WPS7993 Policy Research Working Paper 7993 Growing Markets through Business Training for Female Entrepreneurs A Market-Level Randomized Experiment in Kenya David McKenzie Susana Puerto Development Research Group Finance and Private Sector Development Team March 2017 Policy Research Working Paper 7993 Abstract A common concern with efforts to directly help some small There is no evidence of negative spillovers on the competing businesses to grow is that their growth comes at the expense businesses, and the markets as a whole appear to have grown of their unassisted competitors. This study tests this possi- in terms of number of customers and sales volumes. This bility using a two-stage randomized experiment in Kenya. market growth appears to come from enhanced customer The experiment randomizes business training at the market service and new product introduction, generating more level, and then within markets to selected businesses. Three customers and more sales from existing customers. As a years after training, the treated businesses are selling more, result, business growth in underdeveloped markets is pos- earn higher profits, and their owners have higher well-being. sible without taking sales away from nontreated businesses. This paper is a product of the Finance and Private Sector Development Team, Development Research Group. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The authors may be contacted at dmckenzie@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Growing Markets through Business Training for Female Entrepreneurs: A Market-Level Randomized Experiment in Kenya* David McKenzie, Development Research Group, World Bank Susana Puerto, Employment Policy Department, International Labour Organization Keywords: Business Training, Spillovers, Microenterprise, Market Development JEL codes: O12, O17, J16, L26. * We gratefully acknowledge funding provided to this project from the International Initiative for Impact Evaluation (3ie), Private Enterprise Development in Low-Income Countries (PEDL), the International Labour Organization (ILO), the Strategic Research Program (SRP), and the World Bank Jobs Umbrella multi-donor trust fund. Human Subjects Approval was obtained from Innovations for Poverty Action (IPA, 13February-002) and the Maseno University Ethics Review Committee (MSU/DRPC/MUERC/000006/13). Authority to conduct research in Kenya was provided by the Kenyan Ministry of Science and Technology (NCST/RCD/14/013/553B). A pre- analysis plan and the associated trial were registered on the AEA Social Science Registry on February 21, 2014 (AEARCTR-0000287). We thank Valerie Breda, Faizan Diwan, Bilal Hakeem, Mary Kamore, Virginia Losada, Jane Maigua, Grace Makana, Silvia Paruzzolo, and Miriam Wekesa for their contributions to the study design and implementation, ICRW for their qualitative work, and participants at various seminars for useful comments.       1. Introduction Governments and NGOs around the world promote and offer a variety of direct support to small businesses, with the provision of business training one of the most common services provided. Better business practices are strongly associated with better firm performance across a range of countries in both the cross-section and over time (McKenzie and Woodruff, 2015), and so the hope is that teaching these practices will help firms grow. However, a key concern around public or subsidized provision of business training is whether any growth of trained firms comes at the expense of their competitors (Rotemberg, 2014). This concern is particularly apparent when working with microenterprises in rural markets in developing countries, where it is easy to believe that if firms are all selling similar products in a small market, any extra sales made by trained firms must come from competing away these sales from neighboring untrained firms. Conversely, if better management practices are best thought of as a technology (as in Bloom et al., 2015), then there might be positive spillover benefits as untrained firms copy the practices adopted by the trained firms and also become more productive. Furthermore, if training leads firms to expand the variety of products being offered and the customer service provided, the overall market size may increase, particularly in underdeveloped marketplaces. We conduct a randomized experiment in 157 rural markets in Kenya to test how business training (the International Labour Organization (ILO)’s Gender and Enterprise Together program) affects the profitability, growth and survival of female-owned businesses, and to evaluate whether any gains in profitability come at the expense of other business owners. We work with a large sample of 3,537 firms, and use a two-stage randomization, first randomizing at the market level, and then randomizing the offer of training to individuals within treated markets. A year and a half after the training has taken place, half of the sample assigned to training was then offered a subsequent mentoring intervention intended to test whether additional group-based and in-person support strengthens the impacts of training. Four rounds of follow-up surveys with low attrition are used to measure impacts at one and three years after training. Firms assigned to training are 3 percentage points more likely to survive after three years, earn 18 percent higher sales, and make 15 percent higher profits. Their owners have better mental health and a higher subjective standard of living. These gains are greater at three years than at one year 2    after the training, and are similar for firms assigned to training only as for firms also assigned to a mentor. These gains come with no significant spillover effects on untreated firms operating in the same markets, and total sales and the total number of customers are higher in the treated markets than control markets. We also find no reduction in new entry into these markets after training. This market growth appears to stem from better customer services, better business practices, and the introduction of new products, with no significant impacts on access to finance or input management. We conclude that, in underdeveloped markets, microenterprise growth need not come at the expense of competitors, and business training can help the overall market grow. There are three strands of this literature of particular relevance to this study. The first is evidence on the effectiveness of business training. McKenzie and Woodruff (2014) review this literature and note that overall evidence on effectiveness is mixed, in part because many studies have low statistical power and measure impacts over short durations. This existing literature also suggests that business training may be less effective for female business owners, either because they work in sectors with very low efficient scales or because they face many other constraints that limit the ability of their businesses to grow (e.g. de Mel et al., 2014; Berge et al., 2015; Giné and Mansuri, 2016). Get-Ahead, the training program studied here, was designed especially for women with low education levels. Bulte et al. (2016) evaluate the impact of this same program on female microfinance clients in Vietnam and find some evidence of firm growth 12 months after training. Our results show stronger and longer-term evidence that this training can help female-owned businesses grow. Second, several studies examine whether augmenting standard business training with mentoring can enhance its effectiveness. Valdivia (2015) and Giné and Mansuri (2016) find that individualized hand-holding or mentoring after training does not have sustained impacts relative to training alone. Brooks et al. (2016) find, also in Kenya, that assigning a mentor to a young firm does increase profits in the short-run, but the effect fades over time. Our results are consistent with mentoring not delivering additional gains compared to training alone. Third, almost all of the existing literature to date has been unable to measure impacts at the market level, so cannot tell if any gains come from untrained firms. The one exception is Calderón et al. (2013), who worked with 17 villages in rural Mexico, assigning 7 to treatment and 10 to control. They find no significant spillovers, although this may in part reflect low statistical power given the small number of villages and that they lose 18 percent of their sample to attrition and 41 percent 3    to closure by their second follow-up. Our paper builds on this work and not only shows no significant spillovers, but significant growth at the market level. The remainder of the paper is structured as follows: Section 2 discusses the selection of our sample and randomization procedure; Section 3 the training and mentoring interventions; Section 4 our data collection and estimation approach; Section 5 impacts on primary and secondary outcomes; Section 6 mechanisms for these results; Section 7 cost-effectiveness; and Section 8 concludes. 2. Selection of the Sample, Randomization Procedure, and Baseline Characteristics 2.1 Selecting a Sample A participatory process involving the ILO, government ministries, and organizations serving female entrepreneurs1 was used to select four counties for the study: Kakamega and Kisii in the Western region, and Embu and Kitui in the Eastern region. These regions are largely rural, with an average population for each county of approximately 1 million, and the majority of the population below the poverty line. In each of the four counties, field staff from Innovations for Poverty Action, Kenya, mapped out all market centers deemed as “medium” or “large” outside of the main cities and conducted a market census of all female-owned businesses between June and November 2013 (see timeline in Appendix 1). After listing, an eligibility filter was applied (Appendix 2) and a baseline survey administered to 3,537 individuals located in 157 separate markets. These markets are typically small and remote, largely consisting of women2 operating a limited variety of businesses such as selling fruits, vegetables, grains, and dried fish products from tables; and offering services like hairdressing, dressmaking and small food kiosks. The market is a designated place in the village, and authorities typically build several stalls for the sellers to display                                                              1 Stakeholders consulted included the Department of Micro and Small Enterprise Development (DMSED) of the Ministry of Labour, Ministry of Youth Affairs, Ministry of Cooperative Development and Marketing, Ministry of Youth, Federation of women entrepreneurs associations (FEWA), Women Enterprise Fund (WEF), Youth Employment Development Fund (YEDF), Business Development Service providers, and Inoorero University. 2 While men operate some businesses in these markets, they tend to specialize in other products, and women are the dominant sellers of the products in our sample.  4    their goods and services, and charge them a market fee to sell in the market. The average market in our study has 22 firms surveyed in it, with 75 percent in retail trade and 25 percent in services.3 2.2 Randomization Procedure Individuals were then assigned to treatment and control for the business training intervention in a two-stage process, diagrammed in Figure 1. First, Markets were assigned to treatment (have some individuals in them invited to training) or control (no one in the market would be invited to training) status. Randomization was done within 35 strata defined by geographical region (within county) and the number of women surveyed in the market. The need to ensure sufficient numbers for training meant that more than half the markets were allocated to treatment, with 93 markets assigned to treatment and 64 to control. Then within each market, Individuals were assigned to treatment (be invited to training) or control (not be invited to training) within treated markets by forming four strata, based on quartiles of weekly profits from the census (<=450, 451-800, 801-1500, 1501-4000), and then assigning half the individuals within each stratum to training. This resulted in 1,172 of the 2,160 individuals in treated markets being assigned to treatment, and 988 to control.4 For the follow-up mentoring intervention, we dropped markets with 15 or fewer firms in the study, since these would have too few firms to make mentoring cost-effective. This left 73 of the 93 treated markets. We randomly assigned 37 of these (within the original randomization strata) to mentoring, and the remaining 36 to be training only. Then within the mentoring markets, all those individuals assigned to training were assigned to be offered mentoring services, although in practice only those businesses still surviving were actually offered these services. We then compare the mentoring and training-only markets to the 44 markets in the control group with 16 or more firms. 2.3 Characteristics of the Sample and Verification of Random Assignment Table 1 provides some key characteristics of the women selected in our sample and their firms by treatment assignment (Appendix Table 1 provides a similar table by assignment to mentoring).                                                              