Pitt Et Al 2003 #Does Microcredit Empower Women
Pitt Et Al 2003 #Does Microcredit Empower Women
Pitt Et Al 2003 #Does Microcredit Empower Women
by
Mark M. Pitt
Shahidur R. Khandker
and
Jennifer Cartwright
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the
exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if
the presentations are less than fully polished. The papers carry the names of the authors and should be cited
accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the
authors. They do not necessarily represent the view of the World Bank, its Executive Directors, or the
countries they represent. Policy Research Working Papers are available online at http://econ.worldbank.org.
1. Introduction
In recent years, governmental and non-governmental organizations in many low income countries have
introduced credit programs targeted at the poor. Many of these programs specifically target women
based on the view that they are more likely than men to be credit constrained, have restricted access to
the wage labor market, and have an inequitable share of power in household decision-making. The
Grameen Bank of Bangladesh is perhaps the best-known example of these small-scale production
credit programs for the poor, and over 90 percent of its clients are women. Earlier work (Pitt and
Khandker (1998), Pitt, Khandker, McKernan, and Latif (1999), Pitt, Khandker, Choudhury, and
Millimet (2003), Pitt (2001)) has found that the effects of program participation differ importantly by the
gender of program participant. For example, Pitt and Khandker find that the flow of consumption
expenditure increases 18 taka for every 100 taka borrowed by women, but only 11 taka for every 100
taka borrowed by men. Pitt, Khandker, Choudhury, and Millimet (2003), using a totally different
approach to parameter identification, find that credit provided women importantly improves measures of
health and nutrition for both boys and girls, while credit provided men has no significant effect.
What underlies these gender differences? There are essentially two different mechanisms that
can result in different effects of credit program participation by gender: (i)“empowerment” effects, and
(ii) standard income and substitution effects. Collective models of household decision-making provide
one avenue of understanding “empowerment.” In a simple version of collective decision-making, the
household's social welfare is some function of the individual utility functions. Browning and Chiappori
(1998) have shown that if behavior in the household is Pareto efficient, the household's objective
function takes the form of a weighted sum of individual utilities, with weights t. The weight t can be
thought of as representing the bargaining power of the female household member relative to the male
household member in determining the intra-household allocation of resources. When t is zero, female
preferences are given no weight and the household's social welfare function is identically that of the
male. In much of the literature, t is presumed to be increasing in the relative value of female time and her
money income. In addition, t may be altered through social pressure. The parameter t, which directly
reflects women’s power in household decision-making, is one index of “women’s empowerment.”
The differing credit effects by gender of participant reported by Pitt, Khandker and associates
do demonstrate that an empowerment effect as a consequence of credit program participation; they can,
in principle, be only the result of standard income and substitution effects. In an economy in which
women do not work in the wage labor market, participation in a group-based credit program increases
the shadow value of female time by providing a complementary input for the production of goods for the
market by the self-employed. In contrast, if men still provide time to the wage labor market, the
shadow value of their time is unaffected by program participation. Consequently, the self-employment
activities of women fostered by micro-credit may generate different demand effects than the self-
employment activities of men fostered by micro-credit. If the preference weight t is unaffected by male
participation, such participation does not alter the shadow price of women's time either. The only source
of change in demand when men are the credit program participator arises from the income effect
associated with the rental value of the capital endowment provided by the credit program. Note that
although male participation identifies the income effect conditional on t, this information does not help
disentangle the substitution effect from the bargaining (empowerment) effect induced by women's
participation. Thus a finding that the effect of women's program participation on child health differs from
the effect of men's program participation (as in Pitt et. al. (2003)) cannot be taken to necessarily imply
that women have gained power in the household, even if women are assumed to prefer child quality
more than their husbands.
A modeling strategy that seeks to separate out the income and substitution effects from the
empowerment effect (on t) resulting from micro-credit program participation would make difficult
demands on the data and require strong restrictions on the form of preferences. An alternative
approach is to collect data on attitudes by and towards women, and on their decision-making
autonomy. This data is necessarily self-reported and subjective, but econometric techniques, notably
instrumental variables estimation, are available to correct for the possible confounding effects of
systematic variation in subjective response. Note that self-reported measures of decision-making
power, even if experimentally elicited, do not necessarily imply that women actually have more power
(as measured by t), but they do add one more piece to the accumulated evidence pointing in that
direction.
2
This paper estimates the impact of participation in micro-credit programs on a large set of
qualitative responses to questions that characterize women’s autonomy and gender relations within the
household. The data come from an extensive survey household survey collected in rural Bangladesh in
1999. We test the assertion that participating in micro-credit programs is an empowering experience for
women whose life choices are otherwise restricted through poverty, patriarchy, and societal or religious
norms. In addition, we examine the effect of men’s credit program participation on these same measures
of female empowerment.
2. Previous Studies
Over the past fifteen years or so, a substantial literature has been produced on various aspects of micro-
credit programs in poor communities. A few studies of these studies have focused on the relationship
between credit program participation and some notion of women’s empowerment.
Goetz and Sengupta (1996) present a decidedly negative image of the effect of credit on
women's empowerment. Using a five-level scale reflecting the degree of control that women have over
the loans they take, they conclude that most women have a minimal level of control over their loans, and
that when the time comes for loans to be repaid, this lack of control can have a damaging impact on the
well-being of women. At best, they reason, women who have little or no control over their loans will
also not be held responsible for repaying them and thus they will be left out of the process altogether
and any special impact of lending to women rather than men is neutralized. In cases where men have
appropriated loan funds and are subsequently unable or unwilling to repay the loans, women may suffer
because they are forced to sell assets or go hungry in order to raise the money to repay. Furthermore,
the authors suggest the potential for women's credit participation to worsen the degree of domestic
abuse they suffer.
The focus of the study by Goetz and Sengupta is not on empowerment per se, but on women's
managerial control over loan use. The authors find that, according to their criteria, less than 18% of
women in the sample retained "full control" over the loans they took from credit programs. 39% of
respondents were judged to have very little control or no control at all over their loans. The authors
3
make assertions that credit is fungible within the household, but do not support these assertions
empirically.
Hashemi et. al. (1996) find that membership in Grameen Bank and BRAC have significant
positive effects on empowerment even controlling for women's independent contributions to household
income. They find that even in cases where credit program members do not contribute independently
because their husbands appropriate their loan funds, because the loans they have taken are not
generating income, or for various other reasons, just the experience of being a member of the program is
beneficial for empowerment.
Hashemi et. al. acknowledge the problem of selection bias and the possibility that positive
effects of credit program participation on empowerment are biased upwards. To remedy this, they
control for the respondents' demographic and socio-economic characteristics, specifically age,
education, relative wealth, religion, geographic division and surviving sons and daughters. Unfortunately,
there is no effort to control for the significant unobserved heterogeneity that remains. This unobserved
heterogeneity likely includes the unobserved attitudes and characteristics of husbands, wives and other
family members, including pre-existing women’s empowerment and autonomy. It seems quite possible,
for example, that more empowered women are more likely to be able to join a micro-credit program.
Hashemi et. al. also include a variable representing duration of membership to test whether
there is a change over time in the effect of credit on empowerment. They consider this variable as an
additional means of controlling for selection bias because "a significant duration effect would strongly
suggest that credit programs further empower the women who join them." In making this assertion, the
authors fail to recognize that, just as the decision to join a credit program is endogenous and likely to be
correlated with the unobservable empowerment endowment, so too are (1) how early one joins a
program, and (2) whether (and when) one decides to discontinue membership in the program. Thus it is
likely that unobservable heterogeneity in empowerment is correlated with duration of membership. It
would be plausible to suggest, for example, that women who are initially more empowered might remain
in credit programs while those who are relatively less empowered drop out due to family pressure,
inability to use credit effectively, lack of confidence in one's own ability to invest wisely, or any number
4
of other (empowerment related) factors.
