1
Master Thesis Economics
The Effect of Cash Transfers on Female Empowerment, Evidence
from a Social Security Programme in Pakistan
Amsterdam, 2017
Author: Emma Saskia Scarf Uva id: 110880265
Thesis Supervisor: Prof. Erik Plug Academic year: 2016-2017
2
STATEMENT OF ORIGINALITY
This document is written by Student Emma Saskia Scarf, who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
3
Abstract
The Benazir Income Support Programme offers direct cash transfers to women in households living in extreme poverty in Pakistan. Using cross-sectional data from the Pakistan Social and Living Standards Measurement Survey 2013-2014, this paper will estimate the
relationship between the transfer and any female empowerment benefits. Empowerment can be detected through decision making power over education, employment, fertility and consumption. The paper will use a Regression Discontinuity Design and Propensity Score Matching to estimate the impact of the programme. The findings suggest potential gains in consumption decision making power, however due to issues concerning the estimation method precise conclusions should not be drawn from the results.
4
Contents
1. Introduction……….……5
2. Literature Review………...7
3. Programme description and context………...9
4. Data and Empirical Approach…………...…...…..……..………13
4.1 Regression Discontinuity………...………...……16
4.2 Propensity Score Matching………...20
5. Main Results………..23
6. Discussion and Conclusion……….…………..25
7. References………..………27
5
1. Introduction
This study analyses unconditional cash transfers as a tool for women’s empowerment and the socio-economic factors associated with women's intra-household bargaining power. Cash transfers have been increasingly utilized in developing countries to alleviate the most extreme forms of poverty. Direct cash transfers like Progresa in Mexico and Red de Protección
Social in Nicaragua have successfully relieved resource constraints in households and promoted investment in health and human capital. Additionally, an unintentional impact of such programmes was the empowering effect that came from bestowing women with an income. As a result, it is now the accepted norm amongst the majority of cash transfer programmes to grant them directly to the woman of the household, with increased gender equality becoming a consolidated goal within cash transfers.
The Benazir Income Support Programme is the largest social safety net in Pakistan. It provides a direct unconditional cash transfer to women in households living in extreme poverty. Studies of other programmes where cash transfers were given directly to female members of a household found that it significantly increased a woman's influence in household decision making (Schultz 1990, Thomas 1990, Gitter and Barham 2008). By endowing women with income, it improves their bargaining power and influence within the household, and grants them greater agency over their lives. Transfers have targeted women in the past not only in the hopes of empowering them, but in the belief that women were more likely to invest in human development goals. Preceding literature has found that when women have control over resources they promote larger gains in health, education and the nutritional outcomes of children then when men have full responsibility. These findings have been established in research such as Thomas (1990), Haddad (1999) and Duflo (2003). This study will attempt to find out if this is also the case in Pakistan and whether having access to income affects women in terms of household decisions, hierarchy and consumption patterns.
6 Thus, the research question for this thesis can be defined as “What are, if any, the effects of the Benazir Income Support Programme on Female Empowerment as measured by
bargaining power”.
Cross sectional data from the 2013-14 Pakistan Social and Living Standards Measurement survey is used to provide estimates for the effect of the benefit on women’s decision making power over education, employment, fertility and consumption.
By taking advantage of the discontinuity in eligibility created by the cut-off score from proxy means testing, a regression discontinuity design was applied. The main findings from this estimation approach suggest a potential positive effect from the benefit on women’s decision making power over consumption decisions and employment, however the results for
education and fertility were inconclusive. Due to partial compliance within the programme a second estimation method, propensity score matching, was also utilised. The estimates from this approach were also suggestive of potential gains in bargaining power over consumption decisions.
The remainder of this thesis is organized as follows. Section 2 provides a review of the current literature regarding cash transfers and female empowerment, followed by a description of the Benazir Income Support Programme and its context within Pakistan in Section 3. Subsequently, Section 4 presents a description of the Pakistan Social and Living Standards Measurement dataset and the estimation methods, the main results are reported in Section 5. Whilst section 6 includes a succinct conclusion of the findings of this paper.
7
2. Literature Review
Previous studies have researched the impact that recipient gender has on cash transfer outcomes and intra-household bargaining power. The main premise for the empowering impact of cash transfers is that by giving women the benefit, you increase their bargaining power by bestowing them with more economic independence. Through improving women's bargaining power, their status within the household can be strengthened, (Soares and Silvas, 2010). Adato, Mindek and Quisumbing (2000) looked at the effect of the Mexican Progresa Conditional Cash Transfer on women's status in the family and intra-household relations. The Progresa programme supported conditional cash transfers, maternal and child health care and nutritional supplements. By making women the programme beneficiary on behalf of the family, Progresa also aims to empower them by increasing their control over resources, (Gomez de Leon and Parker 1999). The study by Adato et al, (2000) looked at how the Progresa programme impacted women's bargaining power over two important concerns, primarily the woman's role in decision making within the household, as well as children's schooling achievement. They found that the benefit led to an increase in the probability that the women had a strong influence in decision making within the household as well as changing gender biases within the community.
The second study by Gitter and Barham (2008), looked into the effect a cash transfer handed to women in Nicaragua had on intra-household relationships, education and child nutrition. They studied whether “power structures” within households shifted as a result of the transfer. Power within the household prior to treatment is proxied by relative education between spouses. Previous studies have found five main factors that can predict the balance of bargaining power within the household; relative education (Beegle, Frankenberg and Thomas, 2001), unearned income (Thomas, 1990 and 1994; Hoddinott and Haddad, 1995), length of marriage, age discrepancies and assets brought to marriage (Fafchamps and Quisumbing, 2002). Furthermore, women’s bargaining power in Gitter and Barhams (2008) study was measured through changing consumption patterns. It has long since been
acknowledged that women have different preferences than men. Thomas (1990) and, Fafchamps and Quisumbing (2002) found that women's preferences prioritised health and education of their children higher than that of men. Duflo (2003) also found evidence for these preferences in a South African pension scheme. When women received the pension
8 benefit, which can be viewed as an unconditional cash transfer, their grandchildren,
specifically granddaughters, received better nutrition. This improvement in child nutrition was not seen when pension recipients were men. Such redistributive effects were also discovered in Nicaragua where the Red de Protección Social programme saw female beneficiaries improve children’s educational attainment as well as nutrition. This effect suggests that women had viable bargaining power within the household. These interactions are determined by the individuals’ preferences, altruistic or egoistic, as well as by the mechanism households use for decision making. This may be through a bargaining process, variations of a benevolent dictator game, among other possibilities (Alderman et al, 1995). In this context, the receipt of a cash transfer and who it is awarded to might cause shifts in household dynamics (Abel, 2014).
