Robust-to-endogenous-selection estimators for two-part models, hurdle models, and zero-inflated models

Start date

Thursday, 18 Oct 2018, 16:00

End date

Thursday, 18 Oct 2018, 17:00

Tinbergen Building
Campus Woudestein
Workshop: Machine Learning

David Drukker (STATA Corp)

This paper extends the results in Drukker (2017) to show that two-part models, hurdle models, and zero-inflated models are all robust to the endogeneity of the process that maps outcomes to on-boundary or off boundary values. 

It shows that a two-part-model approach can consistently estimate the mean outcome conditional on covariates without explicitly modeling the the endogenous process that maps the     outcomes to on-boundary or off-boundary values. It also derives new robust estimators for some models in the literature.


  • David M. Drukker is the Executive Director of Econometrics at Stata and has a Ph.D. in Economics from the University of Texas at Austin. His passion for programming took him to Stata in 1999. He has developed many Stata commands for estimating treatment effects and for analyzing panel data, time-series data, cross-sectional data, and spatial data. He played a key role in the initial development of Stata MP, helped integrate Mata into Stata, and has helped develop some of Stata's numerical techniques. He is also an active researcher, publishing papers in The American Economic Review, the Journal of Regional Science, Econometric Reviews, Economics Letters, and the Stata Journal, among other places.  He has been principal investigator on two large research grants. His current research interests are causal inference, spatial econometrics, and the robustness properties of two-part models.

More information

Coordinators: Andreas Alfons, and Wendun Wang,

Contact: Anneke Kop,