Final evaluation report of Benin’s Youth Employment project June 2022 1 Acknowledgements This report results from a close collaboration between the Government of Benin, the World Bank’s Social Protection and Jobs Global Practice and the Africa Gender Innovation Lab at the World Bank. The program was implemented by Benin’s National Employment Agency (Agence Nationale pour l’Emploi) with support from the Youth Employment Project’s Coordination Unit within the Ministry of Small and Medium Enterprises and Employment Promotion. The impact evaluation was funded by the Umbrella Facility for Gender Equality Trust Fund. The Umbrella Facility for Gender Equality (UFGE) is a multi-donor facility designed to strengthen awareness, knowledge, and capacity for gender-informed policy making. Funding is made possible through generous contributions from the governments of Australia, Canada, Denmark, Finland, Germany, Iceland, Norway, Spain, Sweden, Switzerland, the United Kingdom, and the United States. The data was collected by IREEP within the African School of Economics. The report was co-authored by Thomas Bossuroy (Social Protection and Jobs Global Practice) and Julia Vaillant (Africa Gender Innovation Lab). Authors extend particular gratitude to: ANPE’s Director General Urbain Amegbedji as well as Djibril Habib and his team at ANPE; Maxime Sogbossi, Wilfried Gbessi, Jack Joseph Segla and the entire Coordination Unit of the Youth Employment project; Leonard Wantchekon, André Guéguéhoun, Clément Litchegbe and their team at IREEP; Katrina Sharkey, John van Dyck, Joachim Boko, Solène Rougeaux and Saint-Martin Mongan-Agbeshie at the World Bank; Pierre Ouedraogo for outstanding field coordination of the impact evaluation; and Marine Gassier, Nelly Rakoto-Tiana and Ababacar Sedikh Gueye for superb research assistance. 2 Contents List of Figures and Tables............................................................................................................................. 4 Summary ....................................................................................................................................................... 5 1. Introduction ........................................................................................................................................... 6 2. Background ........................................................................................................................................... 8 2.1. The program .................................................................................................................................. 9 2.2. Targeting and enrollment .............................................................................................................. 9 2.3. Intervention: design and implementation .................................................................................... 10 2.3.1. Training ............................................................................................................................... 10 2.3.2. Grant ................................................................................................................................... 12 2.4. Delivery and frontline services ................................................................................................... 12 3. Methods............................................................................................................................................... 13 3.1. Experimental design.................................................................................................................... 13 3.2. Data collection ............................................................................................................................ 14 3.3. Empirical strategy ....................................................................................................................... 15 4. Sample description and balance tests .................................................................................................. 16 4.1. Characteristics of respondents .................................................................................................... 16 4.2. Randomization balance ............................................................................................................... 16 4.3. Attrition ....................................................................................................................................... 17 5. Results ................................................................................................................................................. 17 5.1. Primary outcomes: business performance, earnings and employment ....................................... 17 5.2. Secondary outcomes: expenditures, agency and assets............................................................... 19 5.3. Mechanisms and exploratory analysis ........................................................................................ 20 6. Dealing with attrition .......................................................................................................................... 22 7. Conclusion and discussion .................................................................................................................. 22 8. References ........................................................................................................................................... 26 9. Tables .................................................................................................................................................. 27 Appendix ..................................................................................................................................................... 33 3 List of Figures and Tables Figure 1: Sequence of interventions............................................................................................................ 10 Figure 2: Experimental design .................................................................................................................... 14 Table 1: Training participation rates ........................................................................................................... 11 Table 2: Use of childcare support ............................................................................................................... 12 Table 3: Time between interventions and follow-up data collection .......................................................... 15 Table 4: Characteristics of beneficiaries in the evaluation ......................................................................... 27 Table 5: Impacts on business performance, earnings, and employment ..................................................... 28 Table 6: Impacts on expenditures, agency, and assets ................................................................................ 29 Table 7: Impacts on self-employment and business starts, deaths, and transitions .................................... 30 Table 8: Impacts on labor and capital inputs .............................................................................................. 31 Table 9: Impacts on transfers and intra-household dynamics ..................................................................... 32 Table A1: Baseline balance test of respondent characteristics ................................................................... 33 Table A2: Attrition...................................................................................................................................... 35 Table A3: Impacts on business performance, earnings, and employment (full results) ............................. 36 Table A4: Impacts on expenditures, agency, and assets (full results) ........................................................ 41 Table A5: Impacts on components of agency ............................................................................................. 44 Table A6: Impacts on self-employment and business starts, deaths, and transitions (full results) ............. 47 Table A7: Impacts on primary outcomes conditional on having business at baseline ................................ 49 Table A8: Impacts on labor and capital inputs, full results......................................................................... 51 Table A9: Impacts on transfers and intra-household dynamics (full results) ............................................. 53 Table A10: Inverse Probability Weighing, first stage probit ...................................................................... 56 Table A11: Reweighted impact estimates at follow-up 3 (restricted to common support) ......................... 58 Table A12: Estimation of Lee bounds for each treatment arm ................................................................... 59 4 Summary Youth employment is a burning issue in Africa. With low levels of formal education and a very narrow formal sector, promoting productive self-employment is critical. Several constraints hinder the creation and development of successful businesses, including the lack of management skills and the lack of start-up capital. In response to these challenges, the Government of Benin launched the Youth employment project ( Projet Emploi des Jeunes – PEJ) in 2013. The PEJ focused on helping young beneficiaries start or expand their income-generating activities by delivering business and life skills training and start-up grants. We conducted a randomized control trial to evaluate the relative impacts of relaxing the financial capital constraint, the human capital constraint, and both simultaneously. To this end we evaluated the impacts of either delivering a $400 cash grant, a life and business skills training, or the combination of training and cash to underemployed youth in Benin. The study includes 3,444 young participants (average age 26) with low education from 15 communes and covers three rounds of follow-up data collection. We find that receiving only the training had strong and sustained impacts on business outcomes and earnings. Almost 3 years after delivery, participants who had received only the training had significantly higher profits and earnings than the control group. The increase in monthly earnings of women and men who participated in the training is US$16 and US$29 respectively, a larger amount that the unconditional cash transfer paid out as part of the Benin Safety Net program. The success of the training may be explained by a combination of factors, including i) high-quality delivery, ii) adaptation of the curriculum to the needs of the target population, iii) gender-informed design ensuring high participation of women, iv) inclusion of a substantive socioemotional skills component. Impact of each treatment arm on profits over time The grant, whether delivered in combination with the training or alone, had no effects on business outcomes. This is true for both women and men, whether they started out as business owners or not. In fact, the grant seems to cancel any positive effects of the training on business performance and earnings. However, grant recipients did see some improvements in their welfare through the increase in expenditures and assets. The lack of effect of the grant for women is intriguing, since they are investing in their business by hiring and accumulating productive assets. Suboptimal investment decisions due to pressure to redistribute may explain why women’s profits are not increasing. These results suggest that a well-designed, well-implemented training could be an important policy instrument to improve employment outcomes for the youth, and that capital support should be designed alongside accompanying measures to reduce the risk of capture and maximize impact. 5 1. Introduction Employment, especially for youth, has become a burning priority throughout Africa. Although the current generation of Africans entering the labor force is the most educated ever, many are finding that their prospects for employment and earnings differ very little from those of their parents. In a few countries, they are worse. Youth in urban areas have been vocal about their dissatisfaction. Urban demonstrations consisting primarily of politically active and disaffected youth have become more common in African capitals. Understandably concerned, policy makers in Sub-Saharan Africa are seeking policies and programs that can ameliorate the dissatisfaction of young people and ease their transition into adulthood by encouraging the creation of sustainable, productive employment. But urban youth are only the most visible and audible part of the employment problem. The majority of young people still live in Africa’s rural areas and small towns. Poorer and less educated than their urban counterparts, they too struggle to find pathways to adulthood, especially to stable, remunerative employment that allows them to support a family. For young women, the pathway can be especially treacherous. As they navigate the school-to-work transition, their control over their own destiny and their employment choices may be limited by social norms that define what is acceptable and expected of women and men in society. The challenge of youth employment in Africa may appear daunting, yet Africa’s vibrant youth represent an enormous opportunity, particularly now, when populations in much of the world are aging rapidly. Youth not only need jobs, but also create them. In response to these challenges, the Government of Benin launched the Youth employment project (Projet Emploi des Jeunes – PEJ) in 2013. Besides supporting the apprenticeship and skills certification systems, the PEJ focused on helping young beneficiaries start or expand their income-generating activities. In the absence of formal wage employment, the majority of the population is self-employed by default rather than by choice, and run informal, micro businesses with very low productivity and profitability and limited prospects for growth. The design of the project rests on the assumptions that both skills and capital constraints prevent youth from creating and developing successful small businesses. A multi-part training program was designed to enhance different dimensions of skills: socio-emotional (through a life skills training), entrepreneurial (through a micro-business training), behavioral (through group-based dynamics and several follow-ups by professional trainers). The project also aimed to relax credit constraints. Starting a small business requires some initial investment in equipment, machines, marketing, or inventory. With widespread failures in the credit market, the cash constraint prevents youth who wish to start their own activity from doing so. In this context, a large enough unconditional cash transfer was designed to help participants create and sustain a business. The evidence on the impact of business skills training in developing countries is reviewed in McKenzie and Woodruff (2013) and was updated in 2020. The 2013 review highlights that the impact of traditional business skills training programs on employment is minimal but they do seem to have success in generating business start-up, and a few studies find significant impacts on profits or revenues of enterprises (e.g. Calderon et al., 2012; Valdivia, 2012) but often the effects are too small to be detectable. However, the 6 review notes that most impact evaluations lack the power to detect small effects of business profits or sales. Low take-up is a frequent reason for low statistical power, with take-up rates around 65% on average. In the most recent meta-analysis conducted by McKenzie and Woodruff (2020), traditional business trainings are found to have modest but significant impacts on revenue and profits (5 and 12% average increase respectively). McKenzie and Woodruff (2013) argue that the reason for modest impacts is that entrepreneurs only apply one or two business practices out of the 20 or 30 that are taught in these programs. A growing literature examines the impacts of psychology-based and soft skills-based trainings on entrepreneurial success. Personal initiative training for example has positive effects on profits in Togo (Campos, 2017) but quality of implementation seems to be an important factor of success of the training (Jayachandran, 2020; Alibhai et al, 2019). The evidence base on the impact of adding soft skills and psychology-based elements to traditional business trainings is thin (Jayachandran, 2020). A recent study from Jamaica finds that combining soft-skills training with regular business skills training may have worse effects than delivering soft-skills training only (Ubfal et al, 2022). In Uganda, different combinations of soft and hard skills yielded the same large impacts of profits (Chioda et al, 2021). A related literature considers the impact of cash grants on existing and aspiring microentrepreneurs. The evidence on the effects of these types of programs is even scarcer, as there are only a few rigorous impact evaluations of cash grant programs in developing countries. Studies in Sri Lanka and Ghana of the effect of cash or in-kind grants to microentrepreneurs show positive impacts on profits, although these effects are heterogeneous by the nature of the grant and the gender of the entrepreneur (De Mel et al., 2008; Fafchamps et al., 2014). A few studies evaluate the impact of the combination of business skill training and cash grant, with mixed results: in Sri Lanka the combination of the training and grant has large but short lived effects on business profitability, while the sole training has no effect on business outcomes (De Mel et al, 2014); in Uganda short basic business skill trainings and cash grants to poor women doubled business ownership and income (Blattman et al, 2013). However these impact evaluations do not evaluate the grant and training components separately. Berge et al (2015) evaluate the relative impact of a business training program, business grants, and, the combination of training and grants among existing microentrepeneurs of Dar es Salaam in Tanzania. They find that the combination of training and grants improved male entrepreneur’s business outcomes but either of the interventions alone had no impact. None of the interventions had any effect on women’s business outcomes. Fiala (2018) finds no impact of grants on female or male business owners semi-urban Uganda, whether combined with a business training or not. Giné and Mansuri (2021) study the effect of a business training alone, a microfinance loan, and the combination of the two in Pakistan. The find that the training improved business knowledge and practices, but had no effect of business performance outcomes. There are few studies that compare the relative effects of relaxing the human and financial capital constraints and that examine how they interact with each other. A first motivation of the impact evaluation was therefore to assess whether access to capital or access to skills is the main binding constraint, and how they interact with each other. A second motivation of the impact evaluation was to identify the gender specificities of those impacts. Young women and men are not equally constrained on the labor market. First, young women are on average less educated than men, due to inequalities in access to formal education. In the evaluation sample, 30 percent of women never went to 7 school against less than 10 percent of men, and they have 2.5 fewer years of education on average. Second, gender norms impact non-cognitive skills such as aspirations and self-esteem of young women, which might affect entrepreneurial skills. Third, decision-making power is much lower for young women, who are often subject to decisions made by their husband or parents relative to their participation in the labor force or their choice of occupation. For example, our surveys show a 30 percentage point difference between the share of women and the share of men who report independent decision making on income or large purchases. Fourth, the time spent on household activities such as childcare or daily chores is a major impediment to women’s capacity to seize opportunities and focus their attention on productive activities. In our sample, 85.8 percent of women report childcare as a constraint to economic activities, against 9.9 percent of men. These specific constraints might affect the impact of the training and the grant. This design of this impact evaluation overcomes some of the methodological limitations found in previous studies. Firstly, it is sufficiently powered both to detect small impacts on earnings and profits and to estimate to estimate heterogeneous gender impacts. Previous studies have frequently not been able to detect effects among women or to compare effects between men and women, due to low statistical power. Secondly, by conducting three rounds of follow-up data collection 15, 27, and 45 months after the end of the training, it is able to detect medium to long term effects of the program and to examine the trajectory of impact over time. Many similar studies only collected one round of data collection, and past studies have not typically followed participants beyond 12 to 24 months. 