Employment Effects of Short and Medium Term Further

Employment Effects of Short and Medium Term Further Training Programs in Germany in the Early 2000 s Martin Biewen, University of Mainz, IZA, DIW Bernd Fitzenberger, University of Frankfurt, ZEW, IZA, IFS Aderonke Osikominu, University of Frankfurt Marie Waller, University of Frankfurt, CDSEM, ZEW

Motivation n n Training programs still major part of active labor market programs in Germany (e. g. expenditures 2000: 6, 793 bill. EUR, 2004: 3, 616 bill. ) Traditionally, focus on long, expensive programs Recently, shift towards cheaper short-term training measures Research questions: n To what extend have programs positive effects? n To what extend can cheaper short-term programs substitute the traditional long-term programs?

Literature n n Older studies n Hübler (1998), Lechner (1999), Hujer/Wellner (2000), Fitzenberger/Prey (2000), u. a. n Survey data: SOEP, Arbeitsmarktmonitor Ost More recent studies n 1) Lechner et al. (2005 a, b), Fitzenberger/Speckesser (2005), Fitzenberger et al. (2006) n 2) Lechner/Wunsch (2006), Schneider et al (2006) n Administrative data from 1) 80 s/90 s 2) 2000 s

Contribution of our study n n n New, informative data make possible, for the first time, serious evaluations of recent programs Use of up-to-date econometric methods that address possibility of multiple treatments and dynamic selection into treatment New evidence on effectiveness and comparative effectiveness of short and medium-term programs

Program types n n n Short-Term Training (STT) n 2 – 12 weeks n E. g. computer course, application training Further Training (CFT, PFT) n Several months to one year n Classroom Training (CFT), Practical Training (PFT) n E. g. accounting training in practice firm Retraining (RT) n 2 to 3 years n Leads to formal professional degree

Data (1) n Integrated Employment Biographies (IEB 2. 05) n Administrative Data n 2, 2% random sample drawn from 4 sources n Employment History (Be. H), 01/90 -12/03 n Benefit Recipient Hist. (Le. H), 01/90 -06/04 n Supply of Applicants (Bew. A), 01/00 -07/04 n Program Participation (MTG), 01/00 -07/04 n 1, 4 million individuals, 17 million spells n Validation of data set was part of the project

Data (2) n Be. H Example employed unempl. benefit Le. H Bew. A MTG searching unempl. assistance subsistence allowance registered as unemployed STT PFT Time

Evaluation strategy (1) n Evaluation Problem: n Effect of program is difference of actual employment outcome and employment outcome in case of counterfactual non-participation n Problem: only one outcome observable n Possible solution: use outcomes of comparable control group of non-participants

Evaluation strategy (2) n n Who is a potential participant? n Inflow-sample in non-employment conditioning on previous employment n Advantages n Wide definition of unemployment n Avoid problem of endogenous unemployment Our inflow-sample n Inflow in non-employment 02/2000 - 01/2002 n At least 3 months of previous employment n 25 -53 years old at beginning of non-employment

Evaluation strategy (3) n Multiple Treatments (e. g. Lechner (2001)) n Different Treatments n Here: STT, CFT, PFT or „no treatment“ n Potential outcomes n Average Treatment Effect on the Treated

Evaluation strategy (4) n Dynamic selection into treatment n Program may start at different points of time during unemployment spell n Unemployed individuals who don‘t participate now may participate later n Static approach implicitly conditions on future outcomes (Fredriksson/Johanson (2003)) n Treatment effect may vary with previous unemployment duration (Sianesi (2003, 2004)) n → Distinguish different starting points

Evaluation strategy (5) n Aggregation of potential starting points STT Example: 4 -6 months unemployed CFT PFT STT CFT UN PFT UN 0 -3 months unemployed 4 -6 months unemployed 7 -12 months unemployed Time

Evaluation strategy (6) n Interpretation of treatment effect n Treatment effect reflects decision problem of the case worker: participation now vs. participation later (waiting), or participation in program vs. participation in program

Evaluation strategy (7) n Propensity-score matching n In an experimental sense, individuals are comparable if they had the same propensity to participate in the program n Among all -individuals, estimate propensity to participate in program vs. in program n Estimate the counterfactual employment outcome of participants in if they instead had participated in by a local linear kernel regression on the propensity score and the calendar month of the beginning of the unemployment spell

Evaluation strategy (8) n Estimated treatment effect Actual employment outcome of a particular participant Counterfactual employment outcome of the participant is given by weighted average of the employment outcomes of the control group

Evaluation strategy (9) n Cross-validated bandwidth choice (Bergemann et al. (2004)) Choose bandwidth so that the employment outcome of a particular member of the control group is predicted as good as possible by the employment outcomes of the other members of the control group. Here, the particular member of the control group stands for a particular member of the treatment group whose employment status is to be predicted as good as possible.

Evaluation strategy (10) n Determinants of the propensity score n Individual characteristics: age, qualifications, marital status, nationality, health … n Characteristics of the last job: occupation, industry, wage … n Labor market and transfer receipt history n Assessments of case worker: lack of motivation, lack of cooperation, penalties … n Regional information: regional unemployment rate, federal state …

Evaluation strategy (11) n Validity of Cond. Independence Assumption n Rich set of covariates, typically 20 to 35 statistically significant regressors in propensity score n Even information on typically unobserved factors n Further unobserved factors proxied by labor and transfer receipt history n Assignment to programs contains strong random element due to local availability of courses n „Pre-Program Test“/Balancing-Tests

Evaluation strategy (12) n Further details of estimation procedure n Smith/Todd (2005)-Balancing-Test n n n Extensive specification searches for each PS (program £ East/West £ men/women £ strata) Graphical check of common support assumption Fully bootstrapped standard errors

Results (1): West Germany Short Term Training (STT) 7% 5% M. 0 -3 months unempl. 4 -6 months unempl. 9% F. 7 -12 months unempl. 10 %

Results (2): West Germany Classroom Further Training (CFT) 8% M. Lock-in Effect 0 -3 months unempl. 4 -6 months unempl. 5% 16 % F. 7 -12 months unempl. 10 %

Results (3): West Germany Practical Further Training (PFT) 10 % F. M. 0 -12 months unempl.

Results (4): East Germany Short Term Training (STT) 7% M. 0 -3 months unempl. F. 4 -6 months unempl. 7 -12 months unempl.

Results (5): West Germany Classroom Further Training (CFT) 9% M. 0 -3 months unempl. F. 4 -6 months unempl. 7 -12 months unempl.

Results (6): West Germany Practical Further Training (PFT) 7% F. M. 0 -12 months unempl.

Conclusions n n n n West Germany: both STT and CFT/PFT have sizable positive employment effects (5 -10%) The employment effects of STT are in many cases comparable to those of the longer CFT/PFT Effects for women generally larger than for men Effects larger for the long-term unemployed PFT effective for West German women Almost no positive effects in East Germany To do: 1) cross-evaluate programs, 2) incorporate new data, 3) evaluate RT