86059 Building Women’s Economic and Social Empowerment Through Enterprise An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda April 2013 LOGiCA Study Series No.1 Christopher Blattman Eric Green Jeannie Annan Julian Jamison The International Bank for Reconstruction and Development / The World Bank 1818 H Street, NW Washington, DC 20433 All rights reserved First published November 2013 www.logica-wb.net This Working Papers Series disseminates the findings of work in progress to encourage discussion and ex- change of ideas on gender and conflict related issues in Sub-Saharan Africa. Papers in this series are not formal publications of the World Bank. The papers carry the names of the authors and should be cited accordingly. The series is edited by the Learning on Gender and Conflict in Africa (LOGiCA) Program of the World Bank within the Fragile States, Conflict and Social Development Department. This paper has not undergone the review accorded to official World Bank publications. The findings, interpreta- tions and conclusions herein are those of the author and do not necessarily reflect the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, its Executive Directors, or the governments they represent. To request further information on the series, please contact ereessmith@worldbank.org LOGiCA. Papers are also avail- able on the LOGiCA website: www.logica-wb.net. Cover and layout design: Duina Reyes Photos provided by United Nations photo library Building Women’s Economic and Social Empowerment Through Enterprise An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda LOGiCA Study Series No.1 April 2013 Christopher Blattman Columbia University Eric Green Duke University Jeannie Annan International Rescue Committee & Harvard University Julian Jamison Consumer Financial Protection Bureau with Filder Aryemo, Natalie Carlson, Mathilde Emeriau, and Alexander Segura, Innovations for Poverty Action Table of Contents Executive Summary................................................................................................................... 3 Acknowledgments..................................................................................................................... 5 1. Introduction......................................................................................................................... 6 2. Intervention and Research Design...................................................................................... 8 a. Context: Northern Uganda............................................................................................................8 b. WINGS Intervention.....................................................................................................................8 c. Research Design.............................................................................................................................9 3. Who are the beneficiaries and what do they do?............................................................... 14 a. Participant profiles......................................................................................................................14 b. Businesses proposed and pursued...............................................................................................15 4. Impacts of the core Intervention: Do economic and social empowerment go hand in hand?................................................................................................................ 17 c. Impacts on earnings and earnings opportunities........................................................................17 d. What is the distribution of poverty impacts?..............................................................................20 e. Who succeeds?.............................................................................................................................21 f. Health and social impacts.............................................................................................................22 5. Are these impacts “high”? A cost-benefit analysis............................................................. 28 a. From “impact” to relative return”................................................................................................28 b. Estimating the returns to the WINGS program..........................................................................28 c. The returns to follow-up..............................................................................................................31 d. Conclusions.................................................................................................................................31 6. What are the effects of the WINGS programs on other village members, especially existing traders and entrepreneurs?................................................................. 32 7. Do supervision and mentoring improve performance? The effectiveness (and cost-effectiveness) of follow-up................................................................................. 34 a. The potential gains from follow-up: The “accountability” and “advice” effects..........................34 b. Distinguishing accountability from advice.................................................................................34 c. Short-term impacts of follow-up.................................................................................................35 d. Longer term (one-year) impacts of follow-up.............................................................................38 e. Is follow-up cost effective?...........................................................................................................42 8. The effect of building social and group networks.............................................................. 43 An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 1 9. Does Male Involvement Promote Women’s Empowerment and Well-Being?................................................................................................................. 46 a. Underlying theory of change.......................................................................................................46 b. Did W+ impact the couple’s communication and relationship?.................................................46 c. Did W+ impact the partner’s direct and indirect support for the business?...............................48 d. Did W+ impact the household’s economic security?..................................................................49 e. Did W+ impact the woman’s health and empowerment?............................................................50 10. Discussion and Conclusions............................................................................................. 52 References............................................................................................................................... 58 Building Women’s Economic and 2 Social Empowerment Through Enterprise Executive Summary I nvesting in women is said to be a key to devel- opment. Educate her, buy her a cow or goat, or help her start a business and great things will follow: sustained increases in income, greater em- savings tripled, and short-term expenditures and du- rable assets increased 30 to 50% relative to the con- trol group. While the absolute changes seem small in magnitude, these are huge gains relative to where powerment and social inclusion, health and educa- these women start. tion for the children, and (especially in war-affected regions) mental health and happiness. The treatment is most impactful on the people with the lowest initial levels of capital and access to credit. Testing whether this is true will take a great many This is largely consistent with economic theories of studies and interventions. In this report we study the poverty that argue that the poor have the potential impacts of giving cash grants of approximately $150 to be productive but are constrained by an absence and basic business skills training to the very poorest of cheap capital. The most constrained thus tend to and most excluded women in a war-affected region, benefit most from treatment. Other factors that we northern Uganda. The program was designed and might worry would inhibit success—low education, implemented by an Italian non-governmental or- high levels of emotional distress, or poor health— ganization (NGO), AVSI Uganda, with decades of seem to have little association with the impact of the experience serving this population. intervention. 1800 poor young women (and some men) in 120 Second, although these results suggest the program villages were randomly assigned to a first or second leads to relatively large increases in income and phase of the intervention, allowing us to assess the wealth, we see no effect on women’s independence, impacts after roughly 18 months. In each phase we status in the community, or freedom from intimate also vary core program components—organizing partner violence (though, importantly, the program women in some villages into groups, varying the does not increase a woman’s probability of experi- degree of supervision and advising they receive, and encing partner violence). Perhaps economic success varying the level of involvement of the husband. and empowerment are not closely linked, at least in the short run, for poverty impacts of our magnitude. This report provides provisional answers to these Likewise, we see little evidence that a more inclu- questions based on data collected from April 2009 sive role for males in the household leads to better to August 2012. The questions will continue to be empowerment or economic success, although we explored and analyzed in academic papers in future, see promising improvement in partner support and but we attempt to draw out the key findings and relationships. policy lessons as close to the end of the intervention and data collection as possible. Third, we see little effect on psychological or social well-being from this reduction in poverty. This is First, we see dramatic increases in business and congruent with other experimental studies, includ- reductions in poverty. The women were encour- ing employment programs in northern Uganda, that aged by the NGO to begin with retail trading and show little short-term connection between poverty goods, and most start and sustain small retail busi- relief and either social support or symptoms of dis- nesses with the capital they receive, while continu- tress. ing their farming and other miscellaneous activities. A year after the intervention, monthly cash earnings Fourth, close supervision and advising by the NGO doubled from 16,500 Uganda Shillings (UGX) to leads to slight increases in economic success. Pat- 31,300 ($6.60 to $12.52 U.S. Dollars; USD), cash terns of grant spending shift very little, marginally An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 3 away from durable asset investment into business ings rise. Since they spend some of the grant do- expenditures. This suggests some effect of “account- mestically, demand for locally-produced goods also ability”. Longer-run incomes are higher by about $2 rises, increasing incomes as well. Households which (UGX 5,000) monthly, concentrated among higher are already trading, however, experience more com- earners (the median impact is less than half the av- petition and falling profits, possibly because of a erage impact). There is little difference in capital reduction in market power. Thus net consumers investments. This suggests that the advising aspect tend to benefit from the intervention along with the of the visit may have provided value. We see little direct beneficiaries, and net producers (or traders) psychosocial impact, however, suggesting the gains tend to lose out. are mainly economic. The economic gains for some people are large enough that, if they are sustained Overall this program seems most effective at poverty over a long enough time horizon, they may justify alleviation, and organizations looking to empower the cost of providing this intensive attention. The women or reduce exclusion will need to experiment impact, such as it is, is most apparent for one to two with alternative approaches. To maximize the pov- rounds of supervision. Supervision and advice be- erty alleviation impacts for the most people, inter- yond this first or second visit does not seem to have ventions like this one must strive to become more an economic impact and so probably does not pass cost effective. Costs of disbursement, targeting and a cost-benefit test. follow-up should be streamlined so that they are less Fifth, we see large spillover effects into these small (even far less) than the grant size. Components like village economies. With most women becoming business training should be evaluated more rigor- traders, imports from major trading centers increase, ously. A straight up comparison of these additions and the price of consumer goods fall. This raises the to their value in cash is an important study for the spending power of all households and so real earn- entire humanitarian sector. Building Women’s Economic and 4 Social Empowerment Through Enterprise Acknowledgments W e thank AVSI Uganda and AVSI USA for their cooperation and long partnership in designing and implementing this pro- gram. We are especially thankful to Jackie Aldrette, Ezio Castelli, Federico Riccio, Francesca Oliva, Fa- Trust, the Learning on Gender and Conflict in Af- rica (LOGICA) Trust Fund at the World Bank, and Yale University’s Institution for Social and Policy Studies. bio Beltramini, John Makoha, Samuele Rizzo, Filip- For research assistance, we especially thank Filder po Ciantia, Massimo Zucca and Francesco Frigerio. Aryemo, Natalie Carlson, Mathilde Emeriau, Sara Lowes, Lucy Martin, Godfrey Okot, Alexander Se- We gratefully acknowledge research funding and gura, and Christian Lehmann. support from AVSI Uganda, a Vanguard Charitable An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 5 1. Introduction T his study investigates an attempt to eco- nomically and socially empower some of the poorest and most vulnerable young women in one of the poorest and most fragile places ment. Unfortunately, conflict and crisis-affected countries are not all that rare. Also, in most low and middle-income countries (conflicted or not), the very poorest are similar in many ways to the very in the world: northern Uganda. poorest in northern Uganda. They are often young people, more often than not female, who have few We experimentally evaluate the intervention to an- skills and capital, little prospect of earning a living, swer a series of bigger questions in aid and develop- and, to the extent they are dislocated from their fam- ment: ily or community, have little of the support needed First, does providing inputs like cash and business to survive another bad shock. skills help the very poorest build sustainable sourc- Governments and aid organizations in nearly every es of income? If so, what does this tell us about the country deal with women and men like this every causes of poverty and the constraints holding back day, and how to empower them—literally, to in- the poor? crease their economic and social strength—is a con- Second, does such economic empowerment (like stant question with few clear answers. reduced poverty and sustainable businesses) lead Perhaps the most common approach is to give poor to broader forms of empowerment (like increased people “inputs”—cash, capital, microfinance, skills, independence and reduced risk of violence)? How and other things that go into production. These important is the support and inclusion of important might be small farms, or self-managed “micro” enter- men in a woman’s life, such as partners, fathers, and prises. Such interventions make a couple of crucial brothers? assumptions. One is that the very poor can use these Third, does economic empowerment have social and inputs productively—that there are high returns psychological benefits as well? That is, do poverty to these inputs. A second is that they are somehow relief and sustainable new businesses increase social constrained from obtaining these inputs in the ab- support and decrease psychological distress among sence of the intervention. If one or both of these as- a war-affected, relatively excluded population? sumptions are false, then we should see little change in poverty or empowerment from the endowment Fourth, does supervision and mentoring lead the of inputs. poor to spend or invest differently, and what is the effect on general economic success and poverty re- These are probably fair assumptions to make, or so duction? Should NGOs and governments leave the more and more investigation suggests. Growing evi- poor to decide for themselves how to spend aid, or dence implies that, at least on average, the poor can would the aid be more successfully used with ac- earn high returns to cash, capital, and new skills. This countability and advice? is not altogether surprising if one believes in steep returns at low levels of initial resources, but would Fifth, in small village economies, what are the ef- be more surprising if there were deep complemen- fects of cash transfers on those who don’t receive the tarities across physical, social, economic, and cog- grants? What effect does increased business com- nitive capital. Moreover, the poor are undoubtedly petition and trade have on the welfare of other con- constrained. Business investments and finance are sumers and producers in the village? especially scarce and expensive, partly because fi- Northern Uganda might seem like a special case, nancial markets are grossly underdeveloped. A great emerging from twenty years of conflict and displace- many forces conspire to constrain the poor. Building Women’s Economic and 6 Social Empowerment Through Enterprise More evidence is needed, especially among the economic and social empowerment. poorest, and especially among women. This study adds to that body of evidence, examining a program Specifically, we try to address the following ques- that not only provides grants and skills training to tions: the poorest of the poor, but also intensive advising • Can a program of business skills training, cash and supervision. grants, and intensive follow-up help some of the Advice and supervision is common in interventions poorest women build sustainable enterprises targeting the poor. People rich and poor alike make and raise their earnings? If so, what does this tell bad or shortsighted decisions with their money, es- us about the roots of poverty and how to allevi- pecially (it is feared) with large windfalls sitting in ate it? their pockets. Even the most disciplined may face • Face-to-face meetings with program staff and family or social pressures to share funds, or to pay social workers are often an integral