 2014, we isolate the causal impact of
Gender norms
remittance income by employing a matching estimator to construct counterfactuals that allow us to com-
Pakistan
South Asia
pare women from otherwise identical households who differ solely in their access to remittances. Our
results indicate that women in households with access to remittances are less likely to accept domestic
violence than women in comparable households without remittance income, though there is consider-
able heterogeneity once we differentiate between different contexts for DV. Interestingly, the result per-
sists even when we distinguish between transnational and domestic remittances, access to either type of
transfer reducing the acceptance of domestic violence by comparable magnitudes. This suggests the oper-
ation of mechanisms other than the transnational diffusion of gender norms that has typically been
emphasized in the literature.
Ó 2020 Elsevier Ltd. All rights reserved.

1. Introduction household level data from a Multiple Indicators Cluster Survey


(MICS) conducted in the province of Punjab in the Islamic Republic
There has been increasing scholarly interest in the institutional of Pakistan in 2014, we investigate if the access to remittance
impact of remittance inflows,1 with the literature focusing primarily income on the part of a household helps to change beliefs held
on the transformation of formal institutions, such as the quality of by women in the household regarding the acceptability of domes-
governance (Abdih, Chami, Dagher, & Montiel, 2012), the level of tic violence (DV).
corruption (Tyburski, 2014), the composition of public expenditure Our results substantiate a strong beneficial impact of remit-
(Doyle, 2015), the level of electoral competition (Pfutze, 2014), the tances: using a matching estimator to construct counterfactuals
persistence of autocratic regimes (Ahmed, 2012, 2014) and their that allow us to compare women from otherwise similar house-
potential democratization (Escriba-Folch, Meseguer, & Wright, holds who differ in their access to remittance income, we find that
2015). women in households which receive remittances are approxi-
There is, however, a relative dearth of empirical research on the mately 6.9 percentage points less likely to tolerate domestic vio-
potentially transformative role of remittances on informal institu- lence for any reason whatsoever than women in comparable
tions in recipient societies, particularly culturally constructed households which do not receive remittances. The impact persists
expectations about the social role of women. It is to this underex- even when we differentiate between domestic and transnational
plored yet important area that we direct our contribution. Using remittances: women in households receiving domestic remittances
are about 5.5 percentage points less likely to accept domestic vio-
lence relative to women in otherwise comparable households
⇑ Corresponding author. which do not receive domestic remittances, the corresponding fig-
E-mail addresses: amitra@bard.edu (A. Mitra), bangjamest@sau.edu (J.T. Bang), ure for transnational remittances being marginally higher at about
dr.faisal@s3h.nust.edu.pk (F. Abbas).
1 7 percentage points.
See the survey by Baudassé, Bazillier, and Issifou (2018), for example.

https://doi.org/10.1016/j.worlddev.2020.105149
0305-750X/Ó 2020 Elsevier Ltd. All rights reserved.
A. Mitra et al. World Development 138 (2021) 105149

Unsurprisingly, we observe a heterogeneous impact of remit- barriers that had so far prevented women from reporting violence.
tances if we distinguish between different contexts for DV. The Regardless of how one interprets the data, gender relations in Punjab
MICS questionnaire for individual women asks women if they remain among the most patriarchal in South Asia, prevailing gender
believe a husband is justified in beating his wife is five distinct cir- norms being influenced by the interplay of traditional and Islamist
cumstances, namely, if she goes out without informing him, ideologies that emphasize the subjugation of women to men.
neglects the children, argues with him, refuses sex with him, or At the same time, Punjab has historically contributed the great-
burns the food. Once we look at the impact of receiving remit- est share of international migrants from Pakistan. A 2015 report by
tances on the tolerance for DV in each of these contexts individu- the International Labor Organization (ILO) estimates that approxi-
ally, we obtain a statistically significant impact only in the last mately 52% of all migrant workers who registered with the Bureau
two, namely, refusing sex and burning food. While the reduced of Emigration and Overseas Employment (BEOE) in Pakistan over
acceptance of violence from the husband on occasion of being the period 1981–2015 came from the Punjab.7 As such, it is not
refused sex, for example, makes us hopeful that the access to hard to believe a significant part of transnational remittance flows
remittance income may help to increase the degree of autonomy to Pakistan have accrued to households in Punjab.
women have over their own bodies, the results taken together sug- While estimates of internal labor flows and associated remit-
gest that women in Punjab may not regard all forms of DV as tance transfers are relatively scarce in Pakistan8; one study
equally reprehensible. At the very least, they show that even in (Nenova, Niang, & Ahmad, 2009) puts domestic remittances at about
the specific context of DV, gender norms in certain domains are 90% of the value of international remittances. Given the profound
more resistant to change than others. social changes induced by migration and associated remittance flows
The paper is structured as follows: Section 2 briefly documents regardless of origin, it is surprising that there is a distinct lack of
the scale of gender violence experienced by women in Punjab and studies that investigate the impact of remittances on various aspects
provides a context for our study. Section 3 outlines the theoretical of gender relations in Punjab. The present study is an attempt to fill
foundations of our analysis and locates our contribution in the mul- this void.
tiple strands of literature it contributes to. Section 4 introduces the
data and provides a brief description of the econometric methodol- 3. Theoretical foundations and related literature
ogy, while Section 5 describes our results, and evaluates their
robustness. Section 6 concludes the paper by noting the policy impli- Why should one expect the access to remittance income to
cations of our analysis and indicating directions for further research. reduce the occurrence of DV and its acceptability to women? There
is an emerging literature in economics (Beine, Docquier, & Schiff,
2. Gender violence in Punjab 2013; Bertoli & Marchetta, 2015; Tuccio & Wahba, 2018) which
attributes the transformation of local gender norms as communi-
There has been increasing concern within Pakistan and beyond ties experience emigration to ‘social remittances’ or the set of
about the sheer scale of violence faced by women in the country, ideas, beliefs, identities and practices carried home by migrants
recorded incidents being by far the highest in the largest and most from host societies (Levitt, 1998). As explained subsequently in
populous province of Punjab. A 2014 report by the Aurat Founda- the paper, this does not appear to be an important mechanism in
tion, a private non-profit working for women’s rights in Pakistan, our context.
documents 10,070 cases of violence against women (VAW) Alternatively, or perhaps in conjunction, the access to remit-
reported between January and December of 2014 in the country tances is bound to increase the resources of the household in abso-
as a whole, of which 7548 (75%) were from Punjab alone.2 The pro- lute terms. This should reduce both the occurrence of DV and the
vince also contributed thirteen of the fifteen districts in the country acceptability of it to women via a number of mechanisms: If we
which ranked worst in terms of violent crimes against women. accept the idea that men use DV as a means to obtain and preserve
Years of political activism bore fruit when despite the furor power in the household and that violence is one of multiple
unleashed by a coalition of fundamentalist religious and political resources such as education, income, or social status that can be
interests,3 the provincial government of Punjab passed the landmark used for this purpose (Allen & Straus, 1979; Eswaran & Malhotra,
Protection of Women Against Violence Act of 2016,4 which broadened 2011; Goode, 1971; Vyas & Watts, 2009); then greater availability
the definition of violence to criminalize forms of physical, verbal and of economic resources due to the inflow of remittances should
cyber abuse that were formerly outside the scope of legal intervention; reduce the incentive to use violence on the part of men.
took steps to curb the virtually insurmountable bureaucratic hurdles Second, in a patriarchal society like Punjab where a man regards
that women had to negotiate in order to report violence, let alone himself and is regarded by others as the provider, poverty fosters
obtain legal redress for it; and authorized the creation of protection perceptions of inadequacy that may translate to violence. As
centers for women forced out of their homes due to violent abuse. Jewkes (2002) puts it: ‘‘An inability to meet social expectations of
While most accounts indicate that initiatives introduced as part successful manhood can trigger a crisis of male identity. Violence
of the Act have brought needed solace to hundreds of women;5 against women is a means of resolving this crisis because it allows
reported cases of VAW have continued to increase in Punjab over expression of power that is otherwise denied.” It follows, therefore,
the last three years (Punjab Gender Parity Report 2018),6 though it that an improvement in household wealth due to remittance
is not clear if this indicates an upsurge in the actual occurrence of income should correlate with reduced threats to manhood that
VAW in the region or if it is due to lowered legal and bureaucratic lie at the heart of DV.
Third, it has been argued that men who suffer from a low sense
2 of worth due to perceptions of low social status are more likely to
The Aurat Foundation report identifies ten categories of violence against women
(VAW), namely, abductions, murders, rapes, honor killings, sexual assaults, domestic use violence towards their spouses as a means of extracting the ‘re-
violence, acid throwing, burning, suicide, and miscellaneous. The full text is available spect’ that they feel is denied to them outside the household
at:https://www.af.org.pk/PDF/VAW%20Reports%20AND%20PR/VAW%202014.pdf. (Goodman, Serag, Raimer-Goodman, Keiser, & Gitari, 2017). Given
3
See https://www.dawn.com/news/1245591.
4
that the decision to migrate is, to a great extent, influenced by con-
http://punjablaws.gov.pk/laws/2634.html.
5
See https://www.pbs.org/newshour/world/in-pakistan-abused-women-find-
7
comfort-and-justice-at-this-all-female-run-center. See https://www.ilo.org/islamabad/whatwedo/publications/WCMS_514139/
6
The document is available at: https://pcsw.punjab.gov.pk/system/files/PGPR- lang–en/index.htm.
8
2018_0.pdf. See Ishfaq, Ahmad, Hassan, and Javed (2017) and references therein.

