NBER WORKING PAPER SERIES

HARD AND SOFT SKILLS IN VOCATIONAL TRAINING:

EXPERIMENTAL EVIDENCE FROM COLOMBIA

Felipe Barrera-Osorio
Adriana D. Kugler
Mikko I. Silliman

Working Paper 27548

http://www.nber.org/papers/w27548

NATIONAL BUREAU OF ECONOMIC RESEARCH

1050 Massachusetts Avenue
Cambridge, MA 02138
July 2020

We thank the Carvajal Foundation for partnering with us to run and implement the program, and
in particular, Mario Gonzales, Ana Enriquez and Angela Gonzales, and for providing funding for
research assistants and travel to conduct field work. We are also grateful to the staff at the
Ministry of Health and Social Protection for providing us with the administrative data, and
Milagros O'Diana for research assistance. We thank David Deming, Nada Eissa, Carolina
Gonzales, Carmen Pages, and Ken Wolpin as well as seminar participants including Isaac Mbiti,
Derek Neal, and Lesley Turner for helpful comments. The views expressed herein are those of the
authors and do not necessarily reflect the views of the National Bureau of Economic Research.

NBER working papers are circulated for discussion and comment purposes. They have not been
peer-reviewed or been subject to the review by the NBER Board of Directors that accompanies
official NBER publications.

© 2020 by Felipe Barrera-Osorio, Adriana D. Kugler, and Mikko I. Silliman. All rights reserved.
Short sections of text, not to exceed two paragraphs, may be quoted without explicit permission
provided that full credit, including © notice, is given to the source.
Hard and Soft Skills in Vocational Training: Experimental Evidence from Colombia
Felipe Barrera-Osorio, Adriana D. Kugler, and Mikko I. Silliman
NBER Working Paper No. 27548
July 2020
JEL No. C21,I25,I26,J24,J60,O54

ABSTRACT

We randomly assign applicants to over-subscribed programs to study the effects of teaching hard
and soft skills in vocational training and examine their impacts on skills acquisition and labor
market outcomes using both survey and administrative data. We find that providing vocational
training that either emphasizes social or technical skills increases formal employment for both
men and women. We also find that admission to a vocational program that emphasizes technical
relative to social skills increases overall employment and also days and hours worked in the short
term. Yet, emphasis on soft-skills training helps applicants increase employment and monthly
wages over the longer term and allows them to catch up with those learning hard skills. Further,
through a second round of randomization, we find that offering financial support for
transportation and food increases the effectiveness of the program, indicating that resource
constraints may be an obstacle for individuals considering vocational training.

Felipe Barrera-Osorio Mikko I. Silliman

Department of Leadership, Harvard University
Policy and Organizations Graduate School of Education
Vanderbilt University 13 Appian Way
106 C Payne hall Cambridge, MA 02138
230 Appleton Place [email protected]
Nashville, MA 37203
[email protected]

Adriana D. Kugler
Georgetown University
McCourt School of Public Policy
37th and O Streets NW, Suite 311
Washington, DC 20057
and NBER
[email protected]
1 Introduction
Vocational training programs are often seen as a means to improve the transition between

formal schooling and employment. However, the extent to which vocational training pro-

grams succeed in improving the labor market outcomes of their participants is highly debated

 and empirical evaluations of vocational programs report mixed results. While some pro-

grams suggest positive and sustained impacts (Kugler et al, 2020; Silliman and Virtanen,

2019; Brunner et al., 2019; Attanasio et al., 2017; Attanasio, Kugler and Meghir, 2011),

others report no eects (Hicks et al., 2014). Moreover, even the positive eects of some

vocational programs that succeed in improving labor market outcomes in the short term

can dissipate in only a few years (Acevedo et al., 2017; Alzua et al., 2016; Hirshleifer et

al., 2016). A common explanation for this fadeout is that the later outcomes of vocational

trainees may suer if the narrow training they receive does not provide them the skills to

adapt to changes in the nature of work (Hanushek et al., 2017; Krueger and Kumar, 2004).

In this paper, we use a randomized experiment to study whether vocational training

programs can be designed to provide their participants with sustained benets by exoge-

nously varying the technical and social skills provided in the program. Building o recent

work highlighting the importance of social skills in adapting to labor-market change (Dem-

ing, 2017), we randomize applicants to over-subscribed vocational training programs in Cali,

Colombia to various treatment arms as follows. First, we randomly assign applicants to

either receive vocational training or not. Then, within particular vocational programs, we

randomly assign applicants to programs with varying degrees of training in technical and

social skills. We assess the eects of providing social and technical training in vocational

programs by tracking applicants through both an extensive array of survey data and data

on labor market outcomes from social security administrative records.

By randomly assigning curricular content within otherwise comparable vocational train-

ing programs, our study is among the rst to the best of our knowledge to explicitly ad-

dress how curricular content in vocational training aects labor market performance. While

a number of studies focus on policy-changes to oer some insight into the potential eects

of changes in vocational curricula (e.g., Bertrand et al., 2019; Hall, 2016; and Malamud and

Pop-Eleches, 2010), any eects identied by these non-experimental studies might be partly
1
driven by changes in the composition of students entering vocational training. Further,

comparisons between cohorts are challenging, because, as Field et al. (2019) observe, voca-

tional graduates are particularly sensitive to changes in initial local labor market conditions

1 Meer (2007) notes that selection into vocational programs or tracks make it challenging to interpet
descriptive dierences in outcomes between graduates of dierent programs as causal.

2
due to business cycle uctuations. There are few randomized trials that examine the impacts

of soft skills. Groh et al. (2016) randomly assigned community college women graduates in

Jordan to soft skills training and nd no impact of the program. Acevedo et al. (2017) and

Ibarran et al. (2014) both examine the impact of combining soft skills with internships in
2
the Dominican Republic and nd positive short-term eects but only for women.

In the aggregate, our results show that admission to vocational training through random

assignment increases formal employment for both men and women. Both the social and

technical vocational programs produce positive eects on labor market outcomes. These

eects are robust to a number of specications and they are present regardless of whether

we use our own survey or administrative measures of employment. Our cost eectiveness

estimates suggest that the program pays for itself after about seven months. Our results

also suggest that both programs improve organizational skills.

Next, we test for dierences between the eects of the social and technical vocational

training programs. Our analysis highlights three ndings. First, the initial benets of

vocational training are smaller for those randomly assigned to social skills training than for

those assigned to the technical training program. Second, we nd that despite the smaller

initial premium to the training program emphasizing social skills, those exposed to the social

skills training eventually catch up to their technical program counterparts between six to 12

months after nishing the program. Third, we nd no signicant dierences in social skills
3
between participants exposed to the two types of vocational programs.

Through a further treatment arm that randomizes stipend receipt, we study the potential

role of resource constraints as barriers to participation in vocational training. We nd that

applicants randomized to receive stipends experience a bigger eect of vocational training

on employment than those without stipends. We also examine the heterogeneity of our

results and nd that the benets of vocational training accrue largely to men in terms of the

probability of getting a job but that women also benet similarly to men in terms of increased

formality and hours and days worked. This suggests that skills alone may be insucient for

helping women to overcome other barriers to employment such as lack of access to childcare.

The aggregate results from our study are in line with previous studies from Colombia, the

Dominican Republic, and Mongolia, which suggest that vocational programs can improve

2 This study oers two treatment arms, one that combines soft skills training and the internship, and one
that also adds vocational skills training to the soft skills training and internship. Unfortunately, this study
does not allow disentangling the eect of soft skills from that of the internships. While both papers listed
study the same program, Acevedo et al. (2017) extend the analysis to include labor market outcomes.
3 One potential reason for why there are no observed dierences in the social skills between participants in
the two types of vocational training is that both programs include teaching in hard and soft skills. Likewise,
since the jump in social skills training is not from zero to one-hundred, it is possible that eects on labor
market outcomes are explained by complementarities between hard and soft skills.

3
labor market outcomes in the short term (Field et al., 2019; Acevedo et al., 2017; Attanasio,

Kugler and Meghir, 2011; and Card et al., 2011). Kugler et al. (2020) and Attanasio et

al. (2017), who are able to follow their sample for up to 10 years through administrative

data, nd that the initial benets of vocational training in Colombia may persist in terms

of labor market outcomes into the longer-term and Kugler et al. (2020) also nd that

there are also benets 13 years after participation in the program in terms of further formal

educational attainment. Moreover, Kugler et al. (2020) also document spillover benets

from vocational education to relatives of the treated individual, suggesting that studies of

vocational education in developing or middle-income contexts may understate the benets

of such programs by focusing only on the individual.

Furthermore, our study advances the knowledge on the ongoing debate in the literature on

vocational training by building on insights from economic theory. A common view suggests

that general (vs. specic) skills are important in providing individuals with the exibility

to adapt to changes in the demands of the labor market (Deming and Noray, 2018; Deming,

2017; Goos et al., 2014; Acemoglu and Autor, 2011; and Goldin and Katz, 2009). Applying

this view to the context of vocational education, some scholars have argued that any initial

benets of vocational education, which tends to emphasize technical education specic to

a trade, are likely to disappear with time (Hampf and Woessmann, 2017; Hanushek et al.,

2017; and Heckman and Krueger, 2005; and Krueger and Kumar, 2004).