2. Background The labor force in Benin is young, predominantly rural and working in the informal sector (81 percent of the population is self-employed). While unemployment is low (only 0.7 percent in the strict ILO definition), underemployment is severe, especially among the youth. Underemployed workers are disproportionately female, rural, self-employed, with low levels of education, and engaged mainly in agriculture, livestock, fishing and forestry. Benin’s demographic structure implies that its workforce is very young and that there is therefore very strong pressure on the labor market and the education system. Around 200,000 Beninese youth attain working age every year. Underemployment is a major challenge with 72 percent of employed workers employed less than 35 hours and/or in a lower paid job than the minimum guaranteed wage. The rate of visible underemployment is 35 percent and the invisible underemployment rate is 62 percent. Nine out of 10 young people (15–25 years old) are underemployed, as are nine out of 10 rural women. The agricultural sector remains the main provider of employment (42 percent of jobs) followed by the trade sector (19 percent) and manufacturing activities (15 percent). Women in Benin are particularly vulnerable and consistently work in lower paid jobs than men, and also with incomes on average two times lower than those of men. One-third of working age women remain outside the labor market, mainly due to the burden of domestic chores which they typically manage single- handedly. In rural areas, 44 percent of women work less than 35 hours a week. More than 20 percent of rural women work without pay, compared to only 10 percent rural men and 6 percent urban men. 8 However, employment promotion programs mostly focused on urban, educated and predominantly male youth. In 2010–2011, before the Youth Employment project, the flagship program of the National Employment Agency (ANPE) benefitted about 800 individuals, of which 84 percent were men and 80 percent had some secondary education or more. The programs were all delivered in one of the six urban centers of the country. 2.1. The program The Benin Youth Employment Project (Projet Emploi des Jeunes or PEJ) was a national program run by the Government of Benin from 2014 to 2019 supporting 17,500 youth, ages 18 to 35. The objective of the project was to improve access to employment skills and opportunities for underemployed youth in Benin by developing their technical and business skills through apprenticeships and business training, start-up support, and institutional capacity building. The aim of the PEJ was to deliver impactful employment services to populations deprived of opportunities and with limited capacity to develop robust economic activities on their own, due to a lack of information, skills or capital. The PEJ was therefore designed as a national program focused on all 77 communes in Benin, spanning various local contexts, adapted to the specific needs of the target populations, and delivered at a highly decentralized level to ensure participation of rural beneficiaries. The project has particular sectoral focus on artisanal trades, agricultural transformation, and tourism. The key beneficiaries of the project were the young underemployed men and women of Benin between the ages of 15–35. Underemployed youth in this instance were defined as those who reported either: (i) working less than they would like; or (ii) working full-time but earning less than the minimum wage or poverty level; or (iii) working in a job that does not match their education, training or experience. The project placed emphasis on targeting hard-to-reach, poor youth not served by other programs and most affected by underemployment- that is, female, rural, self-employed and/or those with low levels of education. The geographic distribution of beneficiaries was strictly proportional to the size of the population and the program allotted beneficiary quotas to each commune of the country. 2.2. Targeting and enrollment The eligibility criteria for the interventions evaluated included being 18-35 years old, having an education level equivalent or lower than the middle school certificate, and not being currently in school. The operation started with a national communication campaign on the application process, including local radio messages, roadside billboards, and traditional town criers (gongonneurs) to spread awareness on the eligibility criteria and the enrollment process. Enrollment was highly decentralized. Mobile application booths were set up at the arrondissement1 level to facilitate access by remote populations and women who typically have lower mobility. Over several days, young men and women all over the country were able to submit their application to the program in a nearby locality. As the project aimed to serve 50 percent women, the application process was stratified: at the time of enrollment, young men and women formed two separate lines and were registered alternately until the end of the application period. 1 Arrondissement is the administrative level below commune. 9 A total of 73,700 applicants registered across 77 communes, massively outnumbering the 17,000 open positions in the program. To select candidates into the training, the government used an innovative random selection process. Upon submission of their application, each candidate picked a scratch card with a concealed unique number. The order of the scratch cards was random. The registration agent would record the name, contact and the unique number that the applicant revealed on their card. At that point, applicants didn’t know if they have a winning number, which avoided any pressure on enrollment agents. After full completion of the enrollment process, a lottery determined whether the lowest or the highest numbers would be selected for the program. The lottery was broadcast on national television to highlight the transparency of the selection process. 2.3. Intervention: design and implementation The intervention evaluated here was run by the National Employment Agency and consisted in providing two different trainings and an unconditional cash grant to help youth start or expand a sustainable business. The program was sequenced to leave time for entrepreneurial activities between modules, thereby facilitating links between theory and practice and encouraging experiential learning (Figure 1). Life skills Generate Start Your Improve workshop Your Business Your (ACV) Business and simple Business Follow-up Follow-up Final Idea biz plan home home group 6 days 2 days 3 days 3 days visit 1 visit 2 meeting Cash grant Figure 1: Sequence of interventions 2.3.1. Training The trainings consisted in a socioemotional skills training and a business skills training delivered by private training providers under contract with the Government of Benin. - The first training was a six-day socioemotional skills training called Life Skills Workshop (Atelier Compétences de Vie or ACV), designed specifically for the PEJ. The aim of ACV was to improve trainees’ attitudes (self-awareness, self-esteem, stress management, values and “good� behavior), interpersonal skills (communication, assertiveness, peer-pressure management, networking, leadership, gender relationships, some sexual and reproductive health), and personal initiative (goal-setting, decision making and problem solving, financial literacy and money management, risk taking). - Soon after the end of the ACV, participants underwent the Start and Improve Your Business (SYIB) training program developed by the International Labor Organization and implemented at a large scale globally. The PEJ used the Level One version of this training, which is tailored to non- educated populations and relies on the use of images, vignettes, games and role plays. SIYB is composed of three modules. 10 o The first module, Generate Your Business Idea, takes two days and focuses on carrying out a self-assessment and identifying a promising business idea based on market needs, personal skills and resources available. o The second module, Start Your Business, takes three days and aims to get trainees to think through the technical feasibility of their idea, its potential profitability and the main actions necessary to reach the goals. It ends with the development of a simple business plan, for which trainees receive a half-day individual counselling session by the trainer. - Approximately one month after the grant is received, beneficiaries regrouped and attended the three-day Improve Your Business module which covers more advanced business management notions: marketing strategy, stock management, basic accounting and cash flow management (including managing claims on resources from relatives), sales strategy, planning of activities. - At least two weeks after the Improve Your Business module, the trainer visits the beneficiaries at home or in their workplace to discuss the constraints they are facing, give them advice on progress, coach them on using a simple accounting tool they were trained on, and help them apply skills learned during the training. A second individual visit takes place one month later, with similar objectives. A final session is then organized for the entire training group to debrief on success stories and failures, one month after the second visit. The trainings were delivered in sessions of 25 participants maximum (average 19) organized at the arrondissement level, such that they were accessible locally to a large share of the population with limited travel costs. Midday meals were procured from local catering businesses and provided to all participants to limit half-day dropouts and improve the learning experience. All participants received 2000 FCFA every day (about 4 USD) to cover their transportation costs. Participants also received a small kit made of stationery and training material. To facilitate the participation of women, trainees with young children needing childcare were invited to bring a second person who would look after the children. Both the trainees and their ‘babysitter’ were offered transportation and a midday meal. Table 1: Training participation rates Men Women Attended at least one session 98.8% 98.8% Missed zero session 93.6% 95.7% Missed only one session 3.4% 2.8% Missed two sessions or more 3.0% 1.4% Would recommend training 98.8% 98.4% Received zero visit by trainer 21.0% 21.5% Received one visit by trainer 30.8% 37.8% Received two visits by trainer 32.6% 26.8% Received three or more visits by trainer 15.7% 13.8% Note: impact evaluation sample Table 1 shows participation rates self-reported by survey respondents at the first follow-up data collection. The survey was conducted among the impact evaluation sample only, which covered 15 of the 77 communes. Participation to the training sessions was very high for both men and women, with around 98 percent of participants reporting that they completed the training with zero or one session missed. Participants then report receiving on average two to three home visits by the trainer. These responses 11 broadly correspond to the program data (attendance sheets) that was collected along implementation, thereby alleviating concerns of overreporting. Table 2: Use of childcare support Proportion N Women participants with childcare duties who were aware they could come 90% 704 along with a ‘babysitter’ Women participants with childcare duties who did come along with a 61% 704 ‘babysitter’ Note: impact evaluation sample The childcare support facility was widely used (Table 2). About 90 percent of women participants with childcare duties had received the information that they could come to the training with a second person to attend to their child, and more than 60 percent did take advantage of this opportunity. This contributed to the high level of participation by women. 2.3.2. Grant Participants were then offered an unconditional cash grant to help them start their business. To apply for the grant, they were required to submit the abovementioned business plan, which outlined the use of the grant and specified the amount requested. Participants could ask for up to 200,000 FCFA (about 400 USD). They were eligible to receive the grant as long as their business plan complied with simple submission requirements. The quality or chances of success of the business plans were not assessed by the program and were not used as a condition for disbursing the grant - resources were available to provide grants to all participants. The grant was delivered in one tranche by mobile money after SIM cards were delivered by the payment operator to all cash grant recipients. Although the cash grants were delivered with delay, 98% of individuals offered the cash grant received it in a reasonable timeframe. Not surprisingly, most participants wrote up business plans for an amount close to the ceiling of 200,000 FCFA. 2.4. Delivery and frontline services The design of training curricula was contracted to CESAM, a training firm acting as the technical focal point and coordinator for the ILO in the sub-region. CESAM designed the contents, produced and printed training material, trained the trainers and provided support and quality control during implementation. The delivery of the trainings was contracted out to private training firms. Since the methodology and curricula were based on ILO’s SIYB modules, trainings had to be delivered by ILO-accredited professional trainers. Benin, as most countries in the region, has a vast network of accredited ILO-professionals who work in private training firms. The government therefore launched an open procurement of training firms, split by geographical lots, with the requirement that training providers needed a valid certification in the ILO methodology. To ensure sufficient numbers of certified training providers, firms could include half the 12 cost of re-certifying former trainers in their financial proposals. A total of 18 training firms were contracted to deliver the training. The central training design firm CESAM organized week-long trainings of the trainers in a central location in Cotonou. While the SIYB Level 1 methodology was already familiar to ILO-certified trainers, the ACV curriculum had to be taught and practiced by training providers before implementation. Each training firm then deployed trainers to the communes covered by their contract. The provision of training was done at the arrondissement level, as close to beneficiaries as possible. The cash grant was delivered through mobile money. A mobile payment operator, MTN, was selected by the government through a competitive bidding process. Before payments, MTN fielded outreach agents to collect contact information from beneficiaries and updated phone numbers. MTN provided SIM cards to all beneficiaries didn’t have mobile phones. The delivery was coordinated by Benin’s employment agency ANPE. A team based in Cotonou was responsible for managing contracts of CESAM and frontline training firms. The regional offices of ANPE provided logistical support to the training firms and some supervision of training activities. Additional supervision and follow-up was done by local staff known as ACE (Agents Communaux d’Emploi). ACEs worked as part of municipal services and were responsible for supporting project activities in everyday monitoring and overall quality assurance. They served as an active network of delivery agents who could interact with beneficiaries, training center staff, municipal authorities and training firms to ensure coordination and coherence in the program. 3. Methods 3.1. Experimental design The study sample comes from 15 communes of the 77 communes in which the project was implemented. These communes were selected for the following reasons: i) practicality: only communes belonging to départements in the southern part of Benin were included to keep data collection and monitoring of compliance costs reasonable, ii) contribution to poverty: in each included department, the three communes with the highest contribution to extreme poverty were selected. This ensured that the results would be valid for poorer communes of the central and northern parts of the country. A set of 3,444 eligible applicants was randomly selected in those 15 communes, following stratification by commune, by gender, and by priority status (having been an apprentice with a master craftsman or having completed a technical training or being in one activity of the priority sectors defined by the project). To identify the effect of the grant component, the training component, and the combination of the grant and training, the 3,444 eligible applicants in those 15 communes were randomly assigned to one of four groups: i) grant and training (T1), ii) training only (T2), grant only (T3), control, in which participants received neither the training nor the cash grant ( Figure 2). 13 All eligible applicants Trainings No training T1 = Training + Cash grant T2 = Training only T3 = Cash grant only Grant No grant Grant No grant C = Control T1 T2 T3 C Nt1=861 Nt2=861 Nt3=861 NC=861 Figure 2: Experimental design The first step was assignment to training. Randomization into the training treatment arms was done through the scratch cards, as described above. Unlike the general program though, the order of selection of numbers was not announced on public television but done through a desk randomization. After the training was completed, in person public lotteries were organized to assign applicants to receiving the cash grant or not. Lotteries were stratified by training treatment arm. The design of the cash grant component was simplified for the impact evaluation sample, as compared to the version presented above. To avoid potential selection effects, the following modifications to the general program were made: i) To avoid systematic differential selection into the cash grant component, applicants were not required to submit a business plan or to have completed the training in order to enter the lottery; ii) The amount of the cash grant in the impact evaluation sample was fixed to FCFA 200,000 in order to avoid systematic differences in the amounts requested that would stem from having received or not the training. The evaluation was designed to be able to evaluate heterogeneous impacts by gender of the beneficiary. Pre-baseline power calculations indicated a higher coefficient of variation for women’s earnings. Thus to ensure that the female sample was sufficiently powered, women were oversampled in the experiment, and represent about 60% of the sample. 