American Economic Journal: Applied Economics 2017, 9(2): 131–143

[Link]

Vocational Training for Disadvantaged Youth in Colombia:

A Long-Term Follow-Up†
By Orazio Attanasio, Arlen GuarÍn, Carlos Medina, and Costas Meghir*

We evaluate the ­ long-term impacts of a randomized Colombian

training and job placement program. Following the large ­short-term
effects, we now find that the program effects persist, increasing formal
participation and earnings contributions to social security and
working in larger firms. By using a large administrative source we
are also able to establish that the program improved both male and
female labor market outcomes by a similar amount—a result that was
not apparent with the smaller evaluation sample. The results point to
a cost-effective approach to reducing informality and improving labor
market outcomes in the long run. (JEL I28, J13, J24, J31, O15)

S kill acquisition and participation in the formal labor market is a central policy
concern in developing countries. From a growth perspective, a ­­well-functioning
economy that is able to attract investment requires individuals with sufficient lev-
els of skills and work readiness (see, for instance, Organisation for Economic
Co-operation and Development (OECD) 2012). Moreover, unskilled youth have
particularly low labor market attachment and are vulnerable to poverty. Identifying
interventions that can promote human capital acquisition and improve labor market
outcomes in the long run has been elusive, with generally mixed results, at least in
­high-income countries.
In Colombia, a vocational training program, Jóvenes en Acción (JeA), carried out
in 2005 and evaluated with a randomized controlled trial in Attanasio, Kugler, and
Meghir (2011)—henceforth, AKM—was shown to improve formal sector employ-
ment and pay, particularly for young women for whom the implied internal rate of
return of the program was at least 21.6 percent. For men, however, there was no
discernible effect other than a shift to the formal sector.

* Attanasio: Department of Economics, University College London, Gower Street, London WC1E6BT, UK
(e-mail: [Link]@[Link]); Guarín: Banco de la República, Medellín, Calle 50 N 50-21 (e-mail: aguariga@
[Link]); Medina: Banco de la República, Medellín, Calle 50 N 50-21 (e-mail: cmedindu@[Link].
co); Meghir: Yale University, 37 Hillhouse Avenue, New Haven, CT 06511 (e-mail: [Link]@[Link]). This
is a revised version of NBER working paper 21390 (July 2015) “Long Term Impacts of Vouchers for Vocational
Training: Experimental Evidence for Colombia.” We thank an anonymous referee for helpful comments and sug-
gestions. We also thank Ana María Iregui and participants at an internal seminar at Banco de la República and at
the CEDE of Universidad de los Andes for their comments. We are grateful to the Ministry of Health and Social
Protection and the Department of National Planning for providing us access to the PILA and SISBEN datasets. All
errors are the responsibility of the authors. Meghir was funded by the Cowles Foundation and the ISPS at Yale.
Attanasio was funded by the ESRC Professorial Fellowship ES/K010700/1, “The accumulation of human capital
in developing countries.”
†
Go to [Link] to visit the article page for additional materials and author
disclosure statement(s) or to comment in the online discussion forum.

131
132 American Economic Journal: applied economicsapril 2017

The program, which combined classroom training with an internship, was imple-
mented in the wake of the worst recession Colombia experienced since WWII in
­­1999–2000.1 It was provided mainly by private sector training institutions that could
screen the applicants for suitability. In turn, the payments to the training institutions
depended on successful completion of the program and on successful professional
placement of the applicants.
In this paper, we carry out a long-term follow-up of the AKM study, based on
administrative records for the period July 2008 to December 2014. In addition to the
original evaluation sample, covering a random subset of those who participated in
the experiment, we also analyze the entire cohort of eligible individual applicants
who were part of the randomization, substantially improving precision.2
The nature of the administrative data is such that workers are only observed if
they are in the formal sector. Thus, our evaluation will focus on ­­long-term attach-
ment to the formal sector and on the resulting earnings and other related outcomes,
such as pension contributions. Informality is a major issue around the world and in
Latin America in particular. Firms that operate in the formal sector tend to be higher
productivity, offer higher pay (see Meghir, Narita, and Robin 2015), and often other
benefits such as training and improved professional opportunities. Indeed, reallocat-
ing workers to the formal sector can improve growth. However, informality is hugely
prevalent: according to Medina, Núñez, and Tamayo (2013) since the mid-1980s
informality (based on the International Labour Organization (ILO) definition) has
fluctuated in Colombia from 53 percent to 60 percent and is particularly high among
the young (see Saavedra and Medina 2014) at a time which is important for laying
the foundations of a productive career.3
The results show sustained improvements for trainees over long periods of time,
making it unique (to our knowledge) in the adult training literature. We find that the
JeA program had a positive and significant effect on the proportion working in the
formal sector; it increased contributions to social security and other programs, and
increased formal sector earnings by 12 percent overall. Importantly, using the data
from the entire experimental cohort, we are able to establish that the effects for men are
significant and in some cases significantly larger than those for women. Beneficiaries
of the JeA program were also more likely to work for a large firm, one with at least
200 employees, inducing possibly an overall increase in the quality of jobs. On the
other hand, we have no ­­long-term information on how those not working in the infor-
mal sector are faring and this precludes a complete evaluation of the program.