The methodology used by these authors contrasts program villages to a "comparison sample" of
non-program villages based on region, population density, and village size. Due to the impossibility of
finding two villages that are "identical" in all characteristics that might affect measures of empowerment,
such an approach is inherently problematic because it neglects the potential for village-level
unobservable characteristics to bias the results.1.
Interestingly, Hashemi et. al. note that among non-participants, residence in villages with
Grameen Bank programs has a positive and statistically significant effect on empowerment. They note
that this could be the result of (1) non-random program placement and/or (2) spillover effects (such that
the existence of a Grameen Bank program changes village society in such a way as to effect the
empowerment of non-participants as well as participants) and that it is impossible to disentangle the two
effects. It should be noted that whereas the former creates heterogeneity bias, the latter actually
represents effects of the program, and thus contains information on the ability of programs to empower
women (both members and non-members residing in the same villages). As described in Pitt and
Khandker (1998), the existence of spillover effects does not affect the consistency of any estimate of
the effect of credit on a dependent outcome, but it does alter the interpretation of the estimate.
Hashemi et. al. (1996) create an “index” of empowerment through a linear weighted
combination of individual empowerment indicators. The authors do acknowledge the arbitrariness of this
index approach. They establish a cutoff point at the 30th percentile (again, arbitrarily chosen) such that
women who score above this cutoff are labeled empowered and those who score below it are labeled
unempowered This system reduces the measure of empowerment, previously existing along a
continuum, to a single binary outcome for each of eight categories. These eight categories are further
1
Indeed, as will be discussed below, we find strong evidence that village-level heterogeneity is an important source
of bias exists in this sample. Hausman and Breusch-Pagan tests on all emp owerment factors in this study reveal that
a fixed-effects model at the thana (sub-district) level is insufficient to correct for this bias, implying that village level
heterogeneity is a significant source of bias even controlling for all thana-level heterogeneity.
5
compacted into a "composite empowerment indicator" such that a woman was labeled empowered
overall if she had been labeled "empowered" in at least three of the eight categories and was labeled
unempowered otherwise. Again, the choice of five-out-of-eight as the cutoff mark by which to reduce
the eight categorical binaries to one single binary represents an arbitrary choice on behalf of the
researchers.
A study by Mizan (1993) uses a similar approach to that of Hashemi et al (1996). Mizan also
uses an index, called the Household Decision Making (HHDM) Scale, which is computed from answers
to questions regarding: decisions of food purchase, education and marriage of children, expenses on
medication for self and husband, investment woman's earnings in business, purchase and sale of land,
hiring of outside labor, purchase of agricultural inputs, providing financial support to husband's family,
and purchase of clothes for self and other household members. The coding used is as follows: decision
made by husband only=1, decision made jointly by husband and wife=2, and decision made by wife
only=3. Thus, the DMS registers a higher value for a higher level of female bargaining power. Mizan's
study uses a sample of 100 participating women chosen from two villages (50 women in each village)
and 100 non-borrowers, without control for self-selection into the programs.
Mizan finds that the number of years a woman had borrowed from the Grameen Bank and the
approximate monthly income from the Grameen Bank investment both had a positive and statistically
significant effect on the HHDM score. The conclusion of the study is that Grameen Bank participation
raises women's decision-making power within households because it increases women's employment
and income earnings. The study also finds that participation has a significant effect on fertility control
ability. The variables "Income" and "Years of loan" are both positively correlated with the HHDM score
when the other is controlled for. Also, "Income" (from Grameen Bank) is significant when participation
(dummy) is controlled. Mizan concludes that "this suggests that apart from the financial resources a
woman gains, an effect of the experience with Grameen Bank is important by itself." (120)
3. Data
The data used in this paper come from a large household survey conducted in 1998/99, which is a
6
follow-up survey of an earlier survey conducted in 1991/92. Both household surveys were conducted
by the Bangladesh Institute for Development Studies (BIDS) in collaboration with the World Bank.
Only the follow-up survey (conducted in 1998/99) included a special module on women’s
empowerment.
The base household survey interviewed 1,798 households randomly drawn from 87 villages of
29 thanas in rural Bangladesh. Of these 29 thanas, 24 were program thanas (8 from each of the three
programs: Grameen Bank, BRAC, and BRDB RD-12 project), and 5 were non-program thanas.
Three villages in each program thana were randomly selected from a list of program villages in which a
program had been in operation for at least three years. Three villages in each non-program thana were
also randomly selected from the village census of the Government of Bangladesh. From the village
census list of households, 20 households from each village were drawn using stratified random sampling.
Out of these households, 17 were target (owned land of one-half acre or less, and hence qualified for
program participation) and 3 nontarget (owned land of more than one-half acre, and hence did not
qualify for program participation). To ensure that a sufficient number of program participating
households were included in the target households in program villages, participant households were
overdrawn. 2 Of the 1,798 households selected, 1,538 were target and 260 were nontarget households.
Among the target households, 905 (59 percent) participated in a credit program. The program villages
surveyed had either male and female credit groups, or both: 40 had credit groups for both men and
women, 22 had female-only groups, and 10 had male-only groups. The existence of villages with only
female or only male groups is a key feature of the parameter identification method described below. A
more detailed description of this survey can be found in Khandker (1998).
These households were revisited in 1998/99. The resurvey tried to include all households from
the1991/92 survey, including splits, plus added some new households.3 A sample of 2074 households
2 An additional 58 households were selected from 15 villages of 5 program thanas (covering all three programs),
because a nutrition survey was additionally conducted in those villages and a larger number of target households
was required.
3 After the 1991/92 survey, one or more microcredit programs moved to some control villages of 1991/92 survey,
making them program villages. So three new thanas (with three villages in each thana) were added. In addition, two
more villages were added to previous nonprogram thanas. In the program thanas, six new villages were added.
Altogether 104 villages from 32 thanas were included. 131 were missing during the resurvey. Up to 4 new households
7
with married couples was administered the women’s empowerment questionnaire. Table 1 shows the
distribution of households across the eight categories of program credit, broken down by gender. Table
2 lists all of the empowerment questions asked. The name of the variable, the full (translated) text of the
question, the coding of the variable (“Y” standing for a “yes” answer and “N” standing for a “no”
answer), and an indicator of who was asked the question (husband or wife) are provided.
Approximately 80 percent of the questions were asked only of wives.
The survey questions are grouped into the following headings:
b. Purchasing capacity
For seven categories of common household purchases (food, toiletries, candies for the children, cooking
8
utensils, furniture, children's clothing, and own clothing) women were asked whether or not they (rather
than someone else in the household) make the purchase and, if so, whether or not they make the
purchase without their husbands' permission. The percentage of women who answered that they make
purchases themselves varies widely by category, from less than five percent (for furniture) to more than
sixty percent (for candies and household utensils). When husbands were asked about their wives'
freedom to make purchases, 87% responded that their wives are not able to buy assets on their own
without the husband's permission.
4
See, for instance, Montgomery, Bhattacharya, and Hulme (1996), Goetz and Sengupta (1996), and Pitt and Khandker
(1998).
9
d. Control over income and savings
Traditionally, women in Bangladesh have very little contact with the labor market and generally do not
have significant cash incomes of their own. This reflects customary and religious restrictions on women's
mobility outside the home. 62% of men reported that their wives have no independent source of income.
Over 75% of women reported that they do not operate any income-generating activity of their own and
78% of women reported not having independent income that they could use at their own discretion
(without consulting their husband). A sizeable number (42%) of women reported that they do have their
own independent savings, and if they did, husbands were aware of these savings 91% of the time.
Wives expressed having a low level of control over these savings, with 85% saying that they were not
able to decide autonomously how to utilize them.
Only around 15% of women reported having received money from their parents, siblings, or
other blood relatives in the past 12 months. Of these, 95% said that their husbands knew that they had
received this money. Only 17% reported that they had full control over deciding the use of that money:
62% reported partial control and 21% reported having no control at all.
More the three quarters of women (78%) reported that they had at some point been forced to
cede money to their husbands and 56% of women replied that their husbands had forced them not to
work outside the home. 81% reported that they would not be able to give their own money away at
will.
