Productive Transfers, Intra-household Bargaining and Empowerment Evidence from a Randomized Trial in Nicaragua

Transcription

1 Productive Transfers, Intra-household Bargaining and Empowerment Evidence from a Randomized Trial in Nicaragua Chrysanthi Hatzimasoura*, Patrick Premand** and Renos Vakis** January Abstract The strength of the relationship between economic development and gender empowerment is an important empirical question both at the macro and micro levels. We test whether a productive transfer intervention providing a package of capital and skills to women in poor rural Nicaraguan communities affected their empowerment. The intervention led to changes in individual-level employment and earnings for the main women beneficiary, but not on overall household-level employment, income or consumption. The intervention also led to higher intra-household bargaining power for women. Women s private goods constitute a larger share of household non-food expenditures, consistent with a significantly larger Pareto weight in a non-unitary household model. Decision-making in the household become more participatory and spousal relationships also appear to improve. Keywords: female empowerment, intra-household bargaining, employment, productive transfers, field experiment, Nicaragua. * George Washington University; ** The World Bank 1 We are grateful to Ana María Muñoz Boudet, Soledad Cubas, Amer Hasan, Teresa Suazo and Egda Velez for contributions during the study implementation, as well as to FUMDEC, who implemented the intervention under the leadership of Rosa Adelina Barahona, Marlene Rodriguez and Milton Castillo. Special thanks go to the FUMDEC field teams for their patience and engagement throughout the study. The FUMDEC intervention was partly funded by the Buffett and Trickle-up foundations. The impact evaluation was generously funded by the World Bank Gender-Action Plan Results-Based Initiative (RBI) and a BNPP trust fund. UCW provided additional funding to analyze child-labor data, and we are grateful to Jacobs de Hoop and Furio Rosati for their contributions in analyzing program impacts on child labor and schooling. We are most grateful to Karen Macours for countless contributions during the design and implementation of the study, as well as Veronica Aguilera and Enoe Moncada at CIERUNIC for high-quality data collection. 1

2 1. Introduction There is a potential two-way relationship between economic development and women s empowerment (Duflo, 2013). Economic development can help reduce gender inequalities. Gender inequalities are often deeper among the poor, and economic development tends to also improve women s well-being, including among others through changes in the education, fertility or employment spheres. On the other hand, women s empowerment can contribute to development. Unequal access to land, labor or financial markets by women can generate economic inefficiencies and high welfare costs (World Bank, 2001; World Bank, 2012). Unequal resource endowments (in terms of both assets and human capital) or market failures (such as missing markets for women or asymmetry of information on women s productivity) can affect intra-household bargaining. Societal institutions (including social norms, customs, legal) as well as limited agency or aspiration failures may also contribute. Gender inequalities in intra-household bargaining can limit women s labor-supply or investment opportunities, in turn lowering the average return on a household s income portfolio. At the same time, improvements in women s empowerment are not necessarily efficiency enhancing, and can sometimes come at the expense of men (Duflo, 2013). The relative balance between broad-based policy actions aimed at fostering economic development and more targeted actions focused on reducing gender inequalities has been a matter of discussion. Interventions promoting female empowerment have the potential to contribute to increase productivity and reduce poverty. Improvements in women s empowerment is also often seen as desirable outcomes by themselves (World Bank, 2006), as reflected in the fifth sustainable development goal. Duflo (2013) argues that the mutually reinforcing nature of economic development and women s empowerment may not be as strong as expected, so that specific policy actions on women s empowerment remain necessary. Similar issues on the nature of the interlinkages between economic development and gender empowerment arise at the micro-level. The core question is whether programs seeking to improve women s economic status are conducive to their overall empowerment, either within the household or more broadly in societies. Productive interventions and employment programs are increasingly used with the objective to promote women s economic status and their intra-household empowerment. Productive interventions include programs that provide capital or training to improve women s economic opportunities with the objective to facilitate access to more productive occupations or livelihoods, often in self-employment. In theory, productive interventions targeted at women have the potential to generate economic gains and empowerment effects that are mutually reinforcing. Increased control over resources in the hands of women can increase the intra-household bargaining power of women, which may in turn improve resource allocations and returns on additional household investments. By targeting women, productive interventions can therefore affect intra-household bargaining power in the short-term and promote household income diversification by facilitating women s investments in activities with higher potential returns. In doing so, increased empowerment 2