See Gaviria Uribe and Núñez Mendez (2003); Barrera-Osorio and Corchuelo (2003); Medina and Núñez
1

(2005); Sarmiento et al. (2007); Santa María et al. (2009); and Steiner, Rojas, and Millán (2010), among others, for
issues relating to training in Colombia. According to Székely Pardo (2012), 20 percent of Colombian youths 15–18
­­
years old were not in education, employment, or training by 2005, while Guarín and Medina (2015) show that in
Colombian cities like Medellín, the same figure is as high as 44 percent for youth ­­16–20 years old.
2
Other studies on training in Latin America include Card et al. (2011), and Ibarraran et al. (2014) who provide
experimental results for the ­­short term, while Alzúa, Cruces, and Lopez (2015) present experimental evidence three
years after graduation. This is the first study to look beyond this horizon, to the best of our knowledge. Subsequent
to the first draft of our paper we learned that Kugler et al. (2015) also study the ­­long-term impacts of the same
program but based on the evaluation sample alone.
3
See also Perry at al. (2007), Cárdenas and Mejía (2007), and López Castaño(2010).
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 133

One concern is that some of the observed effects do not represent an increase in
overall economic activity, but could arise from the displacement of other ­workers
who are close substitutes of the trainees. Although evidence in Blundell et al. (2004)
showed no displacement effects in the United Kingdom, a specially designed ran-
domized experiment evaluating a job placement program in France showed evidence
of job displacement (see Crépon et al. 2013). If the training does improve human
capital, such displacement should only be a concern in the short run; in the longer
run the economy should expand in response to increased resources. However, if all
that happens is to create a channel through which some workers get better access
to the formal sector, while others with identical skills do not, displacement can be
a real concern even in the longer run. AKM discuss the possibility of displacement,
and they found no sign of displacement effects. Here, we present further evidence
that suggests that the effects are not due to displacement, although we can only rely
on observational data for this purpose.
The rest of the paper is organized as follow: in Section I, we describe the inter-
vention; in Section II, we describe the data; in Section III, we present the estimated
effects of the program. Section IV concludes.

I. The Intervention

The Jóvenes en Acción program was a training program for urban young unem-
ployed in Colombia. It was targeted to unemployed youths 18 to 25, who belonged to
the poorest population classified in the two lowest levels of a score, called SISBEN,
which is used in Colombia to target all welfare programs.4 The program was imple-
mented in the seven main cities of the country.5 It began to enroll students in 2002,
and, by 2005, it had enrolled 80,000 students.
The goals of the program, which we describe in more detail in online Appendix A,
were to develop the youths’ occupational skills, increasing their employability and
productivity, to promote the private supply of training, and to improve the matches
between workers and firms. Jóvenes en Acción consisted of training courses pro-
vided by private institutions (Entidades de Capacitación (ECAP)). Each course was
expected to train about 30 unemployed youths, chosen by the ECAPs among eli-
gible applicants. For evaluation purposes, in 2005, the ECAPs were encouraged to
select more than 30 applicants; the courses were offered to 30 applicants randomly
selected from this group.
The course had to have three main components: classroom training;
­­on-the-job training; and the youth’s project of life (Fondo de Inversión para la Paz
(FIP) 2001 and Departamento Nacional de Planeación (DNP) 2016). The program
also included a small stipend of about US$2.20 per day for trainees without children

4
SISBEN is the acronym in Spanish for Information System for Beneficiaries Selection, and it consists of six
levels built with the quality of life SISBEN score, used in Colombia to target public subsidies. To apply to the pro-
gram, individuals were additionally required to have a valid ID, and if the applicant was a mother of children under
seven years of age, she must present an official document to prove her maternity. Applicants that had previously
taken training courses at Servicio Nacional de Aprendizaje (SENA) or any training institution were also eligible
for Jóvenes en Acción.
5
Bogotá, Medellín, Cali, Barranquilla, Bucaramanga, Manizales, and Cartagena.
134 American Economic Journal: applied economicsapril 2017

