Policy Research Working Paper 10489 Returns to Soft Skills Training in Rwanda Andrew Brudevold-Newman Diego Javier Ubfal Africa Region & Gender Global Theme June 2023 Policy Research Working Paper 10489 Abstract Young adults seeking to enter the labor market often program facilitated accelerated entry into the labor market confront a skills mismatch with firms reporting difficulty in a period characterized by COVID-19-related disruptions. finding new entrants with appropriate levels of soft skills. These effects dissipated over the following year as more jobs This paper reports findings from a randomized controlled became available in the economy and the control group’s trial in Rwanda in which recent graduates from tertiary employment caught up with that of the treatment group. education were randomly assigned to a two-week inten- The paper finds evidence of significant job network expan- sive soft skills training program developed and delivered by sion for participants of the training, which could have led staff of the University of Rwanda. Results indicate that the to faster labor market entry for the treated youth. This paper is a product of the Gender Innovation Lab, Africa Region and the Gender Global Theme. 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://www.worldbank.org/prwp. The authors may be contacted at dubfal@worldbank.org, and abrudevoldnewman@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 Returns to Soft Skills Training in Rwanda Andrew Brudevold-Newman and Diego Ubfal* JEL Codes: I26, J16, J24, 015 Keywords: soft skills, labor market entry, transition from school to work, networks, RCT. * Brudevold-Newman: World Bank, abrudevoldnewman@worldbank.org. Ubfal: World Bank, dub- fal@worldbank.org. This paper greatly benefited from comments by Sofia Amaral, Edward Kadozi, Estelle Koussoub´ e, David McKenzie, Francois Ngoboka, and Adam Osman. Baba-Ali Mwango and Leodomir Mfura provided excellent research assistance and project coordination. We gratefully acknowledge financial support, through a grant to Innova- tions for Poverty Action (IPA) from Deutsche Gesellschaft fur¨ Internationale Zusammenarbeit (GIZ) GmbH as part of the Special Initiative “Decent Work for a Just Transition” and on behalf of the German Federal Ministry for Economic Cooperation and Development (BMZ). We also acknowledge support from the World Bank’s Umbrella Facility for Gender Equality. The Umbrella Facility for Gender Equality (UFGE) is a multi-donor trust fund administered by the World Bank to advance gender equality and women’s empowerment through experimentation and knowledge creation to help governments and the private sector focus policy and programs on scalable solutions with sustainable outcomes. The UFGE is supported with generous contributions from Australia, Canada, Denmark, Finland, Germany, Iceland, Ireland, Latvia, the Netherlands, Norway, Spain, Sweden, Switzerland, United Kingdom, United States, the Bill and Melinda Gates Foundation, and the Wellspring Philanthropic Fund. This paper is a product of the Gender Innovation Lab, Office of the Chief Economist, Africa Region. The findings, interpretations, and conclusions expressed in this work do not necessarily reflect the views of The World Bank Group, its Board of Executive Directors, or the governments they represent. The project received ethical approval by Innovations for Poverty Action Institutional Review Board and the Rwanda National Ethics Committee. AEA RCT Registry ID: AEARCTR-0009714. All errors and omissions are our own. 1 Introduction The school-to-work transition is a critical and often challenging period for young people, with youth unemployment rates often far exceeding those of adults. Employers are frequently re- luctant to hire recent graduates because the skills acquired at school may not match those de- manded by the firm. Surveys of employers from around the world indicate that firms struggle ˜ to find workers with appropriate levels of soft skills (Cunningham and Villasenor, 2016). These soft skills are often complementary to technical skills, and have been shown to be predictive of labor market success and productivity (Adhvaryu et al., 2023; Ajayi et al., 2022; Borghans et al., 2008; Campos et al., 2017; Heckman and Kautz, 2012). Despite growing recognition of their importance, there remains relatively limited evidence on how to effectively develop soft-skills (Deming, 2022). Dedicated soft skills training may help bridge the skills gap, allowing prospec- tive employees to acquire the skills that are not developed in school but are highly demanded by employers. This project seeks to test whether an intensive soft-skills training course for recent tertiary educa- tion graduates can effectively improve their soft skills and boost their downstream labor market outcomes. The soft skills training was commissioned by the Rwanda Development Board in re- sponse to findings from a survey of Rwandan employers in which they reported that they valued employees with strong communication skills but that these employees were particularly diffi- cult to find.1 The training was prepared and implemented by professors of the University of Rwanda, the largest education institution in the country, and focused on effective communica- tion and networking, including material on both interpersonal (teamwork, collaboration, trust, empathy, and negotiation) and intrapersonal (self-awareness, personal initiative, and persever- ance) skills.2 The training was delivered through an intensive two-week bootcamp (160 hours in total) in the South of Rwanda.3 A group of 450 randomly-assigned, eligible recent graduates (50% women) were invited to attend at no cost under the assumption that recent graduates are either liquidity or credit constrained and thus might not be able to afford the cost of the training.4 1 We conducted the survey online in April 2020 with a sample of 30 large employers who typically hire recent graduates. Employers were given a list of 28 skills, following Cunningham and Villasenor ˜ (2016), and were asked to rate them based on: 1) how much they value the skill when hiring a new graduate and 2) how difficult the skill is to find when hiring a new graduate. Results indicated that the seven skills local employers valued the most were commitment, honesty, integrity, punctuality, responsibility, communication and technical skills. However, when asked about the skills most difficult to find among those they value, they selected communication and technical skills. While universities and vocational schools target technical skills, there seems to be a clear demand for improved communication skills, which was observed both among public and private firms in our sample. 2 See Ajayi et al. (2022) for definitions of each of these interpersonal and intrapersonal skills. 3 The bootcamp took place during the peak of the COVID-19 pandemic and extensive measures were taken to guarantee in-person activities while mitigating health risks (e.g., COVID-19 tests were conducted on all participants). 4 Neither vocational nor regular schools focus on soft skills development, and there are few soft-skill training providers in the country. Moreover, financial institutions do not offer loans that cover post-graduate education. 1 Employment statistics show that in Rwanda, like in many countries in the region, youth face a challenging school-to-work transition: 28% of youth are unemployed, with women facing unem- ployment rates that are 7 percentage points higher than those of men (Rwandan National Institute of Statistics, 2022). Youth who are able to secure an income generating activity overwhelmingly do so in the informal sector, which accounts for 94% of the income generating activities. The data present a mixed picture of the labor market for tertiary-educated Rwandans, who face a higher unemployment rate of almost 34% but are also far more likely to eventually find formal-sector jobs: more than half of working, tertiary-educated Rwandan youth hold formal jobs.5 In this context, it is reasonable to expect that a soft skills training, if effective, could improve not only informal income generating activities, but also access to formal jobs. Moreover, improved soft skills may also help participants adapt to changes in work requirements and retain their jobs (Deming, 2017; Heinrich and Housemann, 2021). The Rwanda Development Board advertised the program on several online job boards and in- vited youth who graduated from university or technical/vocational education programs in the previous two years to apply. Around 2,400 participants applied to the program, and close to 1,800 completed an online baseline survey, many more than the 450 available vacancies. Of these, 500 women and 500 men were randomly selected and invited to attend information sessions where public lotteries to assign the vacancies of the training were conducted. The lotteries selected 450 participants who were assigned to the training (225 women and 225 men) and 450 participants who were assigned to the control group. Approximately 74% of participants from all around Rwanda attended the bootcamp, even though for logistical reasons it was held in the South of the country following strict COVID-19 guidelines. Our results indicate that the program facilitated an accelerated entry into the labor market in a period characterized by COVID-19-related disruptions. Using data from three online follow- up surveys—administered 4, 10 and 15 months after the training, each with very high response rates—we see increases in the probability of earning income, hours worked, and main job earn- ings. There is evidence of significant increases in job search in the short run and persistent in- creases in job networks, increasing interactions with other program participants even 15 months after the training. However, there is no evidence that the training facilitated youth transitioning into formal employment, with no effects on changes in access to formal jobs even in the short run. We validated the lack of effect on formal jobs with a snapshot from administrative data on Social Security records, which indicates no differences between treated and control youth in ever being registered with a formal employer 17 months after the training. The short-run labor 5 All the statistics presented in this paragraph were obtained from authors’ calculations using the microdata of the Rwanda Labor Force Survey. 2 market effects are driven by informal employment and dissipate over the following year, as more jobs become available in the economy and the control group employment catches up with that of treated participants. We do see, however, that 15 months after the training, treated youth are more likely to have a permanent contract in the informal sector and to have informal jobs in the formal sector, indicating that the training might have helped them sustain their jobs but not switch to better jobs. Interestingly, we find large gender gaps in soft skills against women, consistent with a recent paper looking at gender gaps in soft skills for 17 African countries (Ajayi et al., 2022). However, we do not see significant effects of the training for either men or women on interpersonal or intrapersonal soft skills measured 15 months after the training, and there is no heterogeneity in the effects of the training by gender. This paper makes three main contributions. First, we contribute to the growing experimental evidence base examining labor market impacts of soft-skills interventions (Acevedo et al., 2020; Adhvaryu et al., 2023; Allemand et al., 2023; Barrera-Osorio et al., 2020; Groh et al., 2016; Osman and Speer, 2022). The first four papers are different in focus and context, studying either on-the- job soft skills training, training combined with internships or vocational education, or focusing on low-educated samples.6 Our study is more closely related to the last two papers. Groh et al. (2016) evaluate a 45-hour soft skills training for women community college graduates in Jordan, and do not find any significant employment impacts either in the short or medium term. Osman and Speer (2022) compare a 120-hour training program focused on soft skills (mainly on commu- nication skills) with two programs, one focused on technical skills (English and software), and another one mixing soft and technical skills for recent college