2
A. Mitra et al. World Development 138 (2021) 105149

siderations of social status as reflected in the income of the house- the first two mechanisms, the lack of information regarding the
hold relative to a reference group (Stark and Taylor, 1991), it is not identity of the primary beneficiary of remittance income in the
hard to see that the increase in relative household income and household prevents us from exploring the role of the last.
resultant improvement in self-esteem would reduce the incentive Second, our study advances an as yet limited literature (Abiona
to perpetrate DV on the part of men. & Koppensteiner, 2016; Cools & Kotsadam, 2017; Sekhri &
Fourth, and particularly salient for South Asia, Sekhri and Storeygard, 2014) which explores the impact of unanticipated
Storeygard (2014) theorize that men face the choice between labor shocks to household income on the likelihood of DV. To the best
income and appropriative income in the form of resources of our knowledge, the present study is the first to explore the other
extracted from the bride’s family. In this case, violence is a tool side of the question, namely, how will the access to an insurance
for maximizing the latter. Access to remittances should reduce technology in the form of remittances which mitigates the impact
the incentive to appropriate and should, therefore, reduce the inci- of adverse income shocks affect DV or more precisely, the tolerance
dence of DV. for it?
Fifth, and closely related to the previous arguments, unantici- Finally, we contribute to the rich literature that explores various
pated or uninsurable resource shocks experienced by the house- aspects of the remittance experience of Pakistan, ranging from
hold (Abiona & Koppensteiner, 2016; Sekhri & Storeygard, 2014) studies investigating determinants of household access to remit-
increase the occurrence of DV due to the mechanisms discussed tances such as conflict (Ghorpade, 2017) to those documenting
earlier. The access to remittance income should predict a reduction the impact of remittance income on various household outcomes,
in DV since remittances act as a key source of insurance for house- such consumption (Javed, Awan, & Waqas, 2017) and the supply
holds in developing countries that lack both social safety nets and of labor (Mughal & Makhlouf, 2013).
private insurance markets (Docquier and Rapoport, 2006).
Sixth, Buller et al. (2018) observe that DV often occurs in the
4. Data and methods
context of conflict over the disposal of limited household funds,
the operational factors being as much a lack of income as a lack
4.1. MICS 2014
of liquidity. Again, the access to remittance income should reduce
the likelihood of such conflicts and hence, the incidence of DV.
Pakistan is an Islamic Republic consisting of four provinces,
Finally, an influential study by Dutton (2006) notes that the per-
namely Punjab, Sindh, Baluchistan, and Khyber Pakhtunkhwa, of
petration of DV by a man requires an entire ecology consisting of a
which the first is by far the largest and most populous. The 18th
macrosystem in the form of norms and belief systems shared by a
constitutional amendment in 2010 significantly curtailed the
community; an exosystem composed of formal and informal groups
power of the federal government and gave the provinces autonomy
that connect the household to the broader environment; a mi-
over forty distinct domains, including labor, education, environ-
crosystem consisting of the household itself as the immediate con-
mental policy, and contract law, to name but a few. The devolution
text embedding the individual; and ontogenetic factors specific to
of power required the collection of statistically sound sub-national
the perpetrator.9
data that would inform policy design at the provincial level and the
While the mechanisms described so far in this section focus on
MICS is an initiative launched in collaboration with the United
the microsystem of DV, note that the access to remittance income
Nations Children’s Fund (UNICEF) to address this need.
will reduce the dependence of a family on what limited social
The universe for MICS Punjab 2014 consists of all households in
insurance is provided by the local government, the community,
the nine administrative divisions and 36 districts of the province. A
or both. Given that the forces of patriarchy typically operate by
two-stage, stratified cluster sampling approach was used for the
controlling who receives much needed public goods and other ben-
selection of the sample. The primary sampling unit (PSU) was the
efits from the community or the state, a reduced reliance on such
Enumeration Block (EB) in urban areas and the village in rural
welfare should serve to disassociate a household from the exosys-
areas. The PSUs were stratified on the basis of living standards of
tem that helps to propagate DV. Note that given a sufficient mea-
the households in the block and their quality of housing. In order
sure of remittance-receiving households that can afford to spurn
to select the households, the survey first selected PSUs with prob-
the economic aid of the embedding networks, one may expect to
ability proportional to size. Following this, a sample of twenty
see a transformation of the macrosystem itself.
households or secondary selection units (SSUs) was selected with
We end this section by noting that our paper contributes to
equal probability from each first stage PSU
three distinct areas of research. First, it complements an expanding
The entire sample of SSUs was drawn using a systematic sam-
interdisciplinary literature evaluating the efficacy of various condi-
pling technique with random start from 2050 PSUs, of which 774
tional and unconditional transfer programs in reducing the inci-
were urban and 1276 were rural. The final allocation was 2050
dence of DV. As observed in the survey by Buller et al. (2018),
clusters with 20 households in each, giving a total sample of
the literature has typically emphasized three causal mechanisms
41,000 households. The response rate was almost 98 percent
whereby cash transfer programs reduce DV: (a) the pure income
across the province, making the sample representative of the pro-
effect induced by increased household resources reduces the stress
vince as a whole. We refer the reader to the Punjab Bureau of
associated with poverty and eliminates perceptions of inadequacy
Statistics Planning & Development Department for further infor-
on the part of men that often lie at the root of DV; (b) the increased
mation on the methodology.
liquidity experienced by the household reduces the likelihood of
conflict regarding the utilization of the limited family budget;
and (c) transfers targeted at women improve the bargaining posi- 4.2. Outcome variables
tion of women within the household or increase their perceived
worth as a source of income.10 While our results are consistent with We should clarify at the very onset that we focus on the toler-
ance for DV on the part of women rather than the actual occur-
rence of such violence. Note that this helps to avoid the problem
9
See Lawson (2012) for a review of this and other theories of DV.
10
of eliciting accurate information on the intimate partner abuse
Alternatively, if one conceives of DV as a means to preserve male dominance of
the household, such directed transfers may increase the incidence of violence simply
from Pakistani women who are typically reluctant to divulge this
because they threaten male hegemony which is aided by economic dependence of the information due to a combination of shame; helplessness; concern
woman on her partner (Eswaran & Malhotra, 2011; Cools & Kotsadam, 2017). for adverse reputational effects on her parents’ family; and fear of
3
A. Mitra et al. World Development 138 (2021) 105149