When asked if they would be able to get 500 taka in the case of an emergency, two-thirds of
women predicted that they would be able to. The primary sources from which women predicted they
would borrow such emergency money were from own relatives (32%), husbands (29%), and husband's
relatives (28%). Less than 3% of women in the sample replied that they would borrow from
moneylenders.
e. Mobility
In Bangladeshi society, the physical mobility of women is often restricted. Traditions and family-imposed
restrictions may forbid women from leaving the family compound, or may regulate when, where, and
10
with whom they travel. Additionally, issues of safety often prevent women from traveling alone for even
short distances. 83% of husbands reported that their wives never went alone to places such as the
market, bank, health clinic, and so on. Of these, over half (55%) explained that they or their sons
always accompanied the wives when going outside the home and another 18% explained that their
wives were accompanied by neighbors or relatives. Wives responded similarly. 53% said that when
they traveled outside the village they went with their husbands and/or sons, and 22% traveled in the
company of other women. Almost 9% of women reported that they never left the village at all. 82% of
women said that they had never visited their parents without their husband's permission.
11
h. Family planning
In this sample, women were more likely than men to be users of birth control. Among couples in their
reproductive age, over 93% of men reported that they did not use any male birth control method.
Among these men, 65% explained that the reason was that their wives used a female birth control
method, and 16% responded that they simply did not like to use birth control. Women's responses were
similar: over 91% of women reported that they had never been able to make their husbands use a male
birth control method. Of these women, 68% explained that the responsibility of birth control was usually
given to them.
i. Attitudes
The survey also included several questions for both husbands and wives regarding their opinions and
attitudes on gender in society. More than two-thirds of men (68%) replied that they believe their wives
to be less intelligent than themselves. 79% replied that they do not consider their wives capable of
making decisions pertaining to purchase or sale of major household assets. An overwhelming majority of
women (94%) stated that they believe that their husbands are superior to them "in qualities and
education." When asked why, 59% of these women explained that the husband is the earning member
of the household and that this makes him superior, and 34% stated that a woman's lot in life is to be
inferior to her husband. When asked what kind of impact women's empowerment would have (or was
having) on society, men were fairly evenly split between positive and negative reactions. Roughly half
(47%) responded positively by claiming that the primary impact of women's empowerment would either
be the creation of a better society or that it would be economic improvement for the family. The other
53% responded negatively, saying that women's empowerment would cause chaos in society, problems
bringing up children, or a disruption of peace within the household.
When asked to describe what they perceived to be the greatest obstacles to achieving women's
empowerment in Bangladeshi society, 46% of men cited lack of education as the primary obstacle, 23%
cited lack of safety, and 17% cited religious restrictions. As secondary obstacles, men also cited
religious restrictions (30%), lack of income generating activities (22%), lack of safety (21%) and the
12
social structure (18%). The main obstacles cited by women were lack of education (47%), lack of
safety (21%), and religious restrictions (16%).
C ij = X ij β c + Z ijπ + µ j + ε ij
c c
(1)
where Xij is a vector of household characteristics (e.g. age and education of household head), Zij is a set
of household or village characteristics distinct from the X's in that they affect Cij but not other household
5
The quantity of credit is, of course, only one measure of the flow of services associated with participation in any one
of the group-based lending programs. These programs are more than just lending institutions. Nevertheless, the
quantity of credit is the most obvious and well measured of the services provided.
13
behaviors conditional on Cij (see below), ßc, and p are unknown parameters, µ cj is an unmeasured
determinant of Cij that is fixed within a village, and ε cij is a nonsystematic error that reflects unmeasured
where ßy and d are unknown parameters, µ yj is an unmeasured determinant of yij that is fixed within a
village, and ε ijy is a nonsystematic error reflecting, in part, unmeasured determinants of yij that vary over
households. The estimation issue arises as a result of the possible correlation of µ cj with µ yj , and of ε cij
withε ijy . Econometric estimation that does not take these correlations into account may yield biased
estimates of the parameters of equation (2) due to the endogeneity of credit program participation Cij.
The standard approach to the problem of estimating equations with endogenous regressors,
such as equation (2), is to use instrumental variables. In the model set out above, the exogenous
regressors Zij in equation (1) are the identifying instruments. Unfortunately, it is difficult to find any
regressors Zij that can justifiably be used as identifying instrumental variables. Lacking identifying
instruments Zij, the sample survey was constructed so as to provide identification through a quasi-
experimental design.
Our sample of households includes households in villages that do not have access to a group-
based credit program. If credit program placement across the villages of Bangladesh is attentive to the
village effects µj, identifying program effects by comparing households in nonprogram villages with
households in program villages without controlling for the selectivity of program placement will generally
result in biased estimates of program effects. Using a village fixed effects estimation technique may
remove the source of correlation between program placement and the empowerment behavior of
interest, however, without further exogenous variation in program availability, the credit effect is not
14
identifiable from a sample of self-selected households.6 The parameter of interest, d, the effect of
participation in a credit program on the outcome yij, can be identified if the sample also includes
households in villages with treatment choice (program villages) who are excluded from making a
treatment choice by exogenous rule. That exogenous rule is the restriction that households owning more
than 0.5 acres of cultivable land are precluded from joining any of the three credit programs.7
To illustrate the identification strategy, consider a sample drawn from two villages —village 1
does not have the program and village 2 does; and, two types of households, landed (Xij=1) and
landless (Xij=0). Innocuously, we assume that landed status is the only observed household-specific
determinant of some behavior yij in addition to any treatment effect from the program. The conditional
demand equation is:
y ij = C ij δ + X ij β y + µ yj + ε ijy (3)
The exogeneity of land ownership is the assumption that E(Xij, ε ijy ) = 0, that is, that land ownership is
uncorrelated with the unobserved household-specific effect. The expected value of yij for each
household type in each village is:
6
In addition, the effect of any observed village characteristics that are thought to influence y ij , such as prices and
community infrastructure, are not identifiable.
7
The validity of the assumption that landownership is exogenous is defended at length in Pitt and Khandker (1998).
There are a number of households in our sample that were program participants and yet had more than 0.5 acres of
land at the time of program entry, raising the possibility of mistargeting and potential bias in econometric results
relying on this targeting rule. It appears that some of this excess land is either uncultivable or marginally so. Pitt
(1999) demonstrates that the value per acre of land owned by program participating households who also own more
than 0.5 acres of cultivable land at the time of joining is a small proportion of the value per acre of the cultivable land
of program participants owning less than 0.5 acres of cultivable land at the time of joining. This suggests that
program officers are using some notion of “effective” units of cultivable land in determining eligibility rather than of
the type of mistargeting that would result in econometric bias. Pitt (1999) discusses this issue at length and
demonstrates that treating the exogenous targeting rule to be greater than 0.5 acres provides a consistent estimator
for certain types of mistargeting. He finds that application of targeting rules greater than 0.5 acres (up to 2.0 acres)
actually slightly strengthens the qualitative results on the effect of credit by gender on household consumption.
15
where ? is the proportion of landless households in village 2 who choose to participate in the program.
It is clear that all the parameters, including the effect of the credit program, d, is identified from this
design. In particular, the estimator of the program effect d is a variant of the differences-in-the-
differences estimator widely applied in the general program evaluation literature. To see this, note that
an estimate of d is obtained from the following difference-in-the-difference:8
[E(yij | j=2, Xij=0) - E(yij | j=2, Xij=1)] - [E(yij | j=1, Xij=0) - E(yij | j=1, Xij=1)] (4e)
If landed status is a continuous measure of landholding, then the credit effect d is identified from
variation in landholding within the program villages (j=2) and a sample of nonprogram villages is not
required.
Even if land ownership is exogenous for the purposes of this analysis, it is necessary that the
“landless” and the “landed” can be pooled in the estimation. In order to enhance the validity of this
assumption, we restrict the set of nontarget households used in the estimation to those with less than 5
acres of owned land. In addition, we include the quantity of land owned as one of the regressors in the
vector Xij and include a dummy variable indicating the target/nontarget status of the household.
