Life Skills, Employability and Training for Disadvantaged Youth: Evidence from a Randomized Evaluation Design

This paper presents an impact evaluation of a revamped version of the Dominican youth training program Juventud y Empleo. The paper analyzes the impact of the program on traditional labor market outcomes and on outcomes related to youth behavior and life style, expectations about the future and socio-emotional skills. In terms of labor market outcomes, the program has a positive impact on job formality for men of about 17 percent and there is also a seven percent increase in monthly earnings among those employed. However, there are no overall impacts on employment rates. Regarding non-labor market outcomes, the program reduces teenage pregnancy by five percentage points in the treatment group (about 45 percent), which is consistent with an overall increase in youth expectations about the future. The program also has a positive impact on non-cognitive skills as measured by three different scales. Scores improve between 0.08 and 0.16 standard deviations with the program. Although recent progress noted in the literature suggests that socio-emotional skills increase employability and quality of employment, the practical significance of the impacts is unclear, as there is only weak evidence that the life skills measures used are associated to better labor market performance. This is an area of growing interest and relevance that requires further research.


1) Introduction
Youth training programs have been implemented in Latin America and the Caribbean since the early 1990s. These programs target less-educated youth, a group that faces serious difficulties in achieving a successful insertion into the labor market, with the explicit aim of raising participants' job skills and matching them to suitable employers. 2 Drawing on lessons from evaluations of the Job Training Partnership Act in the United States and the Youth Training Scheme in Britain, these programs combine classroom training with a subsequent internship period of on-the-job work experience. 3 The training programs have two basic premises: first, that the lack of basic technical and life skills determine the poor labor market insertion of the targeted youth i.e. low wages, informality and underemployment, and second, that these courses are successful in enhancing those skills. 4 There is also an underlying assumption that the economy has or is creating vacancies to be filled by program graduates. 5 A salient characteristic of these programs is their emphasis on socio-emotional skills, 6 which have gained increasing importance in most of these projects (Ibarraran and Rosas, 2009;Gonzalez, Ripani and Rosas, 2011). Until recently, however, job training programs have included socio-emotional skills components in an ad hoc manner, based on scant qualitative elements and without focusing too much on measuring these skills or the results of training in improving them. Recent evidence on the 2 See Heckman, Lalonde, and Smith (1999) for a general overview of training programs, and Betcherman, Olivas and Dar (2004) for a recent summary that includes some evaluations of developing country training programs. 3 The Job Training Partnership Act program is described extensively by Heckman, Lalonde and Smith (1999). Dolton, Makepeace and Treble (1994) describe the Youth Training Scheme. 4 For a recent assessment of these programs, see Gonzalez, Ripani and Rosas (2011). 5 As implemented in LAC, these programs place a heavy emphasis on the private sector, both as a provider of training and as a demander of trainees. Private training firms compete for public funds with proposals that need to be backed by commitments from local employers to offer internships. 6 While cognitive skills are related to the ability to learn and are related to the intellectual coefficient, socio-emotional or non-cognitive skills (also known as personality traits or life-skills) are related to behaviors and attitudes, and are also referred to as to "soft-skills". 4 importance of non-cognitive skills both from econometric and qualitative analyses of determinants of success in the labor market show that employers value certain behaviors that are linked to highproductivity workers (Heckman, Stixrud and Urzua, 2006;Urzua, 2009;Fazio, 2011).
While conceptually socio-emotional skills are well defined, it has been difficult to measure and analyze them empirically. Recent literature has explored alternative measurements (Brunello and Schlotter, 2011;Felfe, Lechner and Stein, 2011), mostly for developed countries. However, there is a knowledge gap on how to measure socio-emotional skills in the LAC region and about the importance of such skills in explaining the labor market outcomes of youth, particularly disadvantaged youth.
Furthermore, from a policy perspective, it is important to know whether and how socio-emotional skills can be acquired by young people in Latin America and the Caribbean.
One of the most innovative youth-training programs in LAC is the Dominican Republic's Youth and Employment Program, Juventud y Empleo (JE). JE was first designed in 1999 and is the first program of its type to have an experimental evaluation from its inception. This evaluation design has enabled program managers to learn from the implementation of the program and to use the evaluation findings to improve subsequent phases, in a virtuous cycle of evaluation and feedback. This cycle includes rigorous quantitative as well as qualitative evidence. In this way, the program has been able to be modified to test new hypotheses.
While previous evaluations of this program have focused almost exclusively on the labor market impacts-namely, employment rate, labor earnings and quality of employment, which we also report on-this paper also takes a closer look at the mechanisms by which training is supposed to improve participants' labor market performance, specifically, by increasing the non-cognitive and 5 socio-emotional skills with which they join the labor force. We also examine other important outcomes that can be attributed to training, such as the teenage pregnancy rate. Given the high teenage pregnancy rates in the Dominican Republic and the negative labor-market consequences of teenage pregnancy, this is an important outcome from a theoretical and a practical standpoint. Dominican teenagers receive little instruction in sex education. In the country, about 17 percent of females aged 15-19 years old already have children (ONE, 2008).
Evidence from the LAC region confirms the negative effects of teenage pregnancy on various socio-economic variables. In Mexico, in the short run, teenage pregnancy reduces years of schooling, school attendance and hours of work, while it increases marriage rates. In the long run, teenage pregnancy results in a loss in years of education and in lower income. It also contributes to a higher probability of being married and divorced (Acero-Gomez and Campos-Vazquez, 2011).
Our analysis is based on a sample of applicants for the cohort of trainees that participated in a version of the JE program that was modified as a result of the first impact evaluation. The cohort under study applied to receive training during 2008. We show that labor market impacts are mixed, with negligible impact on overall employment and significant impact on job quality for men. We find positive impacts in terms of perceptions and expectations about the future, in particular for young women who simultaneously reduce their pregnancy rates significantly. We also document a positive impact of training on alternative measures of life-skills, particularly leadership skills, conflict resolution, self-organization and persistency of effort. These skills, in particular persistency of effort, have been analyzed and the findings show that they improve labor market outcomes in developed countries (Heckman and Urzua, 2006). The impact of those soft skills on labor market performance in developing countries is a rich area for future research.