Further evidence for the empowering effect of cash transfers was seen in other Latin American countries; Soares and Silva (2010) noted that in studies of programmes across Brazil, Chile and Colombia there was evidence for female recipients having increased
bargaining power over consumption decisions. Furthermore, as well as greater influence over economic decisions the Bolsa Familia, Familias en Acción and Chile Solidario programmes were found to have a positive impact on women’s “esteem, confidence and self-perception” (Soares and Silva 2010).
Cash transfers have also become increasingly popular within Asia, and with current plans to introduce a basic income across India to alleviate poverty, it is necessary to measure the impact such programmes have. A pilot programme conducted on 20 villages in Madhya Pradesh in 2014 used a difference-in-differences approach to measure the impact of an unconditional cash transfer awarded to women. The experiment, which ran for an entire year, was conducted by SEWA (2014); the Self-Employed Woman’s Association with support from UNICEF. The study found that female recipients of the basic income were more likely to have bargaining power over household decision making and income, as well as improving their labour force participation. These positive results were also mirrored in Bangladesh, Ahmed et al, (2009) found that the Rural Maintenance Programme and Food for Asset
Creation Programme in Bangladesh improved woman’s empowerment and gave them greater control over household resources. This thesis will add to the current literature available studying cash transfers in the Indian sub-continent.
9
3. Programme Description and Context
The Benazir Income Support Programme (BISP), was first introduced in 2008 after stagflation led to a food and energy crisis. As the poorest members of society spend the largest proportion of their income on these goods those in poverty saw a significant decrease in their purchasing power (Khan and Qutub, 2010). Thus, the government initiated the Benazir Income Support Programme to provide a social safety net to those worst affected. In the first stage of the programme from 2008-2010 beneficiaries were identified by
parliamentarians and households received Pk Rs. 1,000 per month (25 USD). Members of the National Assembly were given 8,000 application forms and were tasked on handing them out to households that fit the agreed upon criteria. Eligible households were those with a monthly income below a threshold of Pk Rs. 6,000, with a married female household member and dependents (Government of Pakistan, BISP 2015).
Due to the subjective nature of this process and the potential for political bias, the programme was overhauled in 2010. Eligibility of the programme has since been based on a national Poverty Score Card (PSC) survey, using proxy means testing, households were given a score from 0 to 100, and households under 16.71 were deemed “chronically poor” and were eligible for the benefit (Government of Pakistan, BISP 2015). Additionally, the basic income was increased from Pk Rs. 1,000 to Pk Rs. 1,500 per (30 USD). The proxy mean testing approach for the poverty score card in Pakistan was developed by the World Bank using the Pakistan Social and Living Standards Measurement survey (PSLM) data in 2007-2008 in order to capture the poorest 25% of the population. The survey covered 23 variables that predict poverty, for example education, room ratio, and land, with each characteristic given a specific weighting. The full list of characteristics used for the proxy means test can be found in the appendix in Table 4. To be eligible, as well as living in a household with a PSC count lower than 16.71, women had to be in possession of a valid identity card, with no commercial bank account, in a household without a permanent income or receiving any other financial
assistance or pension. The scorecard survey was slowly introduced across the country using a census approach, if a household was deemed eligible they were provided with a BISP
application form. The biggest impediment to women receiving the benefit was the essential requirement of holding a valid form of identification known as a CNIC number. During the
10 test phase of the programme over 300,000 eligible women in 16 districts were unable to collect the benefit due to their lack of CNIC.
During the first period of the programme from 2008-2012 the benefit was delivered to
women through the post office. This method however was prone to petty corruption and theft. Postmen were reported to withhold cash for bribes and demand kickbacks of up to half of the benefit in order to secure delivery. Furthermore, it was reported that very few postmen verified that the benefit was being handed over to the intended female recipient in the household (Khan and Qutub, 2010). In 2012, a new delivery mechanism was introduced whereby women received a BISP card which was loaded up with the quarterly allowance of PK Rs, 4,500 which could be dispensed through an ATM. This method has also been prone to a number of issues regarding corruption and accessibility. First and foremost, in rural areas there may be a significant distance to the nearest ATM which may make the benefit
inaccessible for many women. Secondly, women may be prohibited from travelling alone or leaving the house unaccompanied, and as a result are unable to collect the benefit. This may be a particular obstacle in rural areas which tend to be more conservative, hence women have less independence outside of the house. Lastly, recipients may be unable to operate an ATM, which may further hinder the process of receiving the benefit. One case of corruption
included a school master taking a cut of everyone’s benefit in return for helping them use the ATM (Cheema et al, 2014). Furthermore, reports of corruption by the Department for
International Development included several accounts of individuals paying kickbacks to officials to obtain a BISP card. Such conditions and cases of corruption may have contributed to partial compliance seen in the dataset.
To receive the benefit women must also fit one of the following categories; a woman living with her husband and unmarried children, divorced and living with unmarried children, divorced and living alone or with her parents/relatives, or living alone with her
parents/relatives. Whilst providing women with the benefit may be empowering, it has been suggested that the above conditions create moralistic judgement on who is deemed worthy of the benefit (Gubrium et al, 2013).Any unmarried woman living alone or with subsequently out of wedlock children would not be eligible for the benefit. Furthermore, there has been considerable controversy regarding the targeting of beneficiaries, it has been suggested that the unconditional cash payments are being used as bribes to safeguard a bank of political
11 votes for the Pakistan People Party. This was especially the case when parliamentary
members identified beneficiaries. “There are six or seven people I know that do not deserve (BISP money). They have become the beneficiaries based on political contacts. They have their own house and they earn themselves”. (Male Beneficiary focus group, District
Gujranwala, Punjab, Cheema et al, 2014). However, targeting appears to have improved since the introduction of the poverty score card system with 68% of BISP beneficiaries falling below the poverty line, (Cheema et al, 2014). The Programme in itself takes no measures to promote female empowerment other than the allocation of the transfer to women in the household, as a result any benefits must be reaped from strengthened decision making powers within the household as a result of greater economic independence.