3 can facilitate further welfare improvement in the long-term. Still the mechanisms that link economic empowerment and intra-household bargaining remain unclear in the empirical literature. The literature on the impacts on productive interventions on economic empowerment, as well as on intra-household bargaining, points to varying success. The evidence on programs that aim to promote women s economic empowerment through providing access to capital, training, or savings facilitation has shown mixed results, although interventions that seek to address social constraints have had more positive results (for reviews, see Buvinic and Furst-Nichols, 2014, or Todd, 2012). On the other hand, the literature on intra-household decision-making shows, mostly through observational data, that women s bargaining power can affect economic outcomes. Yet the determinants of intra-household bargaining empowerment, and in particular the extent to which it is malleable through interventions, remains a subject of active research (for a review, see Doss, 2013). There are multiple potential pathways through which productive interventions affect empowerment and intra-household bargaining. Some studies show that economic impacts for women come with changes in a range of life skills and risky behaviors (Bandiera et al., 2012; World Bank, 2012). Macours, Premand and Vakis (2012) show that impacts of productive interventions on the poor may have come with shifts in women s aspirations. Macours and Vakis (2008) show that a productive safety nets programs creates large social spillover effects among beneficiaries by enhancing impacts on investments in human and physical capital. Ashraf (2009) studies financial choices of married individuals in the Philippines, showing that information and communication strongly affects the choice of male spouses to invest in a bank account jointly held with their spouse. Majlesi (2016) studies how changes in labor market opportunities for men and women affect women's decision making. Results show that increases in labor market opportunities improve women's decision-making power. This literature on productive intervention and employment complements research on safety nets, and in particular cash transfer programs. Many transfers are targeted at women on the grounds that they will be used more efficiently. Transfers targeted to women can also affect empowerment, even in contexts where empowerment did not constitute an original program objective (Adato et al., 2000; Duflo, 2003; Rangel, 2005). Almas et al. (2015) find that providing cash transfers to women instead of households head significantly increase women s empowerment and their bargaining position in the household. There is also evidence that transfers given to women are more likely to be invested in children s education, nutrition or housing than transfers given to men (Thomas, 1994; Hoddinott and Haddad, 1995; Duflo, 2003) 2. 2 We do not focus on children s outcomes in this paper, but see de Hoop et al. (2015), for a detailed analysis of the intervention s impacts on children s outcomes. 3

4 A large literature has developed and tested alternative models of intra-household allocation (for reviews see Donni and Chiappori, 2011). Unitary household models have been largely rejected empirically. Non-unitary models consider households with at least two members who have different preferences. Cooperative models assume that household decision leads to Pareto-efficient outcomes. Non cooperative models allow for Pareto inefficiencies, and start from the point that household members maximize their own utility subject to their own budget constraint, while taking the decisions of their partner into account. In such a model, a women s relative power in the household can be captured by her Pareto weight. A key empirical question and testable hypothesis is the degree to which the Pareto weight can be affected by outside shocks. In this paper, we test the degree to which a productive intervention targeted to poor rural women affects intra-household empowerment. We do so by documenting how the intervention affected economic outcomes along a range of dimensions (employment, earnings, ), and how these changes contributed to changes in intra-household bargaining. We measures changes in intra-household bargaining through the Pareto weight derived from a non-unitary household model. We also document impacts on other proxies, including intra-household decision-making, quality of spousal relations and individual agency. The study analyzes a program operated by the Fundación Mujer y Desarollo Communitario (FUMDEC 3 ), a local NGO with multi-year experience in rural Nicaragua on productive transfers and gender. In 2009, the World Bank and FUMDEEC partnered to launch a program the objective of which was to design, implement and evaluate the impact of productive transfers on income generation and women s empowerment. The intervention was targeted women in rural communities of the municipality of Santa Maria de Pantasma in the department of Jinotega, located in the north of Nicaragua. The program offered households with at least one adult female member a package of benefits that included (i) training on community organization and gender awareness, (ii) training in technical or business skills to develop or expand small-scale household enterprises, livestock or agricultural activities, (iii) capital transfers in the form of cash, seeds, or livestock and (iv) follow-up technical assistance. The exact mix of capital transfers, training and technical assistance was adjusted depending on the type of activity each beneficiary wished to start or expand. An experimental impact evaluation design was integrated into the project in order to learn from the program intervention and inform potential scaling-up decisions as well as ongoing productive programs targeting poor rural women in Nicaragua. As a result, eligible communities were randomly assigned to treatment and control groups. The experiment was designed to evaluate the impact of the program both on economic empowerment (including on a range of outcome indicators such as income generation, labor market participation, asset ownership) as well as intra-household bargaining empowerment. 3 The FUMDEC program was selected because of its scale-up potential following after policy consultations involving many Nicaraguan NGOs. 4