part of such for pressing but short-term needs. These concerns programs. Follow-up can provide accountabil- spawn programs that range from the lightest of ity and give clients the incentive to invest their nudges to the heaviest paternalism. There is little ev- grants rather than “eat” them or dissipate them idence on what level of supervision, if any, is needed, and whether self-enforcement at the group or the in- through the kin network. Training and follow- dividual level is possible. up also impart ideas and advice. But this face- time is expensive. Is it cost-effective? And is an Finally, we are seldom concerned narrowly with intensive and face-to-face approach consistent economic empowerment alone, but also empower- with the goal of empowerment? ing people more generally—improving their ability to access the fundamental elements of development: • Does this poverty alleviation lead to social and happiness, health, education, rights, and social and psychological gains? Or, put another way: does political participation. To what extent do anti-pov- economic empowerment lead to social empow- erty programs not only empower people economi- erment? cally, but socially and psychologically as well? In the case of women in male-dominated societies, is the • What role does the social group and the house- best way to achieve this with or without the coop- hold, especially the spouse, play in women’s eration of men? economic success? Can approaches that build group or household cohesion, or that include This study tackles these questions in the context the spouse, increase economic and social suc- of an experimental project with some of the poor- cess and empowerment? est young people, primarily women, in northern Uganda. We collaborated with a non-governmental • Looking even more broadly, what are the effects organization, AVSI Uganda, to study their integrat- on other villagers of giving cash to a subset of ed approach, testing their package of interventions them? In a small economy, does encouraging experimentally as a whole, but also experimentally multiple new businesses have negative (or posi- varying some of the key components of the package tive) spillovers for existing businesses? to answer some of these more general questions of An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 7 2. Intervention and Research Design a. Context: Northern Uganda and the research questions above, we worked with AVSI on the design, implementation and evalua- Twenty years of war and widespread displacement tion of two new phases of the program in 2009 and have left the majority of the population of northern 2011: the Women’s Income Generation Support, or Uganda impoverished. Social networks that tradi- WINGS, program. tionally cared for the most vulnerable in this region are greatly overstretched. Those marginalized from b. WINGS Intervention kin and their communities are at-risk both economi- cally and socially. It will take many years for house- There are three components to the core WINGS holds to build up assets and livestock and achieve program: (i) a few days of business skills training pre-war levels of productivity and income. A major (BST), (ii) an individual start-up grant of roughly worry is that the most vulnerable households will $150, and (iii) regular follow-up by trained com- not be able to develop and maintain livelihoods and munity workers. Optional components include (iv) income without assistance that targets their specific group formation, training and self-support; and (v) needs, including provision of skills, capital, and so- spousal inclusion, training and support. A brief de- cial networks. scription of each follows. Young women and girls in particular have suffered economically and educationally from the war. In Business Skills Training 2007 AVSI and two of the IPA Investigators sur- AVSI provides a brief course in basic business skills veyed more than 600 young females aged 14 to 35 for all participants. This course typically runs for five affected by the conflict in northern Uganda, includ- days and covers topics necessary for the planning, ing more than 200 women formerly abducted by the starting, and managing of simple business activities. armed group. The evidence from the survey, along The curriculum has been adapted for illiterate users with program experience among NGOs in north- and AVSI staff is experienced in effectively working ern Uganda, suggests that the development of new economic opportunities and building social capital with illiterate beneficiaries, who are the majority of will be crucial ingredients in reducing poverty and the target group of this proposed program. Trainers improving the health, education and psychosocial are AVSI staff members with years of experience in well-being of youth. Young women, especially those the psychosocial and livelihoods sector and with with children or orphans to care for, are in most need specific training in business skills, group dynamics of such livelihoods assistance. This includes a dis- and problem solving within the world of business. proportionate number of formerly abducted young Clients submit business plans to the AVSI team af- mothers, most of whom do not return to school. ter the training. Each plan is reviewed and discussed with the client. Upon approval, the client is eligible An international NGO, AVSI Uganda, has been ac- for a start-up grant. tive in Northern Uganda for almost three decades. Over the last 7 years, AVSI developed and refined an Start-Up Grant economic assistance program that targets the most vulnerable members of the community and provides All clients receive a start-up grant of approximately them with extensive psychosocial services and social $150 USD to be used for the implementation of ap- networks alongside business skills and grant (rather proved business plans. In the past, this money has than credit-based) assistance. To assess the program been disbursed all at once and in several tranches. Building Women’s Economic and 8 Social Empowerment Through Enterprise Follow-up tic view of a person’s needs and resources. Typically this leads the organization to prioritize working AVSI understands that the importance of follow-up with households and families rather than individu- visits to the individual and the groups is important als whenever possible. Prior to this evaluation, AVSI from two sides: the inter-personal and the business had allowed individuals to participate in their live- dimension. Many years of experience have demon- lihoods programs with partners, but such inclusion strated to AVSI that on-going support for young, was not systematic. As we describe in a later section, new entrepreneurs is essential to help them succeed AVSI formalized their approach to working with and address the challenges that arise with every partners for the purpose of this evaluation and cre- nascent business endeavor. Clients receive at least three follow up visits. ated new training materials and a follow-up proce- dures to enable them to better support individuals On the business side, AVSI staff maintains close participating in the program with a partner. supervision of business activities for the first few business cycles, providing advice on meeting mar- c. Research Design ket challenges and implementing sound business practices. AVSI staff have been trained in business The WINGS program and evaluation began in Janu- skills and most importantly have years of experience ary 2009 and concluded in 2012. At the start of the within the environment of small enterprises in the program, AVSI worked with leaders in 120 commu- specific geographic districts of this program, with nities in Gulu and Kitgum districts to identify and accumulated links to successful businesses and an screen 2,300 potential beneficiaries. Following this array of formal and informal financial services. initial assessment, AVSI selected 1,800 of the most vulnerable residents between the ages of 14 and 30 Group Training (86% female), approximately 15 per program com- While not a part of the core WINGS intervention, munity. AVSI has the ability to help individual entrepre- neurs in the same community form business sup- The empirical strategy for the evaluation consisted port networks to enable them to effectively share in- of a randomized experimental design and mixed- formation and ideas, to collaborate in activities like methods data collection. Following the baseline savings and investment, and (possibly) to reinforce survey with all 1,800 beneficiaries in mid-2009, farsighted investment decisions and discourage IPA held public lotteries in Gulu and Kitgum to shortsighted consumption. Group support of this randomly assign the 120 program villages to Phase nature is a common feature of savings, enterprise 1 or Phase 2 (stratified by district). In this wait-list development, and microfinance programs partly for control design, all beneficiaries were guaranteed to these reasons. receive the program, but not all at once. By serving 900 beneficiaries per phase, AVSI had to scale-up AVSI has developed a Group Dynamics facilitator’s their program by 300 percent. Therefore, it was not manual based on years of experience in Uganda. possible to serve all intended beneficiaries at once. The manual addresses topics such as the purpose and usefulness of group participation, qualities and Phase 1 started in mid 2009 and Phase 2 began in selection of group leaders, communication skills, early 2011 following the endline survey of all 1,800 record keeping and evaluation of progress; these beneficiaries in November 2010. By comparing the topics are approached through interaction with the beneficiaries in Phase 1 to those in Phase 2, who had participants and frequent small group work and ac- not received the program at the time of the endline tivities. It is one of the added components we will survey, we were able to estimate the medium-run evaluate. impacts of the program on our core outcomes of Spousal inclusion and training interest: sustained livelihoods, poverty, empower- ment, gender-based violence, family education and A central tenant of AVSI’s approach is to take a holis- health, and psychosocial well being. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 9 In addition to the pre and post surveys of all benefi- To the best of our knowledge, the economic and ciaries, the IPA team also conducted in-depth quali- social impacts of follow-up support to recipients of tative data on the process of business development; economic assistance programs like WINGS has not administered additional experimental behavioral been rigorously evaluated. The most relevant litera- economic activities (Interactive Behavioral Mea- ture may be the role and impact of loan officers in surements, or IBMs) to measure beneficiaries’ risk microfinance programs, but the evidence base is very and time preferences that may impact business deci- limited (Siwale & Ritchie, 2011). Given the logisti- sion-making and success; and completed a commu- cal challenges and high cost of facilitating multiple nity survey with non-participants to measure com- home visits and monitoring, it is important to dem- munity-level impact of the program on market prices onstrate the cost-effectiveness of this component and existing entrepreneurs. The qualitative work fed of the program. In the second phase of the study, largely into the research design in Phase 2. Addition- beneficiaries were randomized to receive 0 follow- al qualitative work, and analysis of the IBMs, will be up visits, 2 visits, or 5 visits to estimate the effect of incorporated in future academic papers. follow-up ‘dose’. In addition to studying the impact of dose, we also attempted to tease apart the mecha- Program Variations nism of follow-up impact—‘accountability’ versus longer-term advising and relationship building—by Support for Business Networks. AVSI’s experience examining differences in early spending decisions with this program model suggested that the target based on beneficiaries’ expectations of follow-up. women lack support networks that they could use for business advice, savings and lending, and other Household Approach. The link between economic support. Development programs commonly form assistance programs and women’s empowerment is villagers, especially women, into groups for this pur- mixed. Women’s income tends to increase and ben- pose. It is universal, yet untested. We wanted to test efit household members, particularly children, but whether this was an effective way to increase success women targeted for assistance do not consistently and well-being. Therefore in the 60 Phase 1 villages report increased empowerment, such as greater we instituted a cross-cutting design (CCD), where independence from their male partners, increased women in 30 of the Phase 1 villages were encour- control over household resources, or more partici- aged to form a mutual support group, elect a leader- pation in household decision-making. Increasingly, ship, and hold regular meetings. The groups received researchers, donors, and practitioners have begun to two days of advising and team building exercises. focus on the role of men in women’s empowerment When we conduct the interim survey at the end of (DAW, 2003; Sternberg & Hubley, 2004), but rigor- Phase 1, we will be able to measure the impact of ous evaluations of interventions involving men are these women’s support networks on all of our out- still rare. comes of interest. As we describe in the next section, our main find- Follow-Up Dose. Program experience suggests that ings for the standard WINGS package support this on-going support for young, new entrepreneurs “impact-paradox”1: targeting a vulnerable woman to is essential to help them succeed and address the be the recipient of an economic assistance program challenges that arise with every nascent business benefits the household financially, but on average endeavor, but it is not clear that close monitoring is does not empower the woman or improve her well- cost-effective or essential to business success. AVSI being in any measurable way in the medium-term. staff maintains close supervision of business activi- This finding led the research and program teams to ties for the first few business cycles, providing advice on meeting market challenges and implementing sound business practices. In addition to these eco- 1  Garikipati, S. (2008). The Impact of Lending to Women nomic objectives, the follow-ups were conceived as on Household Vulnerability and Women‘s Empowerment: a means of counteracting the relative marginaliza- Evidence from India. World Development, 36(12), 2620- tion of the target group. 2642. doi:16/j.worlddev.2007.11.008 Building Women’s Economic and 10 Social Empowerment Through Enterprise design and evaluate a slight reframing of the inter- Phase 2 villages to participate in either the standard vention from an individual-approach to a more ho- WINGs program or the W+ variant. Program ben- listic household-approach targeting the woman plus eficiaries in villages assigned to W+ could invite an an important household member. We called this important household member to join them at the new version of the program “Women Plus”, or W+ initial business skills training; beneficiaries in con- for short. trol villages participated as individuals. While we did not require W+ beneficiaries to participate with We wanted to test the hypothesis that we could a partner or dictate who that partner had to be, most maintain (or increase) the positive household effects women participated with husbands, male compan- observed when targeting the woman while also hav- ions, or other important male figures; the male par- ing a positive impact on her sense of empowerment ticipants in W+ villages could also attend the train- and well-being. So we asked every pending pro- ing with a partner. gram recipient to identify an important household member who could participate in the initial phases W+ program beneficiaries and their partners re- of business development—from the business skills ceived the same training as beneficiaries in control training through the process of proposing, launch- villages over the same number of days, but W+ teams ing, and growing the business in the medium term. also completed additional training modules dur- The idea was to support the woman as the principal ing this time that focused on communication, joint business owner while engaging a key member of her problem-solving, and gender relations. We added household who could provide direct or indirect sup- this aspect of the training for W+ because the results port for the business. of a heterogeneity analysis supported our qualitative finding that women with more supportive partners To test this hypothesis, we randomized the 60 at baseline were more successful in the program. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 11 Figure 1. Flow diagram for Phase 1 Building Women’s Economic and 12 Social Empowerment Through Enterprise Figure 2. Flow diagram for Phase 2 An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 13 3. Who are the beneficiaries and what do they do? Who is eligible for the intervention, what do they and 35, with little or no formal education, low in- do prior to the intervention, and how do occupa- come and limited access to credit. tions change? We review some pre-progrm (base- Farming activity, followed by casual labor, are the line) data on the treatment and control group before main econmic activities before the intervention. turning to impacts. Relatively few women engage in trading or trades. Figure 3 below shows the frequency of various ac- a. Participant profiles tivities at baseline (the number of people, out of the sample of 1,800, who answered “yes” to the follow- As shown in Table 1 below, the typical WINGS can- ing questions regarding their activities in the past didate was a young woman between the ages of 20 four weeks). Table 1. Summary statistics of population at baseline Gender Male 13.8% Female 86.1% Age Under 20 13.9% 20-24 20.1% 25-29 28.3% 30-34 21.0% 35+ 16.7% Education Percentage with no formal schooling 39.7% Percentage with 8 or more years of schooling 3.6% Mean years of schoolin 2.8 Economic USD Mean cash earnings (UGX, past 4 weeks) 8,914 $3.58 Mean savings (UGX) 4,839 $1.94 Percentage with access to UGX 15k loan 23.8% Percentage with access to UGX 100k loan 4.1% Farming activity, followed by casual labor, are the main econmic activities before the intervention. Relatively few women engage in trading or trades. Figure 3 below shows the frequency of various activities at baseline (the number of people, out of the sample of 1,800, who answered “yes” to the following questions regarding their activities in the past four weeks). Building Women’s Economic and 14 Social Empowerment Through Enterprise Figure 3. Number of people who reported having done each activity in the past four weeks at baseline ",-./"012"62B"<8"0124"138"B-46/8" #*'" N<8=;26<8B"=-7A"=41:7O>" $!