Women’s empowerment is understood as to give power to women for having not only the access to the resources and opportunities but also the ability to utilize these resources and opportunities for their personal and social change (Sophie 2007). Female empowerment in this context will be defined as the ability of women to make personal decisions regarding her needs and independently choose the allocation of resources and income within the household. The success of the programme in empowering women at the household level is highly
dependent on whether receiving the transfer gives the recipient increased influence in
household decision making. Whilst the Benazir Income Support Programme attempts to give women more bargaining power in the household, it provides no support other than stipulating whom the transfer is awarded to. Therefore strong patriarchal social dynamics within the household may override any potential benefit regarding the gender of the beneficiary. However encouragingly Cheema et al, (2014) found that 64% of the female beneficiaries in their survey got full access and control over the funds in their household. Therefore it is possible to assess how having control of these funds effected female empowerment.
Female empowerment is of particular interest in Pakistan which is a typically patriarchal society, women are often restricted by traditional Islamic social and cultural norms. In fact, Pakistan ranks 125 out of 169 countries on the United Nations Development Programme Gender Inequality Index which is based on health, employment and
empowerment outcomes UNDP (2010). The maternal mortality rate of 320 per 100,000 live births and adolescent fertility rate of 45.7 per 1000 births are still considered high in the developing world (World Bank 2015). Furthermore, participation of women in labour market is still rare with only 21.5 % of women in the labour market, compared to 68.8% for men,
12 cultivating the economic disempowerment of women (World Bank 2015). Girls are also consistently less likely to be enrolled in schools than their male counterparts. Primary Net Enrolment Rates find that only 53% of girls are enrolled compared to 60% of boys
(UNESCO 2014). The discrimination in enrolment of girls in primary schools, low status of reproductive health and low female participation in the labour market are strong indicators for gender inequality. As a result, women have fewer opportunities for their wellbeing, within the household and in social and economic life of the country. It is customary in Pakistan for woman to have little to no say in decision-making at the household and community level, furthermore they can be socially penalized when independently taking part in public activities or representative bodies (Ali, Fani and Saima, 2010).
The UN (1995) defines female empowerment as women’s “right to have and to determine choices; their right to have access to opportunities and resources; their right to have the power to control their own lives, both within and outside the home”. To this end female empowerment in this paper will be proxied through perceived decision making power over fertility, education, employment, health and consumption over food, clothing and medicine. In the PSLM HIES (2013-2014) dataset women were asked who made crucial decisions in the household which impacted the woman herself. The questions were as follows “Who in your household decides whether you can seek or remain in paid employment?”, “Who in your family decides whether you can use birth control methods?”, “Who in your household
decides whether you can start or continue to get education?” and “Who in your household usually makes decisions about purchase of following consumption items? Clothing, Food and Medical treatment”. These questions were then used to create binary variables depending on whether the woman had influence in the decision-making process. The possible answers were Woman herself, Spouse/Father of the household decides alone, Spouse/Father in consultation with the woman concerned or Head/Father and other male members decide. The outcome binary variable was equal to 1 if the woman had decision making power herself or when in consultation with her spouse and 0 if the woman had no influence in the decision-making process.
One notable issue with the data is that the programme shows imperfect compliance when it comes to how the benefit was distributed. The programme is meant to identify beneficiaries based on their poverty score card measure which was calculated using a proxy means test.
13 However, the data shows a large proportion of eligible women did not receive the benefit and a high number of non-eligible individuals did. This could be due to several reasons. Firstly prior to identifying beneficiaries through proxy means testing local governments were in charge of distributing the benefit. As a result, there could have been different criteria for the allocation of the benefit which may have resulted in non-eligible women receiving the transfer. Furthermore, only women with a valid ID card and not receiving other government handouts or transfers could apply, information that was unreported in the dataset. This may explain why only 24% of the eligible women were found to be receiving the benefit, namely 518 of a total of 2,168 below the necessary threshold. This shows a considerable amount of leakage in the programme which can cause issues with the estimation method.
4. Data and Empirical Approach
The data for this study comes from the Pakistan Social and Living Standards Measurement, Household Integrated Economic Survey 2013-2014 (PSLM). The household survey is carried out annually switching between the district and provincial level, it includes extensive
coverage of household characteristics and activities. The survey contains cross-sectional data from 17,989 households and was conducted by the Federal Bureau of Statistic in Pakistan. The questionnaire includes information for each household member over education, consumption, income as well as identifying households who received the BISP transfers which amounted to 1,172 beneficiaries in the sample. Transfer information was asked to male heads of household despite women being the main intended recipient of the transfer, however it is possible to match household heads with their spouse who received the benefit. The female section of the survey asks several self-reported questions over decision making power, which will be used as a proxy for female empowerment and is the outcome variable of
14 Table 1 provides the descriptive statistics for the sample of women studied from the PSLM dataset for the regression discontinuity design as well as the propensity score matching sample. Whilst there is significant difference between treated and non-treated individuals this is to be expected due to the nature of the benefit. Many of these characteristics were used to identify beneficiaries, and as a result should have different sample means across groups. The large difference between PSC sample means across the two groups also shows that the
method used to identify beneficiaries was somewhat successful, with those receiving the cash transfer having a significantly lower PSC score. Characteristics that are thought to define pre-existing decision making power within the households will also be controlled for. Preceding literature has defined these as; relative education between a couple, the length of the
marriage, as well as any age difference between a woman and her spouse (Fafchamps and Quisumbing, 2002). The differences in sample means will also be accounted for by
controlling for PSC which includes all the above variables when calculating the proxy means test score.
During the propensity score matching approach observations were split into 12 comparable groups based on these observational characteristics. The groups were balanced, which states that the mean values for the variables used to calculate the propensity score are the same across matching intervals and within each group. It is important for means to similar across matching groups in order to verify the common support condition it satisfied. The condition states that if observations have similar observable characteristics and thus propensity score it must be that their probability of receiving treatment must also be the same and positive.