3.2. Data collection The baseline survey was conducted in July/August 2017 to collect detailed data among individuals eligible to the project activities. The trainings were conducted in September-December 2017, and the disbursement of cash grant was in March-June 2018. McKenzie (2012) demonstrates that for noisy outcomes with low autocorrelation, such as profits or income, power can be increased by taking multiple measurements on the considered outcome at short intervals, which averages out noise. Therefore, for each round of follow-up data collection that were initial planned, monetary variables were collected twice, first in a face-to-face survey, then in a phone survey 5 months later. The first follow-up data collection was thus split into two phases. The first phase was a face-to-face survey and was conducted in November 2018. It collected detailed information on income, profits, expenditures, agency, investment, employment and skills. The second phase, conducted in March 2019 consisted in a short questionnaire administered by telephone. The short phone survey focused on monetary variables only, such as profits, earnings, and personal expenditures. The same structure was used for the second follow-up 14 survey, for which phase 1 and 2 were conducted in November 2019 and March 2020 respectively. An additional round, not initially planned, of follow-up data was collected by phone in July 2021. The successive rounds of follow-up data collection allow us to assess impacts of the program approximately 15, 27, and 45 months after the training, and 8, 20 and 38 months after the delivery of the grant. Table 3: Time between interventions and follow-up data collection Time after training Time after grant Follow-up 1 +15 months +8 months Follow-up 2 +27 months +20 months Follow-up 3 +45 months +38 months 3.3. Empirical strategy The randomization strategy allows us to estimate the effect of each treatment, training, grant, and training + grant, without selection bias. To estimate the effect of each intervention on our primary and secondary outcomes, we estimate the following ANCOVA specification, controlling for the stratification variables used in the randomization. Standard errors are clustered at the individual level. As noted above, the randomization was stratified by commune, by gender, and by a third variable capturing whether the candidate was considered a priority applicant. The core specification is also designed to disaggregate the effect of the treatment both overtime and by gender, which leads to the following equation: In which: - 𝒀𝒊𝒕 is the outcome measure considered for individual i at follow-up 1, 2 or 3; - 𝒓𝒎 is a categorical variable capturing the phase of data collection (m=1 is phase 1 of follow-up 1, m=2 is phase 2 of follow-up 1, etc.); - Group1/Group2/Group3 capture the treatment assignment; - 𝑹𝒕 is an indicator variable for the round of data collection; - 𝑾𝒊 is an indicator variable equal to 1 if respondent i is a woman; - 𝝅𝒔 is a vector of stratification variables. The specification, primary and secondary outcomes were pre-specified in a registered pre-analysis plan2 (PAP). This paper follows the PAP with some deviations, the most significant ones being the addition of a performance score for business outcomes and a third follow-up data collection. A detailed description of deviations from the PAP will be included in an upcoming working paper. 2 Bossuroy, Thomas and Julia Vaillant. 2019. "Addressing capital and skills constraints to youth self-employment in Benin." AEA RCT Registry. November 19. https://doi.org/10.1257/rct.2327 15 4. Sample description and balance tests 4.1. Characteristics of respondents Key socio-economic and employment characteristics of beneficiaries are summarized in Table 4. Respondents are 26 years old on average, with men and women very close in age. However, women are more often partnered (72.7 percent) and have children (72.8 percent) than men (57.9 and 53.2 percent respectively), reflecting a younger age at marriage. Educational attainments are quite different between men and women. Only 9.5 percent of men in the sample never went to school, against 30.7 percent of women. The average number of years of education is 6.6, meaning that a majority of beneficiaries did not complete primary school and reflecting the targeting criteria (education should be lower than high school). Education levels are significantly different across genders too, with a 2.5 year gap (5.6 for women vs 8.2 for men). Almost two thirds (65.5 percent) of men and women in the sample were, or still are apprentices. Among them, 58% have received some form of certification marking the completion of their apprenticeship, whether it is a traditional diploma from the master craftsman or a formal, government sanctioned diploma.3 In addition, 15.5 percent of the sample received some other professional or technical training. This reflects both the project targeting, which prioritized individuals who had been in apprenticeship and technical training and that apprenticeship is a common way to acquire technical skills for youth who drop out of formal education in Benin. The overwhelming majority of beneficiaries in the sample worked in the past 7 days (85.5 percent). We define an “insufficient employment� variable, which combines unemployment and precarious work situations, that is being underemployed (working less than 40 hours a week) or in vulnerable employment (working as an unpaid apprentice or unpaid contributor to the household's business). When considering this definition, 68 percent of the sample is insufficiently employed. Women are likelier than men to be unemployed or in a precarious form of employment. For the 78.5 percent of the sample that is self-employed, their businesses are small with 1.7 employees on average. Gender differences in business size are important: men have 2.3 employees on average in their business compared to 1.3 for women, and men-run businesses have productive assets worth more than three times the value of productive assets in women-run businesses. Large differences are also observed in total earnings and expenditures. While average earnings amount to 32,000 FCFA per month, men earn 35 percent more than women (37,800 vs 28,000). 4.2. Randomization balance Table A1 presents balance tests for survey respondents of each of the four treatment groups. Globally the randomization was successful. We find a few significant differences in characteristics between treatment 3 In the Beninese traditional apprenticeship system, receiving a certificate marking the completion of an apprenticeship is not sufficient to separate from the master craftsman and start one’s own activity. It is typical for the master craftsman to request that the apprentice pay a “liberation fee�, after which they will be free to become a competitor. Although the government of Benin has attempted to put an end to that practice, it was still in effect when the project started. In addition, apprentices need capital to start their own activity. In the baseline sample, we find that among those who had received an apprenticeship certificate, 15% were still working as unpaid apprentices in the past 6 months. 16 groups. Men in the grant-only group are younger and have fewer dependent children than in the control group. Women in the control group have less years of education on average than women in T1. We see no significant differences between treatment groups in any of the primary outcomes. 4.3. Attrition Of the 3,444 individuals included in the baseline survey, we tracked 93.2% at follow-up 1, 95.4% at follow- up 2, and 83.4% at follow-up 3. Follow-up 1 and 2 were face-to-face surveys, during which the survey firm made significant efforts to find respondents. This effort was supported by the local project staff mentioned above (ACE) who were in close contact with the communities and were able to help enumerators locate and convince respondents to participate in the survey. Due to both the coronavirus pandemic and resource constraints, the third follow-up was a phone survey with no face-to-face interactions with respondents. Enumerators only had their phone numbers from previous phases to track respondent, which explains the higher attrition rate in that phase. The analysis of attrition correlates shows that at follow-up 1 and 2, there was no differential attrition by treatment arm. However, at follow-up 3, we observe significantly higher attrition in T1, the group that received the training and the grant, and in T3, the group that received the grant only. We also see that women are significantly for likely to not have been recontacted and that working in a priority sector reduces the chances of attrition. Given the low rates of attrition in the first two follow-ups, we are not concerned with attrition bias. However, since attrition is correlate with treatment at follow-up 3, we conduct a bonding exercise and re-estimate impacts using inverse probability weighing and find that our results are robust to moderate assumptions about the sample. 5. Results The success of the interventions is determined by considering their effect on primary outcomes, which include business performance indicators, earnings, and employment. Looking into secondary outcomes provides additional information on how these interventions transformed beneficiaries' lives through other channels, such as expenditures, acquisition of assets, and agency. Then we will consider a range of intermediary variables that can help uncover mechanisms of impact of the interventions. These include business entry and exits, labor and capital inputs into the activity, and inter- and intra-household transfers. Results tables Table 5-Table 9 show the estimated impact of each treatment (T1: training + grant, T2: training only, T3: grant only) at follow-up 3 (about 38 months after the end of the training and 45 months after the reception of the grant), separately for women and men. The bottom panel shows the p-values of equality tests of coefficients for the impact of each treatment on men vs. women, and of equality tests of each treatment arm against each other for women and men respectively. Full results for all three rounds of follow-up are shown in the Appendix, along with all pre-specified primary outcomes. 5.1. Primary outcomes: business performance, earnings and employment Table 5 shows the estimated impact of each treatment on business performance indicators, earnings, and employment. We see strong positive and significant impacts of the training alone (T2) on business performance, as measured by the synthetic score, for both men and women (col. 1). There are no significant differences in impact between men and women. When the grant is added to the training (T1), impacts are 17 much smaller in magnitude and no longer significant. Tests of coefficient equality show that T1 and T2 have significantly different impacts for women, however one cannot reject equality of coefficients for T1 and T2 for men. The grant alone (T3) has no impact on business performance and significantly underperforms the training arm for both men and women. Examining profits, revenues and earnings separately (col. 2-4) shows the same pattern: a significant increase for women and men who received the business and life skills trainings. For women, the increase is close to a doubling in profits, compared to the control group, while for men profits increase by about 53 percent. This leads to increases in monthly earnings by 16,000 FCFA (US$ 28.8 as of July 2021) for men and 9,000 FCFA (US$ 16.2) for women. There are some signs of impact of T1 for men: an increase in revenues significant at the 10%, and an increase in total earnings. The grant arm only has no impact on revenues, profits or earnings for either women or men. Examining the dynamics of impacts over time reveals some interesting patterns and gender heterogeneity (Table A3). The grant alone arm (T3) had strong short term negative impacts on women’s business performance, which significantly attenuated over time. These negative impacts are consistent across all components of the business performance score, including profits and revenues, and on earnings, irrespective of the correction method used. By contrast, no negative effects of the grant were observed for men in the short term. Estimates for men are rather close to zero and non significant, and impact coefficients for T3 estimated for men and women are significantly different in the first two follow-ups. When combined with the training (T1), the grant does not generate negative impacts for women. Figure 3 below illustrates those dynamic patterns. It shows the mean value of winsorized profits (at the 99% level) over the three rounds of follow-up, for men and women separately. It is remarkable that the impacts of the training (T2) grow over time, rather than attenuate, leading to a steady growth in profits in that particular treatment arm, both for men and women, even when other groups display some more variability. For women, the differences between coefficients estimated at each follow-up are significantly different from zero. Figure 3: Impact of each treatment on profits over time These striking results do not translate into impacts on our measure of (insufficient) employment. The grant only arm decreases women’s insufficient employment by 5 percentage points. None of the other arms have a significant effect for women or men. It is notable that the grant arm, which had no effect on earnings and business performance for women in the medium term, is the only arm to also increase women’s 18 employment. Examining the dynamics over time show that men’s employment was never impacted by any of the treatments (Table A3 in Appendix). On the other hand, the improving effect of the grant on women’s insufficient employment was observed already at the first follow-up. The training only arm also had an immediate, if short lived, effect on women’s employment. 5.2. Secondary outcomes: expenditures, agency and assets Table 6 shows the impacts of the PEJ on secondary outcomes: personal expenditures, agency and assets. While the first two outcomes were measured in all data collection rounds, the asset module was left out of the phone-based follow-up 3 to shorten survey duration and maximize response rates. Results on assets are therefore shown at follow-up 2. We first look at impacts on personal expenditures, measuring expenses made by the respondent for his/her own use. The combination of training and grant (T1) has significant impacts on personal expenditures for both men and women (col.1). Impacts of the training alone (T2) are also positive and significant, but to a lesser degree than when combined with the grant. The grant alone (T3) has a strong impact on expenditures for women, but no impact for men. For men, these results are very consistent with impacts of each treatment arm on earnings. For women though, the grant, either alone or combined with training, helped improve expenditures whereas it had no impact on earnings. It is possible that women used a portion of the grant for expenditures rather than to invest in their economic activity only. This would imply, though, that women were able to save a large proportion of the grant, in order to still see this impact 3 years after the grant was delivered. We now turn to agency, measured by a score including decision-making power, self-efficacy, locus of control, and depression. Table 6 (col.2) shows that the training arms had a positive and significant impact on women’s agency. The impact is stronger for T2, but also positive for T1. There was no impact for men though. The grant alone (T3) had no effects on agency for either men or women at follow-up 3. The impacts on components of the agency score (Table A5 in Appendix) show that the grant had negative effects on respondent’s mental health measured by a depression score, 3 years after the grant was delivered. For women, the positive impacts of training arms are driven by self-efficacy, decision-making power and control. Impacts on assets, measured at follow-up 2, are only positive for men receiving the combination of training and grant (T1), driven by an increase in both individual and household assets. Full results from all phases displayed in 19 Table A4 in Appendix reveal interesting dynamics of impact. On personal expenditures (col.1), the impacts of arms including training (T1 and T2) appear muted in the short run but increase over time. The impact of T2 even becomes significant only at follow-up 3, for both men and women. By contrast, the impact of the grant appears strong in the early phases and attenuates over time – except for women. The effect of arms including training (T1 and T2) on agency (col.5 of 20 Table A4 in Appendix) was observed at follow-up 1 for women already and was sustained over time. The grant had a short term positive effect on women’s agency but that effect was short lived. The training arms initially improved men’s agency score in the first two follow-ups, but that effect was smaller and no longer significant by follow-up 3. In follow-up 1 we also see a positive effect of T1 on women’s individual assets, which attenuated over time and is not significant anymore at follow-up 2. 5.3. Mechanisms and exploratory analysis Two key findings emerge from the above analysis: the training leads to a substantial increase in profits and earnings for both women and men, and the grant has a null to negative impact, especially for women. In this section we examine intermediary outcomes and investigate potential mechanisms to shed light on these results. To explain what is driving those results, we look at business entry and exit, inputs and investments, and intra-household dynamics. Table 7 shows largely positive results of the program on the development of businesses, as measured by business starts, deaths, and changes. We see a striking increase in the likelihood of owning a business for all three treatment arms, for both women and men (col.1). Receiving the training or the grant or both increases the likelihood of owning a business by 5 to 7 percentage points, depending on the training arm. This is a substantial increase in the extensive margin, as a vast majority of the control group already owns a business at follow-up 3 (around 88 percent of women, and 85 percent of men). In addition, those who own businesses also own more businesses than the control group (col.2). The strongest effect here is seen in T1, the combination of training and grant, in which beneficiaries have on average 0.2 additional businesses than the control group. In other words, for every five beneficiaries, one has an additional business. Only the grant only arm does not increase the number of businesses owned for men. The program also reduced the likelihood that a business identified at baseline is no longer in operation at follow-up (col.3). All three treatments decreased the likelihood of business death for men. The point estimates are all negative for women too, but only the training arm (T2) displays a significant impact. Finally, the program did not lead to more changes in the sector of operation of the business. Those changes are common, with 40 to 44 percent of the control group reporting changes between baseline and follow-up. If anything, the program may have stabilized businesses (with negative point estimates on the likelihood to change for all treatment arms for both men and women). Overall the program has occasioned an expansion of business ownership, both at the extensive and intensive margins, and a stabilization of businesses run by beneficiaries at baseline. It is worth noting that these effects are consistent across treatment arms and gender, implying that the negative impact of the grant on profits for women cannot be explained by the survival of poorly performing businesses, the shift to new businesses that take time to become productive, or potentially disruptive changes introduced to existing businesses. We now analyze how the program affected the level of investment by beneficiaries into their businesses. Table 8 first shows impacts of the program on labor inputs in the business (col. 1 and 2). While the program had no impacts on men’s hiring, the training arms did increase the number of employees in women’s businesses. Women who received the grant and training had on average close to 0.5 additional employees working, and 0.36 of them were paid employees. For every three women beneficiaries, one paid job is created and sustained 38-45 months after the end of the program. 