**FIELD OF SPECIALIZATION:** Causal Inference, Program Evaluation, Nonparametric Techniques, Applied Econometrics, Labor Market

**RESEARCH PROJECTS**

*Bia M., Flores C., Flores-Lagunes A., Mattei A.*

**"Finite Sample Properties of Estimators of Dose-Response Functions based on the Generalized Propensity Score"**

In this paper we propose three semiparametric estimators of the dose-response functions based on kernel and spline techniques. In many observational studies treatment may not be binary or categorical. In such cases, one may be interested in estimating the dose-response function in a setting with a continuous treatment. This approach strongly relies on the uncounfoundedness assumption, which requires the potential outcomes are independent of the treatment conditional on a set of covariates. In this context the generalized propensity score can be used to estimate dose-response functions (DRF) and marginal treatment effect functions. We evaluate the performance of the proposed estimators using Monte Carlo simulation methods. We also apply our approach to the problem of evaluating job training program for disadvantaged youth in the United States (Job Corps program). In this regard, we provide new evidence on the intervention effectiveness by uncovering heterogeneities in the effects of Job Corps training along the different lengths of exposure.

## Bia M. and Van Kerm P.

**"Propensity score reweighted regression in analysis of wage differentials"**

The ubiquitous approach in analysis of wage differentials is the Blinder-Oaxaca (BO) decomposition (Blinder A., 1973; Oaxaca R., 1973). However, misspecification of the regression models may lead to misleading inference about the various components of the BO decomposition. Nonparametric techniques can be used to avoid this problem (Mora R., 2008). Nonetheless, as emphasized in Fortin et al.'s (2010) survey, these approaches are often much more computationally demanding, face the curse of dimensionality when dealing with a large number of covariates, and/or do not allow straightforward singling out the impact of individual covariates on wage differentials. This paper considers a middle way that maintains ease of implementation of the regression-based approach and the possibility to make detailed decompositions, but that is more robust in the presence of misspecification than the classic approach. It involves estimating the coefficients from the reference regression (e.g. on male or natives) by weighted least squares (rather than OLS) where the weights are function of the relative density of the covariates in the reference (e.g. male or natives) and target samples (e.g. women or immigrants). This procedure, suggested in Fortin et al. (2010), has been demonstrated to lead to improved predictive performance when prediction is made out-of-sample and in the presence of model misspecification (Shimodaira H., 2000; Sugiyama M. and Mϋller K., 2005). Specifically, our objective in this paper is to empirically assess the gains of using such a procedure in a case study to the gender pay gap in Luxembourg. We compare estimates against a simple OLS approach and a fully nonparametric local linear model. We also experiment with alternative ways to specify and estimate the weighting function. While the approach is not as robust to misspecification as a nonparametric model can be, we find that it can offer significant improvement over the standard approach.

## Bellani, L. and Bia, M.

**"Measuring intergenerational transmission of poverty”**

This paper examines the causal channels through which growing up poor affects the individual’s economic outcomes as an adult. We contribute to the growing literature on intergenerational transmission, refining the measurement of the causal effects of poverty in childhood, applying different modern statistical techniques. We use a propensity score matching method to select a control group of non-treated individuals. The matched samples of poor and non-poor children are then used to assess impacts on adulthood outcomes, primarily income and poverty. The analysis is based on a wide-ranging cross-country comparison using EU-SILC data. For the specific purpose of this project we use the module on intergenerational transmission of 2005 and 2011, where retrospective questions about parental characteristics (such as education, age, occupation) were asked. We find that being poor in childhood significantly decreases the level of income in adulthood, increasing the average probability of being poor by 4%.

## Bia, M., Mercatanti A. and Li F.

**"Evaluation of Training Programs by exploiting secondary outcomes in Principal Stratification frameworks: the case of Luxembourg"**

The Principal Stratifications (PS) framework is widely adopted in the evaluation of public policies, in that it allows to account for the mediation effect of an intermediate variable that lies in the causal pathway between the intervention and the outcome. To the purpose of evaluating the effect of a Training Program (TP) (intervention variable) on future wages (outcome), the use of a PS framework naturally arises, because wages are observed only for those individuals who are employed after the TP. The peculiarity that wages have to be observed only for employed people implies an adequate ``stratification'' of individuals into homogeneous subsets with respect to the potential employment status (intermediate variable), under alternative values of the TP.

In this paper we investigate the causal effects of a given Training Program on wages under the PS approach in order to address this requirement, using data from the global social security database on labour force in Luxembourg (IGSS) and the administrative data collected by the Employment Agency (ADEM). We also exploit a secondary outcome (hours worked) to the purpose of sharpening inference when the primary outcome is censored by ``death'', since wages are unobserved for people who remain unemployed even after completion of the TP (Mattei et al., 2013; Mercatanti et al., 2014). Finally, a sensitivity analysis is conducted to evaluate how different departures from unconfoundedness assumptions affect the estimated causal effects of interest in our study.

In this paper we investigate the causal effects of a given Training Program on wages under the PS approach in order to address this requirement, using data from the global social security database on labour force in Luxembourg (IGSS) and the administrative data collected by the Employment Agency (ADEM). We also exploit a secondary outcome (hours worked) to the purpose of sharpening inference when the primary outcome is censored by ``death'', since wages are unobserved for people who remain unemployed even after completion of the TP (Mattei et al., 2013; Mercatanti et al., 2014). Finally, a sensitivity analysis is conducted to evaluate how different departures from unconfoundedness assumptions affect the estimated causal effects of interest in our study.