graduates in the Arab Republic of Egypt. They find that the program focused on soft skills does not improve labor market out- comes, whereas the program that combines soft and technical skills has larger effects on income and on the probability of finding better jobs (that require speaking English) even in the long run. One limitation of these two studies is that they do not have measures of soft skills, and thus they cannot test whether the lack of effects is observed even when soft skills are improved.7 We can rule out significant effects of the training on an index of 14 soft-skill measures after 15 6 Adhvaryu et al. (2023) study on-the-job soft skills training among Indian garment workers and find significant productivity gains, which they attribute to improved teamwork and collaboration, but no effects on wages or work retention. Acevedo et al. (2020) test the effects of internships combined with either vocational and soft skills training or only soft skills training for at-risk youth in Dominican Republic. They find lasting improvements in personal skills and expectations for women (stronger for the full treatment) and not for men, and improvements for labor market outcomes for women but only in the short run (similar for both treatments). Allemand et al. (2023) evaluate a training program focused on activating conscientiousness for construction workers in Senegal and find that it reduces job turnover. Barrera-Osorio et al. (2020) estimate the effects of two versions of vocational training emphasizing hard or soft skills for relatively poor vocational training students in Colombia. They find that the program focusing on technical skills does better at improving labor market outcomes in the short run, but that focusing on soft skills helps youth sustain employment over the longer term. 7 Recent evidence has demonstrated that soft skills are malleable in different settings (Acevedo et al., 2020; Ashraf et al., 2020; Campos et al., 2017; Ubfal et al., 2022). 3 months, and we explore a different channel by which soft skills training programs focused on communication could improve labor market outcomes (at least in the short run): improvements in job networks.8 Second, we contribute to a growing literature examining the relationship between training pro- grams and networking outcomes. As Carranza and McKenzie (2023) document with data from seven countries, the vast majority of jobs in developing countries are found by workers learning about jobs through their social networks. Training programs may shift networking behaviors and contribute to job finding through two main mechanisms: a direct effect on networks through interaction with classmates of the training cohort and an indirect effect through skills acquired in the training that could lead to increased networking activities. Evidence for the direct effect comes from papers showing increases in networking independent of training content when com- paring training participants with control group participants (e.g., Campos et al. (2017) find that both a hard-skills training program and a soft-skills training program yielded positive effects on network size among microentrepreneurs in Togo).9 Dimitriadis and Koning (2022) provide evi- dence for the indirect effect of training showing that a 2-hour interpersonal social skills training randomized within participants of a marketing training in Togo increased networking among them. Building on the potential for soft-skills training to boost networking behaviors, the train- ing we study included an explicit networking component. We complement the existing research by showing that trainees boosted networking by both leveraging the expanded network that resulted directly from participating in the training, and also increasing their engagement with existing contacts, which provides evidence for the indirect effect.10 Finally, we contribute in two ways to the nascent literature examining the relationship between gender, soft skills, education, and labor market outcomes. We study the extent to which the soft skills training impacts vary by gender and also examine the extent to which a set of recently identified stylized facts covering soft-skills and women across Sub-Saharan Africa hold within our sample of tertiary educated youth. In many countries, women must overcome a range of gender-specific constraints including restrictive social norms, differential care responsibilities, and different job-search network structures.11 Recent empirical work has documented that these 8 Another channel explored in the literature is the role of soft skill certificates including performance indicators, which can provide relevant signals to both participants of the training and potential employers (Carranza et al., 2022). The intervention we study provided a certificate of completion but did not include assessment results. 9 The direct effect is also obtained with exogenously introduced peer relationships that do not necessarily include training, such us visits to peer entrepreneurs (Cai and Szeidl, 2017), online chatting (Vega-Redondo et al., 2019) or work groups (Sandvik et al., 2020). Afridi et al. (2022) find smaller direct effects for women in a setting with conservative gender norms. 10 Wanberg et al. (2020) also find that an online networking training in the United States increased networking intensity among people the trainees knew. 11 While Rwanda compares favorably to many of its regional peers in terms of gender equity—ranking among the 4 gender-specific constraints also include a dramatic soft-skills gender-gap in Sub-Saharan Africa, with men consistently scoring significantly higher on a range of soft-skills measures (Ajayi et al., 2022).12 Closing the skills gender gap by strengthening inter and intrapersonal skills may help women negotiate within the household, expand their job search practices, and ultimately improve their labor market outcomes. The evidence base is still thin, with none of the papers focused on the effects of soft skills training for highly educated youth cited above conducting heterogeneity by gender, and with papers focused on less-educated samples finding mixed evidence.13 We find similar take-up of soft skills training for women and men, and no significant differences in treatment effects. The paper continues as follows. Section 2 presents the experimental design, data, and COVID- 19 context. Section 3 describes the methodology and presents the results. Finally, Section 4 concludes. 2 Experimental design, data, and COVID-19 context We evaluate a soft skills training for recent tertiary education graduates in Rwanda. The project was implemented by the University of Rwanda in collaboration with the Rwanda Development Board (RBD). RDB advertised the program on several local online job boards and invited youth who graduated from university or technical/vocational education programs after 2018 to apply. Ultimately, these advertisements yielded more eligible applicants for the training than could be accommodated by the implementing organization; we use this over-subscription together with in-person lotteries to randomize 900 eligible applicants—stratified by gender—to either the soft skills training or a control group. Of the 2,390 eligible applicants who completed an initial short application form, 1,779 completed a longer baseline survey. Of these, 500 young men and 500 young women were randomly se- lected and invited to attend one of three regional, in-person lotteries to allocate the vacancies top 5 countries in the 2019 Africa Gender Equality Index—there are still prominent gaps in some areas. In 2021, labor force participation was 62% for men and 47% for women, and the share of youth not in education, employment or training was 41% for women and 30% for men (Rwandan National Institute of Statistics, 2022). 12 Notably, the observed soft-skills gender gap is largest among the most educated and returns to these soft skills vary by gender, with intrapersonal skills strongly correlated with women’s earnings. 13 Groh et al. (2016) have a sample of only educated women and find no effects of soft skills training on labor market outcomes. Among the papers focused on lower-educated samples, Acevedo et al. (2020) find more positive effects of internships combined with soft skills training for women than for men on personal skills, expectations and short-run labor market outcomes; while Barrera-Osorio et al. (2020) find mainly similar effects of vocational training with focus on soft skills for men and women, with some evidence of stronger effects for men. Among papers studying non-labor market outcomes, Campos et al. (2017) find large effects for both men and women of personal initiative training on business outcomes in Togo, and Ashraf et al. (2020) find significant effects on education outcomes of a training focused on negotiation skills for girls in Zambia. 5 for the training. The regional lotteries mirrored the initial implementation plan with separate training sessions scheduled for university campuses in Kigali, Northern Province, and Southern Province. In total, 960 participants attended the in-person lotteries: stratifying by gender, 450 were randomly allocated to the boot-camp training, 450 were randomly allocated to a control group, and 60 were assigned to a non-evaluation wait-list. Several days before the boot-camp, the implementing partners shifted from three planned training locations to a single location in the Southern Province. To boost attendance, the project team contacted all participants to remind them of the training, notify them that the costs of traveling to the training location would be covered, and answer any questions or concerns the participants might have about the late change of venue. 2.1 Intervention The intervention we study is a 13-day (160 hours in total: 102 hours of lectures and an estimated 58 hours of practical exercises and assignments), in-person soft skills training for recent univer- sity graduates implemented by the University of Rwanda. The curriculum focused on effective communication, networking, and both interpersonal (teamwork, collaboration, trust, empathy, and negotiation) and intrapersonal (self-awareness, personal initiative, and perseverance) skills, and was designed to address employers’ perceptions that tertiary education curricula produced technically proficient graduates who lacked the complimentary soft skills to effectively transition into the labor market. More details on the training are presented in Figure A1.14 Participants were offered transportation fees to the training venue, as well as accommodation and subsistence fees. To facilitate attendance of women with young children, the program covered accommoda- tion and transportation fees for a caregiver. Given that the in-person training was implemented during the COVID-19 pandemic, all participants were tested upon arrival at the training venue and symptoms were monitored on a daily basis to minimize contagion risks. In terms of delivery, the training sought to break away from the traditional university lecture setting and incorporated a range of participatory methods including case studies, simulations, role play, group activities, panel discussions, presentations, and games. Moreover, since participants lived at the premises of the university for the two weeks of the training, there was extensive social interaction (e.g., dinners, after class gatherings) that could have further facilitated the development of social net- works. 14 The training was facilitated by the University of Rwanda professors who jointly developed the curriculum. The training comprised 10 units across 3 modules: Communication skills for work readiness (Introduction to communi- cation skills for work readiness; Types of communication skills; Channels of communication; and Communication barriers); Networking skills (Rationale of professional networking; Networking strategies; Building and maintaining professional networks; Job search with people and internet research engines); and intrapersonal and interpersonal skills. 