21 Table 8 shows how the program impacted the capital stock in businesses (col. 3). The training+grant combination had a significantly and positive effect on the stock of capital in men and women’s businesses. The components of capital stock, displayed in Table A8 in Appendix, show that these impacts are driven by the accumulation of productive assets rather than inventory. While the other arms had no impact on capital stock for men, they had significant yet smaller impacts for women. These impacts on capital stock result from cumulated capital investments over time, which are shown at follow-up 3 in Table 8 and for all phases in Table A8 in Appendix. T1 led to capital investments over the three phases of follow-up for men, and over the first two phases for women – accounting for the strong impacts of T1 on capital stock. The training only arm (T2) showed small non-significant impacts on investments at first, but impacts increased across the three phases. Conversely, the grant only arm (T3) had significant impacts on capital investments in the first follow-up for both men and women, followed by more inconsistent impacts in subsequent follow-ups. The inventory component of capital stock increased significantly in the short term for women who received the cash grant only and that effect disappeared over time. These contrasting patterns led to more muted impacts on capital accumulation from T2 and T3. It is noteworthy that men who received only the cash grant did not increase the value of their capital stock relative to the control group, whether in the short or the medium term. Reported business practices improved for men and women who received the training (Table 8, col.5). Results are shown at follow-up 2 as business practices were not collected at follow-up 3 to reduce the duration of the phone survey. The positive results for the arms with training (T1 and T2) correspond to 1.5 and 1.2 additional business practices reported for men and women respectively, or a 20% and 15% increase in the number of practices. Table A8 in Appendix shows that these impacts were observed at follow-up 1 already and were sustained over time, with a slight decrease in magnitude for women between follow-up 1 and 2. We finally discuss impacts of the program on transfers to and from the beneficiary as well as intra- household transfers for the purposes of household expenses (Table 9). The different arms had quite contrasted impacts on transfers, with clear gender differences. The arms with grant (T1 and T3) led to an increase in transfers out, for both men and women (col. 2). They also have a negative impact on transfers received by women, especially visible at follow-up 1, but also on transfers received by men except in follow-up 3 (see Table A9 col. 1). The net impact on transfers (subtracting transfers out to transfers in) is clearly negative for women, and marginally negative for men. The training arm (T2), however, has to a significant impact on the amounts received by men, and a non-significant but marginally positive increase for women at follow-up 3 (but a positive and highly significant impact at follow-up 2, see Table A9). Impacts on transfers out (made by the respondent) in the T2 arm are slightly positive but of low magnitude and marginally significant. Overall, T2 leads to positive net transfers for men, and has no impact on net transfers for women. This suggests that beneficiaries who received the grant generally distributed some of their benefits (especially women) while those who received the training rather benefitted from additional financial support (especially men). Women generally redistribute more than men and benefit from less financial support. At follow-up 3, transfers between spouses (Table 9: Impacts on transfers and intra-household dynamics col. 3 and 4) are mostly unaffected by the program. If anything, respondents tend to receive a little less money from their spouse when they received the program but coefficients are not meaningful or significant except 22 for men who received the grant only. They also do not increase the amount they transfer to their spouse, except for men who received the combination of training and grant (T1). However, results from earlier phases show that women who received the grant alone (T3) transferred greater amounts to their spouse at follow-ups 1 and 2 (Table A9, col. 7). This result does not hold for men, or when women also received the training (T1), pointing to a potential capture of the grant by men and to the mitigating impact of the training. 6. Dealing with attrition We first conduct an inverse probability weighing (IPW) analysis to reweigh estimates correcting for attrition based on observables. Table A10 shows the first stage probit, estimating baseline correlates of the probability of being recontacted at follow-up 3. The predicted probability is then used to estimate program impacts at follow-up 3 only, reweighing observations by the inverse probability of being found. This strategy gives more weight to respondents who, based on their observable characteristics at baseline, were less likely to be located and interviewed at follow-up 3, thus correcting the attrition bias based on observable characteristics. 23 Table A11 shows impacts on primary outcomes at follow-up 3 using the IPW specification. We find that the positive impact of T1 on men’s business outcomes is not robust: the coefficient on the performance zscore and revenues is no longer significant, suggesting that higher attrition in T1 was biasing effects upwards. We also adopt a bounding strategy to control for the potential bias resulting in uneven attrition between treatment group. Lee bounds are shown in Table A12. Lee bounds are calculated by trimming observations from the group that experienced less attrition. The upper (lower) bound drops observations in the group that had less attrition at the bottom (top) of the outcome distribution. For example, the upper bound is the impact of the treatment assuming that all respondents in the control group who are attritors had the lowest outcomes, and vice versa. Lee bounds are estimated separately for each treatment arm and without controls, therefore they are compared to the treatment effects estimated separately for each treatment arm without controls. There are minimal differences between the lower and upper bound of the treatment effect of T2, for both men and women. Attrition in T1 may be underestimating the effects on women’s business outcomes, since the upper bound becomes significant but not the lower bound or the point estimate for the performance score or profits. The bounds on profits and revenues for T3 in the male sample also suggest that the effect of the grant may be underestimated for men. 7. Conclusion and discussion In this paper we examine the relative impacts of relaxing the financial capital constraint, the human capital constraint, and both simultaneously. To this end we evaluated the impacts of either delivering a $400 cash grant, a life and business skills training, or the combination of training and cash to underemployed youth in Benin. We find that the training had strong and sustained impacts on youth’s business outcomes and earnings. Almost four years after delivery, participants who had received only the training had significantly higher profits and earnings than the control group. This is true for both women and men, whether they started out as business owners or not, and is in contradiction with most of the past literature on the impact of business trainings, which has mostly found them to have muted impacts on women. A combination of factors may explain the success of the training in this context. First, the training did not only focus on business skills but also, and equally so, on life skills. It therefore belongs to the promising new generation of psychology-informed trainings, with an additional emphasis placed on social relations. The positive and sustained psychological effects of the trainings, such as increased self-efficacy and decision-making power are likely to be a contributing factor, as suggested by recent reviews of the impact of incorporating a socioemotional component into trainings and other social programs. More research will be conducted on heterogeneity of effect by baseline levels of socioemotional skills and business practices to investigate those pathways. Second, the training was delivered in a well monitored, supervised and decentralized manner, with a master firm overseeing quality of trainers and implementation throughout, and dedicated local staff supporting implementation. Third, the positive impacts on women may be attributed to the training being gender-informed in several different ways: i) accommodations were made for women with young children, ii) training scheduled were compatible with women’s household responsibilities, iii) the life skills training content addressed women’s constraints in particular: decision - making, communication, empowerment, conflict resolution, and gender equality, iv) trainings were delivered locally, enabling women to participate without traveling great distances. Fourth, a large portion 24 of the sample had received previous technical training or had been in apprenticeship before the program. Pre-existing technical skills among participants may have contributed significantly to enhancing the effects of the business skills training, since many participants already had a trade in which they could apply the newly acquired management and socioemotional skills. Future heterogeneity analysis will confirm this point. Not only are the effects of the training significant and sustained over time, they are also quite large in magnitude. The increase in monthly earnings of women and men is US$16 and US$29 respectively. A useful benchmark is the amount of the unconditional cash transfer paid out as part of the Safety Net program in Benin. Beneficiaries of the cash transfer receive the equivalent of US$9 a month for 2 years. Thus effects of the training were much higher than the Safety Net program and are likely to be sustained over time. We also ran a cost analysis of each of the components. The training cost was US$ 1,050 per capita, and the cash grant cost US$ 6294. A simple back-of-the-envelope calculation using earnings tells us that the cost of the training was recouped for men after 36 months, while for women, assuming sustained impacts after the last follow-up, the cost will be recouped after 64 months. Given the upwards trajectory of effects for women, this is not an unreasonable assumption. The results on training point to the importance of longer time horizons when evaluating the impacts of such interventions. The impact of the training increased steadily over time, the first, short-term follow-up leading to very different conclusions on the impact of the training than the examination of the trajectory of impacts and longer-term impacts. Investing in human capital may take longer to bear fruit than financial capital. Past studies have mostly looked at short term effects, potentially not giving enough time for the skills learned in trainings to translate into improvements in monetary outcomes. Another key preliminary finding is that the grant had either null or negative effects on primary business outcomes and earnings, whether it was combined or not with the training. In fact, results suggest that the grant tended to worsen outcomes and cancel any positive effects of the training on profits, revenues, and earnings. One important caveat to this conclusion is that more heterogeneity analysis is needed to identify whether there are segments of the population who in fact did benefit from the grants. The implications of this finding have to be nuanced when examining welfare improvements: beneficiaries of the grant did see some improvements in their expenditures and assets (men only). Although this was not the intended purpose of the grant, increased personal expenditures and assets point to a reduction in poverty and improvement in welfare stemming from receiving the grant. It is not possible to conclude definitely as to the effect of combining the grant and the training on men’s business outcomes, on which the effect tend to be positive or null, depending on the measure and the correction of attrition. However, the null effect is very clear for women. In some cases receiving the grant, with or without the training, even worsened women’s profits and revenues. This is an intriguing result, as women are in fact using the grant to productive ends, hiring more employees and investing in productive assets. 4 The cost per capital was calculated by dividing the total cost of the training and grant components by the total number of beneficiaries. It includes general project management costs, prorated to the specific components considered here, sensitization and communication activities to advertise the project, costs of the enrolment operation, including staff and equipment, cost of conducting lotteries, monitoring and supervision through the hiring of the local employment agents, costs of the life and business skills training, including content design, training of training and delivery, and total amount of the grants delivered and payment fees. 25 We discuss a few plausible reasons why women are not able to transform those investments into higher profits and earnings. Firstly, women’s businesses may be taking a long time to react to the investments that were made initially, in particular for those who received the grant and the training. We do observe a positive trend in business outcomes: the largest negative effects were observed at the first follow-up and attenuated over time. It is possible that positive impacts would eventually be observed in another 12 or 24 months thanks to the accumulation of capital that has taken place over the past 3 years. In that vein it is worth noting that even if women’s business performance is not better in the grant groups than in the control group, they are creating jobs, which appear durable when they also received the training. In fact, for every three women who received the combined training and grant treatment, one paid job was created. Secondly, capture of the grant stemming from intra-household dynamics and redistributive taxation are explanations that are consistent with the data. Women who received the grant increased how much they transferred to their spouse in the short term, and out of the household over time. This suggests that women had to share the grant with their network and kin even if their profits and earnings had not actually increased. The pressure to share may also have led women to make sub-optimal investment decisions, by investing too much, too fast, or into the wrong assets, in order to avoid capture of the grant by their kin network. Such pressures may also be at play in their hiring decisions: women could be creating jobs and foregoing profit increases for themselves. We also see in the second follow-up that the grant has a negative effect on women’s profits only when their husband also owned a business at the previous follow-up. The negative effect of the partner having a business doesn’t stem from a full capture of the grant for the husband’s business: women in the grant group are accumulating capital. However, husbands may be capturing parts of the investment, or women may be making suboptimal investment decisions to avoid diverting the funds to her husband’s business. It is also possible that they are making an optimal intra-household allocation of the grant into the most profitable activity, which would be the husband’s, and only making a small investment in her own activity. These results call for a large uptake and scale-up of business trainings incorporating psychosocial modules, especially in contexts where self-employment remains the main pathway towards more productive jobs. Benchmarking these very positive results against the literature suggests that the design of the training, including adaptations to the literacy and education levels of participants and an appropriate sequencing to maximize experiential learning, and the quality of delivery matter a lot and need to be planned out carefully by policymakers. Finally, the results on grant call for revisiting the accompanying measures deployed around the provision of capital support. In economic inclusion programs, lump-sum cash support had a positive impact, complementary to other measures, and was even necessary to trigger positive impacts on business outcomes in Senegal. It may be necessary to design coaching, savings mechanisms, or community engagement activities (especially husbands) to reduce capture and maximize the productive impacts of grants. 26 8. References Alibhai, S., Buehren, N., Frese, M. D., Goldstein, M. P., Papineni, S., & Wolf, K. (2019). Full Esteem Ahead ? Mindset-Oriented Business Training in Ethiopia. Policy Research Working Paper Series, June, 1– 44. Berge, L. I. O., Bjorvatn, K., & Tungodden, B. (2015). Human and financial capital for microenterprise development: Evidence from a field and lab experiment. Management Science, 61(4), 707–722. https://doi.org/10.1287/mnsc.2014.1933 Blattman, C., N. Fiala, & Martinez S. (2014). Generating skilled self-employment in developing countries: Experimental evidence from Uganda. The Quarterly Journal of Economics. 697-752 Calderon, G., Cunha, J. M., & De Giorgi, G. (2013). Business Literacy and Development: Evidence from a Randomized Controlled Trial in Rural Mexico (No. w19740). National Bureau of Economic Research. Campos, F., Frese, M., Goldstein, M., Iacovone, L., Johnson, H. C., McKenzie, D., & Mensmann, M. (2017). Teaching personal initiative beats traditional training in boosting small business in West Africa. Science, 357(6357), 1287-1290. Chioda, L., Contreras-Loya, D., Gertler, P. J., & Carney, D. R. (2021). Making Entrepreneurs: Returns to Training Youth in Hard Versus Soft Business Skills. SSRN Electronic Journal. De Mel S, McKenzie D, Woodruff C. (2008) Returns to Capital in Microenterprises: Evidence from a Field Experiment. Quarterly Journal of Economics. 123(4):1329–72. De Mel S, McKenzie D, Woodruff C. (2014) Business training and female enterprise start-up, growth, and dynamics: Experimental evidence from Sri Lanka. Journal of Development Economics. 106:199–210. Fafchamps, M., McKenzie, D., Quinn, S., & Woodruff, C. (2014). Microenterprise growth and the flypaper effect: Evidence from a randomized experiment in Ghana. Journal of Development Economics, 106, 211- 226. Fiala, N. (2018). Returns to microcredit, cash grants and training for male and female microentrepreneurs in Uganda. World Development, 105, 189–200. https://doi.org/10.1016/j.worlddev.2017.12.027 Gine, X., & Mansuri, G. (2019). Money or Management? A Field Experiment on Constraints to Entrepreneurship in Rural Pakistan. Economic Development and Cultural Change. https://doi.org/10.1086/707502 Jayachandran, S. (2021). Microentrepreneurship in Developing Countries. In Handbook of Labor, Human Resources and Population Economics (pp. 1–31). McKenzie, D., & Woodruff, C. (2013). What Are We Learning from Business Training and Entrepreneurship Evaluations around the Developing World? The World Bank Research Observer, 29(1), 48–82. McKenzie, D., Woodruff, C., Bjorvatn, K., Bruhn, M., Cai, J., Gonzalez-Uribe, J., Quinn, S., Sonobe, T., & Valdivia, M. (2021). Training Entrepreneurs. In VoxDevLit (Vol. 1, Issue 2). 