under seven years of age, and about US$3.00 per day for women with children under
seven. In 2005 there were 114 ECAPs offering 441 courses to 26,615 trainees, with
their instructors teaching about 7.6 hours per day.6
A large fraction of the overall payment to the ECAPs was conditioned on the stu-
dent completing a three-month
­­ apprenticeship with the participating firms in a timely
fashion (see FIP 2011 and Departamento Administrativo para la Prosperidad Social
2016). Further bonus payments depended on the firms hiring the trainee on formal
contract. In turn, the ECAPs would define suitable candidates for training amongst
those eligible. This allowed them to identify those most likely to succeed. The incen-
tive scheme, combined with ­­prescreening, was an important departure from the stan-
dard practice at SENA—the Colombian government training agency (see Saavedra
and Medina 2013)—and a key feature of this program.
The total cost of the program to the government is the sum of the cost of the
course and the maintenance stipend to the students, which lasted six months (see
FIP 2011 and Departamento Administrativo para la Prosperidad Social 2016). To
this, we need to add any forgone earnings of participants. The total amount invested
was US$22 million (see Ministerio de la Protección Social 2005).

II. The Experiment and the Data

A. The Random Assignment to the Jóvenes en Acción Program

For the purpose of evaluating the intervention by random assignment to treat-

ment, the ECAPs were asked to select up to 50 percent more applicants than the
places available for the courses for the 2005 cohort of trainees. Since the ECAPS
were only paid for those who completed the training, they had strong incentives to
select individuals accordingly. Two-thirds of the selected applicants were then ran-
domly assigned to the program they applied for, while the remaining one-third were
assigned to a control group.

B. The Data

The original evaluation of the program was based on a random sample drawn
from the population that was screened by the ECAPs and thus participated in the
experiment. In this study, we use a combination of this evaluation sample and
administrative data for the long-run follow-up. We are also able to use the entire
experimental cohort, increasing the sample size tenfold.

The Evaluation Sample.—The baseline evaluation data includes 2,066 individuals

in the treatment group and 2,287 in the control, sampled randomly from the entire
population of the eligible. The baseline data was collected in January 2005 and the
follow-up in August to September 2006. Further details can be found in AKM, who
also show that it was well-balanced between treatment and control.

6
Forty percent of the beneficiaries were from Bogotá, 18 percent from Medellín, 16 percent from Cali, 11 per-
cent from Barranquilla, 7 percent from Bucaramanga, 5 percent from Cartagena, and 2 percent from Manizales.
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 135

Administrative Records.—No further follow-up was collected by survey. However,

we can carry out the long-term evaluation using two alternative sources: first, we can
link those in the evaluation sample to administrative data for 2008–2014;
­­ second,
we can measure some outcomes for the entire experimental cohort, which is almost
ten times the size of the evaluation sample, by linking them to administrative records
in 2010.

The SISPRO Data: The administrative data source is the SISPRO, which con-
tains information from the Unified Register of Contributions, known as PILA. This
is the national information system used by firms to file the mandatory contribu-
tions to health, pensions, and disability insurance they pay for their employees. The
Ministry of Social Protection has universal coverage of all the monthly contributions
in the country since July 2008, which provides us with the possibility of following
up individuals in the evaluation survey and constructing a longitudinal database with
monthly frequency from July 2008 to December 2014. Since the PILA contains
the monthly census of all contributions in the country, individuals in the evaluation
database that were not found in the PILA register are either out of work, or working
in the informal labor market. The structure of the data also allows us to observe
individuals entering and exiting the formal labor market based on the date of entry
and exit from the register.
For those in the experimental population who are not in the evaluation sample we
can only link records in 2010. Thus, if an individual is working in the formal sector
in 2010, we can observe them. If they are not matched in that year, we know that in
that year they were either not working or an informal worker.
In Table 1, we show the sample sizes for the evaluation sample and the entire
cohort. The experimental population consisted of 33,929 screened and eligible
applicants. The original evaluation sample included a random sample of those, half
drawn from the controls and half from the treatment. Nearly all have valid baseline
information and are used here. Overall, 54 percent were women; the average age is
21; and the average education is 10 years. For the entire experimental population
we matched 31,054 with the baseline administrative data. Consistent with the origi-
nal randomization, about two-thirds are treated and one-third are controls. Detailed
descriptive statistics are in the online Appendix.

Balance at Baseline: For both women and men, the baseline characteristics are
well-balanced in the evaluation sample. This was originally documented by AKM
and also shown in Table A1a in the online Appendix. Moreover, Table A1b in the
online Appendix compares baseline characteristics measured in the SISBEN data
for the entire cohort. Again, for both women and men, the sample is well-balanced,
except that beneficiaries are slightly younger.