6 2.2 Data Collection Our analysis draws on four main sources of data. First, applicants completed an online short application form and a subsequent longer baseline survey that focused on pre-program employ- ment history and status, labor supply, socio-demographic characteristics, and soft skills.15 The online baseline and the public lotteries were completed in April 2021, prior to the start of the training, which took place from the end of April 2021 through the second week of May 2021. We conducted three online follow-up surveys: the first follow-up was conducted between August and October 2021, an average of 4 months after the conclusion of the training; the second was conducted in February and March 2022, about 10 months after the training; and the final follow- up was completed in July through September 2022, an average of 15 months after the training.16 To mitigate the high risk of respondents not completing a long online survey, the questionnaire was relatively short and focused primarily on labor market modules. Participants were com- pensated for their time with phone credit. The third follow-up included the same soft-skills assessment used in the baseline.17 Attrition rates were remarkably low in all of our follow-up surveys. We successfully surveyed 96 percent of the baseline sample for the 4-month and 10- month follow-ups, and 97 percent for the 15-month follow-up. Attrition is not associated with treatment status in any of our three rounds of data collection.18 We believe that one of the key reasons of this high response rates, unusual for online surveys, is that participants were selected into the study among those who have completed both an online application form, and an online baseline survey, which implied that they were already familiar with this survey methodology. Moreover, the sample consists of highly educated youth who are more likely to be comfortable using digital tools. We also found that conducting phone call reminders to those who had not completed the online survey after a few days was crucial to significantly increase response rates. 2.3 COVID-19 and the Rwandan Labor Market As in many other countries, Rwanda responded to increased COVID-19 cases with a series of full and partial lockdowns in March-June 2020, January-February 2021, and June-August 2021 (International Monetary Fund, 2021). These lockdowns and the subsequent re-opening of the 15 Appendix Figure A2 summarizes the project implementation timeline together with the various data collection rounds. A small subset of the sample completed the soft-skills portion in-person, just before the in-person lotteries that determined program allocation. 16 At each round, the staff from Innovations for Poverty Actions sent the link to participants by email and text mes- sages. Respondents who had not completed the online survey after 3 weeks were called and offered the opportunity to complete the survey over the phone. Only 6% of respondents chose this method, on average, over the three survey rounds. 17 We gave participants the option to stop the survey before the soft skills module, be paid half the participation fee, and complete it at another time. However, very few participants made use of this option and the larger majority completed the full survey at once. 18 Appendix Tables A1-A4 demonstrate sample balance at each survey round while Appendix Table A5 reports regression results testing whether treatment status predicts attrition for each of the survey rounds. 7 economy play a significant role in defining the economic context facing the study participants at each follow-up survey, from a particularly weak labor market at the first follow-up to one that had largely recovered by the third follow-up. Specifically, the employment rates for university- educated individuals averaged 77% over the four quarters leading up to the first set of lockdowns in March 2020. The labor market softened over the following year, dropping to 67.8% by the time the program was advertised in March 2021 and bottoming out at 67.3% at the time of the first follow-up in August 2021 (Rwandan National Institute of Statistics, 2020-2022). The employment rate ticked up over the following year, increasing to 72.5% by the second follow-up in February 2022 and exceeding the pre-COVID-19 level (78.8%) by our final follow-up in August 2022 (Rwandan National Institute of Statistics, 2020-2022). 2.4 Sample Characteristics Appendix Table A1 reports the socio-demographic, education, and labor market characteristics of the sample at baseline, together with the p-value of the difference between the treatment and control groups.19 At baseline, the sample summary statistics describe an educated population in the process of forming families and entering the labor force. In line with the education-eligibility criteria of the program, all participants had recently completed a tertiary education degree: 77% of the sample completed a bachelor’s degree, 21% completed a TVET diploma degree, and 2% completed a master’s degree. Despite an average age of almost 28, only 12% of the sample were married or co-habitating, with a similar proportion having a child. There are important gender differences in these behaviors, with women reporting being almost twice as likely to be married as men (15% vs 8%), and about 2.5 times as likely to have a child (18% vs 7%). On the labor mar- ket front, while almost 85% had ever worked, only 40% had worked for pay in the month before the survey, and only 3% report working in a formal job. Across all of the variables reported in Table A1, the only statistically significant difference between the treatment and control groups is the share of respondents from Southern Province. This broad balance is confirmed in Panel D which reports a joint F-test of equality across the treatment and control groups for the variables in Panels A-C. Recent national surveys demonstrate that our sample is significantly more educated and less likely to have started families than the broader population of Rwandan youth. Specifically, only about 8% of Rwandan women aged 24-30 have attended some tertiary education—one of the eligibility criteria for our program—while 62% are married and 77% have given birth, rates almost 4 times higher than our sample (Rwandan National Institute of Statistics, 2021). Tertiary educated women are generally significantly better off than their less-educated peers, with living 19 Appendix Tables A6-A7 present balance tables of pre-program measures of soft skills for the treatment and control groups. 8 conditions and assets-based wealth index that is over 1 standard deviation higher. Applicants are generally slightly less likely to be working than tertiary educated young women as a whole, of whom 56% are currently working, with wage work and self-employment accounting for 71% and 24% of the jobs, respectively. 2.5 Compliance Despite the late change in announced venue,20 and the significant time commitment of the train- ing, attendance was fairly high with 74% of the individuals assigned to treatment attending at least one session.21 Attendance at each of the daily training sessions was also high: of the 334 individuals who attended at least 1 session, 325 attended 10 or more sessions. A post-training participant evaluation conducted on the last day of the training suggests that participants were satisfied with it: over 90% of attendees indicated that they were either satisfied or very satisfied with the content, delivery, and relevance of the training.22 Participants were slightly less sat- isfied with the duration, pace, and exercises of the training, with respondents suggesting that future iterations be conducted over a longer period. The high satisfaction bears out in ex-post participant-reported willingness-to-pay for the training, with participants answering that they would pay US$208 for the training. We discuss how this compares to the cost of implementation in the Discussion section, below. 3 Analysis In this section, we report intent-to-treat (ITT) estimates of the training. Treatment assignment was random within strata,23 so the pooled impacts of the intervention across three survey rounds on a given outcome Yi,t can be measured using the following ANCOVA regression specification: Yi,t = α + β × Treati + γ × Yi,t0 + ξ t + δstratum + ε i,t where yi,t represents outcome y for household i at follow-up t, Treati is an indicator variable equal to 1 if individual i was invited to the training, Yi,t0 is the baseline value of outcome y for individual i, ξ t is a vector of survey round indicator variables, and δstratum is a series of strata 20 Asmentioned above, one week before the training, the implementation partner shifted from offering venues in three regions of the country as announced in the public lotteries, to having only one venue in the South of the country. To compensate for this shift, there was a significant investment in contacting participants and facilitating their transportation to the venue. 21 Appendix Table A8 examines the correlates of attendance in the treatment group, finding lower attendance among individuals originally assigned to complete the training in Kigali, as well as those who reported positive pre-program wage or casual employee work. 22 Appendix Table A9 presents summary statistics from the participant evaluation. 23 There were 18 strata corresponding to: 6 lotteries in Kigali (three for men and three for women), 6 lotteries in the South (four for men and two for women), and 6 lotteries in the North (four for men and two for women). 9 fixed effects. We cluster standard errors at the individual level. We also examine how treatment effects changed over time. Specifically, we run the following regression allowing for the interaction between treatment and round dummies: Yi,t = α + β × Treati + δ × ξ t × Treati + γ × Yi,t0 + ξ t + δstratum + ε i,t where the vector δ captures round-specific treatment impacts. 3.1 Labor Market Impacts Table 1 presents the intent-to-treat impacts of the training on individuals’ main income generat- ing activity and labor supply. Odd-numbered columns present the pooled impacts over the three survey rounds while even-numbered columns show impact heterogeneity by round.24 Over- all, our results indicate that the soft skills training boosted labor market outcomes: individuals assigned to treatment report a 9 percent increase in the likelihood of earning any income (an increase of 6 percentage points over a control mean of 69%), a 16 percent increase in earnings, and a 12 percent increase in hours worked in the main job, on average over the 15 months after the training. However, these pooled effects obscure significant time trends in the impacts, with large, positive short-term impacts waning over time. Specifically, at the first follow-up (4-months after the training), individuals assigned to the training were 13 percentage points (21 percent) more likely to report an active income generating activity than the control group. This impact wanes over the following year with a marginally significant 5-percentage-point increase in the likelihood of reporting an active income generating activity 10 months after the training and no evidence of a remaining impact 15 months after the training. As mentioned above, the timing of the surveys is important for interpreting impacts: the first follow-up took place at the peak of COVID-19 effects on the labor market, with the employment rate for university-educated in- dividuals increasing by over 10 percentage points over the subsequent follow-ups, potentially facilitating the control group’s catch up. A similar waning pattern of impacts holds for earnings: individuals assigned to treatment report earning 33% more after 4 months, about 18% more after 10 months, and a near zero impact after 15 months. However, the impacts for labor supply do not wane. At each follow-up, treatment-group individuals report working 3-4 more hours over the last 7 days than the control group.25 Taken together, our results suggest that the training fa- cilitated an accelerated entry into the labor market that stagnated and allowed the control group 24 Appendix Table A10 presents results for secondary activities. There is no effect on the probability of having a secondary activity or on the intensive margin (e.g., hours worked in a secondary activity). 25 The persistent increase in hours worked together with the dwindling earnings impacts raise the prospect that the training led to treatment-group individuals taking less remunerative jobs. Appendix Table A11 shows that the training did not have a significant impact on earnings per hour for the first and second follow-ups. However, we do see some negative effects on earnings per hour in the third follow-up, although this effect is not statistically significant. 