%" *($" ",-./"012"62B"<8"719/18/M7"B-46/8>" #&*" '((" ",-./"012":/4H149/6"=-72-;";-E14>" &!#" %+*" ",-./"012"E4/3/6"-;=1A1;PE//4>" %#)" ",-./"012"C-5/8"=-4/"1H"N4-<7<8BO"0124"138" )$" -8<9-;7>" ##" ",-./"012":24=A-7/6"" )&" ')" ",-./"012"A28C/6>" '(" ",-./"012"C-5/8"=-4/"1H"719/18/"/;7/M7" %&" -8<9-;7>" '%" %&" ",-./"012"618/"L2-440"3145>" %&" Q4/-C9/8C" R18C41;" +" ",-./"012"9-6/"E4<=57>" $" ",-./"012"3145/6"-7"-"A/-;CA"14"IJK" )" 3145/4>" !" ",-./"012"G7A/6"14"3145/6"18"-"G7A" &" H-49>" #" &" ",-./"012"/8B-B/6"<8"918/0";/86<8B>" !" ",-./"012"A-6"-"B1./489/8C"D1E"14" '" :1;" !" %" ",-./"012"3145/6"-7"-"?16-@?16-"64<./4>" !" ",-./"012"3145/6"-7"-8"/9:;10//"<8"-" !" =19:-80>" !" !"#$%#&'() Figure 3 looks at the frequency, not the intensity of Figure 4. Hours spent on activities. Figure 4 below shows the breakdown of economic activity at baseline economic activity by hours at baseline, for all par- Formal Brewing Money Selling, Hunting, 0.5% Other ticipants (treatment and control). It tells a similar alcohol/ lending, 2.6% unskilled, 2.4% Sector Wage Work, 0.7% beer, 3.3% 0.1% story to the frequency chart above, in which the larg- Animal raising for est percentage of time is spent on farming and sub- others, 3.9% sistence work. Farming, animal raising and casual Animal raising for self, 12.6% labor make up approximately 90% of hours spent on Farming for self, economic activity. 41.8% Casual Labor, 10.5% b. Businesses proposed and pursued What types of enterprise did people pursue? We Farming for others, look at the activities proposed by Phase 1 benefi- 21.8% ciaries and also their overall pattern of economic activities eighteen months after treatment. We see a An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 15 substantial increase in trading activities. ries are roughly 50 percentage points more likely to be engaged in trading—61% of treated people are Business plans proposed by WINGS recipients were engaged in trading versus 12% of controls. Figure 7 then received and approved by AVSI. Figure 5, taken displays these means, calculated on a series of ques- from an AVSI presentation, shows the breakdown tions concerning whether participants had regularly of these approved businesses by type. Over half the sold fruits, vegetables, livestock, livestock products, businesses centered on the general selling of mixed grains, or groceries that they had either primarily or items, with the rest being dominated by selling of exclusively purchased for resale. livestock, fish and farm prod- ucts. Figure 5. Business approved for AVSI recipients Following the intervention, we see a dramatic shift in hours $#!% "#+% spent selling items (from 2.6% $#*% up to 22.7% of time) and an increase in animal raising and ,-./01-2/34 farming for oneself (below, $$#)% 5-4614/77-89 in Figure 6). This is likely the :;<0=>/ direct impact of the new busi- !"#$% ?-@/42<>A nesses started by the WINGS beneficiaries, in which they B/9/2CD7/4 were selling produce, livestock E26/;1 or various items. &'#(% Eighteen months after the in- tervention, Phase 1 beneficia- Figure 6. Hours spent on Figure 7. Percentage of group buy- economic activity at midline ing and reselling at midline Trade work, 4.4% "&&#$ Hunting, 0.5% Other (&#$ unskilled, !"#$ 8.1% !&#$ Farming for self, 20.6% '&#$ Farming for others, Selling, 22.7% 4.7% %&#$ "%#$ Casual Labor, 6.2% &#$ )*+,-.+/-$ 01/-*12$ Animal raising for self, Money lending, 0.1% 28.1% Brewing alcohol/ beer, 4.9% Formal Sector Wage Work, 0.9% Animal raising for others, 3.3% Building Women’s Economic and 16 Social Empowerment Through Enterprise 4. Impacts of the core Intervention: Do economic and social empowerment go hand in hand? We begin with a simple comparison of the initial ment regressions rather than the simple difference Phase 1 group to the Phase 2 group, roughly 18 between the treatment and control group). months after Phase 1 received the training and grant, but before Phase 2 started the program. c. Impacts on earnings and earnings opportunities Table 2, below, lists a number of indicators of eco- nomic well-being and the average levels among An important goal of the WINGS program is to help those assigned to the WINGS program and those beneficiaries create small businesses that generate assigned to the control group (Phase 2), roughly 18 earnings and earning opportunities for the women months after the start of the intervention. The dif- or household. The most important goal is to reduce ference between these two averages is the impact of poverty and the extreme deprivation of being on the the program. In figures below, we report the level poorest end of an already extremely poor popula- and proportional change this average treatment ef- tion. fect represents (using estimates from average treat- There are a number of ways to Table 2. Indicators of economic well-being in the measure poverty, including cash control andtreatment (assigned to WINGS) groups earnings, the money that house- Indicator Control Treatment holds actually spend on goods Net cash earnings in the past 4 weeks and services (i.e., short term con- 16.5 31.3 (000's of UGX) sumption), and the stock of more Household short-term spending (000s durable assets and other forms of 35.6 46.5 of UGX) wealth (an asset or wealth index). Wealth index (0 to 1) 0.4 0.5 Throughout this report we will rely on all three. Total hours of employment in the past 59 92 4 weeks Comparing the treated to the Total hours spent on chores in the (temporary) control group, the 158 161 past 4 weeks WINGS program had a very large Total hours spent on subsistence work effect on cash earnings relative to 30 41 in the past 4 weeks the control group, essentially dou- Total hours spent on market activities bling individual incomes among 29 50 in the past 4 weeks program recipients compared to Value of transfers out of the the control group. (Note: This household since Christmas (000s of 9.3 18.7 cash income excludes income UGX) that may be received in kind, such Outstanding loans (000s of UGX) 5.7 9.3 as from subsistence farming.) For Perceived access to credit (index) 0.8 1.0 the average WINGS beneficiary, net cash earnings increased UGX Savings (000s of UGX) 40.7 163.4 16,211 in the month before the An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 17 Figure 8. Average treatment effects on poverty 250% Average Trearment Effects in 200% Percentage Change 174% 150% 130% 100% 98% 50% 43% 33% 29% 0% Net cash Net cash Household Household Wealth index Wealth index earnings earnings short-term short-term (average) (50th past 4 weeks past 4 weeks spending spending percentile) (average) (50th (average) (50th percentile) percentile) Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. survey, a 98% increase over controls (See Table 2 their monthly income increased UGX 7,800 (equiv- and Figure 8).2 alent to USD 3.13 at the 2010 market exchange rate). The median control group member earned just UGX In absolute terms, an increase of UGX 16,211 does 6,700 (USD 2.69) a month, so you could read the not seem large (about $6.50 at market exchange median treatment effect as a doubling of incomes rates of 2500). Relative to the average income in the at the median (Figure 8). Again, however, note that control group it is quite substantial, however. The the absolute levels of change are fairly small. low absolute level of profits should be kept in mind when looking at these large relative changes. Income is not the only way to measure material well- being. Net income is volatile, and may not represent This average impact is influenced, and possibly actual change in poverty and material well-being, skewed, by a small number of people who do very because it could be temporary, or shared with others well. For instance, at the 99th percentile, we have or saved for later in life. Thus economists typically people in the treatment group who earn as much as UGX 288,000 in a month, perhaps because of an un- use a measure of “consumption” as a measure of pov- usually good month. To avoid drawing conclusions erty. We have two measures of consumption: about the whole program from high-performers, we i. Household short-term spending: This is can also look at the median treatment effect, which an estimate of the household’s short term is how the person in the precise middle of the treat- spending per person on food and non-food ment group performed. For that median person, items in the previous month. ii. Household wealth index: This is an index, 2  About half the income gains come from additional work; scaled 0 to 1, of the household’s durable beneficiaries use the capital and skills they receive to increase assets (e.g. furniture) and housing quality, hours of employment. Thus hourly earnings increase by and so reflects wealth and potential for long about half. This suggests that the program enables people to increase productivity. This productivity may come from term consumption and spending. the new business skills, or it may represent returns available all along, but that an absence of credit or capital kept them Our household spending measure increases UGX from achieving. 11,741 (USD 4.72) from the program, which is a Building Women’s Economic and 18 Social Empowerment Through Enterprise Figure 9. Average treatment effects on employment 115% Average Trearment Effects in Percentage Change 95% 79% 75% 61% 55% 41% 35% 15% 1% -5% Total hours of Total hours spent on Total hours spent on Total hours spent on employment chores in past 4 subsistence work in market activities in in past 4 weeks weeks past 4 weeks past 4 weeks Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. 33% increase over the control group (Figure 8). grant and from ongoing higher incomes mean that Both the consumption and income impact are clear- the client’s labor is more productive, and so they ly statistically different from zero, meaning we are increase hours of work. This interpretation rests on assured of a positive impact. The median treatment the idea that the clients were capital constrained be- effect is fairly similar in proportion. As with income, fore the grant of cash—they could earn high returns however, the absolute change is smaller. from added capital, if they had it, by growing the size of their enterprise and working more at it. Capital We also see an increase in wealth. The absolute value and labor are complements, and so when the cash of the index does not have an easy or natural inter- is received the clients respond by increasing their la- pretation. It is an approximate ranking of house- bor to use the new capital as effectively as possible. holds. The average and median treatment effects are Economic theory suggests this rise in labor sup- large and positive, and statistically robust, imply- ply is countered, somewhat, by an “income effect” ing that WINGS clients substantially increase their whereby wealthier people want to consume more durable assets relative to the control group. We can of everything, including more leisure time. For the interpret these results as saying that the increased in- capital constrained, this income effect is probably come from the intervention is channeled largely into swamped by the incentives to increase work. short-term as well as durable consumption, raising standards of living. Figure 10 displays financial impacts. Savings triples on average, going from UGX 40,740 (USD 16.36) From Figure 9, we also see that hours of employ- to UGX 169,862 (USD 68.22). Some income is also ment increase substantially. WINGS clients do not transferred to other household members and out- change the hours they spend on chores, but they side the household. Our consumption measure in- do increase their labor supply in earning activities, cludes spending inside the household on education either agriculture (which we classify as subsistence and health. But it does not include transfers outside work) and also market activities (including their en- the household, which were UGX 9,734 (USD 3.90) terprise, plus any other enterprises and wage work). more among the treated than the controls, a 105% One interpretation is that the added capital from the increase over controls. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 19 Figure 10. Average treatment effects on financial outcomes 400% Average Trearment Effects in Percentage 350% 319% 300% 250% 200% Change 150% 100% 105% 67% 50% 26% 0% Value of transfers out Outstanding Loans Perceived Access to Savings of the Household Credit since Christmas Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. eryone was well served by the program. Some had d. What is the distribution of poverty enormous potential, good fortune, and succeeded impacts? tremendously. But the return on investment was very low for many people—so low that we may get We have talked about the average and the median into the area of such low returns that the program impact, but in fact there is a great deal of variation does not serve them merely as well as some alterna- within the treatment group. For instance, while the tive, such as simple cash handouts, or a more effi- median increase in income (i.e. the treated person cient program, or a program with alternative design at the 50th percentile) is UGX 7,800 (USD 3.13), (see Section 4). the person at the 20th percentile saw an increase of just UGX 1,700 (USD 0.68) Figure 11. Net income in the past 4 weeks QTE and the person at the 80th percentile saw an increase of UGX 20,500 (USD 8.23, Figure 11). Income is quite volatile from month to month, and so what we want to look at is a measure of more “permanent” income, like consumption. At the median con- sumption increased UGX 4,601 (USD 1.85), at the 20th percentile it increase UGX 4,225 (USD 1.70), and at the 80th percentile it increased UGX 6,806 (USD 2.73, see Figure 12). The good news is that the program gen- erally had a uniformly positive impact on income and consumption—virtually ev- Building Women’s Economic and 20 Social Empowerment Through Enterprise a regression of success on a treatment Figure 12. Household consumption QTE indicator, the baseline value of the char- acteristic, and an interaction between the two). In general, we see that women in the sam- ple report lower levels of economic suc- cess at endline than males in the sample, though the estimate is not statistically sig- nificant at conventional levels. The largest and most significant correlate of success is initially high access to credit and capi- tal. Wealth begets wealth. Higher levels of education, stronger support networks and being older also contributed to suc- cess. Perhaps surprisingly, having a good partner and a higher level of economic decision-making empowerment showed little effect. Similarly, levels of physical e. Who succeeds? health, exposure to war violence and depression did not have a significant positive or negative effect on AVSI targeted poor, excluded and underprivileged success. These sources of vulnerability do not ap- young women and men. Within this group, how- pear to be associated with success variation, at last in ever, there is fairly wide variation in initial wealth, this already vulnerable group. skills, social support, and other characteristics. Who Second, we look at who responds to treatment with among the target beneficiaries succeed in the ab- more economic success in Figure 14. Rather than sence of treatment? Who responds more success- looking at how these same factors influence average fully to the program? Who fails? A look at these levels of success, we look at how they influence the patterns not only helps explain some of the wide size of the program impact. (Statistically, this is the variations in economic success, but also helps tar- coefficient on the interaction between the character- geting and program design in future. istic and treatment). Again, we see that females have First, we look at what characteristics are associated lower success, except that here we see that, in addi- with more success in the absence of the treatment. tion to having lower average levels of success, their This is analogous to looking within the control group response to the treatment is also more muted than at what characteristics are correlated with later earn- males. ings and wealth. We look at several potential deter- We also see that those with higher levels of access minants of success or failure, one by one, in Figure to credit at baseline see fewer gains from treatment. 