By randomly assigning applicants to intensive vocational training programs emphasizing

either technical or soft skills, we directly investigate how labor market dynamics are aected

by the extent to which vocational curricula include social skills, which are likely to be a

particularly important form of general skills (Deming, 2017). By comparing vocational

programs that dier only in the degree to which they include social skills, our study builds

on the few studies that evaluate vocational programs with components related to social
4
skills (Acevedo et al., 2017; Groh et al., 2016). We nd that vocational training improves

organizational skills  but, perhaps surprisingly, we nd little dierence in the acquisition

of social skills between individuals exposed to dierent types of programs. Nonetheless, we

do see that while those exposed to the program emphasizing technical skills are quicker to

4 Acevedo et al., (2017) compare two vocational programs both of which have the same degree of soft skills
training, stipends, and internships, but vary in the extent to which they have technical training. Groh et al.,
(2016) study an intervention which compares the labor market outcomes of public female community college
graduates in Jordan who are exposed to either an employment voucher, soft skills through an employability
training program, or a combination of the two. Our results contrast with the Acevedo et al.'s (2017) results,
which show positive short-term impacts of social skills programs on women, but not in the medium term,
and negative impacts on men even in the short term. We instead nd positive impacts of vocational and
soft skills training on both women and men in the short and medium terms. Our results are also in sharp
contrast to Groh et al.'s (2016) results, which show no eects at all of a much shorter soft skills training
program in Jordan.

4
nd employment, those exposed to the program emphasizing social skills soon catch up, and

are slightly more successful in maintaining their jobs. These ndings add nuance to the

debate of general and specic skills in vocational training, suggesting that technical skills

can be helpful in helping people nd employment, but social skills may be important in
5
allowing them to maintain their jobs into the longer term. These ndings may also help to

explain why vocational programs that include some training in general skills may succeed

in providing long-term benets (Kugler et al., 2020; Bertrand et al., 2019; and Silliman and

Virtanen, 2019).

Finally, our results highlight obstacles that may continue to prevent people from ben-

eting from vocational training. Existing studies document that resource constraints can

prevent individuals and families from investing in training and education. Further, stud-

ies suggest that vocational training programs can aect men and women in dierent ways.

For Colombia, Kugler et al. (2020) nd that credit constraints were likely behind women's

increase in educational attainment when they gained access to vocational training, while

knowledge of specic skills was likely behind the increase in college education after training

for men.

The rest of the paper proceeds as follows. In Section 2, we describe the program and the

experimental design. In Section 3, we describe both the survey as well as the administrative

data we use for our analysis. We present the results in Section 4. We present cost-benet

analysis in Section 5 and conclude in Section 6.

2 Program Description and Experimental Design

We randomly assigned applicants who enrolled in oversubscribed vocational training classes

which were part of the Inclusive Employment Program (IEP) oered by the Carvajal Foun-
6 7
dation based in Cali, Colombia. First, within specic vocational programs (for example,

cooking or sales), applicants were randomly assigned to receive vocational training versus no

training at all (i.e., treatment versus control groups). Second, those randomly assigned to

5 While consistent with theory suggesting sustained returns to general skills, our experimental design does
not allow us to directly tease-apart whether these results are due to general skills better allowing individuals
to adapt to technical change, or if social skills - either by improving employee-employer relationships or the
increased importance of face-to-face encounters - are driving these results.
6 We partnered with the Carvajal Foundation to run and implement the program. The Carvajal Founda-
tion is a non-prot foundation devoted to help with social programs in Cali including programs to support
entrepreneurship, education initiatives, training, and employment programs.
7 Cali is the third largest city in Colombia with 2.2 million people, after the capital, Bogota, with 7.4
million people and Medellin with 2.4 million people.

5
vocational training were assigned to receive dierent degrees of training in social (Treatment

1) and technical (Treatment 2) skills. Third, half of those assigned to vocational training

were also provided a stipend for transportation and meals.

Between the end of June 2018 and the beginning of December 2018, the Carvajal Foun-

dation oered 18 classes in total with 8 dierent types of courses. While one course ended in

July and one in August of 2018, the vast majority ended in the Fall of 2018 (4 in September,

2 in October, 8 in November and 2 in December). The courses were all in the service sector

in each of the following areas: sales and client services, general services, surveillance and

security services, cashiers, quality control assistant, cooking assistant, delivery assistant and
8
storage assistant. The courses lasted for a period of between 4 and 10 weeks with courses

oering classes between 5 and 8 hours a day.

Participants registered voluntarily into classes in response to a call for registrations by the

Carvajal Foundation. The Foundation established this program to help the poorest in the

community access jobs. Thus, the foundation reaches broadly to enroll participants through

radio, social media, loud-speakers in cars that go through poor neighborhoods, yers, and

through the public employment oce and oces that provide other public services to the

poor. As reported in the next section, most individuals who registered for these courses were

in the lowest socio-economic strata according to the Census of the Poor in Colombia.

Individuals who were interested in registering for the classes attended an informational

meeting and registered for the specic classes they wanted to take. Each class had between

23 and 31 spots and registration in each class ranged from 28 to 47 registrations per class.

Given over-subscription in the classes, the foundation randomly selected individuals to either

receive a spot or not receive a spot in the course. The lotteries for each course were recorded

by video to ensure everyone knew people were allocated into the courses by luck. Those

who did not win a spot in the training courses through the lottery were in the control group

and were not provided other services by the Carvajal Foundation for an entire year following

registration. As shown in the next section, the randomization divided people into groups

that were very similar on average in terms of their characteristics, thus giving credibility

that the lottery worked well in terms of randomly assigning individuals into and out of the

classes. There were initially 710 people who registered in the courses and of these, 483 were

randomly assigned to the training and 227 were assigned to the control group.

The next step of the random assignment consisted of randomly assigning classes to place

greater emphasis in teaching social skills or technical skills. In particular, Treatment 1,

which oered greater emphasis on social skills oered 100 hours of social skills and 60 hours

of technical skills, while Treatment 2, which oered greater emphasis on teaching technical

8 The program oered two of each of these types of courses, except for 4 courses in General Services.

6
skills, oered 60 hours of social skills and 100 hours of technical skills. The weekly hours

in each course ranged between 30 and 60 hours, which meant that courses lasted between

about 3 and 5 weeks. Since there were two classes in each area, a coin toss decided which

of the two classes, for example in security services, gave more emphasis to soft skills and

which one to technical skills. This was done for each type of course and the courses were

oered with the same content in each, just with more time devoted to teaching social skills

or with more time devoted to teaching technical skills. In total, there were 483 individuals

assigned to training with greater emphasis on social skills and 227 assigned to training with

greater emphasis on technical skills. Social workers provided the social skills content of the

courses, which included self-esteem, work ethic, organizational skills, inter-personal skills

and communication skills. The content of the technical skills varied depending on the course

(e.g., security and surveillance services, cashiers, or cooking assistant skills). In all cases,

the content was specic to a job and the skills were taught both in the classroom as well as

through practical hands-on experience in the training center of the Carvajal Foundation.

The last step of the randomization involved either oering or not oering a stipend to

participants. The monetary stipend consisted of US$1.50 per day. This monetary transfer

was meant to help participants cover the costs of transportation and meals while taking the

courses. The oer of a stipend was randomized at the class level to avoid envious comparisons

among participants in the same course. The stipend was paid at the end of the week for the

entire week to ensure that participants attended the classes each day of that week before

receiving the payment. As with the class-level randomization for the skill emphasis, the toss

of a coin determined which classes received (or did not receive) the stipends.

3 Data Description

3.1 Survey Data

We track our sample through survey and administrative data. We collected data on both

treatment and control individuals and their families by conducting our own pre-treatment

surveys. Further, we collected information on labor market characteristics and social skills

in the pre-treatment surveys. Importantly, these pre-treatment surveys were conducted in

November 2017, before people found out if they were randomly assigned into or out of the

program. We, then, collected information between March and April of 2019 using a reference

period of February 2019 for all labor market questions and a reference period of March and

April 2019 for social skills questions for both treatment and control individuals after the

7
courses had concluded.

We were able to re-interview 653 individuals in the post-treatment period or 91.97% of

the initial registered individuals. Of these, 444 individuals were in the treatment group

(91.9% of this group) and 209 in the control group (92.1% of this group), which means that

the attrition was the same among those in the treatment and control groups. Of the 8% who

attrited, 37 people were not interviewed in the follow up survey, 10 people did not conduct

the initial survey, and 10 were excluded because they indicated having a socio-economic level

above that intended for participants of the program.

Table 1 shows the covariate balance check between applicants assigned to treatment and

control groups. The rst column shows descriptive statistics for the control group, while

the second column shows the dierence in characteristics between the treatment and control

groups. Sixty six percent of control group individuals are women. They are on average 26

years old and have 11 years of education. Ninety nine percent have an elementary school and

94% secondary school education and 29% have technical higher education. Most individuals,

55%, are Afro-Colombians and 17% report being Mestizo. Control group households are

relatively poor with an average household income of US$19.23/day and an average house-

hold size of 4.45 individuals. Importantly, Column 2 shows that treatment individuals are

very similar to control group individuals in terms of all their characteristics. None of the

treatment-control dierences are individually or jointly signicant (the F-test is 0.59).