5 Results show that the intervention improved beneficiaries skills (such as business practices) and facilitating access to capital, both through direct transfers as well as facilitation of savings and informal credit. This led to limited impacts on beneficiaries employment, mostly small changes along the intensive margin, and raised the beneficiary s individual earnings. Spill-over effects on other adults household members employment are limited. Ultimately, no impacts on household income or consumption are observed. The intervention led to changes in individual-level employment and earnings for the main women beneficiary, but not on overall household-level employment, income or consumption. The intervention also led to higher intra-household bargaining power for women. Women s private goods constitute a larger share of household non-food expenditures, consistent with a significantly larger Pareto weight in a non-unitary household model. Decision-making in the household become more participatory and spousal relationships also appear to improve. This report describes the intervention, the impact evaluation design, results from the baseline data collected for the impact evaluation and then describes the impact of the intervention on program participants. The report is organized as follows. Section 2 describes the intervention, section 3 describes the profile of women who decided to participate in the program. Section 4 describes the evaluation design. Section 5 documents the economic impact of the program on beneficiary women and their household. Section 6 documents impacts on intra-household bargaining and empowerment. Section 7 concludes. 2. The Intervention 2.1 Objectives and Components In 2009/10, a Nicaraguan NGO (Fundación Mujer y Desarrollo Comunitario, or FUMDEC) 4, implemented a productive transfer program with support from the World Bank. The intervention built on a model in place in other communities in northern Nicaragua since 1996, and had two main objectives: (i) to facilitate income generation and diversification by promoting women s economic activities, and (ii) to foster gender empowerment by improving women s aspirations, their participation in households economic decisions as well as their social participation. To achieve these objectives, the program offered households with at least one female member 16 to 60 years old a package of benefits that included (i) training on community organization and gender awareness, (ii) training in technical or business skills to develop or expand small-scale household enterprises, livestock or agricultural activities of their choice, (iii) capital transfers in the form of cash, seeds, or livestock and (iv) follow-up technical assistance. The exact mix of capital transfers, training 4 For more information about FUMDEC, see 5

6 and technical assistance was adjusted depending on the type of activity that each beneficiary wished to start or expand: non-agricultural household enterprises, livestock, or other small-scale agricultural activities. Specifically, the program included three main phases. (The next section describes the prior beneficiary selection process). First, beneficiaries were offered various training workshops on community organization, social capital and gender awareness. The community organization training included 4 modules for all beneficiaries on how to form and organize women s groups, as well as two additional modules for selected leaders on managing women s groups. The support to community organization aimed at strengthening social dynamics and promoting project sustainability via social interactions, learning and later facilitating access to credit managed by the beneficiaries. The gender awareness training was offered to all beneficiaries and included 8 modules on issues such as gender roles and relations, self-esteem, reproductive health, power relations, intra-household decision making, use and control of resources, conflict resolution, domestic violence, laws that protect women, These trainings were delivered in the community. In addition, a gender awareness training targeted a small number of selected men, usually leaders spouse, who were trained together in a central location on a sub-set of the aforementioned themes. Second, beneficiaries were offered training in technical or business skills to develop or expand smallscale household enterprises, livestock or agricultural activities of their choice. Each beneficiary was offered between 4 and 6 training sessions. The scope of the training depended on the activity chosen by individuals. It focused on technical skills and crop management for individuals who chose agricultural activities. It tackled livestock management and related technical skills for individuals who chose livestock activities. And it covered basic business skills for individuals who chose small business activities. Throughout these activities, beneficiaries formed plans to prepare for the capital transfers. Third beneficiaries received capital transfers in the form of cash, seeds, or livestock. Beneficiaries received start-up capital in both cash and kind. Part of this capital was in the form of low cost credit to be repaid to community banks. The final amount of the productive transfers depended on the package and was a function of the specific material, inputs and cash provided to the beneficiary. Credit came at zero interest, with the exception of: (i) a $200 micro business transfer of which half was in the form of credit at 3% monthly interest rate over the remaining balance; (ii) seeds that were repaid to the community seed bank at a rate of two seeds for each seed given. Repayment periods varied depending on the package and were linked to crop and harvest cycles. They ranged between eight months for the cash credit and up to a year for the rest. After the transfers, follow-up technical assistance visits were organized 5. 5 One additional component of the program, not yet implemented at the time of the follow-up survey, consisted in the creation of community banks. For this purpose, the program would provide training in management and 6