III. Program Effects

We start with a graphical representation of the impacts of the program. Figure 1

shows the difference in formal employment rates between the treatment and control
group over time and by gender; the graph includes a 95 percent point wise confidence
136 American Economic Journal: applied economicsapril 2017

Table 1—Sample Sizes

Evaluation sample Entire cohort

Total Treated Control Total Treated Control
Screened and eligible applicants 3,956 2,040 1,916 33,929 22,438 11,491
Individuals with valid baseline informationa 3,932 2,029 1,903 31,054 20,548 10,506
Proportion female 54 percent 70 percent
a For the entire cohort: matched to baseline SISBEN.

Females (evaluation sample) Females (entire cohort)

0.15

0.10

0.05

0.00

−0.05
0

ch 0

ril 10
ay 0

ne 0

A ly 0

0
be 010

0
y 8
nu y 2 9
y 9
Ja Jul 010

y 0
nu y 2 1
y 1
Ja ul 12

y 2
nu y 2 3
y 3
ly 14
14

pt ust 0
O ber 10
ov ber 10
ar 01

ua 01
ar 01

M 01
Ju 201

Ju 201

D mb 01

01
ar 00
Ja ul 00
ar 00

ar 01
Ja Jul 01
ar 01

ar 01
Ja ul 01

Se ug 201
20

Ju 20
20

Ap 20

0
20
nu y 2
2

M y2

em 2

2
em r 2
r2
nu y 2
2

2
nu 2
2

Fe ary
y

ec e
r
Ja Jul

nu

o
br

ct

e
Ja

N
Males (evaluation sample) Males (entire cohort)
0.15

0.10

0.05

0.00

−0.05
14

0
ch 0

ril 10
ay 0

ne 0

A ly 0
pt ust 0
O ber 10
ov ber 10
0
be 010

0
y 8
nu y 2 9
y 9
Ja Jul 010

y 0
nu y 2 1
y 1
Ja ul 12

y 2
nu y 2 3
y 3
ly 14

ua 201
ar 01

M 01
Ju 201

Ju 201
Se ug 201

D mb 01

01
ar 00
Ja ul 00
ar 00

ar 01
Ja Jul 01
ar 01

ar 01
Ja ul 01
ar 01
20

Ju 20
20

Ap 20

0
20
M y2

em 2

2
em r 2
r2
nu y 2
2

2
nu 2
2

nu y 2
2

y
y

ec e
Fe ar

r
Ja Jul

nu

o
br

ct

e
Ja

Figure 1. Treatment Effect on Formal Employment by Treatment and Gender

Source: ­­JEA-SISPRO

interval around the estimated impact. While the results for the evaluation sample are
much noisier than those of the full sample, they are both implying a positive effect,
which is stable over time.7 The results with the entire cohort show clearly that these
are significant effects. Similar patterns obtain if we plot the effect for formal e­ arnings.
The important conclusion is that the increased attachment to the formal sector caused
by the program seems to be sustained for both women and for men, despite the fact
that the original evaluation did not identify significant effects for men.

7
The evaluation sample and the entire cohort contain a different proportion of women.
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 137

A. Program Estimates and Inference

In what follows, we present average results for the evaluation sample and the
entire cohort, based on regression analysis using the following model:

​Y​ij​​  = α​Di​​​ + ​Xi​​​  β + S​Cj​​​ + ​εij​ ​​​  ,

(1) ​

where ​​Yi​j​​​ is the outcome of person i​ ​in site and course j​ ​; ​​D​i​​​ is the treatment indicator,
which is one for those assigned to treatment and zero otherwise; and S​​C​j​​​ are site by
course fixed effects. ​​X​i​​​ is a vector of characteristics of the applicants at baseline.
The ​​Xi​​​​ differ depending on whether we use the evaluation sample, where we have
detailed baseline information, or the entire cohort, where the baseline characteris-
tics are those reported in the SISBEN8 survey. Characteristics used as controls are
reported in the notes to the tables.
The randomization was at the individual level and it took place within the
population of eligible and screened applicants for each course. We thus include
­­site-by-course fixed effects (S​​C​j​​​). The estimates we present will be a weighted
average (​α​) of intent-to-treat effects across all courses.9 However, since compli-
ance among applicants is 97 percent, our estimates should not differ substantially
from the average treatment effect in this population. We weigh each applicant by​​
P​SCi​​​ ​(­­1 − ​P​SCi​​)​, where ​​P​SCi​​​ is the population probability taken from administrative
records that an applicant ​i​that applied to course ​C​at site ​S​be offered a place.10
Inference is based on 1,000 bootstrap replications stratified by city, gender, and
treatment status. We report standard errors and p-values ­­ adjusted for multiple test-
ing following the step-down ­­ procedure described in Romano and Wolf (2005) and
Romano, Shaikh, and Wolf (2008). Each table of results represents a group being
jointly tested for the entire sample and then also for men and women separately.
p-values are then interpreted as the significance level that would have to be applied
to the entire family of hypotheses if we were to accept the null that the effect is zero.