13. We create a composite measure of “economic This is consistent with the result above, where wealth success” from endline earnings, durable assets, and begets wealth. The more capital rich people would savings than has zero mean and unit standard devia- have advanced anyways—the intervention is less in- tion (a z-score). Each baseline characteristic is like- fluential in their success. Rather, it is those who are wise turned into a z-score or (in a few instances) is the most credit and capital constrained who benefit a binary indicator. A negative value implies the char- the most from the intervention. This is consistent acteristic is associated with lower economic perfor- with economic theories of poverty. mance. The specific number indicates the change (in standard deviations) in economic success from These same theories, however, predict that those a binary change or unit standard deviation change in with the highest potential (higher levels of skills, the characteristics. (The specific results come from ability, patience and good health) and most con- An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 21 Figure 13. Ignoring treatment, the Figure 14. Impact on ATE of a unit correlation between baseline change in the characteristic characteristics and endline economic success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straints (access to capital and credit, in this case) f. Health and social impacts would see the greatest benefits from the interven- tion. We do not see a large positive effect of skills/ We are not merely interested in material well-being. education, patience or good health on response to Lack of stable employment and poverty can be dis- treatment (unlike the case of capital constraints) empowering, depressing, and alienating. The poor suggesting that these may not be strong measures of do care about their next meal, or their material goods, ability and potential. but they also value greater happiness, lower anxiety, closer social relations, and respect and recognition We see little effect of partner relations on ultimate in their community. This is especially the case when success. If we look at income alone (rather than the the target beneficiaries are the most disempowered, aggregate measure of income, savings and assets) we depressed, and alienated in their community. It may see a considerable positive relationship between bet- be especially true among women who typically have ter partner treatment and income. This relationship, less control over their lives and decisions than men. as well as qualitative data suggesting the importance of partner success to well-being, motivated one of In general, however, we see little health and social the Phase 2 cross-cutting interventions (“W+”) dis- effects, positive or negative, of the intervention on cussed below. beneficiaries, despite the evident poverty impact. Building Women’s Economic and 22 Social Empowerment Through Enterprise Table 3. Indicators of health and social impact in the control and treatment (WINGS) groups Indicator Possible Range Desirable scores Control Treatment Health Days sick last 30 days 0-30 Low 3.2 3.7 Went to bed hungry last 7 days 0-1 Low 0.1 0.1 Perception of health status 1-10 High 5.5 5.7 Betancourt index of depression and 0-48 Low 9.0 9.1 anxiety Empowerment Economic decision making index 0-18 High 10.7 10.8 Gender attitudes index 0-15 High 5.6 5.5 Interpersonal violence index 0-21 Low 19.4 19.2 Independence index 0-15 High 5.2 5.3 Household support index 0-9 High 7.2 7.1 Social Capital Groups and Networks Group membership 0-Any High 1.7 2.4 Is a community leader 0-1 High 0.2 0.2 Trust Self-reported trustworthiness 1-10 High 5.8 6.2 Social Cohesion/Inclusion Community hostility index 0-12 Low 0.7 1.0 Social support index 0-21 High 7.2 7.9 Collective Action Mobilized the community for 0-1 High 0.2 0.2 meetings Table 3 lists major health and social indicators and The only statistically significant finding is that the the average levels of each measure or index in the average WINGS beneficiary reported being sick 20 treatment and control groups. percent more days in the past month than did the av- erage member of the control group. The percentage Health of people who went to bed hungry in the past week was measured with error, so we cannot be confident Starting with health, we look at the number of days that the change is different from zero. The same can beneficiaries reported being sick in the last month, be said for perceptions of health status and an index the number of days they reported going to bed hun- of symptoms of depression and anxiety. gry in the past week, and a perception of their health status relative to others in the community (see Fig- Since the impact on the number of days sick is re- ure 15). ported as percentage change, it is important to keep An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 23 Figure 15. Impacts on health. 40% Average Trearment Effects in Percentage 30% 20% 20% 10% Change 4% 5% 0% -10% -20% -20% -30% -40% -50% Days sick Went to bed Perception of Index of depression in last month hungry last week health status and anxiety (low good) (low good) (high good) (low good) Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. sion (such as feeling sad, having worries, Figure 16. Impact on self-reported symptoms of and experiencing nightmares) adapted anxiety and depression. from other work in northern Uganda that derived a ‘local’ understanding of these &$# constructs through qualitative work. !"#$%&'(&#$)*$++,'"&-"#&-"%,$./& &"# As shown in Figure 16, we see no sig- &'# nificant impact of treatment on psycho- logical distress, in spite of the economic &!# 12.342,## success enjoyed by beneficiaries. Both 54+)36+.3## !%# groups report a reduction in psychologi- cal distress over time, which is not sur- !$# prising because the overall quality of life !"# in northern Uganda likely improved with ()*+,-.+# /.0,-.+# every year spent living away from the dis- placement camps.3 in mind the absolute difference. In this case, a 20 percent increase is equal to about a half a day more per month on average since the average member of 3  This finding is a good reminder about the value of having a control group that did not receive the intervention. If we the control group reported being sick 3.2 days in the had only observed the treatment group before and after the past month. intervention, it would have appeared as if the program led to a decrease in distress. What we see here (with the help of our To assess psychological health, we administered a set control group) is that distress declined over time for every- of questions about symptoms of anxiety and depres- one, regardless of whether they received the program. Building Women’s Economic and 24 Social Empowerment Through Enterprise Figure 17. Impacts on recipients’ children. 180% 160% 140% 120% 116% 100% 80% 60% 59% 40% 20% 24% 0% -1% -2% 2% -20% # of biological Proportion of Education 1 if youngest How many Health children bio children expenditure on child sick at days youngest expenditure on under 18 children least one day child sick in children currently at in pas month past month school Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. Children ceal important heterogeneity—some women who do better and some who do worse, and our ongo- Looking at treatment effects on children presents a ing analysis will explore some of the main possible bit of a puzzle. Recipients of the WINGS program sources of heterogeneity. show higher expenditure on their children’s educa- tion and particularly their health, but we see no cor- The finding of no differences between the treatment responding increase in the percentage of children at- and control group members on intimate partner vio- tending school or their children’s health. Incidence lence is significant because of the risk of increased of children’s sickness seems to remain constant, but intimate partner violence as women have more children of WINGS recipients seem to be sick for money. It will be important to continue to analyze more days (below, Figure 17). The reason for this the data for heterogeneous impacts in intimate part- result is not clear. ner violence, but the absence of any main effect is encouraging. Other studies have reported increas- Empowerment es in intimate partner violence, but in loan-based schemes, and not in randomized impact evaluations. Next we look at several measures of empowerment The fact that the WINGS program was grant-based, (Figure 18), including the person’s influence in thus without the pressures of repayment and high household economic decision-making, household interest rates, might have reduced the risk for inti- relations, attitudes toward female empowerment, mate partner violence. and intimate partner violence, and we see almost no difference between treatment and control group Social Capital members. This average effect of zero4 may con- Finally we look at several measures of beneficiaries’ 4  This average effect of zero is very nicely illustrated by dash line for each indicator crossing the zero axis, suggesting that the effect could possibly be zero. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 25 Figure 18. Impact on empowerment. 25% 20% 15% Average Trearment Effects in Percentage Change 10% 8% 5% 1% 1% 0% -1% -3% -5% -10% -15% Economic Gender attitudes Supportive Independence Domestic decision index behavior index violence making index (high good) index (money) (high good) index (high good) (high good) (low good) Diamond represents the Average Treatment Effects in Percentage Change. Average Treatment Effect is equal to the difference in mean between the Treatment Group and the Control Group. The dashed line spans the Lower and Upper bound of the Confidence Interval. social capital. Social capital can take on a variety thiness than did the average person in the control of meanings, but we use it here to mean the value group. This difference is small but statistically signif- found in one’s social network. Increasing the social icant and suggests that participation in the program capital of beneficiaries is both a desired outcome of changed how beneficiaries see themselves relative to WINGS and a potential predictor of economic suc- others in the community. cess. Here we consider whether WINGS impacted beneficiaries’ social capital in terms of group partici- Social cohesion and inclusion are also key compo- pation and leadership, trustworthiness, collective nents of social capital. People who relate well to others and are accepted by others face fewer barri- action, and social cohesion and inclusion. ers to conducting business and accessing resources. Membership and participation in groups helps peo- We measured two aspects of social cohesion and ple to obtain information and other resources, find inclusion: community hostility toward the benefi- social support and companionship, and influence ciary and social support. Examples of community community decision-making. WINGS beneficiaries hostility include having serious conflicts with other are involved in almost 50 percent more groups on community members, having community mem- average than members of the control group. This is a bers say things to insult or hurt you or your chil- large difference in relative terms, but only translates dren, or experiencing unprovoked aggression from to 0.72 more groups on average. WINGS beneficia- other community members. We find that WINGS ries are also more likely than control group members beneficiaries, on average, report 38.1 percent more to be a community leader, but this difference is not hostility than people in the control group. This dif- statistically significant. ference is statistically significant, but it is important to note that this is a very small absolute difference. Trust is another important dimension of social capi- The community hostility index ranges from a pos- tal; trust can facilitate transactions and relationship sible score of 0, meaning no hostility experienced, to building. On average, WINGS beneficiaries rated 12, an indication of high hostility. As shown in Table themselves 8 percent higher on a scale of trustwor- 3, the mean score among the control group is 0.70, Building Women’s Economic and 26 Social Empowerment Through Enterprise a very low hostility score. Therefore, an increase of viduals contribute to the collective well-being of a 38.1 percent among the treatment group results in a community. In Phase 1 we find that a significantly mean score that is still less than 1 on this index. higher percentage of WINGS beneficiaries mobilize the community for meetings compared to members We also find that the average WINGS beneficiary of the control group. This suggests that the WINGS also reports having 10.7 percent more social sup- program had an activating effect on beneficiaries port compared to the average member of the con- that made them more likely to contribute to the trol group, a statistically significant difference. Social well-being of the community. support is indicated by turning to friends or neigh- bors for advice and receiving practical or material Overall we can say that treatment group members help from these people. report both an increase in positive community sup- port and relationships, but also an increase in nega- Finally, collective action and cooperation are also tive community attitudes and relations—both small important elements of social capital. Usually we effects in absolute terms. One hypothesis is that kin think of these constructs as group-level phenomena and community members whom the beneficiary can that characterize communities that work together now help support are brought closer, while others to solve problems. Both require individual input, may be resentful of the assistance, given the great however, so we can examine to what extent indi- need in these communities. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 27 5. Are these impacts “high”? A cost-benefit analysis a. From “impact” to relative return” Of course, few of these studies include within the cost of the grant the cost of targeting and disbursing. Overall, the WINGS program increased earnings If accounted for, these would undoubtedly lower and allowed higher consumption—short and long rates of return downwards. If the cost of disburse- term—for some of the poorest and most vulnerable ment and targeting were 40 to 100 percent, this ad- people in northern Uganda. However, the simple ministration cost could conceivably eat away most magnitude of the economic impacts is not very in- of the excess returns. At some point, one might ar- formative from a policy perspective. gue, it would simply be better to give the cash rather than the program to the poor. We are not solely interested in the benefit, but rather the benefit relative to costs. One way to do this is So far we have focused mainly on the material ben- to estimate the rate of return on the NGO’s “invest- efits of the intervention, especially earnings. What if ment”, which is the cost of delivering the program. an intervention provides some non-pecuniary bene- fit, such as reductions in depression or improvement Most of all, we would like to see if the return is in community support? These benefits are (literally) positive—that the program does not cost more to priceless, and it is difficult to calculate a return. Es- deliver than it produces in benefits to the poor. For timates could be made, but in this instance they do instance, suppose that an intervention results in a not seem necessary. The psychosocial impacts of stream of benefits to the client in future. In the case the AVSI intervention are, it appears, quite mod- of WINGS, for example, suppose that the interven- est. Thus it seems reasonable to evaluate the returns tion increased earnings by UGX 16,200 not just in from this intervention based on the impact on earn- the month we measured, but every month for the ings alone. next 5, 15 or 50 years. What is that worth? Is the present value of that stream of earnings greater than b. Estimating the returns to the WINGS the sum spent by the program? program More than this, we would like to see that a program What kinds of returns do we see from the WINGS is one of the best possible investments in the poor— program? Table 4 above provides one example of a that it provides a higher return than the alternatives. returns calculation. One way to do this is to compare the program and its returns with those from other interventions. That The top panel, A, describes the return from the full is, we’d like to see that the stream of benefits is at intervention (we return to panel B below, when least as great as alternative interventions that cost evaluating the follow-up component). similarly. To estimate the stream of earnings, we can look at Evidence on returns from different interventions the average treatment effect (of roughly 16,200 is slim, but a growing body of evidence from pro- UGX a month) or the median treatment effect (of grams targeting the ultra-poor suggests that positive roughly 9,700 UGX a month). The long term return returns are possible. Recent research on returns on will depend on two considerations. First is the time small grants to the poor suggest annual rates of re- horizon of returns. As a thought experiment, we turn of 40 to 100 percent. look at 5, 15 and 50 year horizons. Building Women’s Economic and 28 Social Empowerment Through Enterprise Table 4: Rate of return calculations Rate of return on investment (by cost of intervention component) + Targeting + Group + Over- Grant + Business & Disburse- dynamics + Follow-up head (i.e. alone training ment costs training Total) Impact on Discount Net present Treatment monthly Horizon rate (of value of 300,000 550,944 798,514 961,705 1,658,537 1,720,063 effect cash (years) future earnings earnings earnings) Median 9,700 50 50% 232,800 -22% -58% -71% -76% -86% -86% Average 16,200 50 50% 388,800 30% -29% -51% -60% -77% -77% Median 9,700 50 15% 775,550 159% 41% -3% -19% -53% -55% Average 16,200 50 15% 1,295,249 332% 135% 62% 35% -22% -25% Median 9,700 50 3% 3,012,633 904% 447% 277% 213% 82% 75% Average 16,200 50 3% 5,031,408 1577% 813% 530% 423% 203% 193% Median 9,700 15 50% 232,650 -22% -58% -71% -76% -86% -86% Average 16,200 15 50% 388,550 30% -29% -51% -60% -77% -77% Median 9,700 15 15% 693,062 131% 26% -13% -28% -58% -60% Average 16,200 15 15% 1,157,484 286% 110% 45% 20% -30% -33% Median 9,700 15 3% 1,404,613 368% 155% 76% 46% -15% -18% Average 16,200 15 3% 2,345,849 682% 326% 194% 144% 41% 36% Median 9,700 5 50% 212,697 -29% -61% -73% -78% -87% -88% Average 16,200 5 50% 355,226 18% -36% -56% -63% -79% -79% Median 9,700 5 15% 407,736 36% -26% -49% -58% -75% -76% Average 16,200 5 15% 680,960 127% 24% -15% -29% -59% -60% Median 9,700 5 3% 539,828 80% -2% -32% -44% -67% -69% Average 16,200 5 3% 901,568 201% 64% 13% -6% -46% -48% The second consideration is the “discount rate”, or with interest rates charged by rural microfinance in- the value we place on returns today versus returns stitutions in different countries, and lower than the tomorrow. This is partly a function of personal pref- rates of time preference we measure in behavioral erence. Most people prefer money today to money games in this sample). tomorrow. It is also an economic consideration. That Note that a high discount rate is akin to having a money has an “opportunity cost”—its value if it had fairly short horizon—at the higher discount rates, it been put into a risk-free investment and paid out a matters little whether the horizon is 5 or 50 years. regular return. As a thought experiment, we look at three alternative discount rates: a “low” rate of 3%, With these discount factors, a stream of 16,200 UGX one more medium-sized, of 15% per year (the same, every months for 15 years is equivalent to UGX roughly, as Uganda’s lowest commercial lending 388,550 at a 50% rate, 1.15 million at a 15% rate and interest rate), and a high rate of 50%, which heav- UGX 2.54 million at 3%. The more you value money ily discounts returns in the future (but is consistent today versus money tomorrow, the lower is the pres- An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 29 B. Follow-up component Estimated per person cost of intervention 2 FU 5 FU Impact on monthly Discount rate (of Net present Horizon (years) 278,733 696,832 cash earnings future earnings) value of earnings Median 1,700 50 50% 40,800 -85% -94% Average 5,000 50 50% 120,000 -57% -83% Median 1,700 50 15% 135,921 -51% -80% Average 5,000 50 15% 399,768 43% -43% Median 1,700 50 3% 527,987 89% -24% Average 5,000 50 3% 1,552,904 457% 123% Median 1,700 15 50% 40,774 -85% -94% Average 5,000 15 50% 119,923 -57% -83% Median 1,700 15 15% 121,464 -56% -83% Average 5,000 15 15% 357,248 28% -49% Median 1,700 15 3% 246,169 -12% -65% Average 5,000 15 3% 724,027 160% 4% Median 1,700 5 50% 37,277 -87% -95% Average 5,000 5 50% 109,638 -61% -84% Median 1,700 5 15% 71,459 -74% -90% Average 5,000 5 15% 210,173 -25% -70% Median 1,700 5 3% 94,609 -66% -86% Average 5,000 5 3% 278,262 0% -60% ent value of that stream of future