We, also, check that the characteristics between the control and treatment groups are

similar for each treatment arm. Columns 3 and 4 report the treatment-control dierences

for Treatment 1 (with emphasis on social skills training) and for Treatment 2 (with emphasis

on technical skills training), respectively. Column 3 shows that the only dierence between

Treatment 1 and the control group is the household size, which is slightly bigger for those

in the treatment. However, the test of joint signicance of all characteristics yields an F-

test of 0.95 and a p-value of 0.52, showing that these are not jointly signicantly dierent

between the social skills treatment and the control group. Similarly, Column 4 shows that

Treatment 2 and the control are only dierent along one dimension. The technical treatment

has slightly less access to electricity but this dierence is only signicant at the 10% level.

Moreover, the joint signicance test shows that the dierences of all the characteristics are

not jointly signicant (F-test is 0.71 and p-value 0.79).

Panel A of Table 2 reports pre-treatment dierences between the treatment and control

groups in terms of labor market outcomes in May of 2018, which is before any of the courses

started. As with other pre-treatment characteristics, we conduct a balancing test for indi-

cators of employment, whether the worker has a contract, whether the worker is a causal

worker. We also include balancing tests of days worked per month and hours worked per

8
week as well as wages per hour. The control group means in Column 1 show that 55% were

employed, but only 28% had a written contract. Control group individuals worked only 13.5

days per month and 24.5 hours a week. Not only did they not work full time, but also their

wages were very low, only US $1.20/hour. Column 2 shows that the treatment group was

very similar. In fact, none of these variables are signicantly dierent between treatment

and control group individuals. Likewise, the dierences between the control and each of the

treatment arms are also insignicant for all labor market outcomes.

Panel A of Table 3 reports measures of self-reported soft skills. We ask individuals to

reportusing a 7-point Likert scalewhether they agree if they possess a list of specic soft
9
or socio-emotional skills. The 1 to 7 scale includes the following options: strongly agree,

agree, somewhat agree, neither agree nor disagree, somewhat disagree, disagree, and strongly

disagree. We ask several questions related to work ethic and responsibility, inter-personal

skills and ability to get along with others, leadership skills, teamwork and communication

skills. Column 1 in Panel A of Table 3 shows the average scores for the control group. The

scores generally show that for the most part, individuals self-report as being neither good nor

bad in terms of their work ethic. The average score for work ethic is 3.78 or close to 4, which

means they neither agree nor disagree. Individuals in the control group report somewhat

agreeing that they possess organizational, interpersonal, leadership and communication skills

(the average Likert scores are 3.6, 3.37, 3.35 and 3.43 and 3 represents somewhat agreeing).

However, on average, they self-report as having less ability to work in teams. The average

score is 5.22, which means they somewhat disagree with possessing teamwork skills. Column

2 shows that treatment group individuals report similar soft skills to the control group,

as none of the dierences in average Likert scales between the two groups are signicant.

The same is true for the dierences between those receiving greater emphasis on technical

treatment and the control group. In addition, the dierence between those receiving more

social training and the control group are mostly insignicant and only the communication

skill dierences are marginally smaller for those in the treatment group.

3.2 Administrative Data

We conducted our follow up surveys between 3 and 7 months after the participants nished

the program. To be able to examine the long-term impacts of the program, we use Social

Security records up until December of 2019. This allows us to examine the labor market

9 Since these skills are distinct from one another, we choose to keep them separate rather than combine
them into an index.

9
outcomes for individuals between 11 months and 17 months after nishing the program. Since

the Social Security records only provide information about formal sector jobs that provide

social security benets, we are not able examine informal sector jobs as we do with our

survey data. In addition, due to condentiality, for the process of merging our experimental

sample with Social Security recordsa process undertaken by the Ministry of Health and

Social Protectionwe have to delete all personal information and only identify who was in

dierent treatment arms and who was in the control group. Therefore, all the analysis with

administrative data does not control for observable characteristics. Nevertheless, given the

baseline balance in characteristics across treatment arms, it is highly likely that controls for

baseline characteristics would not make a dierence in the point estimators.

Figure 1 shows data on employment characteristics from monthly administrative records

for the control and treatment groups. Panel (a) shows the trajectory of days of formal

employment, Panel (b) shows formal employment probabilities, and Panel (c) shows monthly

wages. The solid lines in the gures represent the control group, the dashed lines represent the

social treatment and the discontinuous lines represent the technical treatment. Figure 1(a)

shows that prior to treatment (most courses ended by November 2018 except for 2 which

ended on December 3rd), all groups had on average about 5 days of formal employment

per month. Days of employment jumped for all groups starting in December of 2018, but

the jump was greater for the two treatment groups than for the control group and this

persisted until December of 2019. Similarly, Figure 1(b) shows that the probability of formal

employment was only around 0.2 for the treatment and control groups prior to conclusion

of the courses, the probability of formal employment jumps disproportionately for the two

treatment arms relative to the control group after December of 2018 and remains relatively

higher until December 2019. Finally, Figure 1(c) shows also similar wages prior to the

courses, and relatively higher wages for the treatment individuals compared to the control

individuals after the conclusion of the courses. These gures preview similar ndings in the

next section using our survey data, which allows us to control for other characteristics, as

well as a more detailed event-study analysis of the administrative data.

10
4 Results

4.1 Aggregate Eects of Vocational Training

10
In this section we estimate intention to treat (ITT) eects of the training program. First we

use survey data and simply estimate dierences in average outcomes (Yict ) of those individuals

i assigned to course c at time t to treatment (T ) and those assigned to the control group.

Then, we estimate regressions controlling for baseline characteristics (Xi ), course xed eects

(γc ) eects and time since nishing training. To perform these estimates, we use variations

of a model of the following form:

Yict = β0 + β1 Tit + β2 Xi + γc + ηict (1)

Before running any formal estimates of the eects of vocational training on labor market

or social skill outcomes, we report raw dierences between those assigned to the treatment

and control groups across all measured outcomes. Panel B of Table 2 shows average dier-

ences in post-treatment labor market outcomes for treatment and control groups. Column

2 of Panel B shows increases in employment in the extensive and intensive margin. Af-

ter completing their courses, treatment group individuals experience an increase in average

overall employment and employment with contracts as well as increased days and hours

worked compared to control group individuals. By contrast, the likelihood of casual work

declines for treated relative to control individuals. These results, thus, suggest a shift to-

wards formal and better employment. Column 3 of Panel B in Table 2 shows results from

a dierence-in-dierence specication, comparing treatment and control individuals post-

treatment relative to treatment and control individuals pre-treatment. The results similarly

show increased formal employment and days and hours worked and decreased casual work.

The next four columns in Panel B show similar simple dierence and double dierence re-

sults for those with greater emphasis on social training and those with greater emphasis

on technical training. The results for the social treatment arm show a shift from informal

towards formal employment, while the results for the technical treatment arm show the same

increases on overall employment, days and hours worked and the shift towards formal em-

ployment found when combining both treatments. These preliminary results suggest that

treatment improved labor market outcomes and that the eect was greater for those with

more emphasis on technical training.

10 We focus on ITT rather than average treatment on the treated (ATT) eects, because there were some
of those initially assigned to the control group who were allowed to eventually enter the training courses.
This mainly happened at the beginning of the implementation of the program in the rst few courses oered.

11
 Panel B of Table 3 shows simple and double dierence results on the outcome measures

of social skills. Columns 2 and 3 show that both organizational and communications skills

seem to improve after training. In particular, organizational skills appear to improve for

both those in the social and technical treatment arms, while communication skills are also

improved for those receiving greater emphasis on social skills. These results suggest improved

social skills of applicants assigned into treatment.

We assess the robustness of the dierences between the outcomes of control and treat-

ment groups in our survey data through a series of statistical tests in Table 4. First, we

estimate ITT eects with course-specic xed eects and demographic background variables

(specication 1, Columns 1,4 and 7, Table 4). The baseline demographic characteristics are

gender, age, years of education, race/ethnicity indicators, and the household size. Next,

we re-run these estimates using a dierences-in-dierences strategy, taking advantage of the

pre-period measures of the outcomes (specication 2, Columns 2, 5, and 8). Third, we add a

variable measuring days after graduation to the estimation equation since the post-treatment

survey asked labor market questions about February 2019 as the reference period and about

March/April 2019 for the soft skills questions, but the courses all concluded at dierent times

(specication 3, Column 3). When we adjust these estimates for multiple-hypothesis testing

(Appendix Table 1), our results for increased participation in formal work remain statisti-

cally signicant at the 99% level without controls, but we lack power to detect statistical

signicance when we add our full set of controls (Benjamin et al., 2006).

The results from survey data are largely robust across specications and show that ran-

dom assignment of individuals into vocational training improves their labor market outcomes.

Columns (1) to (3) in Table 4 show that access to training increased the likelihood of being

employed with a formal contract and decreased the likelihood of being a casual worker. The

results are largely robust to the dierent specications and controls. The results in Column

(3) which includes all the controls shows that assignment to training increased the likelihood

of being in formal employment by 12 percentage points (70 percent) and decreased the like-

lihood of being a casual worker by 13 p.p. (28 percent). Additionally, while imprecise, we

also nd positive eects of training on days and hours of work in the survey data. Adding

credence to these estimates, when we translate this increase in hours of work to monthly

wages, our estimate of an increase in US$21 is very close to the estimated US$23.6 that we

estimate precisely with administrative data (Table 5, Panel A  discussed later in the paper).
11
By contrast, we nd no evidence of increased productivity as measured by hourly wages.

