With rapidly changing technologies human capital development is viewed as a key element to ensure employability of individuals in the long run. Training programs are an important tool of active labor market policy in many advanced countries (see e.g. OECD, 2007, table J). Training schemes typically comprise a wide variety of programs, from job application training to basic education and advanced vocational training taking place on the job or in classrooms. In addition, there exist important heterogeneities within a given type of training.
An important dimension of heterogeneity that has received little attention in the literature is how the impact of a given type of training program varies with the length of enrolment. One reason for this is that realized training duration is an endogenous variable that depends on the success of job search during training. On the one hand, lucky participants who receive a suitable job offer during training may drop out early, while the unlucky ones continue until the scheduled program end or even prolong participation for lack of job opportunities. This argument suggests that dropouts are a positively selected subset of participants. On the other hand, individuals may also drop out because they are unable to follow the program (e.g. because of lack of endurance). This would suggest a negative selection of dropouts compared to completers. In any case, program drop out for reasons that are related to employment outcomes after program start raise additional endogeneity issues that are difficult to incorporate in static evaluation approaches that are commonly used in the literature.
Not only the decision to continue or to drop out is dynamic but also the assignment to training often depends on the success of job search. In countries with comprehensive systems of active labor market policies like Germany or Sweden (cf. Sianesi, 2004, on Sweden), participation in active labor market programs may take place at any point in time during unemployment. Eventually, every unemployed who does not manage to find a job on his own becomes eligible for program participation.
The main contribution of this paper is to estimate the impact of training incidence and duration on employment transitions while taking into account the dynamic, endogenous nature of program participation and duration. We focus on a large scale training program in Germany that lasts eight months on average and for which enrolment lengths vary between a couple of weeks and more than one year. We specify a joint model for the transition rates into and out of employment and training using a very flexible bivariate random effects probit model. Our specification allows in a flexible way for state dependence and duration dependence in the transition rates as well as in the treatment effects and it includes interaction effects of these model components with observed covariates. Our rich administrative data allow us to integrate such flexibility into the model while performing separate estimations by gender and region (East and West Germany). Estimating a discrete time model for labor market transitions, we account for the full observed observation vector for each individual over time, irrespective of the number of spells experienced by an individual.
We use Bayesian Markov Chain Monte Carlo (MCMC) techniques that allow a numerically very robust estimation of our flexible model specification. Another advantage of the MCMC technique is that it provides predictions of the individual specific effects. This allows us to assess explicitly the selectivity of the treated and the nontreated individuals. We develop a simulation approach that uses the estimated individual specific effects to calculate the posterior distributions of different treatment effects of interest, such as the average treatment effect on the treated for the employment probability. Furthermore, we use our estimation results to simulate the effects of alternative policy scenarios. In particular, we examine how the impact of training changes when assigning different planned enrolment lengths.
Contents
1 Introduction
2 Institutional Background and Data
- 2.1 Training in Germany
2.2 Constructing a Panel Data Set
2.3 Descriptive Analysis
3 Evaluation Framework
- 3.1 Econometric Model
3.2 MCMC Estimation of a Random Effects Probit Model
3.3 Estimation of Treatment Effects
4 Estimation Results
- 4.1 Model Fit and Selection on Unobservables
4.2 Classical Treatment Effect on Employment and on Transition Rates
4.3 Training versus Waiting
4.4 Treatment Effects for Different Planned Training Durations
5 Conclusion
References
Appendix
- A Detailed Information on the Data
B Algorithm for MCMC Estimation
C Detailed Estimation Results
Download
The Heterogeneous Effects of Training Incidence and Duration on Labor Market Transitions