10 to catch up over the subsequent year (as evidenced by the increasing trend in positive earnings and hours worked that control groups means show in Table 1). Appendix Table A12 examines the occupational choice impacts of the training, showing that the short-term income-generating activity impacts observed above were largely driven by increases in the likelihood of intern/apprentice positions and self-employment, though the shift towards these sectors is short-lived with no impacts persisting past the first follow-up. Overall, our pooled results suggest that the training had limited aggregate impacts on occupational choice: the esti- mated coefficients indicate shifts of 1 percentage point or less on probability of self-employment, employer status, or apprenticeships, none of which are statistically significant. The coefficient on wage work is larger, at 4 percentage points, but this is also not statistically significant. Looking at impact heterogeneity over time, we see some evidence of short-term impacts for selected work categories that diminished over time. Specifically, our results indicate that the training led to short-term increases of almost 30% in both the likelihood of holding an apprentice position and self-employment. Both of these impacts disappear by the second follow-up, with limited evi- dence of any impacts at either the second or third follow-ups. Notably, the limited increases in wage work in later survey rounds suggests that the apprenticeships secured at the first follow-up did not transition to subsequent paid wage work at a faster rate than the control group was able to secure paid work. Table 2 explores whether the training improved job quality. There are no effects on the probabil- ity that youth work a job in the area of their studies.26 The training did increase the probability of having a written contract (a pooled effect of 14%, which is only significant at the 10% level), a permanent contract (a large pooled effect of 38% due to an increase of 3 percentage points over a low control mean of 8%), and of working in a registered firm (a pooled effect of 14%).27 Interestingly, the positive impact on having a permanent contract is driven by effects after 10 and 15 months, which could reflect improvements over time. However, no effects are detected on the probability of receiving job benefits that characterize formal jobs (contributions to Social Insur- ance, Health Insurance, and paid annual leave). The lack of program impacts on the probability of having a formal job is also confirmed when looking at a snapshot of administrative, Social Security records obtained 17 months after training.28 26 There are also no effects on measures of job satisfaction asked only in the second and third follow-ups. 27 Around 26% of employees in the formal sector report having informal jobs in Rwanda (Rwandan National Institute of Statistics, 2022). 28 In November 2022, 17 months after training, we were able to recover information on registration to Social Security records for all 900 youth in our sample. The data indicate that 62% of youth in the control group were registered with Social Security at that time, which means they have had a formal job at some point in their lives. Appendix Table A13 shows a small treatment effect of 2 percentage points that is not statistically significant. 11 3.2 Mechanisms We now shift to explore potential mechanisms for the short-term boost in labor market outcomes. Tables 3 and 4 examine impacts on intermediate outcomes associated with each of the different training modules—soft skills, communication skills, and networking—with the results suggest- ing that networking seems to be the dominant path. Table 3 demonstrates that the program did not have meaningful persistent impacts on self-reported measures of soft skills, showing limited impacts after 15 months on aggregate soft skills or on separate indices for either intrapersonal or interpersonal skills.29 We can rule out impacts on soft skills larger than 0.1 standard deviation. Table 4 examines impacts on job search practices, finding mixed results. First, we see that the program did not consistently boost job search, though it did have small short-term impacts on job search readiness measures including confidence approaching employers and fraction of the sample with a digital version of the CV, albeit on an already high level among the control group. Notably, the training did have significant and sustained impacts on job networking: individuals assigned to the training discussed jobs with 17% percent more people than the control group; an effect that is sustained over the duration of the program. Table 5 shows that the shift in job-search networking stems from both increases in discussions with people that were part of pre-program networks and discussions with an expanded network resulting from participation in the training. The pooled specification shows that the training increased the share of individuals who spoke about jobs with family, friends, schoolmates, and social media connections by 7, 6, 10, and 11 percentage points, respectively. These impacts gen- erally persist across the different survey rounds, with impacts on friends being the only category where impacts wane over the study period. The last column captures the exogenous network effect of the program—that training attendees expanded their network through the program by meeting and engaging with the other trainees (the question ask about mates from any training program)—and shows a significant increase in the share of respondents reporting discussing jobs with training mates, which diminishes over time, but shows significant and large positive effects even after 15 months. In March 2023, we conducted focus groups with 6 participants from the training, and all of them mentioned that 21 months after the training, they were still engaging via WhatsApp with other participants of the training.30 29 Soft skills were only measured at baseline and endline, and we cannot rule out that they were affected but only in the short run. Tables A14-A15 present impacts for each of the individual soft skills we measured demonstrating limited evidence in support of impacts across any of the sub-domains. We chose this list of 14 categories of soft skills following Ajayi et al. (2022). 30 Some participants even showed us the WhatsApp groups in their phones, which were formed by around 30 previous participants of the training program, and they mentioned having received invitations to social events, and also information regarding labor market opportunities via these groups. 12 3.3 Treatment Effect Heterogeneity: Soft Skills, Gender, and Economic Outcomes This section assesses the extent to which several stylized facts related to soft skills, gender, and la- bor market outcomes in Sub-Saharan Africa hold within our sample of highly-educated Rwandan youth before testing for heterogeneous treatment effects from the soft skills training. Specifically, Ajayi et al. (2022) identify: (1) a significant gender gap of around 0.15 standard deviation across a wide range of soft skills with the interpersonal skills gap increasing in education; (2) socio- emotional skills are associated with higher earnings for men and women; and (3) interpersonal skills are more strongly correlated with earnings for women than for men.31 Consistent with the stylized fact, Appendix Table A16 finds significant gender gaps across three measures of soft skills—an aggregate index of all soft-skills measures, an aggregate of intrap- ersonal soft-skill measures, and an aggregate of interpersonal soft-skills measures—though our gaps are slightly smaller than those estimated for the region as a whole using more heteroge- neous samples.32 Appendix Table A17 demonstrates that equivalent results hold when control- ling for degree type and work status while also showing that point estimates support gender gaps favoring men across all sub-measures of intrapersonal and interpersonal skills: results for interpersonal skills are particularly striking where women report significantly lower levels for each of the skills. Building on these gender gaps in soft skills, we next examine the relationship between gender, socio-emotional skills, and economic outcomes. We find a persistent gender gap in whether the respondent had worked in the 30 days before the baseline: women were about 8 percent- age points less likely to be working than men.33 We find limited evidence that either aggregate socioemotional skills, intrapersonal skills, or interpersonal skills close the gap for women (Ap- pendix Table A18). A similar pattern holds for pre-program income: women earn about 40% less than men and socio-emotional skills do not close the gap. Indeed, the results suggest limited returns to socio-emotional skills for tertiary educated women in the Rwandan labor market.34 Appendix 2 explores whether the effects of the training were different for women and men. 31 The authors identify a fourth set of stylized facts detailing how the skills gap varies with education. We are unable to speak to this point as all applicants to the soft-skills program had completed some form of tertiary education. 32 Intrapersonal skills include emotional awareness, self-awareness, emotional regulation, self-control, perseverance, personal initiative, and problem solving. Interpersonal skills include listening, empathy, expressiveness, relatedness, influence, negotiation, and collaboration. We follow Ajayi et al. (2022) in the definition and construction of each index. 33 The administrative records recovered 17 months after training show a significant gender gap in formal jobs: while around 70% of men in both treatment and control groups had had a formal job at some point in their lives, this was true for only 54% of women in the control group (the share in the treatment group was 58%, though the treatment effect is not statistically significant). 34 We should take these results with caution since these are not causal estimates, and there is a literature pointing out the difficulty of measuring soft skills with self-reported questions (e.g., Laajaj and Macours (2021)). 13 Overall, we find no evidence of treatment effect heterogeneity by gender.35 Most of the interaction coefficients for main and intermediate outcomes are small and not statistically significant.36 4 Discussion We evaluate a 2-week soft skills training developed and delivered by the main national university in Rwanda that aimed to boost labor market outcomes for highly-educated individuals. While the program facilitated an accelerated entry into the labor market during the peak of the COVID- 19 crisis, these effects dissipated over the following year as more jobs become available and the control group’s employment caught up with that in the treatment group. We find evidence of sig- nificant job network expansion for participants of the training, which could have helped treated youth find jobs more quickly and explain the short-run impacts of the training. Our results, in conjunction with the literature, raise two important points about the impacts of soft-skills programs. First, soft skills training alone may not be sufficient to yield sustained labor market impacts. Osman and Speer (2022) make the crucial point that there might be significant complementarities between soft skills and technical skills even among college graduates, finding larger returns to a program that bundled soft skills together with English and software skills for university graduates in the Arab Republic of Egypt. Even though 97% of our sample reported intermediate or advanced English skills at the application stage, program facilitators still flagged limited English skills as one of the main challenges during implementation.37 These English skills (or other complementary technical skills) may ultimately represent a separate binding con- straint, holding back sustained labor market impacts from the soft skills training, even among a sample of highly-educated youth. This points to the importance of combining soft skills training with programs enhancing other work-related skills.38 Furthermore, combining soft skills training programs with detailed soft skills assessments could act as a strong signal for both trainees and potential employers and improve job matching (Carranza et al., 2022). Second, despite previous literature demonstrating the malleability of soft skills, we find limited evidence that the training impacted a set of 14 soft skills measures.39 It is possible that programs covering the same amount 35 Gender was the only dimension of heterogeneity pre-specified in the registration of the experiment, and used in the stratification of the randomization. Moreover, the experiment included a quota for 50% of female participants, which increased statistical power for this analysis. We conducted exploratory unreported analysis for other possible sources of heterogeneity such as employment status at baseline and location (given that we found these variables differently affected take up of the training), and we did not observe any significant heterogeneity in training impacts. 