The exclusion restrictions that identify the effects of credit on the outcomes yij are the
interactions of a dummy variable indicating if the household has the choice to join the credit program
(which requires meeting the land ownership rule and residing in a village with a credit program) and all
the exogenous variables of the model, Xij and µj.9 In the results reported below, these instrumental
variable models are estimated by two-stage least squares.
An important question of this research is whether reported empowerment is affected differently
by credit if the program participant is a woman or a man. For that reason, the reduced form credit
equation is disaggregated by gender:
This insensitivity of results to the choice of targeting rule used in estimation to further demonstrated in Pitt (2001).
8
However, as Pitt (1999) points out, since this is a quasi-experiment, not an actual experiment, the direct application of
(4e) would most likely result in a downward biased estimate of d. The regression approach applied here is necessary
to control for differences in other observed and unobserved variables across the four groups identified in equations
(4a) though (4d).
9
Consequently, the model is not nonparametrically identified. That is, if the linear indices Xc ? and (Xy ß + dIc ) in (5)
were replaced by nonparametric functions of the X's, and Ic the model is not identified.
16
C ijf = X ij β cf + µ jf + ε ijf
c c
(6)
C ijm= X ij β cm + µ jm + ε ijm
c c
(7)
where the additional subscripts f and m refer to females and males respectively. The conditional
household outcome equation is then:
where Djf and Djm are village specific indicator variables such that Djf takes the value of one in village j if
there is a female group in village j, and zero otherwise.
Additional identification restrictions are required when there are both male and female credit
programs with possibly different effects on behavior. Identification of gender-specific credit is achieved
by making use of another quasi-experimental attribute of these programs and the survey. All program
groups are single-sex and not all villages have both a male and a female group. The sample includes
some households from villages with only female credit groups, so that males in landless households are
denied the choice of joining a credit program, and some households from villages with only male credit
groups, so that landless females are denied program choice.
17
This paper treats subsets of empowerment variables as containing an underlying latent factor,
estimates index “weights” using the methods of factor analysis, and computes numerical estimates of the
latent factor. Factor analysis is a set of statistical techniques often used when the number of true
“underlying dimensions” that describe a condition (such as empowerment) is smaller than the number of
observed variables. Factor analysis converts a large number of observed variables into a smaller number
of hypothetical variables, called factors, each of which is a linear combination of several observed
variables. The use of factor analysis implies that the relationships between certain types of observed
variables are stronger than those between others, such that if observed variables are arranged into
appropriate groups, the correlation among variables within groups will be higher than the correlation
across groups.
The decision to employ factor analysis is based upon our prior belief about the nature of
empowerment. At one extreme, we could postulate that all the variables in the study are causally
determined by only one factor, which we could call “empowerment.” An alternative approach, which
we follow, is to postulate that there may be more than one type of underlying empowerment factor, but
fewer than the number of observed variables. We think it sensible, for example, to expect that those
questions that pertain to political activism measure a different type of underlying condition than do those
questions that ask about reproductive control. Consequently questions are grouped into 10 thematic
groups to produce factors representative of certain topics.
The women’s empowerment survey provides discrete responses to all questions. The
conditional densities of the responses of person i to question j given the latent empowerment variable of
person i, ui, depend on the linear index given by
η ij = β j + u i λ j (9)
where ?ij is the (linear) index, ßj represents a question-specific threshold for a positive response, and ?j
represents the extent to which question j discriminates between persons having different levels of
empowerment. The ?j is the factor-loading of the latent variable in the linear index. This model is
known as the two-parameter item-response model, and has been used to estimate latent ability using
data from binary (true/false) test questions. Appending a non-systematic error eij such that the ui are the
18
only source of stochastic covariation among responses of any person, and assuming normally distributed
errors, this is essentially a random-effects probit model with varying error correlations. In particular, the
error correlation between question j and question k is proportional to ?j?k , as in Pitt (1997). In the
typical random effects model it is assumed that ?j = ?k for all j and k, so that there is a single error
correlation ? = ?2. Estimation of this model is accomplished by maximum likelihood using Gauss-
Hermite quadrature for numerical integration. After estimating the parameters ßj and ?j, an empirical
Bayes method is used to estimate the latent variable (random effect) for each person sampled. This
estimation was carried out with the gllamm6 package of Rabe-Hesketh and Pickles (2000). This is fairly
demanding computationally. The same estimation was also carried out using standard factor analysis for
models with continuous responses. In every case the simple correlation coefficent between the probit
random effects model and standard continuous variable factor analysis was above 0.95. The results
reported below are based on standard factor analysis, which has the advantage of readily providing
additional statistics on the “fit” of the approach.
Table 3 presents information on the construction of each of the ten factors and the results of the
factor analysis. For each factor, the eigenvalue, which measures the degree to which the variance of
variables is accounted for by the factor, is listed along with the names of all the observed variables
(component variables) that were used to create the factor.10 For each component variable, two values
are presented. The first column of values presents the factor loading for each component variable, which
is the simple correlation coefficient for this variable and the factor. The sign of each factor loading
indicates the sign of this correlation. If all factor loadings are of the same sign, this confirms that all
variables do indeed “fit” in the grouping used to produce the factor. The second column of values
presents the uniqueness of each component variable, which is the portion of the observed variance
unaccounted for by the common factor and hence unique to that variable. It is computed by dividing the
eigenvalue for the individual variable by the sum of all eigenvalues for all variables.
Those variables having at least 1800 observations (out of a potential sample of
10
The component variables for the tenth factor are not listed, for the sake of brevity. All the variables in table 2 are
component variables for this factor.
19
2074) were used in factor analysis. 101 out of 132 variables met this criterion and were thus eligible for
use. The main reason for lack of observations among the remaining 31 variables was that the associated
survey questions were answered contingent on the response to a previous question. For example, the
question "Do you buy household food?" was answered by all 2074 women in the sample, but the
follow-up question "If so, do you buy this food without your husband's permission?" was only answered
by the 372 women who responded affirmatively to the first question.
The selection of variables in the 10 categories of empowerment was based on our prior belief
about which variables contain similar types of information. Out of the 101 eligible variables, only 75
were actually used in factor analysis (most were used only once, but some were used to create several
different factors). The other variables were not used since it was felt that they were not directly relevant
to any of the factor themes.
Since our prior beliefs about which empowerment variables should be grouped together may
not be universally shared, regression analysis was performed not only on the ten factors but also on all
of the observed variables, including those that were not included in any factor grouping. Throughout the
paper, the ten hypothetical variables created through factor analysis are referred to as “factors” and the
observed variables from the women’s empowerment questionnaire are referred to as “individual
variables.”
Estimation of the determinants of the binary responses to individual empowerment questions is
complicated for some variables for which there is little or no variation within villages. The problem is that
the village fixed effect perfectly predicts the outcomes for the village. The village fixed effect goes to plus
infinity if all responses are ‘1' and negative infinity if all responses are ‘0'. This identification problem can
be cured with the additional sample variation resulting by using thana rather than village fixed effects.
There are three or four sample villages in each sample thana. However, it is first important to determine
whether a thana fixed effect/village random effects models eliminates location-specific heterogeneity
bias. Hausman and Breusch-Pagan tests were conducted, comparing a village fixed effects estimate to
thana fixed effect/village random effects. For all factors, the null hypothesis that village random effects
conditional on thana fixed effects provide consistent estimates was rejected. This means that significant
20
correlated heterogeneity exists across villages within thanas than results in bias in the thana fixed effects
model. Consequently, all fixed effects estimates presented in this paper are at the village level rather than
the thana level. It should also be noted that this finding provides strong evidence of the need for the use
of fixed-effects in the model to control for non-random program placement across locations. Indeed, the
finding can be interpreted to conclude that heterogeneity bias could arise from non-random program
placement across villages and within thanas, not only across thanas themselves, suggesting that
program placement in this sample is highly non-random. This further demonstrates the problematic
nature of any methodology (such as that used by Hashemi et. al.) that simply uses nearby non-program
villages as controls for program villages.