earnings. You can intervention. At a 15% discount rate, a continuous see a similar pattern for 50 and 5 year horizons. increase in monthly earnings of 16,200 (the aver- age treatment effect) is worth approximately UGX We then compare this present value to the cost of 1.15 million today. That is more than UGX 500,000 the program. In total, using records provided by lower, or 33% lower, than the per person cost of the AVSI, we estimate that the total cost of the interven- total program. So the rate of return is -33%. tion, per person, was approximately UGX 1.720 mil- lion (last column). To illustrate the cumulative cost With a lower discount rate of 3%, however, the same of the components of the program, we start with the average treatment effect yields a significant positive amount of the grant (UGX 300,000) and gradually return—of 36%. add our estimates of the per person cost of each pro- gram component until we reach the total of 1.720 One note: Both of these results come from average million. treatment effects. The (lower) median treatment effect never reaches the breakeven point even with To calculate the rate of return, we take the return the lower discount rate. This points to the impor- (the difference between the present value and the tance of having a consistently high return in future. cost) and examine it relative to the total cost of the If the return drops over time, the positive returns Building Women’s Economic and 30 Social Empowerment Through Enterprise estimated for the average treatment effect in the 5% treatment effect evaluated using a 3% discount scenario may be exaggerated. On the other hand, if rate for a sufficiently long time horizon—is con- there is growth over time, the returns are undoubt- sistent with positive long term rates of return, edly greater. suggesting that the intervention does better than the crude (but important) alternative— Another note: We see that the less expensive is the simply giving to the client cash equivalent to the program, the higher is the return. Each component cost of the program. This average treatment ef- contributes to the return. Components that cost fect seems robust enough that it is reasonable to more than the benefit they add reduce the average return. Components that provide the highest benefit take this higher estimate as a guide to impact. for their cost increase the average return. Inexpen- • The more that a client values future over pres- sive, high-return components are possibly the best ent consumption (i.e. the more patient they are, ones to focus on. This is a question we will tackle and the lower the discount rate), the more value below in looking at the contribution of follow-up, the WINGS intervention provides relative to experimentally. the alternatives. Analysis of time preference measures in the behavioral games, as it happens, c. The returns to follow-up suggest very high rates of impatience among the population, suggesting negative overall rates of In general, the intervention was not designed to cal- return. culate the cost-effectiveness of each component. In Table 3, the fact that we assume the benefit stays the • Likewise, if the donor or implementer of the same without any of the added components is un- program places a high value on future earnings, likely to be true. Generally it is not possible to say and is farsighted and patient, while the client is which of the above actually contributed to the ob- not, then the donor may still feel this is a worth- served increase in income. The idea that the main ef- while investment. fect is given by the cash transfer comes from results of other studies, but few have looked at the returns • This merely implies that the intervention had a to added components. positive return. This is not necessarily the same thing as the best possible return. In particular, it The randomization of follow-up, however, does is undoubtedly true that some components pro- allow us to test the cost effectiveness of that inter- vide more impact at less cost than others, and vention. Table 3, Panel B calculates returns to the including the lower-impact-higher-cost com- follow-up component. We turn to this below, in the discussion of follow-up. It is worth noting that ponents in the program reduces the potential for low discount rates, the average treatment effect return. Similarly, there may be other interven- from the follow-up component yields a positive and tions available that provide more “bang for the sometimes quite high rate of return. This suggests buck” in terms of higher returns given limited that the assumption of constant benefits regardless resources. of the program components is not an accurate one, • A glance at the incremental cost of some ser- and we ought to concentrate attention on the right- vices and components raises some suspicions. most columns of Table 3. The cost of administering and targeting the cash grant is roughly the same as the amount of the d. Conclusions grant itself. The cost of follow up is as much as the cost of the grant and the targeting/disburse- This kind of cost-benefit analysis is inherently chal- ment costs combined. Whether components lenging and it is difficult to draw strong conclusions. A few lessons are worth emphasizing, however: such as these provide equal relative rates of re- turn is an important question for investigation, • Our higher estimate of returns—the average which we start to examine below. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 31 6. What are the effects of the WINGS programs on other village members, especially existing traders and entrepreneurs? In addition to the comparison of WINGS Phase 1 Theoretically, therefore, the effect could go either clients to future Phase 2 beneficiaries, we might ask way. What happens on balance in the WINGS vil- what the effect of this intervention is on the local vil- lage? We are in the midst of a more formal theoreti- lage economy and (most of all) the men and women cal and empirical analysis, but preliminary results already in the businesses that the WINGS clients suggest that WINGS clients and agricultural house- open. holds in the village both benefit on net, while house- holds that traded in or produce goods and services Many of these villages are quite small, and the suffer from falling prices and competition. WINGS client households often represented 15 to 25% of the households. The cash transfer, therefore, To assess this, the research team visited all the treat- was a sizable and noticeable shock to the village ment and control villages after the completion of economy. Economic theory predicts both positive Phase 1 and surveyed about 2,500 randomly chosen and negative effects, depending on one’s starting po- households who did not participate in the WINGS sition. program. The survey collected detailed information on income, entrepreneurial activity, labor supply, The injection of cash will raise spending by WINGS savings, as well as consumption and expenditure clients, providing a source of business and demand of non-participant households. Comparing the re- for existing traders and producers. This should in- sponses of non-participants in Phase 1 versus Phase crease their profits. At the same time, this may lead 2 villages sheds light on the community-wide effects prices in the village to increase if the demand shock of the WINGS program. is large enough, and the trade in these items is suffi- ciently restricted. These rising prices reduce the pur- WINGS clients primarily became traders of goods chasing power (and hence well-being) of any house- outside the community and producers of other trad- hold that is a net consumer of the goods (i.e. people able goods. Preliminary results suggest that WINGS who don’t produce or trade in that good). benefits net consumers of these products. In the realm of the survey, detailed data on community At the same time, WINGS clients start to trade and prices was collected, and a price index created. In produce themselves. This acts to lower prices in Phase 1 communities this price index is two per- the community, especially to the extent they begin cent lower than in Phase 2 communities. This sug- to bring in tradable goods that were previously ex- gests that WINGS increases the supply of scarce pensive and difficult to obtain. These falling prices traded goods and stimulates competition between benefit all households that are net consumers of the micro-enterprises, driving down consumer prices. good. Net producers and other traders, meanwhile, Lower prices reduce the costs of the consumption can see their profits fall both because of lower prices basket, decreasing consumption expenditure by 11 and also because they have more competition from percent. Most affected by the price decline are gro- the WINGS clients who have started similar enter- ceries: monthly grocery expenditure decreases by prises. about nine percent. Non-participants put the result- Building Women’s Economic and 32 Social Empowerment Through Enterprise ing surplus into savings: grain holdings, for example, drive up agricultural wages. While WINGS partici- increase by four percent among households citing pants spend less time working on other households’ agriculture as main source of livelihood. plots, non-participant households do spend half a day per month more. This is accompanied by an Preliminary results further suggest that WINGS increase in labor income of 1,200 USH per month. crowds out profits of existing micro-enterprises. The labor supply effect varies across non-participant In theory, existing micro-enterprises are affected households. While households not having a micro- through two mechanisms. On the one hand, high- enterprise show an increase of half a day per month, er competition may crowd-out sales. On the other households operating a micro-enterprise spent only hand, because WINGS participants spend part of a quarter of a day/month more working on other their income gains on consumer goods, sales of ex- peoples’ plots. isting micro-enterprises may increase. In practice, we find that the former effect slightly dominates the Overall, the WINGS program leads to a rise in the latter. Profits of existing micro-enterprises fall by on aggregate well-being of the community, but wealth average 2,500USH per month (roughly 12 percent). and well-being are redistributed away from trad- Labor supply of non-participants to micro-enter- ers and net producers of traded goods towards the prise activities, however, is not affected. WINGS clients and agricultural households (net consumers of traded goods). This redistribution, Furthermore, WINGS tends to affect the agricul- and the adverse effects of an aid program on other tural labor market. WINGS participants spend less community members, is not something typically time working on other households’ plots. Lower taken into account in programs of this nature and agricultural labor supply in turn tends to slightly ought to be a consideration in future. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 33 7. Do supervision and mentoring improve performance? The effectiveness (and cost-effectiveness) of follow-up a. The potential gains from follow-up: al success. For instance, AVSI staff maintains close The “accountability” and “advice” supervision of business activities for the first few effects business cycles, providing advice on meeting market challenges and implementing sound business prac- Follow-up is a central part of AVSI’s intervention tices. They seek to help young entrepreneurs to meet package. At the same time, it is an expensive service with experienced businessmen and women in their to provide—it takes a significant amount of staff area, share difficulties and seek practical advice and time, transport, and administrative and logistical mentoring. AVSI staff have been trained in business support. Does it work and why? Is it cost effective? skills and most importantly have years of experience What does this tell us about microenterprise devel- within the environment of small enterprises in the opment for the poor? program region, with accumulated links to success- ful businesses and an array of formal and informal AVSI sees two important advantages of follow-up: financial services. AVSI staff have also been trained one business-related and one interpersonal. Many in psychosocial skills and basic counseling, and one years of experience have demonstrated to AVSI that objective of follow-up was to provide basic psycho- on-going support for young, new entrepreneurs is social support to this highly poor and vulnerable essential to help them succeed and address the chal- population. Staff could work as a listener, advice- lenges that arise with every nascent business endeav- giver, or active mediator in group, community and or. In AVSI’s standard intervention, clients receive at household disputes. least 3 follow up visits, and often more. We can imagine two major theoretical rationales b. Distinguishing accountability from for, and effects of, follow-up. One we call the “ac- advice countability effect”: expecting a program officer to check up on you leads to more focus and disci- How to discern these different theories and effects? plined behavior, such as increased investment, em- Distinguishing business versus interpersonal effects phasis on starting up business operations, effort, and is the most straightforward: we assess each in turn ultimately aggregate profits of the enterprise. These with a variety of different measures. To distinguish accountability effects might be mainly business- the accountability effect from the advice effect, and oriented, but conceivably they could lead to more the returns to increased advice, is more complicated. independence and empowerment for the woman, or “good” behavior by household members (e.g. sup- First, we worked with AVSI in Phase 2 of the inter- port for the enterprise, or reduced intimate partner vention to vary the amount of follow-up experimen- violence). tally, at the level of the individual. Phase 2 clients were placed in a lottery. Within each village, a third A second rationale we call the “advice effect”: hav- were told they would receive no further follow-up, ing a trained program officer visit and give you sub- and that AVSI would not visit them again. A third stantive advice may affect your business and person- were told that AVSI would visit them twice more, Building Women’s Economic and 34 Social Empowerment Through Enterprise mainly to check on progress on stated goals and the We should be able to see the advice effect most business plan, and provide advice. The final third clearly in the long run endline survey, by compar- were told that AVSI would visit them roughly five ing the effects of receiving 5 versus 2 follow-up vis- times in the coming months, both to check on prog- its on average. The comparison of 2 follow-ups to ress and provide more extensive advice. All were no follow-up will indicate in large part the effects told that the assignment was random, for the pur- of accountability in the longer run, but will also in- poses of evaluation, and not linked to the quality of corporate some effects of advice. Thus it will help us their proposal. distinguish the impact of follow-up, but it might not tell us the mechanism so clearly. To distinguish the accountability from the advice effect, we surveyed clients twice after the interven- c. Short-term impacts of follow-up tion: the first time roughly a month after they re- ceived the grant, but before any follow-up visits oc- Follow-up assignment and follow-through curred; while the second survey occurred roughly a year after the grant, after all follow-ups were long was largely successful completed. We call these the “short term” and “long To pick up any accountability effect, it’s important term” Phase 2 endline surveys. The important dis- that clients believe and remember their treatment tinction is that, at the time of the short-term survey, assignment. That is, the clients assigned to no follow- no one had received follow-ups but they had been ups must be told and remember and believe this. The told whether or not to expect one. clients assigned to any follow-up must remember and expect this as well. To a large extent this seems We should see the effects of the accountability treat- to be the case, as can be seen in Figure 20 below. Of ment most clearly in the short term endline survey: those assigned to either 2 or 5 follow-ups, 97 to 98% those who expect follow-up visits and being held ac- expected a return visit from AVSI at the time we ran countable for their decisions may be more inclined the short survey. Of those assigned to no follow-ups, to invest rather than consume the grant, work hard- just 10% expected (erroneously) a visit from AVSI er, or have made more progress in business startup for accountability and advice. and profitability. To pick up any advice ef- Figure 20. Expectations and results of follow-ups. fect, the follow-ups assigned must more or less match the actual follow-ups received. %!#$ This too appears to be the 6.>,?@/9A$89A$A-?-,1->$D$ case, as seen in Figure 20. Of "#$ /9$E$89++97B:;2$ "#$ those assigned to receive 5 follow ups, 91% received at (#$ least three to six follow-ups 6.>,?@/9A$89A$A-?-,1->$!$ 322,F.->$/9$E$ %(#$ G9++97:;2$ (with the vast majority of 9A$C$89++97B:;2$ "#$ these receiving 5). None of 322,F.->$/9$C$ G9++97:;2$ the people assigned to 0 or '#$ 2 follow-ups received several 6.>,?@/9A$89A$A-?-,1->$"$ 322,F.->$/9$"$ '#$ G9++97:;2$ follow-ups accidentally. Of 89++97B:;2$ !""#$ those assigned to two fol- low-ups, 96% received 1-2 %)#$ follow-ups (most receiving *+,-./$0-+,-1-2$3456$7,++$ 2). None of those assigned %&#$ 89++97$:;$7,/<$/<-=$ !"