3 In addition, markets typically have a market day once a week, where outside vendors come to sell goods such as clothing, plastic housewares, and shoes. These vendors who are only in the market one day a week are not included in our study. 4 When the number of firms within a stratum was odd, the additional unit was also randomly allocated to training. 5    The sample looks similar across the three groups and we cannot reject the joint orthogonality of baseline characteristics when comparing the treatment group to the spillover group, the treatment group to the pure control, or the spillover group to the pure control. The average woman in our sample is 36 years old, has 9 years of schooling, and has been running her firm for just over 6 years. Two-thirds of the women are currently married. The modal firm has no employees (only 20 percent have one or more employees). The mean firm earns 1,100 KSH (US$13) per week in profits on sales of 5,500 KSH (US$65), and has capital stock of 31,000 KSH (US$370).5 One- quarter have ever received financing from a bank or microfinance organization and 45 percent are registered at the local level. Only 35 percent of firms keep business records at baseline, and on average firms are using just over half of the 26 business practices in the McKenzie and Woodruff (2015) index. This suggests scope for improvement from business training. 3. The Training and Mentoring Interventions 3.1 The Get-Ahead Program The training provided is the ILO’s Gender and Entrepreneurship Together – Get-Ahead for Women in Enterprise program. This program “differs from conventional business training materials by highlighting essential entrepreneurial skills from a gender perspective... It addresses the practical and strategic needs of low-income women in enterprise by strengthening their basic business and people management skills. It shows women how to develop their personal entrepreneurial traits and obtain support through groups, networks and institutions dealing with enterprise development” (Bauer et al., 2004). The program began in Thailand in 2001, and has now been used in at least 21 countries, serving approximately 400,000 women. The program has four key modules, with the following themes:  Module 1: Basics on Gender and Entrepreneurship. The module introduces Get-Ahead and delivers basic concepts on the promotion of gender equality between men and women and the life cycle of people and enterprises.  Module 2: The Business Woman and Her Environment. The module focuses on raising awareness among women about their strengths and weaknesses as well as current or future working environment and its effect on the business.                                                              5 The exchange rate was approximately 1 US$=84 KSH in 2013 at the time of baseline. 6     Module 3: The Business Project. The module offers trainees information on the development of business ideas, opportunities and challenges, and the basics around marketing, production, services and technology, and financing, including costing and book keeping.  Module 4: People, Organization and Management. It is a soft skills module around managing a business (or a family business), with information on management of self and others, business support and networking, and action planning. Some of the topics covered throughout these modules are not often emphasized in general business training programs, for example the conversations about the difference between sex and gender, and the role of cultural constraints in shaping women in business; dividing household and business tasks; and how to network with other women and the role of women’s associations. In addition, it covers a number of topics more typical of standard programs such as recordkeeping and bookkeeping; separating business and household finances; marketing; financial concepts; costing and pricing; generating and fine-tuning new business ideas; setting smart objectives; and traits needed for business success. The program seeks to create a “business mind” among low-income women engaged in small-scale businesses. The training methodology is participatory, with practical exercises to teach concepts. Modules expose women to different situations and environments simulating real business life. For example, women learn about the different types of costs involved in production, and how to account for their own costs through making lemonade. They have role play exercises to practice different sales strategies for customers, and make necklaces to discuss a production process and the importance of different factors in product design. The course is a five-day course, as shown in Appendix 3. All trainers had at least five years of experience in training small firms, and had tertiary qualifications. Training took place in two to three locations per county. The locations were chosen to be relatively central to clusters of marketplaces, and were typically held in local hotels or church buildings. Training was offered for free, and participants were provided transport subsidies of approximately US$6 per day to cover the costs of traveling from their residences to these locations (an average of 14 kilometers). The cost of providing the training is estimated at between US$222 and US$333 per woman trained.6                                                              6  The smaller number reflects workshops with 20 attendees, while the larger is for workshops with 30 attendees. 7    3.2 Training Take-up Training took place immediately after the baseline surveys in each county, between June and November 2013. Of the 1,172 individuals assigned to training, 77.7 percent attended at least one day of training. Of the individuals who attended at least one day, 94.6% attended all five days. In Diwan et al. (2014) we report on a choice structure experiment intended to increase training attendance, and discuss the correlates of attendance. Age and marital status are strong and statistically significant predictors of attendance: all else equal, women aged above 35 are 35 percentage points more likely to attend training than those below 35, while married women are 24 percentage points less likely to attend than unmarried women. This potentially reflects the competing demands on their time from other household tasks. Women are also more likely to attend if they have previously participated in training (perhaps reflecting greater perceived benefits from attending), have a large household (potentially providing more people to undertake household and business tasks in their absence), and are located closer to the training venue (reducing travel time). Women who earn more profits are less likely to attend, perhaps reflecting a higher opportunity cost of time, or that they think there is less need to improve. 3.3 Mentoring Intervention After seeing the one-year follow-up results and emerging evidence suggesting that many business training programs may be too short to show sizeable impacts (McKenzie and Woodruff, 2014, 2015), we decided to add a mentoring component for half the sample assigned to training. Mentors were female business owners of a similar average age to our study sample, with 75 percent having had post-secondary education, and 68 percent having studied business or accounting, and a median of 5.5 years’ business experience. They were recruited and trained by the implementing partners Women Enterprise Fund (WEF) and Kenya Industrial Estates (KIE). Each mentor was assigned a group of three to six mentees. The mentor program then involved the mentor meeting in a group with the mentees for 10 sessions, comprising of one session every two weeks for five months, following a standardized structure (Appendix 5). In addition, the mentor would meet with the individual mentees once a month over this period to provide individualized guidance. The mentoring process aimed to reinforce the business training by having the business owner identify core goals for their business, consider where there are gaps, explore options to fill these gaps, and then take action towards meeting these goals. 8    There were 524 firms assigned to training in the mentoring markets. In practice, only the 446 women whose businesses were surviving at the time of the first follow-up were contacted about mentoring. Out of these, 392 signed up for the program, and 388 attended at least one session. Conditional on attending, the median went to 77 percent of all sessions. Take-up is thus 388/524 (74 percent) of those assigned, and 388/446 (87 percent) of those in business at the time of the intervention. The cost of providing the mentoring is estimated at US$553 per woman trained.7 3.4 How might training and mentoring affect treated and spillover firms? The aim of these programs is to help those participating to grow their businesses. Consider a standard firm production decision, in which a firm owner with entrepreneurial ability θ, and production function f(.), chooses inputs of labor L and capital K to maximize profits: , , s.t. (1) Where p is the market price of their product, w and r are the prevailing market input prices for labor and capital, and τ reflects how binding credit constraints on capital acquisition are, given wealth level A. Business training and mentoring can then potentially increase profit levels through several channels. Increasing entrepreneurial ability may allow the business owner to produce more with the same inputs, giving the business more to sell. Better record-keeping and financial accounts may make it easier for the owner to obtain financing, alleviating credit constraints and again allowing the business to produce more. If markets are perfectly competitive, then firm owners will be able to sell all they produce at the market price p, and there will be no spillover impact on other firms in the market. However, if firms are competing with one another for customers, then the price p need not remain constant as firms produce more. Better marketing practices, better customer service, and more efficient cost control may enable trained firms to compete away customers from other firms in the market. As a consequence, any gains to treated firms could come at the expense of lower sales and lower profits for non-treated firms in these markets. Conversely, one of the hopes of the training providers is the possibility of positive spillovers to other firms in the market. A first potential channel is that the trained business owners might teach other in the market the skills learned, or that others in the market may be able to observe and adopt the new practices irrespective of whether or not they are directly taught. Second, the Get Ahead                                                              7 The estimate is based on the number of women that attended at least one session. 9    training emphasizes the role of networks and women working together to seek financing or purchase products together to obtain bulk discounts, which could result in lower costs and higher profits for both treated and untreated. Implicit in this discussion is the assumption that the production function f(.) is itself unchanged. But business training may encourage firm owners to consider producing new products or offering new services. This offers another means for treated firms to grow without negatively impacting on untreated firms – they can expand the market through increasing variety. 4. Follow-up Surveys and Estimation Approach 4.1 Follow-up Surveys Four rounds of follow-up surveys were conducted, in order to measure outcomes approximately one year and three years after training occurred (see timeline Appendix 1). Two types of surveys were used. A comprehensive long-form survey collecting data on a wide range of business outcomes was used in rounds 2 and 4. These were supplemented by much shorter surveys in rounds 3 and 5. These short surveys were conducted two or three months after the long surveys, and were intended to provide a second observation on volatile business outcomes like sales and profits, as well as an additional opportunity to gather data from individuals who could not be found at the time of the long survey rounds. Appendix 6 describes how the key outcomes are measured. Appendix Table 2 details response rates. Overall, we were able to interview 95.0 percent of the sample in at least one of round 2 or 3, and 92.3 percent in at least one of round 4 or 5. In addition, in cases where we were unable to interview someone due to refusal, travel, death, or other reasons, we collected information from other household members or close contacts on whether the individual in our sample was currently operating a business. This enables us to have data on survival status for 99.3 percent of the sample at one year, and 97.2 percent at three years. There is no significant difference in data availability with treatment status at the three-year horizon, although those assigned to treatment are 1 to 2 percentage points more likely to have data available at the one year horizon. In addition to the survey data, we have two other sources of information that aid in assessing impact. The first are photos of the inventories of the businesses, which were taken at the time of the baseline survey and in rounds 2 and 4. We had two independent field staff value these inventories based on the market prices of the different items, and average these values to get a 10    photo-based measure of the size of the firm. They also visually compared the firms in photos to determine whether the business looked bigger in relative terms to what it was at baseline. Secondly, intensive qualitative work for impacts over the first year of the study was carried out by ICRW (2015). 