Personality Traits, Preferences and Economic Success
Final Report Abstract
This project was funded in the first funding phase. No new project proposal was submitted for the second funding phase as the PI took over the position of the Vice-Rector for Research and Young Researchers of the University of Konstanz. The initial research goal of this primarily methodological project was to develop appropriate estimators for nonlinear models which can cope with high-dimensional settings and which provide a consistent model selection strategy in high-dimensional settings. As an empirical application the role of noncognitive skills represented by numerous different and competing skill measures were to be studied. The initial idea was to develop Bayesian factor models to cope with this problem of high dimensional settings. Over the course of time modern regularization techniques appeared to be the more attractive way to tackle this problem, so that the focus of the project switched from Bayesian techniques to Machine Learning approaches. Marečková & Pohlmeier (2020) study the predictive quality of noncognitive skill measures by means of machine learning techniques. Unlike previous empirical studies on the impact of noncognitive skills on individual unemployment focusing on the in-sample explanatory power, they study the power of skills measured at childhood to predict individual unemployment over a long-run horizon of 20 years and more. Based on the group lasso a novel strategy of constructing target-oriented indices for personality traits is proposed where survey items with higher predictive relevance concerning unemployment receive larger weights. Their novel machine learning approach is capable to cope not only with the challenge of selecting the most relevant factors from data with a large number of skill measures but also leads to a sparse set of skill measures which is economically and psychologically interpretable. Based on data from the British Cohort Study (BCS) the predictive power of different noncognitive skill measures is studied. In particular, the authors propose a machine learning strategy on how to optimize the assignment mechanisms for manpower training programs and psychological intervention schemes for youths and young adults. Heiler & Marečková (2020) propose a flexible regularization approach that reduces point estimation risk for the general problem of estimating the mean of observations characterized by distinct group. The method can also be used for smoothing or weak aggregation of both information within and across different time-series. It nests methods established in the literature on ridge regression, discretized support smoothing kernels and model averaging methods. They derive risk-optimal penalty parameters and propose a plug-in approach for estimation. Monte Carlo simulations reveal robust improvements over standard methods in finite samples. The method is applied to a re-evaluation of the effects of minimum wages on employment and to smooth information from multiple timeseries in a panel on parental behavior in a field experiment on the deterrence hypothesis.
Publications
- (2019), How well can Noncognitive Skills Predict Unemployment? A Machine Learning Approach
Marečková, J. and W. Pohlmeier
(See online at https://doi.org/10.13140/rg.2.2.33573.96488) - (2020), Shrinkage for Categorial Regressors, Journal of Econometrics
Heiler, P. and J. Marečková
(See online at https://doi.org/10.1016/j.jeconom.2020.07.051)