#$ to zero follow-ups appeared to have been followed up by AVSI according to our re- An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 35 cords (though it is possible that a small number of AVSI field staff strongly encouraged investment of unrecorded follow-ups occurred). the grant in the business. Within a month of receiv- ing the grant, those expecting a follow-up visit had All this suggests we have some of the best condi- spent 34% of the grant on business-related expendi- tions possible to observe accountability and advice tures, compared to 27% of those not expecting any effects, if they exist. follow-up or accountability (see Figure 19). This difference of about 7 percentage points is not large We see a moderate accountability effect on in absolute terms—about 26,000 UGX, or USD grant investment, though not on patterns of 10.44—but in relative terms is large. It represents a investment roughly on-quarter increase in business investment compared to the no follow-up group, and is statisti- Figure 21 shows the proportion of the grant that cally significant. clients spent on the following four categories within the first month of receiving the grant: The increase in business investment comes mostly at the expense of consumer durables and other long- a) business expenditures (e.g. raw materials, term consumption, which falls 4 percentage points tools, inventory, etc.) among those expecting a follow up, from 15% to 11% of the grant. This difference too is statistically b) “long-term” consumption (e.g. durable as- significant. sets, home improvements, etc.) There are smaller decreases in short-term consump- c) “short-term “ consumption (e.g. food, small tion and saved/unspent grant funds, which are not household items, etc.) statistically significant. Expectations of accountabil- ity do not seem to increase the pace of investment d) unspent or saved funds or spending. We presented clients with a pile of stones and a sheet While we see an increase in the level of business ex- with pictures of twelve kinds of expenditures (which penditures, we do not necessarily see a shift in the we classify later on into these four categories). Cli- pattern of investment and expenses. ents were asked to allocate the stones according to how they spent the grant, and the proportions are Figure 22 presents the results of a business expen- calculated from the relative balance of stones. ditures survey, which asked clients to estimate total expenses in various categories in the previous Figure 21.Grant allocation 4 weeks after receipt. 4 weeks. (These expenses might or might not come from the grant, and are not ("#$ constrained to be less than )*+,+*-+.$+/$0*1.2$3,4.2$+.$=:3>.433$ %'#$ the grant amount—they are )*+,+*-+.$+/$0*1.2$3,4.2$+.$;+.0<24*8$ &&#$ merely the client’s recollec- 9+.3:8,-+.$ &!#$ tion of aggregate spending in every category.) )*+,+*-+.$+/$0*1.2$3,4.2$+.$37+*2$24*8$ &#$ 9+.3:8,-+.$ %#$ We see very small, statisti- cally insignificant differ- !%#$ ences in the amounts spent )*+,+*-+.$+/$0*1.2$.+2$3,4.2$+*$31546$ !"#$ within each business cat- egory. ?33>0.46$2+$@+;;+A<:,$ B+$@+;;+A<:,$ What is surprising is that we also see little difference Building Women’s Economic and 36 Social Empowerment Through Enterprise in the aggregate mount Figure 22. Estimated business expenses since grant, reported spent on busi- in 000s of UGX ness items. Both treatment and control groups report about 36,000 UGX (USD G374H&I774H&06:&./@7;&J767;01&78976474& %#%+& +#!+& 14.49) in spending. This is F/;3?/3;01&5>9;.B7>76/4&06:&>0/7;5014& )#(+& remarkable because it di- 78976474& )#($& verges from the response we !#*)& received from the allocation -..14D7E359>76/&93;?@0474& !#%+& of grant spending questions !#$+& above. A6B76/.;C&93;?@0474& *#'(& The results differ in the lev- <0=&>0/7;5014&93;?@0474& '"#+,& '"#()& el—if the clients had spent a quarter to a third of the -./01&23456744&78976:5/3;74& !*#,%& !"#$%& grant on business, we would expect 90,000 to 120,000 K445J67:&/.&L.11.=M39& N.&L.11.=M39& UGX in spending (USD 36.15 – 48.20). The results also differ by treatment status—those expecting a in some respects probably represent reality. We can follow-up visit said they spent a higher proportion imagine systematic measurement error in either of their grant but their level of spending is about the case. We suggest the divergent result should lead us same at 36,000 UGX. to be cautious about inferring a solid accountability It is difficult to say which estimate is right—both effect. Figure 23. Perceived control over grant. L9/?/9M/2%/B%:9;26%5?426%/2%02E;2647% !"!'% 164N5%/9%:1342%6/%/6<495% !"!'% -A17%8/0%<4C?%74D174%956%?9/?/547@-% !"#$% H551:247%6/%O/CC/EP0?% Q/%O/CC/EP0?% An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 37 Those assigned to follow up (accountabil- cash earnings, a statistically significant increase. In ity) do not appear to report lower levels of absolute terms the impact is small—about 5,000 UGX per month, or USD 2.01. Since cash earnings control over the grant. are so low in this group, however, the relative impact The intervention seeks empowerment of the poor, on the business is substantial. but at the same time holds them accountable for in- This average effect on earnings is the same for assign- vestment. Do clients feel they have limited control ment to 2 and 5 follow-ups, suggesting that the mar- over the grant as a consequence of being followed ginal effect of the 3rd, 4th or 5th follow-up visit (for up? AVSI required clients to produce a coherent profitability, at least) is low. Two visits seems to be business plan, and strongly encouraged the buy- enough to encourage higher profits—either because ing and selling of items (though clients were free more of the grant was invested (i.e. they have a high- to change afterwards). How did clients respond to er capital stock, perhaps because of the accountabil- these guidelines, and did they feel constrained by ity effect) or because of the substantive value of the them? advice. Looking at the median capital stock shows Figure 23 records responses to a number of empow- a fairly large increase, indicating that the former is erment questions asked four weeks after grant re- probably a factor. The absence of an earnings effect ceipt. In general, most clients, regardless of whether from the third to fifth visits, however, diminishes the they are assigned to follow up or not, report that case for the substantive vale of advice. they spent little of the grant on unwanted items, This average impact, however, is somewhat sensi- played as role in grant spending, had control over tive. Like the Phase 1 economic impacts above, this the spending, and felt free to spend the grant how treatment effect of follow-up on income seems to they pleased. Those who were assigned to follow up be driven especially by a handful of very high im- felt slightly less free to make changes, and tended to pacts. Alternative measures that are not as affected follow through with the business in their proposal by these extreme values, like the median treatment (rather than an alternative). But in general the cli- effect, are substantially lower. Other transforma- ents expressed a fair amount of independence and tions of income that compress these extreme values control. (not shown) show virtually no average impact on in- come. This suggests that we should take the income d. Longer term (one-year) impacts of ATE with some caution. follow-up With higher income we would expect an increase Table 5 displays indicators of wellbeing as measured in short-term consumption, long-term consump- at the long-term endline survey, roughly one year tion (through durables) and also savings. Those after the start of Phase 2, for those assigned to the who receive the two follow-ups, however, tend to WINGS program with and without any follow-up. show a small or no change in current consumption, a large increase in savings stock (by 25%, or about The one-year economic impact of follow- 50,000 UGX, USD 20.08), and a modest (but not up: increased profits and savings, but lower significant) increase in asset wealth. This increase in asset wealth savings and increase in durable assets is most pro- nounced among those assigned to 5 follow-ups. Table 5 and Figure 24 display the impacts of receiv- ing any follow-up on four different measures of eco- The evidence here points to modest improvements nomic success: net cash earnings in the past 4 weeks from both accountability and advice. To the ex- (previous to the survey), a measure of short term tent that net cash earnings is the best measure of spending (consumption), an index of durable assets returns, however, we do not see an added benefit and wealth, and the stock of savings. from the third to fifth follow-up. Rather, the third to fifth follow-up, if anything, mainly seems to lead Those assigned to any follow-up report 27% higher to a decision to save more—a reallocation of how Building Women’s Economic and 38 Social Empowerment Through Enterprise Table 5. Indicators of well-being in the control and treatment (assigned to follow-up) groups Indicator Possible Range Desirable scores Control Treatment Poverty and Financial Net cash earnings in the past 4 weeks 0 to Any High 18.6 24.1 (000s of UGX) Household short-term spending 0 to Any - 53.2 55.2 (000s of UGX) Wealth index (0 to 1) 0-1 High 0.4 0.5 Savings stock (000s of UGX) 0 to Any High 192.8 250.2 Health Days sick last 30 days 0-30 Low 3.3 2.7 Went to bed hungry last 7 days 0-1 Low 0.1 0.1 Perception of health status 1-10 High 6.3 6.4 Betancourt index of depression and 0-48 Low 9.9 8.4 anxiety Empowerment Economic decision making index 0-18 High 11.5 11.3 Gender attitudes index 0-15 High 5.8 5.9 Interpersonal violence index 0-21 Low 20.6 20.6 Independence index 0-15 High 6.6 7.1 Household support index 0-9 High 7.4 7.4 Social Capital Groups and Networks Group membership 0-Any High 2.6 2.7 Is a community leader 0-1 High 0.3 0.2 Trust Perceived trustworthiness in 1-10 High 7.0 6.9 community Collective Action/Cooperation Mobilizes the community for 0-1 High 0.1 0.1 meetings Social Cohesion/Inclusion Community hostility index 0-12 Low 0.6 0.5 Social support index 0-21 High 6.5 6.0 An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 39 Figure 24. Impact of assignment to any follow up. 100% Average Trearment Effects in Percentage Change 80% 60% 51% 45% 40% 27% 25% 20% 21% 20% 11% 6% 7% 3% 0% -20% -40% Net earnings in Household short- Wealth index Savings stock Capital Stock past 4 weeks term spending (Average) (Average) (Average) (Average) (Average) income is used rather than an increase in income. up condition reported that they went to bed hungry The confidence intervals are so wide, however, that in the past week compared to people in the no fol- it is impossible to distinguish an independent and low-up condition. Those who receive follow-up also significant effect of the third to fifth follow-ups. The report 11 percent fewer symptoms of depression evidence is, frankly, ambiguous and sometimes in- and emotional distress. The impact of the 2 follow- consistent. It is difficult to see much evidence in ups (14.2%) is larger than the impact of 5 follow- favor of many follow-ups, however, especially given ups (9.6%), though the difference between the two the cost-effectiveness analysis below. is not statistically significant, so must be taken with caution. Follow-up leads to modest, though not al- ways significant, improvements in health, Despite these apparent effects on hunger and psy- empowerment, and social well-being. chological distress, at least with intensive follow-up in the case of hunger, there was no evidence of an Looking across a variety of measures of health, em- effect on perceived health status. powerment, and social well-being, we see small to moderate improvements among those assigned to Empowerment follow-up, although these improvements are not al- Figure 26 examines measures of independence and ways statistically significant. empowerment, and relations with the spouse/main Health male in the household, and the household in gener- al. Those who receive follow-up report little change As shown in Figure 25, beneficiaries who received in independence in the household or ability to make follow-up support reported fewer days of illness on decisions on their own. These empowerment mea- average than those who did not receive this support, sures are nearly unchanged between those assigned but the effect was small in absolute terms and not to receive follow up and those that were not. significant. Any number of follow-up visits led to a decrease in the percentage of beneficiaries who went We see modest increases, of the order of 5 to 7%, in to bed hungry, but this effect was only significant for spousal support, independence in the household, those who were assigned to receive five visits; 43 and in rates of marriage. None of these are statisti- percent fewer beneficiaries in the intensive follow- cally significant, however. Building Women’s Economic and 40 Social Empowerment Through Enterprise Figure 25. Impacts on health. 10% Average Trearment Effects in Percentage Change 0% 1% -10% -11% -16% -20% -30% -28% -40% -50% -60% -70% Days sick Went to bed Perception of Index of depression in last month hungry last week health status and anxiety (low good) (low good) (high good) (low good) Figure 26. Impacts on empowerment. 15% Average Trearment Effects in Percentage Change 10% 6% 5% 5% 2% 0% 0% -2% -5% -10% Economic Gender attitudes Supportive Independence Domestic decision index behavior index violence making index (high good) index (money) (high good) index (high good) (high good) (low good) Social Capital (positive or negative) are larger, surprisingly, for two follow-ups than five follow-ups, though the dif- Figure 27 examines core measures of social capital. ference between the two is not statistically signifi- We observe reductions in community hostility (of cant. We are hesitant to draw a conclusion from this 22%) as well as community support (of 8%) though puzzling pattern, as it could be a result of statistical only the latter is statistically significant. All impacts noise. An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 41 Figure 27. Impacts on social capital 20% Average Trearment Effects in Percentage Change 10% 0% 1% -1% -8% -8% -10% -20% -22% -30% -40% -50% -60% Group Is a community Perceived Community Social support membership leader (indicator) trustworthiness hostility index (high good) (self) in index (high good) community (low good) (high good) e. Is follow-up cost effective? follow-up in Phase 1, or roughly 279,000 UGX per participant (USD 111.65). In this case the return to To assess whether the added follow-up breaks even follow-up is 43% at a 15% discount rate and a (large) (i.e. has a positive return), we can return to the rate 160% at a 3% discount rate, looking at a horizon of of return analysis above, in Table 4. 15 years. These returns are greater than the average return to the entire WINGS intervention in Phase Again, we see that whether or not the return is posi- 1, suggesting that follow-up may have raised the tive depends a great deal on the size of the treatment overall cost effectiveness of the intervention rela- effect and the value placed on current versus future tive to some other components. The follow-up cost money. does not necessarily represent its administrative and The average treatment effect of roughly 5,000 UGX overhead cost, however, and so may overstate some- (USD 2.01) is somewhat fragile, but in magnitude what the relative returns, in addition to the lack of seems to be large enough to justify the cost of two robustness regarding alternative measures in our es- follow-ups, estimated as 2/5ths of the total cost of timate of the treatment effect. Building Women’s Economic and 42 Social Empowerment Through Enterprise 8. The effect of building social and group networks Before examining whether the group dynamics not running joint microenterprises or helping each intervention had an impact on business or social other with the new business so much as they are en- success, first we should look whether it had any gaging in more communal farming (a traditionally impact on group dynamics. Survey responses collaborative activity in the area) and forming sav- suggest it did. ings groups together. These communal farming and savings groups could simply be a substitute for other The immediate impact of the group dynamics train- groups (i.e. those without the group dynamics train- ing was to increase group participa- tion, leadership, and cohesiveness, Figure 29. Group dynamics effect on as seen in Figure 28 below. The aver- meeting frequency age participant in a group dynamics program was involved in 2.8 groups, -%./%0123%4567%8292:;<=><23% !"$'% compared to 2.1 groups in the 122?%<9%=%1.9?@%/.>%8A3<9233% !"(,% WINGS program alone and 1.72 in =;0D<023% !"!)% the control group. -%./%0123%4567%8292:;<=><23% !"(!% 122?%<9%=%1.9?@%/.>%3.;<=B% !"##% F7GH6%I% 3AEE.>?% !"!(% HJ% Participants in the group dynamics -%./%0123%4567%8292:;<=><23% ("(*% training also met far more often with 122?%<9%=%1.9?@%/.>%3=D<9C3% !"+(% F7GH6% their groups for different activities. =;0D<023% !"!'% K9BL% Figure 29 below shows the frequen- -%./%0123%4567%8292:;<=><23% #",,% 122?%<9%=%1.9?@%/.>%;.11A9=B% !"+!% cy of meetings per month for various /=>1<9C% !"!&% activities. -%./%0123%4567%8292:;<=><23% ("')% #"!*% 122?%<9%=%1.9?@% The responses suggest that WINGS !"