27 Ubfal, D., Arráiz, I., Beuermann, D. W., Frese, M., Maffioli, A., & Verch, D. (2022). The impact of soft-skills training for entrepreneurs in Jamaica. World Development, 152, 105787. Valdivia, M. (2011). Training or technical assistance? A field experiment to learn what works to increase managerial capital for female microentrepreneurs. Mimeo GRADE. 9. Tables Table 4: Characteristics of beneficiaries in the evaluation MEN - MEN WOMEN TOTAL WOMEN N Mean/SD N Mean/SD N Mean Difference 26.313 25.905 26.062 Age 1321 2115 3436 0.407** [4.413] [4.624] [4.548] 0.579 0.727 0.670 Partnered 1322 2118 3440 -0.148*** [0.494] [0.446] [0.470] 0.532 0.728 0.652 Has one child or more 1324 2119 3443 -0.196*** [0.499] [0.445] [0.476] 0.095 0.307 0.226 Never attended school 1323 2119 3442 -0.212*** [0.294] [0.461] [0.418] 8.177 5.629 6.609 Years of education 1324 2117 3441 2.549*** [4.751] [4.993] [5.055] 0.669 0.646 0.655 Is/was an apprentice 1323 2118 3441 0.023 [0.471] [0.478] [0.476] 0.187 0.118 0.145 Received a professional or technical training 1323 2119 3442 0.068*** [0.390] [0.323] [0.352] 0.915 0.824 0.859 Worked in the past 7 days 1321 2118 3439 0.091*** [0.279] [0.381] [0.348] Unemployed, underemployed or vulnerable 0.604 0.727 0.680 1317 2099 3416 -0.124*** employment [0.489] [0.445] [0.467] 0.735 0.819 0.785 Main activity is self-employed 1222 1781 3003 -0.084*** [0.442] [0.385] [0.411] Total number of employees, paid or 2.338 1.344 1.726 913 1463 2376 0.994*** not [3.730] [2.919] [3.289] 3.73e+05 1.15e+05 2.15e+05 Total value of productive assets 888 1406 2294 2.57e+05*** [1.20e+06] [1.09e+06] [1.14e+06] 37842.543 28043.719 32014.043 Total earnings in the past 30 days 1235 1813 3048 9798.824** [72043.402] [1.39e+05] [1.17e+05] 61659.815 36151.877 45960.913 Total monthly expenditures 1324 2119 3443 25507.938*** [75464.343] [47557.886] [61112.322] 28 Table 5: Impacts on business performance, earnings, and employment (1) (2) (3) (4) (5) Revenues from Profits from own Total earnings Z-score profits own business in business in the over the past Insufficient and revenues the past month - past month - month winsorized employment VARIABLES winsorized 1% winsorized 1% 1% woman -0.104*** -17.882*** -7.403*** -15.498*** 0.059*** (0.039) (5.309) (1.389) (2.387) (0.019) T1: training+grant x FU3 x man 0.100 19.398* 1.610 15.302*** 0.038 (0.070) (10.639) (3.507) (4.863) (0.036) T2: training only x FU3 x man 0.216*** 43.739*** 11.542*** 15.780*** 0.057 (0.081) (13.109) (4.457) (5.455) (0.036) T3: grant only x FU3 x man -0.029 17.138 -3.869 0.719 0.000 (0.071) (10.823) (3.601) (4.765) (0.037) T1: training+grant x FU3 x woman 0.051 7.331 -0.587 3.109 -0.036 (0.047) (6.365) (2.179) (2.977) (0.031) T2: training only x FU3 x woman 0.211*** 22.599*** 7.741*** 9.054*** 0.007 (0.047) (7.410) (2.393) (2.925) (0.030) T3: grant only x FU3 x woman 0.005 8.638 -0.328 0.471 -0.051* (0.045) (6.093) (2.163) (2.783) (0.030) Constant 0.013 58.280*** 13.679*** 56.677*** 0.408*** (0.058) (8.716) (2.097) (4.170) (0.030) Observations 12,853 13,089 13,095 13,458 15,200 R-squared 0.086 0.143 0.116 0.121 0.087 Control group mean BL, Man 0.0980 52.47 15.33 34.50 0.594 Control group mean BL, Woman -0.0810 37.92 8.977 21.63 0.726 Control group mean FU3, Man 0.0710 70.60 21.79 41.64 0.603 Control group mean FU3, Woman -0.0450 44.70 7.986 15.24 0.657 T1: Man - Woman (p-value) 0.536 0.312 0.560 0.0220 0.0854 T2: Man - Woman (p-value) 0.957 0.151 0.426 0.249 0.239 T3: Man - Woman (p-value) 0.665 0.480 0.358 0.962 0.231 Woman: T1 - T2 (p-value) 0.00176 0.0592 0.00155 0.0667 0.202 Woman: T1 - T3 (p-value) 0.356 0.850 0.915 0.397 0.651 Woman: T2 - T3 (p-value) 3.95e-05 0.0760 0.00202 0.00516 0.0796 Man: T1 - T2 (p-value) 0.201 0.115 0.0534 0.940 0.662 Man: T1 - T3 (p-value) 0.116 0.867 0.215 0.0121 0.369 Man: T2 - T3 (p-value) 0.00733 0.0874 0.00307 0.0171 0.178 Note: Program impacts for the first two follow-ups are estimated but not shown in this table (see appendixTable A3). The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Columns 1, 2, 3 also control for indicators of having a business. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1. Insufficient employment: indicator variable equal to 1 if respondent is unemployment, underemployed (working less than 40 hours a week), or in vulnerable employment (unpaid apprentice or unpaid contributor to the family business). Columns (1)-(4) are estimated for the sample of participants who are currently working. Column (5) is estimated on the entire sample. 29 Table 6: Impacts on expenditures, agency, and assets (1) (2) (3) Z-score total Individual asset monthly Agency score score (0-17) * VARIABLES expenditures Follow-up 3 Follow-up 3 Follow-up 2 woman -0.434*** -0.238*** -1.563*** (0.039) (0.028) (0.140) T1: training+grant x FU x man 0.195*** 0.044 0.429** (0.071) (0.043) (0.177) T2: training only x FU x man 0.122* 0.046 0.199 (0.072) (0.044) (0.179) T3: grant only x FU x man 0.061 -0.065 0.122 (0.065) (0.041) (0.181) T1: training+grant x FU x woman 0.179*** 0.064* 0.097 (0.051) (0.033) (0.108) T2: training only x FU x woman 0.117** 0.099*** 0.123 (0.052) (0.033) (0.115) T3: grant only x FU x woman 0.196*** 0.022 0.111 (0.052) (0.036) (0.102) Constant 0.187*** 0.081** 2.098*** (0.050) (0.036) (0.176) Observations 15,614 9,315 6,451 R-squared 0.199 0.160 0.539 Control group mean BL, Man 0.324 0.213 5.135 Control group mean BL, Woman -0.158 -0.147 1.496 Control group mean FU2, Man 5.224 Control group mean FU2, Woman 1.754 Control group mean FU3, Man 0.281 0.108 Control group mean FU3, Woman -0.183 -0.134 T1: Man - Woman (p-value) 0.835 0.687 0.0963 T2: Man - Woman (p-value) 0.955 0.304 0.712 T3: Man - Woman (p-value) 0.0723 0.0845 0.954 Woman: T1 - T2 (p-value) 0.259 0.315 0.833 Woman: T1 - T3 (p-value) 0.749 0.249 0.901 Woman: T2 - T3 (p-value) 0.154 0.0380 0.918 Man: T1 - T2 (p-value) 0.387 0.969 0.223 Man: T1 - T3 (p-value) 0.0878 0.0200 0.107 Man: T2 - T3 (p-value) 0.450 0.0197 0.690 Notes: * Information to compute individual and household asset scores was not collected at the 3rd follow-up survey. For the individual and household asset scores, results shown are those from the 2nd follow-up. Program impacts for the first follow-up are estimated but not shown (see appendix Table A4). For the expenditure z-score and agency score, the program impacts for the first two follow-ups are estimated but not shown in this table. The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Standard errors in parentheses are clustered at the individual level *** p<0.01, ** p<0.05, * p<0.1 30 Table 7: Impacts on self-employment and business starts, deaths, and transitions (1) (2) (3) (4) Have changed Number of own Death of Own business business since businesses * business VARIABLES BL woman 0.029 0.043 -0.024 0.082** (0.018) (0.050) (0.016) (0.032) T1: training+grant x FU3 x man 0.064*** 0.229*** -0.038* -0.000 (0.023) (0.070) (0.020) (0.044) T2: training only x FU3 x man 0.071*** 0.126* -0.050*** 0.042 (0.022) (0.068) (0.019) (0.045) T3: grant only x FU3 x man 0.071*** 0.045 -0.052*** -0.035 (0.022) (0.066) (0.019) (0.044) T1: training+grant x FU3 x woman 0.071*** 0.238*** -0.023 -0.027 (0.017) (0.058) (0.015) (0.038) T2: training only x FU3 x woman 0.056*** 0.177*** -0.030** -0.019 (0.017) (0.057) (0.014) (0.038) T3: grant only x FU3 x woman 0.053*** 0.137** -0.015 -0.027 (0.018) (0.058) (0.015) (0.037) Constant 0.664*** 1.184*** 0.149*** 0.320*** (0.024) (0.066) (0.023) (0.043) Observations 9,336 9,336 6,519 6,213 R-squared 0.107 0.083 0.024 0.036 Control group mean BL, Man 0.700 0 0 Control group mean BL, Woman 0.667 0 0 Control group mean FU3, Man 0.853 1.554 0.0740 0.405 Control group mean FU3, Woman 0.882 1.645 0.0500 0.446 T1: Man - Woman (p-value) 0.775 0.913 0.516 0.622 T2: Man - Woman (p-value) 0.569 0.537 0.359 0.270 T3: Man - Woman (p-value) 0.482 0.264 0.0947 0.882 Woman: T1 - T2 (p-value) 0.320 0.306 0.548 0.833 Woman: T1 - T3 (p-value) 0.234 0.0971 0.599 0.999 Woman: T2 - T3 (p-value) 0.839 0.505 0.261 0.830 Man: T1 - T2 (p-value) 0.735 0.160 0.492 0.379 Man: T1 - T3 (p-value) 0.724 0.0108 0.402 0.456 Man: T2 - T3 (p-value) 0.989 0.253 0.877 0.105 Notes: Program impacts for the first two follow-ups are estimated but not shown in this table (see appendix XX). The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. * Baseline outcome is not controlled in column (4). Information to compute the number of businesses was not collected at baseline. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 All specifications use data from the first phase of follow-ups 1 and 2 and follow-up 3, hence the number of observations being smaller than the profits estimation in Table 5 which also uses data from the second phases of follow-ups 1 and 2. Columns (3)-(4) are conditional on having a business at baseline. 31 Table 8: Impacts on labor and capital inputs (1) (2) (3) (4) (5) Number of Value of K investment Business Number of paid Value of K stock - employees paid or (flow 12 month) - practice score employees winsorized 1% VARIABLES unpaid winsorized 1% * woman -0.667*** -0.497*** -214.553*** -39.077*** 0.054 (0.180) (0.121) (31.836) (8.297) (0.262) T1: training+grant x FU3 x man -0.380 -0.047 320.482*** 120.351*** 1.452*** (0.295) (0.210) (86.567) (27.975) (0.375) T2: training only x FU3 x man 0.186 -0.032 59.932 35.956* 1.565*** (0.305) (0.187) (66.122) (20.889) (0.393) T3: grant only x FU3 x man -0.180 0.095 80.775 48.677** 0.388 (0.283) (0.207) (68.755) (24.614) (0.364) T1: training+grant x FU3 x woman 0.483** 0.368** 78.933*** 15.024 1.213*** (0.199) (0.147) (27.810) (10.384) (0.277) T2: training only x FU3 x woman 0.326* 0.159 57.819* 9.120 1.246*** (0.186) (0.103) (29.865) (10.948) (0.288) T3: grant only x FU3 x woman 0.263 0.110 55.634* 7.825 -0.146 (0.172) (0.076) (31.258) (10.776) (0.277) Constant 1.525*** 0.973*** 334.763*** 66.280*** 5.058*** (0.261) (0.192) (48.229) (12.214) (0.374) Observations 6,210 6,215 6,215 6,218 4,214 R-squared 0.249 0.188 0.173 0.102 0.148 Control group mean BL, Man 2.206 0.716 289.9 126.7 7.816 Control group mean BL, Woman 1.212 0.246 97.97 33.22 7.448 Control group mean FU2, Man 7.902 Control group mean FU2, Woman 7.936 Control group mean FU3, Man 2.332 0.927 484.6 88.95 Control group mean FU3, Woman 1.082 0.176 150 25.35 T1: Man - Woman (p-value) 0.0107 0.101 0.00587 0.000234 0.593 T2: Man - Woman (p-value) 0.676 0.352 0.975 0.224 0.497 T3: Man - Woman (p-value) 0.153 0.941 0.726 0.110 0.224 Woman: T1 - T2 (p-value) 0.497 0.221 0.494 0.611 0.910 Woman: T1 - T3 (p-value) 0.310 0.0876 0.468 0.534 2.39e-06 Woman: T2 - T3 (p-value) 0.759 0.655 0.949 0.914 3.00e-06 Man: T1 - T2 (p-value) 0.0956 0.946 0.00672 0.00848 0.798 Man: T1 - T3 (p-value) 0.535 0.569 0.0147 0.0380 0.0104 Man: T2 - T3 (p-value) 0.266 0.579 0.796 0.662 0.00640 Notes: * Information to compute the business practice score was not collected at the 3rd follow-up survey. For the the business practice scores, results shown are those from the 2nd follow-up. Program impacts for the first follow-up are estimated but not shown (see appendix XX). For the other outcomes, the program impacts for the first two follow-ups are estimated but not shown in this table (see appendix XX). The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 32 Table 9: Impacts on transfers and intra-household dynamics (1) (2) (3) (4) (5) Amount of Amount of Total amount Total amount monthly monthly transferred TO transferred BY transfers TO transfers BY Spouse has a respondent in respondent in respondent by respondent to business * past 30d - past 30d - their spouse - their spouse - winsorized 1% winsorized 1% winsorized 1% winsorized 1% VARIABLES * * woman 0.658** -1.276*** 9.006*** -7.453*** 0.012 (0.331) (0.223) (0.565) (0.628) (0.028) T1: training+grant x FU3 x man 0.590 1.723*** -0.008 3.769** -0.005 (0.549) (0.513) (0.928) (1.513) (0.038) T2: training only x FU3 x man 1.815*** 0.600 -0.402 0.101 0.019 (0.659) (0.418) (0.864) (1.264) (0.037) T3: grant only x FU3 x man 1.302** 1.176** -1.651** -0.801 -0.019 (0.608) (0.489) (0.700) (1.223) (0.037) T1: training+grant x FU3 x woman -0.624* 0.955*** 0.879 -0.401 -0.039 (0.327) (0.308) (1.151) (0.672) (0.031) T2: training only x FU3 x woman 0.274 0.429* -0.893 -0.555 -0.049 (0.401) (0.260) (1.084) (0.665) (0.031) T3: grant only x FU3 x woman -0.099 0.783*** -1.311 0.297 -0.029 (0.388) (0.291) (1.067) (0.723) (0.030) Constant 1.600*** 2.401*** -2.116*** 7.676*** 0.515*** (0.430) (0.360) (0.665) (0.825) (0.038) Observations 9,336 9,333 9,336 9,336 9,336 R-squared 0.051 0.072 0.132 0.134 0.043 Control group mean BL, Man 2.526 3.205 Control group mean BL, Woman 1.665 1.431 Control group mean FU3, Man 2.497 2.760 3.222 12.32 0.645 Control group mean FU3, Woman 1.385 1.045 13.62 4.184 0.634 T1: Man - Woman (p-value) 0.0405 0.175 0.510 0.00657 0.466 T2: Man - Woman (p-value) 0.0357 0.705 0.693 0.611 0.142 T3: Man - Woman (p-value) 0.0400 0.464 0.761 0.391 0.829 Woman: T1 - T2 (p-value) 0.0288 0.0976 0.161 0.828 0.766 Woman: T1 - T3 (p-value) 0.187 0.616 0.0805 0.359 0.764 Woman: T2 - T3 (p-value) 0.419 0.238 0.725 0.260 0.549 Man: T1 - T2 (p-value) 0.0841 0.0497 0.709 0.0266 0.550 Man: T1 - T3 (p-value) 0.284 0.381 0.0773 0.00476 0.722 Man: T2 - T3 (p-value) 0.496 0.297 0.150 0.517 0.333 Notes: Program impacts for the first two follow-ups are estimated but not shown in this table (see appendix XX). The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. * Baseline outcomes are not controlled in columns (3), (4) and (5). Information to compute those outcomes was not collected at baseline. Standard errors in parentheses are clustered at the individual level *** p<0.01, ** p<0.05, * p<0.1 33 Appendix Table A1: Baseline balance test of respondent characteristics PANEL A: MEN (1) (2) (3) (4) (5) t-test t-test t-test T1: T2: T3: trainin Differe Differe Differe trainin cash Control Total Baseline variables g+ nce nce nce g only only cash Mean/S Mean/S Mean/S Mean/S Mean/S N N N N N (4)-(1) (4)-(2) (4)-(3) D D D D D Age 314 26.62 323 26.78 332 25.59 352 26.24 1321 26.30 -0.39 -0.55 0.65** [4.55] [4.41] [4.20] [4.39] [4.41] Live with a partner 316 0.61 322 0.61 332 0.54 352 0.56 1322 0.58 -0.05 -0.05 0.02 [0.49] [0.49] [0.50] [0.50] [0.49] Number of dependent children 316 2.01 323 1.99 332 1.43 352 1.72 1323 1.78 -0.29 -0.27 0.29* [2.68] [2.42] [2.12] [2.34] [2.40] Household size 316 4.80 323 4.91 332 4.42 352 4.63 1323 4.69 -0.17 -0.29 0.20 [3.31] [3.13] [3.34] [3.04] [3.21] Years of formal education 282 9.06 295 9.21 307 9.12 313 8.78 1197 9.04 -0.28 -0.42 -0.34 [4.35] [4.17] [3.99] [4.05] [4.13] Z-score profits and revenues 298 0.10 297 0.09 301 0.06 319 0.10 1215 0.09 -0.01 0.01 0.04 [0.80] [0.77] [0.80] [0.72] [0.77] Revenues from own business in the past month 298 57.71 295 52.51 301 52.49 317 51.71 1211 53.58 -6.00 -0.80 -0.78 - winsorized 1% (in thousands) [117.77] [101.03] [116.43] [101.06] [109.17] Profits from own business in the past month 293 17.94 295 16.58 299 15.81 317 15.03 1204 16.31 -2.91 -1.55 -0.78 - winsorized 1% (in thousands) [31.50] [28.30] [30.59] [24.85] [28.84] Total earnings over the past month 301 40.44 300 35.39 305 31.87 329 34.32 1235 35.47 -6.12 -1.06 2.45 winsorized 1% (in thousands) [59.90] [52.17] [47.46] [48.95] [52.28] Unemployment, vulnerable employment OR 316 0.57 322 0.61 329 0.63 351 0.60 1318 0.60 0.03 -0.01 -0.03 underemployment [0.50] [0.49] [0.48] [0.49] [0.49] 34 PANEL B: WOMEN (1) (2) (3) (4) (5) t-test t-test t-test T1: T2: T3: trainin Differe Differe Differe trainin cash Control Total Baseline variables g+ nce nce nce g only only cash Mean/S Mean/S Mean/S Mean/S Mean/S N N N N N (4)-(1) (4)-(2) (4)-(3) D D D D D Age 520 26.03 506 25.76 527 26.03 562 25.84 2115 25.91 -0.19 0.07 -0.19 [4.73] [4.61] [4.57] [4.61] [4.63] Live with a partner 521 0.75 508 0.71 527 0.73 562 0.72 2118 0.73 -0.03 0.01 -0.01 [0.44] [0.45] [0.45] [0.45] [0.45] Number of dependent children 521 2.08 508 2.05 528 2.12 562 2.07 2119 2.08 -0.00 0.02 -0.05 [1.96] [1.92] [2.09] [1.93] [1.98] Household size 521 5.33 508 5.49 528 5.39 562 5.33 2119 5.38 0.00 -0.15 -0.05 [3.03] [2.79] [2.82] [2.62] [2.81] Years of formal education 361 8.48 349 8.07 361 8.12 395 7.88 1466 8.13 -0.60** -0.19 -0.24 [4.00] [3.92] [4.05] [3.88] [3.96] Z-score profits and revenues 438 -0.08 436 -0.07 446 -0.07 447 -0.08 1767 -0.07 0.00 -0.01 -0.00 [0.61] [0.63] [0.54] [0.57] [0.59] Revenues from own business in the past month 436 40.06 428 43.50 441 37.30 441 38.24 1746 39.75 -1.83 -5.27 0.94 - winsorized 1% (in thousands) [85.41] [99.53] [81.08] [84.63] [87.82] Profits from own business in the past month 435 9.05 433 8.95 445 8.66 440 9.00 1753 8.91 -0.05 0.04 0.33 - winsorized 1% (in thousands) [16.44] [17.75] [15.09] [17.09] [16.60] Total earnings over the past month 448 24.76 440 25.24 461 20.03 464 21.69 1813 22.89 -3.07 -3.56 1.66 winsorized 1% (in thousands) [47.67] [51.60] [34.17] [39.88] [43.74] Unemployment, vulnerable employment OR 515 0.73 506 0.72 521 0.72 557 0.73 2099 0.73 0.00 0.01 0.01 underemployment [0.45] [0.45] [0.45] [0.44] [0.45] Notes: The value displayed for t-tests are the differences in the means across the groups. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 35 Table A2: Attrition Attrition follow-up 1 Attrition follow-up 2 Attrition follow-up 3 (1) (2) (3) (4) (5) (6) T1: training + cash 0.00305 0.00383 0.00721 0.00795 0.0347** 0.0416** (0.0114) (0.0114) (0.00973) (0.00973) (0.0176) (0.0174) T2: training only -0.00012 0.00044 -0.00810 -0.00795 0.0107 0.0149 (0.0113) (0.0113) (0.00883) (0.00882) (0.0171) (0.0169) T3: cash only 0.00836 0.00758 0.00247 0.00277 0.0241 0.0294* (0.0116) (0.0115) (0.00939) (0.00935) (0.0172) (0.0171) Woman -0.0109 -0.00841 0.0302** (0.00859) (0.00702) (0.0126) Priority sector -0.00152 -0.00873 -0.0616*** (0.00862) (0.00723) (0.0137) Constant 0.0591*** 0.0639*** 0.0394*** 0.0252** 0.143*** 0.224*** (0.00780) (0.0201) (0.00644) (0.0121) (0.0116) (0.0326) Control for district dummies No Yes No Yes No Yes Number of observations 3442 3435 3442 3435 3442 3435 R2 0.000205 0.00936 0.000780 0.00846 0.00130 0.0220 Standard errors in parentheses * p<0.1, ** p<0.05, *** p<0.01" 36 Table A3: Impacts on business performance, earnings, and employment (full results) PANEL A: Profits and Revenues (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Profits from Revenues Revenue DFITS IHS Profits Profits from IHS IHS own IHS Profits IHS Profits from own from all Profits from from own all businesses Revenue Revenue Z-score business in from own from all business in businesses in own business in in the past from own from all profits and the past business in businesses the past the past business in the past month business in businesses revenues month - the past in the past month - month the past month - winsorized the past in the past winsorized month month winsorized winsorized month Calculated 1% month month VARIABLES 1% 1% 1% woman -0.104*** -7.403*** 0.513* -6.260*** -2.212*** -10.241*** 0.264 -17.882*** 0.469*** -23.140*** 0.207 (0.039) (1.389) (0.262) (1.047) (0.402) (1.733) (0.249) (5.309) (0.177) (6.646) (0.154) T1: training+grant x FU1 x man -0.057 -3.109 -0.654 -4.175** -0.931* -4.081 -0.660 -10.958 0.141 -9.379 0.050 (0.057) (2.390) (0.453) (1.792) (0.549) (2.755) (0.423) (7.138) (0.234) (9.094) (0.210) T1: training+grant x FU2 x man 0.162*** 4.069** 0.993*** 3.058* 0.894* 3.717 0.869*** 10.943 0.623** 12.812 0.365* (0.057) (1.925) (0.323) (1.595) (0.537) (2.484) (0.303) (7.740) (0.242) (10.515) (0.221) T1: training+grant x FU3 x man 0.100 1.610 0.011 0.122 0.188 6.226 0.578 19.398* 0.399 30.731** 0.537* (0.070) (3.507) (0.596) (2.779) (0.762) (3.955) (0.550) (10.639) (0.340) (13.092) (0.299) T2: training only x FU1 x man 0.149*** 3.704 0.865** 1.717 0.872* 4.981* 0.702* 17.289** 0.746*** 15.516 0.577*** (0.057) (2.442) (0.399) (1.849) (0.501) (2.936) (0.376) (8.461) (0.220) (9.509) (0.183) T2: training only x FU2 x man 0.162*** 4.132* 0.719** 5.294** 0.595 5.804** 0.696** 18.793** 0.448* 27.453** 0.399* (0.062) (2.116) (0.326) (2.073) (0.545) (2.701) (0.295) (8.869) (0.260) (11.739) (0.220) T2: training only x FU3 x man 0.216*** 11.542*** 0.360 0.212 1.180 14.137*** 0.260 43.739*** 0.652** 56.368*** 0.462 (0.081) (4.457) (0.608) (3.119) (0.743) (5.115) (0.578) (13.109) (0.332) (16.873) (0.303) T3: grant only x FU1 x man 0.026 -1.546 0.009 -3.028* -0.661 -2.325 -0.162 5.775 0.459** 2.107 0.285 (0.057) (2.240) (0.403) (1.644) (0.513) (2.686) (0.385) (7.940) (0.221) (9.260) (0.197) T3: grant only x FU2 x man 0.078 2.381 0.106 2.299 -0.793 2.339 0.038 14.556* 0.314 11.633 0.157 (0.061) (2.006) (0.345) (1.847) (0.550) (2.575) (0.322) (8.839) (0.247) (10.794) (0.218) T3: grant only x FU3 x man -0.029 -3.869 -1.071* -6.670*** -1.199 -1.791 -0.737 17.138 0.100 17.200 -0.100 (0.071) (3.601) (0.629) (2.420) (0.761) (4.101) (0.597) (10.823) (0.348) (13.054) (0.325) T1: training+grant x FU1 x woman -0.031 -2.368* -0.829** -2.049** 0.225 -2.789* -0.462 -3.401 -0.103 -1.231 0.028 (0.034) (1.298) (0.322) (0.998) (0.445) (1.537) (0.302) (4.437) (0.148) (5.303) (0.130) T1: training+grant x FU2 x woman -0.028 -0.297 -0.519** -0.365 0.507 -0.622 -0.345 1.977 -0.304* 3.391 -0.253 (0.037) (1.048) (0.257) (0.949) (0.478) (1.321) (0.239) (4.941) (0.180) (6.231) (0.164) T1: training+grant x FU3 x woman 0.051 -0.587 -0.472 -1.693 0.880 3.107 0.257 7.331 -0.112 18.367** 0.195 (0.047) (2.179) (0.476) (1.710) (0.648) (2.652) (0.429) (6.365) (0.256) (8.062) (0.227) T2: training only x FU1 x woman 0.045 2.149 0.152 2.314** 1.285*** 2.470 0.529** -3.211 -0.179 -1.634 -0.026 (0.034) (1.320) (0.294) (1.167) (0.437) (1.561) (0.269) (4.486) (0.148) (5.256) (0.133) T2: training only x FU2 x woman 0.077** 3.490*** 0.373* 2.709*** 0.670 4.291*** 0.335 7.610 -0.077 11.040* -0.063 (0.038) (1.223) (0.214) (0.944) (0.454) (1.520) (0.210) (5.511) (0.166) (6.675) (0.157) T2: training only x FU3 x woman 0.211*** 7.741*** 0.926** 3.635** 2.691*** 9.455*** 1.219*** 22.599*** 0.397* 31.034*** 0.487** (0.047) (2.393) (0.421) (1.714) (0.627) (2.671) (0.383) (7.410) (0.232) (8.938) (0.212) T3: grant only x FU1 x woman -0.116*** -4.258*** -1.532*** -3.655*** -0.945** -4.327*** -1.346*** -8.613** -0.305** -9.542** -0.307** (0.034) (1.223) (0.333) (1.011) (0.449) (1.449) (0.321) (4.063) (0.149) (4.739) (0.138) T3: grant only x FU2 x woman -0.100*** -1.561 -0.714*** -1.295 -1.315*** -2.092* -0.597** -2.454 -0.585*** -2.531 -0.467*** (0.035) (0.979) (0.250) (0.828) (0.473) (1.235) (0.239) (4.146) (0.184) (5.307) (0.168) T3: grant only x FU3 x woman 0.005 -0.328 -0.802* -1.208 -0.525 0.177 -0.515 8.638 -0.067 9.510 0.029 (0.045) (2.163) (0.472) (1.646) (0.648) (2.439) (0.449) (6.093) (0.247) (7.175) (0.226) Constant 0.013 13.679*** 5.632*** 12.597*** 5.299*** 21.281*** 7.044*** 58.280*** 8.211*** 75.183*** 9.068*** (0.058) (2.097) (0.443) (1.584) (0.557) (2.530) (0.407) (8.716) (0.313) (10.892) (0.273) Observations 12,853 13,095 13,095 12,412 13,089 13,096 13,096 13,089 13,089 13,090 13,090 37 R-squared 0.086 0.116 0.122 0.092 0.061 0.135 0.145 0.143 0.350 0.146 0.419 Control group mean BL, Man 0.0980 15.33 6.472 12.90 1.597 18.07 6.611 52.47 8.070 57.29 8.102 Control group mean BL, Woman -0.0810 8.977 6.696 7.914 -0.503 10.20 6.814 37.92 8.165 40.95 8.256 Control group mean FU1, Man 0.0920 17.41 5.639 14.29 4.534 24.37 6.520 67.92 8.645 83.45 9.121 Control group mean FU1, Woman -0.0600 9.758 5.938 9.079 2.386 13.73 6.468 43.45 8.811 54.51 9.145 Control group mean FU2, Man 0.114 18.77 6.565 16.80 4.038 25.77 7.251 62.31 8.028 82.66 8.640 Control group mean FU2, Woman -0.0710 10.46 7.255 9.644 1.762 14.22 7.738 41.96 8.972 53.99 9.368 Control group mean FU3, Man 0.0710 21.79 5.049 19.30 3.191 29.11 5.721 70.60 8.297 89.42 8.901 Control group mean FU3, Woman -0.0450 7.986 6.272 8.667 0.875 11.77 7.082 44.70 9.105 57.71 9.524 FU1,T1: Man - Woman 0.687 0.773 0.735 0.272 0.0855 0.669 0.682 0.346 0.363 0.427 0.926 FU1,T2: Man - Woman 0.102 0.555 0.113 0.774 0.513 0.433 0.680 0.0267 0.000301 0.107 0.00567 FU1, T3: Man - Woman 0.0252 0.256 0.00133 0.729 0.660 0.491 0.0104 0.0936 0.00301 0.252 0.0111 FU2,T1: Man - Woman 0.00317 0.0429 0.000190 0.0615 0.566 0.113 0.00125 0.312 0.00107 0.417 0.0167 FU2,T2: Man - Woman 0.222 0.791 0.363 0.253 0.910 0.620 0.304 0.272 0.0708 0.207 0.0680 FU2,T3: Man - Woman 0.00795 0.0738 0.0501 0.0727 0.445 0.113 0.105 0.0732 0.00195 0.216 0.0159 FU3,T1: Man - Woman 0.536 0.560 0.476 0.541 0.438 0.469 0.596 0.312 0.183 0.405 0.305 FU3,T2: Man - Woman 0.957 0.426 0.385 0.297 0.0794 0.387 0.110 0.151 0.479 0.176 0.939 FU3,T3: Man - Woman 0.665 0.358 0.702 0.0350 0.450 0.649 0.738 0.480 0.666 0.595 0.717 Woman,T1: FU2 - FU1 0.924 0.126 0.397 0.164 0.588 0.160 0.726 0.251 0.317 0.426 0.110 Woman,T2: FU2 - FU1 0.426 0.375 0.504 0.757 0.221 0.310 0.528 0.0473 0.583 0.0488 0.827 Woman,T3: FU2 - FU1 0.674 0.0366 0.0304 0.0392 0.472 0.136 0.0381 0.136 0.161 0.161 0.375 Man,T1: FU2 - FU1 5.11e-05 0.00168 0.000232 5.24e-05 0.00136 0.00309 0.000257 0.000523 0.0370 0.00894 0.137 Man,T2: FU2 - FU1 0.814 0.846 0.706 0.0811 0.619 0.767 0.985 0.838 0.217 0.179 0.357 Man,T3: FU2 - FU1 0.336 0.0711 0.819 0.00286 0.813 0.0691 0.612 0.179 0.535 0.246 0.527 Woman,T1: FU3 - FU2 0.113 0.897 0.924 0.469 0.596 0.147 0.181 0.368 0.488 0.0379 0.0676 Woman,T2: FU3 - FU2 0.00390 0.0585 0.202 0.593 0.00191 0.0377 0.0262 0.0247 0.0621 0.0124 0.0157 Woman,T3: FU3 - FU2 0.0209 0.553 0.855 0.958 0.246 0.335 0.859 0.0377 0.0467 0.0560 0.0309 Man,T1: FU3 - FU2 0.350 0.460 0.0974 0.293 0.354 0.484 0.592 0.360 0.506 0.0978 0.568 Man,T2: FU3 - FU2 0.465 0.0650 0.531 0.102 0.441 0.0661 0.422 0.0278 0.557 0.0513 0.839 Man,T3: FU3 - FU2 0.108 0.0865 0.0480 0.000422 0.591 0.280 0.172 0.790 0.509 0.630 0.392 Woman,T1: FU3 - FU1 0.0922 0.445 0.504 0.850 0.341 0.0277 0.134 0.0805 0.971 0.0110 0.483 Woman,T2: FU3 - FU1 0.00102 0.0263 0.105 0.514 0.0348 0.0143 0.116 0.000364 0.0233 0.000237 0.0266 Woman,T3: FU3 - FU1 0.0167 0.106 0.190 0.197 0.557 0.0923 0.124 0.00384 0.370 0.00594 0.162 Man,T1: FU3 - FU1 0.0173 0.175 0.285 0.114 0.146 0.00714 0.0337 0.000878 0.414 0.000294 0.0874 Man,T2: FU3 - FU1 0.396 0.0683 0.413 0.631 0.687 0.0567 0.451 0.0433 0.764 0.0124 0.697 Man,T3: FU3 - FU1 0.412 0.525 0.0923 0.135 0.490 0.889 0.335 0.239 0.270 0.180 0.202 FU3,Woman: T1 - T2 0.00176 0.00155 0.00563 0.00563 0.00957 0.0356 0.0275 0.0592 0.0608 0.207 0.221 FU3,Woman: T1 - T3 0.356 0.915 0.546 0.794 0.0499 0.298 0.119 0.850 0.873 0.296 0.507 FU3,Woman: T2 - T3 3.95e-05 0.00202 0.000564 0.00958 4.35e-06 0.00105 0.000146 0.0760 0.0780 0.0209 0.0532 FU3,Man: T1 - T2 0.201 0.0534 0.620 0.981 0.246 0.168 0.621 0.115 0.503 0.192 0.825 FU3,Man: T1 - T3 0.116 0.215 0.133 0.0309 0.112 0.0995 0.0467 0.867 0.445 0.410 0.0756 FU3,Man: T2 - T3 0.00733 0.00307 0.0501 0.0456 0.00540 0.00639 0.145 0.0874 0.152 0.0457 0.120 38 PANEL B: Employment PANEL C: Earnings (1) (2) (3) (4) (5) (6) (7) (8) Total earnings Hourly IHS Total DFITS Total over the past earnings Insufficient Hours worked Worked in the Worked in the earnings over earnings over month below employment in past 30d past 6 months past 7 days the past the past winsorized minimum month month VARIABLES 1% wage woman 0.059*** -6.262 -0.012** -0.028*** -15.498*** -0.353** -12.644*** 0.104*** (0.019) (4.671) (0.005) (0.008) (2.387) (0.176) (1.742) (0.021) T1: training+grant x FU1 x man 0.034 -5.883 0.009** 0.023** -0.409 -0.040 -2.411 -0.022 (0.030) (6.220) (0.004) (0.009) (4.624) (0.214) (2.933) (0.036) T1: training+grant x FU2 x man 0.033 -2.271 0.005 -0.003 2.741 0.346 1.427 -0.038 (0.030) (6.389) (0.005) (0.011) (2.810) (0.211) (2.042) (0.032) T1: training+grant x FU3 x man 0.038 -8.993 -0.000 0.005 15.302*** 1.197** 9.513*** -0.146*** (0.036) (8.413) (0.007) (0.015) (4.863) (0.476) (3.441) (0.038) T2: training only x FU1 x man -0.001 0.868 0.007 0.012 7.441 0.347* 8.519** -0.002 (0.030) (6.244) (0.005) (0.011) (4.707) (0.204) (3.870) (0.035) T2: training only x FU2 x man 0.008 0.101 0.003 0.002 6.953** 0.468** 6.323** 0.006 (0.029) (5.904) (0.005) (0.011) (2.939) (0.206) (3.111) (0.031) T2: training only x FU3 x man 0.057 2.458 -0.003 -0.010 15.780*** 0.536 2.407 -0.046 (0.036) (8.687) (0.008) (0.017) (5.455) (0.511) (3.705) (0.036) T3: grant only x FU1 x man -0.015 4.728 0.009** 0.022** 2.494 -0.061 -1.693 0.060* (0.030) (6.198) (0.004) (0.010) (4.817) (0.210) (3.462) (0.034) T3: grant only x FU2 x man 0.028 2.461 0.002 -0.014 3.616 0.065 2.910 -0.029 (0.029) (6.027) (0.006) (0.012) (2.847) (0.217) (2.555) (0.032) T3: grant only x FU3 x man 0.000 -11.378 0.004 -0.001 0.719 -0.195 -5.600* -0.008 (0.037) (8.174) (0.006) (0.016) (4.765) (0.538) (3.163) (0.034) T1: training+grant x FU1 x woman -0.011 2.199 0.012** 0.006 -0.954 -0.087 -0.522 -0.007 (0.025) (5.467) (0.006) (0.012) (2.875) (0.168) (2.227) (0.028) T1: training+grant x FU2 x woman -0.018 2.473 0.015*** -0.011 0.538 -0.088 0.347 -0.016 (0.024) (5.284) (0.005) (0.012) (1.600) (0.166) (1.261) (0.022) T1: training+grant x FU3 x woman -0.036 -1.802 0.016*** -0.004 3.109 0.270 1.120 -0.043* (0.031) (7.409) (0.006) (0.016) (2.977) (0.416) (2.285) (0.025) T2: training only x FU1 x woman -0.059** 8.426 0.012** 0.022* 2.475 0.125 3.202 0.014 (0.025) (5.267) (0.006) (0.011) (3.119) (0.164) (3.041) (0.027) T2: training only x FU2 x woman -0.015 5.150 0.006 -0.014 4.419*** 0.211 4.028*** -0.006 (0.024) (5.233) (0.006) (0.012) (1.693) (0.158) (1.297) (0.021) T2: training only x FU3 x woman 0.007 -3.184 0.011 -0.003 9.054*** 1.025*** 8.171*** -0.054** (0.030) (7.205) (0.007) (0.016) (2.925) (0.377) (2.459) (0.025) T3: grant only x FU1 x woman -0.058** 7.034 0.008 -0.007 -4.657* -0.730*** -2.048 0.009 (0.025) (5.374) (0.006) (0.013) (2.685) (0.176) (2.324) (0.028) T3: grant only x FU2 x woman -0.005 1.320 -0.013 -0.035*** -0.774 -0.541*** -0.874 0.003 (0.024) (5.240) (0.008) (0.013) (1.509) (0.177) (1.176) (0.021) T3: grant only x FU3 x woman -0.051* 10.032 0.011* 0.017 0.471 -0.615 -0.448 -0.006 (0.030) (7.325) (0.007) (0.015) (2.783) (0.448) (2.233) (0.023) 39 Constant 0.408*** 217.342*** 0.966*** 0.905*** 56.677*** 8.992*** 48.179*** 0.567*** (0.030) (6.460) (0.008) (0.014) (4.170) (0.258) (3.126) (0.033) Observations 15,200 15,058 15,560 15,469 13,458 13,458 12,912 7,353 R-squared 0.087 0.130 0.021 0.021 0.121 0.063 0.108 0.087 Control group mean BL, Man 0.594 156.4 0.956 0.901 34.50 9.329 31.16 0.680 Control group mean BL, Woman 0.726 120 0.896 0.796 21.63 8.765 20.37 0.755 Control group mean FU1, Man 0.477 212.6 0.992 0.958 48.02 9.842 39.67 0.686 Control group mean FU1, Woman 0.545 199.6 0.975 0.914 29.15 9.094 26.60 0.769 Control group mean FU2, Man 0.363 221.1 0.991 0.962 33.64 9.709 30 0.801 Control group mean FU2, Woman 0.440 210.4 0.978 0.928 17.48 9.035 15.96 0.905 Control group mean FU3, Man 0.603 254.7 0.987 0.941 41.64 7.592 36.51 0.723 Control group mean FU3, Woman 0.657 246 0.983 0.933 15.24 7.977 15.13 0.916 FU1,T1: Man - Woman (p-value) 0.208 0.303 0.542 0.203 0.913 0.860 0.560 0.715 FU1,T2: Man - Woman (p-value) 0.102 0.328 0.406 0.482 0.338 0.389 0.247 0.682 FU1, T3: Man - Woman (p-value) 0.230 0.767 0.938 0.0488 0.157 0.0128 0.925 0.196 FU2,T1: Man - Woman (p-value) 0.155 0.549 0.0939 0.575 0.508 0.0971 0.664 0.558 FU2,T2: Man - Woman (p-value) 0.509 0.501 0.711 0.294 0.467 0.311 0.504 0.739 FU2,T3: Man - Woman (p-value) 0.346 0.881 0.101 0.227 0.187 0.0273 0.190 0.361 FU3,T1: Man - Woman (p-value) 0.0854 0.474 0.0423 0.650 0.0220 0.0763 0.0248 0.0171 FU3,T2: Man - Woman (p-value) 0.239 0.577 0.120 0.746 0.249 0.352 0.159 0.855 FU3,T3: Man - Woman (p-value) 0.231 0.0288 0.314 0.337 0.962 0.486 0.136 0.944 Woman,T1: FU2 - FU1 (p-value) 0.794 0.962 0.681 0.221 0.585 0.995 0.694 0.766 Woman,T2: FU2 - FU1 (p-value) 0.138 0.553 0.336 0.00778 0.516 0.643 0.782 0.513 Woman,T3: FU2 - FU1 (p-value) 0.0686 0.297 0.0204 0.0788 0.130 0.346 0.600 0.864 Man,T1: FU2 - FU1 (p-value) 0.965 0.562 0.519 0.0430 0.435 0.0626 0.165 0.689 Man,T2: FU2 - FU1 (p-value) 0.795 0.901 0.534 0.452 0.902 0.566 0.499 0.827 Man,T3: FU2 - FU1 (p-value) 0.211 0.720 0.284 0.00972 0.795 0.566 0.155 0.0195 Woman,T1: FU3 - FU2 (p-value) 0.634 0.576 0.940 0.704 0.361 0.395 0.726 0.363 Woman,T2: FU3 - FU2 (p-value) 0.529 0.270 0.524 0.542 0.0823 0.0341 0.0725 0.0877 Woman,T3: FU3 - FU2 (p-value) 0.197 0.253 0.0106 0.00378 0.621 0.870 0.839 0.741 Man,T1: FU3 - FU2 (p-value) 0.896 0.421 0.503 0.660 0.00301 0.0720 0.0119 0.00634 Man,T2: FU3 - FU2 (p-value) 0.260 0.782 0.412 0.525 0.0650 0.890 0.300 0.165 Man,T3: FU3 - FU2 (p-value) 0.530 0.113 0.851 0.457 0.505 0.609 0.0118 0.579 Woman,T1: FU3 - FU1 (p-value) 0.511 0.623 0.666 0.585 0.264 0.406 0.571 0.277 Woman,T2: FU3 - FU1 (p-value) 0.0737 0.135 0.845 0.156 0.0912 0.0235 0.193 0.0420 Woman,T3: FU3 - FU1 (p-value) 0.848 0.707 0.730 0.175 0.143 0.808 0.587 0.667 Man,T1: FU3 - FU1 (p-value) 0.924 0.718 0.277 0.231 0.00254 0.00779 0.00212 0.00548 Man,T2: FU3 - FU1 (p-value) 0.178 0.858 0.267 0.263 0.146 0.702 0.168 0.284 Man,T3: FU3 - FU1 (p-value) 0.716 0.0619 0.458 0.184 0.748 0.797 0.325 0.0800 FU3,Woman: T1 - T2 (p-value) 0.202 0.864 0.415 0.957 0.0667 0.0783 0.00692 0.713 FU3,Woman: T1 - T3 (p-value) 0.651 0.146 0.423 0.231 0.397 0.0725 0.513 0.163 FU3,Woman: T2 - T3 (p-value) 0.0796 0.0971 0.979 0.247 0.00516 0.000367 0.000818 0.0784 FU3,Man: T1 - T2 (p-value) 0.662 0.241 0.703 0.409 0.940 0.231 0.0922 0.0205 FU3,Man: T1 - T3 (p-value) 0.369 0.798 0.567 0.721 0.0121 0.0158 5.93e-05 0.000848 FU3,Man: T2 - T3 (p-value) 0.178 0.149 0.349 0.641 0.0171 0.227 0.0455 0.330 40 Notes: The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Indicators of having a business are controlled for profits and revenues outcomes. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1. Insufficient employment: indicator variable equal to 1 if respondent is unemployment, underemployed (working less than 40 hours a week), or in vulnerable employment (unpaid apprentice or unpaid contributor to the family business). 41 Table A4: Impacts on expenditures, agency, and assets (full results) (1) (2) (3) (4) (5) (6) (7) Z-score total Total monthly DFITS Total IHS Total monthly Individual asset Household asset monthly expenditures - monthly Agency score expenditures score (0-17) * score (0-17) * VARIABLES expenditures winsorized 1% expenditures woman -0.434*** -15.046*** -0.851*** -13.361*** -0.238*** -1.563*** 0.150 (0.039) (1.532) (0.079) (1.299) (0.028) (0.140) (0.162) T1: training+grant x FU1 x man 0.098* 3.653 0.157** 0.719 0.109*** 0.130 0.096 (0.053) (2.349) (0.068) (1.701) (0.040) (0.180) (0.206) T1: training+grant x FU2 x man 0.106* 4.148* 0.132 2.354 0.098** 0.429** 0.380* (0.055) (2.292) (0.108) (1.705) (0.041) (0.177) (0.194) T1: training+grant x FU3 x man 0.195*** 11.388*** 0.207** 12.630*** 0.044 - - (0.071) (3.488) (0.083) (3.749) (0.043) T2: training only x FU1 x man 0.046 1.112 0.084 0.983 0.127*** 0.150 0.080 (0.053) (2.283) (0.073) (1.720) (0.040) (0.174) (0.195) T2: training only x FU2 x man 0.094 4.102* 0.108 1.920 0.160*** 0.199 0.111 (0.058) (2.440) (0.107) (1.799) (0.041) (0.179) (0.187) T2: training only x FU3 x man 0.122* 8.536** 0.078 6.220* 0.046 - - (0.072) (3.510) (0.098) (3.322) (0.044) T3: grant only x FU1 x man 0.118** 4.832** 0.159** 1.847 0.048 -0.210 -0.239 (0.053) (2.352) (0.069) (1.704) (0.038) (0.178) (0.205) T3: grant only x FU2 x man 0.224*** 9.648*** 0.218** 4.980*** -0.001 0.122 -0.027 (0.059) (2.526) (0.109) (1.767) (0.041) (0.181) (0.202) T3: grant only x FU3 x man 0.061 5.211 0.063 3.873 -0.065 - - (0.065) (3.182) (0.087) (3.045) (0.041) T1: training+grant x FU1 x woman 0.056 0.606 0.278*** 1.075 0.104*** 0.210** -0.109 (0.037) (1.342) (0.084) (1.028) (0.032) (0.101) (0.174) T1: training+grant x FU2 x woman 0.087** 2.736** 0.160 2.142** 0.091*** 0.097 0.194 (0.038) (1.316) (0.109) (1.052) (0.031) (0.108) (0.180) T1: training+grant x FU3 x woman 0.179*** 5.431** 0.365*** 2.418 0.064* - - (0.051) (2.136) (0.105) (2.092) (0.033) T2: training only x FU1 x woman 0.048 0.705 0.246*** 0.543 0.103*** 0.084 0.204 (0.038) (1.397) (0.085) (1.024) (0.032) (0.105) (0.177) T2: training only x FU2 x woman 0.019 0.360 0.078 0.363 0.094*** 0.123 0.335* (0.037) (1.279) (0.107) (1.037) (0.033) (0.115) (0.178) T2: training only x FU3 x woman 0.117** 3.235 0.229** 1.519 0.099*** - - (0.052) (2.226) (0.102) (2.233) (0.033) T3: grant only x FU1 x woman 0.113*** 2.731** 0.333*** 3.005*** 0.055* 0.101 0.049 (0.038) (1.381) (0.086) (1.084) (0.033) (0.096) (0.167) T3: grant only x FU2 x woman 0.178*** 6.376*** 0.304*** 4.261*** -0.009 0.111 0.117 (0.041) (1.506) (0.107) (1.130) (0.032) (0.102) (0.174) T3: grant only x FU3 x woman 0.196*** 6.441*** 0.368*** 4.955** 0.022 - - (0.052) (2.159) (0.106) (2.238) (0.036) Constant 0.187*** 29.681*** 8.976*** 26.752*** 0.081** 2.098*** 3.024*** (0.050) (2.120) (0.215) (1.651) (0.036) (0.176) (0.219) Observations 15,614 15,614 15,614 14,862 9,315 6,451 6,451 42 R-squared 0.199 0.176 0.111 0.190 0.160 0.539 0.319 Control group mean BL, Man 0.324 57.90 11.19 48.89 0.213 5.135 6.387 Control group mean BL, Woman -0.158 35.36 10.30 32.82 -0.147 1.496 6.282 Control group mean FU1, Man 0.392 44.94 11.01 40.67 0.202 5.303 6.588 Control group mean FU1, Woman -0.214 25.14 9.963 22.50 -0.162 1.721 6.611 Control group mean FU2, Man 0.402 40.91 10.70 36.66 0.164 5.224 6.367 Control group mean FU2, Woman -0.175 21.48 9.646 20.27 -0.186 1.754 6.513 Control group mean FU3, Man 0.281 45.04 10.92 45.42 0.108 Control group mean FU3, Woman -0.183 26.23 10.15 26.86 -0.134 FU1,T1: Man - Woman (p-value) 0.498 0.247 0.236 0.856 0.912 0.686 0.429 FU1,T2: Man - Woman (p-value) 0.975 0.875 0.128 0.823 0.620 0.736 0.622 FU1, T3: Man - Woman (p-value) 0.937 0.429 0.0969 0.560 0.876 0.113 0.257 FU2,T1: Man - Woman (p-value) 0.758 0.581 0.842 0.914 0.868 0.0963 0.465 FU2,T2: Man - Woman (p-value) 0.253 0.162 0.828 0.447 0.179 0.712 0.367 FU2,T3: Man - Woman (p-value) 0.503 0.254 0.536 0.728 0.868 0.954 0.578 FU3,T1: Man - Woman (p-value) 0.835 0.111 0.200 0.00768 0.687 FU3,T2: Man - Woman (p-value) 0.955 0.164 0.251 0.180 0.304 FU3,T3: Man - Woman (p-value) 0.0723 0.722 0.0167 0.738 0.0845 Woman,T1: FU2 - FU1 (p-value) 0.411 0.115 0.265 0.316 0.702 0.320 0.0649 Woman,T2: FU2 - FU1 (p-value) 0.419 0.796 0.112 0.859 0.802 0.745 0.437 Woman,T3: FU2 - FU1 (p-value) 0.0919 0.0139 0.781 0.260 0.0716 0.925 0.666 Man,T1: FU2 - FU1 (p-value) 0.870 0.820 0.813 0.311 0.792 0.0562 0.118 Man,T2: FU2 - FU1 (p-value) 0.339 0.162 0.814 0.557 0.412 0.744 0.862 Man,T3: FU2 - FU1 (p-value) 0.0223 0.0190 0.556 0.0406 0.235 0.0256 0.245 Woman,T1: FU3 - FU2 (p-value) 0.0669 0.191 0.0987 0.890 0.497 Woman,T2: FU3 - FU2 (p-value) 0.0443 0.162 0.204 0.579 0.905 Woman,T3: FU3 - FU2 (p-value) 0.729 0.976 0.600 0.745 0.442 Man,T1: FU3 - FU2 (p-value) 0.187 0.0230 0.505 0.00372 0.247 Man,T2: FU3 - FU2 (p-value) 0.683 0.163 0.794 0.159 0.0228 Man,T3: FU3 - FU2 (p-value) 0.0118 0.154 0.164 0.695 0.167 Woman,T1: FU3 - FU1 (p-value) 0.0164 0.0251 0.423 0.508 0.316 Woman,T2: FU3 - FU1 (p-value) 0.172 0.240 0.870 0.647 0.913 Woman,T3: FU3 - FU1 (p-value) 0.120 0.0955 0.758 0.379 0.447 Man,T1: FU3 - FU1 (p-value) 0.146 0.0182 0.528 0.000589 0.184 Man,T2: FU3 - FU1 (p-value) 0.270 0.0220 0.951 0.0936 0.0963 Man,T3: FU3 - FU1 (p-value) 0.362 0.900 0.251 0.482 0.0159 FU3,Woman: T1 - T2 (p-value) 0.259 0.346 0.228 0.677 0.315 FU3,Woman: T1 - T3 (p-value) 0.749 0.655 0.980 0.240 0.249 FU3,Woman: T2 - T3 (p-value) 0.154 0.173 0.222 0.135 0.0380 FU3,Man: T1 - T2 (p-value) 0.387 0.508 0.149 0.141 0.969 FU3,Man: T1 - T3 (p-value) 0.0878 0.126 0.0620 0.0349 0.0200 FU3,Man: T2 - T3 (p-value) 0.450 0.412 0.877 