The long run effects of job training

I am always on the lookout for impact evaluations that give us the long term effects of interventions.   I recently came across a paper by Pablo Ibarraran, Jochen Kluve, Laura Ripani and David Rosas Shady looking at the effects of a youth training program in the Dominican Republic.    While we have some evidence on the long term effects of these kind of programs from developed countries, this is quite possibly the first in a developing context.   

So what’s the program?   Ibarraran and co. are looking at the effects of the Juventud y Empleo program.   This program targets out of school youth, age 16-29 who are living in poor areas.   These folks get 150 hours of vocational training (in low-skill jobs) plus 75 hours of soft-skills training (self esteem and the like).   And then there is a three month internship with a private firm.   Throughout, the participants get a $3 per day stipend.  

This was originally one of the first experimental evaluations of a labor training program in Latin America.   Card and coauthors, in a 2011 paper, find modest effects, with potentially some increases in earnings conditional on working.   But this work was hampered by compliance issues (it’s an interesting paper to read to see how they dealt with them).   Ibarraran and coauthors took another stab at evaluation, with a later cohort, in a 2014 paper.    Again, there was no overall impact on employment, but they did find an increase in formal employment for men and higher earnings conditional on being employed, as well as reduced pregnancy and increased non-cognitive skills.   

Now let’s take a look at the long term.    Ibarraran and co. have a program randomly assigned at the individual level.   The setup is interesting, from both a compliance and analysis point of view.   In the initial lottery, training centers identify groups of 35 eligible youths.  20 are offered training, and 15 are assigned to the control.   But of these 15, 5 are randomly selected (by a central team) to be replacements in case any of the original treatment 20 don’t show up or drop out right away.    This gives Ibarraran and co. not only an (original) treatment group, but some replacements, which they can use for ATT and LATE estimates.   Since those results aren’t markedly different, I’ll skip them in what follows.  

The baseline survey for this evaluation was in 2008, the first follow-up (for the other Ibarraran and co. paper) was in late 2010-2011, and the most recent survey was in late 2014 – allowing the authors to estimate effects after 6 years.   Attrition isn’t horrible: both of the follow up surveys find about 80% of the sample, equally balanced across treatment and control.   Ibarraran and co. look at the baseline characteristics and it looks like the attriters aren’t significantly different (there is a significant difference in attending school, which might be interesting, but it’s really quite small).   Looking at the characteristics of folks at the endline one interesting difference is that males assigned to treatment are significantly more likely to be married relative to males in the control group.   Ibarraran and co. speculate that this could be due to the employment differences.  

Speaking of which, what do they find?    First, after 6 years, still nothing on employment and earnings for the average participant.    But there is something going on with the quality of employment, and principally for men.   Men are 8 percentage points (or 26 percent) more likely to have a job with medical insurance.  Ibarraran and co. highlight this as an indicator of increased formality.   Turning to results in Santo Domingo (the country’s “most important urban labor market”), things get interesting.   The average worker in Santo Domingo is more likely to have a job with health insurance, and now this is significant (at 10 percent) for both women and men.   Earnings for treated women in Santo Domingo are significantly higher, about 25% more than the control group.   Returning to the overall sample, and the fact that the follow up surveys had recall data (with one wee gap), Ibarraran and co. can trace out the treatment effect over time.   Their figure for male jobs with health care looks like this:

It’s interesting to see the early dip as those assigned to treatment participate in the program, but then the steady climb over time away from the control group.   As Ibarraran and co. note, it’s interesting to see that there are a number of time periods where the result is not significant.   So without this recall data, the timing of the follow up survey would really matter in picking up this effect.

So, all in all, the long run is showing something of an uptick from the shorter term.    Overall, there isn’t a net employment effect here, it’s more of a move to more formal employment.   Here’s looking forward to other longer term results – for labor market programs and others.