6
The paper is organized as follows. After this introduction, Section 2 provides the specifics of the program, as well as a description of its previous evaluation. Section 3 describes the basic design features of this evaluation and the data collected. Section 4 presents the results, followed by conclusions in Section 5.

Description of the Intervention
The JE program -is a Dominican active labor market program (ALMP) that aims to improve the labor market entry of youth between 16 and 29 years of age who did not complete high school. It has been in operation since 2001 and was the first job training program in Latin America to incorporate a randomized evaluation component when the project was designed.
The program offers a wide range of job training courses such as administrative assistant, baker, hair stylist, clerk, auto mechanic, bartender, and so on. The Ministry of Labor outsources the provision of training services to private training institutions (Centros Operadores del Sistema, COS) that are registered and approved by the national training institution (Instituto Nacional de Formación Técnico Profesional, INFOTEP). Courses of 225 hours are conducted in the COS facilities and split into two parts: 75 hours of basic or life skills training, and 150 hours of technical or vocational training. Basic skills training is meant to strengthen trainees' self-esteem and work habits, while vocational training is meant to address the technical training needs of local employers. Training at the COS is followed by an internship in a private sector firm, which should be contacted by the COS in order to develop training programs tailored to the firm´s labor demand. Young people are identified by the COS according to their preferred vocation and the availability of the desired course. Once they reach 35 potential 7 participants, the COS sends the names and identification numbers to the program coordinating unit (PCU), which randomly selects those who are offered the training course.