7 The program had an additional component, which was not implemented prior to the follow-up data collection, related to the creation and support to community banks. The NGO intended to provide training in management and organization to community leaders and initial technical support to set-up and manage beneficiary-led communal banks. It was envisioned that these banks would eventually become as a sustainable source of credit for the community. Tables A1.1 and A1.2 in Appendix contain a detailed break-down of the intervention content. The package had an average value of US$602 per beneficiary (the exact value depending on the type of activity being supported) 6. It included US$316 in direct capital transfers (in the form of a mix of cash, seeds and livestock) and US$286 that covered the costs of training and technical assistance. More than 80 percent of the targeted households live on average with less than US$2 per capita per day, and the transfer amounted to around 24% of pre-transfer annual household consumption, a rather sizeable magnitude. 2.2 Eligibility and beneficiary selection process The program was implemented in rural communities in the municipality of Santa Maria de Pantasma, located in the department of Jinotega in the Northern region of Nicaragua. Nicaragua is the 2nd poorest country in Latin America after Haiti. Santa Maria de Pantasma is one of the municipalities with the highest incidence of poverty in Nicaragua, with an incidence of extreme and general poverty estimated to be 34% and 79% respectively based on a 2005 census eligible communities were selected following criteria used by the implementing NGO including: (a) being located in a rural area (b) not have benefited from other related interventions/programs, (c) having more than 20 households (d) be in a territory known for the implementing NGO, and (e) having local authorities agreement to the implementation of the programs in these communities. The community selection was undertaken by the implementing NGO (FUMDEC), in coordination with municipal authorities and the World Bank. 8 Baseline census data was collected in all households in eligible communities from mid-june to mid-july 2009 (Table A1.3). The baseline instrument was designed to identify a potential program beneficiary (titular) in each household. The potential beneficiary in each household was identified organization to community leaders and initial technical support. It was envisioned that these banks would eventually become as a sustainable source of credit for the community. 6 In addition, administrative costs for the pilot amounted to US$225 per beneficiary, for a ratio of administrative costs to total transfers of 37%. 7 World Bank (2008) 8 Narrow geographic targeting was required by constraints on the implementing NGO s current operations. Nonetheless, indicators for the selected municipality remain comparable to indicators for many other municipalities in Nicaragua, including those in the Atencion a crisis programs (Macours, Premand, Vakis; 2013). 7

8 through the following question: 9 Of all women between 16 and 60 in the households, who is the one who most participates the most in activities in the community, for example related to health, education, production or organization. Following the baseline survey, all the identified potential beneficiaries were invited to a first information assembly organized by the implementing NGO (see Table A1.4). The invitation process was undertaken in collaboration with community leaders and including door-to-door invitations of each household. A total of 78% of invited potential beneficiaries participated in the first assemblies. During this first assembly, the program objectives and components were presented. The process that was going to be used to choose the final list of beneficiaries was explained. All communities were also made aware that only half of the total targeted communities would be randomly chosen to receive the intervention. All potential beneficiaries were asked to consider signing up for the program and were given a week to do so. A few days after the first assembly, one community dropped out. The community leader delivered a letter to the implementing NGO announcing that women in the community were already participating in another program and as a result could not commit to participating in the FUMDEC program. The community dropped out before the public lottery to assign control and treatment communities (see below) and so it did not affect the randomization process. It was also the smallest community, with only 21 households (see table A1.3). Approximately a week after the first assembly, a second assembly was held in all remaining eligible communities and potential beneficiaries were asked to enroll in an intent-to-participate list (conditional on their community being selected). These assemblies took place between mid-july and mid-august Tables A1.5 and A1.6 present enrollment data of self-selected intent-to-participate women (and Table A1.7 by community). Out of the 1956 households with potential beneficiaries in the 24 target communities, 877 (45%) enrolled and 980 (50%) did not enroll. The remaining 99 households (5%) were not eligible as they did not have a woman between the ages of 16 and 60 who could serve as a beneficiary. Once this process was complete, a public lottery was organized at the municipal center in order to make the final community selection for the intervention. Communal and municipal leaders were invited along with representatives from beneficiaries in each community. The implementing NGO and a World Bank counterpart chaired the assembly (on August 14th 2009). Target communities where divided in 10 blocks of 2 or 3 neighboring communities. 10 The lottery was organized so that a community (or pair of communities) was selected within each group (one allocated as treatment and 9 The exact Spanish phrasing was: De todas las mujeres del hogar, quien es la que más participa en actividades comunitarias, por ejemplo en salud, educación, producción o organización? 10 Specifically, the communities were divided in 6 pairs of 2 communities and 4 with 3 communities. Communities were grouped by proximity. When 3 communities were close to each other, the smallest two were paired within the group. 8