4.2 Estimation Approach A pre-analysis plan and the associated trial were registered on the AEA Social Science Registry on February 21, 2014. The assigned registry number is AEARCTR-0000287.8 This plan pre- specified the primary and secondary outcomes of the study, the estimation approach, and the causal chain to be investigated prior to the collection of any follow-up data. Subsequent funding enabled us to conduct the short follow-up surveys to collect additional data on profits and sales immediately following the long-form surveys, and to add the mentoring intervention. Following McKenzie (2012) we pool these short and long-run follow-up data to get average effects at one and three years, and use an ANCOVA specification where the baseline data are available in order to maximize power. We then examine outcomes at the individual firm level by estimating for firm i in market j at time t=2,..,5: , ,, , ∗1 , ∗3 , ∗1 , ∗3 , , ,, ,, ∑ 1 , , (2) Where , , is the given outcome variable measured in round t , , , is its baseline value and , , a dummy variable indicating whether or not this baseline value is missing, , is an indicator for being in a treatment market and being assigned to treatment, , is the spillover term, measuring whether firm i is a control firm in a market assigned to treatment; 1 is a dummy taking the value one in follow-up rounds 2 and 3; 3 is a dummy taking the value one in follow-up rounds 4 and 5; we follow Bruhn and McKenzie (2009) in using as a vector of randomization strata dummy variables (geographic region*market size*profit range), are survey round dummies, and , , is the error term, which we cluster at the market level to account for the market level random assignment. and provide the intent-to-treat effects at one- and three-year horizons, which is the effect of being assigned to treatment relative to being a firm in                                                              8 http://www.socialscienceregistry.org/trials/287 11    the control markets. Note that the three-year impact represents the impact of having been assigned to receive business training three years earlier, followed by having a 50 percent chance of being assigned to receive mentoring approximately 9 months earlier. We test = to determine whether the impacts differ over time, and = =0 to determine whether we can reject that there is no treatment effect in any period. One of the main questions of interest in this study is also whether there are spillovers from the treated individuals to other individuals within the same villages. These spillovers could be positive (e.g. treated women share knowledge with control women in the same markets) or negative (e.g. treated women compete away the sales of control women from the same markets). and measure these spillover effects at the one and three year horizons by comparing control firms in treated markets to control firms in control markets. We then test whether being assigned to the mentoring treatment had a differential impact from training alone via the following regression, estimated using only rounds 4 and 5: , ,, , , , , , , ,, ,, 1 4 , , (3) Where Mentor indicates being in a market assigned to mentorship and being assigned a mentor, Trainonly indicates being in a market assigned to training only and being assigned training, and Spillovermentor and SpilloverTrainonly are the spillover groups in the mentor and training only markets. We estimate this equation only for markets eligible for the mentoring intervention (those with more than 15 firms in them at baseline). The standard errors are again clustered at the market level. We then test in order to determine whether mentoring has a different effect from training alone. We focus on the intention-to-treat effects in our main analysis. In Appendix 8 we also report the local average treatment effects (LATE) of receiving training on our primary outcomes, by instrumenting training attendance with training assignment in equation (2). None of the control group attended training, so the LATE is the same as the average treatment effect on the treated. The LATE impacts are approximately 24 percent higher than the ITTs. 5. Impacts on Primary and Secondary Outcomes 12    We begin by examining impacts on our primary and secondary outcomes at the individual level, then examine impacts on the markets as a whole. 5.1 Impacts on Primary Outcomes at the Individual Level Our primary outcomes are whether the business training succeeds in helping the firm survive, and increases firm sales, profits, and size (as measured by the photo value of inventories). We consider several measures of these variables, as specified in the pre-analysis plan, along with an aggregate index, which is the average of standardized z-scores of these outcomes and provides an overall measure of whether the intervention has succeeded in increasing firm performance as well as a way for controlling for multiple hypothesis testing. Table 2 provides the results. Panel A shows the one and three year impacts of assignment to training, while panel B separates the three-year results by whether the firm is in a market also assigned to mentoring or to training only. Over the one-year horizon the direct impacts are all positive, but only the impact on daily sales is statistically significant at the 5 percent level, and there is a 0.05 standard deviation increase in the aggregate index, significant at the 10 percent level. These estimates are all larger in magnitude by year three, with statistically significant impacts on survival (3 percentage points), daily sales (171 KSH, or 14.6% of the control mean), weekly sales (1,038 KSH, or 18.0% of the control mean), weekly sales of the main product the business sells (454 KSH, or 13.4% of the control mean), and weekly profits (221 KSH, or 15.4% of the control mean). The overall aggregate index shows a 0.088 standard deviation increase, which is significant at the 1 percent level. The year three results are statistically different from the year one results for the key outcomes of survival, weekly sales, and weekly profits, and we can reject the null hypothesis of no treatment effect in either period. The impact on our objective and independent measure of business size: the value of inventories as photographed in the firm is positive (818 KSH, or 9.5% of the control mean), but not statistically significant. Appendix 9 considers a second measure from the photographs, which comes from comparing the round 4 business to the baseline business and classifying it as bigger, smaller, or the same size. We do find treated businesses are 4.2 percentage points more likely to be classified as bigger, which is significant at the 5 percent level. Appendix 9 also considers two additional sets of robustness checks. Our main analysis codes profits and sales as zero for firms which have closed down (since they are not earning profits or sales). Appendix 9 shows that firms assigned to training have significantly higher weekly profits, weekly sales, and a higher aggregate index even when 13    we condition on survival. Appendix 9 also shows the results remain significant after using an alternative definition of firm survival, and when using alternative transforms of profits and sales such as the inverse hyperbolic sine and logarithmic transformations. Finally, it also addresses the potential concern that business training affects how profits and sales are reporting by showing no treatment effect on the number of reporting errors, nor on the difference between the reported inventories and the size based on valuing photographs of inventories. Figures 2 and 3 show the quantile treatment effects on profits and sales for the three-year time horizon. The quantile effects increase with quantile, showing that the impacts were larger at the top of the distribution than the bottom, but are statistically significant from the 20th percentile onwards until at least the 80th percentile. Consider next the evidence for spillover effects. Over a one-year horizon we see negative spillovers on weekly profits and weekly sales, with this effect significant at the 10 percent level for sales. However, we also see positive and insignificant spillover estimates for several of our other profits and sales measures and, as a result, the impact on the aggregate index is small, with the point estimate being a 0.01 standard deviation reduction, which is not statistically significant. Over the three-year horizon, none of the outcomes shows a significant spillover impact, and the aggregate index has a 0.002 standard deviation increase, which is small and not statistically significant. Likewise, the quantile treatment effects in Figures 2 and 3 show the spillover impacts to be much smaller than the direct effects, and not statistically significant. As a result, we cannot reject that there are no spillover effects (either positive or negative) on untreated firms operating in the same markets as the treated. This picture of either small or no spillovers in sales and profits is consistent with the qualitative work, with some respondents saying they saw no change in the market competition as a result of some women being trained, while others noting a mix of positive (passing on knowledge to other firms) and negative (treated women providing better customer service to attract customers) spillovers. Panel B examines whether the impacts differ for those assigned to mentoring versus training alone. The impacts are similar in magnitude for most outcomes, with the one exception being survival, where training alone appears to have a larger impact than training followed by mentoring. When we consider our overall index, which accounts for multiple testing, we find a 0.098 standard 14    deviation increase for mentoring and 0.104 standard deviation increase for training alone. These impacts are very similar in magnitude, and we cannot reject equality (p=0.901). 5.2 Impacts on Secondary Outcomes at the Individual Level To examine the extent to which these improvements in profits and sales translate into broader individual wellbeing, our pre-analysis plan specified four domains of secondary outcomes to examine. The impacts on these outcomes are shown in Table 3. The first set of secondary outcomes concerns employment, whether in self-employment or wage work. Column 1 of Table 3 shows that 88.5 percent of the pure control group are engaged in some form of employment for pay at the time of the three-year follow-up surveys, and the increase in firm survival also translates into an increase in any employment. Column 2 looks at all income from work, which combines profits from self-employment with any earnings from wage labor. We see the higher profits are not being offset by lower wage earnings, as total income from work increases by 389 KSH per week, or 18.1% of the control mean. The training emphasized a number of topics and approaches that had the goal of empowering women in terms of decision-making around finances and business. We measure 10 different outcomes in this domain (e.g. are they compelled to spend money on their husband or family, do they need someone’s permission to travel to sell a business asset, do they have money they have sole control over, etc.). The average individual in the control group is able to do 7 out of 10 of these decisions, and Column 3 shows that training is not found to have any sizeable or significant impact on this measure of empowerment at either time horizon. This is consistent with the qualitative assessment, which noted that training did not appear to change individual or household decision-making dynamics (ICRW, 2015). The third domain we examine is subjective well-being and mental health. We measure subjective well-being today and anticipated subjective standard of living in 5 years’ time on a Cantril ladder, and mental health using the MHI-5 index of Veit and Ware (scored so that higher scores indicate better mental health). Respondents show a great deal of optimism about the future, seeing themselves as being on step 5.0 out of 10 on the life ladder currently, but expecting to be on step 8.0 in five years’ time. Training increases both current and future subjective well-being by 0.2 to 15    0.3 steps. The impact on mental health is positive and also statistically significant in both time periods. Finally, we examine the impact on ownership of household durable assets. There is a positive, but marginally significant impact in year one, and positive but insignificant impact in year three. The only evidence for a significant difference between the mentoring and training alone groups comes for the any employment outcome, which is consistent with the differential survival effect. Taken together, these results show the higher profits and sales do appear to have translated into higher overall well-being for the women given training. 