We also examine the impacts of training on the acquisition of soft skills in Panel B of

Table 4. Similar to the labor market outcomes, we increasingly add more controls in dierent

11 Note that we impute zero earnings for those who are not employed to avoid selection biases.

12
specications. Like in the simple and double dierence specications, the specications with

course xed eect and baseline demographics and outcomes as well as days since graduation

show a positive impact on organizational skills for both the groups that receive more emphasis

on social and technical skills training. This means that part of the reason why the training

worked was by providing soft skills that are helpful in getting a new job.

We then estimate the eects of vocational training on labor market outcomes using ad-

ministrative data. Since we are unable to link the administrative and survey data, these

estimates are run without any control variables, but have the advantage of allowing for mea-

surement over several months for a much longer period of time (from between 11 months

and 17 months after the courses nish). The results from administrative data also sug-

gest that assignment to vocational training increased the employment probabilities, days of

employment and wages (as proxied for by social security contributions).

We rst run our administrative-data estimates using an event-study framework where

we estimate dierences between treatment and control groups month-by-month, and include

month xed-eects (πm ). Standard errors are clustered at the individual level.

31
X
Yit = β0 + βt Tit + πm + ηit (2)
t=1

Figure 2 shows an event study analysis of the training program on these three outcomes.

While the rst course nished on July 20, 2018 and the last ended on December 3, 2018, most

courses ended by November of 2018. Figure 2(a) shows that days worked were very similar

between treatment and control individuals up to November 2018, but that they increased

sharply for those assigned to treatment relative to control individuals starting in December

2018 and were much higher for this group during the year 2019. Figure 2(b) shows a similar

increase in employment probabilities for treatment relative to the control group individuals

starting in December. Also, monthly wages show similarly a jump in December 2018 that
12
remains for the rest of the 2019. Figures 3 and 4 show separate eects for the two treatment

arms.

To provide a single estimate of the eects of the training programs we pool the monthly

administrative records (Table 5, Panel A). We run these estimates using a dierences-in-

dierences model where we compare the outcomes of individuals before taking part in the

experiment (β1 ) to those after all coursework ended (β3 ) (Equation 3). Our model allows for

changes in employment outcomes while treated individuals are exposed to the training (β2 ),

12 Note that we examine monthly wages instead of wages per hour because we do not have hours in the
administrative data. When we examine monthly wage in the survey data we also nd the same eect as in
the administrative data, indicating that this increase in wages is due to increased days and hours worked.

13
 13
but these are not signicant for any of our outcomes and are not reported in Table 5.

Yit = β0 + β1 Ti + β2 (Ti ∗ Duringt ) + β3 (Ti ∗ P ostt ) + πm + ηit (3)

These results suggest that the vocational training program increased employment by 2.16

days per month, the likelihood of being employed each month by nearly 8 percentage points,

and social security contributions (a proxy for wages) by US$21 per month.

Next, we take advantage of the dynamic nature of the monthly administrative data to

test whether or not the benets of vocational training fade-out over time (Table 5, Panel

B). To estimate the dynamic eects of training on employment, we add an additional term

to our equation by interacting the treatment in the post-period with the number of months

following treatment (Equation 4).

Yit = β0 + β1 Ti + β2 (Ti ∗ Duringt ) + β3 (Ti ∗ P ostt ) + β4 (Ti ∗ M onthsP ostt ) + πm + ηit (4)

The third row in Panel B measures any dynamic eects of training on employment. These

coecients are small and statistically insignicant, suggesting that we are unable to detect

any evidence of fadeout the post-training labor market outcomes between those assigned to

treatment and control groups.

4.2 Hard and Soft Skills in Vocational Training

We then examine the dierences in the dynamics between the labor market outcomes of

people admitted to the social and technical tracks using administrative data. Figure 5

shows the ITT employment dynamics between social and technical treatment groups using

the administrative data. Figures 5(a)-(c) show clear negative eects of social training (vs.

technical training) in the initial post-treatment period on days, employment probabilities and

monthly wages. As time passes, however, the dierences in labor market outcomes between

the social and technical treatment groups disappear. That is, the dierences between the

two groups go from being negative in December 2018 and the rst few months of 2019 to

being zero in the second half of 2019.

Panels C and D of Table 5 examine impacts separately for those that receive more em-

phasis of social and technical training, respectively. As with the survey data, these results

13 Including the During period to our analysis avoids any mechanical relation between post-period results
and dips in employment due to participation in the training program.

14
show a larger impact for those who received more emphasis on technical training. For those

with more social skills training, there is an increase in the probability of employment of 10

pp and an increase in wages of US$20. Those who receive technical skills training experience

an increase of 3.3 days worked, an increased probability of employment of 0.13 pp and an

increase in monthly wages of US$34, all signicant at the 1 percent level.

We formally test if the post-training dynamics are aected by assignment to the social

vs. technical training programs in Table 5 (Panel E). We include a social treatment dummy,

social treatment dummy interacted with a post-treatment dummy and a social treatment

dummy interacted with both the post-treatment dummy and the months since November.

The negative sign in the double interaction term suggests that there may be an initial cost

to social skills training, but the positive sign in the triple interaction term suggests that

social skills help improve workers' performance in the labor market over time. This is in line

with the positive impact of soft skills found in Deming (2017). While we nd corresponding

dierences between social and technical training using survey data in Tables 6 and 7, these

results are much less precise. Also, since we are unable to follow participants for a period

of time using the survey data, we cannot capture the dynamics we are able to see in the

administrative data.

4.3 Does Receiving a Stipend Make Vocational Training More Ef-

fective?
In Table 8 we test for whether stipend receipt magnies or reduces the eects of admission

to vocational training. On the one hand, receiving a stipend may make it more likely that

individuals attend the training sessions and magnify the potential impacts of the program.

On the other hand, receiving the stipend may make people less likely to attend the courses.

As explained below, however, the money was disbursed at the end of the week after people

had attended all weekly classes, so this is unlikely to be the case. Table 8 includes an

interaction of the treatment with the stipend to see if the training has a bigger or smaller

eect when a stipend is provided to the students. The results show that providing training

together with the stipend receipt had large positive and signicant eects on employment in

the extensive and intensive margins. The eects on employment, days and hours worked are

larger when training and the stipend are provided together. Without stipend receipt, the

eects of vocational training are zero, suggesting that resource constraints exhibit a serious

obstacle for people, preventing them from vocational training programs. This is in line with

the ndings in Kugler et al. (2020) who also nd evidence of credit constraints in another

15
training program in Colombia.

4.4 Dierential Impacts by Gender

Tables 9 and 10 report results of models that interact a female dummy with the treatment

dummy to identify if there are dierential eects of the program by gender. In order to

run these pooled models, we completely saturate the regression with interactions of all de-

mographics with the gender indicator. Table 9 shows no dierential eect for women and

men (the interaction term is insignicant) for most of the outcomes, except for employment.

While assignment to treatment has a positive impact on the employment of men, there is

no impact of assignment to treatment on employment for women. However, assignment to

treatment does benet women equally to men in terms of oering them access to formal em-

ployment and increasing the number of days worked. Table 10 also shows a smaller eect of

training on women in terms of gaining organizational skills compared to men. Thus, Tables

9 and 10 suggest that men were more likely to benet from vocational training than their
14
female counterparts. Note that these results contrast with previous evaluations of training

programs for young people, which nd bigger eects of training on young women (Attanasio

et al., 2011; Card et al., 2011). These results also contrast to the study by Acevedo et al.

(2017), which nd that soft skills training combined with internships and vocational training

has eects on women in the short term but not in the medium term. By contrast to our

nding, they also nd no eects on young men, although the baseline imbalance for men

may be driving those results.

4.5 Cost-Benet Analysis

Although the program may aect areas outside the labor market, wage returns represent a

lower-bound of a program's cost benets. The ITT estimates from the administrative data

suggest that, in aggregate, program participants experience a monthly US$21 wage gain by

participating in the program for the rst year following program participation. While lacking

statistical power, the estimates from the survey data are of a similar magnitude, suggesting

that participants are not merely shifted into the formal sector of the labor market. Given

the average age of program participants of 26, we assume that they will work another 35

14 We also test for heterogeneity by education and social strata, but we are unable to detect any dierences
by prior education or social strata probably partly due to the degree of homogeneity in our sample across
these measures.

16
years. Since we can only follow participants for one year after the program, the overall cost-

benet analysis will hinge on assumptions regarding whether or not the benets individuals

experience are permanent, or whether they fade-out over time. In Panel D of Table 11, we

present two scenarios: one in which the benets of the program persist, and another where

we assume 10 percent annual depreciation. Under the rst scenario, participants experience

a benet of US$4,262 over their lifetimes, whereas under the second scenario participants
15
experience a US$1,186 benet over their lifetimes.

Including the stipend, the direct marginal costs of operating the program are US$176. As

such, the two scenarios represent lifetime wage gains of US$4,086 and a US$1,010 respectively.

The cost-eectiveness of the vocational programs emphasizing social versus technical skills

are similar. Put another way, our results suggest that the vocational training program pays

for itself in about eight months.

5 Discussion and Conclusion

We use a randomized experiment in Cali, Colombia to study the eects of vocational training

on labor market outcomes. In aggregate, we nd that vocational training shifts people to

the formal sector of the labor market and increases their monthly earnings and employment.