9 the other as control). This block randomization procedure between neighboring communities maximizes power when the number of clusters to randomize is low (Bruhn and McKenzie, 2008). Table A1.7 contains results from the public lottery. Block randomization led to selection of 13 treatment communities and 11 control communities. The intervention began in late August 2009, when the implementing NGO visited beneficiaries in treatment communities to finalize the allocation of specific benefit packages and organize the implementation of the various program components. Between August and October 2009, the first training workshops were held, including on community organization, leaders identification and gender awareness. Implementation of the productive packages started soon thereafter including the training related to the various packages. Agricultural benefit packages were provided in the fall of 2009 (in time for the last crop cycle of 2009) and between March and June 2010 (before the main crop cycle of 2010). For micro-businesses, start-up transfers were conditional on business plan development and training and final payments took place by May All benefits were delivered by September 2010, after which activities from FUMDEC ended in the participating communities until after the follow-up survey. 3. Study Design 3.1 Data Baseline census data was collected in all households in the eligible communities before the implementing NGO entered the territory (from mid-june to mid-july 2009). Data collection was led by a local data collection firm (CIERUNIC). The baseline was designed based on maps of enumeration areas from the National Institute of Statistics (INIDE). Local leaders were asked to certify that community boundaries were respected and that the final list of surveyed households was complete. A short questionnaire aimed to capture basic variables to build a profile of households, as well as measure key outcomes for the impact evaluation (see below). The instrument also included a basic mathematical test to proxy individual ability. As mentioned above, the baseline instrument also allowed the identification of a potential program beneficiary (titular) in each household. The census instrument contains socio-demographic variables as well as a limited set of key outcome variables measuring intra-household decision-making and economic activities for potential beneficiaries. The first column of Table A1.3 summarizes the scope of baseline census data collection. Across eligible communities, 1956 complete or marginally incomplete interviews were realized. 3 9

10 households refused to participate, 28 households were absent throughout the data collection period and 159 dwellings appearing on the official map of the enumeration areas were un-occupied 11. The follow-up survey took place between July and December 2011, approximately two years after the baseline, and approximately one year after the end of the program. The follow-up instrument aimed at capturing a broader set of outcomes, thus allowing a thorough evaluation of the program s impact including: (i) welfare outcomes such as consumption and income; (ii) employment and economic activities (such as involvement in agriculture, livestock, or non-agricultural selfemployment); (iii) assets and access to credit, as well as (iv) female empowerment (such as intrahousehold decision-making, aspirations, and self-esteem). 3.2 Balance between treatment and control groups This report focuses on the direct program impact on beneficiary women one year after the end of the program. In order to estimate this, program impacts can be estimated using simple intentto-treat procedures by comparing outcomes at follow-up between beneficiaries in treatment communities (TSI in Figure A1.1) to those of women who selected into the program in control communities (CSI in Figure A1.1, who did not receive the program because they were randomized out). The difference in outcomes between these two groups yields the average treatment effect on the treated, i.e. the direct effect of the intervention on comparable households who selected into the program. Randomized assignment of the program between the 24 target communities ensures that there are no differences between the treatment and control communities. This section uses baseline data to test the validity of this impact evaluation design. As the results show, while the number of clusters to randomize is small, randomization did achieve balance between the control and treatment groups. As would be expected, some marginal differences between the two groups remain but are marginal and do not invalidate the impact evaluation design. First, intended program participation between treatment and comparison communities is the same. As discussed in section 2.1, 43 percent of households in control communities selected in the program (but did not receive), 52 percent selected out and 5 percent where not eligible. This compares with 47 (selected in), 48 (selected out) and 5 percent for ineligible in treatment communities. These differences are not statistically significant. Tables A2.1 to A2.8 in Annex 2 present comparisons of households who selected in the program in both treatment and control communities (comparing groups TSI and CSI in Figure A1.1). 11 The high number of un-occupied households partly reflects the fact that many households temporarily migrate to the target communities during the coffee harvest season. 10