retribution, including escalated violence, divorce, and losing access holds with access to remittance income to those held by women
to her children (Andersson et al., 2009). Compounding the problem in households similar to the former in every respect except the
is the reluctance to turn to formal institutions for help, which are access to remittances. The key methodological challenge in this
justifiably perceived as unsympathetic, uncaring, overtly bureau- context is that the access to migration and hence, remittance
cratic, and more likely to exacerbate the problem (Zakar, Zakar, & income, is not randomly distributed over the population. It is
Krämer, 2012). almost universally accepted that migration is essentially a house-
As importantly, how women perceive DV is a critically impor- hold decision undertaken to overcome constraints on income gen-
tant subject of study in its own right: there is a wealth of evidence eration imposed by a lack of liquidity and the imperfection of
that women who consider such violence acceptable are more likely credit and insurance markets, which also exacerbate the impact
to blame themselves for it, less likely to seek redress, less likely to of unanticipated shocks to household income (Taylor & Wyatt,
provide solace to other victims of DV, and more likely to suffer the 1996). As such, any attempt to identify a causal impact of remit-
debilitating physical and mental health conditions that result from tances requires one to address the selection bias inherent in
long term exposure to such trauma.11 migration.
Unsurprisingly, the perpetrators of DV often do not see it as vio- Our response to the problem of selection employs a matching
lent abuse but legitimize it as ‘a disciplinary action when a woman estimator, which yields causal estimates under the Conditional
does not comply with certain expected gender roles’ (MICS 2014 Final Independence Assumption (CIA) or the assumption that all vari-
Report, pp. 176).12 As such, MICS 2014 asked all women aged ables which influence the decision to migrate and hence, selection
between 15 and 49 years if they believe a husband to be justified into the ‘treated group’ of households receiving remittances, are
in hitting or beating his wife in each the following situations: (i) if perfectly observable. 13As a first step to the empirical analysis,
she goes out without telling him, (ii) if she neglects the children, therefore, we match each of the households in our sample which
(iii) if she argues with him, (iv) if she refuses to have sex with received remittances (2114 households with 8370 women) one-to-
him, and (v) if she burns the food. one with demographically similar households which did not receive
Our primary outcome variable is coded as 1 if a woman remittances (23,003 households with 55,581 women). Recall that at
responds yes to at least one of the five hypothetical scenarios given a minimum, matching seeks to ‘balance’ or equate the characteristics
above. It is coded as 0 if she responds no or don’t know to all of the of the treated and control groups on the average, as would be the
five hypotheticals. Subsequently, for the set of exercises reported case if households were randomly assigned access to remittance
in Table 3, we construct a set of five categoric variables corre- income. But it also allows us to create matched pairs with two units
sponding to the five different scenarios, each of which is coded of otherwise similar characteristics differing solely in terms of treat-
as 1 if a woman thinks DV to be justified in that scenario and 0 ment status. In other words, matching allows a pairwise comparison
otherwise. of closely comparable households differing only with respect to the
receipt of remittance income.
Two approaches to balancing or creating comparable treatment
4.3. Variable of interest
and control groups dominate the literature, namely, Propensity
Score Matching (PSM) and Mahalanobis Distance Matching
The primary variable of interest is an indicator variable equal to
(MDM), with a third alternative in the form of Coarsened Exact
one if the household under observation received any remittances
Matching (CEM) having emerged of late.14 While PSM is perhaps
and equal to zero if it did not (or if the respondent did not know).
the most widely used method – not the least because of its simplicity
Hence, what we are essentially modeling is the marginal impact of
– it suffers from several deficiencies. In particular, reducing the
receiving any remittance income whatsoever rather than the effect
covariates for the treatment variable to a unidimensional score
of receiving an additional rupee of remittances.
ignores a lot of potential variation in the characteristics we would
Note that this is consistent with current practice (Adams and
like to match on. Doing so can therefore worsen the balance, effi-
Cuecuecha, 2010, 2013; Bang, Mitra, & Wunnava, 2016). The mea-
ciency, bias, and model dependence of the match (King & Nielsen,
surement of household income in a developing country like Pak-
2019). MDM and CEM remedy this problem by matching directly
istan is subject to considerable error, not the least because of the
on the characteristics believed to be correlated with treatment.
fundamental distrust of the state prevailing among low income
Iacus, King, and Porro (2012) show that both procedures usually
households. Compounding the problem, some households, espe-
result in better matches than PSM, though the relative desirability
cially the very poor, receive much of their ‘remittance income’ in
of MDM and CEM depends on the context. In practice, research often
the form of durable goods brought back to the family by return
uses a combination of MDM and CEM and we follow this approach as
migrants, which would not show up were we to consider the mon-
well.
etary value of remittance income.
We match remittance receiving households both on household
It should also be mentioned that we have explored whether the
characteristics such as size, region, and mobile telephone owner-
perception of DV by women responds differently to domestic and
ship and on characteristics of the household head such as gender,
international remittances. Since the two types of remittances are
age, education, and source of income. Of these variables, we match
neither mutually exclusive nor nested, we could not account for
exactly on region, mobile phone ownership (Yes = 1), and gender
them in a single model and had to estimate two separate models,
(Female = 1); we coarsely exact match on head’s age (in 10-year
respectively comparing households receiving domestic remit-
increments), education (preschool, primary, middle, matriculation,
tances to those that do not and households receiving international
remittances to those without such access.
13
This is not a tenuous assumption given the richness of the data.
14
4.4. Matching method and variables PSM calculates the estimated probability of receiving the treatment and matches
each treated observation to the untreated observation with the nearest propensity. By
contrast, MDM and CEM match observations directly on the observed determinants of
In order to obtain a causal impact of remittances, we would like the treatment variable: MDM does so based on the Mahalanobis distance measure,
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
to compare the attitudes towards DV held by women in house-
DM ðx1 ; x2 Þ ¼ ðx1  x2 Þ0 S1 ðx1  x2 Þ (i.e. the Euclidean distance weighted by the
covariance of the variables), whereas coarsened exact matching does so by cutting
11
See Tran, Nguyen, and Fisher (2016) and references therein. each of the variables into ranges and then finding matches that fall into the same
12
The MICS 2014 Final Report is available at: http://bos.gop.pk/finalreport. ranges on all of the variables.