B. Effects of the Jóvenes en Acción on Participation in the Formal Labor Market

We focus on three outcome variables: whether an individual is working in a for-

mal job in any one month, whether they are working in a large formal firm, and their
earnings. As before, and in everything that follows, the outcomes are set to zero for
those who have no formal job recorded.
The results for men and women combined are reported in Table 2, while results
by gender are shown in Table 3. In Table 2, the ­­p-values are adjusted for testing the
three outcomes in that table. In Table 3, the ­­Romano-Wolf p-values
­­ account for test-
ing of all six outcomes reported there. All levels of significance are reported based
on the ­­Romano-Wolf ­­p-values and not on the standard t-statistics.
­­

8
We explain the nature of the SISBEN survey in the online Appendix.
9
The randomization probability is in principle the same across courses, so it is not strictly necessary to include
such fixed effects. They have little or no influence on the results.
10
While we report the results obtained using this weighting scheme, we also compute unweighted estimates,
which are available on request. Weighting does not make any substantial difference to our results.
138 American Economic Journal: applied economicsapril 2017

Table 2—Labor Market Outcomes

Evaluation sampleb Entire cohortc

Coefficient on Coefficient on
Control being offered Control being offered
means training means training
Formal incomea 258,922 35,331 220,126 26,825
(432,480) (10,766) (393,121) (4,267)
[0.025] [0.000]

Working in the 0.36 0.042 0.33 0.038

formal sector (0.012) (0.005)
[0.009] [0.000]

Working in 0.189 0.032 0.174 0.027

large formal (0.010) (0.005)
  sector firm [0.025] [0.000]

Observationsd 148,746 306,696 126,072 372,648

Notes: Romano-Wolf p-values (in square brackets) for this group (three coefficients). Standard
errors (in parentheses) are clustered at the applicant level. All regressions control for site-by-
course fixed effects.
a COP$ of 2013.
b These regressions control for the following pretreatment characteristics interacted
with gender: age, education, marital status, employment, paid employment, salary,
­self-employment earnings, whether working in the formal sector, whether working with a
contract, days worked per month, and hours worked per week. Romano-Wolf p-values for
this group. Standard errors are clustered at the applicant level.
c These regressions control for the following pretreatment characteristics interacted with
gender: whether the applicant is in a low socioeconomic stratum, whether living in a
house or apartment, whether living at home without threats, age in 2005, homeownership,
household size, education of the head of the household, age in 2005 of the head of the
household, number of children under 5 years old and number of adults over 65 years old
in the household, SISBEN score and its square, whether the applicant is the head of the
household, and whether the applicant is the spouse/partner of the head of the household.
d Observations for evaluation sample: 78 months × individuals; observations for entire
cohort: 12 months × individuals.

Table 3—Labor Market Outcomes by Gender

Evaluation sampleb Entire cohortc

Women Men Women Men
Control Control Control Control
means Coeffic. means Coeffic. means Coeffic. means Coeffic.
Total formal 200,103 35,495 327,673 35,126 187,022 23,178 287,686 35,632
incomea (366,562) (12,421) (489,732) (18,597) (373,297) (4,805) (422,794) (8,814)
[0.065] [0.251] [0.000] [0.000]

Working in 0.29 0.047 0.43 0.036 0.28 0.03 0.42 0.05

formal sector (0.015) (0.019) (0.451) (0.006) (0.493) (0.011)
[0.052] [0.251] [0.000] [0.000]

Work in large 0.146 0.038 0.24 0.025 0.15 0.020 0.223 0.044
formal firm (0.013) (0.016) (0.357) (0.005) (0.416) (0.009)
[0.065] [0.251] [0.000] [0.000]

Observations 81,588 165,750 67,158 140,946 84,612 259,788 41,460 112,860

Notes: Standard errors are in parentheses, clustered at the applicant level. Romano-Wolf p-values (in square brack-
ets) for six hypotheses in each sample, respectively. Other notes as in Table 2.
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 139

Formal Earnings.—We find that formal earnings are about COL$35,000 higher
among the individuals who were randomly assigned to training, which corresponds
to a 13.6 percent increase and is significant at the 2.5 percent level. This shows a
remarkable persistence of the effects of the program. In terms of pesos, the effect
is similar for males and females, at about 35,000 (Table 3), although for women it
represents a higher percentage increase: female earnings in the control group are
COL$200,000, while those for males are COL$327,000. Thus, the respective per-
centage increases are 17.5 percent for females and 10.7 percent for males. Once we
control for multiple testing the female effect is only significant at the 6.5 percent
level, while the male effect is not significant. However, when we turn to the larger
sample from the entire cohort we get much more precise results: they confirm the
large and significant effect overall and, crucially for the value of the program, we
now can establish that there was a large and significant effect for both females and
males. While the point estimate for males is substantially larger than that of women,
the difference is not significant. This result is key because it puts the entire program
in a different perspective, showing that it also improved male outcomes.