0.533 0.0197 FU2,Woman: T1 - T2 (p-value) 0.833 0.448 FU2,Woman: T1 - T3 (p-value) 0.901 0.672 FU2,Woman: T2 - T3 (p-value) 0.918 0.226 FU2,Man: T1 - T2 (p-value) 0.223 0.176 FU2,Man: T1 - T3 (p-value) 0.107 0.0559 FU2,Man: T2 - T3 (p-value) 0.690 0.503 43 Notes: * Information to compute individual and household asset scores was not collected at the 3rd follow-up survey. The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 44 Table A5: Impacts on components of agency (1) (2) (3) (4) (5) (6) Standardized Standardized Standardized independent Standardized Standardized Agency score decision making self efficacy decision making control score depression score score score VARIABLES score woman -0.238*** -0.620*** -0.469*** -0.229*** -0.094* -0.181*** (0.028) (0.048) (0.048) (0.048) (0.052) (0.049) T1: training+grant x FU1 x man 0.109*** -0.010 0.016 0.096 0.074 0.165*** (0.040) (0.074) (0.081) (0.068) (0.076) (0.064) T1: training+grant x FU2 x man 0.098** -0.096 0.040 0.073 0.099 0.131** (0.041) (0.078) (0.081) (0.072) (0.079) (0.064) T1: training+grant x FU3 x man 0.044 0.003 -0.091 0.069 0.180** -0.052 (0.043) (0.074) (0.085) (0.077) (0.083) (0.074) T2: training only x FU1 x man 0.127*** -0.059 -0.003 0.171*** 0.093 0.222*** (0.040) (0.070) (0.080) (0.065) (0.074) (0.062) T2: training only x FU2 x man 0.160*** 0.088 0.060 0.143** 0.249*** 0.091 (0.041) (0.069) (0.077) (0.070) (0.075) (0.063) T2: training only x FU3 x man 0.046 -0.018 -0.148* -0.073 0.325*** 0.015 (0.044) (0.078) (0.081) (0.080) (0.079) (0.071) T3: grant only x FU1 x man 0.048 0.061 0.064 0.007 0.024 0.077 (0.038) (0.067) (0.082) (0.065) (0.072) (0.064) T3: grant only x FU2 x man -0.001 -0.047 0.000 -0.040 0.029 -0.027 (0.041) (0.077) (0.081) (0.072) (0.073) (0.065) T3: grant only x FU3 x man -0.065 -0.081 0.056 -0.245*** 0.090 -0.178** (0.041) (0.076) (0.075) (0.081) (0.079) (0.074) T1: training+grant x FU1 x woman 0.104*** 0.034 0.038 0.176*** 0.086 0.122** (0.032) (0.064) (0.061) (0.054) (0.057) (0.055) T1: training+grant x FU2 x woman 0.091*** 0.101 0.126** 0.116** 0.084 0.036 (0.031) (0.063) (0.062) (0.055) (0.054) (0.058) T1: training+grant x FU3 x woman 0.064* 0.076 0.085 0.126** 0.019 0.112* (0.033) (0.067) (0.067) (0.064) (0.058) (0.061) T2: training only x FU1 x woman 0.103*** 0.015 0.026 0.181*** 0.105* 0.115** (0.032) (0.067) (0.060) (0.055) (0.057) (0.058) T2: training only x FU2 x woman 0.094*** 0.088 0.154** 0.128** 0.064 0.010 (0.033) (0.065) (0.062) (0.058) (0.056) (0.060) T2: training only x FU3 x woman 0.099*** 0.158** 0.016 0.217*** 0.130** 0.001 (0.033) (0.071) (0.069) (0.063) (0.061) (0.061) T3: grant only x FU1 x woman 0.055* 0.060 0.054 0.095* -0.024 0.086 (0.033) (0.064) (0.062) (0.055) (0.057) (0.056) T3: grant only x FU2 x woman -0.009 0.025 0.039 0.050 -0.087 -0.063 45 (0.032) (0.065) (0.064) (0.059) (0.053) (0.060) T3: grant only x FU3 x woman 0.022 0.054 0.149** 0.137** -0.025 -0.122* (0.036) (0.068) (0.066) (0.067) (0.061) (0.064) Constant 0.081** 0.575*** -0.209*** 0.060 -0.200*** 0.312*** (0.036) (0.066) (0.074) (0.063) (0.070) (0.065) Observations 9,315 7,047 7,119 15,415 15,177 15,342 R-squared 0.160 0.137 0.121 0.075 0.125 0.132 Control group mean BL, Man 0.213 0.329 0.317 0.132 0.0510 0.247 Control group mean BL, Woman -0.147 -0.193 -0.257 -0.0950 -0.0660 -0.109 Control group mean FU1, Man 0.202 0.443 0.392 0.203 0.0320 0.181 Control group mean FU1, Woman -0.162 -0.220 -0.200 -0.133 -0.0880 -0.114 Control group mean FU2, Man 0.164 0.437 0.333 0.139 0.0320 0.183 Control group mean FU2, Woman -0.186 -0.283 -0.278 -0.189 -0.0930 -0.126 Control group mean FU3, Man 0.108 0.377 0.237 -0.0600 -0.0130 0.120 Control group mean FU3, Woman -0.134 -0.291 -0.204 -0.0530 -0.136 -0.0400 FU1,T1: Man - Woman (p-value) 0.912 0.618 0.812 0.318 0.893 0.586 FU1,T2: Man - Woman (p-value) 0.620 0.396 0.757 0.895 0.892 0.177 FU1, T3: Man - Woman (p-value) 0.876 0.983 0.921 0.268 0.578 0.914 FU2,T1: Man - Woman (p-value) 0.868 0.0311 0.348 0.604 0.866 0.239 FU2,T2: Man - Woman (p-value) 0.179 0.993 0.293 0.858 0.0369 0.316 FU2,T3: Man - Woman (p-value) 0.868 0.430 0.678 0.301 0.174 0.665 FU3,T1: Man - Woman (p-value) 0.687 0.421 0.0781 0.525 0.0889 0.0659 FU3,T2: Man - Woman (p-value) 0.304 0.0676 0.0935 0.00140 0.0344 0.871 FU3,T3: Man - Woman (p-value) 0.0845 0.144 0.303 5.29e-05 0.215 0.538 Woman,T1: FU2 - FU1 (p-value) 0.702 0.390 0.268 0.340 0.981 0.161 Woman,T2: FU2 - FU1 (p-value) 0.802 0.368 0.109 0.411 0.518 0.0907 Woman,T3: FU2 - FU1 (p-value) 0.0716 0.658 0.849 0.490 0.302 0.0149 Man,T1: FU2 - FU1 (p-value) 0.792 0.344 0.805 0.766 0.732 0.588 Man,T2: FU2 - FU1 (p-value) 0.412 0.0707 0.478 0.709 0.0427 0.0473 Man,T3: FU2 - FU1 (p-value) 0.235 0.216 0.507 0.536 0.949 0.120 Woman,T1: FU3 - FU2 (p-value) 0.497 0.775 0.640 0.900 0.335 0.277 Woman,T2: FU3 - FU2 (p-value) 0.905 0.438 0.128 0.245 0.326 0.905 Woman,T3: FU3 - FU2 (p-value) 0.442 0.742 0.212 0.269 0.364 0.419 Man,T1: FU3 - FU2 (p-value) 0.247 0.313 0.226 0.967 0.291 0.0154 Man,T2: FU3 - FU2 (p-value) 0.0228 0.266 0.0476 0.0222 0.353 0.326 Man,T3: FU3 - FU2 (p-value) 0.167 0.731 0.574 0.0237 0.455 0.0498 Woman,T1: FU3 - FU1 (p-value) 0.316 0.624 0.607 0.510 0.367 0.885 Woman,T2: FU3 - FU1 (p-value) 0.913 0.129 0.914 0.637 0.743 0.0918 Woman,T3: FU3 - FU1 (p-value) 0.447 0.947 0.292 0.600 0.988 0.00328 Man,T1: FU3 - FU1 (p-value) 0.184 0.895 0.326 0.779 0.212 0.00642 Man,T2: FU3 - FU1 (p-value) 0.0963 0.665 0.165 0.00588 0.00781 0.00723 Man,T3: FU3 - FU1 (p-value) 0.0159 0.127 0.938 0.00526 0.460 0.00148 FU3,Woman: T1 - T2 (p-value) 0.315 0.281 0.360 0.164 0.0702 0.0830 46 FU3,Woman: T1 - T3 (p-value) 0.249 0.767 0.380 0.868 0.475 0.000503 FU3,Woman: T2 - T3 (p-value) 0.0380 0.176 0.0747 0.247 0.0158 0.0649 FU3,Man: T1 - T2 (p-value) 0.969 0.808 0.551 0.0977 0.112 0.404 FU3,Man: T1 - T3 (p-value) 0.0200 0.308 0.108 0.000302 0.320 0.134 FU3,Man: T2 - T3 (p-value) 0.0197 0.459 0.0197 0.0528 0.00681 0.0171 Notes: The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 47 Table A6: Impacts on self-employment and business starts, deaths, and transitions (full results) (1) (2) (3) (4) Nb. of own Have changed Own business Death of business VARIABLES businesses business since BL woman 0.029 0.043 -0.024 0.082** (0.018) (0.050) (0.016) (0.032) T1: training+grant x FU1 x man 0.143*** 0.341*** -0.075*** 0.015 (0.020) (0.063) (0.017) (0.041) T1: training+grant x FU2 x man 0.094*** 0.163** -0.048** -0.014 (0.022) (0.067) (0.021) (0.043) T1: training+grant x FU3 x man 0.064*** 0.229*** -0.038* -0.000 (0.023) (0.070) (0.020) (0.044) T2: training only x FU1 x man 0.065*** 0.169*** -0.052*** 0.039 (0.024) (0.065) (0.020) (0.041) T2: training only x FU2 x man 0.028 0.030 -0.017 -0.000 (0.025) (0.070) (0.024) (0.044) T2: training only x FU3 x man 0.071*** 0.126* -0.050*** 0.042 (0.022) (0.068) (0.019) (0.045) T3: grant only x FU1 x man 0.116*** 0.187*** -0.043** -0.033 (0.022) (0.062) (0.021) (0.039) T3: grant only x FU2 x man 0.055** 0.006 -0.030 0.006 (0.024) (0.066) (0.023) (0.043) T3: grant only x FU3 x man 0.071*** 0.045 -0.052*** -0.035 (0.022) (0.066) (0.019) (0.044) T1: training+grant x FU1 x woman 0.096*** 0.221*** -0.039*** 0.004 (0.017) (0.048) (0.015) (0.035) T1: training+grant x FU2 x woman 0.105*** 0.156*** -0.047*** 0.030 (0.016) (0.050) (0.014) (0.036) T1: training+grant x FU3 x woman 0.071*** 0.238*** -0.023 -0.027 (0.017) (0.058) (0.015) (0.038) T2: training only x FU1 x woman 0.065*** 0.203*** -0.031** -0.014 (0.018) (0.049) (0.015) (0.034) T2: training only x FU2 x woman 0.051*** 0.126** -0.015 -0.019 (0.019) (0.053) (0.017) (0.036) T2: training only x FU3 x woman 0.056*** 0.177*** -0.030** -0.019 (0.017) (0.057) (0.014) (0.038) T3: grant only x FU1 x woman 0.078*** 0.254*** -0.020 0.030 (0.018) (0.050) (0.016) (0.035) T3: grant only x FU2 x woman 0.032 0.039 -0.008 0.013 (0.019) (0.053) (0.017) (0.036) T3: grant only x FU3 x woman 0.053*** 0.137** -0.015 -0.027 (0.018) (0.058) (0.015) (0.037) Constant 0.664*** 1.184*** 0.149*** 0.320*** (0.024) (0.066) (0.023) (0.043) Observations 9,336 9,336 6,519 6,213 R-squared 0.107 0.083 0.024 0.036 Control group mean BL, Man 0.700 0 0 Control group mean BL, Woman 0.667 0 0 Control group mean FU1, Man 0.830 1.327 0.0780 0.263 Control group mean FU1, Woman 0.825 1.313 0.0730 0.317 Control group mean FU2, Man 0.812 1.452 0.104 0.369 Control group mean FU2, Woman 0.857 1.485 0.0590 0.463 Control group mean FU3, Man 0.853 1.554 0.0740 0.405 Control group mean FU3, Woman 0.882 1.645 0.0500 0.446 FU1,T1: Man - Woman (p-value) 0.0496 0.116 0.0701 0.835 FU1,T2: Man - Woman (p-value) 0.981 0.672 0.361 0.306 FU1, T3: Man - Woman (p-value) 0.144 0.383 0.347 0.215 FU2,T1: Man - Woman (p-value) 0.664 0.926 0.933 0.407 FU2,T2: Man - Woman (p-value) 0.433 0.245 0.961 0.725 FU2,T3: Man - Woman (p-value) 0.429 0.686 0.418 0.900 FU3,T1: Man - Woman (p-value) 0.775 0.913 0.516 0.622 48 FU3,T2: Man - Woman (p-value) 0.569 0.537 0.359 0.270 FU3,T3: Man - Woman (p-value) 0.482 0.264 0.0947 0.882 Woman,T1: FU2 - FU1 (p-value) 0.626 0.193 0.652 0.455 Woman,T2: FU2 - FU1 (p-value) 0.494 0.143 0.399 0.878 Woman,T3: FU2 - FU1 (p-value) 0.0167 2.45e-05 0.533 0.647 Man,T1: FU2 - FU1 (p-value) 0.0162 0.00352 0.185 0.463 Man,T2: FU2 - FU1 (p-value) 0.112 0.0223 0.119 0.335 Man,T3: FU2 - FU1 (p-value) 0.00608 0.00185 0.540 0.314 Woman,T1: FU3 - FU2 (p-value) 0.0556 0.165 0.162 0.139 Woman,T2: FU3 - FU2 (p-value) 0.801 0.387 0.410 0.992 Woman,T3: FU3 - FU2 (p-value) 0.297 0.0862 0.714 0.300 Man,T1: FU3 - FU2 (p-value) 0.202 0.317 0.612 0.762 Man,T2: FU3 - FU2 (p-value) 0.0676 0.170 0.112 0.285 Man,T3: FU3 - FU2 (p-value) 0.478 0.537 0.329 0.341 Woman,T1: FU3 - FU1 (p-value) 0.191 0.769 0.339 0.451 Woman,T2: FU3 - FU1 (p-value) 0.687 0.646 0.948 0.899 Woman,T3: FU3 - FU1 (p-value) 0.226 0.0484 0.771 0.151 Man,T1: FU3 - FU1 (p-value) 0.000350 0.0999 0.0496 0.742 Man,T2: FU3 - FU1 (p-value) 0.812 0.542 0.892 0.943 Man,T3: FU3 - FU1 (p-value) 0.0433 0.0282 0.658 0.958 FU3,Woman: T1 - T2 (p-value) 0.320 0.306 0.548 0.833 FU3,Woman: T1 - T3 (p-value) 0.234 0.0971 0.599 0.999 FU3,Woman: T2 - T3 (p-value) 0.839 0.505 0.261 0.830 FU3,Man: T1 - T2 (p-value) 0.735 0.160 0.492 0.379 FU3,Man: T1 - T3 (p-value) 0.724 0.0108 0.402 0.456 FU3,Man: T2 - T3 (p-value) 0.989 0.253 0.877 0.105 Notes: The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. * Baseline outcome is not controlled in column (4). Information to compute the number of businesses was not collected at baseline. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 49 Table A7: Impacts on primary outcomes conditional on having business at baseline (1) (2) (3) (4) (5) Revenues from own Profits from own business Z-score profits and Total earnings over the past business in the past month - in the past month - Insufficient employment revenues month - winsorized 1% VARIABLES winsorized 1% winsorized 1% woman -0.108** -18.594*** -7.876*** -15.200*** 0.041* (0.046) (6.183) (1.700) (2.799) (0.023) T1: training+grant x FU1 x man -0.023 -9.848 -2.548 0.853 0.042 (0.067) (8.757) (2.962) (5.650) (0.035) T1: training+grant x FU2 x man 0.187*** 13.781 4.101* 3.061 -0.019 (0.065) (9.074) (2.352) (3.359) (0.034) T1: training+grant x FU3 x man 0.134 27.878** 2.321 18.679*** 0.020 (0.085) (13.437) (4.317) (5.946) (0.043) T2: training only x FU1 x man 0.111* 7.900 1.734 1.583 -0.011 (0.063) (9.474) (2.770) (4.855) (0.034) T2: training only x FU2 x man 0.139** 18.090* 3.087 5.591* -0.015 (0.070) (10.344) (2.496) (3.354) (0.033) T2: training only x FU3 x man 0.299*** 48.647*** 16.364*** 21.114*** 0.086** (0.094) (15.273) (5.316) (6.498) (0.041) T3: grant only x FU1 x man 0.041 8.004 -1.367 4.799 -0.022 (0.067) (9.818) (2.736) (5.933) (0.036) T3: grant only x FU2 x man 0.066 14.567 1.473 3.057 0.010 (0.069) (10.734) (2.363) (3.411) (0.035) T3: grant only x FU3 x man -0.018 21.044 -3.830 2.059 0.034 (0.085) (13.872) (4.507) (5.761) (0.043) T1: training+grant x FU1 x woman -0.031 -5.630 -1.891 -1.757 0.001 (0.037) (4.819) (1.472) (3.353) (0.030) T1: training+grant x FU2 x woman -0.046 -0.760 -0.906 -1.076 -0.031 (0.039) (5.179) (1.144) (1.813) (0.029) T1: training+grant x FU3 x woman 0.075 11.212 -0.030 3.562 -0.023 (0.051) (7.359) (2.498) (3.400) (0.036) T2: training only x FU1 x woman 0.055 -3.795 2.718* 2.382 -0.053* (0.037) (5.080) (1.534) (3.626) (0.029) T2: training only x FU2 x woman 0.105** 11.592* 4.105*** 3.870** -0.041 (0.042) (6.510) (1.441) (1.963) (0.029) T2: training only x FU3 x woman 0.220*** 28.063*** 7.551*** 9.500*** 0.000 (0.053) (8.668) (2.663) (3.279) (0.036) T3: grant only x FU1 x woman -0.121*** -11.528*** -4.179*** -6.006* -0.074** (0.037) (4.378) (1.395) (3.179) (0.029) T3: grant only x FU2 x woman -0.104*** -2.934 -1.653 -0.973 -0.030 (0.038) (4.681) (1.125) (1.733) (0.029) T3: grant only x FU3 x woman 0.030 10.222 0.506 1.776 -0.060* (0.051) (6.914) (2.476) (3.136) (0.036) Constant -0.009 59.667*** 12.716*** 0.422*** (0.064) (9.977) (2.376) 57.907*** (0.035) (4.712) 50 Observations 10,440 10,455 10,461 10,714 R-squared 0.094 0.150 0.120 10,621 0.103 Control group mean BL, Man 0.0980 52.47 15.33 0.128 0.594 Control group mean BL, Woman -0.0810 37.92 8.977 34.50 0.726 Control group mean FU1, Man 0.0920 67.92 17.41 21.63 0.477 Control group mean FU1, Woman -0.0600 43.45 9.758 48.02 0.545 Control group mean FU2, Man 0.114 62.31 18.77 29.15 0.363 Control group mean FU2, Woman -0.0710 41.96 10.46 33.64 0.440 Control group mean FU3, Man 0.0710 70.60 21.79 17.48 0.603 Control group mean FU3, Woman -0.0450 44.70 7.986 41.64 0.657 FU1,T1: Man - Woman (p-value) 0.920 0.657 0.835 15.24 0.329 FU1,T2: Man - Woman (p-value) 0.424 0.257 0.743 0.665 0.303 FU1, T3: Man - Woman (p-value) 0.0284 0.0583 0.331 0.884 0.217 FU2,T1: Man - Woman (p-value) 0.00131 0.153 0.0522 0.0812 0.766 FU2,T2: Man - Woman (p-value) 0.661 0.589 0.722 0.292 0.527 FU2,T3: Man - Woman (p-value) 0.0249 0.128 0.227 0.667 0.342 FU3,T1: Man - Woman (p-value) 0.518 0.263 0.608 0.306 0.396 FU3,T2: Man - Woman (p-value) 0.437 0.232 0.117 0.0197 0.0788 FU3,T3: Man - Woman (p-value) 0.602 0.474 0.362 0.0937 0.0654 Woman,T1: FU2 - FU1 (p-value) 0.716 0.346 0.517 0.963 0.358 Woman,T2: FU2 - FU1 (p-value) 0.262 0.0149 0.430 0.827 0.724 Woman,T3: FU2 - FU1 (p-value) 0.681 0.0576 0.0839 0.667 0.195 Man,T1: FU2 - FU1 (p-value) 0.000591 0.00135 0.0155 0.0934 0.135 Man,T2: FU2 - FU1 (p-value) 0.654 0.228 0.598 0.653 0.909 Man,T3: FU2 - FU1 (p-value) 0.687 0.396 0.289 0.338 0.448 Woman,T1: FU3 - FU2 (p-value) 0.0234 0.0611 0.732 0.744 0.856 Woman,T2: FU3 - FU2 (p-value) 0.0282 0.0365 0.176 0.140 0.334 Woman,T3: FU3 - FU2 (p-value) 0.00711 0.0266 0.358 0.0596 0.487 Man,T1: FU3 - FU2 (p-value) 0.505 0.219 0.659 0.321 0.452 Man,T2: FU3 - FU2 (p-value) 0.0519 0.0170 0.00455 0.00250 0.0427 Man,T3: FU3 - FU2 (p-value) 0.293 0.595 0.253 0.00528 0.660 Woman,T1: FU3 - FU1 (p-value) 0.0482 0.0146 0.486 0.848 0.591 Woman,T2: FU3 - FU1 (p-value) 0.00328 0.000146 0.0846 0.199 0.234 Woman,T3: FU3 - FU1 (p-value) 0.00632 0.00122 0.0878 0.109 0.742 Man,T1: FU3 - FU1 (p-value) 0.0428 0.00105 0.251 0.0516 0.661 Man,T2: FU3 - FU1 (p-value) 0.0307 0.00614 0.00296 0.00493 0.0536 Man,T3: FU3 - FU1 (p-value) 0.468 0.292 0.595 0.00191 0.292 FU3,Woman: T1 - T2 (p-value) 0.0106 0.0786 0.00941 0.684 0.555 FU3,Woman: T1 - T3 (p-value) 0.422 0.901 0.845 0.106 0.358 FU3,Woman: T2 - T3 (p-value) 0.000789 0.0522 0.0147 0.613 0.128 FU3,Man: T1 - T2 (p-value) 0.129 0.268 0.0240 0.0244 0.168 FU3,Man: T1 - T3 (p-value) 0.128 0.698 0.266 0.756 0.781 FU3,Man: T2 - T3 (p-value) 0.00343 0.147 0.00145 0.0216 0.280 Notes: This table includes only individuals with a business at baseline. The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Columns 1, 2, 3 also control for indicators of having a business. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1. Insufficient employment: indicator variable equal to 1 if respondent is unemployment, underemployed (working less than 40 hours a week), or in vulnerable employment (unpaid apprentice or unpaid contributor to the family business). 51 Table A8: Impacts on labor and capital inputs, full results (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) Value of K Value of K Value of Number of Value of K investment Number of investment productive Value of Value of Value of Business employees stock - Value of K (flow 12 paid (flow 12 assets - productive stock/inventory - stock/inventory practice paid or winsorized stock (ihs) month) - employees month) winsorized assets (ihs) winsorized 1% (ihs) score * unpaid 1% winsorized (ihs) 1% VARIABLES 1% woman -0.667*** -0.497*** -214.553*** -0.862*** -39.077*** -1.863*** -191.362*** -1.209*** -13.576** 1.273*** 0.054 (0.180) (0.121) (31.836) (0.149) (8.297) (0.292) (27.836) (0.166) (6.145) (0.307) (0.262) T1: training+grant x FU1 x man -0.121 0.098 67.588 0.718*** 60.232*** 2.223*** 54.725 0.895*** 10.267 0.379 1.547*** (55.357) (0.238) (0.194) (65.532) (0.204) (16.010) (0.431) (0.216) (10.573) (0.481) (0.361) T1: training+grant x FU2 x man -0.411 -0.009 -2.723 0.362* 26.881** 0.771 10.784 0.401* -8.192 -0.573 1.452*** (0.252) (0.187) (49.379) (0.201) (13.035) (0.475) (45.156) (0.209) (9.231) (0.479) (0.375) T1: training+grant x FU3 x man -0.380 -0.047 320.482*** 0.573*** 120.351*** 1.120** 341.133*** 0.544** -6.763 -0.497 - (0.295) (0.210) (86.567) (0.211) (27.975) (0.521) (83.069) (0.223) (11.019) (0.513) T2: training only x FU1 x man 0.243 0.172 56.602 0.607*** 25.956 -0.010 44.764 0.735*** -2.209 0.045 2.154*** (0.292) (0.248) (69.892) (0.199) (17.262) (0.470) (57.349) (0.216) (9.226) (0.482) (0.402) T2: training only x FU2 x man -0.337 -0.139 -57.627 0.122 -15.317 -0.499 -56.641 -0.112 6.134 1.263** 1.565*** (0.238) (0.148) (47.363) (0.203) (10.639) (0.477) (41.686) (0.237) (9.979) (0.499) (0.393) T2: training only x FU3 x man 0.186 -0.032 59.932 0.208 35.956* -0.014 56.282 0.061 16.708 -0.099 - (0.305) (0.187) (66.122) (0.211) (20.889) (0.514) (63.011) (0.233) (13.467) (0.518) T3: grant only x FU1 x man -0.188 -0.203 35.321 0.164 66.530*** 1.949*** 12.650 0.231 9.167 0.775 0.654* (0.224) (0.164) (63.718) (0.257) (18.142) (0.441) (52.142) (0.273) (9.718) (0.490) (0.344) T3: grant only x FU2 x man -0.618*** -0.214 -58.623 0.079 -3.134 -0.072 -35.516 -0.038 -11.086 -0.578 0.388 (0.211) (0.154) (46.870) (0.214) (10.891) (0.485) (43.443) (0.244) (9.306) (0.482) (0.364) T3: grant only x FU3 x man -0.180 0.095 80.775 0.208 48.677** -0.174 73.148 0.142 10.537 -0.293 - (0.283) (0.207) (68.755) (0.204) (24.614) (0.512) (64.157) (0.218) (11.573) (0.525) T1: training+grant x FU1 x woman 0.615*** 0.235** 98.175*** 1.059*** 65.345*** 4.218*** 73.113*** 1.227*** 14.833*** 1.274*** 2.235*** (0.168) (0.102) (28.240) (0.164) (9.313) (0.340) (22.521) (0.214) (5.344) (0.370) (0.283) T1: training+grant x FU2 x woman 0.327** 0.247*** 56.491*** 0.528*** 30.544*** 1.489*** 59.046*** 0.741*** -3.341 -0.057 1.213*** (0.160) (0.082) (19.009) (0.147) (6.651) (0.377) (17.829) (0.179) (4.308) (0.375) (0.277) T1: training+grant x FU3 x woman 0.483** 0.368** 78.933*** 0.615*** 15.024 1.235*** 63.883*** 0.855*** 10.332 0.126 - (0.199) (0.147) (27.810) (0.150) (10.384) (0.412) (24.265) (0.170) (7.522) (0.428) T2: training only x FU1 x woman 0.276* 0.064 -7.549 0.115 -2.083 0.889** -5.912 0.277 -1.987 -0.243 1.611*** (0.155) (0.084) (23.405) (0.177) (7.001) (0.366) (20.844) (0.223) (4.695) (0.375) (0.285) T2: training only x FU2 x woman 0.075 0.082 18.703 0.199 6.243 -0.168 18.332 0.273 -2.590 -0.364 1.246*** (0.164) (0.075) (18.842) (0.145) (5.511) (0.368) (17.141) (0.187) (4.960) (0.378) (0.288) T2: training only x FU3 x woman 0.326* 0.159 57.819* 0.333** 9.120 0.620 40.491 0.354* 7.922 -0.224 - (0.186) (0.103) (29.865) (0.155) (10.948) (0.405) (26.144) (0.189) (7.627) (0.427) T3: grant only x FU1 x woman 0.293* 0.101 29.393 0.630*** 27.259*** 2.671*** 15.130 0.283 18.672*** 1.248*** 0.483* (0.153) (0.078) (20.955) (0.174) (7.341) (0.365) (18.754) (0.240) (5.049) (0.377) (0.252) T3: grant only x FU2 x woman 0.223 0.103* 65.942*** 0.326** 15.858*** 0.883** 49.221*** 0.151 9.001* 0.434 -0.146 (0.169) (0.058) (21.902) (0.161) (6.083) (0.374) (18.760) (0.202) (5.241) (0.386) (0.277) T3: grant only x FU3 x woman 0.263 0.110 55.634* 0.347** 7.825 0.099 48.509* 0.496*** 2.800 0.320 - (0.172) (0.076) (31.258) (0.165) (10.776) (0.404) (27.875) (0.184) (6.101) (0.418) 52 Constant 1.525*** 0.973*** 334.763*** 10.604*** 66.280*** 6.964*** 281.489*** 10.008*** 40.370*** 3.620*** 5.058*** (0.261) (0.192) (48.229) (0.238) (12.214) (0.433) (41.921) (0.260) (8.116) (0.413) (0.374) Observations 6,210 6,215 6,215 6,215 6,218 6,218 6,218 6,218 6,215 6,215 4,214 R-squared 0.249 0.188 0.173 0.109 0.102 0.095 0.171 0.133 0.074 0.066 0.148 Control group mean BL, Man 2.206 0.716 289.9 11.66 126.7 7.843 259.4 11.28 24.84 3.728 7.816 Control group mean BL, Woman 1.212 0.246 97.97 10.17 33.22 6.100 75.31 9.282 18.87 4.395 7.448 Control group mean FU1, Man 2.193 0.883 401.9 11.90 73.58 7.418 346.8 11.65 35.70 4.177 8.119 Control group mean FU1, Woman 1.032 0.158 104.5 10.81 25.72 5.884 84.78 10.01 19.74 5.460 8.134 Control group mean FU2, Man 1.996 0.552 316.3 12.14 42.79 6.548 270.5 11.84 39.11 4.834 7.902 Control group mean FU2, Woman 1.090 0.0940 116.6 11.13 15.94 4.824 95.29 10.51 20.96 5.486 7.936 Control group mean FU3, Man 2.332 0.927 484.6 12.51 88.95 6.947 435.1 12.26 40.53 4.178 Control group mean FU3, Woman 1.082 0.176 150 11.55 25.35 4.995 111.1 10.93 35.97 6.021 FU1,T1: Man - Woman (p-value) 0.0100 0.519 0.650 0.142 0.767 5.33e-05 0.746 0.209 0.690 0.108 0.123 FU1,T2: Man - Woman (p-value) 0.918 0.668 0.360 0.0391 0.107 0.0995 0.384 0.0916 0.982 0.609 0.260 FU1, T3: Man - Woman (p-value) 0.0677 0.0677 0.925 0.103 0.0327 0.164 0.962 0.877 0.368 0.407 0.679 FU2,T1: Man - Woman (p-value) 0.0112 0.225 0.256 0.475 0.801 0.199 0.312 0.182 0.613 0.355 0.593 FU2,T2: Man - Woman (p-value) 0.136 0.172 0.132 0.738 0.0705 0.551 0.0933 0.177 0.413 0.00492 0.497 FU2,T3: Man - Woman (p-value) 0.00103 0.0463 0.0155 0.327 0.125 0.0916 0.0717 0.528 0.0476 0.0751 0.224 FU3,T1: Man - Woman (p-value) 0.0107 0.101 0.00587 0.866 0.000234 0.848 0.000874 0.239 0.173 0.305 FU3,T2: Man - Woman (p-value) 0.676 0.352 0.975 0.613 0.224 0.286 0.806 0.305 0.555 0.838 FU3,T3: Man - Woman (p-value) 0.153 0.941 0.726 0.572 0.110 0.645 0.709 0.188 0.529 0.316 Woman,T1: FU2 - FU1 (p-value) 0.0659 0.915 0.195 0.00347 0.000564 2.65e-09 0.596 0.0340 0.000909 0.00194 0.000679 Woman,T2: FU2 - FU1 (p-value) 0.162 0.854 0.356 0.663 0.271 0.0234 0.337 0.986 0.902 0.787 0.221 Woman,T3: FU2 - FU1 (p-value) 0.663 0.980 0.144 0.142 0.151 0.000107 0.0980 0.593 0.125 0.0840 0.0221 Man,T1: FU2 - FU1 (p-value) 0.195 0.525 0.227 0.0789 0.0387 0.00879 0.384 0.0227 0.0458 0.0764 0.791 Man,T2: FU2 - FU1 (p-value) 0.0404 0.221 0.0756 0.0212 0.00835 0.361 0.0513 0.000913 0.375 0.0319 0.137 Man,T3: FU2 - FU1 (p-value) 0.0431 0.945 0.0766 0.749 6.90e-05 0.000479 0.271 0.364 0.0519 0.0154 0.441 Woman,T1: FU3 - FU2 (p-value) 0.449 0.414 0.470 0.641 0.194 0.632 0.866 0.604 0.0824 0.728 Woman,T2: FU3 - FU2 (p-value) 0.152 0.508 0.240 0.458 0.813 0.126 0.462 0.732 0.200 0.784 Woman,T3: FU3 - FU2 (p-value) 0.823 0.940 0.750 0.916 0.490 0.136 0.981 0.169 0.366 0.828 Man,T1: FU3 - FU2 (p-value) 0.914 0.875 0.000166 0.366 0.00140 0.585 9.48e-05 0.565 0.897 0.903 Man,T2: FU3 - FU2 (p-value) 0.0567 0.562 0.0396 0.707 0.0117 0.416 0.0490 0.532 0.460 0.0305 Man,T3: FU3 - FU2 (p-value) 0.0743 0.111 0.0264 0.559 0.0448 0.879 0.0713 0.477 0.0808 0.645 Woman,T1: FU3 - FU1 (p-value) 0.583 0.452 0.634 0.0198 0.000526 2.21e-09 0.790 0.115 0.578 0.0289 Woman,T2: FU3 - FU1 (p-value) 0.782 0.407 0.0927 0.273 0.367 0.607 0.181 0.763 0.218 0.972 Woman,T3: FU3 - FU1 (p-value) 0.872 0.932 0.488 0.188 0.138 9.41e-07 0.319 0.437 0.0293 0.0695 Man,T1: FU3 - FU1 (p-value) 0.403 0.565 0.00717 0.540 0.0502 0.0878 0.00128 0.172 0.173 0.161 Man,T2: FU3 - FU1 (p-value) 0.870 0.444 0.968 0.0746 0.674 0.996 0.872 0.00787 0.168 0.815 Man,T3: FU3 - FU1 (p-value) 0.979 0.175 0.447 0.867 0.513 0.00102 0.302 0.748 0.912 0.100 FU3,Woman: T1 - T2 (p-value) 0.497 0.221 0.494 0.0687 0.611 0.163 0.368 0.00805 0.783 0.452 FU3,Woman: T1 - T3 (p-value) 0.310 0.0876 0.468 0.104 0.534 0.0101 0.581 0.0517 0.312 0.670 FU3,Woman: T2 - T3 (p-value) 0.759 0.655 0.949 0.934 0.914 0.230 0.786 0.484 0.498 0.233 FU3,Man: T1 - T2 (p-value) 0.0956 0.946 0.00672 0.112 0.00848 0.0567 0.00219 0.0484 0.118 0.494 FU3,Man: T1 - T3 (p-value) 0.535 0.569 0.0147 0.101 0.0380 0.0295 0.00434 0.0805 0.195 0.728 FU3,Man: T2 - T3 (p-value) 0.266 0.579 0.796 0.999 0.662 0.784 0.826 0.737 0.689 0.743 FU2,Woman: T1 - T2 (p-value) 0.910 FU2,Woman: T1 - T3 (p-value) 2.39e-06 FU2,Woman: T2 - T3 (p-value) 3.00e-06 FU2,Man: T1 - T2 (p-value) 0.798 FU2,Man: T1 - T3 (p-value) 0.0104 FU2,Man: T2 - T3 (p-value) 0.00640 Notes: * Information to compute the business practice score was not collected at the 3rd follow-up survey. The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1 53 Table A9: Impacts on transfers and intra-household dynamics (full results) (1) (2) (3) (4) (5) (6) (7) (8) Amount of IHS Amount of monthly IHS Total amount IHS Total Total amount IHS Total Amount of monthly popote Amount of transferred TO amount transferred BY amount monthly popote transfers monthly respondent in transferred TO respondent in transferred BY popote transfers TO FROM popote past 30d - respondent in past 30d - respondent in transfers spouse - spouse - transfers winsorized 1% past 30d winsorized 1% past 30d FROM winsorized winsorized TO spouse spouse 1% VARIABLES 1% woman 0.658** 1.431*** -1.276*** -1.137*** 9.006*** 5.031*** -7.453*** -3.725*** (0.331) (0.181) (0.223) (0.179) (0.565) (0.194) (0.628) (0.229) T1: training+grant x FU1 x man 0.049 0.621** 1.090** 0.595* 0.626 0.116 0.362 0.638* (0.585) (0.292) (0.495) (0.328) (0.663) (0.229) (1.045) (0.374) T1: training+grant x FU2 x man -0.261 -1.592*** 0.616 0.760** 0.558 -0.073 2.325** 1.130*** (0.572) (0.304) (0.399) (0.316) (0.702) (0.228) (1.012) (0.370) T1: training+grant x FU3 x man 0.590 0.685** 1.723*** 1.315*** -0.008 0.054 3.769** -0.080 (0.549) (0.280) (0.513) (0.345) (0.928) (0.315) (1.513) (0.382) T2: training only x FU1 x man 0.020 0.794*** 0.880* 0.606* 0.475 0.410* 1.953* 0.733** (0.580) (0.295) (0.456) (0.321) (0.574) (0.243) (1.151) (0.371) T2: training only x FU2 x man -0.397 -1.106*** 0.118 0.076 1.454* 0.415* 0.998 0.777** (0.479) (0.310) (0.378) (0.297) (0.792) (0.250) (0.990) (0.362) T2: training only x FU3 x man 1.815*** 1.253*** 0.600 0.841** -0.402 0.384 0.101 -0.122 (0.659) (0.301) (0.418) (0.332) (0.864) (0.316) (1.264) (0.362) T3: grant only x FU1 x man -0.976* -0.087 0.476 0.030 0.135 -0.019 -0.715 0.229 (0.500) (0.258) (0.457) (0.305) (0.605) (0.219) (1.020) (0.362) T3: grant only x FU2 x man -0.883* -1.428*** 0.678* 0.490 -0.190 -0.125 0.917 0.461 (0.459) (0.297) (0.397) (0.302) (0.567) (0.220) (0.987) (0.367) T3: grant only x FU3 x man 1.302** 1.109*** 1.176** 0.704** -1.651** -0.550* -0.801 -0.749** (0.608) (0.290) (0.489) (0.330) (0.700) (0.295) (1.223) (0.373) T1: training+grant x FU1 x woman -1.084** -0.914*** -0.005 0.170 -0.907 0.410 0.621 0.255 (0.435) (0.228) (0.232) (0.199) (0.814) (0.276) (0.491) (0.191) T1: training+grant x FU2 x woman 1.019** 0.938*** 0.417* 0.439** 0.330 0.131 1.298*** 0.246 (0.436) (0.239) (0.214) (0.200) (0.943) (0.284) (0.448) (0.206) T1: training+grant x FU3 x woman -0.624* -0.549** 0.955*** 0.855*** 0.879 -0.192 -0.401 0.348 (0.327) (0.226) (0.308) (0.234) (1.151) (0.292) (0.672) (0.284) T2: training only x FU1 x woman -0.505 -0.391 0.251 0.434** 1.208 0.616** -0.019 0.017 (0.459) (0.244) (0.249) (0.209) (0.958) (0.282) (0.422) (0.181) T2: training only x FU2 x woman 1.413*** 1.242*** 0.199 0.203 0.399 0.432 0.322 -0.135 (0.466) (0.232) (0.204) (0.189) (0.944) (0.278) (0.356) (0.190) T2: training only x FU3 x woman 0.274 -0.202 0.429* 0.570** -0.893 -0.629** -0.555 0.087 (0.401) (0.242) (0.260) (0.223) (1.084) (0.297) (0.665) (0.278) 54 T3: grant only x FU1 x woman -1.159*** -0.959*** -0.378* -0.148 -0.566 0.221 1.151** 0.495** (0.425) (0.227) (0.207) (0.189) (0.838) (0.277) (0.539) (0.196) T3: grant only x FU2 x woman 0.740* 0.873*** 0.438** 0.513*** 0.376 0.086 0.935** 0.137 (0.432) (0.236) (0.212) (0.198) (0.954) (0.279) (0.419) (0.202) T3: grant only x FU3 x woman -0.099 -0.575** 0.783*** 0.751*** -1.311 -0.735** 0.297 0.340 (0.388) (0.225) (0.291) (0.227) (1.067) (0.293) (0.723) (0.280) Constant 1.600*** 1.038*** 2.401*** 2.104*** -2.116*** 0.254 7.676*** 4.151*** (0.430) (0.248) (0.360) (0.276) (0.665) (0.254) (0.825) (0.316) Observations 9,336 9,336 9,333 9,333 9,336 9,336 9,336 9,336 R-squared 0.051 0.190 0.072 0.065 0.132 0.279 0.134 0.225 Control group mean BL, Man 2.526 1.718 3.205 3.294 Control group mean BL, Woman 1.665 1.659 1.431 1.998 Control group mean FU1, Man 2.666 2.123 2.944 2.541 1.691 0.851 8.687 4.971 Control group mean FU1, Woman 3.206 2.529 1.172 1.258 9.611 5.980 1.171 0.945 Control group mean FU2, Man 2.066 2.346 2.043 2.039 1.624 0.947 7.935 5.420 Control group mean FU2, Woman 4.202 6.393 1.003 1.105 10.33 6.284 1.227 1.301 Control group mean FU3, Man 2.497 1.802 2.760 2.689 3.222 2.760 12.32 6.545 Control group mean FU3, Woman 1.385 1.498 1.045 1.038 13.62 7.292 4.184 3.584 FU1,T1: Man - Woman (p-value) 0.0817 5.67e-06 0.0337 0.237 0.104 0.380 0.814 0.340 FU1,T2: Man - Woman (p-value) 0.429 0.000745 0.197 0.632 0.472 0.556 0.0933 0.0689 FU1, T3: Man - Woman (p-value) 0.746 0.00461 0.0688 0.593 0.449 0.467 0.0896 0.499 FU2,T1: Man - Woman (p-value) 0.0593 0 0.649 0.367 0.832 0.547 0.345 0.0283 FU2,T2: Man - Woman (p-value) 0.00378 0 0.846 0.703 0.354 0.962 0.513 0.0185 FU2,T3: Man - Woman (p-value) 0.00538 0 0.582 0.945 0.573 0.523 0.986 0.415 FU3,T1: Man - Woman (p-value) 0.0405 0.000195 0.175 0.240 0.510 0.530 0.00657 0.330 FU3,T2: Man - Woman (p-value) 0.0357 5.09e-05 0.705 0.469 0.693 0.0107 0.611 0.617 FU3,T3: Man - Woman (p-value) 0.0400 7.18e-07 0.464 0.901 0.761 0.623 0.391 0.0112 Woman,T1: FU2 - FU1 (p-value) 0.000153 2.38e-09 0.135 0.274 0.236 0.334 0.296 0.971 Woman,T2: FU2 - FU1 (p-value) 0.00145 2.37e-07 0.853 0.353 0.472 0.503 0.527 0.509 Woman,T3: FU2 - FU1 (p-value) 0.000380 3.18e-09 0.00157 0.00501 0.370 0.631 0.747 0.153 Man,T1: FU2 - FU1 (p-value) 0.668 2.23e-09 0.356 0.643 0.938 0.494 0.0751 0.154 Man,T2: FU2 - FU1 (p-value) 0.505 1.00e-07 0.111 0.126 0.278 0.987 0.417 0.891 Man,T3: FU2 - FU1 (p-value) 0.874 8.72e-05 0.684 0.182 0.663 0.692 0.121 0.487 Woman,T1: FU3 - FU2 (p-value) 0.000827 1.30e-06 0.126 0.148 0.669 0.307 0.0366 0.765 Woman,T2: FU3 - FU2 (p-value) 0.0374 5.00e-06 0.452 0.174 0.308 0.000841 0.243 0.486 Woman,T3: FU3 - FU2 (p-value) 0.118 2.76e-06 0.295 0.377 0.150 0.0104 0.442 0.524 Man,T1: FU3 - FU2 (p-value) 0.197 1.11e-10 0.0513 0.162 0.608 0.708 0.325 0.00116 Man,T2: FU3 - FU2 (p-value) 0.00201 1.41e-09 0.329 0.0442 0.106 0.932 0.483 0.0140 Man,T3: FU3 - FU2 (p-value) 0.000441 0 0.358 0.581 0.0932 0.205 0.143 0.00188 Woman,T1: FU3 - FU1 (p-value) 0.357 0.210 0.00984 0.0177 0.153 0.0684 0.223 0.778 Woman,T2: FU3 - FU1 (p-value) 0.160 0.542 0.596 0.633 0.104 0.000212 0.500 0.830 Woman,T3: FU3 - FU1 (p-value) 0.0423 0.182 0.000475 0.000927 0.518 0.00465 0.331 0.635 Man,T1: FU3 - FU1 (p-value) 0.446 0.845 0.283 0.0692 0.571 0.863 0.0307 0.0577 Man,T2: FU3 - FU1 (p-value) 0.0271 0.217 0.599 0.524 0.370 0.943 0.197 0.0290 Man,T3: FU3 - FU1 (p-value) 0.000686 0.000213 0.237 0.0645 0.0447 0.124 0.945 0.0114 FU3,Woman: T1 - T2 (p-value) 0.0288 0.184 0.0976 0.277 0.161 0.174 0.828 0.402 FU3,Woman: T1 - T3 (p-value) 0.187 0.916 0.616 0.694 0.0805 0.0879 0.359 0.980 FU3,Woman: T2 - T3 (p-value) 0.419 0.151 0.238 0.480 0.725 0.742 0.260 0.410 55 FU3,Man: T1 - T2 (p-value) 0.0841 0.0784 0.0497 0.239 0.709 0.363 0.0266 0.918 FU3,Man: T1 - T3 (p-value) 0.284 0.178 0.381 0.127 0.0773 0.0792 0.00476 0.110 FU3,Man: T2 - T3 (p-value) 0.496 0.665 0.297 0.725 0.150 0.00698 0.517 0.117 Notes: The specification controls for dummies for each follow-up surveys, baseline outcomes, priority sectors, and district dummies.* Baseline outcomes are not controlled in columns (5) to (9). Information to compute those outcomes was not collected at baseline.Standard errors in parentheses are clustered at the individual level *** p<0.01, ** p<0.05, * p<0.1 56 Table A10: Inverse Probability Weighing, first stage probit (1) Observed at VARIABLES FU3 Woman -0.062 (0.063) Age 0.008 (0.008) Priority sector 0.175*** (0.057) Unemployment, vulnerable employment OR underemployment -0.050 (0.061) Total earnings over the past month winsorized 1% -0.000 (0.001) Revenues from own business in the past month - winsorized 1% 0.000 (0.000) Profits from own business in the past month - winsorized 1% -0.001 (0.002) Household asset score (0-17) 0.022** (0.009) Agency score 0.073 (0.051) Own business 0.058 (0.066) Marital status (reference=Simgle Married -0.025 (0.082) Cohabiting couple 0.180* (0.105) Divorced -0.251 (0.239) Widow/widower -0.563 (0.441) Number of children -0.051 (0.032) Number of dependent children 0.046** (0.022) Household size 0.010 (0.011) Speak French (reference= No) Yes, some sentences 0.148* (0.081) Yes, intermediate level -0.042 (0.114) Yes, fluent -0.093 (0.148) Highest level of education 0.018* (0.011) Currently enrolled in school 0.197 (0.236) Training in business 0.266** (0.133) Received assistance in last 12 months (reference=no) From the government -0.402 (0.316) From an NGO, foundation -0.027 (0.310) From others -0.333 (0.438) Missed meals in last 7 days 0.028 57 (0.059) Constant 0.249 (0.321) Observations 3,403 The model controls for ethnic groups, religion groups and district dummies. Standard errors in parentheses are clustered at the individual level *** p<0.01, ** p<0.05, * p<0.1 58 Table A11: Reweighted impact estimates at follow-up 3 (restricted to common support) (1) (2) (3) (4) (5) Revenues Total from own Z-score Profits from earnings business in profits own business in over the Insufficient the past and the past month - past month employment month - revenues winsorized 1% winsorized winsorized 1% 1% woman -0.064 -17.313** -10.275*** -22.382*** 0.046 (0.063) (7.550) (3.059) (4.163) (0.036) T1: training+grant x FU3 x man 0.119 18.023 -1.235 10.997* 0.015 (0.082) (11.780) (4.184) (5.735) (0.042) T2: training only x FU3 x man 0.237*** 43.507*** 9.798** 11.350* 0.048 (0.091) (14.224) (4.984) (6.266) (0.041) T3: grant only x FU3 x man 0.008 21.170* -5.259 -3.672 -0.016 (0.082) (12.002) (4.332) (5.690) (0.041) T1: training+grant x FU3 x woman 0.043 8.293 1.122 6.033** -0.021 (0.048) (6.649) (2.241) (2.920) (0.033) T2: training only x FU3 x woman 0.199*** 21.306*** 9.326*** 11.665*** 0.023 (0.047) (7.527) (2.428) (2.800) (0.032) T3: grant only x FU3 x woman -0.016 8.593 1.102 3.112 -0.043 (0.047) (6.608) (2.311) (2.758) (0.032) Constant -0.067 41.081*** 5.384 29.072*** 0.571*** (5.806) (0.086) (12.827) (4.344) (0.052) 2,372 Observations 2,376 2,354 2,354 0.098 2,737 R-squared 0.082 0.136 0.106 34.50 0.036 Control group mean BL, Man 0.0980 52.47 15.33 21.63 0.594 Control group mean BL, Woman -0.0810 37.92 8.977 41.64 0.726 Control group mean FU3, Man 0.0710 70.60 21.79 15.24 0.603 Control group mean FU3, Woman -0.0450 44.70 7.986 0.657 Notes: Program impacts are only estimated for the 3rd follow-up survey. The specification controls for dummies for baseline outcomes, priority sectors, district dummies, and indicators of having a business for columns 1, 2, 3. Standard errors in parentheses are clustered at the individual level. *** p<0.01, ** p<0.05, * p<0.1. Insufficient employment: indicator variable equal to 1 if respondent is unemployment, underemployed (working less than 40 hours a week), or in vulnerable employment (unpaid apprentice or unpaid contributor to the family business). 59 Table A12: Estimation of Lee bounds for each treatment arm (1) (2) (3) (4) (5) Revenues Profits from from own own business Total earnings Z-score business in in the past over the past Insufficient profits and the past month - month employment revenues month - winsorized winsorize 1% winsorized 1% VARIABLES 1% T1: training+grant x FU3 x man Standard specification with controls 0.139* 20.720* -0.677 11.740** 0.025 (0.080) (11.315) (4.031) (5.572) (0.041) Specification without controls 0.094 21.000* 1.536 11.189* 0.017 (0.081) (12.143) (4.312) (5.761) (0.041) Lee bounds - lower 0.060 20.006 0.051 9.883 0.008 (0.093) (12.434) (4.673) (6.112) (0.045) Lee bounds - upper 0.139 30.629** 4.724 13.806** 0.023 (0.099) (13.357) (5.207) (6.943) (0.043) T2: training only x FU3 x man Standard specification with controls 0.255*** 45.520*** 9.587** 12.183** 0.055 (0.089) (13.662) (4.877) (6.093) (0.040) Specification without controls 0.171** 34.705** 7.999* 9.773 0.050 (0.087) (13.573) (4.785) (6.084) (0.041) Lee bounds - lower 0.163* 34.312** 7.199 9.303 0.046 (0.096) (13.870) (4.976) (6.447) (0.043) Lee bounds - upper 0.192* 39.002*** 9.378* 10.689 0.053 (0.099) (14.836) (5.361) (7.155) (0.044) T3: grant only x FU3 x man Standard specification with controls 0.006 18.990 -5.947 -3.631 -0.011 (0.080) (11.579) (4.118) (5.508) (0.041) Specification without controls -0.052 12.840 -6.293 -6.056 -0.013 (0.079) (11.945) (4.146) (5.466) (0.041) Lee bounds - lower -0.111 10.445 -9.145** -9.271 -0.033 (0.089) (11.960) (4.412) (5.874) (0.044) Lee bounds - upper 0.040 29.858** -0.108 1.284 0.001 (0.090) (13.633) (4.923) (6.768) (0.042) T1: training+grant x FU3 x woman Standard specification with controls 0.032 7.157 0.879 5.755** -0.028 (0.046) (6.466) (2.174) (2.814) (0.032) Specification without controls 0.043 13.717** 2.989 6.961*** -0.027 (0.043) (6.053) (2.011) (2.564) (0.033) Lee bounds - lower -0.032 11.230* -0.180 3.615 -0.065* (0.051) (6.326) (2.078) (2.755) (0.038) Lee bounds - upper 0.110** 27.944*** 6.548*** 11.379*** -0.008 (0.045) (6.510) (2.158) (2.728) (0.036) T2: training only x FU3 x woman Standard specification with controls 0.192*** 21.942*** 9.420*** 11.661*** 0.011 (0.047) (7.411) (2.403) (2.767) (0.032) Specification without controls 0.173*** 23.156*** 8.937*** 11.118*** 0.015 (0.043) (7.023) (2.156) (2.492) (0.032) Lee bounds - lower 0.153*** 22.621*** 7.530*** 9.881*** 0.005 (0.048) (6.790) (2.203) (2.627) (0.035) Lee bounds - upper 0.197*** 29.115*** 10.507*** 12.975*** 0.020 (0.048) (7.256) (2.261) (2.608) (0.033) 60 T3: grant only x FU3 x woman Standard specification with controls -0.016 8.037 1.115 3.166 -0.046 (0.046) (6.162) (2.175) (2.599) (0.032) Specification without controls -0.033 8.947 1.020 1.824 -0.039 (0.041) (5.798) (1.959) (2.373) (0.032) Lee bounds - lower -0.069 8.047 -1.021 -0.528 -0.053 (0.049) (5.864) (2.061) (2.539) (0.036) Lee bounds - upper 0.002 17.417*** 3.094 4.849* -0.032 (0.044) (6.721) (2.066) (2.588) (0.034) Constant -0.086 41.770*** 5.557 27.983*** 0.557*** (0.084) (12.269) (4.138) (5.495) (0.051) Observations 2,450 2,428 2,427 2,448 2,805 R-squared 0.082 0.133 0.105 0.101 0.036 Control group mean BL, Man 0.0980 52.47 15.33 0.0340 0.594 Control group mean BL, Woman -0.0810 37.92 8.977 0.0220 0.726 Control group mean FU3, Man 0.0710 70.60 21.79 0.0420 0.603 Control group mean FU3, Woman -0.0450 44.70 7.986 0.0150 0.657 Notes: Program impacts are only estimated for the 3rd follow-up survey. The standard specification controls for gender, dummies for baseline outcomes, priority sectors, district dummies, and indicators of having a business for columns 1, 2, 3. Standard errors in parentheses are clustered at the individual level. For the estimation of the Lee bounds, standard errors in parentheses are bootstrapped with 1000 replications. *** p<0.01, ** p<0.05, * p<0.1. Insufficient employment: indicator variable equal to 1 if respondent is unemployment, underemployed (working less than 40 hours a week), or in vulnerable employment (unpaid apprentice or unpaid contributor to the family business). 61