15
Table 1. Sample means of key variables on the household level
Regression Discontinuity Propensity Score Matching Treatment Control Treatment Control
Variable Mean N Mean N Mean N Mean N
Women Characteristics Age 35.57 (0.213) 1,172 33.83 (0.074) 10,654 35.55 (0.222) 1,095 34.00 (0.082) 9,195 Education 0.11 (0.012) 1,172 0.67 (0.010) 10,654 0.11 (0.012) 1,095 0.49 (0.009) 9,195 No Education 0.92 (0.008) 1,172 0.65 (0.005) 10,654 0.97 (0.005) 1,095 0.83 (0.004) 9,195 Live Births 5.60 (0.008) 1,172 4.16 (0.023) 10,654 5.62 (0.082) 1,095 4.35 (0.025) 9,195 Employment 0.43 (0.014) 1,172 0.27 (0.004) 10,654 0.43 (0.015) 1,095 0.28 (0.005) 9,195 Age Married 19.00 (0.094) 1,172 19.77 (0.032) 10,654 19.00 (0.098) 1,095 19.52 (0.034) 9,195 Age Difference Education Difference 4.80 (0.150) 0.51 (0.026) 1,172 1,172 5.08 (0.048) 0.51 (0.011) 10,654 10,654 4.81 (0.155) 0.53 (0.027) 1,095 1,095 5.03 (0.052) 0.58 (0.012) 9,195 9,195 Household Characteristics Region 0.14 (0.010) 1,172 0.37 (0.004) 10,654 0.144 (0.011) 1,095 0.33 (0.005) 9,195 Household Size 8.25 (0.104) 1,172 7.19 (0.031) 10,654 8.35 (0.110) 1,095 7.39 (0.034) 9,195 PSC 19.14 (0.292) 1,172 33.45 (0.166) 10,654 19.44 (0.303) 1,095 30.34 (0.142) 9,195 Landowner Agriculture 0.02 (0.004) 0.38 (0.014) 1,172 1,172 0.06 (0.002) 0.30 (0.004) 10,654 10,654 0.02 (0.005) 0.38 (0.015) 1,095 1,095 0.05 (0.002) 0.31 (0.005) 9,195 9,195 Car 0.00 (0.001) 1,172 0.06 (0.002) 10,654 0.00 (0.001) 1,095 0.02 (0.001) 9,195 Stove 0.14 (0.011) 1,172 0.44 (0.005) 10,654 0.14 (0.011) 1,095 0.38 (0.005) 9,195 Latrine 0.37 (0.014) 1,172 0.67 (0.005) 10,654 0.39 (0.015) 1,095 0.65 (0.005) 9,195 Electricity 0.84 (0.010) 1,172 0.90 (0.003) 10,654 0.86 (0.010) 1,095 0.91 (0.003) 9,195 Female Empowerment Contraceptives 0.59 (0.014) 1,172 0.68 (0.004) 10,654 0.59 (0.015) 1,095 0.67 (0.005) 9,195 Children 0.57 (0.013) 1,172 0.65 (0.004) 10,654 0.58 (0.013) 1,095 0.64 (0.005) 9,195 Clothing 0.73 (0.013) 1,172 0.55 (0.005) 10,654 0.73 (0.013) 1,095 0.56 (0.005) 9,195 Medicine 0.78 (0.012) 1,172 0.70 (0.004) 10,654 0.77 (0.013) 1,095 0.70 (0.005) 9,195 Education 0.05 (0.006) 1,172 0.15 (0.003) 10,654 0.05 (0.006) 1,095 0.13 (0.004) 9,195 Food 0.77 (0.012) 1,172 0.59 (0.005) 10,654 0.77 (0.013) 1,095 0.60 (0.005) 9,195 Employment 0.84 (0.010) 1,172 0.78 (0.004) 10,654 0.46 (0.007) 1,095 0.46 (0.003) 9,195
Note: Education is measured from 0-3 denoting no education, primary, middle school and higher education
respectively. PSC is the poverty score card measured from 1-100 and Age gap denotes the discrepancy in ages between marital couples.
16 Nevertheless, this cannot guarantee that women in the treatment group don’t have unobserved characteristics that could be driving the results. Especially due to the high level of leakage in the programmes distribution, the cut off appears to have minimal explanatory power over how the programme was allocated as a result there may be a common underlying trait which explains how the cash benefit was handed out.
In the appendix, there is additional preliminary graphical analysis for the outcomes of interests; the indicators for female bargaining power. Figures 3a, 3b onwards show the outcomes of interest by the running variable; PSC score. The graphs can help identify if there is any discontinuity near the cut off value which could suggest a potential treatment effect. The average values for the female empowerment variables for each PSC score are smoothed by a fourth-degree polynomial. When interpolating the graphs however there doesn’t appear to be any initial noticeable discontinuity near the threshold.
4.1 Regression Discontinuity
Due to the nature of the cross-sectional data it was not possible to observe households before the benefit was introduced. However, we can observe the relation between the Benazir
Income Support Programme and certain empowerment indicators. This can be done by taking advantage of the eligibility cut off at score at 16.71. This discontinuity in the probability of being eligible for the benefit can help identify if there was any corresponding jump in the outcome of interest at the cut off. Due to partial compliance when distributing the benefit some eligible families did not receive BISP whilst some non-eligible families did. As a result, the relation between eligibility and receiving the benefit is probabilistic instead of a distinct cut off which will create a fuzzy regression discontinuity. In a Fuzzy RD design, treatment is instrumented by a binary indicator for eligibility status. A two stage least squared approach can then be used to estimate the effect of the BISP transfer on female empowerment as a function of the running variable PSC.
17 One identifying assumption is that conditional on a function of PSC and other background characteristics, eligibility is correlated with the outcomes of interest only through treatment. As the eligibility threshold score is exogenously determined by the programme this should be the case. Furthermore, poverty score weighting was calculated post interview so subjects would have been unable to manipulate their programme eligibility status.
The first stage models the BISP transfer as a function of the poverty score PSC, while the second stage estimates the outcome the BISP transfer has on female empowerment.
Formally, the first and second stage equations are the following,
First Stage; 𝑇𝑇𝑖𝑖 = 𝛼𝛼0+ 𝛼𝛼1𝑍𝑍𝑖𝑖+ 𝜌𝜌1𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖+ 𝜋𝜋1𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑍𝑍𝑖𝑖 + 𝛼𝛼2𝑋𝑋𝑖𝑖 + 𝑣𝑣𝑖𝑖 (1) Second stage; 𝑌𝑌𝑖𝑖 = 𝛽𝛽0+ 𝛽𝛽1𝑇𝑇𝑖𝑖+ 𝛾𝛾1𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖 + 𝛿𝛿1𝑃𝑃𝑃𝑃𝑃𝑃𝑖𝑖𝑇𝑇𝑖𝑖 + 𝛽𝛽2𝑋𝑋𝑖𝑖 + 𝜀𝜀𝑖𝑖 (2)
Treatment (𝑇𝑇𝑖𝑖) in equation (1) is predicted by the instrumental variable (Z𝑖𝑖), a flexible function of normalized PSC and additional interaction terms. (Z𝑖𝑖) is a binary variable for eligibility, it takes the value of 1 when a woman is eligible for the transfer, which occurs when her PSC score is below 16.71. PSC is the normalized poverty score for the woman, which represents the difference between an individual’s score and the eligibility threshold. Including a function of normalized PSC and interactions is necessary if the relationship between the running variable, namely PSC and the endogenous variable is nonlinear. The terms 𝑣𝑣𝑖𝑖 and 𝜀𝜀𝑖𝑖 account for any econometric errors which may be present.