5.3 Market-Level Impacts Our results show that firms assigned to treatment have grown, with no negative spillover for the untrained in the same market. This suggests that the overall markets have grown. To formally test this hypothesis, we aggregate outcomes to the market-level, and estimate the following equation at the level of the 157 markets: , ∗1 ∗3 , ∑ 1 , (4) Where MarketTrained indicates market j was assigned to training, TotalY is the total of outcome Y (profits, sales, or customers) in market j at time t, and the standard errors are clustered over time at the market level. Table 4 shows the results of estimating equation (4). By the third year, overall profits have grown 2,554 KSH per week (9.2% of the control mean, not statistically significant), overall sales in the market by 15,353 KSH per week (13.7% of the control mean, significant at the 5 percent level), and the total number of customers in the market per week by 318 (16.7% of the control mean). These results are consistent with the overall market growing, at least over the longer-term. In rounds 2 and 4 we also conducted a market census to determine whether there were new entrants to the market since the time of training, and whether these new entrants along with existing firms that did not pass our eligibility filters had cut back on profits due to some firms being trained. Columns 3 through 6 show positive and insignificant impacts on the numbers of new entrants in these markets, the total number of firms not in our survey, and the weekly profits of these firms. The success of those trained therefore also does not come from crowding out new entrants. 6. Why do these impacts arise? Causal Chain and Mechanisms 16    Our surveys and pre-analysis plan enable us to trace out the causal chain from providing training through to changes in business outcomes, and to examine the different mechanisms through which training may or may not have an effect. Since we examine a number of different mechanisms, in Appendix 10 we provide sharpened q-values that hold constant the false discovery rate when reporting results for specific outcomes. All outcomes at the three-year horizon that have p-values below 0.05 also have sharpened q-values below 0.05, whereas only the one year impacts that are significant at the 1 percent level have sharpened q-values below 0.10. 6.1. Changes in Business Knowledge, Practices, and Use of a Mentor The first step in the causal chain is for training to lead to changes in the business knowledge and business practices of the women taking training. Business knowledge was assessed through giving respondents a description of a business and then asking them seven questions that involve calculating the revenue, value of stock on hand, variable costs, total expenses, profits, fixed costs, and break-even point. This proved very difficult for most participants, with the median respondent only getting 2 out of 7 questions right, and only 0.5 percent getting all the answers correct. This question was only asked in the first long follow-up survey and was asked of both those with surviving businesses as well as those whose business had closed down. Column 1 of Table 5 shows that there is no significant treatment effect or spillover effect on business knowledge. This is consistent with the financial literacy results of Carpena et al. (2011) who find that financial literacy training does not improve performance on questions involving numerical calculations. We measure business practices through a set of 26 questions that measure the marketing, record- keeping, buying and stock control, and financial planning of the firm. These questions are only measured in the long follow-up survey rounds and only for firms that survive. These questions have been shown to correlate strongly with business performance in a range of countries by McKenzie and Woodruff (2015), and to predict future survival and growth of the firm. The mean firm in the pure control group is employing 53 percent of these practices. Column 2 shows that the impact of being invited to training is a statistically significant increase in the use of business practices at both the 1 and 3 year horizons: treated firms have a 0.05 to 0.07 increase in the proportion of practices used. This is approximately a 10-13 percent increase on the control mean. There is a marginally significant positive spillover in business practices to untreated firms in the same marketplaces over 1 year, however the magnitude is very small (0.01), and the three-year impact is smaller still and not significant. The increase in practices is three times as 17    large for those assigned to mentoring (0.096), compared to those assigned to training alone (0.033), suggesting the mentoring did help reinforce business practices. Finally, column 3 confirms that those assigned to the mentoring treatment are indeed much more likely to say they have used a mentor. However, we also see a significant impact on using a mentor from those firms assigned only to training, suggesting that training did lead to a doubling in the use of a mentor, even without the separate mentorship intervention. 6.2 Impact on Dealing with Customers Increased marketing and better presentation of the business may enable the firm to increase sales through gaining more customers and being better able to retain existing customers. We examine this in the first three columns of Table 6. The women in our sample serve a median of 60 and mean of 100 customers per week in the control markets (after top-coding at the 99th percentile to reduce the influence of outliers). Column 1 shows that those assigned to training are serving an additional 14 customers a week after three years, significant at the 1 percent level. Columns 2 and 3 show treated firms are more likely to have gained a new customer during the last 3 months, and less likely to have lost a regular customer during the same time period. There is no significant spillover impact, which is consistent with the previous evidence that the total number of customers in the market increased. Total sales of the business increased, as did the number of customers. Column 4 considers the sales per customer. The point estimates are positive, suggesting more revenue per customer, but not statistically significant. If we instead consider outcomes in log form, so that we have the identity: log / (5) Then the treatment effects are 17.6 percent on sales (p=0.006), 9.0 percent on sales per customer (p=0.107) and 8.5 percent on the number of customers, suggesting that the increase in sales volume is coming approximately evenly from expanding the number of customers as well as from obtaining more revenue per customer. Why might business owners be able to get more customers? The qualitative work in part points to a better customer experience, with shops being cleaner, more likely to open on time, and owners being nicer to the customers. For example, the qualitative work noted one woman saying “There are those who didn’t know how to attract customers, but now I can see they have been able to attract customers…They are talking to them nicely unlike before where they would talk rudely,” 18    and another noting that now “I try to talk [with] customers in a nice and professional way. I try to make delicious meals and make my place of work look attractive to them…I make sure [my cooking] it's up to the customers’ standard, not watery food” (ICRW, 2015). In qualitative interviews with those who had gone through mentoring, one participant said “I used to not care much about my business, I could open late, and sometimes the shop would remain closed if I don’t feel like opening. Now I know that it is important to put effort into my business.” We incorporated this potential channel into the three-year follow-up survey, asking whether the business always opened at a set time each day. Column 5 shows only 34 percent of the control group do this, and training increases this by a significant 6 percentage points. Mentoring appears to have a greater impact than training alone on having regular opening hours. Appendix 11 shows that the route to more customers does not appear to be through price competition, with no significant impact of treatment on the prices charged (or input costs of) the most commonly sold goods. A second reason firms were able to attract more customers was by diversifying the range of products they sold. Banerjee and Duflo (2008) note the preponderance of small businesses in developing countries all selling similar things to one another, with no reason for customers to seek them out in terms of either product line or shopping environment. Column 6 shows that firms assigned to training were 9 to 11 percentage points more likely to have introduced a new product to their business, significant at the 1 percent level. There was tremendous variety in the new products introduced, with almost every business that said it had introduced a new product giving a different answer for what product it had introduced. Examples included avocados, oranges, carrots, garlic, fertilizer, body oil, cowpeas, belts, hair oil, tea, tobacco and Weetabix. Since most firms were selling only a handful of products, this diversification could have attracted customers that would not have otherwise purchased from them. 6.3 Impact on Financial Control, Access to Finance, and Capital Levels We next examine the extent to which training enables firm owners to produce more or earn higher profits by managing resources more effectively, and overcoming financial constraints. Several studies have emphasized the possibility that business training may have its strongest impact on sales during a bad month by helping participants identify strategies to reduce downward fluctuations in sales through diversifying the products they offer, as well as being more proactive about alternative activities during slow months. McKenzie and Woodruff (2014) note however 19    that the evidence for this has been mixed in existing studies. We examine this channel in columns 1 and 2 of Table 7. We see that although businesses are more likely to regularly use business records to know if sales of a particular product are increasing or decreasing, this does not translate into any impact on reducing fluctuations in profits in bad months relative to usual months. Columns 3, 4, and 5 of Table 7 examine the inventory management and purchasing outcomes of the firms. We see no significant impacts of treatment on the fraction of stock lost to spoilage, the receipt of bulk discounts in purchasing, or in the cost of producing the firm’s most profitable item. The latter is measured with considerable noise, so the insignificant point estimate admits a wide confidence interval. Better business practices may enable firms to obtain more financing. This may work through either the demand or supply side for formal finance. On the demand side, individuals who have gone through training may have more plans for expansion or feel more confident approaching banks. On the supply side, banks may be more willing to lend to firms which keep better records, or which have higher profitability and sales. However, column 6 of Table 7 shows that access to finance is limited in our sample, with only 17 percent of the pure control group receiving a loan from a bank or microfinance organization in the past year, and that treatment does not have a significant impact on this. This is consistent with the qualitative work, which found a number of challenges on both the demand and supply side for formal finance: women were often afraid of what might happen if they fell behind on payments, found the loan application process cumbersome and time- consuming, and also faced challenges in terms of ability to provide documentation and/or collateral in some cases (ICRW, 2015). We do find firms to be 5.8 percentage points more likely to be using business bank accounts (column 7). Columns 8 and 9 then look at whether firms have grown in size according to their inventory levels and capital stock. By year three both the inventory levels and capital stocks are higher in the treatment group, which is consistent with them having more to sell, and with building up the size of the firm. The inventory impact is larger after mentoring than after training alone. Since firms did not receive more finance, this build-up in inventories and capital is likely to reflect re-investment of profits. This may also explain why impacts are stronger after three years than one year, since it takes time for firm owners to slowly reinvest profits into more merchandise to sell. 6.4 Impacts on Owner Hours, Attitudes, and Social Capital 20    The final set of intermediate outcomes and mechanisms are examined in Table 8. We start by examining whether women change the amount of time they are devoting to their business. The qualitative work suggested this might be the case, as evidenced by this quote from one participant “Then I used to open [my business] any time I wished…I would open much later…but these days it’s better since I constantly open at nine and close at night at around eight. Those days I just used to do a little work, I could not stay for long. Whenever I got some money to pay for my merry-go- round, then I would just close my business for the day. Also, then if I got someone who was buying five bags of maize, then I could just close business and