11 The objective is to test whether women that self-select into the program do not differ between control and treatment communities. Only some marginal differences appear. For instance, treatment households who selected in have a higher rate of owning poultry than control households who selected in (6.5 points higher), are more active in agriculture (5.7 points higher) and were more likely to report having participated in Nicaragua s CCT program 10 years earlier (0.7 points higher). The main noteworthy difference relates to differences in empowerment between households in treatment and control communities. This implies that the impact evaluation will be conservative and only identify large post-intervention differences in empowerment between the two groups. The empirical analysis will control for such imbalances. Annex A3 contains a detailed discussion of comparisons of characteristics between all households in beneficiary and control villages (whether they are selected in or not). Overall, the results confirm the robustness of the experimental design of the study in that there are very few differences between control and treatment communities. For the few significant differences, treatment control communities appear slightly worse off than control communities in the specific dimensions. This suggests that, at worse, estimated program impacts will be lower-bound, conservative estimates. 3.3 Estimation Strategy: Intent to Treat Given that the randomization achieved balance, program impacts can be estimated through first-difference in outcomes at follow-up between beneficiaries in treatment communities and women in control communities who initially signaled an interest in participating in the program. Formally, we estimate the intent-to-treat for those selected in (ITT), the primary outcome of interest. The ITT outcome captures how random assignment to the treatment group affects the outcome of interest and provides an estimate of the program s average effect on participants who had selected in: (1) Y = α + δ * T1 +Χ+C+ ε where Y is the outcome of interest (such as income, employment, etc ), T1 is an indicator function that takes on the value 1 if a household expressed interest in participating in the program and was assigned to the treatment group, and 0 if it expressed interest in participating in the program but was assigned to the control group. X is a fixed effect controlling for the variation within the blocks of communities on which randomization was based. The intent-to-treat estimate of program impact is given by δ. 12 C includes a large set of baseline control variables to address the few observed imbalances. We use fixed effects to maximize power. The block randomization of the design allows us to control for outcomes that are a function of pre-program characteristics of the blocks of communities that also affect the outcome of interest. In our estimations, standard errors are clustered at the community level. 12 Standard errors are clustered at the community level. 11

12 The attrition rate for beneficiaries is 5.2 % which is balanced across the communities that received the treatment and those that did not. 4. Profile of women who decide to enroll Before we turn to the impact results, this section discusses a key finding from the baseline data in relation to the self-selection process to enroll into the program. As a reminder, this self-selection occurred before the lottery to assign communities between treatment and control. Tables A4.1 to A4.8 compare baseline characteristics of households that selected into the program after the 2 nd assembly (both in treatment and control communities, groups TSI and CSI in figure A1.1) to those that selected out of the program (both in treatment and control communities, groups TSO and CSO). The analysis suggests that potential beneficiaries who selected in the program are better-off than those who selected out. They have better indicators of dwelling quality (floor, number of rooms), dwelling ownership, access to services (water), durables (table ownership) and are located closer to the local primary school (often the community center) (table A4.1). Women who selected in come from larger households (table A4.2), and have higher levels of education (table A4.3). Potential beneficiaries who selected in are also less likely to be poor, or have consumption below $1 a day or $3 a day (table A4.4). Based on predicted values from the proxy-mean formula, households who selected in are 4.8 points less likely to be poor, 3.8 points less likely to have consumption below $1 a day, and 2.3% less likely to have consumption below $3 a day. Similar patterns also arise in relation to economic activities (tables A4.5 and A4.6). Women who selected in are more likely to work and come from a household with more adult workers and more diversified economic activities. Specifically, households who selected in are less likely to be agricultural workers, and more likely to be self-employed in agricultural or non-agricultural activities. This is reflected in the fact that households who selected in are more likely to own land, assets (pigs or poultry), take credit, and own a small household business. Women who selected in have better empowerment indicators than potential beneficiaries who selected out (Table A4.7). In other words, it is mainly the most empowered women who self-selected into the program. They are more likely to take decisions alone on children s school attendance (by 4.9 points), on buying their own clothes (by 6.8 points), on household chores (by 4 points) even though they are less likely to solely take decisions on participation to social programs (by 3.2 points). They are also more likely to be involved in decisions on buying children s clothes (by 5.8 points), taking a loan (by 4.8 points), participating in community meetings (6.9 points), buying own clothes (6.8 points) or participating in social programs (by 3.6 points). Other indicators measured at the level of the beneficiary (table A4.8) also show that potential beneficiaries who selected into the program are more 12

13 educated and more likely to own assets compared to potential beneficiaries who decided to select out of the program. Also noteworthy is the fact that female beneficiaries themselves are more likely to own assets or credit, another source of evidence that self-selection primarily attracted more empowered and better-off women. These descriptive statistics reveal some strong differences in terms of who self-selects in the program. These results are important, because many productive interventions target based on-demand such as the FUMDEC pilot, but the implications of such targeting are not well-understood. The results suggest that better off, hence possibly more motivated, women are more likely to self-select in the program in this case. Another study from Nicaragua suggests that such targeting based on demand does not reach the poor, who turn out to be those who could benefit most from productive interventions (Macours, Premand and Vakis, 2013). The self-selection patterns in the FUMDEC program are consistent with lower demand for productive interventions among the poor in the Atención a crisis program. These patterns are consistent with the experience of the implementing NGO in communities where they had been operating for several years, which suggests that better-off and more empowered women are more likely to enroll in the program initially, while relatively poorer and less empowered women eventually join later after the program is more established in the community Economic Impacts 5.1 Program Impacts for Beneficiary Women Table 1 presents program impacts on a range of employment outcomes for the main beneficiary. The first set of indicators relates to employment in the last 12 months: self-employment in agricultural activities, self-employment in non-agricultural activities, employment in agricultural wage jobs and employment in non-agricultural wage jobs. Results show very limited changes in overall employment portfolios for beneficiaries at the extensive margin. One notable exception is that women in treatment communities are almost six percentage points more likely to engage in commerce (an increase of more than 40 percent relative to the control group). There are, however, some differences along the intensive margin, and in the composition of employment. Specifically, over the last year, beneficiaries work more days as self-employed in the commerce (12 days) as well as wage employed in private sector jobs (5), but fewer days as selfemployed in food production (13), manufacturing (2) and services (4). This is also partially reflected 13 In the context of the study, enrolment were closed after the initial enrolment period, and there was no new intake of beneficiaries until the follow-up survey. 13