𝑌𝑌 in equation (2) is the outcome of interest for woman i and 𝛽𝛽1 is the causal effect from the BISP transfer. The variable 𝑋𝑋𝑖𝑖 is a vector of background characteristics such as age, the region the woman is from; either rural or urban, the age difference between the woman and her spouse as well as the difference in education between the woman and her spouse.
18 Treatment (𝑇𝑇𝑖𝑖) is an endogenous binary variable which is equal to 1 if the woman is a BISP recipient.
Due to the fact some women above the threshold received treatment and many below did not the estimates include never-takers and always-takes as a result the Fuzzy RD design does not reflect intention to treat or (ITT) estimates. Instead by using an instrumental variable
estimator the results show the average treatment effect (ATE) for compliers.
Figure 1. Probability Of treatment by distance from cut off value
To illustrate the fuzzy regression discontinuity design Figure 1 shows the relationship between having a PSC score below the necessary threshold and receiving the transfer. The circles on the graph depict the average number of women receiving the BISP transfer within a bin-width of five points of the PSC score, whilst the solid and dashed lines show fitted values from the first stage regression seen in equation (1) and differing polynomial functions of PSC. It is evident from the graph that the lower a woman's PSC score the higher the
-. 1 0 .1 .2 .3 -20 0 20 40 60 80 Local Average Normalized PSC
Linear Linear with Interaction
19 probability is of her receiving the BISP transfer, which is to be expected. However, what is less clear is the effect of the threshold on probability of treatment. There appears to be a very limited jump at the threshold in BISP allocation. The discontinuity in treatment at the
threshold is limited, the probability increases by 9 percentage points. Furthermore, this reduces as less restrictive functional forms of equation (1) are applied and the sample is confined around the threshold. This may result in estimation problems due to a weak instrument which could severely impact the ability of the empirical method to discern meaningful effects.
Table 2 shows the results from the first stage regressions, the point estimates for the linear form show an increase of treatment by 0.09 (Column 1, s.e 0.01), when an individual is eligible, this result is highly significant. The F-stat of 113.8 for the instrument used suggests that the cut-off score is a strong and relevant predictor for treatment. This finding remains fairly constant when applying background characteristics and including an interaction term between PSC score and eligibility status. The table also includes balancing tests for additional controls (columns 6-9) including age, age difference, region and education difference. For the most part these observable characteristics seem unrelated to eligibility status, however
education difference appears to be significant, this is likely due to the fact education level is a contributing factor to the PSC score and thus eligibility. However, the effect of eligibility on
Table 2. First stage estimates and balancing tests
BISP Transfer Balancing
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Age Age Diff Region Education Diff
Z 0.09*** (0.01) 0.09*** (0.01) 0.06*** (0.02) 0.02* (0.01) -0.00 (0.02) 2.14*** (0.21) 0.18 (0.15) -0.00 (0.01) -0.23*** (0.03) Controls Interaction terms Polynomial Order First First First Second Second
N 11826 11826 11826 11826 11826 11826 11826 11826 11826 P-value 0.000 0.000 0.000 0.035 0.797 0.000 0.212 0.725 0.000 F-Stat 113.848 102.554 23.255 4.431 0.063 90.317 1.557 0.123 52.514
R2 0.07 0.08 0.08 0.08 0.08 0.007 0.005 0.000 0.011
Note: standard errors specified are reported in brackets and *, **, *** indicate significance at a 10%, 5%, 1% confidence level,
respectively. Columns 1 - 5 present regressions of BISP Programme status on Z, controlling for normalized PSC either linearly (1-3), or quadratically (4-5), Columns (6-9) present separate regressions of the controls age, age and education difference with spouse and region on Z and normalized PSC.
20 treatment falls to insignificant levels when applying a quadratic function of PSC score and further interaction terms, which appears to be the best fit for the data. Furthermore, the F-stat is reduced to 4.43 in column 4, this is indicative of a weak instrument and may cast doubt on the estimation method. Nevertheless, continuing onto the second stage the estimation method will use the linear specification including interactions to allow for flexibility whilst remaining significant, suggesting discontinuity in allocation of treatment.
4.2 Propensity Score Matching
In order to counter some of the issues in the first estimation method a following empirical approach will be used to corroborate any results. The secondary method utilised will be propensity score matching, where you can construct comparable treatment and control groups based on observable characteristics. The comparable groups need to be as statistically similar as possible so that the only discrepancy between them is treatment, here being BISP benefits. Therefore, reproducing a counterfactual that describes the outcome had households not received the BISP cash transfer. Treated individuals are matched with a control group based on similar covariates which predict their probability of treatment. As a result, we can
compare the effect of receiving and not receiving the benefit between households that were eligible and thus have similar characteristics. Therefore, the results will display treatment effect on the treated, instead of intention to treat. Conditional independence needs to hold for the propensity score matching method, this assumes treatment was based solely on
observable characteristics and that all variables that influence treatment and outcomes simultaneously are observed. As a result, all the differences between the treatment and control group which may affect the outcome other than treatment can be controlled for (Heckman et al, 1998). Furthermore, the common support condition needs to be satisfied, it states that if observations have similar observable characteristics and thus propensity score it must be that their probability of receiving treatment must also be the same and positive. In addition, all treated individuals must have a comparable match in the control group, so that the average treatment effect on the treated can be detected within the common support region (Caliendo and Kopeinig, 2008).Propensity score matching eliminates selection bias so that the outcomes of control and treatment groups are independent of participation given the
21 probability of treatment. As a result, any differences between treatment and controls is the effect of treatment. Which then gives the average treatment effect of treated (ATT) (Caliendo and Kopeing, 2008).