Importantly, our results show positive impacts both in the short and medium term, with

eects persisting up to 12 and 17 months after the program.

The key novelty of this study is that we examine how providing dierent intensities of soft

or technical skills as part of vocational training aects labor market dynamics. We randomize

vocational admits to receive varying degrees of social skills training. Our results show that

both programs with an emphasis in social and technical skills have a positive impact on

labor market outcomes, the program with emphasis on technical skills has a bigger short

term eect. However, we nd that those with an emphasis on soft skills catch up to those

who receive more technical training about 6 to 12 months later in terms of employment,

earnings and hours worked. The catch up of those receiving more soft skills could be due to

the ability of those with soft skills to learn on the job due to the acquisition of organizational

and other soft skills. Alternatively, organizational and communication skills gained through

training may help these individuals retain their jobs and negotiate better salaries.

Our paper, thus, provides the rst experimental evidence showing that while technical

skills may improve immediate labor market outcomes, social skills help to sustain employ-

ment benets over time. Our paper contributes to the literature on soft skills in the work-

15 We nd no evidence of a fadeout in monthly returns a year after the program ended (Table 5).

17
place. First, we provide some evidence that soft skills can be taught even to older people.

Second, there are only a handful of randomized trials on soft skills training. Ours is the rst

one to show that soft skills training can have long-lasting eects for both men and women.

Groh et al. (2016) nd no eect, but their training takes place in a 9-day period and only

lasts a total of 45 hours. Acevedo et al. (2017) nd eects only short term impacts on

young women but not on young men in the Dominican Republic. However, not only was soft

training oered after hours and it is possible that young men did not attend, but it is not

possible to disentangle whether the eect of this program was due to an internship oered

in conjunction with the soft skills training. Thus, our paper is the most convincing evidence

to date on the eectiveness of a well-designed soft skills intervention.

Further, we randomize stipend receipt to study how resource constraints aect partici-

pation in vocational training programs in developing country contexts. Our results suggest

that resource constraints may be an important reason why individuals are unable to follow

through with their investments in vocational training.

Taken together, our results suggest that the extent to which vocational programs include

general skills such as social skills, can help sustain their benets into the longer term.

References
Acevedo, Paloma, Guillermo Cruces, Paul Gertler and Sebastian Martinez. 2017. How Job
Training Made Women Better O and Men Worse O, NBER Working Paper No. 23264.
Forthcoming in Labour Economics.
Acemoglu, Daron and Autor, David. 2011. Skills, Tasks and Technologies: Implications for
Employment and Earnings, in Handbook of Labor Economics, Volume 4, pages 1043-1171.
Elsevier.

Alfonsi, Livia, Oriana Bandiera, Vittorio Bassi, Robin Burgess, Imran Rasul, Munshi Su-
laiman, and Anna Vitali. 2017. Tackling Youth Unemployment: Evidence from a Labor
Market Experiment in Uganda, STICERD-Development Economics Papers.
Attanasio, Orazio, Arlen Guarin, Carlos Medina and Costas Meghir. 2017. Vocational
Training for Disadvantaged Youth in Colombia: A Long Term Follow Up. American Eco-
nomic Journal: Applied Economics 9(2): 131-143.
Attanasio, Orazio, Adriana Kugler., and Costas Meghir. 2011. Subsidizing vocational train-
ing for disadvantaged youth in Colombia: Evidence from a Randomized Trial, American
Economic Journal: Applied Economics, 3(3):188-220.
18
Benjamini, Yoav., Abba M. Krieger, and Daniel Yekutieli. 2006. Adaptive Linear Step-up
Procedures that Control the False Discovery Rate, Biometrika, 93(3): 491-507.
Bertrand, Marianne, Magne Mogstad, and Jack Mountjoy. 2019. Improving Educational
Pathways to Social Mobility: Evidence from Norway's Reform 94, NBER Working Paper
No. 26092. Cambridge, MA: NBER.

Brunner, Eric, Shaun Dougherty, and Steven Ross, S. 2019. The Eects of Career and Tech-
nical Education: Evidence from the Connecticut Technical High School System, EdWorking
Paper No. 19-112, Annenberg School, Brown University.

Card, David, Pablo Ibarrarán, Ferdinando Regalia, David Rosas-Shady, and Yuri Soares.
2011. The Labor Market Impacts of Youth Training in the Dominican Republic, Journal
of Labor Economics, 29(2): 267-300.
Deming, David J. 2017. The Growing Importance of Social Skills in the Labor Market,
Quarterly Journal of Economics, 132(4): 1593-1640.
Deming, David J. and Kadeem Noray. 2018. Stem Careers and Technological Change,
NBER Working Paper No. 26092. Cambridge, MA: NBER.

Dhaliwal, Iqbal, Esther Duo, Rachel Glennerster, and Caitlin Tulloch. 2013. Comparative
cost-eectiveness analysis to inform policy in developing countries: a general framework with
applications for education, Education policy in developing countries, 285-338.
Field, Erica M., Leigh L. Linden, Ofer Malmud, Daniel Rubenson and Shing-Yi Wang. 2019.
Does Vocational Education work? Evidence from a Randomized Experiment in Mongolia.
NBER Working Paper No. 26092. Cambridge, MA: NBER.

Goldin, Claudia and Larry Katz. 2009. The Race Between Education and Technology.
Cambridge, MA: Harvard University Press.

Goos, Maarten, Alan Manning, and Anna Salomons. 2014. Explaining Job Polarization:
Routine-biased Technological Change and Oshoring, American Economic Review, 104(8):
2509-26.

Groh, Matthew, Nandini Krishnan, David McKenzie and Tara Vishwanath. 2016. The
Impact of Soft Skills Training on Female Youth Employment: evidence form a Randomized
Experiment in Jordan, IZA Journal of Labor and Development, 5:9.

19
Hall, Caroline. 2016. Does More General Education Reduce the Risk of Future Unemploy-
ment? Evidence from an Expansion of Vocational Upper Secondary Education, Economics
of Education Review, 52: 251-271.
Hampf, Franziska. and Ludger Woessmann. 2017. Vocational vs. General Education and
Employment Over the Life Cycle: New Evidence from PIACC, CESifo Economic Studies,
63(3): 255-269.

Hanushek, Eric A., Guido Schwerdt, Ludger Woessmann, and Lei Zhang. 2017. General
Education, Vocational Education, and Labor-market Outcomes Over the Lifecycle, Journal
of Human Resources, 52(1): 48-87.
Heckman, James and Alan Krueger. 2005. Inequality in America: What Role for Human
Capital Policies? Cambridge, MA: MIT Press.

Hirshleifer, Sarojini, David McKenzie, Rita Almeida, and Cristobal Ridao-Cano, C. 2016.
The Impact of Vocational Training for the Unemployed: Experimental Evidence from
Turkey, Economic Journal, 126(597): 2115-2146.
Ibarraran, Pablo, Laura Ripani, Bibiana Taboada, Juan Miguel Villa, and Brigida Garcia.
2014. Life skills, employability and training for disadvantaged youth: Evidence from a
randomized evaluation design. IZA Journal of Labor & Development, 3(1), 10.
Kemple, James J. and Cynthia J. Willner. 2008.Career Academies: Long-term Impacts on
Labor Market Outcomes, Educational Attainment, and Transitions to Adulthood. New York,
NY: MDRC.

Krueger, Dirk and Krishna B. Kumar. 2004. Skill-specic Rather than General Education:
A Reason for US-Europe Growth Dierences? Journal of Economic Growth, 9(2): 167-207.
Kugler, Adriana, Maurice Kugler, Juan Saavedra and Luis Omar Herrera-Prada. 2020. The
Long-term Impacts and Spillovers of Training for Disadvantaged Youth, Journal of Human
Resources, forthcoming.
Malamud, Ofer and Cristian Pop-Eleches. 2011. School Tracking and Access to Higher
Education among Disadvantaged Groups, Journal of Public Economics, 95(11-12): 1538-
1549.

Malamud, Ofer and Cristian Pop-Eleches. 2010. General Education versus Vocational
Training: Evidence from an Economy in Transition, Review of Economics and Statistics,
92(1): 43-60.

20
Meer, Jonathan. 2007. Evidence on the returns to secondary vocational education, Eco-
nomics of Education Review, 26(5), 559-573.
Silliman, Mikko and Hanna Virtanen. 2019. Labor Market Returns to Vocational Secondary
Education, ETLA Working Papers. No 65.

