Heterogeneous Employment Effects of Job Search Programmes: A Machine Learning Approach

Abstract

Abstract:We systematically investigate the effect heterogeneity of job search programmes for unemployed workers. To investigate possibly heterogeneous employment effects, we combine nonexperimental causal empirical models with Lasso-type estimators. The empirical analyses are based on rich administrative data from Swiss social security records. We find considerable heterogeneities only during the first six months after the start of training. Consistent with previous results in the literature, unemployed persons with fewer employment opportunities profit more from participating in these programmes. Finally, we show the potential of easy-to-implement programme participation rules for improving average employment effects of these active labour market programmes.

Priority to unemployed immigrants? A causal machine learning evaluation of training in Belgium

“…This fact also explains the result inKnaus et al (2017) who find larger effects of job search programs for foreigners, but only during the lock-in phase.…”

“…In both the study of Knaus, Lechner and Strittmatter (2017) and Bertrand, Crépon, Marguerie and Premand (2017) effect heterogeneity is essentially found during the lock-in phase and not so much post treatment. In our evaluation we confirm that heterogeneity is more important during this initial phase, but also find evidence of heterogeneity in the post-treatment effect.…”

“…As opposed to a standard Monte Carlo analysis, such an approach informs the data generating process (DGP) as much as possible by real data and reduces the synthetic components in the DGP to a minimum. The real data were taken from the Swiss social security records that were used to evaluate a job search programme for the unemployed (Knaus, Lechner and Strittmatter 2017). Knaus, Lechner and Strittmatter (2018) conclude that the Causal Forest based ML methods, particularly the Generalized Forest by Athey, Tibshirani, and Wager (2018), belong to the best performing estimators if explicitly adjustedin another stepto take account of confounding.…”

“…It appears also to be the first paper to perform such analysis in a multiple treatment context. Knaus, Lechner and Strittmatter (2017) use Lasso based methods to evaluate the effect heterogeneity of a (single) job search programme in Switzerland using administrative data from 2003. They find substantial effect heterogeneity, but only during the first 6 months after the start of programme participation.…”

Three Essays on Health Economics and Policy Evaluation

“…However, ad-hoc searches for the responsive subgroups may lead to false discoveries or may mistake noise for a true treatment effect (Davis and Heller, 2017). Knaus et al (2017) points out that for large-scale investigations of effect heterogeneity, standard p−values of standard (single) hypothesis tests are no longer valid because of the multiple hypothesis testing problems (Lan et al, 2016;List et al, 2019) and leads to so-called "ex-post selection" problem which is widely recognized in the program evaluation literature. For example, for fifty single hypotheses tests, the probability that at least one test falsely rejects the null hypotheses at the 5% significance level (assuming independent test statistics as an extreme case) is 1 − 0.95 50 = 0.92 or 92%.…”