The treatment group and the control group are identified as the women that received the treatment BISP denoted Di =1 for woman i in equation (3), and women who did not receive the BISP treatment Di =0 for woman i. Then a probit model is used to estimate the
probability of an observation to be assigned treatment, this is based on Xi, a vector of background covariates. Thus, the propensity score is the conditional probability of receiving treatment given pre-treatment characteristics (Katchova 2013).
𝑃𝑃�𝑋𝑋𝑖𝑖� = 𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃(𝐷𝐷
𝑖𝑖 = 1|𝑋𝑋𝑖𝑖) = 𝐸𝐸(𝐷𝐷𝑖𝑖|𝑋𝑋𝑖𝑖) (0 < 𝑃𝑃(𝑋𝑋𝑖𝑖) < 1 ) (3)
Table 4 in the appendix shows the full list of variables used to calculate propensity score. Variables used for the propensity score were also used to calculate the PSC score by the World Bank as well as other covariates used to match women in terms of human capital, demographics and social factors suggested in the use of propensity score matching by Caliendo and Kopeing, (2008).
Figures 2a and 2b show the propensity score across the treatment and control groups. The tables show that the observable characteristics used are a good predictor for BISP treatment status.
Figure 2a. Figure 2b.
Bisp Beneficiary 0 1 2 3 4 D en s it y 0 .2 .4 .6
Probability of Receiving Bisp
Non-Beneficiary 0 5 10 15 20 D en s it y 0 .2 .4 .6
22 Control and treatment observations are then matched by their propensity score and the
treatment effect can be calculated by comparing the outcomes y between the matched observations.
𝑦𝑦 = �𝑦𝑦𝑦𝑦1 𝑖𝑖𝑖𝑖 𝐷𝐷 = 1
0 𝑖𝑖𝑖𝑖 𝐷𝐷 = 0 (4)
𝐴𝐴𝑇𝑇𝑇𝑇 = 𝐸𝐸(𝑦𝑦1| 𝐷𝐷𝑖𝑖 = 1) − 𝐸𝐸(𝑦𝑦0| 𝐷𝐷𝑖𝑖 = 0) (5)
There are several different matching algorithms used to create treatment and controls. The first one used in this study is nearest neighbour matching where an observation with the closest propensity score is matched with the closest equivalent score in the non-treatment group. The second method is radius matching whereby five of the closest propensity scores in the non-treatment group are used within a range of 0.1 of a score, in order to avoid bad matches. Furthermore, replacement is allowed so that a nontreated observation can be used for more than one match of treatment. Lastly for robustness kernel matching is applied, where weighted averages from the observations in the control group are used to establish the
counterfactual outcomes and are compared with the treatment group. This method is used because by using the complete control group the variance is lowered, however as a result there is the risk that bad matches are included when creating the counterfactuals, (Smith 2000). As a result for this method it is imperative that the common support condition holds.
23
5. Main Results
The results in Table 3 show the estimates for the effect of the BISP transfer on certain female empowerment indicators. The regression results come from the linear specification with interaction, using eligibility status as an instrument and controlling for PSC and range of background characteristics. The female empowerment indicators are all binary variables based on decision making power, where the variable takes a value of 1 if women have decision making power and influence over certain aspects of their life and within the household.
Column 2 shows the results for the regression discontinuity approach; significant results can only be seen in decision making power over employment, food and clothing expenditure. The BISP cash transfer can be seen to increase the likelihood of decision making power by
women in food expenditures by 0.46 (column 2, s.e 0.099) and in clothing expenditure 0.28
Table 3. Main Results from Regression Discontinuity and Propensity Score Matching
OLS RD Nearest Neighbor
Matching
Radius Matching Kernel Matching Fertility -0.011 (0.015) -0.149 (0.098) 0.004 (0.026) 0.035** (0.016) -0.009 (0.015) Employment 0.034*** (0.013) 0.138* (0.084) 0.017 (0.019) 0.055*** (0.012) 0.044*** (0.013) Education -0.041*** (0.011) -0.008 (0.069) -0.044*** (0.013) -0.070*** (0.008) -0.051*** (0.007) Food 0.115*** (0.015) 0.464*** (0.099) 0.141*** (0.024) 0.158*** (0.014) 0.132*** (0.014) Clothing 0.099*** (0.015) 0.277** (0.098) 0.103*** (0.025) 0.152*** (0.014) 0.122*** (0.016) Medicine 0.051*** (0.014) 0.032 (0.092) 0.041* (0.025) 0.066*** (0.014) 0.048*** (0.014) Controls N 11826 11826 10290 10290 10290 P-value 0.000 0.000 0.797 0.000 0.035 F-Stat 32.071 102.554 R2 0.016 0.08
Note: standard errors specified are reported in brackets and *, **, *** indicate significance at a 10%, 5%, 1%
confidence level, controls include age, age difference, region, education difference and normalized PSC respectively. Column 1 presents an OLS regression of variables on BISP, Column 2 is an instrumental variable regression where BISP Programme status is instrumented by Z. Columns 3-5 show different Propensity Score Matching methods.
24 (column 2, s.e 0.098). Furthermore, the binary variable indicating whether a woman would be allowed to seek employment if she chooses is shown to increases by 0.14 (column 2, s.e 0.084). For the rest of the results none of the estimates are found to be significant, showing no association between treatment and decision making power over fertility, education or medical care.
The estimates for the propensity score matching can be found in columns 3, 4 and 5. At first glance the results show a significant increase in bargaining power over consumption across all matching methods. Nearest neighbour matching found influence over food purchases increases by 0.14 (column 3, s.e 0.024) given treatment and clothing increases by 0.10
(column 3 s.e 0.025) with medicine also having a positive but smaller impact of 0.04 (column 3 s.e 0.025). The results for the other two matching methods, radius and kernel matching in columns 4 and 5 respectively, also showed positive significant results for clothing, medicine and food decision making. Whilst these are far more modest results then those found when using the regression discontinuity approach the estimates for both are positive and significant. There appears to be a negative relationship between treatment and autonomy over education choices, with significant negative results of -0.044 (column 3, s,e 0.013) in the nearest neighbour matching and the other matching methods, however these results were not
reflected in the previous estimation method. Furthermore, as the women in the sample are all married with children it is likely that they have already realised their education, as a result the possibility of them returning to education is rare.
