Refine
Document Type
- Doctoral Thesis (1)
- Working Paper (1)
Language
- English (2)
Is part of the Bibliography
- yes (2)
Keywords
- Start-Up Subsidies (2)
- Active Labor Market Programs (1)
- Causal Inference (1)
- Instrumental Variables (1)
- Matching (1)
- Policy Reform (1)
Institute
- Wirtschaftswissenschaften (2) (remove)
The German start-up subsidy (SUS) program for the unemployed has recently undergone a major make-over, altering its institutional setup, adding an additional layer of selection and leading to ambiguous predictions of the program’s effectiveness. Using propensity score matching (PSM) as our main empirical approach, we provide estimates of long-term effects of the post-reform subsidy on individual employment prospects and labor market earnings up to 40 months after entering the program. Our results suggest large and persistent long-term effects of the subsidy on employment probabilities and net earned income. These effects are larger than what was estimated for the pre-reform program. Extensive sensitivity analyses within the standard PSM framework reveal that the results are robust to different choices regarding the implementation of the weighting procedure and also with respect to deviations from the conditional independence assumption. As a further assessment of the results’ sensitivity, we go beyond the standard selection-on-observables approach and employ an instrumental variable setup using regional variation in the likelihood of receiving treatment. Here, we exploit the fact that the reform increased the discretionary power of local employment agencies in allocating active labor market policy funds, allowing us to obtain a measure of local preferences for SUS as the program of choice. The results based on this approach give rise to similar estimates. Thus, our results indicating that SUS are still an effective active labor market program after the reform do not appear to be driven by “hidden bias”.
This thesis offers new insights on the effects of Start-Up Subsidies (SUS) for unemployed individuals as a special kind of active labor market program (ALMP) that aims to re-integrate individuals into the labor market via the route of self-employment. Moreover, this thesis contributes to the literature on methods for causal inference when the treatment variable is continuous rather than binary. For example, this is the case when individuals differ in their degree of exposure to a common treatment.
The analysis of the effects of SUS focuses on the main current German program called “Gründungszuschuss” (New Start-Up Subsidy, NSUS) after its reform in 2011. Average Effects on participants' labor market outcomes - as measured by employment and earnings - as well as subjective well-being are estimated mainly based on propensity score matching (PSM) techniques. PSM aims to achieve balance in terms of observed characteristics by matching participants with at least one comparable non-participant in terms of their probability to receive the treatment. This estimation strategy is valid as long as all relevant characteristics that explain selection patterns into treatment are observed and included in the estimation of the propensity score. To make our analysis as credible as possible, we control for a large vector of characteristics as observed through the combination of rich administrative data from the Federal Employment Agency as well as through survey data.
Chapters two to four of this thesis puts special emphasis on aspects regarding (the evaluation of) SUS programs that have received no or only limited attention thus far. The first aspect relates to the interplay of institutional details of the program and its effectiveness. So far, relatively little is known about the importance of SUS program features such as the duration of support. Second, there is no experimental benchmark evaluation of SUS available and thus, the reliability of non-experimental estimation techniques such as PSM is of crucial importance as estimates are biased when relevant confounders are omitted from the analysis. Third, there may be potentially detrimental effects of transitioning into (relatively risky) self-employment on subjective well-being among subsidized founders out of unemployment. These were to remain undetected if the analysis would focus exclusively on labor market outcomes of participants. The results indicate positive long-term effects of SUS participation on employment and earnings among participants. These effects are substantially larger than what estimated before the reform, indicating room for improvement in program design via changes in institutional details. Moreover, non-experimental estimates of treatment effects are remarkably robust to hidden confounding. Regarding subjective well-being, this thesis finds a positive long-run impact on job satisfaction and a detrimental effect on satisfaction with social security. The latter appears to be driven by adverse effects on social insurance contributions.
In chapter five, a novel automated covariate balancing technique for the estimation of causal effects in the context of continuous treatments is derived and assessed regarding its performance compared to other (automated) balancing techniques. Although binary research designs that only differentiate between participants and non-participants of some treatment remain the most-common case in empirical practice, many applications can be adapted to include continuous treatments as well. Often, this will allow for more meaningful estimates of causal effects in order to further improve the design of programs. In the context of SUS, one may further investigate the effects of the size of monetary support or its duration on participants' labor market outcomes. Both Monte-Carlo investigations and analysis of two well-known datasets suggests superior performance of the proposed Entropy Balancing for continuous treatments (EBCT) compared to other existing estimation strategies.