leave for home, but these days I stay until I am convinced that it’s time to leave.” (ICRW, 2015). This qualitative finding does not show up in the one year horizon, but is there over three years for both mentoring and training alone, with women working 4 hours more per week. The training was intended to also increase the confidence of women in their ability to perform business tasks. We measure entrepreneurial self-efficacy through 10 questions that measure the owner’s confidence in their ability to perform key business activities such as coming up with ideas for new products, sell a product to a customer they are meeting for the first time, and persuade a bank to lend them money for their business. The mean control group individual is very confident in their ability to do 4.5 out of these 10 tasks, and Column 2 shows that we find no significant effect of training on self-efficacy. We also ask eleven questions intended to measure the types of attitudes that Get Ahead tries to encourage. These include willingness to take risks to get ahead in business, planning for the future, feeling confident that one can find solutions to problems that arise, and keeping your eyes open for ways to improve the business. These are scored on a 5-point scale, where 5 indicates the most agreement. The mean score across these questions is 3.8 for the pure control group. Column 3 shows no significant effect of treatment on these. The training also emphasized and encouraged cooperating with other women in the marketplace. The remaining columns of Table 8 examine aspects of this. Columns 4, 5, and 6 consider membership of women’s associations, discussing business with other women in the market, and working together with other women to obtain bulk discounts or to purchase goods together. We find no significant treatment effects on any of these measures over the one-year horizon, but do find treated individuals to be more likely to belong to associations and discuss business with others over the three-year horizon. Despite the hope of the trainers that treated women would band 21    together with untreated women to purchase inputs at discounts together, there is no direct effect nor spillover effect on this outcome. 7. Cost-Benefit Our analysis shows that, over a three-year horizon, the training has grown these underdeveloped markets, increasing the total volume of sales and number of customers. As a result, gains to the treated firms do not come at the expense of lost customers for the untreated firms in the same markets. The cost of training was approximately US$200 per person assigned to training, and the gain in weekly profits was 221 KSH ($2.60). The gain in profits would therefore need to last for at least 76 weeks, or approximately 1.5 years, for the benefits of the program to exceed the costs. Although we do not have a long time series to examine these benefits, we do see them being greater at three years than one year, suggesting this appears plausible. The additional mentoring treatment cost approximately $553 USD per individual assigned to mentoring. Since we are unable to reject that the impacts are the same for mentoring as for training alone for our primary outcomes, the evidence here suggests that adding mentoring does not pass a cost-benefit test. 8. Conclusions We find the Get Ahead business training program does improve the survival, profitability, and growth of businesses that receive training. These impacts are much clearer over a three-year horizon than one year after training, and suggest the scope for small businesses run by women to expand. Moreover, we find that this growth does not appear to come from taking sales away from competitor businesses, but instead arises from growing the overall market. Many markets in developing countries are underdeveloped, with poorly managed firms selling a narrow variety of products. The results of this paper show that training can be one way to develop these markets. 22    References Banerjee, Abhijit and Esther Duflo (2008) “What is Middle Class about the Middle Classes Around the World?”, Journal of Economic Perspectives 22(2): 3-28. Bauer, Susanne, Gerry Finnegan and Nelien Haspels (2004) Gender and Entrepreneurship Together: GET Ahead for Women in Enterprise: Training Package and Resource Kit, Bangkok, Berlin and Geneva, International Labour Office. Berge, Lars, Kjetil Bjorvatn and Bertil Tungodden (2015) “Human and Financial Capital for Microenterprise Development: Evidence from a Field and Lab Experiment”, Management Science 61(4): 707-22. Bloom, Nicolas, Raffaella Sadun and John Van Reenen (2015) “Management as a Technology”, Mimeo. Stanford. Brooks, Wyatt, Kevin Donovan and Terence Johnson (2016) “The Dynamics of Inter-Firm Skill Transmission among Kenya Microenterprises”, Mimeo. Bruhn, Miriam and David McKenzie (2009) “In Pursuit of Balance: Randomization in Practice in Development Field Experiments”, American Economic Journal: Applied Economics, 1(4): 200-232. Bulte, Erwin, Robert Lensink, Rosine van Velzen and Nhung Thi Hong Vu (2016) “Do Gender and Business Trainings Affect Business Outcomes? Experimental Evidence from Vietnam”, Mimeo. Calderon, Gabriela, Jesse Cunha, and Giacomo de Giorgi (2013) “Business Literacy and Development: Evidence from a Randomized Trial in Rural Mexico.”, NBER Working Paper Carpena, Fenella, Shawn Cole, Jeremy Shapiro and Bilal Zia (2011) “Unpacking the Causal Chain of Financial Literacy”, World Bank Policy Research Working Paper no. 5798. Chandy, Rajesh, Stephen Anderson-McDonald and Bilal Zia (2016) “Pathways to Profits: Identifying Separate Channels of Firm Growth Through Business Training”, World Bank Policy Research Working Paper no. 7774. De Mel, Suresh, David McKenzie and Christopher Woodruff (2014) “Business Training and Female Enterprise Start-up, Growth, and Dynamics: Experimental evidence from Sri Lanka”, Journal of Development Economics, 106: 199-210. Diwan, Faizan, Grace Makana, David McKenzie, and Silvia Paruzzolo (2014) “Invitation Choice Structure Has No Impact on Attendance in a Female Business Training Program in Kenya” PLOS ONE, 9(10): e109873 Giné, Xavier and Ghazala Mansuri (2016) “Money or Management? A Field Experiment on Constraints to Entrepreneurship in Rural Pakistan”, Mimeo. World Bank. ICRW (2015) ““I am determined, no matter how hard it becomes, I will not give up”: A qualitative assessment of the ILO’s GET Ahead Business Training Program”, ICRW 2015. McKenzie, David (2012) "Beyond Baseline and Follow-up: The Case for More T in Experiments", Journal of Development Economics 99(2): 210-221. McKenzie, David and Christopher Woodruff (2015) “Business Practices in Small Firms in Developing Countries”, Management Science, forthcoming. McKenzie, David and Christopher Woodruff (2014) “What are we learning from business training evaluations around the developing world?”, World Bank Research Observer, 29(1): 48-82. Rotemberg, Martin (2014) “Equilibrium Effects of Firm Subsidies”, Mimeo. Valdivia, Martin (2015) “Business training plus for female entrepreneurship? Short and medium term experimental evidence from Peru”, Journal of Development Economics 113:33-51. 23    Veit, C.T. and Ware, Jnr, J.E. (1983). “The structure of psychological distress and well-being in general populations”, Journal of Consulting and Clinical Psychology, 51, 730-742 24    Figure 1: Randomization Design                                                          Baseline Sample                                                      157 Markets, 3537 Firms                                                                                                                                            Market‐level randomization                Treatment Markets              Control Markets         93 Markets, 2160 Firms        64 Markets, 1377 Firms  GET AHEAD    Business Training                                                                 Individual‐level randomization  Intervention  Treated Firms        Control Firms  1172 Firms         988 Firms          Mentor              Training Only      Training Only                          Control                            Control  Additional  Markets   Markets  Small Markets                         Markets     Small Markets  37 Markets            36 Markets        20 Markets                            44 Markets                20 Markets  Mentoring  983 Firms             963 Firms            214 Firms                              1158 Firms                  219 Firms  Intervention  524 treated          521 treated  459 control           442 controls  Notes: Graphic shows assignment to treatment. The first intervention of business training assigned firms to the GET  AHEAD business training program using a two‐stage randomization: first markets were assigned to be treatment or  control, and then, within markets, firms were randomly selected to be assigned to training. Then in a second stage,  markets  with  fewer  than  16  firms  at  baseline  (“small  markets”)  were  dropped,  and  half  the  training  treatment  markets were assigned to also have those assigned to training within the market receive mentoring. Mentoring was  only offered conditional on the firm surviving.   25    Figure 2: Quantile Treatment Effects on Profits After Three Years Treatment Effect Spillover Effect 1000 1000 500 500 Weekly Profits Weekly Profits 0 0 -500 -500 5 10 20 30 40 50 60 70 80 9095 5 10 20 30 40 50 60 70 80 9095 Percentile Percentile Notes: Intent-to-treat effects from quantile regressions controlling for randomization strata and baseline profits, combining round 4 and round 5 survey data. Straight dashed line shows OLS estimate. 95 percent confidence intervals displayed allowing for clustering at market level. 26    Figure 3: Quantile Treatment Effects on Sales After Three Years Treatment Effect Spillover Effect 6000 6000 4000 4000 Weekly Sales Weekly Sales 2000 2000 0 0 -2000 -2000 5 10 20 30 40 50 60 70 80 9095 5 10 20 30 40 50 60 70 80 9095 Percentile Percentile Notes: Intent-to-treat effects from quantile regressions controlling for randomization strata and baseline sales, combining round 4 and round 5 survey data. Straight dashed line shows OLS estimate. 95 percent confidence intervals displayed allowing for clustering at market level. 27    Table 1: Individual Characteristics and Verification of Randomization  Means by Treatment Assignment Treatment Spillover Pure Pure  Group Group Controls Control P‐value P‐value P‐value (1) (2) (3) S.D. (1) vs (2) (1) vs (3) (2) vs (3) Age 36.0 35.6 35.7 9.05 0.482 0.454 0.690 Years of Education 8.92 8.91 9.09 2.92 0.910 0.569 0.515 Married 0.67 0.66 0.67 0.47 0.404 0.638 0.518 Household Size 4.97 4.85 4.85 2.10 0.188 0.262 0.499 Age of Firm 6.39 6.57 6.27 6.92 0.574 0.741 0.403 Number of Employees 0.27 0.27 0.27 0.63 0.747 0.989 0.953 Weekly Profits 1128 1140 1091 834 0.987 0.395 0.322 Weekly Sales 5220 5885 5401 8048 0.065 0.266 0.288 Capital Stock 30571 34092 29370 60530 0.248 0.863 0.101 Ever Received Bank/MFI Loan 0.24 0.25 0.24 0.43 0.891 0.482 0.934 Keeps Records 0.37 0.34 0.34 0.47 0.235 0.094 0.974 Business Practices Score 0.53 0.53 0.52 0.19 0.934 0.487 0.598 Retail Firm 0.77 0.76 0.75 0.43 0.446 0.121 0.474 Registered with City Council 0.44 0.45 0.42 0.49 0.328 0.447 0.100 Joint orthogonality test p‐value 0.597 0.435 0.691 Sample Size 1172 988 1377 Notes: Tests of treatment versus spillover group control for individual‐level randomization strata and are based on robust standard errors. Tests of the treatment or spillover group compared to the pure control group control for market level randomization strata and use standard errors clustered at the market level. 28    Table 2: Impacts on Primary Outcomes Firm Daily Weekly Main Product Weekly   Main Product Photo Aggregate   Survival Sales Sales Sales Profits Profits Inventories Index Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 0.006 161** 284 150 78 52 482 0.049* (0.010) (68) (255) (297) (69) (91) (343) (0.026) Assigned to Training* 3 Years 0.030** 171** 1038*** 454* 221** 155 818 0.088*** (0.013) (85) (349) (268) (88) (97) (879) (0.032) Spillover Group * 1 Year 0.002 32 ‐476* 157 ‐65 ‐16 337 ‐0.011 (0.011) (70) (270) (328) (66) (93) (345) (0.026) Spillover Group * 3 Years 0.013 1 27 182 ‐29 5 671 0.002 (0.014) (85) (330) (270) (79) (95) (874) (0.030) Mean of Pure Control Group 0.852 1173 5763 3368 1439 1137 8567 0.005 Sample Size 13508 12943 12909 12064 12881 11985 5598 12923 P‐value: Training 1 year=3 years 0.056 0.889 0.006 0.380 0.051 0.399 0.679 0.148 P‐value: Spillover 1 year=3 years 0.403 0.726 0.103 0.943 0.650 0.866 0.688 0.647 P‐value: Training effect zero both years 0.068 0.047 0.008 0.239 0.043 0.269 0.344 0.024 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 0.011 172 1216** 586 279** 133 1909 0.098** (0.018) (115) (486) (366) (116) (129) (1339) (0.041) Assigned to Training Alone 0.052*** 209* 954** 523 300*** 236* 256 0.104** (0.016) (110) (427) (357) (114) (132) (941) (0.042) Spillover Group to Mentoring 0.024 ‐75 138 391 ‐16 27 1972 ‐0.001 (0.018) (112) (451) (340) (99) (120) (1299) (0.037) Spillover Group to Training Alone 0.003 122 ‐3 231 53 69 ‐192 0.034 (0.020) (105) (401) (380) (97) (129) (1095) (0.037) Sample Size 5822 5625 5606 5535 5591 5530 2655 5608 P‐value: Mentoring = Training Alone 0.030 0.765 0.605 0.886 0.877 0.522 0.184 0.901 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects, for the baseline value of the outcome, and for survey round fixed effects. Panel B uses only survey rounds 4 and 5, since mentoring intervention was carried out after round 3. Aggregate index is the average of standardized z‐scores of the other variables. See data appendix for variable definitions. 