Formal Sector Employment.—The increase in formal earnings is driven by an

increase in formal sector employment: the probability of working in the formal sec-
tor is increased by 4 percentage points (Table 2). In both datasets the effects are
highly significant. In Table 3, we find that for females the probability of being a
formal employee increases by 5 percentage points (Romano-Wolf ­­p-value of 0.052).
For males, the point estimate is similar but not significant in the evaluation data.
But again, when we turn to the larger data, we find a significant and larger effect on
formal employment for men (5 percent, ­­p-value 0). In both cases it represents an
impact of about 12 percent higher probability of being formal with respect to the
control applicants. This confirms that the program was beneficial for men as well
as for women, and importantly, the effects are persisting well after the program.
From the evaluation sample we find that participants in JeA worked on average
3.33 months longer in the formal sector (standard error of 1.19).

Probability of Working in a Large Firm.—Finally, as a proxy for quality of jobs,

we estimate the effect of the program on working for a large firm. About 15 percent
of control applicants do so. Overall, both datasets show an increase of about 3 per-
centage points, which is highly significant. As before the evaluation sample is not
large enough to provide precise enough estimates for men and women separately,
but when we turn to the larger dataset, we find a 2 percentage points increase for
women and a 4.4 percentage points increase for men. Both have ­­p-values of zero
and are significantly different from each other. Thus, the program increased partic-
ipation in the formal sector of both men and women, and seems to have improved
the quality of jobs (as measured by working in a large firm) particularly for men.

Contributions to Social Security.—In Table A2 of the online Appendix, we also

present impacts of the program on the amounts of mandatory contributions paid
to various programs based on the data for the whole cohort. These increased sig-
nificantly by 10–15
­­ percent for men and women, confirming directly the greater
140 American Economic Journal: applied economicsapril 2017

a­ ttachment to the formal sector. As before, the entire cohort helps us in establishing
that the effect for men was large and significant.

Displacement.—As discussed in the introduction, a concern with programs such

as JeA, particularly when associated with job placement, is that at least part of the
effect is due to displacement of workers who are not treated but are otherwise close
substitutes and could have obtained these jobs if the trainees had not been actively
placed.11 Moreover, in areas with a high level of treatment, the competition for jobs
can increase, decreasing everyone’s arrival rates. Based on Crépon et al. (2013), one
way to gauge the extent to which this is the case is to use the varying proportion
of treated (out of the unemployed population) for each city and occupation. In the
presence of negative externalities, an increased proportion of treated individuals
should reduce overall formal employment rates. Moreover, to the extent that treated
individuals are being placed at the expense of control individuals, the treatment
effect will increase with the proportion treated as long as the negative externality
on the controls was larger than that on the treated. We can repeat the exercise by
also considering the proportion treated out of the total number of unemployed and
employed in that occupation.
To construct proportion treated, we use the entire experimental cohort of the JeA
program and a Colombian household survey,12 which allows us to estimate how
many people are unemployed and looking for a job in a specific occupation and
city. We also know how many were trained by occupation (from the type of course
they participated in). Combining the two, we derive the proportion treated out of the
unemployed in each city and occupation for the baseline period. This varies substan-
tially. In Table 4, we show the proportions treated averaged across occupations in
each of the cities. Bogotá has the lowest proportion at 12 percent and Bucaramanga
the highest, where 61 percent of the unemployed were treated. Once we go to the
occupation level, the variation is much higher.
To test for negative externalities, we run two sets of regressions. In one, we
include the proportion of the unemployed that are treated as well as the interaction
of this proportion with the treatment dummy. The coefficient on the proportion has
a coefficient of 0.01 (standard error 0.007); in the presence of a negative externality
this coefficient would be negative. The coefficient on interaction of the treatment
dummy with the proportion is −0.015 (standard error 0.01).
In the second regression, we add site-by-course fixed effects (making the effects
of the proportions unidentifiable) and include only the interaction with the treatment
dummy. We obtain an estimate of the interaction term of −0.008 (standard error
0.014), which is not significant. Using the proportion treated out of all those in the
same occupation (whether employed or unemployed) the conclusions are the same.
In both cases the impact on formal sector participation remains about the same at
0.044 and 0.040, respectively, and significant at the 1 percent level.

See Blundell et al. (2004) and Crépon et al. (2013).