Previous evaluations
The first impact evaluation of this program (Card et al., 2011) was based on a sample of applicants of the second cohort of the JE program who applied to receive training in early 2004. Baseline data were collected from applicants prior to random assignment through registration forms completed at the COS.
A follow-up survey was administered from May to July 2005; some 10 to 14 months after most trainees had finished their initial coursework. Simple comparisons between trainees in the follow-up survey and members of the control group show little impact on employment, although there is some evidence of a modest impact on wages and formality for men. Unfortunately, however, the randomized design of the JE evaluation was potentially compromised by the failure to include in the follow-up survey people who were originally assigned to receive training but failed to show up or attended only briefly.
Moreover, as is often the case in voluntary programs even under a well-implemented random assignment, compliance was not perfect: some of the lottery winners (intended to be treated) did not participate in the training either because they did not show up or they dropped out at some point. Some who were selected for the control group ended up taking the training as replacements of drop-outs and no-shows or for some other reason. Card et al. (2011) addressed the problem caused by the failure to follow up on no-shows through selection correction models and by showing with the baseline data that the characteristics of no-shows were similar to those of the replacements. They also excluded the reassigned controls from 8 alternative specifications and the results held. The estimated impacts on employment are all fairly close to zero, and there are no significant differences by gender, age, education, or geographic location. The estimated impacts on monthly earnings are fairly similar for men and women, and for younger and older workers, but they show interesting patterns by education and geographic location. If one compares better-educated applicants in Santo Domingo to all others the results are striking: this subgroup accounts for virtually all of the observed positive impact on monthly earnings. 7 The only other impact evaluation with randomized design of a similar training program in Latin America --the Colombian Jóvenes en Acción program--was done by Attanasio et al. (2011). They conclude that the program, which was contemporary to JE and had the same components, raised earnings and employment, especially for women. Women offered training earned 18% more and had a 0.05 higher probability of employment than those not offered training, mainly in formal sector jobs. 8

3) Evaluation Design
This second evaluation focuses on a modified version of the program and its evaluation design. While the core of the project --two-stage training followed by an internship--is maintained and the evaluation is still based on random assignment, there are some important changes: • COS are supposed to work closer to the firms that provide the internship in order to develop tailored courses to train people for real vacancies.

9
• The life-skills section of the training was revamped as firms argued that what they valued most from training were the general job-readiness/life-skills rather than the technical training. 9 • Random assignment was done on a larger sample for each course (20 treatments, 15 controls).
• Follow-up was improved in terms of sample size, survey instruments and quality controls of the field work.
Random assignment was applied on a group of potential participants identified by the COS that applied to the program and met the eligibility criteria. 10 The program received the information from the COS and verified that none of the applicants had been registered before. For each course, the COS submitted data on 35 eligible and interested young people, and the program managers at the PCU randomly selected and then divided them into two groups. The first one is formed by 20 young people who were offered the program. The second group is composed by the remaining 15 young people, who were assigned to the control group.
If young people offered the program did not respond or dropped out before the tenth day of ongoing classes, the COS could replace up to five slots with members of the control group. The replacements were supposed to be randomly selected by the PCU within the control group, and the PCU provided the names directly to the COS. However, in practice, the COS experienced some degree of discretion in selecting the replacements, which might have lead to selection bias. This is why we focused on the original random assignment to estimate the intention-to-treat effect, where there were no concerns about selection bias. 9 A complementary evaluation based on another cohort of trainees led by the World Bank analyzed the impact of providing only life-skills versus the traditional training, and their preliminary findings suggest that there is no valued added of the technical training. A qualitative analysis by Fazio (2011) presents additional evidence that firms value more the life skills component than the technical training. 10 The eligibility criteria are that participants should be 16 to 29 years old, living in poor neighborhoods; not attending school; with incomplete high school or less; unemployed, underemployed or inactive; and should hold an identity card.
Despite having an ideal initial configuration of the treatment and control groups due to successful randomization, there was imperfect compliance due to (non-random) decisions by COS and participants. Introducing a general notation, let Zi represent the random assignment of each young person i where Zi = 1 are those randomly assigned to the treatment group and Zi = 0 are those randomly assigned to the control group. Similarly, let Di represent the final treatment status were Di = 1 are those who attended the course and Di = 0 those who did not do so. The following table clarifies the setting: During the registration process, the program identified 10,309 applicants who met the selection criteria, with the following distribution according to the administrative data: Note that the number of never-takers equals the number of always-takers because the PCU tried to maintain the number of participants per course.