4
A. Mitra et al. World Development 138 (2021) 105149

Table 1
Descriptive Statistics.

(1) (2) (3)


VARIABLES Full Sample Matched Women in Remittance Households Matched Women in Non-Remittance Households
Dependent Variables: Acceptance of Violence
Any of the Surveyed Reasons 0.409 0.367 0.508
(0.492) (0.482) (0.5)
Going Out without Permission 0.273 0.242 0.344
(0.446) (0.428) (0.475)
Neglecting the Children 0.276 0.237 0.32
(0.447) (0.425) (0.467)
Argues with her Husband 0.284 0.245 0.321
(0.451) (0.43) (0.467)
Refuses Sex with her Husband 0.211 0.171 0.296
(0.408) (0.377) (0.456)
Burns the Food 0.152 0.111 0.189
(0.359) (0.314) (0.391)
Remittances Variables
Received Any Remittances 0.129 – –
(0.335)
Received Domestic Remittances 0.0397 0.308 –
(0.195) (0.462)
Received Overseas Remittances 0.0915 0.711 –
(0.288) (0.453)
Total Amount of Remittances 27,717 219,182 –
(140,144) (337,636)
Amount of Domestic Remittances 3773 29,915 –
(29,225) (77,597)
Amount of Overseas Remittances 23,944 189,268 –
(136,713) (342,151)
Controls
Age of Woman 28.76 28.4 33.73
(9.681) (9.539) (9.111)
Children Ever Born 2.247 1.88 3.451
(2.6) (2.389) (2.874)
Woman’s Combined Wealth Score 0.0815 0.494 0.0639
(0.957) (0.888) (0.918)
Woman’s Education Level = Preprimary 0.388 0.258 0.494
(0.487) (0.437) (0.5)
Woman’s Education Level = Primary 0.176 0.174 0.169
(0.381) (0.379) (0.375)
Woman’s Education Level = Middle 0.105 0.122 0.0968
(0.306) (0.327) (0.296)
Woman’s Education Level = Secondary 0.162 0.213 0.135
(0.368) (0.41) (0.342)
Woman’s Education Level = Higher 0.17 0.233 0.104
(0.375) (0.423) (0.306)
Woman’s Education Level = Missing 0.0003 0.000297 0.000744
(0.0173) (0.0172) (0.0273)
TV Exposure: Almost Every Day 0.529 0.593 0.421
(0.499) (0.491) (0.494)
TV Exposure: At Least Once/Week 0.108 0.126 0.134
(0.311) (0.332) (0.341)
TV Exposure: Less than Once/Week 0.0506 0.0515 0.0449
(0.219) (0.221) (0.207)
TV Exposure: Not at All 0.312 0.23 0.4
(0.463) (0.421) (0.49)
Husband’s Age, 15–24 0.038 0.0296 0.0268
(0.191) (0.169) (0.161)
Husband’s Age, 25–34 0.199 0.218 0.236
(0.399) (0.413) (0.425)
Husband’s Age, 35–44 0.201 0.175 0.178
(0.401) (0.38) (0.382)
Husband’s Age, 45–54 0.142 0.124 0.167
(0.349) (0.329) (0.373)
Husband’s Age, 55–64 0.0259 0.0268 0.0559
(0.159) (0.161) (0.23)
Husband’s Age, 65+ 0.00407 0.00387 0.00297
(0.0637) (0.0621) (0.0545)
Husband’s Age: Not Currently Married 0.391 0.422 0.334
(0.488) (0.494) (0.472)
Age at First Marriage: Child (Under 18) 0.112 0.0947 0.188
(0.315) (0.293) (0.391)
Age at First Marriage: Adult (18–29) 0.518 0.507 0.638
(0.5) (0.5) (0.481)
Age at First Marriage: Late (30 + ) 0.0129 0.0144 0.0123

(continued on next page)

5
A. Mitra et al. World Development 138 (2021) 105149

Table 1 (continued)

(1) (2) (3)


VARIABLES Full Sample Matched Women in Remittance Households Matched Women in Non-Remittance Households
(0.113) (0.119) (0.11)
Age at First Marriage: Never Married 0.357 0.384 0.162
(0.479) (0.486) (0.368)
Ever Used Contraception Yes 0.289 0.235 0.283
(0.454) (0.424) (0.45)
Ever Used Contraception No 0.32 0.342 0.384
(0.466) (0.474) (0.486)
Relationship to Head: Head 0.0278 0.123 0.198
(0.164) (0.329) (0.398)
Relationship to Head: Wife 0.397 0.145 0.306
(0.489) (0.352) (0.461)
Relationship to Head: Daughter 0.324 0.349 0.175
(0.468) (0.477) (0.38)
Relationship to Head: Daughter-in-Law 0.161 0.263 0.29
(0.367) (0.44) (0.454)
Relationship to Head: Other 0.0905 0.12 0.0318
(0.287) (0.325) (0.176)
Woman Employment: Employed 0.108 0.0909 0.11
(0.311) (0.287) (0.314)
Woman Employment: No/Housewife 0.778 0.771 0.867
(0.416) (0.42) (0.34)
Woman Employment: Student 0.114 0.138 0.0228
(0.318) (0.345) (0.149)
Government Benefits: Received 0.0806 0.047 0.1
(0.272) (0.212) (0.301)
Zakat Payments: Received 0.0104 0.0106 0.0349
(0.101) (0.102) (0.184)
Zakat Payments: Don’t Know 0.08 0.0467 0.0986
(0.271) (0.211) (0.298)
Matching Variables
Household Size 7.637 8.129 7.897
(3.647) (4.456) (3.839)
Head of Household Age 48.55 52.47 52.34
(12.83) (13.79) (13.42)
Head’s Age = 15–24 0.0176 0.0153 0.0153
(0.132) (0.123) (0.123)
Head’s Age = 25–34 0.127 0.0947 0.0947
(0.333) (0.293) (0.293)
Head’s Age = 35–44 0.227 0.164 0.164
(0.419) (0.37) (0.37)
Head’s Age = 45–54 0.32 0.266 0.266
(0.467) (0.442) (0.442)
Head’s Age = 55–64 0.187 0.258 0.258
(0.39) (0.438) (0.438)
Head’s Age = 65+ 0.121 0.202 0.202
(0.326) (0.401) (0.401)
Head Employment: Employed 0.409 0.204 0.204
(0.492) (0.403) (0.403)
Head Employment: Unemployed 0.395 0.329 0.329
(0.489) (0.47) (0.47)
Head Employment: Student 0.197 0.467 0.467
(0.397) (0.499) (0.499)
Head’s Education Level = Preprimary 0.385 0.406 0.406
(0.487) (0.491) (0.491)
Head’s Education Level = Primary 0.178 0.182 0.182
(0.382) (0.386) (0.386)
Head’s Education Level = Middle 0.135 0.137 0.137
(0.342) (0.344) (0.344)
Head’s Education Level = Secondary 0.189 0.193 0.193
(0.392) (0.395) (0.395)
Head’s Education Level = Higher 0.112 0.0812 0.0812
(0.316) (0.273) (0.273)
Head’s Education Level = Missing 0.000375 0.000446 0.000446
(0.0194) (0.0211) (0.0211)
Female Head of Household 0.0785 0.291 0.291
(0.269) (0.454) (0.454)
Region = Urban 0.108 0.0778 0.0778
(0.31) (0.268) (0.268)
Region = Other 0.267 0.251 0.251
(0.442) (0.434) (0.434)
Region = Rural 0.626 0.671 0.671
(0.484) (0.47) (0.47)
Mobile Telephone = Yes 0.953 0.979 0.979
(0.212) (0.142) (0.142)
Observations 53,295 6724 6724

6
A. Mitra et al. World Development 138 (2021) 105149

Table 2
Average Marginal Effects (AME) on the Acceptance of DV by Women.