11
12
The Colombian household survey is used by the Administrative Department of National Statistics of
Colombia, DANE, to obtain Colombia’s labor market indicators.
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 141

Table 4—Proportion of the Unemployed Treated by City (average across occupations)

Barranquilla Bogotá Bucaramanga Cali Cartagena Manizales Medellín

0.275 0.118 0.607 0.242 0.131 0.383 0.166

Thus, the evidence based on these simple regressions suggests that displacement
was not a major issue in this intervention and that we can take the treatment effects
as implying a direct increase on economic activity without negative externalities on
other labor market participants. However, these results must be treated with some
care because the treatment intensities are potentially endogenous. Moreover, this
does not mean that such effects (or indeed other general equilibrium effects) would
not occur on an expanded and permanent version of this program.

Cost-Benefit Analysis.—The administrative data does not provide enough infor-

mation for a complete ­­cost-benefit analysis without any further assumptions because
we do not observe outcomes for those working in the informal sector. However, if we
make the conservative assumption that overall employment did not increase (i.e., the
increase in the proportion of those working in the formal sector results in an equal
decrease in those working in the informal) and that earnings in the informal sector
remain unaffected, we can obtain a lower bound to the benefits of the program under
the assumption of no displacement. The costs of the program, including loss of ten-
ure, were calculated in AKM to be US$812. Based on the entire sample, and on an
annual depreciation rate for the benefits of 6 percent and a discount rate of 6 percent
as well, the lifecycle benefits for men are US$1,046 and for women US$1,010,13
implying a gain of approximately US$200 for men and women, which is equivalent
to an internal rate of return of about 10 percent for both men and women. With no
depreciation, which seems reasonable given the evolution of the gains we observed,
the IRR is as high as 16 percent.

IV. Conclusions

The Jóvenes en Acción program was focused on training and work experience for
young people. Its original evaluation by AKM showed positive effects on women,
for whom earnings and formal employment increased; for men, formal employment
increased but not earnings. In this paper, we consider the longer
­­ run effects of the
program. Moreover, by using a dataset that includes ten times the number of indi-
viduals in the original evaluation sample, we are able to obtain much more precise
estimates.
The results are remarkable: the increase in formal employment is sustained for
both men and women with no indication of it fading away, and average formal
earnings increase for both men and women. Moreover, we also find that the prob-
ability of working in a large firm (a proxy for better quality jobs) increases, as do

13
We are using an exchange rate of 2,000 COP to the US dollar.
142 American Economic Journal: applied economicsapril 2017

c­ ontributions to pensions and other programs for both men and women. In testing
whether the impacts can be attributed to displacement of workers who may other-
wise have obtained jobs instead of our subjects, we find no such evidence. Finally,
the internal rates of return, which are 10 percent with the most pessimistic scenario,
far exceed interest rates. Unfortunately, the lack of data on the informal sector pre-
cludes a ­complete evaluation of the program. Indeed, the positive impacts in the for-
mal sector could in principle be counteracted from negative impacts in the informal
one; however, the short run results in AKM do not point to any such negative effects
in the short run.
By all accounts this has been a successful intervention and offers a rare example
of a vocational program with ­­long-term sustained effects, if not for overall employ-
ment, at least in improving the levels of formal employment relative to the informal
one—this is of key importance in developing countries. Evidence on the importance
of the complementarity between the classroom component and work experience
would further enhance our ability to design successful vocational education.

REFERENCES

Alzúa, María Laura, Guillermo Cruces, and Carolina López. 2015. “Youth Training Programs Beyond
Employment: Experimental Evidence from Argentina.” Centro de Estudios Distributivos, Laborales
y Sociales (CEDLAS) Working Paper 177.
Attanasio, Orazio, Arlen Guarín, Carlos Medina, and Costas Meghir. 2017. “Vocational Training for
Disadvantaged Youth in Colombia: A Long-Term Follow-Up: Dataset.” American Economic Jour-
nal: Applied Economics. [Link]
Attanasio, Orazio, Adriana Kugler, and Costas Meghir. 2011. “Subsidizing Vocational Training for
Disadvantaged Youth in Colombia: Evidence from a Randomized Trial.” American Economic Jour-
nal: Applied Economics 3 (3): 188–220.
Barrera-Osorio, Felipe, and Alejandra Corchuelo. 2003. “SENA’s returns: A re-evaluation.” Unpub-
lished.
Blundell, Richard, Monica Costa Dias, Costas Meghir, and John van Reenen. 2004. “Evaluating the
Employment Impact of a Mandatory Job Search Program.” Journal of the European Economic
Association 2 (4): 569–606.
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 Econom-
ics 29 (2): 267–300.
Cárdenas, Mauricio, and Carolina Mejía. 2007. “Informalidad en Colombia: Nueva evidencia.” La
Fundación para la Educación Superior y el Desarrollo (Fedesarrollo) Working Paper 35.
Crépon, Bruno, Esther Duflo, Marc Gurgand, Roland Rathelot, and Philippe Zamora. 2013. “Do
Labor Market Policies have Displacement Effects? Evidence from a Clustered Randomized Exper-
iment.” Quarterly Journal of Economics 128 (2): 531–80.
Departamento Administrativo para la Prosperidad Social. 2016. Manual Operativo: Programa
Jóvenes en Acción. Prosperidad Social. Bogotá, June.
Fondo de Inversíon para la Paz (FIP) and Departamento Nacional de Planeación (DNP). 2001. “Red
de Apoyo Social, Subprograma Jóvenes en Acción.” Unpublished.
Gaviria Uribe, Alejandro, and Jairo A. Núñez Mendez. 2003. “Evaluating the impact of SENA on
earnings and employment.” Departamento Nacional de Planeación (DNP) Archivos de Economía
Documento 220.
Guarín, Arlen, and Carlos Medina. 2015. “Jóvenes que ni estudian ni trabajan, y jóvenes que delin-
quen: El caso de Medellín.” Observar, January, 8–15.
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 and Development 3: Article 10.
Kugler, Adriana, Maurice Kugler, Juan Saavedra, and Luis Omar Herrera Prada. 2015. “Long-term
Direct and Spillover Effects of Job Training: Experimental Evidence from Colombia.” National
Bureau of Economic Research (NBER) Working Paper 21607.
Vol. 9 No. 2 ATTANASIO ET AL.: VOCATIONAL TRAINING IN COLoMBIA 143