14 in questions regarding employment during the last week, where beneficiaries report working longer hours in agricultural self-employment but fewer hours in agricultural wage jobs. Overall, these results suggest that the program did affect employment composition for beneficiary women, although the effects on overall employment are limited. In addition, the program led to some impacts on small household businesses. Beneficiaries are 5.4% more likely to sell non home-made items and 5.6 percent more likely to own a business. Of those who report that they have a business, 8.5% are more likely to report they expect to have their business in the future. Overall, these results are consistent with results above on changes in employment activities along the intensive margins. Table 1: Program impacts on Beneficiary s Employment Variable Description N Control Treatment Difference SE Last 12 Months Was self-employed in agricultural activity (0.032) Was self-employed in agricultural activity (including livestock) (0.011) Was self-employed in non-agricultural activity (0.031) Worked in agricultural wage job (0.021) Worked in non-agricultural wage job (0.018) Was self-employed in livestock (0.018) Worked as wage employed in agriculture (0.021) Worked as self-employment in food production (0.027) Worked as self-employment in manufacturing (0.006) Worked as self-employment in commerce *** (0.020) Worked as self-employment in services (0.011) Worked as wage employed in private jobs (0.012) Worked as wage employed in public jobs (0.011) Total no of days worked (excluding agricultural activity) (9.998) Total # days worked in non-agricultural self-employment (10.742) Total # days worked in agricultural wage job (1.271) Total # days worked in non-agricultural wage job (4.727) Total # days worked as self-employment in food production * (6.920) Total # days worked as self-employment in manufacturing * (1.315) Total # days worked as self-employment in commerce ** (6.113) Total # days worked as self-employment in services ** (1.869) Total # days worked as wage employed in private jobs * (2.598) Total # days worked as wage employed in public jobs (3.687) Last Week 14

15 Worked last week -including unpaid work (0.038) Hours worked (1.047) Hours worked as self-employed in agriculture ** (0.497) Hours worked as a wage employed in agriculture *** (0.225) Hours worked in non-agricultural self-employment (0.693) Hours worked as in non-agricultural wage jobs (0.535) Small Business Outcomes (last 12 months) Household has a small food production business (0.026) Household has a small trading business ** (0.023) Household has a small service business (0.013) Household has any types of business *** (0.022) Number of businesses owned by the household (0.067) Household still has this business (0.025) Value of equipment/tools (log) (0.217) Expects the business will run in December (peak season) *** (0.032) Table 2 presents program impacts on beneficiary income. The results show that beneficiary income increased by 39.6%. In absolute value, the beneficiary s income increased by C$1830 a year, or approximately $72. This increase in income is driven by large increases in income from livestock, (48.1%). Income from backyard agricultural production and commerce are not significant but positive (and large) relative to the control group. Program benefit packages included seeds to diversify agricultural production in households backyard, which is traditionally cultivated by women, as well as asset transfers including livestock such as goats or chicken. These results suggest that the changes in employment patterns documented above led to some productivity gains for the beneficiaries. These shifts are mostly observed through changes in occupations within agricultural activities (the lack of significant results for commerce and backyard production can also be due to the limited power of the study). Table 2: Program impacts on Beneficiary s Income (Logged) Variable Description N Control Treatment Difference SE Total beneficiary income ** (0.177) Total annual income from agricultural wage employment (0.137) Total annual income for non-agricultural self-employment (0.267) Total annual income for non-agricultural wage employment (0.145) Income from sale of titular's own animals ** (0.195) Income from backyard agricultural production (0.261) Income from temporary migration (0.195) 15