29    Table 3: Impacts on Secondary Outcomes Employed   All work Empowerment   Life Ladder Life Ladder Mental Household   for pay income Index Today 5 Years Health Durables Index Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 0.009 177* 0.109 0.254*** 0.253*** 0.265* 0.108* (0.008) (96) (0.110) (0.050) (0.064) (0.151) (0.058) Assigned to Training* 3 Years 0.020* 389*** ‐0.037 0.221*** 0.322*** 0.356** 0.091 (0.011) (141) (0.094) (0.049) (0.054) (0.162) (0.065) Spillover Group * 1 Year 0.003 ‐112 0.013 ‐0.011 0.000 0.115 0.048 (0.009) (93) (0.113) (0.056) (0.068) (0.143) (0.059) Spillover Group * 3 Years 0.011 24 0.070 0.062 0.029 0.272 0.073 (0.012) (133) (0.105) (0.046) (0.058) (0.174) (0.070) Mean of Pure Control Group 0.885 2144 7.034 5.015 7.982 17.707 ‐0.063 Sample Size 13508 12881 5873 12609 12608 5873 5823 P‐value: Training 1 year=3 years 0.323 0.114 0.289 0.581 0.420 0.659 0.813 P‐value: Spillover 1 year=3 years 0.533 0.283 0.689 0.297 0.740 0.460 0.721 P‐value: Training effect zero both years 0.169 0.019 0.543 0.000 0.000 0.034 0.137 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 0.005 488** ‐0.062 0.283*** 0.417*** 0.477** 0.226* (0.015) (187) (0.133) (0.063) (0.057) (0.206) (0.121) Assigned to Training Alone 0.035** 477*** 0.133 0.204*** 0.271*** 0.169 0.132 (0.014) (181) (0.115) (0.056) (0.064) (0.217) (0.106) Spillover Group to Mentoring 0.017 30 ‐0.019 0.192*** 0.093 0.236 0.306*** (0.015) (180) (0.142) (0.052) (0.067) (0.232) (0.117) Spillover Group to Training Alone 0.005 178 0.293** 0.042 0.064 0.049 ‐0.000 (0.016) (153) (0.128) (0.052) (0.062) (0.221) (0.128) Sample Size 5822 5591 2470 5511 5511 2470 2465 P‐value: Mentoring = Training Alone 0.062 0.958 0.174 0.270 0.037 0.237 0.450 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects, for the baseline value of the outcome where available, and for survey fixed effects. Panel B uses round 4 and 5 data only. 30    Table 4: Impacts on Market Level Total Market Total Market Total Market # New Total Other Other Firm Profits Sales Customers Entrants Firms Profits Market Assigned to Training*1 Year 792 2672 131 0.193 1.473 4439 (1378) (5211) (104) (0.509) (1.375) (2940) Market Assigned to Training *3 Years 2554 15353** 318** 0.833 1.290 2662 (1568) (6613) (124) (0.785) (1.784) (3685) Mean of Control Markets 27651 110944 1916 8.59 22.48 34870 Sample Size 628 628 628 301 301 301 P‐value: 1 Year=3 Years 0.201 0.021 0.038 0.517 0.933 0.684 Notes: Standard errors in parentheses, clustered at the market level. Unit of observation is market‐survey round. *, **, and *** denote significance at the 10, 5, and 1 percent levels respectively. Regressions include controls for randomization strata, survey round, and baseline value of outcome where available. 31    Table 5: Impact on Business Knowledge, Business Practices, and Use of a Mentor Business Business   Worked with Knowledge Practices Mentor Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year ‐0.018 0.049*** (0.089) (0.008) Assigned to Training* 3 Years 0.068*** 0.335*** (0.009) (0.037) Spillover Group * 1 Year 0.124 0.014* (0.091) (0.008) Spillover Group * 3 Years 0.006 0.018 (0.009) (0.020) Mean of Pure Control Group 2.000 0.533 0.106 Sample Size 3059 5404 2520 P‐value: Training 1 year=3 years 0.067 P‐value: Spillover 1 year=3 years 0.389 P‐value: Training effect zero both years 0.000 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 0.096*** 0.666*** (0.012) (0.028) Assigned to Training Alone 0.033*** 0.108*** (0.010) (0.020) Spillover Group to Mentoring ‐0.006 0.009 (0.011) (0.023) Spillover Group to Training Alone 0.004 0.047* (0.012) (0.026) Sample Size 2236 2212 P‐value: Mentoring = Training Alone 0.000 0.000 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects. Business knowledge only collected in round 2, mentoring only collected in round 4. Business practices collected in round 2 and 4, with regression controlling for  survey round and for baseline business practices. Outcomes are conditional on operating a firm 32    Table 6: Impact on Dealings with Customers Weekly Gained new   Lost regular Sales per Open Business   Introduced Customers Customer Customer Customer at set time New Product Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 1.7 0.023* ‐0.038** 25.5 0.116*** (3.8) (0.013) (0.017) (30.0) (0.022) Assigned to Training* 3 Years 14.5*** 0.034** ‐0.041*** 38.7 0.062*** 0.089*** (4.4) (0.016) (0.015) (25.9) (0.024) (0.021) Spillover Group * 1 Year ‐2.2 0.009 ‐0.020 0.7 0.007 (3.9) (0.014) (0.018) (20.6) (0.021) Spillover Group * 3 Years ‐0.7 0.008 ‐0.009 2.2 0.030 0.008 (4.6) (0.018) (0.016) (17.1) (0.023) (0.021) Mean of Pure Control Group 100 0.767 0.804 134 0.342 0.269 Sample Size 12867 12977 12978 11186 2398 6216 P‐value: Training 1 year=3 years 0.004 0.586 0.878 0.734 0.327 P‐value: Spillover 1 year=3 years 0.753 0.964 0.666 0.951 0.970 P‐value: Training effect zero both years 0.003 0.049 0.003 0.247 0.000 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 12.1* 0.004 ‐0.048*** 50.3 0.103*** 0.102*** (6.1) (0.021) (0.017) (38.0) (0.033) (0.028) Assigned to Training Alone 12.0** 0.054*** ‐0.054*** 24.6 0.030 0.084*** (5.2) (0.020) (0.017) (24.6) (0.031) (0.028) Spillover Group to Mentoring 0.8 0.000 ‐0.035* ‐5.3 0.026 0.021 (6.4) (0.022) (0.020) (19.1) (0.030) (0.026) Spillover Group to Training Alone ‐4.3 0.009 ‐0.002 ‐0.0 0.039 0.022 (6.0) (0.023) (0.019) (20.5) (0.032) (0.029) Sample Size 5602 5636 5636 4801 2090 2636 P‐value: Mentoring = Training Alone 0.992 0.020 0.758 0.558 0.071 0.597 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects. Opening at set time only asked in round 4 survey and is conditional on business operating. 33    Table 7: Impacts on Financial Control, Access to Finance, and Capital Levels Ratio of Monitors Fraction   Received a Production   Received   Has  Capital Worse/Current Sales Stock Lost bulk  Cost Loan from Business Inventory Stock Profit Trends to spoilage discount Change (%) Bank/MFI Bank Account Value Value Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 0.014 0.112*** ‐0.007 0.009 15.536 0.017 2475 ‐52 (0.009) (0.023) (0.014) (0.023) (15.561) (0.017) (2125) (882) Assigned to Training* 3 Years ‐0.007 0.141*** ‐0.014 0.034 13.954 0.013 0.058** 5889** 3219*** (0.011) (0.022) (0.021) (0.023) (19.084) (0.019) (0.022) (2978) (1085) Spillover Group * 1 Year ‐0.004 0.025 0.037** ‐0.028 7.825 0.014 ‐339 ‐828 (0.009) (0.024) (0.017) (0.025) (17.121) (0.018) (1966) (917) Spillover Group * 3 Years ‐0.011 0.000 ‐0.005 0.013 ‐5.558 ‐0.017 ‐0.003 3350 672 (0.012) (0.023) (0.024) (0.024) (20.019) (0.021) (0.023) (2829) (1065) Mean of Pure Control Group 0.450 0.269 0.194 0.423 160.4 0.172 0.354 19918 13410 Sample Size 10609 6214 4989 5879 5541 5404 3005 6183 6203 P‐value: Training 1 year=3 years 0.136 0.263 0.766 0.447 0.935 0.850 0.107 0.002 P‐value: Spillover 1 year=3 years 0.622 0.355 0.102 0.255 0.526 0.139 0.093 0.174 P‐value: Training effect zero both years 0.252 0.000 0.767 0.308 0.569 0.583 0.137 0.005 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring ‐0.024* 0.194*** ‐0.038 0.029 ‐8.074 0.004 0.048 15371*** 3180** (0.012) (0.029) (0.026) (0.030) (23.297) (0.026) (0.034) (4523) (1504) Assigned to Training Alone ‐0.015 0.070*** ‐0.009 0.002 ‐15.671 0.044* 0.099*** 4705* 2561* (0.013) (0.024) (0.028) (0.030) (25.575) (0.026) (0.025) (2793) (1330) Spillover Group to Mentoring ‐0.010 ‐0.012 0.002 ‐0.012 ‐3.198 ‐0.023 0.010 9707** 628 (0.014) (0.031) (0.032) (0.030) (25.752) (0.027) (0.035) (3770) (1365) Spillover Group to Training Alone ‐0.029** ‐0.014 ‐0.029 0.010 ‐41.846 0.006 0.023 2649 ‐133 (0.014) (0.032) (0.034) (0.032) (26.112) (0.028) (0.025) (3198) (1451) Sample Size 4563 2636 2078 2497 2389 2236 2636 2634 2635 P‐value: Mentoring = Training Alone 0.515 0.000 0.353 0.432 0.791 0.157 0.137 0.021 0.720 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** denote significance at the 10, 5, and 1 percent levels. All regressions control for randomization strata fixed effects. Mentoring regressions only for three year follow‐up. Business bank account not asked in round 2 follow‐up survey. 34    Table 8: Attitudes and Associations Own Labor Entrepreneurial   GET Ahead   Women's # Discuss Works together   Hours Self‐Efficacy Attitudes Association Business with other women Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 0.059 ‐0.037 ‐0.018 0.003 0.181 0.035 (1.106) (0.181) (0.021) (0.017) (0.210) (0.024) Assigned to Training* 3 Years 4.596*** 0.057*** 0.584** 0.033 (1.274) (0.018) (0.277) (0.026) Spillover Group * 1 Year ‐2.060* ‐0.180 ‐0.047** 0.026 0.032 ‐0.000 (1.138) (0.171) (0.022) (0.019) (0.224) (0.025) Spillover Group * 3 Years ‐0.192 ‐0.013 0.066 0.010 (1.356) (0.017) (0.308) (0.024) Mean of Pure Control Group 42.6 4.457 3.802 0.115 4.672 0.490 Sample Size 6215 3059 3059 5400 5398 5403 P‐value: Training 1 year=3 years 0.001 0.042 0.259 0.959 P‐value: Spillover 1 year=3 years 0.189 0.155 0.934 0.731 P‐value: Training effect zero both years 0.001 0.005 0.072 0.242 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 4.777*** 0.055** 0.920** 0.027 (1.642) (0.023) (0.362) (0.039) Assigned to Training Alone 4.300** 0.062*** 0.303 0.030 (1.677) (0.023) (0.329) (0.032) Spillover Group to Mentoring ‐0.429 ‐0.029 0.040 ‐0.016 (1.758) (0.020) (0.417) (0.032) Spillover Group to Training Alone 0.101 0.002 ‐0.003 0.041 (1.659) (0.025) (0.390) (0.031) Sample Size 2635 2234 2235 2236 P‐value: Mentoring = Training Alone 0.814 0.795 0.135 0.940 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** denote significance at the  10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects. Mentoring regressions only for three year follow‐up. Self‐efficacy and Get Ahead attitudes only asked in round 2. 35    ONLINE APPENDICES Appendix 1: Timeline Appendix 2: Additional Details on Listing and Sample Selection Appendix 3: Model for a GET Ahead 5-day workshop for entrepreneurs Appendix 4: Sample Means by Assignment to Mentoring Appendix 5: Mentoring design and implementation Appendix 6: Measurement of Key Outcomes Appendix 7: Data Availability Appendix 8: LATE Impacts for Primary Outcomes Appendix 9: Robustness Checks on Primary Outcomes Appendix 10: Sharpened Q-values for Mechanism Impacts Appendix 11: No Impact on Market Prices or Purchase Costs 36    Appendix 1: Timeline Listing and Baseline Survey: June-November 2013 Business Training Intervention: June-November 2013 Round 2 Follow-up: June-October 2014 (One-Year Post-Training) Round 3 Short Follow-up: November 2014-February 2015 Mentoring Intervention: July-November 2015 Round 4 Follow-up: February-July 2016 (Three-Years Post-Training; 6-10 Months Post- Mentoring) Round 5 Follow-up: May-October 2016. 37    Appendix 2: Additional Details on Listing and Sample Selection Altogether 6,296 female-owned businesses in 161 markets were listed. After the census, three markets in Kakamega county were dropped because the number of women in these markets was too few. We then applied an eligibility filter to determine which women to include in the baseline survey. This filter required the women to have reported profits, and not to have reported profits that exceeded sales; to have a phone number that could be used to invite them for training; to be 55 years or younger in age; to not be running a business that only dealt with phone cards or m- pesa, or that was a school; that the person responding not be an employee; that the business not have more than 3 employees; that the business have profits in the past week between 0 and 4,000 KSH; that sales in the past week be less than or equal to 50,000 KSH; and that the individual had at least one year of schooling. These criteria were chosen to reduce the amount of heterogeneity in the sample (thereby increasing our ability to detect treatment effects), and to increase the odds of being able to contact and find individuals again. Applying this eligibility filter reduced the 6,296 individuals to 4,037 individuals (64%). Baseline surveys took place soon after the listing surveys in each county, between June and November 2013. Out of a target of 4,037 individuals, we were able to interview 3,537 (87.6%) in time to consider them for inviting to training. 