(1) (2) (3) (4)


VARIABLES Unmatched Matched Matched Domestic Matched International
Unadjusted Differences
Remittances (Difference) 0.0497*** 0.136*** 0.110*** 0.153***
(0.00633) (0.00842) (0.0146) (0.00983)
Risk-Adjusted Logit Model
Remittances 0.013 0.0693*** 0.0548** 0.0707***
(0.00889) (0.0202) (0.0272) (0.0236)
Woman’s Age 5.51E-05 0.000697 0.000235 0.00132
(0.000489) (0.00229) (0.00302) (0.00242)
Children Ever Born 0.00557*** 0.00635 0.00626 0.00663
(0.00156) (0.00722) (0.00867) (0.00767)
Education = Primary 0.0314*** 0.00172 0.0387 0.0197
(0.0073) (0.0297) (0.0355) (0.0344)
Education = Middle 0.0646*** 0.0168 0.0208 0.0214
(0.00911) (0.0321) (0.0412) (0.0361)
Education = Secondary 0.109*** 0.0777** 0.101** 0.0741**
(0.00877) (0.031) (0.043) (0.0347)
Education = Higher 0.186*** 0.149*** 0.128*** 0.148***
(0.0096) (0.0316) (0.0478) (0.0359)
Education = Missing 0.0409 0.341** 0.333**
(0.106) (0.139) (0.137)
TV Exposure = At least once/week 0.00264 0.0215 0.0208 0.026
(0.00684) (0.0257) (0.0336) (0.0284)
TV Exposure = Less than once/week 0.0626*** 0.126*** 0.104*** 0.130***
(0.00888) (0.029) (0.0387) (0.0324)
TV Exposure = None 0.0290*** 0.0374 0.0261 0.0319
(0.0109) (0.0422) (0.0529) (0.0501)
Child Marriage 0.0113 0.0444 0.049 0.0399
(0.00746) (0.0312) (0.0369) (0.0344)
Late Marriage 0.0191 0.126* 0.0586 0.135*
(0.0184) (0.0649) (0.0834) (0.0731)
Not Married 0.0311** 0.0189 0.0624 0.017
(0.0154) (0.0585) (0.0761) (0.0654)
Relation to head = wife 0.0122 0.00778 0.00352 0.00915
(0.0155) (0.0378) (0.0496) (0.041)
Relation to head = daughter 0.00234 0.0029 0.0558 0.0101
(0.0176) (0.054) (0.0683) (0.058)
Relation to head = daughter-in-law 0.0034 0.0346 0.0234 0.0442
(0.0166) (0.0415) (0.0575) (0.0431)
Relation to head = other 0.000825 0.0254 0.0398 0.0234
(0.0169) (0.0472) (0.0666) (0.0494)
Employment = housewife/none 0.0228*** 0.0301 0.0243 0.0345
(0.00784) (0.0326) (0.039) (0.037)
Employment = student 0.0184* 0.0629 0.0348 0.0905**
(0.0111) (0.0413) (0.0601) (0.0451)
Wealth Score 0.0819*** 0.0621*** 0.0547*** 0.0632***
(0.00449) (0.0144) (0.0197) (0.0166)
Household Size 0.00153 0.00496** 0.0125*** 0.0026
(0.000938) (0.00235) (0.00409) (0.00247)
Government Benefits 0.0171 0.0805 0.217 0.0525
(0.055) (0.14) (0.166) (0.167)
Zakat Benefits: Received 0.0405* 0.119 0.150* 0.0988
(0.0234) (0.075) (0.0822) (0.0829)
Zakat Benefits: No answer 0.0298 0.0774 0.179 0.0398
(0.0552) (0.134) (0.16) (0.18)
Observations 53,295 13,448 4142 9564
Log Pseudo-Likelihood 33,751.48 8611.55 2695.37 6063.31
Chi-Square 1942.53 289.71 134.76 244.71
P-Value 0.000 0.000 0.000 0.000
Pseudo-R2 0.0639 0.0655 0.0585 0.0697

Standard errors in parentheses


*** p < 0.01, ** p < 0.05, * p < 0.1.

or higher), and source of income (employed, self-employed, or not 4.5. Main specification and variables
employed/student); and we distance match on household size
(number of persons residing in the household who eat together). Once households receiving remittances have been matched
We further restrict our matching to only consider matches that with demographically-similar households without access to such
contain full observations of the variables in our subsequent regres- transfers, we are in a position to estimate the treatment effect of
sion model. remittances on the acceptance of DV by women in the household.

7
A. Mitra et al. World Development 138 (2021) 105149

Table 3
Average Marginal Effects (AME) on the Acceptance of DV by Reason.

(1) (2) (3) (4) (5)