16 The program provided extensive training and technical capacity workshops for beneficiaries. Table 3 presents program impact on a range of business practices. There are a number of small but significant improvements in business practices among program beneficiaries. For example, beneficiaries are more likely to record their transactions and to separate business and household accounts. They are also more likely to report that they believe that profits should be reinvested in merchandise. In this sense, the program does seem to have provided beneficiaries with additional business skills and experience. Table 3: Program Impacts on Business practices Variable Description N Control Treatment Difference SE Business Practices How many customers do you have (1.759) Proportion of customers that are allowed to borrow (0.022) Are activities recorded *** (0.022) Titular records accounts ** (0.018) Do you calculate profits of main activity *** (0.034) Both inputs and sales in profit calculation (0.022) Separate accounts within the household * (0.026) Household and business money saved separately * (0.026) Are household accounts written down (0.018) Business Practice Perceptions It's important to diversify (0.022) It's important to practice accounting ** (0.019) Businesses fail because owners don't work hard enough * (0.019) Customer service can always improve (0.015) Businesses should discourage credit (0.029) Business profits should be reinvested in merchandise * (0.026) Table 4 presents program impacts on saving. Beneficiaries were 11.5 percentage points more likely to have saved in the past 12 months. The amount they saved was 33.8 percentage points more than the average amount saved in the control group (more than doubling savings rates). In addition, beneficiary women are 5.7% more likely to have savings separately from their spouses. Table 4: Program Impacts on Savings Variable Description N Control Treatment Difference SE Did you save in the past 12 months? *** (0.021) How much did you save last month? (log) *** (0.099) 16

17 Does titular have separate household savings from her savings? *** (0.019) The intervention seems to have helped women improve their access to credit. For example, beneficiaries are 20.5 percentage points more likely to have taken out a loan in the past 12 months (Table 5). 31 percent of women took a loan out of a communal bank set-up during the program. Beneficiaries are also 8.2% more likely to have taken a loan from other informal sources. Beneficiaries also are more likely to use the loans for income generating activities. Finally, they are 35.7% more likely to report that they desire to take a loan out in the coming months. Table 5 Credit Variable Description N Control Treatment Difference SE Did you use a seed bank in the last 12 months? (0.008) In past 12 months has anyone taken out a loan *** (0.035) Did you get loan from the communal bank *** (0.039) Did you get loan from elsewhere (formal) ** (0.040) Did you take a loan from an informal source *** (0.018) Total amount of loans taken out from communal bank *** (33.591) Total amount of formal loans taken out ** ( ) Total amount of informal loans taken out (14.805) Purpose of loan was agricultural activities *** (0.046) Purpose of loan was non-agricultural activities *** (0.024) Purpose of loan was for the household (0.015) Do you desire a loan in the coming months? *** (0.037) 17

18 5.2 Economic Impacts on the household In table 6 we report employment results for other adults in the household, excluding the beneficiary woman. The idea here is to test whether the program had an indirect impact on employment of other members of the household. In terms of employment, we do not find impacts except from agricultural self-employment where other household members are 3.4% more likely to work there in the treatment group relative to those in the control. This is also reflected in the hours worked last week where they spend an additional 1.6 hours more in these activities. As such, the within-household spill-overs on employment outcomes are limited. Table 6: Employment of household adults (excluding: the beneficiary and children less than 16 years old) Variable Description N Control Treatment Difference SE Last 12 Months Was self-employed in agricultural activity 1, (0.019) Was self-employed in agricultural activity (including livestock) 1, ** (0.016) Was self-employed in non-agricultural activity 1, * (0.015) Worked in agricultural wage job 1, (0.032) Worked in non-agricultural wage job 1, (0.013) Was self-employed in livestock 1, (0.021) Worked as wage employed in agriculture 1, (0.032) Worked as self-employment in food production 1, (0.007) Worked as self-employment in manufacturing 1, (0.002) Worked as self-employment in trading 1, (0.012) Worked as self-employment in services 1, (0.008) Worked as wage employed in private jobs 1, (0.011) Worked as wage employed in public jobs 1, (0.006) Total no of days worked (excluding agricultural activity) 1, (7.868) Total # days worked in non-agricultural self-employment 1, (4.287) Total # days worked in agricultural wage job 1, (6.404) Total # days worked in non-agricultural wage job 1, (1.804) Total # days worked as self-employment in food production 1, (1.822) Total # days worked as self-employment in manufacturing 1, (0.194) Total # days worked as self-employment in commerce 1, (2.637) Total # days worked as self-employment in services 1, (1.217) Total # days worked as wage employed in private jobs 1, (1.640) Total # days worked as wage employed in public jobs 1, (1.307) Last Week Worked last week -including unpaid work 1, ** (0.018) 18