#$% clients trained in group dynamics are Figure 28. Impacts of group dynamics training on group participation 100% Average Trearment Effects in 80% Percentage Change 60% 53.2% 40% 42.4% 37.1% 27.4% 20% 0% # of groups involved Is a group leader Rating of group Rating of how well in (indicator) cooperation (1-10) members know each other (1-10) An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 43 ing might easily find other community members to savings groups went up for those who received the save and farm with). These may also be new social group dynamics training. This participation seems bonds and joint economic activities, in which case to have resulted in much greater farming productiv- we might expect an increase in farm productivity ity, as can be seen by the increase in farming income and income, as well as an increase in participation accompanied by a much smaller increase in hours in savings groups, the amount of savings and access (below, in Figure 30). to credit. The impacts on poverty and employment are We see results that indicate that the latter case is true, less clear. As noted in the earlier draft, we see a as participation in external communal farming and dramatic increase in income, which persists when Figure 30. Group dynamics impact on farming and savings activities 140% Average Trearment Effects in Percentage Change 120% 100% 89% 80% 60% 63% 52% 40% 20% 0% 5% 0% 4% -20% -40% -60% -80% -100% Hours worked Earnings from Regularly sold Are you Are you Current in all farming/ all farming/ your own involved with a involved with a savings stock animal raising animal raising products farmers group credit or cash in past month in past month or box (VSLA) cooperative? group? Figure 31. Group dynamics training impacts on poverty 100% Average Trearment Effects in Percentage 80% 69% 70% 60% 40% 44% Change 20% 13% 0% -4% -3% -20% Net cash Net cash Household Household Wealth index Wealth index earnings in earnings in short-term short-term (Average) (50th the past 4 the past 4 spending spending Percentile) weeks weeks (50th (Average) (50th (Average) Percentile) Percentile) Building Women’s Economic and 44 Social Empowerment Through Enterprise looking at the median treatment effect as well as Finally, the increase in group cohesiveness and group the average. Other measures of wealth, however, meetings for social support do not correspond with such as household consumption and a normalized the positive effect in psychosocial measures one wealth index, do not see the same impact, although might expect (seen below in Figure 32), although the median treatment effects show a slight increase we do see a non-significant drop in community (seen in Figure 31). hostility. Figure 32. Group dynamics impact on psychosocial outcomes 40% 20% 0% -2% 1% Percentage Change -6% -20% -35% -40% -60% -80% -100% Friends/Neigbors Community Hostility Community Support Symptoms of Support Index index Index depression/distress An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 45 9. Does Male Involvement Promote Women’s Empowerment and Well-Being? Consistent with other studies of economic assistance joint problem-solving skills into the business skills programs targeting poor women, our results suggest training. that WINGS had a positive impact on economic ac- tivity, household wealth, and economic security, but We hypothesized that this ‘nudge’ would lead to a did not improve women’s health and empowerment. measurable impact in the partner’s direct and indirect What could account for this impact-paradox, as it support for the business—from relaxing constraints has been called in the literature? placed on the woman, to providing emotional and indirect support that helps the woman to juggle all One hypothesis is that any personal gain women of her responsibilities, to actively participating in derived from an increase in their household’s eco- business operations. We further hypothesized that nomic well-being was offset by the stressors asso- increased partner support would lead to more busi- ciated with planning, launching, and maintaining a ness success and begin to change gender attitudes in new business. A complementary hypothesis is that the household. the lack of male partner involvement in the program limited opportunities for changes in gender attitudes Our results partially supported these hypotheses. and behaviors that would lead to women’s empower- ment. b. Did W+ impact the couple’s communication and relationship? We decided to test these ideas by running the pro- gram again, this time experimentally varying the Yes, the impact was small but significant (see Fig- framing from an individual approach to a more in- ure 33). Despite only spending a few hours learning clusive household approach. Women in the stan- and practicing communication and joint problem- dard program participated as individuals; women solving skills in the business skills training, the aver- in the W+ variant of the program participated with age W+ participant reported statistically significant household partners. Would this slight reframing— gains in these skills more than a year later. For in- essentially a zero-cost variation of the standard pro- stance, when asked to assess her partner’s listening gram—have a positive impact on women’s empow- skills, the rating of the average W+ beneficiary was erment and well-being? 9.2 percent higher than the rating of the average par- ticipant in the standard WINGS program (about a a. Underlying theory of change half step on a 10-step rating ladder). Our measures mainly come from the longer one-year evaluation, as Our theory of change was based on three key ele- the immediate post-program evaluation contained ments leading to change in women’s empowerment: few questions of this nature. (a) re-framing the program as an opportunity to receive training and start-up capital for a female- We also see a significant impact on communication led business that would involve the household; (b) frequency. On average, W+ participants rated their including male partners from the start; and (c) in- frequency of communication with their partner 8.1 corporating discussions about gender relations and percent higher than the rating of the average partici- exercises to build the couple’s communication and pant in the standard program. A small but significant Building Women’s Economic and 46 Social Empowerment Through Enterprise Table 6. Indicators of well-being in the control and treatment (assigned to W+) groups Indicator Possible Range Desirable Scores W normal W+ Poverty and Financial Net cash earnings in the past 4 weeks 0 to Any High 25.9 18.6 (000s of UGX) Household short-term spending 0 to Any - 57.4 51.8 (000s of UGX) Wealth index (0 to 1) 0-1 High 0.4 0.5 Savings stock (000s of UGX) 0 to Any High 226.4 233.9 Health Days sick last 30 days 0-30 Low 3.3 2.5 Went to bed hungry last 7 days 0-1 Low 0.13 0.09 Perception of health status 1-10 High 6.2 6.4 Betancourt index of depression and 0-48 Low 9.5 8.4 anxiety Empowerment Economic decision making index 0-18 High 11.4 11.4 Gender attitudes index 0-15 High 5.8 5.9 Interpersonal violence index 0-21 Low 20.6 20.7 Independence index 0-15 High 7.2 6.7 Household support index 0-9 High 7.4 7.4 Social Capital Groups and Networks Group membership 0-Any High 1.8 1.7 Is a community leader 0-1 High Trust Perceived trustworthiness in 1-10 High 7.0 6.9 community Collective Action/Cooperation Mobilizes the community for 0-Any High 127. 5 303.5 meetings Social Cohesion/Inclusion Community hostility index 0-12 Low 0.6 0.5 Social support index 0-21 High 6.5 6.0 An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 47 Figure 33. Impact on relationship skills and quality Average Trearment Effects in Percentage Change 25% 20% 15% 10% 9% 9% 8% 6% 5% 5% 0% -5% Rating of Rating of Proportion Hours per day Rating of partner's couple's discussed (awake) relationship listening skills communication business with spent with quality frequency partner partner difference of 5 percentage points was observed in Figure 34. Partner contribution to household chores the proportion of wom- en who reported having talked with her partner ,-./01.$21345$67/2$931-0708$ "(#$ "'#$ about the business since receiving the grant, add- ,-./01.$21345$67/2$9:3319=08$ %+#$ @A$ >.16::?$ %!#$ B/-0?-.?$@CDEB$ ing further support that W+ training led to the ac- ,-./01.$21345$67/2$<1/92708$ )*#$ 6-/1.$ &"#$ quisition of new skills. ,-./01.$21345$67/2$9::;708$ &"#$ Lastly, it appears that %"#$ couples are experiencing ,-./01.$21345$67/2$6-52708$ %(#$ improvements in their 93:/215$ !"#$ relationship as a result of the program: the average woman in the W+ program rated her relationship helped with many household chores, as shown in with her partner 6.2 percent higher than did the av- Figure 34 below. Thirty-six percent of women who erage woman in the standard WINGS program. participated in W+ reported that their partners be- came more supportive after the business skills train- c. Did W+ impact the partner’s ing compared to only about one quarter of women direct and indirect support for the who participated in the standard WINGS program. business? When asked to rate her partner’s support for the We observed no differences between the groups of business, the rating of the average W+ beneficiary women in terms of hours spent on the business and was 8.1 percent higher than the rating of the aver- domestic chores, but a significantly greater propor- age participant in the standard WINGS program. tion of W+ participants reported that their partners Similarly, her rating of her partner’s involvement in Building Women’s Economic and 48 Social Empowerment Through Enterprise Figure 35. Impacts on poverty and savings 140% 120% 100% 80% 70% 60% 40% 20% 15% 10% 0% -7% -10% -12% -20% -17% -28% -40% -60% -80% Net Net Household Household Wealth Wealth Savings Savings earnings in earnings in short-term short-term index index stock stock past 4 past 4 spending spending (Average) (Median) (Average) (Median) weeks weeks (Average) (Median) (Average) (Median) the business was 7.8 percent higher than her coun- It is also notable that one month after receiving the terpart’s in the standard WINGS program. These start-up capital, the average W+ participant report- differences show small but statistically significant ed 72.6 percent less in household savings compared change in direct support. to the average participant in the standard program. This difference evened out, however, by the end of d. Did W+ impact the household’s the program. This initial difference in savings cannot economic security? be explained by differences between the groups in terms of business investments, borrowing, or trans- The results are mixed (see Figure 33). On average, fers. women in the W+ program reported more than a One possible explanation is that AVSI reports that quarter less profit in the past four weeks and nearly it observed that in the W+ households the husband 10 percent less non-durable household consump- and wife would often divide the money to run dif- tion compared to women in the standard WINGS ferent businesses, and therefore the profits of each program. Non-durable goods are consumables that business could be smaller than the total would be do not last, like food and medication. The fact that registered in case a bigger capital was invested in one both of these indicators moved together makes business. Unfortunately this information came too sense because households with less income have less late to be measured in the profits-making section of money to spend. the survey. But the household assets and consump- tion figures should reflect the husband’s investments These statistically significant negative findings are and profits indirectly. The evidence here is mixed— challenged by the observation that W+ participants an increase in asset wealth but a decrease in con- accumulated 1.5 times more wealth on average than sumption and savings. Overall, it is difficult to see a did their counterparts in the control group. We mea- pure economic case for the W+ intervention. sured wealth by durable assets—items like farm im- plements that can be used over and over again—so Interestingly, the W+ program seems to have had a this significant increase in wealth suggests a shift of positive impact on economic security in a very lit- grant spending or investment into durable assets. eral sense. On average, women in the W+ program An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 49 reported experiencing about one quarter less com- The results regarding impact on empowerment are munity victimization compared to women in the mixed. Consistent with Phase 1, we find that women standard WINGS program. This statistically signifi- in W+ do not report experiencing any statistically cant difference can be accounted for by lower rates of higher (or lower) rates of intimate partner violence robberies, property destruction, and stolen goods. on average. Similarly, the impact on indices of eco- nomic decision-making and gender attitudes were e. Did W+ impact the woman’s health statistically indistinguishable from zero. and empowerment? We do see an impact on an index of supportive be- havior. The average W+ participant’s score was 13 When surveyed 30 days after receiving the start-up percent higher than the average participant in the capital, the average W+ participant’s score on an in- standard program. This index measures how well the dex of symptoms of depression and anxiety was 15 beneficiary is treated by her partner and the extent percent lower than the score of the average woman to which the partner supports her business and al- in the control group. When surveyed again at the lows her access to financial resources for household end of the program, evidence of a positive impact needs. This finding is consistent with the evidence persisted, with W+ participants reporting 8.6 per- we see regarding improved relationships and coop- cent lower scores. Both differences were statistically eration among W+ participants and their partners. significant, but small in absolute terms—the rough equivalent of endorsing 1 out of 8 possible symp- We also observe a possible contraction of a woman’s tom less frequently, from often to sometimes, from independence; the average W+ participant’s score sometimes to rarely, or from rarely to never. on an index of independence was 6.1 percent lower than her counterpart’s. A negative interpretation We also observe positive and statistically signifi- would be that the involvement of her partner actu- cant impacts on the number of days sick in the past ally decreased empowerment. A positive or neutral month and the number of hunger days in the past interpretation of this finding would be that the W+ week (Figure 36). While positive, these improve- participants were working more closely with their ments only led to a small increase in perceived health partners on the business, which led to real or per- status that is not statistically significant. ceived differences in freedom. Figure 36. Impacts on health 20% Average Trearment Effects in Percentage Change 10% 4% 0% -10% -9% -20% -26% -30% -33% -40% -50% -60% -70% Days sick Went to bed Perception of Index of depression in last month hungry last week health status and anxiety (low good) (low good) (high good) (low good) Building Women’s Economic and 50 Social Empowerment Through Enterprise Figure 36. Impacts on health Average Trearment Effects in Percentage Change 20% 10% 4% 0% -10% -9% -20% -26% -30% -33% -40% -50% -60% -70% Days sick Went to bed Perception of Index of depression in last month hungry last week health status and anxiety (low good) (low good) (high good) (low good) An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 51 10. Discussion and Conclusions The poorest, most capital and credit- more cost effectively, perhaps relying on more lo- constrained women stand to benefit most cal expertise that do not come with the high wage and transport costs of sending highly trained social from microenterprise start-up assistance workers to the field. The intervention not only doubled incomes among the very poorest women, but had the largest impact Cost-effectiveness in poverty alleviation is on those with the least assets and access to credit. best pursued through further experimenta- One reason is that, in the absence of the interven- tion tion, the women with initial assets and credit access, however meager, manage to invest and advance. Overall, more research is recommended to tease out the effect of the different components, and the ex- This suggests that future interventions ought to tar- tent of spillovers if the intensity of treatment is var- get the poorest and most constrained people with ied. Perhaps the most important variation to test is capital. It meets two of the objectives of any humani- the impact of a simple, low-cost cash transfer versus tarian program: target the poorest and most vulner- the more complete intervention. Two cash transfers able, and have the highest impact at the lowest cost. could be tested: one equal to the regular program cash transfer (but without all the benefits of the full The need for cost-effective service delivery program such as training and supervision) and one equal to the full cost of delivering all components of While the impacts of the intervention were large, in the intervention. general they struggle to pass a simple cost-benefit test. Assisting the poorest women from extreme Another clear candidate for further interrogation is poverty may be so important that such cost-benefit the business skills training. A recent report by McK- considerations are secondary, or must better take enzie and Woodruff (2012) notes that there is little into account the greater value we place on helping evidence so far of its effectiveness, let alone cost- the very poorest. A case could be made for valuing effectiveness. Few have investigated the impact of movements away from the direst poverty over other training on very low-skill populations like this one, impacts of aid. in combination with a cash grant. This situation of- fers the highest potential for training impacts, since We doubt this case needs to be made, however, since skills are so meager in the population. The fact that we believe it is possible to deliver nearly the same we see little relation between skills/education and anti-poverty impacts at less cost. The grant to the economic success, however, suggests that training’s women was likely the most impactful element of impact might be modest. It can and should be in- the program. It represents, however, less than a sixth vestigated whether this process is worth its cost by of the per-person cost of the intervention. Increas- experimentally varying whether training is provided ing the ratio of grant to total cost is likely to be one and the length and content of that training. While of the best investments in poor women in Uganda. it may not be possible or desirable to do away with Halving the cost of the intervention would likely re- training entirely, it is possible it could be streamlined duce the impacts of the grant by far less than half, or economized (training currently takes 4-5 days). enabling more ultra-poor women to be reached and increasing the aggregate welfare in the region. Another focus of future research could be on reduc- ing the cost of disbursement. While it is difficult Additions like follow-up narrowly passed a cost- to allocate and attribute costs, presently it looks as benefit test. These additions need to be delivered though it costs as much or more to target clients and Building Women’s Economic and 52 Social Empowerment Through Enterprise disburse grants than the grants themselves. Stream- tivities and isolating them from sources of support. lining and economizing here is likely to have some On the other hand, the social causation hypothesis of the greatest impact on value, and allow the most claims that the experience of poverty can lead to in- people to be served by such an intervention. One creased stress and social isolation, thereby heighten- potential way to reduce costs is to shift from costly ing the risk of mental health problems such as de- in-person grant distribution to remote disburse- pression and anxiety. ment via a mobile phone-based banking system like M-Pesa in Kenya or Mobile Money in Uganda. This If this latter pathway exists, then economic programs would not work for all clients presently, but mobile would indirectly improve a woman’s health by re- phone penetration rates are increasing every year ducing her daily stressors. In humanitarian settings, and telecom providers continue to expand the geo- this pathway has been referred to as a psychosocial graphical reach of their networks. approach. While common to nearly all post-conflict and humanitarian settings, there is limited evidence Additionally, since both the economic and non-eco- of the effectiveness of this approach, with a recent nomic impacts of the interventions could change systematic review showing no rigorous studies that over time (non-economic impacts may take espe- examined the impact of broader services and secu- cially long to emerge) we recommend a research rity interventions (see Tol et al., 2011) and very lim- design that allows measurement of 3 to 5 year im- ited evidence on