VARIABLES Goes Out Child Neglect Argues Refuses Sex Burns Food
Unadjusted Differences
Remittances (Difference) 0.102*** 0.0832*** 0.0760*** 0.125*** 0.0777***
(0.00771) (0.00769) (0.0077) (0.00711) (0.00611)
Risk-Adjusted Logit Model
Remittances 0.0402* 0.0265 0.0244 0.0402** 0.0321**
(0.0207) (0.0191) (0.0186) (0.018) (0.0148)
Woman’s Age 0.000437 0.000314 0.000833 0.00221 0.000134
(0.00225) (0.0022) (0.00208) (0.00181) (0.00172)
Children Ever Born 0.0062 0.0113* 0.000627 0.00383 0.00287
(0.00627) (0.00577) (0.00607) (0.00542) (0.00443)
Education = Primary 0.0202 0.0404 0.0282 0.0174 0.0338
(0.0303) (0.0295) (0.0279) (0.0273) (0.0256)
Education = Middle 0.0131 0.0235 0.00317 0.0358 0.024
(0.036) (0.0342) (0.0355) (0.033) (0.0307)
Education = Secondary 0.0535 0.0361 0.0725** 0.0407 0.102***
(0.0339) (0.0327) (0.0312) (0.0324) (0.023)
Education = Higher 0.145*** 0.0793** 0.147*** 0.0676** 0.119***
(0.0314) (0.0319) (0.0291) (0.0289) (0.0237)
Education = Missing 0.169 0.125 0.182 0.0658 0.0499
(0.159) (0.16) (0.147) (0.146) (0.15)
TV Exposure = At least once/week 0.0333 0.0133 0.00063 0.00529 0.0117
(0.0245) (0.0247) (0.0231) (0.0234) (0.0196)
TV Exposure = Less than once/week 0.0755** 0.0650* 0.0895*** 0.0986*** 0.0283
(0.0333) (0.0333) (0.0321) (0.031) (0.03)
TV Exposure = None 0.0502 0.0415 0.0396 0.0118 0.0275
(0.0429) (0.0381) (0.0368) (0.0333) (0.0301)
Child Marriage 0.0363 0.0141 0.000277 0.0469* 0.0132
(0.0273) (0.0262) (0.0282) (0.0245) (0.0212)
Late Marriage 0.0563 0.0875* 0.131*** 0.0853* 0.0168
(0.0626) (0.0485) (0.0485) (0.0462) (0.0474)
Not Married 0.0178 0.0196 0.0301 0.0898** 0.0211
(0.0589) (0.0611) (0.0551) (0.044) (0.0432)
Relation to head = wife 0.0236 0.000102 0.0304 0.0552 0.0289
(0.0397) (0.0355) (0.0351) (0.0354) (0.0251)
Relation to head = daughter 0.00921 0.00523 0.0108 0.00453 0.0557
(0.0566) (0.0512) (0.0501) (0.0412) (0.0458)
Relation to head = daughter-in-law 0.0448 0.0193 0.0301 0.015 0.0042
(0.0444) (0.0401) (0.0391) (0.0375) (0.026)
Relation to head = other 0.0077 0.0326 0.00851 0.0454 0.0403
(0.0469) (0.0445) (0.041)7 (0.0374) (0.0309)
Employment = housewife/none 0.0206 0.012 0.0132 0.0351 0.0145
(0.0294) (0.0289) (0.0289) (0.0255) (0.0227)
Employment = student 0.0138 0.0308 0.0306 0.0437 0.0308
(0.0374) (0.04) (0.0393) (0.0412) (0.0355)
Wealth Score 0.0439*** 0.0572*** 0.0523*** 0.0735*** 0.0331**
(0.0149) (0.014) (0.0135) (0.0144) (0.0131)
Household Size 0.00101 0.00236 0.00579** 0.00258 0.000657
(0.00233) (0.0022) (0.00238) (0.00224) (0.0018)
Government Benefits 0.202 0.17 0.192 0.143 0.312*
(0.142) (0.155) (0.148) (0.136) (0.16)
Zakat Benefits: Received 0.0653 0.0392 0.0278 0.00649 0.0242
(0.0865) (0.086) (0.0816) (0.0719) (0.0662)
Zakat Benefits: No answer 0.126 0.0999 0.122 0.11 0.120***
(0.0899) (0.102) (0.0915) (0.078) (0.0398)
Observations 13,448 13,448 13,448 13,448 13,448
Log Pseudo-Likelihood 5326.51 5270.28 5300.30 4593.78 3584.83
Chi-Square 172.85 156.57 155.94 223.16 158.6
P-Value 0.000 0.000 0.000 0.000 0.000
Pseudo-R2 0.0571 0.0482 0.048 0.0964 0.074

Standard errors in parentheses


*** p < 0.01, ** p < 0.05, * p < 0.1.

Clearly, the simplest option is to conduct a difference in propor- rows of Tables 2 and 3 as a first pass at obtaining the causal impact
tions test of women accepting DV in the matched treatment and of remittances.
control samples. Note, however, that while the estimates so To address the concern of omitted variable bias, we estimate a
obtained are free of selection bias based on which households logit model on the balanced sample obtained in the first step con-
received remittances, they may still be subject to omitted variable trolling for a number of characteristics of the women which may
bias since we have not controlled for the personal characteristics of influence their perception of DV and a few household characteris-
women that influence their attitudes towards DV. Nevertheless, we tics. Specifically, we estimate the following equation for the accep-
report the results of the difference-in-proportion tests in the first tance of domestic violence by woman i living in household j:

8
A. Mitra et al. World Development 138 (2021) 105149

!
PðJustifiedij ¼ TrueÞ remittances at all. Column 4 presents the analogous comparison
ln ¼ cRemittancesij þ b0 þ b1 Ageij for international remittances.26
PðJustifiedij ¼ FalseÞ
The first row of each column reports the results of simple differ-
þ b2 Childrenij þ b3 Wealthij þ b4 HHSizej ence in proportions tests for the relevant samples, without control-
X
þ d Educationijh ling for the characteristics of women which influence the
h2H h
X perception of DV. Subsequent rows follow current praxis in report-
þ m2M m
d TVExposureijm ing the causal impact of the explanatory variables in the form of
X average marginal effects (AME).27 Note that regardless of the sam-
þ r2R
dr Relationijr
X ple, these tests yield impacts that are almost twice as large as what
þ p2P
dp MarriageAgeijp we get when we undertake a full regression controlling for the rele-
X vant characteristics of women. This provides an idea of how omitted
þ w2W
dw WorkStatusijw variables distort our estimates.
þ dg Gov ernmentBenefitsj Note also that when we consider the raw association of remit-
tances with the tolerance for DV on the basis of the unmatched
þ dz ZakatBenefitsj þ eij :
model in column 1, it appears that remittances actually increase
acceptance of DV by a small and statistically insignificant margin.
The controls include demographic characteristics of the woman
However, once we balance the sample by matching the women
(age,15 coarsened level of education,16 relation to household head,17
in households with access to remittance income to women in sta-
and coarsened age at first marriage18); economic characteristics (the
tistically similar households without such access and thereby
woman’s relative wealth score,19 and work status20); the number of
purge our estimates of selection bias, we see that remittances re-
children born21 and whether the woman has ever used contracep-
duce the acceptance of DV by almost 7 percentage points. As can
tion22 as measures of the woman’s maternal role in the household;
be seen from columns 3 and 4 respectively, the result is robust
and frequency of exposure to TV to capture the woman’s access to
both to matching women in households that received domestic
media.23 We also include whether a household received any social
remittances who are about 5.5 percentage points less likely to
safety nets benefits or not to control for the effects of government
accept DV and to matching women in households with access to
transfers,24 and Zakat benefits25 influenced by religious doctrines.
international remittances who are about 7 percentage points less
Below we will describe the construction of variables used in our
likely to tolerate DV.
analyses; and household size to control for variations in household
For the most part, the control variables exhibit impacts consis-
size that persist even after the matching.
tent with theory. In particular, we see that women with greater
Dropping observations due to missing data in the women’s
access to wealth (Bhattacharyya, Bedi, & Chhachhi, 2011), greater
characteristics leaves us with a full sample of 53,296 women across
education in general, and more exposure to television media tend
22,541 households. Of these, 6864 lived in one of 1889 households
to excuse domestic violence less. Interestingly, we do not find any
that received remittances and about 40% of these women accepted
impact of a household receiving Zakat benefits (transfers overseen
domestic violence in at least one of the five contexts considered by
by Bait-ul-Mal and potentially tied to stricter adherence to patriar-
the survey. We were able to match 6724 (about 98%) of women in
chal socio-religious norms) on attitudes towards DV.28
remittance receiving households. Descriptive and demographic
As mentioned previously in Section 4.2, the survey asks women
statistics for our sample are reported in Table 1.
to distinguish between five different contexts for DV, namely, (i) if
she goes out without informing the husband, (ii) if she neglects the
5. Results children, (iii) if she argues with the husband, (iv) if she refuses to
have sex with him, and (v) if she burns the food. It is reasonable
The four columns of Table 2 report results obtained on the basis to ask if the impact of remittances on the perception of DV depends
of four different samples. For comparison with subsequent results, on the context in which violence occurs.29 To address this, we repli-
the results presented in column 1 are based on the full unmatched cate our basic specification from column 2 of Table 2, replacing the
sample of all observations, without taking into account the selec-
tion bias inherent in receiving remittances. Column 2 presents 26
In matching domestic and international remittance-receiving households sepa-
the analogues computed on the basis of the matched sample to rately we have been careful to only match controls that received no remittances at all.
correct for selection bias. Columns 3 and 4 investigate if the causal That way none of the ‘controls’ for the domestic remittance-receiving households
impact of remittances on the acceptance of DV differs according to could have received international remittances and vice-versa.
27
the source of these transfers. Thus, column 3 compares attitudes Following Bartus (2005), Consider a regression modelEðyÞ ¼ FðbxÞ, where bx
denotes the linear combination of explanatory variables and F () is the cumulative
towards DV of women who live in households that received remit- distribution function mapping the values of bx to the [0, 1] interval. Then the AME of
P  
tances from within Pakistan to matched women in households that the categorical explanatory variable xi is defined as AMEi ¼ 1n nk¼1 fF bxk jxki ¼ 1 
 