Whilst propensity score matching can eliminate much selection bias it must be recognized that it can only do so with observed characteristics, there may still be considerable
differences between matched individuals because of unobserved characteristics. Although as many covariates were included as possible to create the propensity score, the estimates are highly sensitive to which controls are included. Unobserved characteristics may be a particular problem with the dataset used due to the considerable leakage in treatment. It is possible that eligibility was determined by other factors not available in the data or made known by the programme. Furthermore, standard errors do not take into account that
propensity scores were estimated. The estimated variance of the treatment effect should also include the variance due to the estimation of the propensity score (Caliendo and Kopeinig,
25 2008), as a result standard errors can often be underestimated. Especially when using nearest neighbour matching whereby only using one treated observation the matched results will understate the standard errors (Smith and Todd, 2005). In order to minimise this effect, the standard errors were bootstrapped as suggested by Lechner (2002), however the bias in the results must still be taken into account when attempting to interpret the results.
As a result of the imperfect compliance in the programme and the weakness in the instrument when using certain specifications, precise estimates and causation should not be drawn from the results. Furthermore, it must be noted that the mean values of different background characteristics are statistically different depending on eligibility and treatment status. These observed characteristics are often related to variables used to calculate PSC score and thus by definition are meant to differ across eligibility. Whilst these observed characteristics are controlled for there is always the risk that results are driven by unobserved characteristics that differ across treated and untreated individuals.
6. Discussion and Conclusion
The Benazir Income Support Programme was set up to provide cash transfers to women in Pakistan, the programme intended to cover the poorest 25% of households with a direct transfer of Pk Rs. 1,500 rupees per month to the woman of the household. However, as a result of several conditions placed on the transfer it was difficult to precisely isolate the intended recipient group. Whilst as many conditions as possible were upheld to find a relevant sample group there was still considerable leakage. Firstly, many women who
appeared to be eligible for the benefit were not recipients, this may be due to the programmes limited funding and oversubscription. Likewise, it may be the result of women receiving an unreported benefit or not being in possession of a valid identity card, information that was not included in the dataset and resulted in a woman being ineligible. In the final sample, only 24% of eligible women were recipients. Furthermore, many households with a poverty score
26 card count substantially above the eligible level were receiving the benefit, over half of recipients were deemed over the eligible PSC score. This may be the result of households previously having been identified through parliamentarians with no guidance from proxy means testing. Local officials determined who received the benefit from 2008 till 2011, with these women continuing to receive the funds. However even though they were above the 16.71 eligibility threshold these women tended to have a lower PSC score than average and the majority were still in poverty, suggesting the initial selection process was somewhat successful. In some cases though, it may have been evidence of the apparent corruption detected within the programme. The Federal Investigation Agency reported frequent cases of irregular payments throughout the programme's history. This highlights several shortcomings of the programme design and implementation that should be further improved, however it must be noted that the majority of recipients were found to be below the national poverty line.
As a result of the substantial leakage in the programme the cut-off PSC score had limited effect on whether women were recipients, any jump detected in the probability of treatment around the threshold was correspondingly small. This created significant problems with the estimation method used in the study and the findings should not be viewed as precise estimates. Whilst some of the results from the propensity score matching showed there may be positive spillovers from the benefit in terms of female empowerment these were limited to consumption choices and did not take into account previous existing trends. However, they do suggest that women receiving the benefit may have greater decision making power over how income is spent in terms of food and clothing. Furthermore, there is evidence that
women receiving treatment have control over the benefit income received and how it is spent.
Finally, it must be noted that the main aim of the programme was to relieve poverty and help severely resource constrained households. Despite the benefit being unconditional, it intended to help families invest in nutrition and human capital. To this end the programme can be deemed a success, previous studies have confirmed that the Benazir Income Support
Programme helped reduce extreme poverty and increase wellbeing. Any positive spillovers in terms of female empowerment would be an additional benefit.
27
7.References
Abel, Martin (2014). Unintended labour supply effects of cash transfer Programmes: Evidence from South Africa’s Old Age Pension. Harvard University Working Paper.
Adato, Mindek, and Quisumbing (2000) The Impact of PROGRESA on Women’s Status an Intra-household Relations.
Ahmed, Akhter U, Agnes R Quisumbing, Mahbuba Nasreen, John F Hoddinott and Elizabeth Bryan (2009), ‘Comparing Food and Cash Transfers to the Ultra Poor in Bangladesh’, IFPRI Research Monograph, No. 163 (Washington DC: IFPRI)
Alderman, Chiappori, Haddad, Hoddinott, and Kanbur (1995). Unitary versus collective models of the household: Is it time to shift the burden of proof? The World Bank Research Observer, vol 10, no.1, pp 1-19.
Ali, and Fani and Saima (2010) Cultural Barriers in Women Empowerment: A Sociological Analysis of Multan, Pakistan, European Journal of Social Sciences, 18(1), 147-155.
Asian Development Bank (2009) Cash Transfer Programmes for Poor Women in Pakistan. Accessed April
2011<www.adb.org/Documents/Brochures/Social-Protection-Project-Briefs/PAK-Cash-Transfer.pdf>.
Benazir Income Support Programme (2015) Government of Pakistan, BISP, at a Glance. Accessed 5 April 2016<http://www.bisp.gov.pk>.
Beegle, K., Frankenberg, E., and Thomas, D., (2001). Bargaining power within couples and use of prenatal and delivery care in Indonesia. Studies in Family Planning, 32(2), 130–146.
Caliendo, M. and Kopeinig, S. (2008). ‘Some practical guidance for the implementation of propensity score matching’. Journal of Economic Surveys, 22(1), pp. 31-72.
Cheema Iftikhar, Farhat Maham, Hunt Simon, Javeed Sarah, Pellerano Luca, O’Leary Sean (2014), Benazir Income Support Programme, First follow-up impact evaluation report, Oxford Policy Management Limited, United Kingdom; p: 1-109
Duflo, Esther (2003). Grandmothers and granddaughters: Old age pension and intra-household
allocation in South Africa. National Bureau of Economic Research Working Paper 8061. Available at: hhttp://www.nber.org/papers/w8061
Fafchamps, M., and Quisumbing, A. (2002). Control and ownership of assets within rural Ethiopian households. The Journal of Development Studies, 38(6), 47-82
Gomez de Leon, J. and S. Parker (1999), “The Impact of Anti-Poverty Programmes on Labor Force Participation,” Progresa, Insurgence Sur 1480, Col. Barrio de Actipan, 03250, Mexico, DF
Gitter, S. R., and B. L. Barham (2008). "Women's Power, Conditional Cash Transfers, and Schooling In Nicaragua." World Bank Economic Review 22(2): 271‐290.