21
 Table 1: Covariate balance check

Treatment Treatment arm 1: Treatment arm 2:

-control Social-control Technical-control
Control mean dierence dierence dierence
Male 0.34 0.03 0.02 0.04
(0.03) (0.04) (0.05) (0.05)
Age 26.21 -0.25 -0.43 -0.08
(0.44) (0.51) (0.58) (0.61)
Years of education 11.31 -0.01 -0.14 0.12
(0.11) (0.14) (0.16) (0.15)
Black 0.55 0.01 0.00 0.02
(0.03) (0.04) (0.05) (0.05)
Mestizo 0.17 -0.01 0.01 -0.03
(0.03) (0.03) (0.04) (0.03)
Indigenous 0.03 -0.00 -0.01 -0.00
(0.01) (0.01) (0.02) (0.02)
Disability 0.02 -0.01 0.00 -0.01
(0.01) (0.01) (0.02) (0.01)
Primary education 0.99 -0.01 -0.01 0.00
(0.01) (0.01) (0.01) (0.01)
Secondary education 0.94 -0.02 -0.03 -0.02
(0.02) (0.02) (0.03) (0.02)
Technical higher education 0.29 0.02 -0.02 0.05
(0.03) (0.04) (0.04) (0.04)
Professional higher education 0.02 0.01 0.00 0.01
(0.01) (0.01) (0.02) (0.02)
Enrolled in school 0.07 -0.01 -0.02 -0.00
(0.02) (0.02) (0.02) (0.02)
Using Public Employment Service 0.37 -0.04 -0.06 -0.01
(0.03) (0.04) (0.05) (0.05)
Household size 4.45 0.10 0.36 ** -0.15
(0.13) (0.15) (0.17) (0.17)
HH income per day (USD) 19.23 2.84 7.48 -1.23
(4.25) (5.65) (7.65) (4.99)
HH with electricity 1.00 -0.01 -0.01 -0.01 *
(0.00) (0.01) (0.01) (0.01)
HH with water 0.99 0.00 -0.00 0.01
(0.01) (0.01) (0.01) (0.01)
HH with sanitation 0.98 -0.00 -0.01 0.00
(0.01) (0.01) (0.02) (0.01)
Joint signicance F-test = 0.59 F-test = 0.95 F-test = 0.71
p-val = 0.9104 p-val = 0.5183 p-val = 0.7941

Observations 212 663 434 441

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

22
 Table 2: Pre- and post- : Labor market outcomes

Treatment Treatment 1: Treatment 2:

Control -control Soc.-control Tech.-control
mean dierence DD dierence DD dierence DD
Panel A: Pre-treatment
Employed 0.55 0.01 0.01 0.01
(0.03) (0.04) (0.05) (0.05)
With contract 0.28 0.01 0.04 -0.01
(0.03) (0.04) (0.04) (0.04)
Casual worker 0.44 -0.01 -0.01 -0.01
(0.03) (0.04) (0.05) (0.05)
Days worked per month 13.48 0.06 1.16 -1.04
(0.93) (1.10) (1.27) (1.26)
Hours worked per week 24.43 0.17 0.46 -0.11
(1.78) (2.17) (2.47) (2.51)
Wage per hour (USD) 1.19 0.14 0.63 -0.35
(0.48) (0.59) (0.81) (0.49)
Panel B: Post-treatment
Employed 0.40 0.09 ** 0.08 0.08 0.07 0.10 ** 0.10
(0.03) (0.04) (0.05) (0.05) (0.06) (0.05) (0.06)
With contract 0.17 0.14 *** 0.13 *** 0.11 *** 0.07 0.17 *** 0.17 ***
(0.03) (0.04) (0.05) (0.04) (0.05) (0.04) (0.05)
Casual worker 0.47 -0.16 *** -0.15 ** -0.16 *** -0.15 ** -0.16 *** -0.16 **
(0.03) (0.04) (0.06) (0.05) (0.07) (0.05) (0.07)
Days worked per month 9.54 2.73 ** 2.63 * 2.28 * 1.02 3.17 *** 3.17 ***
(0.85) (1.06) (1.41) (1.21) (1.55) (1.21) (1.62)
Hours worked per week 18.67 4.98 ** 4.81 * 3.76 3.25 6.17 ** 6.17 *
(1.68) (2.11) (2.83) (2.38) (3.10) (2.43) (3.27)
Wage per hour (USD) 0.44 0.01 -0.11 0.01 -0.60 0.01 0.01
(0.05) (0.05) (0.59) (0.06) (0.82) (0.06) (0.50)

Observations 212 663 663 434 434 441 441

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

23
 Table 3: Pre- and post- : Social skill outcomes

Treatment Treatment 1: Treatment 2:

Control -control Soc.-control Tech.-control
mean dierence DD dierence DD dierence DD
Panel A: Pre-treatment
Work ethic 3.78 -0.00 -0.03 0.03
(0.03) (0.03) (0.04) (0.03)
Organizational 3.60 -0.10 -0.11 -0.09
(0.05) (0.07) (0.08) (0.08)
Interpersonal 3.37 0.02 0.02 0.03
(0.03) (0.03) (0.04) (0.04)
Leadership 3.35 0.02 -0.01 0.04
(0.04) (0.04) (0.05) (0.05)
Teamwork 5.22 0.16 0.12 0.20
(0.10) (0.11) (0.13) (0.13)
Communication 3.43 -0.03 -0.09 * 0.04
(0.03) (0.04) (0.05) (0.04)
Panel B: Post-treatment
Work ethic 3.80 0.02 0.02 -0.00 0.03 0.04 0.04
(0.02) (0.02) (0.04) (0.03) (0.04) (0.03) (0.04)
Organizational 3.54 0.11 * 0.21 ** 0.08 0.19 * 0.13 ** 0.13 **
(0.05) (0.06) (0.08) (0.07) (0.10) (0.07) (0.09)
Interpersonal 3.46 0.04 0.02 0.02 -0.00 0.07 ** 0.07
(0.02) (0.03) (0.04) (0.03) (0.04) (0.03) (0.04)
Leadership 3.42 0.06 * 0.04 0.06 0.07 0.06 0.06
(0.03) (0.03) (0.05) (0.04) (0.06) (0.04) (0.05)
Teamwork 5.15 0.03 -0.13 0.04 -0.09 0.03 0.03
(0.08) (0.09) (0.13) (0.11) (0.16) (0.11) (0.14)
Communication 3.43 0.07 * 0.10 ** 0.06 0.15 *** 0.08 * 0.08
(0.03) (0.04) (0.05) (0.04) (0.06) (0.04) (0.05)

Observations 212 663 663 434 434 441 441

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

24
 Table 4: ITT estimates (Survey data)

Any treatment-control Social treatment-control Technical treatment-control

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Labor market outcomes

Employed 0.07 0.08 0.08 0.05 0.05 0.05 0.09 0.11 0.09
(0.04) (0.06) (0.06) (0.05) (0.06) (0.06) (0.06) (0.07) (0.07)
543 543 543 393 393 393 322 322 322
With contract 0.13*** 0.12** 0.12** 0.10** 0.09 0.09 0.17*** 0.17*** 0.16**
(0.04) (0.05) (0.05) (0.04) (0.06) (0.06) (0.05) (0.06) (0.06)
543 543 543 393 393 393 322 322 322
Casual worker -0.16*** -0.13** -0.13** -0.16*** -0.13* -0.14* -0.16*** -0.13* -0.13
(0.04) (0.06) (0.06) (0.05) (0.07) (0.07) (0.06) (0.08) (0.08)
543 543 543 393 393 393 322 322 322
Days worked per month 1.95* 2.25 2.23 1.59 1.06 0.97 2.53* 5.15*** 4.74**
(1.14) (1.54) (1.53) (1.26) (1.61) (1.61) (1.47) (1.97) (1.95)
542 523 523 393 384 384 321 305 305
Hours worked per week 3.44 4.66 4.61 2.56 3.91 3.78 4.99* 7.13* 6.32*
(2.26) (3.01) (2.99) (2.48) (3.22) (3.22) (2.94) (3.79) (3.76)
541 522 522 392 383 383 321 305 305
Wage per hour (USD) -0.04 -0.40 -0.39 -0.03 -0.88 -0.89 -0.05 -0.24 -0.28
(0.06) (0.73) (0.73) (0.07) (0.93) (0.93) (0.08) (0.72) (0.72)
541 498 498 392 366 366 321 288 288
Panel B: Social skill outcomes
Work ethic -0.01 0.00 0.00 -0.02 -0.00 -0.00 0.00 0.01 0.01
(0.02) (0.04) (0.04) (0.03) (0.05) (0.05) (0.03) (0.05) (0.05)
546 542 542 396 392 392 325 322 322
Organizational 0.08 0.24** 0.23** 0.07 0.20* 0.20* 0.12 0.31*** 0.28**
(0.07) (0.10) (0.10) (0.08) (0.11) (0.11) (0.09) (0.11) (0.11)
547 543 543 397 393 393 325 322 322
Interpersonal 0.02 -0.01 -0.01 0.01 -0.03 -0.03 0.03 -0.02 -0.02
(0.03) (0.04) (0.04) (0.03) (0.05) (0.05) (0.04) (0.05) (0.05)
547 543 543 397 393 393 325 322 322
Leadership 0.05 0.03 0.03 0.06 0.04 0.04 0.03 0.01 0.01
(0.04) (0.05) (0.05) (0.04) (0.06) (0.06) (0.05) (0.06) (0.06)
547 543 543 397 393 393 325 322 322
Teamwork -0.03 -0.24* -0.22 0.04 -0.17 -0.13 -0.16 -0.38** -0.36**
(0.10) (0.14) (0.14) (0.11) (0.17) (0.17) (0.13) (0.17) (0.17)
547 512 512 397 362 362 325 312 312
Communication 0.05 0.06 0.06 0.04 0.09 0.09 0.03 -0.02 -0.03
(0.04) (0.05) (0.05) (0.05) (0.06) (0.06) (0.05) (0.07) (0.07)
547 543 543 397 393 393 325 322 322
Controls
Baseline demographics Yes Yes Yes Yes Yes Yes Yes Yes Yes
Course FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Pre-treatment outcome No Yes Yes No Yes Yes No Yes Yes
Days from graduation No No Yes No No Yes No No Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