either received international remittances or did not receive any F bxk jxki ¼ 0 g,where xki is the kth observation on xi . We should also note that AMEs
are, in general, preferred to marginal effects at the mean (MEM) on the following
grounds (Williams, 2012): First, an AME is calculated using all observed values of the
15
Numerical between 15 and 49 years. variables rather than just their means. As such, we are not discarding data. Second, it
16
Pre-primary (baseline), primary (1–5 years), middle (6–8 years), secondary (9–10 allows us to compare two hypothetical populations which differ solely in terms of the
years), tertiary (11–18 years), or missing. variable in question, all other characteristics held constant. As such, it is analogous to
17
Head (baseline), wife, daughter, daughter-in-law, or other. a matching model, which is the theoretically ideal situation for isolating the causal
18
Early adult marriage (18–29, baseline), child marriage, late marriage, or not impact of a variable. For readers interested in comparing the results from these two
married. approaches to calculating margins, we have included the MEM margins in the
19
Computed using Principal Component Analysis (PCA) as a composite indicator of supplement.
wealth possessed by each individual. 28
As a robustness check, we have also estimated our models using instrumental
20
Combined into three categories: employed (baseline), housewife, or student. variables, specifically with a probit model. These results demonstrate a similar
21
Numerical between 0 and 18 children. direction of bias, including the change in sign of the impact of remittances from
22
Yes, no, not currently married, or missing. positive (more acceptance) to negative (less acceptance). However, as one might
23
Almost every day (baseline), at least once per week, less than once per week, or expect the IV estimation involved a loss in efficiency, as demonstrated by the
not at all. decrease in the value of the likelihood function and pseudo-R2. We have included the
24
Yes or no. results of this robustness check in the supplemental appendix.
25 29
Yes, no, or missing We are grateful to a referee for suggesting this exercise.

9
A. Mitra et al. World Development 138 (2021) 105149

previous outcome variable with a set of five categoric variables, each Acknowledgements
indicating agreement with the justifiability of DV in one of the five
contexts. It should also be clarified that we have not disaggregated The authors thank two anonymous referees, Phanindra Wun-
the remittance variable by source in this exercise. nava, and participants of the Sustainability and Development Con-
The results, reported in columns 1 – 5 in Table 3, reveal a ference held at the University of Michigan – Ann Arbor in
heterogeneous impact: while all coefficients have the theoretically November, 2018 for valuable feedback on the paper. All remaining
expected negative sign, the access to remittances reduces the tol- errors are our own. We received no funding for the project. To the
erance for DV by a statistically significant amount when DV occurs best of our knowledge, none of us have any conflict of interests in
in retribution for a woman going out of the house without inform- writing this manuscript. Dr. Faisal Abbas was an Experienced
ing the husband, when she refuses to have sex, and when she burns Research Fellow at the time of writing this paper and wishes to
the food. By contrast, receiving remittances does not have a statis- thank Professor Stephan Klasen and Center for Poverty, Equity
tically significant impact on the acceptance of DV when a woman and Growth in Developing and Transition Countries, George August
neglects her children and argues with the husband. University Goettingen, Germany for facilitating his research stay.
Note that this adds a note of caution to our general optimism
regarding the gender impact of remittances. One of the most insid-
ious effects of patriarchy is to condition woman into believing that Appendix A. Supplementary data
DV is the right of a husband exercised in order to discipline his
spouse for transgressions in various domains (Yount, Halim, Supplementary data to this article can be found online at
Schuler, & Head, 2013). The results presented in Table 3 are consis- https://doi.org/10.1016/j.worlddev.2020.105149.
tent with the idea that social conditioning about the justifiability of
DV may be ‘stickier’ in certain domains than others.
References
6. Concluding observations Abdih, Y., Chami, R., Dagher, J., & Montiel, P. (2012). Remittances and institutions:
Are remittances a curse?. World Development, 40, 657–666.
We find that household access to remittance income reduces Abiona, O., & Koppensteiner, M. F. (2016). The impact of household shocks on
domestic violence: Evidence from Tanzania. Working Paper No. 16/14. UK:
the tolerance for domestic violence on the part of women in the University of Leicester.
household, regardless of whether the remittances come from Adams, R. H., Jr., & Cuecuecha, A. (2010). Remittances, household expenditure and
within the country or outside. While transnational remittances investment in Guatemala. World Development, 38(11), 1626–1641.
Adams, R. H., Jr., & Cuecuecha, A. (2013). The impact of remittances on investment
have a marginally greater impact on the perceived acceptability and poverty in Ghana. World Development, 50, 24–40.
of DV than domestic remittances, the difference is not as much Ahmed, F. Z. (2012). The perils of unearned foreign income: Aid, remittances, and
as would support the hypothesis of remittances being accompa- government. American Political Science Review, 106, 146–165.
Ahmed, F. Z. (2014). The perils of unearned foreign income: Aid, Remittances, and
nied by the transfer of more equitable gender norms from abroad,
government survival—CORRIGENDUM. American Political Science Review, 108,
even if we ignore the fact that much of Pakistani migration has tra- 478.
ditionally been to Saudi Arabia and other Gulf states which have at Allen, C., & Straus, M. A. (1979). Resources, power, and husband-wife violence. In M.
least as patriarchal gender norms as the Indian subcontinent.30 A. Straus & G. Hotaling (Eds.), Social causes of husband-wife violence.
Minneapolis, MN: University of Minnesota Press.
Indeed, a study by Tuccio and Wahba (2018) finds that women in Andersson, N., Cockcroft, A., Ansari, N., Omer, K., Chaudhry, U. U., Khan, A., &
Jordanian households with return migrants from the Gulf are more Pearson, L. (2009). Collecting reliable information about violence against
likely to internalize discriminatory gender norms than women in women safely in household interviews: Experience from a large-scale
national survey in South Asia. Violence Against Women, 15(4), 482–496.
households without migrants. Bang, J. T., Mitra, A., & Wunnava, P. V. (2016). Do remittances improve income
As such, while our results are consistent with the literature in inequality? An instrumental variable quantile analysis of the Kenyan case.
that we find a robust positive impact of remittance inflows on a Economic Modelling, 58, 394–402.
Bartus, T. (2005). Estimation of marginal effects using margeff. The Stata Journal, 5
particular aspect of gender relations in Punjabi society, the causal (3), 309–329.
impact is more likely to operate via reducing poverty and provid- Baudassé, T., Bazillier, R., & Issifou, I. (2018). migration and institutions: Exit and
ing the household with a source of insurance for negative income voice (From Abroad)?: Migration and institutions. Journal of Economic Surveys,
32(3), 727–766.
shocks than through the import of less patriarchal gender norms. Beine, M., Docquier, F., & Schiff, M. (2013). International migration, transfer of
This suggests that policies taken to improve the governance of norms and home country fertility. Canadian Journal of Economics, 46(4),
migration such as providing potential migrants with information 1406–1430.
Bertoli, S., & Marchetta, F. (2015). Bringing it all back home – Return migration and
on employment prospects both abroad and in domestic urban cen-
fertility choices. World Development, 65, 27–40.
ters, subsidizing the cost of travel and making it safer, providing Bhattacharyya, M., Bedi, A. S., & Chhachhi, A. (2011). Marital violence and women’s
migrants with legal and other means of institutional support, and employment and property status: Evidence from North Indian villages. World
Development, 39(9), 1676–1689.
reducing the cost of sending remittances may have significant pos-
Buller, A. M., Peterman, A., Ranganathan, M., Bleile, A., Hidrobo, M., & Heise, L.
itive impacts on gender relations in addition to the poverty allevi- (2018). A mixed-method review of cash transfers and intimate partner violence
ation impacts emphasized in the literature.31 in low and middle-income countries. World Bank Research Observer, 33,
218–258.
Cools, S., & Kotsadam, A. (2017). Resources and intimate partner violence in sub-
Saharan Africa”. World Development, 95, 211–230.
Declaration of Competing Interest Doyle, D. (2015). Remittances and social spending. American Political Science Review,
109, 785–802.
The authors declare that they have no known competing finan- Dutton, D. (2006). Rethinking domestic violence. Vancouver, BC: University of British
Columbia Press.
cial interests or personal relationships that could have appeared
Escriba-Folch, A., Meseguer, C., & Wright, J. (2015). Remittances and
to influence the work reported in this paper. democratization. International Studies Quarterly, 59, 571–586.
Eswaran, M., & Malhotra, N. (2011). Domestic violence and women’s autonomy in
30
developing countries: Theory and evidence. Canadian Journal of Economics, 44
We should clarify that our data does not tell us where the migrant members of (4), 1222–1263.
the household are located. It merely tells us if they are inside or outside the country. Ghorpade, Y. (2017). Extending a lifeline or cutting losses? The effects of conflict on
31
The Migrant Resource Centers established by the government of Pakistan with household receipts of remittances in Pakistan. World Development, 99, 230–252.
support from the International Labour Organization (ILO) exemplifies the type of Goode, W. J. (1971). Force and violence in the family. Journal of Marriage and the
initiatives taken so far. Family, 33, 624–636.