For each of the ten factors created to encompass thematic information on empowerment, a Wu-
Hausman test was conducted to determine whether male and/or female credit could be treated as
exogenous, that is, whether credit is uncorrelated with the residuals of the factor regression. The results
of these tests are presented in Table 4. For each factor, the results (P-value and t-statistic) are
presented for the three tests: for female credit exogeneity, male credit exogeneity, and joint (male-
female) credit exogeneity. Based on these results, the appropriate model is listed for each factor.
a. For six of the ten factors, the null hypothesis could not be rejected at the .05 level for a Wu
test of joint exogeneity of male and female credit. Thus the model adopted for these six factors was one
of exogenous male credit and exogenous female credit. For these six factors, both male and female
credit is treated as exogenous and the appropriate model is village fixed effects regressions with no
instrumental variables.
b. For three of the ten factors, the null hypothesis under the Wu test for joint
exogeneity of male and female credit could be rejected at the .05 level. This model is consistent with a
scenario in which person-specific unobservables are correlated with credit use, suggesting some degree
of self selection into credit programs based on unobserved traits which also affect empowerment as
measured by the factors. For these three factors, both male and female credit is treated as endogenous
and the model is village fixed effects with instrumental variables for both male and female credit.
21
c. For one of the factors, the null hypothesis for the Wu test of joint exogeneity of male and
female credit could not be rejected; however, the null hypothesis for the test of male credit only could
be rejected at the .05 level. For this factor, male credit is treated as endogenous while female credit is
treated as exogenous, and the model is village-fixed effects with instrumental variables for male credit
only.
In order to determine the appropriate regression model for the set of individual empowerment
variables, a set of Wu tests was also conducted. For each empowerment variable, tests were conducted
of male credit exogeneity, female credit exogeneity, and joint male and female credit exogeneity.11
Those variables for which the null hypothesis for all three tests could not be rejected at the .05 level are
treated as being fully exogenous: meaning that no instrumental variables are used at all, and the specified
model is village fixed effect with instrumental variables.
In cases where the null hypothesis in the test of female credit exogeneity could be rejected at the
.05 level but the null for the test of male credit exogeneity could not be, a model of female-credit
endogeneity is used. In this model, instrumental variables are used to correct for endogeneity of the
female credit variable but not of the male credit variable. Likewise for those variables in which the null
hypothesis for the test of male credit exogeneity could be rejected at the .05 level but the it could not for
the test of female credit exogeneity.
Finally, some variables are treated under a model of endogeneity for both male and female
credit, and instrumental variables are used for both. This model is applied to those empowerment
variables for which, either (1) the null hypothesis for the Wu test of joint exogeneity of male and female
credit could be rejected at the .05 level; or (2) both null hypotheses—for the test of male credit
exogeneity and the test of female credit exogeneity—could be rejected at the .05 level.
In all cases, estimated village fixed effects from the model with the corresponding factor are
included as independent variables to correct for heterogeneity bias resulting from non-random program
placement across villages. Several questions in the women’s empowerment questionnaire are only asked
11
For the sake of brevity, the statistics of all these tests (numbering more than 300) are not presented to the reader.
Instead, the presentation of final results for the individual variables in Table 6 indicates which model was employed.
22
conditional upon the response of a prior question. For example, one question asks women “Do you
have your own savings?” and the next question asks “If so, do you control these savings yourself?” Only
those respondents who answer affirmatively to the first question respond to the second question.
In cases such as this, the question was recoded to apply to the full sample. Using the example
above, the recoded question now asks “Do you have any savings which you yourself control?”
Respondents who have no savings and those who have savings that they do not control are both coded
as answering “no”. The only respondents with savings they control are coded as answering “yes”.
There were three conditional questions in the questionnaire for which this was not possible
because the conditionality of these questions did not rely on any other question in the questionnaire but
rather on some other (unmeasured) endogenous condition. These questions were the following: (1) “If
your wife has a loan, do you spend this money yourself?” (asked of men); (2) “If you have a loan, who
controls this money?” (asked of women), and (3) “If you have received any remittance from your
relatives in the past 12 months, do you yourself have control over this money?” (asked of women).
In order for a respondent to be in the sample for questions (1) and (2) it is necessary that the
wife have taken some loan. The determinants of whether this event occurs (and thus, whether
respondents are in the sample for questions (1) and (2)) are likely to be correlated with the determinants
of credit program participation. By definition, the determinants of “having a loan (from any source)”
must overlap with those of “having a loan (from a formal credit program)” because any woman who
“has a loan (from a credit program)” also “has a loan (from any source)”. For question (3), respondents
are only present in the sample if they received remittances in the past 12 months from their relatives. The
economic literature on transfer behavior suggests that it is highly unlikely that receiving family remittances
is orthogonal to credit program participation. 12
The problem of potential bias for these three questions was resolved by using a two-stage-
ordered-probit model (similar to that used to correct for self-selection bias) in which a the inverse Mill
The choice of the model was based on the Wu tests described and was conducted along the same lines as for the
factors.
12
The results of this paper demonstrate that female credit program participation has a statistically significant negative
effect on the likelihood that a respondent has received family remittances. Pitt and McKernan (2000) have also shown
that credit use causes a fall in net remittances from relatives of program participants to the participants.
23
ration corresponding to “being present in the sample” was predicted from a first stage probit and added
as a regressor in the second stage. Accordingly, the results from these three items may be interpreted as
“the likelihood of answering “yes” to question X, conditional upon being eligible to answer question
X.” Thus, for example, the results for question (1) should be interpreted as “the likelihood of a husband
spending his wife’s loan conditional upon the wife having a loan.”
6. Results
Below, the affects on latent empowerment factors of the male and female credit variables are presented
in Table 5 and on the response of individual questions within each factor are presented in Table 6. The
first column of both tables presents the effects of female and male credit under the assumption of
exogeneity (as in the work of Hashemi and others), that is, without instruments or fixed effects. Table 5
also present estimates with village fixed effects but without instruments. The last column of Table 5
imposes exogeneity wherever warranted by the Wu-Hausman tests presented in Table 4. Table 6
presents estimates without imposing exogeneity, when warranted, in its second column. The results of
these Wu-Hausman tests are indicated next to the parameter estimates.
Purchasing
24
Female credit use positively and significantly (t=3.15) affects the latent empowerment factor describing
women’s autonomy with purchasing. In addition, female credit significantly augments women’s ability to
purchase all seven questionnaire items in this category. Female credit also increases the likelihood both
that a husband states that his wife could buy assets on her own and that she could buy them without his
permission. In contrast, husbands credit program participation has a statistically significant (t=-3.55)
negative effect on the women’s purchasing autonomy factor.
Resources
Female credit significantly (t=10.24) increases the latent factor representing a woman’s access to and
control over economic resources. It also affects several individual indicators, including the likelihood that
a man says his wife has her own income, the likelihood that a wife reported having her own income, and
the likelihood of her reporting having her own savings (it did not, incidentally, affect the likelihood that a
woman had savings which she herself could control). In addition, female credit increases the likelihood
that a woman responds that she would be able to raise emergency funds from any source, and that she
would be able to raise them specifically from (1) selling off assets, (2) getting money from her husband,
and (3) borrowing from other people. Female credit decreases the chances that a household reports
that it fights about money. In contrast, male credit is significantly (t=-3.25) associated with lower latent
resource empowerment, and with reduced likelihood that wives have independent savings and access to
emergency funds.
Finance
Women’s credit significantly increases (t=5.84) the latent empowerment factor associated with finance.
In contract, men’s credit reduces this factor (t=-2.63).