Identification Strategy
Follow-up data were collected on a representative sample of all those who participated in the lottery and, thus, is suitable for the estimation of the impact of the program on those who won the lottery --to whom a course was randomly offered--. This is the Intention to Treat Effect (ITT) that estimates the impact of offering the JE program, regardless of what happened after the random assignment. That is, some young people finally decided not to attend the courses or dropped out during the first week of classes, while some of those assigned to the control group ended up receiving the treatment.
Because the ITT yields the causal effect of Z i (Duflo et al., 2006), its estimation includes all the group of young people that participated in the random assignment, including those for whom Di≠Zi , the pairs D i = 1 | Z i = 0 and D i = 0 | Z i = 1, formed by those who took the course although they were randomly assigned to the control group and those who did not show up or dropped out although they belonged to the treatment group, respectively. Therefore, one may expect that the effect of offering JE becomes smaller to the extent that the proportion of those with Di≠Zi increases. 11 Under certain conditions, it is also possible to estimate the impact of the program on the compliers, i.e. those who took the course because they were selected in the lottery. 12 Without loss of 11 If the program has a positive impact and we compare Zi=1 with Zi=0 but less than 100 percent of the former took the course and more than 0 percent of the latter participated in the program, then the comparison yields an underestimation of the "true" program impacts. This is why under imperfect compliance of the random assignment the ITT leads to the dilution of the impact of the program. 12 Although we can estimate the size of the compliers (as the difference in the participation rate among those that won and those that lost the lottery), we cannot identify them individually. The impact on this group is known as the Local Average Treatment Effect (LATE), which yields a larger impact than the ITT since it assumes that any difference in the average outcome between the Zi=1 group and the Zi=0 group is due to the larger fraction of the former that participated in the training. Thus, the estimation of the LATE for JE compliers yields the impact of the program on those who were treated because they won the lottery; otherwise they would not have been recruited to attend the courses. In the instrumental generality, in this paper we discuss the ITT estimates computed with standard ordinary least squares . 13

Baseline Data
The baseline data were collected at the registration stage at each COS. They were available for all those eligible and interested to participate in the program, a total of 10,309 young people. Table 3 shows some characteristics from the baseline survey and a t-statistic for equality of means between Z i groups as an evidence of randomness. Most participants -62 percent-are women, and nearly a half had not completed high school. Ninety percent of participants live in urban areas and about a quarter live in Santo Domingo. The average age is 22 years. About 22 percent of individuals were attending school at the time of the baseline. As shown in the table's last column, random assignment was well implemented as most of the characteristics are balanced. variables jargon, the lottery is an instrument, unrelated with the outcome but related to participation in the program (it increases the probability of participation by about fifty percentage points). 13 We also estimated the LATE using Two Stage Least Squares (2SLS): the estimates are simply the OLS reduced form scaled up by the difference in the participation rate between those with Zi=1 and Zi=0 (the first stage of the 2SLS Source: JE baseline data and administrative records. Note: Means, differences and t-statistics are calculated by linear regression with robust standard errors. ***: significant at 1%; **: significant at 5%; *: significant at 10%.
It is interesting to note that only 4 percent of young people declared being employed at the baseline prior to the beginning of the courses, and that 52 percent of them were unemployed, meaning that 44 percent were inactive. This may be due to the requirements of the selection process, which demanded that they be inactive or unemployed, and may also be an expression of the Ashenfelter's dip, i.e. that both groups received a shock that increased unemployment levels right before the program started. According to the National Labor Force Survey (known as ENFT, Encuesta Nacional de Fuerza de Trabajo), in 2008 the employment rate in the Dominican Republic for young people 16 to 29 years of age with less than a complete high school education was 43 percent.