38    Appendix 3: Model for a GET Ahead 5-day workshop for entrepreneurs Day 1 Day 2 Day 3 Day 4 Day 5  Opening  The business  Production,  Finance  Management or  Gender equality environment: services and  Manage of self networking promotion: life she is not alone technology and others  Action cycle of people  Business ideas  Marketing  Business Planning and enterprises  Marketing support and  Closing  The business networking woman: she can do it 39    Appendix 4: Sample Means by Assignment to Mentoring or Training Only Appendix Table 1: Means by Treatment Status for Mentoring Intervention Assigned Training Spillover   Spillover for Pure Mentor Only for Mentor Training Only Control P‐value P‐value (1) (2) (3) (4) (5) (1) vs (2) (3) vs (4) Age 35.73 36.39 35.38 35.70 35.80 0.262 0.807 Years of Education 9.08 8.76 9.00 8.89 9.08 0.019 0.321 Married 0.69 0.65 0.68 0.64 0.68 0.166 0.999 Household Size 4.96 4.99 4.93 4.83 4.86 0.962 0.841 Age of Firm 6.36 6.82 6.53 6.73 6.31 0.310 0.491 Number of Employees 0.27 0.26 0.27 0.27 0.26 0.985 0.422 Weekly Profits 1124 1115 1132 1155 1085 0.782 0.643 Weekly Sales 5657 5054 6028 6016 5247 0.596 0.558 Capital Stock 37859 24222 39042 27324 26733 0.001 0.005 Ever Received Bank/MFI Loan 0.26 0.26 0.23 0.27 0.23 0.652 0.061 Keeps Records 0.36 0.37 0.33 0.36 0.34 0.967 0.318 Business Practices Score 0.53 0.52 0.53 0.53 0.52 0.153 0.920 Retail Firm 0.77 0.81 0.76 0.79 0.75 0.223 0.115 Registered with City Council 0.48 0.36 0.49 0.39 0.40 0.001 0.023 Sample Size 524 521 459 442 1158 Notes: Means and Sample Sizes are shown for Sample in Non‐Small Markets, for which mentoring intervention applied. 40    Appendix 5: Mentoring Design and Implementation The mentoring intervention provided personalized, hands-on problem-solving support and peer learning to women who had previously received the Get Ahead program with the goal of reinforcing intended business training outcomes – from improved management skills to business growth. The intervention targeted 446 women who had been exposed to the business training in 2013 and expressed interest in further support through mentorship. Two local, public service providers, Kenya Industrial Estate (KIE) and the Women Enterprise Fund (WEF), were identified as partners in the delivery of the mentoring services. KIE implemented the mentoring in Kakamega and Kisii, while WEF did it in Embu and Kitui. The design of the mentoring intervention combined group and individual sessions for a period of five months (July to November 2015). Each female-owned firm (or mentee) received 15 mentoring sessions: 10 through group sessions and five through one-to-one meetings with the mentor. Group sessions occurred twice a month, every two weeks, while individual sessions took place once a month. Each mentor was assigned a group of five mentees. The table below summarizes the topics covered during the 10 group mentoring sessions. Individual sessions deepened discussions on the above topics based on the needs of the mentee and her business. The ILO contracted a mentoring expert for program development and curriculum design, which was subsequently discussed and delivered to mentors from Kenya Industrial Estate and the Women Enterprise Fund.  110 mentors were recruited, only 100 were subsequently trained, and 89 selected for the program.  The program reached out to 446 women who had participated in the Get Ahead Program, 392 signed up, were inducted and received training. However, only 361 women stayed in the program throughout the five months.  Therefore, while originally 89 mentors were linked to 392 mentees, by the end of the program the number of mentees per mentor ranged from 3 to 6.  Drop out reasons included (i) in most cases lack of interest after realizing there were no grants involved in the offer, and (ii) in few cases considerations about the mentoring program not being helpful in enhancing skills and business growth. Monitoring and evaluation of the mentoring was done through (i) monthly meetings with mentors and implementing partners, (ii) monthly reports per mentor, (iii) tracking of journals filled by mentors after each session, (iv) visits and phone calls with mentors for follow up and support, (v) two meetings between the ILO and the implementing partners throughout the duration of the program, and (vi) one closing forum in each county to gather feedback from all parties. 41    Mentoring approach: group sessions Session Objective/Module Outcome Time Session 1:  Introduction of mentor mentees  Knowing each other 2 hrs Introduction, objectives and  Definition of objectives  Being clear on the mentorship objective and expectations agreement  Understanding the mentoring  Understanding and signing the mentoring agreement agreement Session 2: Goals review Mentees goals established 2 hrs GROW goal  How the mentees will look like when they are successful  How will the business look like after growing, as regards to sourcing, production, marketing, packaging, sales, networking, access to finance and general business management Session 3: Reality analysis Current status established 2 hrs GROW reality  Where is the mentee right now in view of her personal and business goals  Business performance established in terms of; sourcing, production, marketing, packaging, sales, networking, access to finance and general business management Session 4: Identifying key gaps (goals vis-à- Gaps identified. This will be done by exploring the difference 2 hrs GROW gap analysis vis current status) between the goal and the current reality. Session 5: Exploring options to addressed A list of options to close each of the gaps identified (in terms of 2 hrs GROW exploring options identified gaps capabilities, skills and assets). Session 6: Financial analysis  Exploring the feasibility of filling in the gaps 2 hrs Feasibility assessment  What is the cost of implementing the options  What will be the source of finance required to close the gaps Session 7: Plan of action to address identified  Analysing options and agree on way forward 2 hrs GROW way forward /action gaps  Drawing an action plan with clear timelines and budget points Session 8: Evaluating progress  Exploring what the mentee has done on the agreed Action Points 2 hrs Implementing the way forward  Identifying successes, challenges, way forward Session 9: Evaluating progress  Exploring what the mentee has done on the agreed action points 2 hrs Implementing the way forward  Identifying successes, challenges, way forward Session 10: End of program evaluation, review  Documenting the impact of the mentorship, i.e. how has the 2 hrs Evaluation, review of action of action plan and agreement on mentee and business benefited from the mentoring sessions plan and sustainability sustainability plan  Reviewing the action plan based on sessions 8 and 9’s outcomes  Agreeing on a sustainability plan after the end formal mentorship 42    Appendix 6: Measurement of Key Outcomes All nominal values were converted into real (August 2013) Kenyan shillings using the consumer price index for the midpoint of each survey round. Firm Survival is measured as whether the owner still operates a business, regardless of whether or not they have changed the business line. For individuals who could not be interviewed, survival was measured by asking family members and neighbors the status of the owner. Daily sales are sales of the business in the last day, coded as zero if the business was closed that day, or is closed for good. It is truncated at the 99th percentile. Weekly sales are total sales of the business in the last week, coded as zero if the business was closed that week, or is closed for good. It is truncated at the 99th percentile. When aggregated across all firms in the market, this forms Total market sales. Main product sales are obtained by multiplying the number of units of the main product sold in the last week by the unit price, truncated at the 99th percentile and coded as zero if the business is closed. Business Profits are measured using the direct question of de Mel et al. (2009)9, asked about the last week as a reference period since pre-testing found a weekly rather than monthly recall was easier for business owners to answer: “what was the total income the business earned during last week after paying all expenses including wages of employees, but not including any income you paid yourself. That is, what were the profits of your business during last week?” This is coded as zero if the business is closed, and truncated at the 99th percentile. When aggregated across all firms in the market, this forms Total market profits. Main product profits are obtained by multiplying the mark-up on the main product sold by the number of units sold of this product in the past week, truncated at the 1st and 99th percentiles, and coded as zero if the business is closed. Photo inventories are the value of inventories as assessed by valuing a photograph of the business inventories. A common set of market prices are used to aggregate products. Two independent enumerators would count the number of each product they see in the photo (e.g. 53 tomatoes) and then aggregate by the price per product. If the two valuations differed by more than 5,000 KSH, they would iterate again until they agreed on a valuation. Aggregate index of profits and sales is the average of standardized z-scores of the primary profits and sales measures. Employed for pay is coded as one if they are self-employed or worked for wages in the past week. Income from work is the sum of weekly profits and income from wage work in the past week.                                                              9  De Mel, Suresh., McKenzie, David., Woodruff, Christopher., (2009). “Measuring microenterprise profits: Must we ask how the sausage is made?” Journal of Development Economics 88(1): 19-31.   43    Empowerment index is the sum of the following outcomes: Compelled to spend money on husband or family (coded 1 if they answer no); not the only person with access to their firms’ money (coded as 1 if only they have access); has some money which they have sole control over and can spend how they like; do not need anyone’s permission to visit a friend, to travel to sell a business asset, to travel to a new location to work, to stay overnight in a different town, to work later than usual hours, to take out a loan, or to spend money on an investment for their business. Life Ladder Today and Life Ladder 5 Years are measured by a standard 10-step Cantril ladder, where individuals are asked to imagine the best (step 10) and worst (step 1) possible lives for themselves, and then say which step represents their current position and where they will be in 5 years. Mental health is measured by the MHI-5 index of Veit and Ware (1983), coded so higher scores denote better mental health. Household durables index is the first principal component of dummy variables for ownership of 10 household assets (iron and heaters, fridge or freezer, fan, sewing machine, radio or CD player, TV or DVD player, motorcycle or scooter, car or van, oven, and gas cooker) and of the number of cows owned and number of goats owned. Number of new entrants is the number of firms operating in the market outside of our experimental sample that have opened since the baseline survey and training intervention. Total other firms is the total number of other firms operating in the market outside of our experimental sample (new entrants plus those pre-existing firms that were not included due to being absent from the market at the time of listing, or being dropped by our eligibility filters). Other firm profits is the total profits in the market of the other firms outside the survey sample. Weekly customers is the number of customers the firm has in the past week, truncated at the 99th percentile. This is coded as zero for firms that are closed. When aggregated to the market level, this forms Total market customers. Business knowledge is the number correct out of 7 questions intended to measure whether the firm owner can calculate sales, expenses, and profits. Business practices is the proportion of 26 practices in marketing, record-keeping, buying and stock control, and financial planning used by the firm (McKenzie and Woodruff, 2015). This is only measured for firms that are surviving at the time of the survey. Worked with a mentor is a dummy variable for whether they have worked with a mentor to try to improve their business in the past year (only asked in round 4). Entrepreneurial self-efficacy is the number out of 10 of business activities that the owner rates themselves as “very confident” in their ability to do (only asked in round 2). This includes entrepreneurial tasks like “estimate customer demand for a new product,” “persuade a bank to lend you money” and “identify good employees.” Get Ahead Attitudes is the sum of scores on 11 questions designed to measure attitudes Get Ahead training is meant to encourage. These are scored 1 through 5, where 1=strongly disagree, 5 = 44    strongly agree. Questions will be coded so that higher scores indicate better entrepreneurial attitudes. Examples include “Even when my business is going well, I keep my eyes open in case I find a way to improve it,” “I don’t worry about where my business will be in the future – I just plan week to week based on what comes up” (negatively coded), and “My business provides about the same as others/is doing about the same as others, so there’s no need to make it better.” (negatively coded). 