Transaction Management
Female credit significantly (t=5.11) increases the factor representing a woman’s power to
oversee and conduct major household economic transactions, and male credit reduces this factor (t=-
2.53). Data from the questionnaire describes decision-making and implementation arrangements
25
(ranging from full power in the wife’s hands to full power in the husband’s) and the likelihood that a wife
spends money, for four major categories: housing repair, livestock purchase, household loans, and
land/equipment transactions. In all four categories, female credit affects women’s autonomy regarding
decision-making and project implementation. The same is true for the likelihood that a woman spends
money in every category except land/equipment transactions. Notably, the t-statistics for female credit
effects on women’s autonomy in deciding and implementing household finance decisions are especially
high. Male credit had a negative effect on wives implementing housing repair projects, livestock
purchase projects, and land/equipment purchase or sale projects.
Activism
Female credit positively affects the factor relating to women’s awareness and activism (t=3.20). Female
credit affects the odds that a woman will be informed of (meaning able to list) the ways in which
kabinnama (a pre-marital bridal contract) can be used to help a woman in the event of divorce. Female
credit also affects the probability that a woman knows the name of the Member of Parliament in her
26
area, the probability that she voted in the last election, and the probability that she voted independently
(rather than under advice/pressure from her husband).
Male credit reduces this factor, although not significantly (t=–1.14), and reduces the probability
that his wife will vote independently.
27
this factor. Female credit increases the likelihood that a woman initiates discussions with her husband
about birth control use, birth control methods, and birth numbers. In addition, female credit increases
the likelihood both that husbands will initiate discussion and that wives will initiate the same discussion
for issues of birth control use and children’s education (implying a positive effect on the total likelihood
of spousal communication on these two issues). Male credit had a negative effect on both the odds that
a wife initiated discussion regarding birth control use with her husband and on the odds that she initiated
discussion about birth control methods.
28
herself could control. Unfortunately, because of the way the question on the questionnaire was
phrased, we have no way of knowing who controlled this savings in cases where the respondent herself
did not. It is possible, then, that a lack of respondents’ control over their savings does not imply that
the respondents’ husbands controlled the savings, but rather that someone else (a third party) had
control. This is plausible in the context of the credit programs in question, which require women to make
periodic savings (thus, just as one would predict, highly increasing the odds that a woman has her own
savings) but which do not allow women complete control over their savings. This is because savings
become working capital within credit groups and are lent out to other members. As a result, it is quite
plausible that a woman would have her own savings as a result of joining a credit program, but would
not have savings that she herself (solely) controls.
Female credit has a negative effect on the odds that a woman reports having “received money
from parents/brothers/sisters or other relatives outside the household in last 12 months.” These results
are in agreement with Pitt and McKernan’s (2000) evidence that net remittances from participants in
credit programs to their relatives (meaning the excess of remittance from participants to families over the
remittance from families to participants) fall as a result of program participation. This result is actually not
overly surprising, and should not necessarily be interpreted as being “bad for empowerment.” Although
receiving remittances from her own relatives can be financially beneficial to a woman (and thus
“empowering”), the need to accept money from one’s kin (rather than, say, earning it oneself) may
actually be a sign of a low degree of command over economic resources. Thus, the explanation for
these results could be that participation in a credit program allows a woman to earn her own income and
thus reduce her dependency on her own family in order to get money. When the story is told in this way,
this result suggests an empowering rather than a disempowering effect of credit. It is also possible that it
is not women’s need for parental remittances that declines as a result of credit program participation,
but rather the willingness of the parents to remit money to the daughter. Since we do not have the
necessary data to discern whether the decrease in kin remittances caused by female credit are the result
of decreased demand or decreased supply, the result is best described as “ambiguous” in terms of its
potential effects on women’s empowerment or well-being.
29
Female credit use positively affects the chances that a woman cited “household chores” as a
subject over which members of her household argue. The question was phrased in such a way that in
order to answer the question, the respondent had to pick one argument topic out of a list, and “not
arguing about anything” was not an option. Thus, the female credit effect on “arguing about household
chores” is more appropriately interpreted as the effect on “arguing about household chores as opposed
to the other argument topics on the list of possibilities,” and thus is not an unambiguous measure of
empowerment.
7. Summary
This paper examines the effects of men’s and women’s participation in group-based micro-credit
programs on various indicators of women’s empowerment using data from a special survey carried out
in rural Bangladesh in 1998/99. The results are consistent with the view that women’s participation in
micro-credit programs helps to increase women’s empowerment. Credit program participation leads to
women taking a greater role in household decision making, having greater access to financial and
economic resources, having greater social networks, having greater bargaining power vis-à-vis their
husbands, and having greater freedom of mobility. Female credit also tended to increase spousal
communication in general about family planning and parenting concerns. The effects of male credit on
women’s empowerment were, at best, neutral and at worst, decidedly negative. Male credit had a
negative effect on several arenas of women’s empowerment, including physical mobility, access to
savings and economic resources, and power to manage some household transactions.
30
References
31
Pitt, Mark M., Shahidur R. Khandker, S-M. McKernan, and M. A. Latif. 1999 "Credit
Programs for the Poor and Reproductive Behavior in Low Income Countries: Are the Reported
Causal Relationships the Result of Heterogeneity Bias?" Demography. February 1999: 1-21.
Pitt, Mark M. 1997. "Estimating the Determinants of Child Health When Fertility and
Mortality are Selective.” Journal of Human Resources. Winter 1997: 127-158.
Pitt, M. and Shahidur R. Khandker. 1998. "The Impact of Group-Based Credit Programs
on Poor Households in Bangladesh: Does the Gender of the Participant Matter?" Journal of
Political Economy 106: 958-996.
Pitt, M., Khandker, S., Choudhury, O., Millimet, D. 2003. “Credit Programs for
the Poor and the Health Status of Children in Rural Bangladesh”, International Economic
Review, February 2003.
Rabe-Hesketh, Sophia and Andrew Pickles. 2000. “glamm6: A Stata program to fit
generalized linear latent and mixed models”, July 12, University of London.
32
Table 1. Number of households borrowing from credit programs, by gender of
borrower*
BRAC BRDB GB ASA PROSHIKA GSS Youth Dev. Other NGO
Male 16 54 121 4 9 2 0 35
* In some households, both men and women borrowed. Also, some women borrowed from more than one program
Table 2. Legend for full text and coding of individual empowerment variables
Name of variable Full text from questionnaire Coding* Asked of:
Food purchase Do you buy the family's daily consumable Y=1, N=0 Wife
food items?
Cosmetics purchase Do you buy toiletries and cosmetics for your Y=1, N=0 Wife
own use?
Candy purchase Do you buy ice-creams, candies, or cookies Y=1, N=0 Wife
for your children?
Utensils purchase Do you buy utensils, pots and pans for the Y=1, N=0 Wife
household?
Furniture purchase Do you buy household furniture? Y=1, N=0 Wife
Children's clothing purchase Do you buy clothing for your children? Y=1, N=0 Wife
Own clothing purchase Do you buy clothing for yourself? Y=1, N=0 Wife
Wife initiates discussion (birth Do you initiate discussion of birth control Y=1, N=0 Wife
control methods) methods?
Husband initiates discussion Does your husband initiate discussion of birth Y=1, N=0 Wife
(birth control methods) control methods?
Wife initiates discussion (birth Do you initiate discussion of birth control use? Y=1, N=0 Wife
control use)
Husband initiates discussion Does your husband initiate discussion of birth Y=1, N=0 Wife
(birth control use) control use?
Wife initiates discussion (kids’ Do you initiate discussion of son’s or Y=1, N=0 Wife
marriage) daughter’s marriage?
Husband initiates discussion Does your husband initiate discussion of son’s Y=1, N=0 Wife
(kids’ marriage) or daughter’s marriage?
33
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Wife initiates discussion Do you initiate discussion of children's Y=1, N=0 Wife
(children's education) education?
Husband initiates discussion Does your husband initiate discussion of Y=1, N=0 Wife
(children's education) children's education?
Wife initiates discussion (birth Do you initiate discussion of birth timing? Y=1, N=0 Wife
timing)
Husband initiates discussion Does your husband initiate discussion of birth Y=1, N=0 Wife
(birth timing) timing?