36 Interestingly, we see no effects on discussing job opportunities with friends and family for women, while effects are positive for men. The effects are large and significant for both men and women on discussing job opportunities with social media contacts, with classmates or other training participants. 37 We find limited evidence of heterogeneity by self-reported English competency as shown in Appendix Table A19. 38 For example, programs that proved effective include on-the-job skills training (Adhvaryu et al., 2023) or programs that help youth accumulate early work experience while they study (Le Barbanchon et al., 2023). 39 As noted above, several papers point to the difficulty of measuring soft skills with self-reported questions and 14 of hours but more spread over time (our program offered 160 hours of training in two weeks) could have more sustained impacts on soft skills, or that follow-up activities are required to help participants implement the changes recommended in the training to more sustainably build their soft skills. Finally, it is important to note that even the relatively short-lived earnings impacts we observe could still make a sufficiently low-cost intervention cost-effective. However, the in-person imple- mentation evaluated here had high costs, partly due to the measures taken to mitigate COVID-19 risk: $622 per intended participant and $835 per actual training attendee.40 One potentially less-costly implementation modality would integrate the training more systematically into exist- ing tertiary education curricula; while this would offset many of the large implementation costs (classroom facilities, transportation, and catering), the fact that the impacts seem to stem from networking suggests that the in-person interaction with new training mates (separate from their earlier university classmates) may be particularly important in driving the impacts observed here. Additionally, the high self-reported willingness to pay for training reported by participants could represent another avenue for reclaiming some of the training costs. A topic for future research is why the longer-term network impacts of the training did not translate into higher earnings and better job opportunities after the peak of the COVID-19 pandemic. even when behavioral tasks are available (Laajaj and Macours, 2021; Danon et al., 2023). 40 About 10% of the costs were direct COVID-19 testing and equipment costs. Facilities, trainers, and catering were the main other costs, representing 29%, 22%, and 21% of the implementation cost, respectively. 15 References Acevedo, P., G. Cruces, P. Gertler, and S. Martinez (2020): “How vocational education made women better off but left men behind,” Labour Economics, 65, 101824. Adhvaryu, A., N. Kala, and A. Nyshadham (2023): “Returns to On-the-job Soft Skills Training,” Journal of Policital Economy. In Press. Afridi, F., A. Dhillon, S. Roy, and N. Sangwan (2022): “Social Networks, Gender Norms and Women’s Labor Supply: Experimental Evidence Using a Job Search Platform,” IZA DP No. 15767. Ajayi, K., S. Das, C. Delavallade, T. Ketema, and L. Rouanet (2022): “Gender Differences in Socio-Emotional Skills and Economic Outcomes: New Evidence from 17 African Countries,” World Bank Policy Research Working Paper 10197. Allemand, M., M. Kirchberger, S. Milusheva, C. Newman, B. Roberts, and V. Thorne (2023): “Conscientiousness and Labor Market Returns: Evidence from a Field Experiment in West Africa,” World Bank Policy Research working paper 1010378. Impact Evaluation series Wash- ington, D.C. Ashraf, N., N. Bau, C. Low, and K. McGinn (2020): “Negotiating a better future: How inter- personal skills facilitate intergenerational investment,” The Quarterly Journal of Economics, 135, 1095–1151. Barrera-Osorio, F., A. Kugler, and M. Silliman (2020): “Hard and Soft Skills in Vocational Training: Experimental Evidence from Colombia,” Working Paper 27548, National Bureau of Economic Research. Borghans, L., A. L. Duckworth, J. J. Heckman, and B. Ter Weel (2008): “The economics and psychology of personality traits,” Journal of human Resources, 43, 972–1059. Cai, J. and A. Szeidl (2017): “Interfirm Relationships and Business Performance,” The Quarterly Journal of Economics, 133, 1229–1282. Campos, F., M. Frese, M. Goldstein, L. Iacovone, H. C. Johnson, D. McKenzie, and M. Mensmann (2017): “Teaching personal initiative beats traditional training in boosting small business in West Africa,” Science, 357, 1287–1290. Carranza, E., R. Garlick, K. Orkin, and N. Rankin (2022): “Job Search and Hiring with Limited Information about Workseekers’ Skills,” American Economic Review, 112, 3547–3583. Carranza, E. and D. McKenzie (2023): “Job Training and Job Search Assistance Policies in Developing Countries,” Working Paper. ˜ Cunningham, W. V. and P. Villasenor (2016): “Employer voices, employer demands, and im- plications for public skills development policy connecting the labor and education sectors,” The World Bank Research Observer, 31, 102–134. 16 Danon, A., J. Das, A. De Barros, and D. Filmer (2023): “Cognitive and Socioemotional Skills in Low-Income Countries: Measurement and Associations with Schooling and Earnings,” World Bank Policy Research Working Paper 10309. Deming, D. (2017): “The Growing Importance of Social Skills in the Labor Market,” Quarterly Journal of Economics, 132, 1593–1640. Deming, D. J. (2022): “Four facts about human capital,” Journal of Economic Perspectives, 36, 75– 102. Dimitriadis, S. and R. Koning (2022): “Social skills improve business performance: evidence from a randomized control trial with entrepreneurs in Togo,” Management Science, 68, 8635– 8657. Groh, M., N. Krishnan, D. McKenzie, and T. Vishwanath (2016): “The impact of soft skills training on female youth employment: evidence from a randomized experiment in Jordan,” IZA Journal of Labor & Development, 5. Heckman, J. J. and T. Kautz (2012): “Hard evidence on soft skills,” Labour economics, 19, 451–464. Heinrich, C. and S. Housemann (2021): “Worker Hard and Soft Skills and Labor Market Out- comes: A Lens through the Temporary Help Industry,” Working Paper. Vanderbilt University. International Monetary Fund (2021): “Policy Responses to COVID 19,” Tech. rep., Interna- tional Monetary Fund. Laajaj, R. and K. Macours (2021): “Measuring Skills in Developing Countries,” Journal of Human Resources, 56, 1254–1295. Le Barbanchon, T., D. Ubfal, and F. Araya (2023): “The Effects of Working While in School: Evidence from Employment Lotteries,” American Economic Journal: Applied Economics, 15, 383– 410. Osman, A. and J. Speer (2022): “Are Soft Skills Enough? Experimental Evidence on Skill Complementary for College Graduates,” Working PAper. Available at SSRN: https://ssrn.com/abstract=417856. Rwandan National Institute of Statistics (2020-2022): “Quarterly Rwanda Labour Force Survey Reports,” Tech. rep., National Institute of Statistics of Rwanda. ——— (2021): “Rwanda demographic and health survey 2019-20.” Tech. rep., National Institute of Statistics of Rwanda - NISR, Ministry of Health - MOH, and ICF. ——— (2022): “Rwanda Labour Force Survey,” Tech. rep., National Institute of Statistics of Rwanda. Sandvik, J. J., R. E. Saouma, N. T. Seegert, and C. T. Stanton (2020): “Workplace Knowledge Flows,” The Quarterly Journal of Economics, 135, 1635–1680. Ubfal, D., I. Arraiz, D. Beuermann, M. Frese, A. Maffioli, and D. Verch (2022): “The Impact of Soft-Skills Training for Entrepreneurs in Jamaica,” World Developmet, 152, 105787. 17 Vega-Redondo, F., P. Pin, D. Ubfal, C. Benedetti-Fasil, C. Brummitt, G. Rubera, D. Hovy, and T. Fornaciari (2019): “Peer networks and entrepreneurship: A pan-African RCT,” IZA DP No. 12848. Wanberg, C. R., E. A. van Hooft, S. Liu, and B. Csillag (2020): “Can job seekers achieve more through networking? The role of networking intensity, self-efficacy, and proximal benefits,” Personnel Psychology, 73, 559–585. 18 5 Tables and Figures Table 1: Impacts on Earnings and Work (1) (2) (3) (4) (5) (6) Earned Income Main Job Earnings Main Job Hours Worked Last 7 Days Last 7 Days (USD) Last 7 days Offered Training 0.06∗∗∗ 0.13∗∗∗ 5.61∗∗ 9.84∗∗∗ 3.41∗∗∗ 3.54∗∗ (0.02) (0.03) (2.83) (3.76) (1.04) (1.47) Offered * Round 2 -0.08∗∗ -3.37 -0.85 (0.04) (4.81) (1.96) Offered * Round 3 -0.15∗∗∗ -9.36∗ 0.44 (0.04) (4.79) (1.93) Control Mean 0.69 34.90 27.65 Control Mean R1 0.61 29.44 23.90 Control Mean R2 0.69 35.96 29.67 Control Mean R3 0.75 39.38 29.44 Observations 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.07 0.12 0.09 p-value Round 3 0.53 0.90 0.01 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. Table 2: Impacts on Job Quality (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Area Contract Contract Job Registered Studies Written Permanent Benefits Firm Offered Training 0.01 0.01 0.04∗ 0.04 0.03∗∗ -0.00 0.01 -0.00 0.06∗∗ 0.07∗∗ (0.02) (0.03) (0.02) (0.03) (0.01) (0.02) (0.02) (0.02) (0.02) (0.03) Offered * Round 2 0.02 -0.00 0.05∗∗ 0.03 -0.03 (0.04) (0.04) (0.03) (0.03) (0.04) Offered * Round 3 -0.01 0.02 0.05∗ 0.00 -0.01 (0.04) (0.04) (0.03) (0.03) (0.04) Control Mean 0.44 0.29 0.08 0.15 0.41 Control Mean R1 0.39 0.23 0.09 0.07 0.35 Control Mean R2 0.45 0.33 0.07 0.16 0.44 Control Mean R3 0.48 0.33 0.10 0.21 0.43 Observations 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.32 0.31 0.01 0.37 0.21 p-value Round 3 0.98 0.11 0.02 0.95 0.05 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow- up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. Job Benefits is a dummy variable equal to one for employees that receive contributions to social insurance, health insurance and paid holidays. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 19 Table 3: Impacts on Socio-Emotional Skills (1) (2) (3) Aggregate soft-skills Intrapersonal Interpersonal index skills index skills index Offered Training 0.03 0.04 0.03 (0.03) (0.03) (0.04) Baseline soft-skills index 0.51∗∗∗ (0.05) Baseline intrapersonal index 0.44∗∗∗ (0.04) Baseline interpersonal index 0.50∗∗∗ (0.05) Control Mean -0.02 -0.02 -0.02 Observations 849 849 849 Notes: OLS regressions controlling for strata dummies used in the randomization and baseline value of the outcome. Soft skills were only measured at baseline and endline. Robust standard errors shown in parenthesis. Intrapersonal skills include emotional awareness, self awareness, emotional regulation, self control, perseverance, personal initiative, and problem solving. Interpersonal skills include listening, empathy, expressiveness, relatedness, influence, negotiation, and collaboration. Table 4: Impacts on Search (1) (2) (3) (4) (5) (6) (7) (8) Searched For Job Confidence Approaching CV in People Discusses in Last Month Employers (0-3) Digital Format Jobs with Offered Training 0.01 -0.03 0.06∗ 0.08∗ 0.03∗∗ 0.04∗∗ 0.91∗∗∗ 1.07∗∗∗ (0.03) (0.03) (0.03) (0.05) (0.01) (0.02) (0.18) (0.23) Offered * Round 2 0.00 -0.01 -0.02 -0.25 (.) (0.06) (0.02) (0.25) Offered * Round 3 0.08∗∗ -0.06 -0.01 -0.22 (0.04) (0.06) (0.02) (0.27) Control Mean 0.64 2.52 0.90 5.42 Control Mean R1 . 2.52 0.89 5.51 Control Mean R2 0.66 2.50 0.89 5.50 Control Mean R3 0.62 2.53 0.92 5.26 Observations 1,728 1,728 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.40 0.13 0.38 0.00 p-value Round 3 0.08 0.64 0.09 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, dummies for stratification and survey wave dummies. Standard errors are clustered at the respondent level. 