Get Training or Wait? Long-Run Employment Effects of Training Programs for the

1. Get Training or Wait? Long-Run Employment Effects of Training Programs for the Unemployed in West Germany Bernd Fitzenberger, Aderonke Osikominu, and Robert Völter Comments October 2006

2. Summary • Applies matching methods to examine program impact over time • Considers 3 programs with different durations • Practice Firm (PF): around 6 months • Provision of Specific Professional Skills and Techniques (SPST): around 6 months • Retraining (RT): up to 2 years (average 1 year) • Evaluation is for quarters up to 30 quarters (nearly 8 years) after program entry • Dependent variable is employment

3. Summary • For each kind of training, it is divided into three strata: • training that begins after 1-2 quarters of unemployment • training that begins after 3-4 quarters of unemployment • training that begins after 5-8 quarters of unemployment • Comparison of “training” vs. “wait” is between those who receive a given training type within a particular stratum and those not receiving training during that stratum • Those not receiving training could receive some kind of training at a later point

4. Overall Comments • Matching methods are well executed • Methods require numerous decisions & although one might quibble about some, each decision is thoughtfully discussed and justified • Extremely valuable to obtain a set of estimates indicating impact on earnings over time • Long follow-up period • Since I have few complaints with the analyses presented, I will provide some suggestions for additional analyses

5. Suggestions • Current paper looks at “cumulative” effect, looking at effects for • the first 8 quarters • the first 16 quarters • the first 24 quarters • I would like comparisons of effects at various points in time, i.e., • quarters 9-16 • quarters 17-24

6. Suggestions • Benefits may well extend beyond 8 years; this is an important aspect of evaluating the program • Estimating average effects for the last 4-12 quarters might be a reasonable estimate of long-term impact on flow of benefits • Assumed constant for the remainder of the working life, 10 years?

7. Effects for strata 1 and 2 appear fairly stable after 20 quarters 1-2 quarters 3-4 quarters 5-8 quarters Strata: Unemployment Prior to Program Entry Figure 2: Treatment Effect for 1986-1987 Cohort Provision of Specific Professional Skills and Techniques (PSPT)

8. Suggestions • Of interest: • Do programs that require more extended training (longer “lock-in” periods) ultimately produce higher benefit flows? • If not, there is little justification for them

9. Suggestions • Might it be useful to calculate average treatment effects combining strata? • Are differences across strata important, or even statistically significant? 1-2 quarters 3-4 quarters 5-8 quarters Strata: Unemployment Prior to Program Entry Figure 2: Treatment Effect for 1986-1987 Cohort Provision of Specific Professional Skills and Techniques (PSPT)

10. Suggestions • Paper comparison question: • Compares those who get a given kind of training with those who don’t get training in at that point • Comparison group may get training (of some kind) at a future point or may get job • Most clearly policy-relevant question, since “waiting” is the alternative • Alternative question: • Do individuals who wait longer get smaller benefits from the program? • This comparison would control for characteristics of those obtaining a particular kind of training at different point in time, which none of the current analyses do

11. Suggestions • Specification test • Paper presents data on employment 10 quarters prior to unemployment, comparing treated with comparison group • Control and treated groups are never different

12. Suggestions • Specification test • Paper presents data on employment 10 quarters prior to unemployment comparing treated with comparison group • Control and treated groups are never different • A weak “test” of model assumptions • Variables identifying employment during this period are included among the matching measures • This is a test of whether the matching is working: Doesn’t test basic conditional independence assumption • Specification test would have to test behavior not explicitly included in matching variables.

13. Suggestions • As in other European studies, no attempt is made to determine effects on earnings for those employed • Justification: High minimum earnings in Europe; interest in individuals at bottom; concern with unemployment • Is it possible that some kinds of training (e.g., retraining) have benefits beyond employment? • I think it would be useful to do some analyses predicting expected earnings, especially in comparisons on programs.