Follow-up Survey
After the completion of the courses, a follow-up household survey was carried out between November 2010 and February 2011 (18 to 24 months after graduation) on a random sample of 5000 out of the 10,309 young people who had initially registered. 14 This sample has 3,250 individuals from the treatment group and 1,750 from the control group.
The questionnaire for the follow-up household survey was put together by an interdisciplinary team from the Ministry of Labor of the Dominican Republic, the Inter-American Development Bank and the World Bank. It includes 15 modules that collect data on household composition and socioeconomic characteristics, labor force participation 15 , labor history, assets, time use, courses and the internship, consumption, health status, risk aversion, future expectations, pregnancy history, dwelling materials and basic skills, including non-cognitive skills and self-esteem.
About 80 percent of the sample were located at their households for the follow up survey, with virtually no difference between those selected and those not selected in the lottery (in the case of Z i = 1 and Z i = 0, 80.8 and 80.4 percent of them were interviewed, respectively). This compares favorably to the first evaluation of JE, where the attrition rate was larger and unbalanced between beneficiaries and members of the comparison group (the attrition rates were 35 and 45 percent, respectively).
We verified the similarity of the interviewed treatment and control groups. We also compared the basic characteristics of those that were interviewed and those that were not. Table A1 in the Appendix compares the sample of those intended to be interviewed and those finally interviewed. The original and the realized samples have similar characteristics at the baseline. Nonetheless, comparing the original and realized groups on the same random assignment (Z i ) there is a small imbalance in the poverty indicator (ICV) for the control group, but the differences are close to zero. Regarding the treatment group (column (a) -(c)), the most significant differences emerged in school attendance and location in urban areas, for which disparities are statistically significant but very small. Hence, we assume that attrition was random and that it affected equally both those selected and those not selected 15 Most of the questions that measure labor market outcomes are based on the ENFT that is carried out twice a year by the Central Bank of the Dominican Republic. Some questions were modified and adapted to the JE evaluation and youth population; the basic indicators of labor force participation are generated following the ENFT and allow performing an external validation of our data.
15 in the lottery.
The follow-up survey includes questions to confirm whether the respondent actually participated in the program. This is important because although the PCU has administrative data on this, the COS enjoyed some degree of control over who took the course. Also, in the process of replacing no-shows and dropouts some members of the Zi=0 group were contacted as replacements and, even if some of them may have declined to participate, they are still classified as control group compliers.
During the follow-up survey, enumerators did not have information on the classification of the youth into the treatment or control groups according to the administrative data. Young people were asked if the COS contacted them after the registration to notify them that they had been selected and to inform them the date and time when the course would start. Table 4 presents the answers to this question, showing that a large fraction of those who were not selected in the lottery were contacted by the COS in order to participate in the program:  to administrative data and information from the follow-up survey.  In this paper, the Di variable is defined based on information from the follow-up survey.
Individuals for whom Di=1 are those who reported having been contacted by the COS to begin the course and having accepted to start it. 16 Alternative measures of Di=1 were also used in order to determine if there were any differences between those who started the program, those who completed 16 Another filter was introduced. We wanted to make sure that those that were not contacted by the COS did not manage to participate in the program by other methods. Hence, we include in the Di=1 group those that report having taken the course, despite the fact that they were never contacted to begin the courses. Approximately 30 individuals from the control group with these characteristics were accepted by the COS to take the course and complete the internship. the classroom training, and those who also completed the internship. 17 The results are robust to alternative definitions. 18

4) Results
The outcomes explored in this paper can be classified into the following three categories: • labor market outcomes

• outcomes related to youth behavior and life style, perceptions and expectations
• measurements of socio-emotional skills.
We estimated the impact of the program on all of these outcomes for the complete sample and for various subgroups defined by gender, age, education, region and course. 19 In addition, in order to interpret the results, we looked at the relationship between the non-cognitive measures and labor market outcomes.

Labor Market Outcomes
Selected labor market outcomes in the follow-up survey by random assignment status are presented in Tables 6 and 7. In terms of employment rate, Table 6 shows that there are only minor differences between Zi=0 and Zi=1: employment is 62.5 and 61.6 percent respectively. 17 A decomposition of the initial Di=1 group by level of participation in the program is presented in Chart A1 in the Appendix. 18 These estimations are available upon request. 19 We present the most relevant outcomes in the paper. The complete set of regressions is available from the authors upon request. The estimation for the different subpopulations compares individuals from the treatment and control groups within each specific subpopulation. So, for example, the ATT coefficients of Santo Domingo compare individuals from the treatment group in Santo Domingo with individuals from the control group in Santo Domingo. This applies even to the type of course subpopulations, as there are individuals both in the treatment and the control groups for each type of course. Labor force participation is examined in Table 7. Female inactivity is larger for individuals assigned to the control group, while the opposite is true for employment and unemployment. Male inactivity and employment are higher for individuals assigned to the control group, while unemployment is higher for individuals assigned to the treatment group. Graphs 1 and 2 below show the employment history of individuals assigned to the treatment and control groups in the overall sample and in Santo Domingo. Taking into account the findings from the previous evaluation, program operators expected to find larger impacts in Santo Domingo because of its labor market dynamics. As expected, there were no significant differences in the months before the program started, particularly in the two to three months before registration took place. There were negative differences while trainees were taking the courses, and there was a catch-up after the courses.
Overall, there were no impacts on employment, and in Santo Domingo there seems to be a positive difference in the months closer to the follow-up survey.