López Castaño, Hugo. 2010. “El Mercado Laboral Colombiano: Tendencias de largo plazo y sugeren-
cias de política.” Banco de la República Colombia Borradores de Economía 606.
Medina, Carlos, and Jairo Núñez. 2005. “The Impact of Public and Private Job Training in Colombia.”
Inter-American Development Bank Research Network Working Paper R-484.
Medina, Carlos, Jairo Núñez, and Jorge Andrés Tamayo. 2013. “The Unemployment Subsidy Program
in Colombia: An Assessment.” Inter-American Development Bank Working Paper 369.
Meghir, Costas, Renata Narita, and Jean-Marc Robin. 2015. “Wages and Informality in Developing
Countries.” American Economic Review 105 (4): 1509–46.
Ministerio de la Protección Social. 2005. Informe de actividades 2004–2005 al Congreso de la
República. Bogotá: Grupo de Estudios y Planeación Sectorial.
Organisation for Economic Co-operation and Development (OECD). 2012. Education at a Glance
2012: Highlights. Paris: OECD Publishing.
Perry, Guillermo E., William F. Maloney, Omar S. Arias, Pablo Fajnzylber, Andrew D. Mason, and
Jaime Saavedra-Chanduvi. 2007. Informality: Exit and Exclusion. Washington, DC: World Bank.
Romano, Joseph P., Azeem M. Shaikh, and Michael Wolf. 2008. “Formalized Data Snooping Based on
Generalized Error Rates.” Econometric Theory 24 (2): 404–47.
Romano, Joseph P., and Michael Wolf. 2005. “Stepwise Multiple Testing as Formalized Data Snoop-
ing.” Econometrica 73 (4): 1237–82.
Saavedra, Juan E., and Carlos Medina. 2014. “Formación para el trabajo en Colombia.” In Equidad
y Movilidad Social: Diagnósticos y Propuestas para la Transformación de la Sociedad Colombi-
ana, edited by Armando Montenegro and Marcela Meléndez, 331–414. Bogotá: Universidad de los
Andes Press.
Santa María, Mauricio, Alexander Estacio, Natalia Millán, Mauricio Olivera, and Mónica Parra.
2009. Evaluación de impacto de tres programas y línea de base de un programa del Servicio Nacio-
nal de Aprendizaje—SENA. Bogotá: La Fundación para la Educación Superior y el Desarrollo.
Sarmiento, Alfredo, Jorge Iván Gonzalez, Lucía Mina, Darwin Marcelo, Sandra Álvarez, Carlos
Alonso, and Elsa Plazas. 2007. “Evaluación del impacto del SENA en el capital social.” http://
[Link]/transparencia/gestion-de-planeacion/Documents/Evaluaciones/evalua_capital_
[Link].
Steiner, Roberto, Norberto Rojas, and Natalia Millán. 2010. Evaluación del impacto del programa
jóvenes rurales emprendedores del Servicio Nacional de Aprendizaje—SENA. Bogotá: La Fun-
dación para la Educación Superior y el Desarrollo.
Székely Pardo, Miguel. 2012. “Jóvenes que ni Estudian ni Trabajan: Un Riesgo para la Cohesión Social
en América Latina.” In Violencia y Cohesión Social en América Latina, edited by Francisco Javier
Díaz and Patricio Meller, 163–208. Santiago, Chile: Cieplan.