25
 Table 5: Vocational training and employment dynamics (Administrative data)

Days formally employed Months formally employed Monthly SS contributions

Panel A: Treatment-control
Treatment -0.33 -0.01 -5.00
(0.83) (0.03) (7.28)
Treatment X Post 2.16 ** 0.08 ** 20.66 **
(1.09) (0.04) (10.17)
Observations 663 663 663
Panel B: Treatment-control
Treatment -0.33 -0.01 -5.00
(0.83) (0.03) (7.28)
Treatment X Post 2.79 ** 0.13 *** 29.70 **
(1.25) (0.04) (11.78)
Treatment X Months-post -0.09 -0.01 * -1.29
(0.12) (0.00) (1.21)
Observations 663 663 663
Panel C: Social treatment-control
Social treatment -0.58 -0.02 -7.43
(0.96) (0.03) (8.24)
Soc. treat. X Post 1.72 0.11 ** 20.67
(1.39) (0.05) (12.92)
Soc. treat. X Months-post 0.05 -0.00 -0.13
(0.14) (0.00) (1.38)
Observations 441 441 441
Panel D: Technical treatment-control
Technical treatment -0.08 0.01 -2.49
(0.95) (0.03) (8.27)
Tech. treat. X Post 3.88 *** 0.15 *** 39.02 ***
(1.44) (0.05) (13.59)
Tech. treat. X Months-post -0.23 * -0.01 ** -2.48 *
(0.13) (0.00) (1.30)
Observations 434 434 434
Panel E: Social-technical
Social treatment -0.82 -0.04 -7.04
(0.68) (0.03) (5.77)
Soc. treat. X Post -1.83 -0.03 -16.26
(1.13) (0.04) (10.82)
Soc. treat. X Months-post 0.28 ** 0.01 * 2.35 **
(0.12) (0.00) (1.18)
Observations 451 451 451
Month FE Yes Yes Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

26
 Table 6: Social vs. technical treatment: Labor market outcomes

Technical treatment Social-technical Specication

mean dierence DD (1) (2) (3)
Employed 0.50 -0.02 -0.03 -0.03 -0.06 -0.10
(0.03) (0.05) (0.06) (0.06) (0.08) (0.08)
With contract 0.34 -0.06 -0.11 * -0.03 -0.04 -0.08
(0.03) (0.04) (0.06) (0.05) (0.07) (0.07)
Casual worker 0.31 0.00 0.00 -0.00 -0.01 -0.01
(0.03) (0.04) (0.06) (0.05) (0.08) (0.08)
Days worked per month 12.71 -0.89 -3.23 * -0.73 -2.38 -3.98 *
(0.86) (1.22) (1.67) (1.44) (2.04) (2.05)
Hours worked per week 24.83 -2.40 -3.14 -1.86 -1.73 -5.07
(1.74) (2.42) (3.34) (2.84) (3.99) (4.00)
Wage per hour (USD) 0.45 -0.00 -0.97 -0.01 -1.58 * -1.75 *
(0.04) (0.06) (0.67) (0.07) (0.91) (0.93)

Controls
Baseline demographics No No No Yes Yes Yes
Course FE No No No Yes Yes Yes
Pre-treatment outcome No No Yes No Yes Yes
Days from graduation No No No No No Yes
Observations 229 451 451 451 451 451

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

Table 7: Social vs. technical treatment: Social skills outcomes

Technical treatment Social-technical Specication

mean dierence DD (1) (2) (3)
Work ethic 3.84 -0.04 * 0.02 -0.01 0.00 0.01
(0.02) (0.02) (0.04) (0.03) (0.05) (0.05)
Organizational 3.67 -0.06 -0.04 -0.01 -0.09 -0.15
(0.04) (0.06) (0.10) (0.08) (0.13) (0.13)
Interpersonal 3.53 -0.05 -0.04 -0.03 -0.06 -0.07
(0.02) (0.03) (0.04) (0.04) (0.05) (0.05)
Leadership 3.47 -0.00 0.06 0.01 0.02 0.01
(0.03) (0.04) (0.05) (0.05) (0.07) (0.07)
Teamwork 5.18 0.01 0.07 0.13 0.05 0.00
(0.07) (0.10) (0.14) (0.13) (0.16) (0.17)
Communication 3.51 -0.02 0.12 ** -0.04 0.04 0.01
(0.03) (0.04) (0.05) (0.05) (0.07) (0.07)

Controls
Baseline demographics No No No Yes Yes Yes
Course FE No No No Yes Yes Yes
Pre-treatment outcome No No Yes No Yes Yes
Days from graduation No No No No No Yes
Observations 229 451 451 451 451 451

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

27
 Table 8: Heterogeneity by stipend receipt: Labor market outcomes

Employment Formal contract Casual worker Days worked Hours Wages

Treatment -0.04 0.08 -0.10 -0.50 -1.47 -1.04
(0.07) (0.06) (0.08) (1.90) (3.71) (0.93)
StipendXTreatment 0.28*** 0.08 -0.05 6.12** 13.65*** 1.39
(0.10) (0.09) (0.11) (2.56) (4.99) (1.22)
Observations 543 543 543 523 522 498

Controls
Baseline demographics Yes Yes Yes Yes Yes Yes
Course FE Yes Yes Yes Yes Yes Yes
Pre-treatment outcome Yes Yes Yes Yes Yes Yes
Days from graduation Yes Yes Yes Yes Yes Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

Table 9: Heterogeneity by gender: Labor market outcomes

Employment Formal contract Casual worker Days worked Hours Wages

Panel A: Main eects by gender
Treatment 0.26** 0.20* -0.21* 5.21* 8.85 -0.36
(0.11) (0.10) (0.12) (2.90) (5.68) (1.37)
Female 1.20** 0.09 -1.07** 19.04 44.10* -11.16*
(0.49) (0.44) (0.54) (12.64) (24.75) (5.95)
FemaleXTreatment -0.26* -0.11 0.11 -4.23 -6.02 -0.05
(0.14) (0.12) (0.15) (3.50) (6.85) (1.67)
Observations 543 543 543 523 522 498
Panel B: Social treatment by gender
Treatment 0.26** 0.19* -0.16 4.57 8.81 -1.26
(0.12) (0.11) (0.14) (2.98) (5.95) (1.71)
Female 1.06* 0.08 -0.87 18.07 39.52 -13.35*
(0.55) (0.50) (0.64) (13.97) (27.96) (7.99)
FemaleXTreatment -0.29** -0.15 0.03 -5.29 -7.39 0.54
(0.15) (0.13) (0.17) (3.67) (7.35) (2.12)
Observations 393 393 393 384 383 366
Panel C: Technical treatment by gender
Treatment 0.19 0.16 -0.28* 6.83* 6.66 0.18
(0.15) (0.13) (0.17) (3.96) (7.62) (1.45)
Female 1.53** -0.37 -1.41* 24.39 45.49 5.86
(0.67) (0.58) (0.74) (17.20) (33.16) (6.23)
FemaleXTreatment -0.14 -0.00 0.20 -2.76 -0.44 -0.61
(0.18) (0.15) (0.19) (4.54) (8.75) (1.67)
Observations 322 322 322 305 305 288

Controls
Baseline demographics Yes Yes Yes Yes Yes Yes
Course FE Yes Yes Yes Yes Yes Yes
Pre-treatment outcome Yes Yes Yes Yes Yes Yes
Days from graduation Yes Yes Yes Yes Yes Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

28
 Table 10: Heterogeneity by gender: Social skill outcomes

Work ethic Organizational Interpersonal Leadership Teamwork Communication

Panel A: Main eects by gender
Treatment -0.08 0.52*** 0.06 0.02 0.04 0.11
(0.08) (0.19) (0.08) (0.10) (0.28) (0.10)
Female 0.10 -1.16 0.11 -0.13 -1.12 -0.47
(0.33) (0.81) (0.35) (0.45) (1.24) (0.45)
FemaleXTreatment 0.11 -0.40* -0.11 0.02 -0.35 -0.07
(0.09) (0.22) (0.10) (0.12) (0.33) (0.12)
Observations 542 543 543 543 512 543
Panel B: Social treatment by gender
Treatment -0.09 0.41** 0.04 -0.00 0.14 0.13
(0.09) (0.20) (0.09) (0.12) (0.32) (0.11)
Female 0.05 -1.30 -0.21 -0.25 -1.44 -0.52
(0.41) (0.94) (0.41) (0.55) (1.52) (0.52)
FemaleXTreatment 0.13 -0.30 -0.10 0.06 -0.37 -0.05
(0.11) (0.25) (0.11) (0.14) (0.39) (0.14)
Observations 392 393 393 393 362 393
Panel C: Technical treatment by gender
Treatment -0.06 0.59** 0.03 0.05 -0.33 -0.08
(0.10) (0.24) (0.11) (0.14) (0.36) (0.14)
Female -0.29 -1.97* -0.07 -0.19 -0.29 -0.63
(0.42) (1.03) (0.46) (0.59) (1.56) (0.60)
FemaleXTreatment 0.09 -0.41 -0.06 -0.05 -0.04 0.07
(0.11) (0.27) (0.12) (0.15) (0.41) (0.16)
Observations 322 322 322 322 312 322