Graph 1. Work History According to Retrospective Declaration
Source: Follow-up survey.

Graph 2: Work History According to Retrospective Declaration in Santo Domingo.
Source: Follow-up survey. Tables 8 shows the ITT estimates for the standard labor market outcomes. Although the program had no statistically significant impact on employment, for men it had significant impacts on formality, measured as employer provided health insurance or as having a written contract. Males assigned to the treatment group are four percentage points more likely to get a job in the formal sector than males assigned to the control group. This represents an impact of 17 percent over the control group average. 20 As shown in the last column, the results are particularly strong for males in Santo Domingo.
There is no impact on monthly earnings for the complete sample. However estimates for monthly earnings conditional on being employed show that the program does have a positive statistically significant impact. Among those employed, individuals assigned to the treatment group have monthly earnings seven percent higher than individuals assigned to the control group. This holds for women in general, and for men in Santo Domingo.
There also seems to be a relation with the duration of unemployment, which is longer for males assigned to the treatment group than for males assigned to the control group. This could reflect the fact that males assigned to the treatment group are searching for better quality jobs even if these are harder to find, or that they have higher reservation wages.

Risky Behavior and Pregnancy, Perceptions and Expectations
An important element of the life skills component is to enable young people to plan and think about their future in a more serious and organized manner. By giving them elements to increase their selfesteem and develop their personal abilities to compete in the labor market, young people can become more optimistic about their future and realize the importance of making the adequate decisions today, which may influence their risky behaviors. Hence, a first step in the success of the life skills component is raising young people' expectations about their future, as this should encourage them to engage in positive behaviors (which become more "profitable") and prevent them from undertaking negative ones (which become more "costly"). Table 9 presents, for all females in the sample regardless of their marital status, the impact of the program on pregnancy at the time of the follow-up survey. The estimates indicate that the program had a statistically significant negative impact on the probability of being pregnant. Females in the treatment group were on average two percentage points less likely to be pregnant than females in the control group. It is worth noting that this effect was driven by the group of females between 16 and 19 years old. Hence, the program was effective in reducing teenage pregnancy, which as discussed in the introduction has impact on their future labor market outcomes. ITT estimates for other risky behaviors such as consumption of alcohol, cigarettes, drugs and lottery were also computed. However, while the individuals assigned to the treatment group seemed to spend on average less money on consumption of these goods than the individuals assigned to the control group, the differences were not statistically  criminal activities were also estimated. However, there was no statistically significant impact on these 21 The control group mean reveals that Dominican youth are very optimistic to begin with. Hence, improvements in the treatment group are with respect to a very high starting point. 22 In this context in which youth have not even completed high school, expectations of a better education level refer to going back to school or continuing to take training courses of the sort of the JE program.
expectations, which is in line with a positive outcome bias in which respondents are more likely to see themselves reflected in positive than in negative situations. 26