20 Table 5: Impacts on Networks (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) Family Friends Social Media Schoolmates Train mates Offered Training 0.07∗∗∗ 0.03 0.06∗∗∗ 0.06∗∗ 0.11∗∗∗ 0.11∗∗∗ 0.10∗∗∗ 0.10∗∗∗ 0.30∗∗∗ 0.39∗∗∗ (0.02) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) (0.03) (0.02) (0.03) Offered * Round 2 0.07∗ 0.04 -0.02 0.01 -0.15∗∗∗ (0.04) (0.04) (0.04) (0.04) (0.04) Offered * Round 3 0.06 -0.05 0.01 0.00 -0.13∗∗∗ (0.04) (0.04) (0.04) (0.04) (0.04) Control Mean 0.42 0.71 0.44 0.52 0.17 Control Mean R1 0.43 0.72 0.47 0.51 0.17 Control Mean R2 0.40 0.66 0.43 0.51 0.15 Control Mean R3 0.44 0.76 0.42 0.54 0.19 Observations 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.01 0.00 0.01 0.00 0.00 p-value Round 3 0.01 0.78 0.00 0.00 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. The dependent variable measures whether the participant dis- cussed job opportunities with each of the respective groups. We control for the baseline value of the dependent variable, dummies for stratification and survey wave dummies. Standard errors are clustered at the respondent level. 21 6 Appendix 1: Additional Tables and Figures 22 Figure A1: Soft-skills training characteristics Figure A1: Soft-skills training characteristics Delivery and costs of delivery Length 160 hours (102 lecture hours, 58 practical/assignment hours) Costs per USD 622 per intended participant, USD 835 per attendee participant Methodology Action-oriented methodology (lectures, individual and group exercises, presentations including subsequent feedback). Language English Logistics and attendance Venue University of Rwanda Huye campus Groups 9 groups Size of groups Average of 38 students per group, with 2 teachers per group Spacing of Classes ran every day from 7:30am – 5:30pm classes Practical exercises and assignment were interspersed and after regular class hours Attending at 74% least 1 class Attending at 72% least 10 classes Content Content The curriculum was developed by the University of Rwanda staff tasked with delivering development the training. The trainers were split into 3 groups with each group responsible for developing one of the three modules. Each module was subsequently validated with the rest of the group before finalized into the training. Content: • Intro to communication skills for work readiness Each sub-module communication • Types of communication skills comprised: skills • Channels of communication 8 lecture hours • Communication barriers 4 practical hours Content: • Rationale of professional networking Each sub-module networking • Networking strategies comprised: skills • Building and maintaining professional networks 8 lecture hours • Job search with people and internet research engines 4.5 practical hours Content: • Intrapersonal: Each sub-module intrapersonal • Personal Initiative and Perseverance comprised: and • Goal setting, adaptability, and attitude 6 lecture hours interpersonal • Problem solving, critical and creative thinking 3.5 practical hours skills • Interpersonal: • Teamwork and collaboration • Trust building and dependability Empathy and negotiation 23 Figure A1: Soft-skills training characteristics (cont.) Trainers Trainers 17 experienced University of Rwanda lecturers, all with at least a master’s degree Training of The trainers co-developed the curriculum before delivering the training. trainers Selection Selected by the University of Rwanda based on the following eligibility criteria: criteria for • Team leader and deputy team leader: trainers • At least masters’ degree with four years of professional experience in delivering professional training and module development • Training facilitators: • Must have at least a minimum of masters' degree or bachelors' degree with professional certification in related soft skills. to be proven by certificates • Should have at least 3 years' practical working experience in training, coaching, capacity building in areas of soft skills; to be proved by certificates • Having specific practical working experience for each training module proposed • A proof of at least 3 similar good assignment completed; to be proved by certificates • Fluent in English 24 Figure A2: Study Timeline 25 Table A1: Balance check (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Socio-Demographic Female 0.50 0.50 0.50 0.50 0.32 Married 0.11 0.32 0.14 0.34 0.32 Has Children 0.12 0.33 0.12 0.32 0.74 Age 27.50 2.83 27.55 3.08 0.71 Province: East 0.08 0.27 0.09 0.28 0.64 Province: Kigali 0.49 0.50 0.44 0.50 0.12 Province: North 0.14 0.35 0.17 0.37 0.28 Province South 0.16 0.37 0.20 0.40 0.07 Province West 0.13 0.34 0.11 0.31 0.21 Mother is employee 0.04 0.21 0.06 0.23 0.42 Father is employee 0.07 0.25 0.07 0.25 0.98 Wealth Index -0.07 1.43 0.05 1.43 0.19 Panel B. Education Years since graduation 1.92 0.92 1.86 0.95 0.32 High Education Diploma 0.21 0.41 0.20 0.40 0.75 Bachelor Degree 0.77 0.42 0.77 0.42 0.83 Masters Graduate 0.02 0.14 0.02 0.15 0.78 Public School 0.74 0.44 0.76 0.43 0.44 Good English Language 0.97 0.17 0.97 0.16 0.72 Good French Language 0.44 0.50 0.44 0.50 0.96 Good Swahili Language 0.14 0.35 0.13 0.34 0.73 Panel C. Employment Has ever worked 0.84 0.37 0.86 0.35 0.55 Not paid job last month 0.59 0.49 0.60 0.49 0.79 Works as employee 0.10 0.30 0.09 0.28 0.43 Works as self-employed 0.04 0.21 0.06 0.25 0.18 Works as family employee 0.09 0.28 0.08 0.28 0.79 Works as casual employee 0.18 0.38 0.17 0.37 0.73 Has formal job 0.03 0.18 0.02 0.14 0.21 Job accordance to degree 0.31 0.46 0.27 0.45 0.13 Labor earnings (monthly USD) 43.05 82.27 35.59 78.00 0.14 Panel D. Training Willingness to pay for course (USD) 173.17 971.41 140.21 333.41 0.51 Training: get job faster 0.58 0.49 0.56 0.50 0.60 Training: get higher pay job 0.29 0.45 0.27 0.45 0.54 Training: get better job 0.35 0.48 0.39 0.49 0.23 Prefers to attend Kigali 0.66 0.47 0.67 0.47 0.33 Prefers to attend North 0.19 0.39 0.18 0.38 0.12 Prefers to attend South 0.15 0.36 0.15 0.36 0.90 Panel D. Joint test for panels A-D p-value (joint F-test) 0.39 Observation 450 450 900 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. We do not test for differences in means for females since the lottery was randomized for each gender separately. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 26 Table A2: Balance check: Follow-up 1 (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Socio-Demographic Female 0.50 0.50 0.51 0.50 0.32 Married 0.11 0.32 0.14 0.35 0.24 Has Children 0.12 0.33 0.12 0.32 0.89 Age 27.53 2.84 27.57 3.09 0.65 Province: East 0.07 0.26 0.09 0.28 0.39 Province: Kigali 0.49 0.50 0.44 0.50 0.09 Province: North 0.14 0.35 0.17 0.37 0.30 Province South 0.16 0.37 0.20 0.40 0.07 Province West 0.13 0.34 0.10 0.30 0.18 Mother is employee 0.04 0.21 0.06 0.23 0.46 Father is employee 0.06 0.25 0.07 0.25 0.86 Wealth Index -0.07 1.44 0.04 1.44 0.25 Panel B. Education Years since graduation 1.94 0.91 1.87 0.95 0.32 High Education Diploma 0.21 0.41 0.20 0.40 0.86 Bachelor Degree 0.77 0.42 0.78 0.42 0.87 Masters Graduate 0.02 0.14 0.02 0.14 0.98 Public School 0.74 0.44 0.76 0.43 0.43 Good English Language 0.97 0.18 0.97 0.16 0.72 Good French Language 0.43 0.50 0.43 0.50 0.92 Good Swahili Language 0.14 0.35 0.12 0.33 0.53 Panel C. Employment Has ever worked 0.84 0.37 0.86 0.35 0.45 Not paid job last month 0.59 0.49 0.60 0.49 0.82 Works as employee 0.10 0.30 0.08 0.28 0.33 Works as self-employed 0.05 0.21 0.06 0.24 0.28 Works as family employee 0.08 0.28 0.08 0.28 0.95 Works as casual employee 0.18 0.38 0.17 0.38 0.80 Has formal job 0.03 0.18 0.02 0.13 0.18 Job accordance to degree 0.32 0.47 0.27 0.45 0.13 Labor earnings (monthly USD) 42.99 81.69 34.69 76.01 0.12 Panel D. Training Willingness to pay for course (USD) 173.18 989.14 135.13 318.61 0.46 Training: get job faster 0.59 0.49 0.56 0.50 0.45 Training: get higher pay job 0.30 0.46 0.28 0.45 0.55 Training: get better job 0.35 0.48 0.39 0.49 0.20 Prefers to attend Kigali 0.67 0.47 0.67 0.47 0.48 Prefers to attend North 0.19 0.39 0.18 0.38 0.16 Prefers to attend South 0.15 0.35 0.16 0.36 0.73 Panel D. Joint test for panels A-D p-value (joint F-test) 0.40 Observation 433 436 869 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. We do not test for differences in means for females since the lottery was randomized for each gender separately. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 27 Table A3: Balance check: Follow-up 2 (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Socio-Demographic Female 0.51 0.50 0.50 0.50 0.32 Married 0.12 0.32 0.14 0.35 0.37 Has Children 0.13 0.33 0.12 0.32 0.72 Age 27.51 2.84 27.60 3.09 0.61 Province: East 0.08 0.27 0.08 0.28 0.87 Province: Kigali 0.49 0.50 0.44 0.50 0.15 Province: North 0.14 0.35 0.17 0.37 0.33 Province South 0.16 0.37 0.20 0.40 0.05 Province West 0.13 0.34 0.11 0.31 0.26 Mother is employee 0.04 0.20 0.05 0.22 0.51 Father is employee 0.06 0.24 0.07 0.25 0.69 Wealth Index -0.10 1.41 0.04 1.42 0.15 Panel B. Education Years since graduation 1.92 0.92 1.86 0.95 0.33 High Education Diploma 0.21 0.41 0.20 0.40 0.72 Bachelor Degree 0.77 0.42 0.77 0.42 0.78 Masters Graduate 0.02 0.14 0.02 0.15 0.85 Public School 0.74 0.44 0.76 0.43 0.46 Good English Language 0.97 0.18 0.97 0.16 0.72 Good French Language 0.43 0.50 0.43 0.50 0.83 Good Swahili Language 0.14 0.34 0.13 0.33 0.65 Panel C. Employment Has ever worked 0.84 0.37 0.86 0.35 0.43 Not paid job last month 0.60 0.49 0.60 0.49 0.73 Works as employee 0.10 0.30 0.09 0.28 0.36 Works as self-employed 0.05 0.21 0.06 0.24 0.31 Works as family employee 0.08 0.27 0.08 0.27 0.93 Works as casual employee 0.18 0.38 0.17 0.37 0.68 Has formal job 0.03 0.18 0.02 0.14 0.27 Job accordance to degree 0.32 0.47 0.27 0.45 0.11 Labor earnings (monthly USD) 43.14 81.93 35.12 78.29 0.13 Panel D. Training Willingness to pay for course (USD) 174.95 992.66 140.98 338.18 0.50 Training: get job faster 0.59 0.49 0.57 0.50 0.61 Training: get higher pay job 0.30 0.46 0.27 0.45 0.41 Training: get better job 0.34 0.47 0.39 0.49 0.20 Prefers to attend Kigali 0.67 0.47 0.66 0.47 0.45 Prefers to attend North 0.19 0.39 0.18 0.39 0.16 Prefers to attend South 0.15 0.35 0.16 0.36 0.79 Panel D. Joint test for panels A-D p-value (joint F-test) 0.46 Observation 430 435 865 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. We do not test for differences in means for females since the lottery was randomized for each gender separately. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 28 Table A4: Balance check: Follow-up 3 (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Socio-Demographic Female 0.49 0.50 0.49 0.50 0.32 Married 0.11 0.32 0.14 0.34 0.34 Has Children 0.13 0.33 0.12 0.32 0.70 Age 27.56 2.84 27.57 3.10 0.75 Province: East 0.08 0.27 0.08 0.28 0.69 Province: Kigali 0.48 0.50 0.43 0.50 0.14 Province: North 0.14 0.35 0.17 0.37 0.36 Province South 0.17 0.37 0.21 0.41 0.06 Province West 0.13 0.34 0.11 0.31 0.26 Mother is employee 0.04 0.21 0.05 0.22 0.68 Father is employee 0.06 0.24 0.06 0.24 0.93 Wealth Index -0.06 1.44 0.03 1.42 0.33 Panel B. Education Years since graduation 1.93 0.92 1.84 0.96 0.19 High Education Diploma 0.22 0.41 0.21 0.41 0.62 Bachelor Degree 0.76 0.43 0.77 0.42 0.69 Masters Graduate 0.02 0.14 0.02 0.15 0.81 Public School 0.74 0.44 0.75 0.43 0.72 Good English Language 0.97 0.17 0.97 0.16 0.83 Good French Language 0.44 0.50 0.43 0.50 0.84 Good Swahili Language 0.14 0.35 0.13 0.33 0.56 Panel C. Employment Has ever worked 0.84 0.37 0.85 0.35 0.65 Not paid job last month 0.59 0.49 0.61 0.49 0.62 Works as employee 0.11 0.31 0.08 0.28 0.24 Works as self-employed 0.04 0.20 0.06 0.25 0.15 Works as family employee 0.08 0.28 0.08 0.28 0.95 Works as casual employee 0.18 0.38 0.16 0.37 0.58 Has formal job 0.03 0.18 0.02 0.14 0.26 Job accordance to degree 0.32 0.47 0.28 0.45 0.11 Labor earnings (monthly USD) 43.41 82.61 33.98 75.02 0.07 Panel D. Training Willingness to pay for course (USD) 177.53 996.74 142.08 338.15 0.50 Training: get job faster 0.58 0.49 0.57 0.50 0.74 Training: get higher pay job 0.30 0.46 0.28 0.45 0.49 Training: get better job 0.35 0.48 0.39 0.49 0.26 Prefers to attend Kigali 0.66 0.47 0.66 0.47 0.45 Prefers to attend North 0.19 0.39 0.18 0.38 0.16 Prefers to attend South 0.15 0.36 0.16 0.37 0.75 Panel D. Joint test for panels A-D p-value (joint F-test) 0.50 Observation 427 436 863 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. We do not test for differences in means for females since the lottery was randomized for each gender separately. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 29 Table A5: Attrition (1) (2) (3) Attrited from Attrited from Attrited from 4-month follow-up 10-month follow-up 15-month follow-up Offered Training 0.01 0.01 0.02 (0.01) (0.01) (0.01) Observations 900 900 900 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. We do not test for differences in means for females since the lottery was randomized for each gender separately. 30 Table A6: Balance check. Soft Skills (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Likert Scales Emotional Awareness -0.01 0.66 0.01 0.66 0.56 Self Awareness 0.01 0.64 -0.01 0.64 0.73 Emotional Regulation 0.01 0.61 -0.01 0.58 0.69 Self Control -0.01 0.68 0.01 0.63 0.53 Perseverance 0.01 0.66 -0.01 0.66 0.55 Personal Initiative 0.00 0.61 -0.00 0.66 0.93 Problem Solving -0.03 0.63 0.03 0.58 0.17 Listening -0.00 0.61 0.00 0.60 0.84 Empathy -0.01 0.57 0.01 0.61 0.53 Expressiveness 0.01 0.72 -0.01 0.71 0.74 Relatedness 0.01 0.63 -0.01 0.64 0.67 Influence -0.00 0.74 0.00 0.66 0.99 Negotiation -0.01 0.68 0.01 0.70 0.69 Collaboration -0.02 0.63 0.02 0.66 0.35 Panel B. Alternative Measures Self Awareness -0.03 0.51 0.03 0.54 0.10 Emotional Regulation 0.00 0.40 -0.02 0.48 0.46 Self Control 0.03 0.57 -0.04 0.73 0.15 Perseverance 0.00 0.55 -0.02 0.63 0.61 Personal Initiative -0.01 0.57 -0.01 0.68 0.95 Problem Solving -0.02 0.63 -0.02 0.70 0.90 Networking -0.00 0.68 -0.07 0.79 0.18 Empathy -0.04 0.61 -0.04 0.72 0.85 Expressiveness 0.01 0.53 -0.05 0.67 0.13 Relatedness 0.01 0.58 -0.04 0.70 0.20 Influence -0.01 0.58 -0.07 0.92 0.21 Negotiation -0.00 0.64 -0.05 0.63 0.20 Collaboration -0.03 0.60 0.02 0.56 0.25 Panel B. Joint test for panels A-B p-value (joint F-test) 0.23 Observation 447 448 894 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an assigned to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 31 Table A7: Balance check. Soft Skills (Indexes 1-5) (1) (2) (3) (4) (5) Control Offered Mean S.D. Mean S.D. p-value Panel A. Likert Scales Emotional Awareness 3.91 0.48 3.93 0.49 0.65 Self Awareness 4.13 0.49 4.11 0.48 0.63 Emotional Regulation 4.01 0.47 3.99 0.45 0.54 Self Control 3.53 0.73 3.56 0.68 0.55 Perseverance 4.19 0.52 4.17 0.51 0.55 Personal Initiative 4.29 0.43 4.29 0.46 0.90 Problem Solving 4.14 0.48 4.18 0.44 0.19 Listening 4.09 0.50 4.10 0.48 0.80 Empathy 4.17 0.43 4.19 0.46 0.56 Expressiveness 4.02 0.54 4.01 0.53 0.73 Relatedness 4.23 0.46 4.22 0.47 0.61 Influence 4.00 0.53 4.01 0.48 0.99 Negotiation 4.06 0.50 4.08 0.51 0.63 Collaboration 4.30 0.43 4.33 0.45 0.36 Panel B. Joint test for panels A p-value (joint F-test) 0.42 Observation 447 448 895 Notes: the p-value reported in Column 5 is obtained from a regression of each variable on an as- signed to treatment dummy with standard errors robust to heteroskedasticity, controlling for lottery dummies. p-value (joint F-test): corresponds to the orthogonality test in a regression of assigned to treatment dummy on covariates, the regression also controls for lottery dummies. 32 Table A8: Correlates of Training Attendance (1) (2) (3) (4) (5) (6) (7) (8) Attend Training Total Attendance Selected for Kigali -0.13∗∗∗ -0.14∗∗∗ -0.07 -1.77∗∗ (0.05) (0.05) (0.04) (0.69) Selected for North -0.07 -0.13∗ -1.65∗ (0.06) (0.07) (0.90) Female 0.06 0.02 0.05 (0.04) (0.04) (0.56) Married -0.06 0.00 -0.09 (0.08) (0.08) (1.01) Has Children 0.01 0.00 0.06 (0.08) (0.09) (1.08) Age 0.01 0.01 0.01 0.10 (0.01) (0.01) (0.01) (0.09) Province: East -0.07 0.07 1.21 (0.08) (0.09) (1.16) Province: Kigali -0.10∗∗ -0.03 -0.03 -0.16 (0.05) (0.06) (0.05) (0.75) Province: North -0.07 0.01 0.21 (0.07) (0.07) (0.99) Province West 0.02 0.10 0.07 1.50∗ (0.07) (0.07) (0.06) (0.90) Wealth Index -0.01 0.00 0.10 (0.01) (0.01) (0.19) Years since graduation -0.03 -0.03 -0.03 -0.39 (0.02) (0.02) (0.02) (0.27) Bachelor Graduate 0.04 -0.00 -0.20 (0.05) (0.05) (0.66) Public School -0.03 -0.02 -0.38 (0.05) (0.05) (0.60) Good English Language -0.10 -0.06 -0.63 (0.11) (0.12) (1.55) Has ever worked -0.06 -0.04 -0.06 -0.65 (0.05) (0.05) (0.05) (0.70) Works as employee -0.21∗∗ -0.20∗∗ -0.17∗ -2.72∗∗ (0.09) (0.09) (0.09) (1.18) Works as self-employed -0.07 -0.06 -0.80 (0.09) (0.10) (1.25) Works as family employee 0.09 0.09 0.10 1.43∗ (0.06) (0.07) (0.06) (0.85) Works as casual employee -0.18∗∗∗ -0.18∗∗ -0.15∗∗ -2.21∗∗ (0.07) (0.07) (0.07) (0.96) Job accordance to degree -0.09 -0.10∗ -0.11∗∗ -1.33∗ (0.06) (0.06) (0.06) (0.74) Willingness to pay for course (USD) 0.00∗ 0.00∗ 0.00 (0.00) (0.00) (0.00) Training: get job faster 0.07 0.07 0.07∗ 0.89 (0.04) (0.04) (0.04) (0.56) Training: get higher pay job -0.01 0.01 0.22 (0.05) (0.05) (0.60) Training: get better job -0.01 0.02 0.24 (0.05) (0.04) (0.57) Observations 450 450 450 450 450 450 450 450 R-squared 0.014 0.022 0.008 0.080 0.011 0.118 0.104 0.115 Mean dep. var. 0.74 0.74 0.74 0.74 0.74 0.74 0.74 9.44 Offered Training is the coefficient from an OLS regression controlling for the randomization stratum fixed effects. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1 33 Table A9: Course Evaluations: Satisfaction and Knowledge Test (1) (2) Soft-skills training Mean SD Panel A. Course satisfaction (1-7) Satisfaction: content 6.44 1.17 Satisfaction: delivery 6.41 0.79 Satisfaction: exercises 6.44 1.07 Satisfaction: duration 5.80 1.31 Satisfaction: pace 5.36 1.43 Satisfaction: relevance 6.40 0.91 Satisfaction: overall 6.39 0.92 Likelihood to use the skills 6.76 0.68 Likelihood to recommend the course 6.74 0.62 Willingness to pay for course (1000s RWF) 218.72 685.84 Panel B. Knowledge test Correct answer: effective communication 0.79 0.41 Correct answer: CV length 0.65 0.48 Correct answer: communication style 0.63 0.48 Correct answer: networking strategy 0.88 0.33 Correct answer: job-search engine 0.23 0.42 Correct answer: SMART goal 0.63 0.48 Correct answer: teamwork skills 0.76 0.43 Observations 320 The evaluation was carried out during the last day of classes. Out of 335 students who attended at least one class of the course, 320 were present during the evaluation and complete it. Table A10: Results: Secondary Activity Details (1) (2) (3) (4) (5) (6) (7) (8) Has Secondary Work hours Worked Days Worked Hours Work Activity usual 7 days Last 7 days Last 7 days Offered Training 0.02 0.02 0.07 0.06 0.14∗ 0.19∗ 0.64 0.91 (0.02) (0.02) (0.54) (0.80) (0.08) (0.11) (0.55) (0.80) Offered * Round 2 0.01 -0.32 -0.05 -0.40 (0.03) (1.06) (0.15) (1.11) Offered * Round 3 -0.01 0.33 -0.11 -0.42 (0.03) (1.11) (0.15) (1.17) Control Mean 0.18 4.38 0.66 4.33 Control Mean R1 0.15 4.06 0.56 3.60 Control Mean R2 0.17 4.41 0.67 4.23 Control Mean R3 0.21 4.68 0.75 5.17 Observations 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.16 0.76 0.24 0.55 p-value Round 3 0.66 0.64 0.52 0.61 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 34 Table A11: Results: Income per hour (1) (2) Earned Income/Hours Worked Last 7 Days (USD) Offered Training 0.10 0.28 (0.19) (0.30) Offered * Round 2 -0.06 (0.37) Offered * Round 3 -0.47 (0.37) Control Mean 1.82 Control Mean R1 1.71 Control Mean R2 1.79 Control Mean R3 1.94 Observations 2,093 2,093 p-value Round 2 0.42 p-value Round 3 0.51 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being of- fered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table A12: Impacts on Occupational Choice (1) (2) (3) (4) (5) (6) (7) (8) Employee Apprentice/Intern Employer Self-Employed Offered Training 0.04 0.02 0.01 0.06∗∗ -0.00 -0.01 0.01 0.05∗ (0.02) (0.03) (0.02) (0.03) (0.01) (0.01) (0.02) (0.03) Offered * Round 2 0.03 -0.06∗ -0.00 -0.05 (0.04) (0.04) (0.02) (0.03) Offered * Round 3 0.03 -0.08∗∗ 0.02 -0.08∗∗ (0.04) (0.04) (0.02) (0.04) Control Mean 0.36 0.22 0.03 0.18 Control Mean R1 0.27 0.20 0.03 0.18 Control Mean R2 0.39 0.23 0.04 0.15 Control Mean R3 0.41 0.22 0.02 0.21 Observations 2,597 2,597 2,597 2,597 2,597 2,597 2,597 2,597 p-value Round 2 0.20 0.90 0.40 0.94 p-value Round 3 0.16 0.37 0.28 0.20 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 35 Table A13: Registered for social security (1) Ever Had Formal Job Offered Training 0.02 (0.03) Control Mean 0.62 Observations 900 Notes: Offered Training is the coeffi- cient from a regression of an indicator for whether the respondent is registered for so- cial security (meaning that she had a for- mal job at some point in her life) 17 months after the training on an indicator for being offered the training, controlling for stratifi- cation dummies. Table A14: Results: Mechanisms - Intrapersonal Socio-Emotional Skills (1) (2) (3) (4) (5) (6) (7) Emotional Self Emotional Self Perseverance Personal Problem Awareness Awareness Regulation Control Initiative Solving Offered Training 0.06 0.05 0.06 0.02 0.03 0.04 0.03 (0.05) (0.04) (0.04) (0.05) (0.04) (0.04) (0.04) Emotional Awareness 0.33∗∗∗ (0.04) Self Awareness 0.36∗∗∗ (0.04) Emotional Regulation 0.31∗∗∗ (0.04) Self Control 0.35∗∗∗ (0.04) Perseverance 0.37∗∗∗ (0.04) Personal Initiative 0.35∗∗∗ (0.05) Problem Solving 0.34∗∗∗ (0.05) Control Mean -0.03 -0.02 -0.03 -0.02 -0.01 -0.02 -0.02 Observations 849 849 849 849 849 849 849 Notes: OLS regressions controlling for strata dummies used in the randomization and baseline value of the outcome. Soft skills were only measured at baseline and endline. Robust standard errors shown in parenthesis. The controls presented in the table are the baseline measures of each index. 36 Table A15: Results: Mechanisms - Interpersonal Socio-Emotional Skills (1) (2) (3) (4) (5) (6) (7) Listening Empathy Expresiveness Relatedness Influence Negotiation Collaboration Offered Training 0.08∗ -0.01 0.06 0.03 0.06 0.03 0.04 (0.04) (0.04) (0.05) (0.04) (0.04) (0.05) (0.05) Listening 0.34∗∗∗ (0.04) Empathy 0.34∗∗∗ (0.05) Expressiveness 0.36∗∗∗ (0.04) Relatedness 0.41∗∗∗ (0.05) Influence 0.41∗∗∗ (0.04) Negotiation 0.31∗∗∗ (0.04) Collaboration 0.41∗∗∗ (0.04) Control Mean -0.04 0.00 -0.03 -0.01 -0.03 -0.01 -0.03 Observations 849 849 849 849 849 849 849 Notes: OLS regressions controlling for strata dummies used in the randomization and baseline value of the outcome. Soft skills were only measured at baseline and endline. Robust standard errors shown in parenthesis. The controls presented in the table are the baseline measures of each index. Table A16: Gender and Socio-Emotional Skills (1) (2) (3) Aggregate Intrapersonal Interpersonal SES index SES index SES index Gender gap (female) -0.08∗∗∗ -0.06∗ -0.10∗∗∗ (0.03) (0.03) (0.03) Observations 895 895 895 Notes: Gender gap is an indicator variable for women. *** p<0.01, ** p<0.05, * p<0.1 Robust standard errors reported. 37 Table A17: Gender, soft skills, and employment Gender gap Obs No controls + degree type + work status Aggregate indices Aggregate SES Index 895 -0.08∗∗ -0.10∗∗∗ -0.09∗∗ (0.03) (0.03) (0.03) Intrapersonal 895 -0.06∗ -0.08∗∗ -0.07∗∗ (0.03) (0.03) (0.03) Interpersonal 895 -0.10∗∗∗ -0.12∗∗∗ -0.11∗∗∗ (0.03) (0.04) (0.04) Intrapersonal skills Emotional Awareness 895 -0.01 -0.01 -0.01 (0.04) (0.05) (0.05) Self Awareness 895 -0.13∗∗∗ -0.14∗∗∗ -0.14∗∗∗ (0.04) (0.04) (0.04) Emotional Regulation 895 -0.05 -0.05 -0.05 (0.04) (0.04) (0.04) Self Control 895 -0.03 -0.07 -0.07 (0.04) (0.04) (0.04) Perseverance 895 -0.07 -0.08∗ -0.08 (0.04) (0.05) (0.05) Personal Initiative 895 -0.09∗∗ -0.11∗∗ -0.11∗∗ (0.04) (0.05) (0.05) Problem Solving 895 -0.04 -0.05 -0.05 (0.04) (0.04) (0.04) Interpersonal skills Listening 895 -0.08∗ -0.10∗∗ -0.11∗∗ (0.04) (0.04) (0.04) Empathy 895 -0.08∗ -0.10∗∗ -0.09∗∗ (0.04) (0.04) (0.04) Expressiveness 895 -0.16∗∗∗ -0.15∗∗∗ -0.16∗∗∗ (0.05) (0.05) (0.05) Relatedness 895 -0.11∗∗ -0.11∗∗ -0.11∗∗ (0.04) (0.05) (0.05) Influence 895 -0.13∗∗ -0.13∗∗ -0.12∗∗ (0.05) (0.05) (0.05) Negotiation 895 -0.14∗∗∗ -0.15∗∗∗ -0.15∗∗∗ (0.05) (0.05) (0.05) Collaboration 895 -0.08∗ -0.09∗∗ -0.09∗ (0.04) (0.05) (0.05) Note: Each coefficient is a result from a separate regression of a soft-skill index on an indicator variable for female. Robust standard errors reported in parenthesis. *, **, and *** indicate significance at the 90, 95, and 99% confidence intervals. 