Gubrium, E., Pellissery S., Lødemel , I. eds (2014) The Shame of it: Global Perspectives on Anti Poverty Policies. Bristol: Policy Press
28 Hoddinott, John, and Lawrence Haddad (1995) “Does Female Income Share Influence Household Expenditure? Evidence from Cote d’Ivoire”, Oxford Bulletin of Economics and Statistics, LVII, Heckman, J. J., Ichimura, H., and Todd, P. (1998). ‘Matching as an econometric evaluation estimator’. The Review of Economic Studies, 65(2), pp. 261-294
Katchova, A. (2013). Propensity score matching. Retrieved from
https://sites.google.com/site/econometricsacademy/econometrics-models/propensity-scorematching Khan S. and S. Qutub (2010) The Benazir Income Support Programme and Zakat Programme: A Political Economy Analysis of Gender, ODI Research Paper. London: Overseas Development Institute.
Kabeer, N. (2005). ‘Gender equality and women's empowerment: A critical analysis of the third millennium development goal 1’. Gender & Development, 13(1), pp. 13-24.
Lechner, Michael (2002), “Programme Heterogeneity and Propensity Score Matching: An
Application to the Evaluation of Active Labor Market Policies,” Review of Economics and Statistics 84 (May 2002), 205–220.
Ravallion, M. (2003). ‘Assessing the poverty impact of an assigned Programme’. The Impact of Economic Policies on Poverty and Income Distribution: Evaluation Techniques and Tools, 1. Retrieved 20/05/2015
Schultz, T. (1990). "Testing the Neoclassical of Family Labor and Fertility." Journal of Human Resources 25: 599‐634.
SEWA Bharat (2014). A Little More, How Much It is– Piloting Basic Income Transfers in Madhya Pradesh, India. New Delhi: SEWA Bharat and UNICEF India Office
Smith, J. (2000): “A Critical Survey of Empirical Methods for Evaluating Active Labor Market Policies,” Schweizerische Zeitschrift fr Volkswirtschaft und Statistik, 136(3), 1–22.
Soares, F. V. and Silva, E. (2010). Conditional cash transfer Programmes and gender vulnerabilities: case studies of Brazil, Chile and Colombia. No. 69. International Policy Centre for Inclusive Growth. Sophie C. (2007) The women empowerment approach: A methodological guide,
<http://www.atol.be/docs/publ/gender/women_empowerment_approach_CVO.pdf>.
Statistics Division (2014) Pakistan Social and Living Standards Measurement Survey 2013-14, Islamabad: Federal Bureau of Statistics, Government of Pakistan.
Thomas, D. (1990). "Intra‐household allocation: an inferential approach." Journal of Human Resources25: 635‐664.
Todd, P. E. (2007). ‘Evaluating social Programmes with endogenous Programme placement and selection of the treated’. Handbook of Development Economics, No. 4, pp. 3847-3894.
UN. (1995). Beijing Declaration and Platform of Action, adopted at the Fourth World Conference on Women. at http://www.un.org/womenwatch/daw/beijing/platform/declar.htm
UNESCO (2014) Why Gender Equality in Basic Education in Pakistan? Accessed 9 September 2016 http://unesco.org.pk/education/documents/publications/
29
8. Appendix
Table 4. Variables for Propensity Score Matching
Variable Name Type Unit
PSC Poverty Score 1-100 Score
Age The Woman’s Age Years
Region Urban or Rural Dummy
Province Province of Household Categorical
Age Difference Age difference with spouse Years
Age Married Age at Marriage Years
Education Woman’s Education Categorical
Head Education Spouses Education Categorical
Agriculture Household Head Working in Agriculture Dummy House Occupation Home owner or rental Number
Number of Rooms Rooms in House Number
Electricity House with electricity connection Dummy
Gas House with gas connection Dummy
Telephone House with telephone connection Dummy Air cooler Air conditioning or Heating Dummy
Toilet House with a Flush toilet Dummy
Table 5a. Nearest Neighbor Matching
Variables Treatment Control ATT S.E T-stat
Fertility 1095 913 0.004 0.026 0.141 Employment 1095 913 0.017 0.019 0.894 Education 1095 913 -0.044*** 0.013 -3.469 Food 1095 913 0.141*** 0.024 5.834 Clothing 1095 913 0.103*** 0.025 4.129 Medicine 1095 913 0.041* 0.025 1.647
Note: Standard Errors are bootstrapped, and *, **, *** indicate significance at a 10%, 5%, 1% confidence
level, respectively. The numbers of treated and controls refer to actual nearest neighbor matches
Table 5b. Radius Matching
Variables Treatment Control ATT S.E T-stat
Fertility 1095 9195 0.035** 0.016 -2.188 Employment 1095 9195 0.055*** 0.012 4.539 Education 1095 9195 -0.070*** 0.008 -9.164 Food 1095 9195 0.158*** 0.014 11.414 Clothing 1095 9195 0.152*** 0.014 10.510 Medicine 1095 9195 0.066*** 0.014 4.861
Note: The numbers of treated and controls refer to actual matches within radius, Standard Errors do not take into account that the propensity score is estimated.
30 .5 .6 .7 .8 .9 1 -20 0 20 40 60 80 Normalized PSC
Employment Decision Making
.2 .4 .6 .8 1 -20 0 20 40 60 80 Normalized PSC
Food Decision Making
Average Outcomes of Interest by Forcing Variable
Figure 3a. Figure 3b.
Figure 3c Figure 3d
Table 5c. Kernel Matching
Variables Treatment Control ATT S.E T-stat
Fertility 1095 9195 -0.009 0.015 -0.589 Employment 1095 9195 0.044*** 0.013 3.497 Education 1095 9195 -0.051*** 0.007 -6.892 Food 1095 9195 0.132*** 0.014 9.627 Clothing 1095 9195 0.122*** 0.016 6.582 Medicine 1095 9195 0.048*** 0.014 3.381
Note: Standard Errors are bootstrapped, and *, **, *** indicate significance at a 10%, 5%, 1% confidence
level, respectively .4 .6 .8 1 -20 0 20 40 60 80 Normalized PSC
Fertility Decision Making
0 .2 .4 .6 -20 0 20 40 60 80 Normalized PSC
31 0 .2 .4 .6 .8 1 -20 0 20 40 60 80 Normalized PSC
Clothing Decision Making
.6 .7 .8 .9 -20 0 20 40 60 80 Normalized PSC
Healthcare Decision Making