Life Skills
A contribution of this paper is to present empirical evidence on the impacts of training on life skills.
For this purpose, the Social and Personal Competencies Scale, CPS for its Spanish acronym ( We also used the Rosenberg and Grit scales, which are standard and proven methodologies in psychology to measure personality traits and socio-emotional competence. The Rosenberg Scale is a professional instrument used in clinical practice to measure self-esteem levels. 23 The test is composed of 10 questions that should take between one to three minutes to answer. A higher score on the scale is associated with a higher level of self-esteem (Brea, 2010). The Grit Scale focuses on determination and strength of mind. 24 It measures persistency of effort, enthusiasm about long term goals, consistency of interests, and ambition. The instrument consists of 13 questions that can be completed in one to four minutes. Higher scores on the Grit Scale are associated with higher levels of determination and motivation during long periods of time despite failure or adversity (Brea, 2010).
Scores for our sample were all estimated as described in Brea (2010). Results are presented in 27 terms of standard deviations in order to ease interpretation. As shown in table 11, the program has a positive and statistically significant impact in most CPS measures. On average, the total CPS score of individuals assigned to the treatment group is 0.11 standard deviations higher than the score of individuals assigned to the control group. Impacts of similar magnitudes hold for CPS scores on leadership, behavior in situations of conflict, self-esteem, and order and self-organization. There are no impacts on the CPS scores on abilities to relate with others and empathy and communication skills. 25 For most of the indicators these results hold for both males and females, but the impact on males tends to be larger. By groups, the impact concentrates on the youngest individuals, on the better educated -incomplete high school is the highest education level to be eligible for the program--and in Santo Domingo.
While the program did not have an impact on the Rosenberg Scale for the complete sample, it increases the Rosenberg Scale of males in about 0.11 standard deviations. Looking at the impacts on the Grit Scale, the estimates show that the program had a positive and statistically significant effect in the total scale and in the sub-scales for persistency of effort and ambition. On average, the total Grit Scale of individuals assigned to the treatment group was 0.08 standard deviations higher than the score of individuals assigned to the control group. The impact on the subscales mentioned above is comparable in magnitude. As in the results for the CPS scale, impacts on the total Grit Scale and the different subscales concentrated and were stronger on the youngest individuals, the ones with higher educational attainment, and on those living in Santo Domingo. However, there was a sharp contrast on the results by gender, as in this case the impacts were only statistically significant for females. 25 The results for the full set of CPS dimensions are available upon request.
28 While there is a statistically significant impact on several of the measures of non-cognitive development analyzed above, it is hard to determine the practical significance of these results.

29
Literature on the relationship between non-cognitive skills and professional success is relatively new and it is still uncertain which specific competencies positively relate to participation in the labor market. In addition, there is also little evidence on the magnitude of the changes in non-cognitive abilities that are required to generate observable changes on the levels of youth employability.
However 30 exercise to further explore the relation between each of the scales used in this paper to measure life skills and the probability of being employed. We estimate standard regressions in which the dependent variable is the employment status of the individual and the explanatory variable of interest is each of the above scales for the control group. 28 The regressions include a set of additional covariates to control for gender, age, education, experience, and whether the individual is currently attending school. Table 12 presents the results from this exercise, which are rather disappointing: there does not seem to be any meaningful association between having higher levels of life skills, as measured by our instruments, and employment. 29 . 28 We use only the individuals assigned to the control group. The objective of limiting the sample is to avoid endogeneity that would arise when using the entire sample, as individuals assigned to the treatment could have both higher scores in the non-cognitive scales and higher levels of employment. 29 We also tried using additional dependent variables, such as formality and earnings, and did not find any impacts.

5) Conclusions
Juventud y Empleo is a labor training program for disadvantaged youth in the Dominican Republic. It is one of the first programs of its type in Latin America to incorporate a randomized design that allows the implementation of rigorous impact evaluations, which provide feedback to the program in its different phases. The first impact evaluation of JE showed limited impacts on employment and wages, which lead to changes in the program that focused on working closer with the private sector and This evaluation also shows positive impacts of JE in different measures of non-cognitive skills.
Using the CPS, the Rosenberg and the Grit scales, estimations show a consistent improvement in scores, ranging between 8 and 16 percent of a standard deviation. It is hard to interpret the practical significance of these results as evidence on this topic is very limited. Further analyses carried out in this paper show no correlation between some of these measures and employment. These results should encourage further research on the relationship between non-cognitive skills and labor market performance in order to understand the mechanisms through which life skills training can contribute to youth insertion into the labor market. Research on the instruments used to measure such skills is also needed.