the effect of economic programs on pacts. To the extent an organization like AVSI does mental health (Lund et al., 2011). not wish to maintain an ultra-poor and vulnerable control group for that period, the cash versus full Our results are disappointing: we find little evidence program comparison is possibly the most instruc- of secondary effects of improving economic status tive, especially to the extent that cash grants can be on women’s physical and mental health in the me- varied to get at the returns to different size transfers. dium term. On average, women who participated in WINGS were less likely than women in the control These conclusions all stem from the emphasis on group to go to bed hungry, but this effect was not poverty-alleviation as a goal. To the extent that statistically significant at conventional levels. There non-economic impacts are the objective, a different was no measurable difference between the groups program may be required. We revisit this point after in perceived health status or reported symptoms of reviewing the evidence on these non-economic ef- anxiety and depression. If anything, women who fects. participated in the program were sick about a half a day more in the past month. Improvements in economic well-being do not necessarily have secondary effects on a We do see higher expenditure on children’s health woman’s health and empowerment and education among the treatment group, but this does not appear to lead to improved health or educa- While the anti-poverty impact is an important find- tional status, at least in medium term. The incidence ing on its own, we came to the study with the hy- of sickness among the beneficiaries’ youngest chil- pothesis that improving economic status would dren remains constant, but the children of WINGS have large secondary effects on outcomes such as a participants seem to be sick about a half a day more woman’s health and empowerment. per month. The reason for this is not clear, and the difference is unlikely to be meaningful. Physical and Mental Health Empowerment Current thinking suggests that the relationship be- tween poverty and mental health is bidirectional, Poor rural woman are typically targeted in micro- meaning that the causal pathway can flow in both finance programs because they are believed to be directions. On the one hand, the social drift hypoth- more likely than men to repay the loan and use esis says that ill mental health can lead to poverty by the profits to benefit the household, particularly limiting a person’s ability to engage in productive ac- through spending on children’s education and health An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 53 (Garikipati, 2008). The conventional wisdom also First, involving male partners and training the assumes that lending to women will enhance their couples on communication and joint-problem status in the household, thereby empowering them solving led to more partner involvement and sup- through a set of “mutually-reinforcing virtuous spi- port for the business, both direct and indirect, rals” (Mayoux, 1999). While ample evidence sug- and had a lasting positive impact on the couples’ gests that lending to women does lead to reliable interactions. While these effects are small, we be- improvements in individual and household income, lieve they are meaningful. Our theory of change for the data on empowerment are not as consistent or W+ begins with the hypothesis that skill building clear. Households benefit from economic assistance and a more inclusive approach would lead to chang- targeting women, but women may not be empow- es in male behavior and, ultimately, attitudes. ered as a result, a pattern that Garikipati (2008) re- fers to as the “impact-paradox”. Second, the combination of income generation plus partner involvement and support had small This paradoxical pattern describes our results but positive impacts on women’s physical and well: women report benefits to the household mental health. Compared to women who partici- in terms of income, consumption, savings, and pated in the standard WINGS program—which did investment in children, but not greater levels of not lead to any health impacts in the first phase— empowerment. Specifically, we detect no effect of women who participated in W+ were sick less of- WINGS on participation in household decision- ten, were more food secure, and experienced fewer making, independence, gender attitudes, or rates of symptoms of depression and anxiety. Again, we intimate partner violence. While this finding of no must caution that these effects are small: less than effect on intimate partner violence is important— 1 day difference in the number of days sick in the suggesting that participation in WINGS did not in- past month, 4 percent difference in the proportion crease a woman’s risk of violence—we had hoped to of women who sometimes go to bed hungry, and the see a decline. rough equivalent of endorsing 1 out of 8 possible symptoms of anxiety and depression less frequently, Involving men does not solve this paradox, from often to sometimes, from sometimes to rarely, but some results are promising or from rarely to never. Why do we observe economic success but not em- While small, these effects are important from a pol- powerment or health? With respect to the lack of icy and practice perspective because we find that a secondary effects on physical and mental health, one simple reframing—a ‘nudge’—can have a positive possibility is that the personal gain women derived impact on women’s health and couples’ relation- from an increase in their household’s economic ships about a year after the brief training. Recall that well-being was offset by the stressors associated with W+ did not involve couples’ counseling or booster planning, launching, and maintaining a new busi- training sessions. Participants in the W+ program re- ness. As for the lack of an impact on empowerment, ceived the same amount of program contact as par- it could be the case that not involving male partners ticipants in the standard WINGS program. The only in the program limited opportunities for changes in difference being that W+ participants spent part of gender attitudes and behaviors that would lead to their training time learning and practicing commu- women’s empowerment. The second phase of this nication and problem-solving skills with partners study examined the question: Would taking a more and participated in the program as a pair with the inclusive household approach that encouraged part- goal of creating and sustaining a successful female- ner involvement reduce her daily stressors, improve led business. For practically no additional cost, we her health, and make her more empowered? see important gains in a woman’s well-being. We observed four important outcomes of experi- Third, the impact-paradox with respect to wom- mentally varying partner involvement in the busi- en’s empowerment continues: we do not observe ness: secondary effects on women’s empowerment. Building Women’s Economic and 54 Social Empowerment Through Enterprise Gender norms and attitudes are learned over a long tial profit. This might not be a bad strategy per se, time and are resistant to change. By having men par- but we cannot make a determination at this time. ticipate in the initial training, watch role-plays, and practice communicating with their partners in front Social Capital of the group, we aimed to stimulate social learn- In the last three decades, scholars and practitioners ing that would lead to behavior change. We found have generated a wealth of evidence that points to modest evidence that this process began for the W+ the importance of social networks in poverty re- couples. duction efforts (e.g., Grootaert, 1999; Narayan & Principles of operant conditioning suggest that, over Pritchett, 1997). The basic idea is that people who time, men will be reinforced for supporting their are more involved in community life have greater wives as their behaviors are reinforced by the wife sources of social support and a louder voice in com- and by the observation that their collective effort is munity decision-making. For someone engaging in benefitting the household. This process is hypoth- business activities, these connections may lead to esized to create a situation of cognitive dissonance greater economic success and act as a buffer against in which the male partner develops conflicting be- hard times. We hypothesized that WINGS would liefs (e.g., women should not be allowed to travel lead to increased social capital and that greater so- freely outside of the village vs. giving women free- cial capital—generated through the formation of dom of movement helps them to be more produc- business groups—would lead to increased business tive business partners) and then seeks to resolve the success. The evidence is mixed. dissonance by updating his beliefs. However, social WINGS beneficiaries experienced a small in- norms about gender roles and relations are power- crease in positive community support and rela- ful guides for behavior so change within households tionships, but also a small increase in commu- may depend on the strength of norms in their com- nity hostility. It could be that the program helps munities. to strengthen ties to close community members who can benefit from the beneficiary’s new ability Overall, while we do not observe changes in wom- to provide support, but strains relations with the en’s empowerment in the medium-term, it is rea- broader community due to feelings of resentment sonable to think that a program like W+ can lay the and the crowding out of existing businesses. We find foundation for longer-term change. An alternative that WINGS benefits net consumers of retail goods explanation of the results is that the ‘nudge’ wasn’t a by reducing prices, but that this reduction in prices strong enough dose of an intervention to bring last- leads to a 12 percent reduction in profits for existing ing attitudinal and behavioral change for women’s micro-enterprises. Overall, the WINGS program led empowerment. More testing of intensity and combi- to a rise in aggregate well-being of the community, nation of programs targeting communication, skills but there were adverse effects for existing traders. and norms is needed. It is not clear whether stimulating social capital Fourth, women participating with partners are by organizing beneficiaries into business groups not more economically successful. In fact they had an effect on business success: we see a dra- might be worse off in some ways. Women in W+ re- matic increase in income among those who par- ported earning one quarter less profit and consum- ticipated in the group dynamics training, but no ing 10 percent less in non-durable household goods. impact on consumption or wealth. At the same time, they also reported accumulating 1.5 times more wealth. It could be the case that part- Focusing future program objectives and ners are influencing a shift in spending towards du- targeting support rable goods. Program experience also suggests that couples might be ‘splitting’ the start-up capital rather The intervention was predicated on psychosocial than working together on one business activity. This impact and reintegration of the most vulnerable as could have the effect of reducing the woman’s poten- much as poverty alleviation for the poorest. The An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 55 findings suggest that, as designed, the intervention most while expanding the reach of the program to is mainly effective at tackling poverty and ineffective more people in need. For instance, AVSI could shift at changing levels of social integration and empow- to mobile money transfers and remote monitoring erment. This is no fault of the program – the conven- to lower costs and add beneficiaries while at the tional wisdom in humanitarian aid is that economic same time limiting expensive in-person strategies empowerment and more general psychosocial well- (e.g., accountability, advising, counseling) to those being and independence are closely tied. What this with above average expected or exhibited needs. study (and others) are beginning to show is that the relationship is not so simple. If health outcomes are key, then evidence-based medicine and mental health programming is most This implies future programs need to rethink and be promising. If the main objective of a program is explicit about their aims, and tailor their interven- women’s empowerment, then interventions that tion to that. If poverty alleviation is the primary out- target strongly held norms are likely needed. Eco- come of interest, more streamlined programming nomic-focused interventions do not seem to deliver dealing with the most binding constraint (capital) is secondary effects on empowerment or health or so- probably optimal. Efforts should be made to lower cial connectedness, but more research is needed in program delivery costs for everyone in order to de- this area. liver more intensive support to those who need it Building Women’s Economic and 56 Social Empowerment Through Enterprise Appendix: A statistics primer We set out to write a report that is free of exces- would be less than 1%. Since the probability is so sive technical jargon. Whenever possible, we tried small, we can reject the default hypothesis that the to present findings in meaningful terms, our prose program had no impact on earnings. We would say aided by several tables and figures. While we believe that the result is “significant.” This is different than that the lessons of the report can be understood saying that the magnitude of the result is large; mag- without a background in statistics, there are a few nitude refers to effect size or impact as explained concepts that are good to review. above. Taken together, significance and magnitude help us know what to make of our results. Impact or effect size Confidence intervals Simply put, impact refers to the size of the difference between groups when evaluating outcomes. In this Another way to evaluate our results is to report con- report, impact will often be stated as the average dif- fidence intervals. Statistical tests are never exact, ference between people who received the WINGS and we always have some degree of error in what we program in Phase 1 (treatment) and people who did set out to measure. Confidence intervals tell us the not (control). This difference can be framed as an ab- range of values that our finding could take. For in- solute value (e.g., increase of UGX 10,000, roughly stance, we might report that the average effect size USD 4.02) for the average WINGS beneficiary) or is UGX 10,000, but because there is some error in as a percentage (e.g., increase of 100% for the aver- this estimate, we would also report that the “true” age WINGS beneficiary). effect size falls somewhere between UGX 8,000 and 12,000 (USD 3.21 – 4.82). Often we report the 95% Statistical significance confidence interval, meaning that the “true” value would be fall in this interval in 95 out of 100 stud- You will notice in our tables that we report prob- ies. If we wanted to be more confident, say 99%, our abilities known as p-values. When p-values are less confidence interval would expand (e.g., UGX 6,000 than the generally accepted cutoffs (1%, 5%, 10%), to 14,000 – USD 2.41 - 5.62). We always want the we indicate this with one or more asterisks. P-values confidence interval to be small; the smaller the inter- help us to evaluate our study hypotheses. The de- val the more precise our estimate of impact (better fault hypothesis is always that there are no differ- surveys lead to more precise measurement, which is ences between the treatment and control groups, or why we take so long preparing!). that the program has no impact. When we detect a difference, we use the p-value to evaluate whether Average, mean and median we should reject the default hypothesis that there are no differences and conclude that the program The term “average” refers to the central value of a had an impact. group of numbers, in our case the central value for an outcome among members of the treatment group or For instance, let’s say that the average WINGS ben- the control group. The arithmetic mean and the me- eficiary earned UGX 10,000 more than the average dian are two measures of the central value. When we member of the control group and the associated p- say “average,” we are referring to the mean. Because value is less than 0.01. We would indicate this with means can be distorted by extreme values—people three asterisks in the table (***). This p-value means who do really well or really poorly—we sometimes that, if there were truly no differences in earnings be- report the median, the precise middle value in the tween the treatment and control groups, the prob- group (the 50th percentile). ability of finding impacts of UGX 10,000 or larger An Experimental Assessment of the Women’s Income Generating Support (WINGS) Program in Uganda 57 References Banerjee, A., E. Duflo, et al. (2010). “The Miracle of Microfinance? Evidence from a Randomized Evalua- tion.” Unpublished working paper, MIT. Banerjee, A. V. and E. Duflo (2005). “Growth theory through the lens of development economics.” Hand- book of Economic Growth. P. Aghion and S. Durlauf. Amsterdam, Elsevier Press. Blattman, C., Fiala, N., Martinez, S. et al. (2011). “Employment Generation in Rural Africa: Mid-term Re- sults from an Experimental Evaluation of the Youth Opportunities Program in Northern Uganda.” Innovations for Poverty Action policy report. de Mel, S., D. J. McKenzie, et al. (2008). “Returns to Capital in Microenterprises: Evidence from a Field Ex- periment.” Quarterly Journal of Economics 123(4): 1329-1372. Garikipati, S. (2008). “The Impact of Lending to Women on Household Vulnerability and Women‘s Empow- erment: Evidence from India.” World Development, 36(12), 2620-2642. McKenzie, David John and Woodruff, Christopher. “What Are We Learning from Business Training and Entrepreneurship Evaluations Around the Developing World? (September 1, 2012).” World Bank Policy Research Working Paper No. 6202. Siwale, J. and Ritchie, J. “Failure by Design: The Rise and Fall of a Microfinance Institution in Zambia – a case of Pride Zambia.” Udry, C. and S. Anagol (2006). “The return to capital in Ghana.” The American Economic Review 96(2): 388-393. Vyas, S & Watts, C. (2008) “How does economic empowerment affect women’s risk of intimate partner vio- lence in low and middle income countries? A systematic review of published evidence.” Journal of International Development 21, 577–602. Building Women’s Economic and 58 Social Empowerment Through Enterprise THE WORLD BANK 1818 H, Street N.W. Washington, D.C. 20433 www.logica-wb.org Building Women’s Economic and 60 Social Empowerment Through Enterprise