Controls
Baseline demographics Yes Yes Yes Yes Yes Yes
Course FE Yes Yes Yes Yes Yes Yes
Pre-treatment outcome Yes Yes Yes Yes Yes Yes
Days from graduation Yes Yes Yes Yes Yes Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

29
 Table 11: Cost-eectiveness

Aggregate Social treatment Technical treatment

Percent Percent Percent
Total Per 100 USD per 100 USD Total Per 100 USD per 100 USD Total Per 100 USD per 100 USD
(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Costs
Program cost 164.44 161.22 168.00
Marginal cost 157.65 154.43 160.86
Stipend (x0.5) 15.00 15.00 15.00
Marginal cost w/ stipend 172.65 100 169.43 100 175.86 100
Panel A: ITTs Estimates from Survey Data
Employed 0.08 - - 0.05 - - 0.09 - -
With contract 0.12** 0.07 41 0.09 - - 0.16** 0.09 53

30
Casual worker -0.13** -0.08 -16 -0.14* -0.08 -18 -0.13 - -
Days worked per month 2.23 - - 0.97 - - 4.74** 2.70 28
Hours worked per week 4.61 - - 3.78 - - 6.32* 3.59 19
Wage per hour (USD) -0.39 - - -0.89 - - -0.28 - -
Wage per month (USD) 21.94 - - 18.00 - - 30.08* 17.11 19
Panel C: ITTs Estimates from Adminstrative Data
Employed 0.08*** 0.05 14 0.09** 0.05 16 0.08*** 0.04 13
Days of contributions 2.16*** 1.25 14 2.05* 1.21 13 2.27*** 1.29 14
Contributions 20.66*** 11.97 13 19.73** 11.64 13 21.63*** 12.30 14
Panel D: Present Discount Value for 35 Years
Discount rate of 5% 4,262 2,468 4,070 2,402 4,462 2,537
Discount rate of 5% 1,186 687 1,132 669 1,242 706
with 10% depreciation

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01). Columns 1 for all three groups come from the main specications in Tables 4 and 5.
 Figure 1: Descriptive graphs from administrative registers

15
Training
period

Days of formal employment

10
Baseline
survey

5
Endline
0 survey

17

18

19
7

9
1

1
0

0
20

20

20
/2

/2

/2
6/

6/

6/
12

12

12
Control Soc. treatment Tech. treatment

(a) Days of employment

.5
Formal employment
.3 .2
.1 .4

9
17

18

19
01

01

01
20

20

20
/2

/2

/2
6/

6/

6/
12

12

12

Control Soc. treatment Tech. treatment

(b) Employment
150
100
USD
50
0

9
17

18

19
01

01

01
20

20

20
/2

/2

/2
6/

6/

6/
12

12

12

Control Soc. treatment Tech. treatment

(c) Contributions

Note s: Figure 1 displays mean outcomes by treatment group. As shown in Figure 1(a), the gray vertical

regions mark the baseline and endline data collection, and the red vertical lines indicate the beginning and

end of the treatment periods.

31
 Days of formal employment Days of formal employment Days of formal employment
6/
20 -2 0 2 4 6 8 6/
20 -4 -2 0 2 4 6 6/
20 -2 0 2 4 6
17 17 17

12 12 12
/2 /2 /2
01 01 01
7 7 7

6/ 6/ 6/
20 20 20
18 18 18

12 12 12
/2 /2 /2
01 01 01
8 8 8

6/ 6/ 6/
20 20 20
19 19 19

(a) Days of employment

(a) Days of employment
(a) Days of employment
12 12 12
/2 /2 /2
01 01 01
9 9 9

Formal employment Formal employment Formal employment

6/
20 -.1 0 .1 .2 .3 6/
20 -.1 0 .1 .2 .3 6/
20 -.1 0 .1 .2 .3
17 17 17

12 12 12
/2 /2 /2
01 01 01
7 7 7

6/ 6/ 6/
20 20 20
18 18 18

32
12 12 12
/2 /2 /2
01 01 01
8 8 8

(b) Employment
(b) Employment
(b) Employment

6/ 6/ 6/
20 20 20
19 19 19

12 12 12
/2 /2 /2
01 01 01
9 9 9

USD USD USD

6/
20 -20 0 20 40 60 80 6/
20 -40 -20 0 20 40 60 6/
20 -20 0 20 40 60 80
17 17 17

12 12 12
/2 /2 /2
01 01 01
7 7 7

6/ 6/ 6/
20 20 20
18 18 18

12 12 12
/2 /2 /2
01 01 01
8 8 8
Figure 2: ITT graphs from administrative registers: Treatment-control

Figure 3: ITT graphs from administrative registers: Social treatment-control

(c) Contributions
(c) Contributions
(c) Contributions

6/ 6/ 6/
20 20 20
19 19 19
Figure 4: ITT graphs from administrative registers: Technical treatment-control

12 12 12
/2 /2 /2
01 01 01
9 9 9
 Days of formal employment
6/
20 -6 -4 -2 0 2 4
17

12
/2
01
7

6/
20
18

12
/2
01
8

6/
20
19

(a) Days of employment

12
/2
01
9

Formal employment
6/
20 -.2 -.1 0 .1
17

12
/2
01
7

6/
20
18

33
12
/2
01
8
(b) Employment

6/
20
19

12
/2
01
9

USD
6/
20 -60 -40 -20 0 20
17

12
/2
01
7

6/
20
18

12
/2
01
8
(c) Contributions

6/
20
19

12
/2
01
9
Figure 5: ITT graphs from administrative registers: Social treatment - technical treatment
Appendix

Table 1: ITT estimates (Survey data): FDR q-values

Any treatment-control Social treatment-control Technical treatment-control

(1) (2) (3) (4) (5) (6) (7) (8) (9)

Panel A: Labor market outcomes

Employed 0.07 0.08 0.08 0.05 0.05 0.05 0.09 0.11 0.09
(0.04) (0.06) (0.06) (0.05) (0.06) (0.06) (0.06) (0.07) (0.07)
543 543 543 393 393 393 322 322 322
With contract 0.13*** 0.12 0.12 0.10** 0.09 0.09 0.17*** 0.17** 0.16*
(0.04) (0.05) (0.05) (0.04) (0.06) (0.06) (0.05) (0.06) (0.06)
543 543 543 393 393 393 322 322 322
Casual worker -0.16*** -0.13 -0.13 -0.16*** -0.13 -0.14 -0.16** -0.13 -0.13
(0.04) (0.06) (0.06) (0.05) (0.07) (0.07) (0.06) (0.08) (0.08)
543 543 543 393 393 393 322 322 322
Days worked per month 1.95 2.25 2.23 1.59 1.06 0.97 2.53 5.15** 4.74*
(1.14) (1.54) (1.53) (1.26) (1.61) (1.61) (1.47) (1.97) (1.95)
542 523 523 393 384 384 321 305 305
Hours worked per week 3.44 4.66 4.61 2.56 3.91 3.78 4.99 7.13* 6.32
(2.26) (3.01) (2.99) (2.48) (3.22) (3.22) (2.94) (3.79) (3.76)
541 522 522 392 383 383 321 305 305
Wage per hour (USD) -0.04 -0.40 -0.39 -0.03 -0.88 -0.89 -0.05 -0.24 -0.28
(0.06) (0.73) (0.73) (0.07) (0.93) (0.93) (0.08) (0.72) (0.72)
541 498 498 392 366 366 321 288 288
Panel B: Social outcomes
Work ethic -0.01 0.00 0.00 -0.02 -0.00 -0.00 0.01 0.01 0.01
(0.02) (0.04) (0.04) (0.03) (0.05) (0.05) (0.03) (0.05) (0.05)
542 542 542 392 392 392 322 322 322
Organizational 0.10 0.24 0.23 0.09 0.20 0.20 0.12 0.31* 0.28
(0.07) (0.10) (0.10) (0.08) (0.11) (0.11) (0.09) (0.11) (0.11)
543 543 543 393 393 393 322 322 322
Interpersonal 0.02 -0.01 -0.01 0.01 -0.03 -0.03 0.03 -0.02 -0.02
(0.03) (0.04) (0.04) (0.03) (0.05) (0.05) (0.04) (0.05) (0.05)
543 543 543 393 393 393 322 322 322
Leadership 0.06 0.03 0.03 0.06 0.04 0.04 0.05 0.01 0.01
(0.04) (0.05) (0.05) (0.04) (0.06) (0.06) (0.05) (0.06) (0.06)
543 543 543 393 393 393 322 322 322
Teamwork -0.03 -0.24 -0.22 0.04 -0.17 -0.13 -0.17 -0.38* -0.36
(0.10) (0.14) (0.14) (0.12) (0.17) (0.17) (0.14) (0.17) (0.17)
543 512 512 393 362 362 322 312 312
Communication 0.03 0.06 0.06 0.02 0.09 0.09 0.00 -0.02 -0.03
(0.04) (0.05) (0.05) (0.05) (0.06) (0.06) (0.05) (0.07) (0.07)
543 543 543 393 393 393 322 322 322
Controls
Course FE Yes Yes Yes Yes Yes Yes Yes Yes Yes
Baseline measures No Yes Yes No Yes Yes No Yes Yes
Days from graduation No No Yes No No Yes No No Yes

Notes: Signicance levels (* = 0.10, **= 0.05, *** = 0.01).

34