Wife initiates discussion (birth Do you initiate discussion of birth numbers? Y=1, N=0 Wife
numbers)
Husband initiates discussion Does your husband initiate discussion of birth Y=1, N=0 Wife
(birth numbers) numbers?
House repair decision Who decides issues of repair/construction of Husband Wife
the house? alone=0,
Husband and
wife together=1,
Wife alone=2
House repair implementation Who implements issues of repair/ construction Wife
of the house?
House repair spending Do you spend on repair/construction of the Y=1, N=0 Wife
house?
Livestock purchase decision Who decides issues of sale/purchase of Husband Wife
livestock? alone=0,
Husband and
wife together=1,
Wife alone=2
Livestock purchase Who implements issues of sale/purchase of Wife
implementation livestock?
Livestock spending Do you spend on sale/purchase of livestock? Y=1, N=0 Wife
Household loans decision Who decides issues of borrowing money? Husband Wife
alone=0,
Husband and
wife together=1,
Wife alone=2
34
Household loans implementation Who implements issues of borrowing money? Wife
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Household loans spending Do you spend on issues of borrowing money? Y=1, N=0 Wife
Wife has independent savings Do you have your own savings? Y=1, N=0 Wife
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Wife has independent savings Do you have your own savings which you can Y=1, N=0 Wife
which she herself controls decide how to utilize?
Emergency funds access If you needed 500 taka in an emergency, Y=1, N=0 Wife
could you get it (from any source)?
Emergency funds access (asset If you needed 500 taka in an emergency, Y=1, N=0 Wife
sale) could you get it by selling own assets?
Emergency funds access (from If you needed 500 taka in an emergency, Y=1, N=0 Wife
husband) could you get it from your husband?
Emergency funds access If you needed 500 taka in an emergency, Y=1, N=0 Wife
(husband's relatives) could you get it by borrowing from your
husband's relatives?
Emergency funds access (own If you needed 500 taka in an emergency, Y=1, N=0 Wife
relatives) could you get it by borrowing from your own
relatives?
Emergency funds access If you needed 500 taka in an emergency, Y=1, N=0 Wife
(moneylenders) could you get it by borrowing from
moneylenders?
Emergency funds access (other If you needed 500 taka in an emergency, Y=1, N=0 Wife
people) could you get it by borrowing from other
people?
Wife's control over loans If you have income generating loans in your Husband Wife
name, who has control over that? alone=0, Wife &
husband (or
another male)
together=1, Wife
& another
female=2, Wife
alone=3
Remittance Have you received money from Y=1, N=0 Wife
parents/brothers/sisters or other relatives
outside the household in the last 12 months?
Wife can decide how to use Can you decide yourself how to use that N=0, Partially=1, Wife
remittance remittance? Y=2
36
Money seizure by husband Has your husband ever compelled you to give Y=0, N=1 Wife
him money/asset if you don't want to?
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Freedom to remit Can you give away your money/asset at will Y=1, N=0 Wife
to somebody?
Husband forbids work outside Has your husband ever forced you not to Y=0, N=1 Wife
home work outside home even if you want to?
Visits relatives (without Have you ever visited your parents or other Y=1, N=0 Wife
husband’s permission) relatives without your husband’s permission?
Marriage has kabinnama Does your marriage have any kabinnama Y=1, N=0 Wife
(prenuptial bride price agreement)?
Awareness of kabinnama Can kabinnama help a woman in the event of Y=1, N=0 Wife
a divorce?
Awareness of inheritance laws Can a widow establish her legal claim over Y=1, N=0 Wife
her dead husband’s property?
Has prevented husband Have you ever been successful in stopping Y=1, N=0 Wife
remarrying your husband from remarrying?
Voted (at all) Did you vote in the last election? Y=1, N=0 Wife
Voted independently Did you vote in the last election without your Y=1, N=0 Wife
husband telling you who to vote for?
Protested against domestic abuse Did you ever protest against any incidents of Y=1, N=0 Wife
wife-beating?
Thinks dowry is good Do you think dowry is good? Y=0, N=1 Wife
Protested against corruption Did you ever protest against any favoritism by Y=1, N=0 Wife
a chairman or a member who distributes
government relief?
Confidant within bari With anybody outside your immediate Y=1, N=0 Wife
family/household, but within your bari, are you
close enough to share your feelings?
Interval of contact within bari With anybody outside your immediate Monthly=0, Wife
family/household, but within your bari, how Weekly=1,
often do you interact with this person? Daily=2
Confidant outside bari With anybody outside your bari, are you close Y=1, N=0 Wife
enough to share your feelings?
Interval of contact outside bari With anybody outside your bari, how often do Monthly=0, Wife
you interact with this person? Weekly=1,
Daily=2
37
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Severity of spousal arguments When you and your husband argue, how bad Physical Wife
does the argument get? abuse=0, Verbal
abuse=1, Loud
arguments=2,
Mild
Arguments=3
Occurrence of physical spousal When you and your husband argue, does Y=0, N=1 Wife
abuse physical abuse occur?
Own relatives in same village Do your parents or any sibling live in the same Y=1, N=0 Wife
village as you do with your husband?
Wife thinks husband is superior Is your husband superior to you in qualities Y=0, N=1 Wife
and education?
Husband uses male birth control Do you yourself use any male birth control Y=1, N=0 Husband
method?
Husband says women’s Does women's empowerment lead to a better Y=1, N=0 Husband
empowerment leads to better society?
society
Husband says women’s Does women's empowerment lead to chaos in Y=0, N=1 Husband
empowerment leads to chaos in society?
society
Husband says women’s Does women's empowerment lead to Y=0, N=1 Husband
empowerment leads to problems problems bringing up the children?
with kids
Husband says women’s Does women's empowerment lead to loss of Y=0, N=1 Husband
empowerment leads to loss of family peace?
peace
Husband says women’s Does women's empowerment lead to the Y=1, N=0 Husband
empowerment leads to better family being better off economically?
economically
Husband cites positive impact of Does women's empowerment have a good Y=1, N=0 Husband
women’s empowerment impact?
Husband cites negative impact of Does women's empowerment have a bad Y=0, N=1 Husband
women’s empowerment impact?
38
Husband’s assessment of What is your general assessment of women's Only negative=0, Husband
women’s empowerment empowerment Mixed=1, Only
positive=2
Husband views lack of education Is lack of education an obstacle to women's Y=1, N=0 Husband
as obstacle empowerment?
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Husband views lack of safety as Is lack of safety an obstacle to women's Y=1, N=0 Husband
obstacle empowerment?
Husband views lack of IGA as Is lack of Income Generating Activities an Y=1, N=0 Husband
obstacle obstacle to women's empowerment?
Husband views social structure Is the social structure an obstacle to women's Y=1, N=0 Husband
as obstacle empowerment?
Husband views law as obstacle Is inheritance law an obstacle to women's Y=1, N=0 Husband
empowerment?
Wife has made husband use birth Have you ever succeeded in making your Y=1, N=0 Wife
control husband adopt a male birth-control method?
Wife has Income Generating Do you have any Income Generating Activity? Y=1, N=0 Wife
Activity
Wife has Income Generating Do you have any Income Generating Activity Y=1, N=0 Wife
Activity which she herself which you yourself operate?
operates
Degree of mobility How do you go to banks, markets, health Doesn't go at Wife
centers or places outside the village (except all=0, Goes with
for your parent's place)? husband or
son=1, Goes
with women=2,
Goes alone=3
Wife ever travels Do you ever go to these places at all? Y=1, N=0 Wife
Wife ever travels alone Do you ever go to these places alone? Y=1, N=0 Wife
Prevent remarriage (threaten How can a wife prevent her husband from Y=1, N=0 Wife
divorce) remarrying…by threatening divorce?
Prevent remarriage (family How can a wife prevent her husband from Y=1, N=0 Wife
pressure) remarrying…by creating family pressure?
Prevent remarriage (local govt) How can a wife prevent her husband from Y=1, N=0 Wife
remarrying…by pressing change in the local
administration?
39
Prevent remarriage (parishad) How can a wife prevent her husband from Y=1, N=0 Wife
remarrying… by pressing change in the Union
Parishad?