10
A. Mitra et al. World Development 138 (2021) 105149

Goodman, M. L., Serag, H., Raimer-Goodman, L., Keiser, P., & Gitari, S. (2017). Sekhri, S., & Storeygard, S. A. (2014). Dowry deaths: Response to weather variability
Subjective social standing and conflict tactics among young Kenyan Men. in India. Journal of Development Economics, 111(C), 212–223.
American Journal of Community Psychology, 60(1–2), 257–266. Stark, O., & Taylor, J. E. (1991). Migration Incentives, Migration Types: The Role of
Iacus, S. M., King, G., & Porro, G. (2012). Causal inference without balance checking: Relative Deprivation. Economic Journal, 101(408), 1163–1178.
Coarsened exact matching. Political Analysis, 20(1), 1–24. Taylor, J. E., & Wyatt, T. J. (1996). The shadow value of migrant remittances, income
Ishfaq, S., Ahmad, V., Hassan, D., & Javed, A. (2017). Internal Migration and Labor and inequality in a household-farm economy. Journal of Development Studies, 32
Mobility in Pakistan. In S. Irudaya Rajan (Ed.), South Asia Migration Report 2017: (6), 899–912.
Recruitment, Remittances, and Reintegration. London and New York: Routledge. Tran, T. D., Nguyen, H., & Fisher, J. (2016). Attitudes towards intimate partner
Javed, M., Awan, M. S., & Waqas, M. (2017). International migration, remittances violence against women among women and men in 39 low- and middle-income
inflow and household welfare: An intra village comparison from Pakistan. Social countries. PLoS ONE, 1(11) e0167438.
Indicators Research, 130(2), 779–797. Tuccio, M., & Wahba, J. (2018). Return migration and the transfer of gender norms:
Jewkes, R. (2002). Intimate partner violence: Causes and prevention. Lancet, 359, Evidence from the Middle East. Journal of Comparative Economics, 46(4),
1423–1429. 1006–1029.
King, G., & Nielsen, R. (2019). Why propensity scores should not be used for Tyburski, M. D. (2014). Curse or cure? Migrant remittances and corruption. The
matching. Political Analysis, 27(4), 435–454. https://doi.org/ Journal of Politics, 76(3), 814–824.
10.1017/pan.2019.11. Vyas, S., & Watts, C. (2009). How does economic empowerment affect women’s risk
Lawson, J. (2012). Sociological theories of intimate partner violence. Journal of of intimate partner violence in low- and middle-income countries? A
Human Behavior in the Social Environment, 22, 572–590. systematic review of published evidence. Journal of International Development,
Levitt, P. (1998). Social remittances: Migration driven local-level forms of cultural 21(5), 577–602.
diffusion. International Migration Review, 32(4), 926–948. Williams, R. (2012). Using the margins command to estimate and interpret adjusted
Mughal, M., & Makhlouf, F. (2013). Labour effects of foreign and domestic predictions and marginal effects. The Stata Journal, 12(2), 308–331 https://
remittances – evidence from Pakistan. International Review of Applied www.stata-journal.com/article.html?article=st0260.
Economics, 27(6), 798–821. Yount, K. M., Halim, N., Schuler, S. R., & Head, S. (2013). A survey experiment of
Nenova, T., Niang, C. T., & Ahmad, A. (2009). Bringing finance to Pakistan’s Poor: women’s attitudes about intimate partner violence against women in rural
Access to finance for small enterprises and the underserved. Washington, DC: Bangladesh. Demography, 50(1), 333–357.
World Bank. Zakar, R., Zakar, M. Z., & Krämer, A. (2012). Voices of strength and struggle:
Pfutze, T. (2014). Clientelism versus social learning: The electoral effects of Women’s coping strategies against spousal violence in Pakistan. J Interpers
international migration. International Studies Quarterly, 58, 295–307. Violence, 27(16), 3268–3298.
Rapoport, H., & Docquier, F. (2006). The Economics of Migrants Remittances. In S.
Kolm & J. M. Ythier (Eds.). Handbook of the Economics of Giving, Altruism and
Reciprocity (Vol 2, pp. 1135–1198). Amsterdam: Elsevier.

11

You might also like