45    Appendix 7: Data Availability Appendix Table 2 shows data availability by treatment status. Appendix Table 2: Data Availability by Treatment Status Round 2 Round 3 R2 or R3 Round 4 Round 5 R4 or R5 Panel A: Interviewed Assigned to Training 0.034*** 0.030*** 0.023*** 0.009 0.023* 0.010 (0.012) (0.011) (0.008) (0.013) (0.013) (0.011) Spillover Group 0.014 ‐0.013 ‐0.003 ‐0.026* 0.002 ‐0.009 (0.013) (0.014) (0.010) (0.014) (0.015) (0.012) Pure Control Mean 0.886 0.889 0.943 0.894 0.876 0.923 Panel B: Data on Survival Available Assigned to Training 0.011* 0.016** 0.002 0.005 0.010 0.003 (0.006) (0.006) (0.003) (0.009) (0.010) (0.007) Spillover Group 0.015** ‐0.003 0.002 0.000 0.008 0.004 (0.006) (0.008) (0.004) (0.010) (0.010) (0.007) Pure Control Mean 0.968 0.962 0.991 0.947 0.924 0.970 Panel C: Data on Weekly Sales and Profits Available Assigned to Training 0.027** 0.031*** 0.016** 0.013 0.031** 0.011 (0.012) (0.009) (0.006) (0.013) (0.013) (0.010) Spillover Group 0.015 ‐0.006 ‐0.003 ‐0.007 0.011 0.006 (0.013) (0.012) (0.008) (0.014) (0.014) (0.011) Pure Control Mean 0.907 0.913 0.964 0.903 0.881 0.939 Sample Size 3537 3537 3537 3537 3537 3537 Notes: Robust Standard Errors in Parentheses, Clustered at the Market Level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. Data on weekly sales and profits availability codes data as available if firm is known to be closed (since then sales and profits are known to be zero). 46    Appendix 8: LATE Impacts for Primary Outcomes Appendix Table 3: LATE Impacts on Primary Outcomes Firm Daily Weekly Main Product Weekly   Main Product Photo Aggregate   Survival Sales Sales Sales Profits Profits Inventories Index Impact of Receiving Training, Allowing for Spillovers Received Training*1 Year 0.007 202** 355 187 97 65 597 0.061* (0.012) (85) (318) (370) (86) (113) (421) (0.032) Received Training* 3 Years 0.038** 212** 1283*** 562* 273** 192 1017 0.108*** (0.016) (104) (428) (329) (108) (119) (1087) (0.039) Spillover Group * 1 Year 0.002 32 ‐476* 157 ‐65 ‐16 336 ‐0.011 (0.011) (70) (268) (327) (66) (93) (343) (0.026) Spillover Group * 3 Years 0.013 1 25 181 ‐30 5 670 0.002 (0.014) (85) (328) (269) (79) (94) (867) (0.029) Sample Size 13508 12943 12909 12064 12881 11985 5598 12923 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects, the baseline value of the outcome, and  survey round fixed effects. Receipt of training instrumented with assignment to training. 47    Appendix 9: Robustness Checks on Primary Outcomes We carry out three robustness checks on our primary outcomes. First, we use the objective visual photographic evidence as to whether the business looks bigger or not than it was at baseline (Appendix Table 4). Second, we report results conditional on survival (Appendix Table 5). Third, we examine transforms of profits and sales which are less susceptible to outliers (Appendix Table 6). Since our initial screening was intended to create relatively homogeneous firms, our pre- analysis plan specified that our primary analysis would focus on the outcome variables measured in terms of levels. However, it indicated that as a robustness check, we would also examine the impact of treatment on the inverse hyperbolic sine transformation of total profits in the last month, and total sales in the last week: log(y+(y2+1)1/2). This is similar to the log transformation, but allows for zeros and negative values. We also consider the log transformation, which conditions on survival. The first column of Appendix Table 6 also considers an alternative definition of survival, in which we assume which firms that were observed closed in round T, and then not observed in subsequent waves, remain closed. Our results are robust to this alternative. In the last two columns of Appendix Table 6, we examine whether training leads to more accurate reporting. Column 6 measures the treatment impact on the number of reporting errors made. We consider six errors: reporting sales less than profits, daily sales less than weekly sales, worst week profits greater than current week’s profits, profits and sales for their main product higher than total profits and sales, and unit costs higher than prices for their most profitable product. We find no significant treatment effect on the number of errors made, suggesting that training does not affect reporting accuracy. Column 7 considers the difference (in log terms) between the value of inventories reported by the firm, and the value assessed by our team based on their photo. If treated participants were deliberately reporting their businesses to be bigger, we should expect a systematic bias in this difference with treatment status. However, we cannot reject that the difference does not vary with treatment status, suggesting reporting accuracy does not change with treatment. Together with the physical evidence in Appendix Table 4, this suggests the greater size of the treatment firms is a real, and not just a reporting, effect. 48    Appendix Table 4: Is the Business Visually Bigger After Three Years? All firms Survivors All Firms Survivors Assigned to Training 0.042** 0.037 (0.021) (0.023) Spillover Group 0.019 0.019 (0.022) (0.023) Assigned to Mentoring 0.059** 0.058* (0.028) (0.031) Spillover Group for Mentoring 0.022 0.017 (0.030) (0.032) Assigned to Training Only 0.034 0.022 (0.028) (0.029) Spillover Group for Training 0.028 0.023 (0.026) (0.027) Sample Size 2864 2571 2527 2265 P‐value: Mentoring=Training Only 0.440 0.286 Notes: Robust standard errors in parentheses, clustered at the market level.  *, **, and *** denote significance at the 10, 5, and 1 percent levels respectively. Outcome is whether photograph of business inventories in Round 4 survey is shows business to be larger than baseline photograph shows.  Columns 2 and 4 condition on the firm surviving, columns 1 and 3 code closed firms as not being bigger. 49    Appendix 5: Impacts on Primary Outcomes Conditional on Survival Daily Weekly Main Product Weekly   Main Product Photo Aggregate   Sales Sales Sales Profits Profits Inventories Index Panel A: Impact of Assignment to Training, Allowing for Spillovers Assigned to Training*1 Year 165** 294 126 69 46 604 0.051* (77) (287) (329) (75) (102) (392) (0.029) Assigned to Training* 3 Years 145 967** 380 182* 124 487 0.080** (94) (382) (294) (98) (107) (1017) (0.035) Spillover Group * 1 Year 30 ‐557* 168 ‐79 ‐23 357 ‐0.012 (78) (290) (368) (70) (105) (404) (0.028) Spillover Group * 3 Years ‐24 ‐46 169 ‐61 ‐17 607 ‐0.001 (95) (354) (303) (86) (108) (982) (0.033) Mean of Pure Control Group 1386 6818 3993 1702 1348 10694 0.107 Sample Size 11339 11305 10460 11277 10381 4775 11319 P‐value: Training 1 year=3 years 0.804 0.027 0.509 0.177 0.567 0.902 0.324 P‐value: Spillover 1 year=3 years 0.574 0.138 0.998 0.840 0.964 0.797 0.733 P‐value: Training effect zero both years 0.097 0.034 0.435 0.180 0.494 0.307 0.068 Panel B: Impact of Mentoring Compared to Training Alone, Allowing for Spillovers Assigned to Mentoring 171 1289** 628 295** 132 2081 0.107** (132) (529) (395) (132) (142) (1560) (0.045) Assigned to Training Alone 165 792* 360 233* 183 ‐658 0.087* (118) (456) (383) (119) (140) (1093) (0.044) Spillover Group to Mentoring ‐124 ‐28 355 ‐68 1 2020 ‐0.015 (120) (467) (367) (102) (133) (1414) (0.039) Spillover Group to Training Alone 139 26 243 62 66 ‐460 0.046 (118) (438) (441) (110) (150) (1220) (0.042) Sample Size 4862 4843 4772 4828 4767 2193 4845 P‐value: Mentoring = Training Alone 0.967 0.354 0.559 0.674 0.761 0.058 0.693 Notes: Robust standard errors in parentheses, clustered at the market level  *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects , for the baseline value of the outcome and for  survey round fixed effects. Panel B uses only survey rounds 4 and 5, since mentoring intervention was carried out between round 3 and round 4. Aggregate index is the average of standardized z‐scores of the other variables. See data appendix for  description of the different outcome variables. 50    Appendix Table 6: Robustness of Primary Outcome Results Alternative Inverse Hyperbolic Log  Log   Number Inventory Survival Sales Profits Sales Profits Errors Reporting Assigned to Training*1 Year 0.006 0.124 0.098 0.094** 0.068 ‐0.014 0.060 (0.010) (0.116) (0.101) (0.037) (0.047) (0.026) (0.063) Assigned to Training* 3 Years 0.031** 0.569*** 0.428*** 0.172*** 0.138*** 0.006 0.078 (0.014) (0.136) (0.119) (0.048) (0.051) (0.030) (0.063) Spillover Group * 1 Year 0.001 ‐0.015 ‐0.008 ‐0.027 0.010 0.004 ‐0.053 (0.011) (0.120) (0.103) (0.040) (0.045) (0.029) (0.066) Spillover Group * 3 Years 0.013 0.237 0.182 0.009 0.043 ‐0.014 0.028 (0.014) (0.151) (0.130) (0.047) (0.047) (0.029) (0.074) Mean of Pure Control Group 0.835 7.270 6.240 8.265 6.889 0.542 0.849 Sample Size 13666 12909 12881 10420 10790 11541 4391 P‐value: Training 1 year=3 years 0.046 0.001 0.003 0.036 0.112 0.607 0.836 P‐value: Spillover 1 year=3 years 0.374 0.069 0.110 0.385 0.463 0.638 0.375 P‐value: Training effect zero both years 0.074 0.000 0.001 0.002 0.027 0.847 0.344 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects, the baseline value of the outcome where available, and for survey round fixed effects. See test for variable definitions 51    Appendix 10: Sharpened Q-values for Mechanism Impacts To control for multiple hypothesis testing when examining multiple mechanisms, we to construct sharpened q-values following Anderson (2008) and Benjamini et al. (2006). This process uses a two-stage procedure to control the false discovery rate when reporting results for specific outcomes.10 Appendix Table 7 reports the original p-values and corresponding sharpened q-values. We see that all three year outcomes that have p-values below 0.05 also have sharpened q-values below this level. In contrast, over the one year horizon, only the impacts on business practices, introducing new products, and monitoring sales trends are significant after this adjustment. 1 Year 3 Years Table   Measure P‐value Sharpened Q P‐value Sharpened Q Table 5 Business Knowledge 0.839 1.000 0.175 0.096 Table 5 Business Practices 0.000 0.001 0.000 0.001 Table 5 Worked with Mentor 0.000 0.001 Table 6 Weekly Customers 0.648 1.000 0.001 0.004 Table 6 Gained New Customer 0.082 0.387 0.040 0.039 Table 6 Lost New Customer 0.025 0.127 0.006 0.011 Table 6 Sales per customer 0.396 0.847 0.138 0.087 Table 6 Open set time 0.010 0.015 Table 6 Introduce new product 0.000 0.001 0.000 0.001 Table 7 Profit ratio 0.131 0.537 0.530 0.249 Table 7 Monitors Sales Trends 0.000 0.001 0.000 0.001 Table 7 Fraction stock spoiled 0.620 1.000 0.521 0.249 Table 7 Received bulk discount 0.700 1.000 0.141 0.087 Table 7 Production cost change 0.320 0.847 0.466 0.249 Table 7 Received loan 0.323 0.847 0.506 0.249 Table 7 Business bank account 0.010 0.015 Table 7 Inventory value 0.246 0.847 0.050 0.046 Table 7 Capital stock 0.953 1.000 0.003 0.007 Table 8 Own labor 0.957 1.000 0.000 0.001 Table 8 Self‐efficacy 0.838 1.000 0.293 0.152 Table 8 Get Ahead attitudes 0.396 0.847 0.032 0.036 Table 8 Women's Association 0.870 1.000 0.002 0.006 Table 8 Number Discuss Business 0.388 0.847 0.037 0.039 Table 8 Works with others 0.154 0.544 0.195 0.102                                                              10  Anderson, Michael (2008), "Multiple Inference and Gender Differences in the Effects of Early Intervention: A  Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects", Journal of the American Statistical  Association, 103(484), 1481‐1495; and Benjamini, Yoav, Abba M. Krieger, and Daniel Yekutieli (2006) “Adaptive  Linear Step‐Up Procedures That Control the False Discovery Rate.” Biometrika 93 (3): 491–507.  52    Appendix 11: No Impact on Market Prices or Input Costs The businesses in our sample sell a diverse range of products, making overall analysis of price changes in the market difficult. We therefore focus on the five most commonly sold products, which are sold by between 20.9 percent (tomatoes, price per single tomato) and 11.9 percent (price per kilogram of sugar) of firms. Firms were asked if they sell these products, and if so, the price they charge for a specified unit, and the purchase price they pay for these inputs. Appendix Table 7 estimates the treatment impacts on these prices. We see no significant treatment impacts or spillover impacts on these prices and unit costs. Moreover, the point estimates for the training group are positive for four out of five products. These results are consistent with firms not growing sales by lowering prices of commonly sold products, but instead by improving customer service and then increasing the variety of other products sold. Appendix Table 7: Impacts on Key Market Prices and Unit Costs Maize Kale Sugar Tomatoes Beans Panel A: Sale Price Charged  Assigned to Training ‐0.217 1.852 2.879 4.971 2.380 (0.952) (2.509) (3.671) (4.766) (3.283) Spillover Group 0.941 ‐1.993 4.196 1.809 2.407 (1.066) (1.941) (3.297) (4.829) (3.636) Mean of Pure Control Group 67.289 14.382 36.661 59.774 93.435 Sample Size 1042 1436 1171 1120 1170 Panel B: Unit Cost of Item Assigned to Training ‐0.491 ‐0.888 2.664 2.976 1.807 (0.788) (1.116) (3.254) (4.039) (2.976) Spillover Group 0.476 ‐1.584 3.523 1.255 2.462 (0.911) (1.062) (2.944) (4.002) (3.181) Mean of Pure Control Group 56.179 8.547 30.053 48.368 69.409 Sample Size 1036 1420 1169 1115 1169 Notes: Robust standard errors in parentheses, clustered at the market level. *, **, and *** indicate significance at the 10, 5, and 1 percent levels respectively. All regressions control for randomization strata fixed effects. Maize and Beans are price per 2 kilogram container, Kale (Sukuma) is price per bunch, Sugar is price per kilogram, and tomatoes is price per single tomato. Prices and Costs truncated at the 1st and 99th percentiles. 53