Prevent remarriage (deny How can a wife prevent her husband from Y=1, N=0 Wife
permission) remarrying…By not giving permission?
40
Table 2. Legend for full text and coding of individual empowerment variables (continued)
Name of variable Full text from questionnaire Coding* Asked of:
Household fights about kids Does your household argue about the Y=0, N=1 Wife
children?
Household fights about money Does your household argue about money? Y=0, N=1 Wife
Household fights about in-laws Does your household argue about your in- Y=0, N=1 Wife
laws?
Household fights about going Does your household argue about going Y=0, N=1 Wife
outside outside?
Household fights about loans Does your household argue about loans? Y=0, N=1 Wife
Household fights about chores Does your household argue about household Y=0, N=1 Wife
chores?
Wife views lack of ed as obstace Is lack of education an obstacle to women's Y=1, N=0 Wife
empowerment?
Wife views lack of safety as Is lack of safety an obstacle to women's Y=1, N=0 Wife
obstacle empowerment?
Wife views lack of jobs as Is lack of jobs an obstacle to women's Y=1, N=0 Wife
obstacle empowerment?
Wife views social structure as Is the social structure an obstacle to women's Y=1, N=0 Wife
obstacle empowerment?
Wife views laws as obstacle Is inheritance law an obstacle to women's Y=1, N=0 Wife
empowerment?
Wife views religion as obstacle Is religion an obstacle to women's Y=1, N=0 Wife
empowerment?
*Most variables are coded with 0=No and 1=Yes. Those variables which are coded differently are shaded.
41
Table 3. Factor analysis compone nts and statistics
42
Table 3. Factor analysis components and statistics (continued)
43
Table 3. Factor analysis components and statistics (continued)
44
Table 3. Factor analysis components and statistics (continued)
45
Table 3. Factor analysis components and statistics (continued)
46
Table 4. Wu Tests for Exogeneity of Female and Male Credit Variables
Factor Test for female Test for male Test for joint Appropriate model
credit credit exogeneity
exogeneity exogeneity
Purchasing t = -0.210 t = -0.480 F =0.350 Exogenous female credit
P-val = 0.835 P-val = 0.632 P-val = 0.7016 Exogenous male credit
Resources t = 0.600 t = -2.250 F =2.760 Exogenous female credit
P-val = 0.552 P-val = 0.024 P-val = 0.0634 Endogenous male credit
Finance t = -1.270 t = -1.390 F =3.930 Endogenous female credit
P-val = 0.205 P-val = 0.164 P-val = 0.0198 Endogenous male credit
Transaction t = -1.260 t = -1.450 F =4.800 Endogenous female credit
P-val = 0.207 P-val = 0.147 P-val = 0.0083 Endogenous male credit
Mobility and t = 0.760 t = -0.790 F =0.390 Exogenous female credit
networks P-val = 0.446 P-val = 0.431 P-val = 0.6783 Exogenous male credit
Activism t = 0.870 t = -0.180 F =0.420 Exogenous female credit
P-val = 0.382 P-val = 0.859 P-val = 0.6569 Exogenous male credit
Household t = -0.940 t = -0.580 F =1.000 Exogenous female credit
attitudes P-val = 0.346 P-val = 0.559 P-val = 0.3686 Exogenous male credit
Husband's t = -1.030 t = -0.030 F =0.580 Exogenous female credit
behavior P-val = 0.302 P-val = 0.975 P-val = 0.5604 Exogenous male credit
Fertility and t = 0.590 t = 1.870 F =3.300 Endogenous female credit
parenting P-val = 0.553 P-val = 0.062 P-val = 0.0371 Endogenous male credit
All Variables t = 0.270 t = -0.820 F =0.330 Exogenous female credit
P-val = 0.791 P-val = 0.415 P-val = 0.7160 Exogenous male credit
47
Male credit -.0177 -.0301 .0163
(-1.74) (-3.25) (0.69)
48
-.0308 -.0446 -
(-1.78) (-2.99)
49
Husband initiates Female credit .14298343 (6.100405) .1306935 (5.7529255) a
discussion (children's Male credit -.01911873(-1.3296468) .05740227 (.71737438) a
education)
Wife initiates Female credit .01643459 (.83961744) .00608107 (.06797522) c
discussion (birth Male credit .00820515 (.68634246) -.02811081(-.31096221) c
timing)
Husband initiates Female credit .03155623 (1.6648596) .04205182 (2.2494966) a
discussion (birth Male credit .02181647 (1.8607394) -.00796965(-.12990483) a
timing)
Wife initiates Female credit .08293512 (4.0749984) .08064387 (4.0464672) a
discussion (birth Male credit -.0060319 (-.49479264) -.00941715(-.15586521) a
numbers)
Table 6. Male and female credit effects on individual empowerment questions (continued)
Name of variable Female/male Exogenous credit model Appropriate model
credit
Husband initiates Female credit .08366957 (4.298626) .0851692 (4.4497355) a
discussion (birth Male credit .01287064 (1.0768375) .06302056 (1.076337) a
numbers)
House repair decision Female credit .08907876 (4.2620781) .
Male credit -.00975032(-.78694676) .
House repair Female credit .07207265 (3.3856953) .05257216 (2.5858211) a
implementation Male credit -.05098127(-4.2001659) -.14109895(-2.0796276) a
50
Land/equipment Female credit .09766105 (5.1815832) .24828216 (2.8976129) c
implementation Male credit -.03660103(-3.3099373) -.22945787(-2.7228374) c
Table 6. Male and female credit effects on individual empowerment questions (continued)
Name of variable Female/male Exogenous credit model Appropriate model
credit
Husband says wife Female credit .12570957 (5.7274842) .
travels alone Male credit -.01651619(-1.3198174) .
51
access (own relatives) Male credit -.02508447(-2.0300173) .
Emergency funds Female credit -.03163719(-.65054464) .
access (moneylenders) Male credit .01068736 (.36908294) .
Emergency funds Female credit .08425008 (2.4472459) .
access (other people) Male credit -.00779705(-.37519001) .
Wife’s control over Female credit -.01703254 (-.60133783) -.01808199 (-.63718212) a
loans Male credit -.0205718 (-1.1283077) -.04399091 (-.61355432) a
Table 6. Male and female credit effects on individual empowerment questions (continued)
Name of variable Female/male Exogenous credit model Appropriate model
credit
Visits relatives Female credit .01911277 (.91810268) .
(without husband's Male credit -.04657823(-3.3794914) .
permission)
Marriage has Female credit .02275705 (1.1305345) .
kabinnama Male credit -.00044701(-.03591271) .
52
Voted (at all) Female credit .13131678 (5.1633827) .
Male credit -.00194159(-.12746871) .
Voted independently Female credit .0414437 (2.0471695) .
Male credit -.02997352(-2.3834331) .
Protested against Female credit .03678351 (1.8633071) .
domestic abuse Male credit .01145688 (1.027746) .
Table 6. Male and female credit effects on individual empowerment questions (continued)
Name of variable Female/male Exogenous credit model Appropriate model
credit
Wife thinks husband is Female credit .1450791 (4.872342) .
superior Male credit -.01446311(-.87795812) .
53
Reason = lack of Female credit -.06532019(-2.328394) .
safety Male credit -.00085579(-.05026752) .
Table 6. Male and female credit effects on individual empowerment questions (continued)
Name of variable Female/male Exogenous credit model Appropriate model
credit
Husband views lack of Female credit .06036897 (2.8999283) .07183463 (3.5440582) a
IGA as obstacle Male credit .02357819 (2.0845775) .02536636 (.39209182) a
54
structure as obstacle Male credit .00915417 (.76310031) -.05232476(-.86429722) a
Table 6. Male and female credit effects on individual empowerment questions (continued)
55
Name of variable Female/male Exogenous credit model Appropriate model
credit
Household fights Female credit .1881154 (3.5742584) .
about going outside Male credit -.02962608(-1.0165987) .
56