38 Table A18: Gender, labor, and socio-emotional skills (1) (2) (3) (4) (5) (6) Worked in last 30 days Labor Income Woman -0.08∗∗ -0.08∗∗ -0.08∗∗ -18.54∗∗∗ -18.73∗∗∗ -18.44∗∗∗ (0.03) (0.03) (0.03) (4.80) (4.82) (4.78) Agg. SES Index 0.05 10.62 (0.05) (8.80) Woman x Agg. SES -0.06 -8.45 (0.07) (10.93) Intrapersonal SES 0.04 7.69 (0.05) (8.87) Woman x Intra SES -0.05 -4.57 (0.07) (11.06) Interpersonal SES 0.04 10.92 (0.05) (7.64) Woman x Inter SES -0.05 -10.00 (0.06) (9.36) Mean for men 0.45 0.45 0.45 47.59 47.59 47.59 Observations 895 895 895 895 895 895 p-value Women 0.87 0.94 0.82 0.74 0.64 0.87 Notes: Gender gap is an indicator variable for women. *** p<0.01, ** p<0.05, * p<0.1 Robust standard errors reported. Table A19: Results: Aggregate Soft Skills by English Ability (1) (2) (3) Aggregate soft-skills Intrapersonal Interpersonal index skills index skills index Offered Training 0.01 0.03 -0.01 (0.05) (0.05) (0.06) Offered * Advanced English 0.05 0.02 0.08 (0.06) (0.06) (0.07) Aggregate SES Index 0.51∗∗∗ (0.05) Intrapersonal 0.44∗∗∗ (0.04) Interpersonal 0.50∗∗∗ (0.05) Control Mean -0.02 -0.02 -0.02 Observations 849 849 849 Notes: OLS regressions controlling for strata dummies used in the randomization and baseline value of the outcome. Soft skills were only measured at baseline and endline. Robust standard errors shown in parenthesis. The controls presented in the table are the baseline measures of each index. 39 7 Appendix 2: Gender Heterogeneity Table B1: Results Pooled: Earnings and Work - Heterogeneity by gender (1) (2) (3) Earned Income Main Job Earnings Main Job Hours Worked Last 7 Days Last 7 Days (USD) Last 7 days Offered Training 0.04 8.34∗ 3.01∗∗ (0.03) (4.46) (1.49) Offered * Female 0.04 -5.40 0.75 (0.04) (5.65) (2.08) Control Mean: Men 0.77 40.27 30.34 Control Mean: Women 0.60 29.57 24.99 Observations 2,597 2,597 2,597 p-value female 0.02 0.40 0.01 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. Table B2: Results Pooled: Work Categories - Heterogeneity by gender (1) (2) (3) (4) Employee Apprentice/Intern Employer Self-Employed Offered Training 0.02 0.00 -0.01 0.01 (0.03) (0.03) (0.01) (0.02) Offered * Female 0.03 0.02 0.01 -0.01 (0.05) (0.04) (0.01) (0.03) Control Mean: Men 0.42 0.22 0.04 0.17 Control Mean: Women 0.30 0.21 0.03 0.19 Observations 2,597 2,597 2,597 2,597 p-value female 0.11 0.49 0.92 0.89 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 40 Table B3: Results Pooled: Job Benefits - Heterogeneity by gender (1) (2) (3) (4) (5) Area Contract Contract Job Registered Studies Written Permanent Benefits Firm Offered Training -0.00 0.04 0.03 0.01 0.09∗∗∗ (0.04) (0.03) (0.02) (0.03) (0.03) Offered * Female 0.03 -0.00 0.01 -0.02 -0.06 (0.05) (0.05) (0.03) (0.04) (0.05) Control Mean: Men 0.53 0.32 0.10 0.18 0.42 Control Mean: Women 0.35 0.27 0.07 0.12 0.39 Observations 2,597 2,597 2,597 2,597 2,597 p-value female 0.40 0.23 0.03 0.94 0.41 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indi- cator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table B4: Results Pooled: Channels - Heterogeneity by gender (1) (2) (3) (4) Searched For Job Confidence Approaching CV in People Discusses in Last Month Employers (0-3) Digital Format Jobs with Offered Training 0.03 0.06 0.03 0.98∗∗∗ (0.04) (0.05) (0.02) (0.25) Offered * Female -0.04 -0.01 -0.00 -0.17 (0.05) (0.07) (0.03) (0.35) Control Mean: Men 0.64 2.51 0.88 5.47 Control Mean: Women 0.64 2.53 0.92 5.37 Observations 1,728 2,597 2,597 2,597 p-value female 0.91 0.23 0.12 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow- up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. 41 Table B5: Results: Network Details by Gender (1) (2) (3) (4) (5) Family Friends Social Media Schoolmates Train mates Offered Training 0.13∗∗∗ 0.10∗∗∗ 0.12∗∗∗ 0.14∗∗∗ 0.32∗∗∗ (0.04) (0.03) (0.03) (0.03) (0.03) Offered * Female -0.11∗∗ -0.08∗∗ -0.03 -0.08 -0.05 (0.05) (0.04) (0.05) (0.05) (0.04) Control Mean: Men 0.37 0.71 0.46 0.52 0.17 Control Mean: Women 0.48 0.72 0.43 0.52 0.17 Observations 2,597 2,597 2,597 2,597 2,597 p-value female 0.67 0.57 0.02 0.09 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 42 8 Appendix 3: Additional Heterogeneity Table C1: Results Pooled: Earnings and Work - Heterogeneity by wealth (1) (2) (3) Earned Income Main Job Earnings Main Job Hours Worked Last 7 Days Last 7 Days (USD) Last 7 days Offered Training 0.04 3.61 2.63∗ (0.03) (3.95) (1.54) Offered * High Ubudehe 0.03 4.67 1.59 (0.04) (5.75) (2.11) Control Mean: High Ubudehe 0.72 34.95 29.79 Control Mean: Low Ubudehe 0.65 34.86 25.46 Observations 2,597 2,597 2,597 p-value high Ubudehe 0.02 0.04 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. Table C2: Results Pooled: Work Categories - Heterogeneity by wealth (1) (2) (3) (4) Employee Apprentice/Intern Employer Self-Employed Offered Training -0.00 0.04 -0.01 0.00 (0.03) (0.03) (0.01) (0.02) Offered * High Ubudehe 0.07 -0.05 0.02 0.01 (0.05) (0.04) (0.01) (0.03) Control Mean: High Ubudehe 0.40 0.21 0.04 0.19 Control Mean: Low Ubudehe 0.32 0.23 0.03 0.17 Observations 2,597 2,597 2,597 2,597 p-value high Ubudehe 0.03 0.53 0.43 0.74 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 43 Table C3: Results Pooled: Job Benefits - Heterogeneity by wealth (1) (2) (3) (4) (5) Area Contract Contract Job Registered Studies Written Permanent Benefits Firm Offered Training -0.00 0.02 0.04∗∗ -0.02 0.06∗ (0.04) (0.03) (0.02) (0.03) (0.03) Offered * High Ubudehe 0.04 0.05 -0.01 0.04 -0.00 (0.05) (0.05) (0.03) (0.04) (0.05) Control Mean: High Ubudehe 0.45 0.32 0.08 0.17 0.41 Control Mean: Low Ubudehe 0.43 0.27 0.09 0.13 0.41 Observations 2,597 2,597 2,597 2,597 2,597 p-value high Ubudehe 0.34 0.06 0.17 0.25 0.11 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table C4: Results Pooled: Channels - Heterogeneity by wealth (1) (2) (3) (4) Searched For Job Confidence Approaching CV in People Discusses in Last Month Employers (0-3) Digital Format Jobs with Offered Training 0.02 0.05 0.05∗∗ 0.96∗∗∗ (0.03) (0.05) (0.02) (0.25) Offered * High Ubudehe -0.01 0.02 -0.04 -0.11 (0.05) (0.07) (0.03) (0.35) Control Mean: High Ubudehe 0.66 2.52 0.87 5.49 Control Mean: Low Ubudehe 0.62 2.52 0.93 5.35 Observations 1,728 2,597 2,597 2,597 p-value high Ubudehe 0.80 0.13 0.53 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. 44 Table C5: Results: Network Details by wealth (1) (2) (3) (4) (5) Family Friends Social Media Schoolmates Train mates Offered Training 0.07∗∗ 0.07∗∗ 0.10∗∗∗ 0.15∗∗∗ 0.30∗∗∗ (0.03) (0.03) (0.04) (0.03) (0.03) Offered * High Ubudehe -0.01 -0.02 0.02 -0.10∗ -0.00 (0.05) (0.04) (0.05) (0.05) (0.04) Control Mean: High Ubudehe 0.39 0.71 0.45 0.50 0.16 Control Mean: Low Ubudehe 0.46 0.72 0.44 0.54 0.18 Observations 2,597 2,597 2,597 2,597 2,597 p-value high Ubudehe 0.06 0.08 0.00 0.15 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table C6: Results Pooled: Earnings and Work - Heterogeneity by Public/Private (1) (2) (3) Earned Income Main Job Earnings Main Job Hours Worked Last 7 Days Last 7 Days (USD) Last 7 days Offered Training 0.09∗∗ 3.65 5.87∗∗∗ (0.05) (5.71) (2.11) Offered * Public -0.05 3.18 -3.36 (0.05) (6.56) (2.42) Control Mean: Public 0.59 36.25 22.95 Control Mean: Private 0.72 34.43 29.31 Observations 2,597 2,597 2,597 p-value public 0.07 0.03 0.04 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. 45 Table C7: Results Pooled: Work Categories - Heterogeneity by Public/Private (1) (2) (3) (4) Employee Apprentice/Intern Employer Self-Employed Offered Training 0.00 0.00 0.02 0.05 (0.04) (0.04) (0.02) (0.03) Offered * Public 0.05 0.01 -0.03 -0.06 (0.05) (0.04) (0.02) (0.04) Control Mean: Public 0.34 0.19 0.03 0.15 Control Mean: Private 0.36 0.23 0.03 0.19 Observations 2,597 2,597 2,597 2,597 p-value public 0.08 0.58 0.20 0.60 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table C8: Results Pooled: Job Benefits - Heterogeneity by Public/Private (1) (2) (3) (4) (5) Area Contract Contract Job Registered Studies Written Permanent Benefits Firm Offered Training 0.04 -0.01 0.00 -0.03 0.03 (0.05) (0.05) (0.03) (0.03) (0.05) Offered * Public -0.04 0.07 0.04 0.04 0.04 (0.06) (0.05) (0.03) (0.04) (0.06) Control Mean: Public 0.36 0.30 0.11 0.16 0.39 Control Mean: Private 0.47 0.29 0.08 0.14 0.41 Observations 2,597 2,597 2,597 2,597 2,597 p-value public 0.91 0.04 0.00 0.39 0.01 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, strat- ification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 46 Table C9: Results Pooled: Channels - Heterogeneity by Public/Private (1) (2) (3) (4) Searched For Job Confidence Approaching CV in People Discusses in Last Month Employers (0-3) Digital Format Jobs with Offered Training 0.01 0.06 0.03 1.04∗∗∗ (0.05) (0.07) (0.03) (0.33) Offered * Public 0.00 0.01 -0.01 -0.18 (0.06) (0.08) (0.03) (0.39) Control Mean: Public 0.64 2.55 0.90 5.24 Control Mean: Private 0.64 2.51 0.90 5.49 Observations 1,728 2,597 2,597 2,597 p-value public 0.59 0.12 0.08 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow- up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. Table C10: Results: Network Details by Public/Private (1) (2) (3) (4) (5) Family Friends Social Media Schoolmates Train mates Offered Training 0.05 0.02 0.09∗ 0.05 0.34∗∗∗ (0.05) (0.04) (0.05) (0.05) (0.04) Offered * Public 0.03 0.05 0.02 0.07 -0.06 (0.06) (0.04) (0.06) (0.06) (0.05) Control Mean: Public 0.45 0.76 0.44 0.49 0.13 Control Mean: Private 0.41 0.70 0.44 0.53 0.18 Observations 2,597 2,597 2,597 2,597 2,597 p-value public 0.01 0.00 0.00 0.00 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 47 Table C11: Results Pooled: Earnings and Work - Heterogeneity by University (1) (2) (3) Earned Income Main Job Earnings Main Job Hours Worked Last 7 Days Last 7 Days (USD) Last 7 days Offered Training 0.06∗∗∗ 5.62∗ 3.58∗∗∗ (0.02) (3.11) (1.23) Offered * University degree -0.03 1.26 -0.69 (0.05) (7.01) (2.35) Control Mean: University degree 0.67 33.65 27.10 Control Mean: Diploma 0.74 39.00 29.47 Observations 2,597 2,597 2,597 p-value University 0.39 0.27 0.14 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. Table C12: Results Pooled: Work Categories - Heterogeneity by University (1) (2) (3) (4) Employee Apprentice/Intern Employer Self-Employed Offered Training 0.02 0.01 0.00 0.01 (0.03) (0.02) (0.01) (0.02) Offered * University degree 0.06 -0.01 -0.02 -0.04 (0.06) (0.04) (0.02) (0.04) Control Mean: University degree 0.36 0.22 0.03 0.17 Control Mean: Diploma 0.36 0.21 0.06 0.20 Observations 2,597 2,597 2,597 2,597 p-value University 0.09 0.97 0.29 0.53 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 48 Table C13: Results Pooled: Job Benefits - Heterogeneity by University (1) (2) (3) (4) (5) Area Contract Contract Job Registered Studies Written Permanent Benefits Firm Offered Training 0.05∗ 0.03 0.03 -0.00 0.05∗ (0.03) (0.03) (0.02) (0.02) (0.03) Offered * University degree -0.17∗∗∗ 0.04 0.04 0.05 0.04 (0.06) (0.06) (0.04) (0.04) (0.06) Control Mean: University degree 0.39 0.30 0.08 0.15 0.41 Control Mean: Diploma 0.60 0.28 0.10 0.13 0.39 Observations 2,597 2,597 2,597 2,597 2,597 p-value University 0.03 0.17 0.05 0.27 0.07 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Table C14: Results Pooled: Channels - Heterogeneity by University (1) (2) (3) (4) Searched For Job Confidence Approaching CV in People Discusses in Last Month Employers (0-3) Digital Format Jobs with Offered Training 0.04 0.03 0.02 0.82∗∗∗ (0.03) (0.04) (0.01) (0.20) Offered * University degree -0.10 0.12 0.03 0.35 (0.06) (0.08) (0.04) (0.43) Control Mean: University degree 0.63 2.53 0.91 5.53 Control Mean: Diploma 0.68 2.48 0.85 5.07 Observations 1,728 2,597 2,597 2,597 p-value University 0.26 0.03 0.12 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. Earnings are top winsorized at the ninety-ninth percentile; we replace earnings with 0 for participants who do not work. 49 Table C15: Results: Network Details by University (1) (2) (3) (4) (5) Family Friends Social Media Schoolmates Train mates Offered Training 0.07∗∗ 0.04∗ 0.12∗∗∗ 0.09∗∗∗ 0.31∗∗∗ (0.03) (0.02) (0.03) (0.03) (0.02) Offered * University degree 0.02 0.07 -0.05 0.03 -0.07 (0.06) (0.05) (0.06) (0.06) (0.05) Control Mean: University degree 0.43 0.72 0.44 0.53 0.15 Control Mean: Diploma 0.42 0.69 0.44 0.49 0.21 Observations 2,597 2,597 2,597 2,597 2,597 p-value University 0.10 0.01 0.24 0.01 0.00 Notes: Offered Training is the coefficient from a pooled OLS regression of the outcome (including three rounds of follow-up data, 4, 10, and 15 months after the end of the program) on an indicator for being offered the training. We control for the baseline value of the dependent variable, stratification dummies and survey wave dummies. Standard errors are clustered at the respondent level. 50