Menu Content/Inhalt
Seminars Print
previous year previous month next month next year
See by year See by month See by week See Today Search Jump to month
Herman van Dijk, Rotterdam University Print
Thursday, 03 November 2011, 12:15 - 13:15

Herman van Dijk, Rotterdam University

A Predictive Likelihood Approach to Possible Endogeneity Using IV Models with Non-Normal Distributions Applied to US Income-Education Data

Paper 1 - Paper 2

Abstract : The purpose of this paper is to measure the effect of education on earned income using a predictive likelihood approach when the model (endogeneity or exogeneity) is in doubt and the data are non-normal. The technical issue that arises is the non-standard/non-elliptical shapes of likelihoods  and posteriors due to weak instruments and weak priors. A new flexible Monte Carlo algorithm which is based on the EM method using mixtures of student t distributions (labeled partial MitISEM) is introduced to approximate the posterior densities and predictive probabilities that are evaluated for assessing the degrees of endogeneity and instrument strength in IV models.
Results from simulated data  with varying degrees of endogeneity and instrument strength indicate that a combination of models is useful when the model is in doubt.  The Angrist and Krueger data for individual US states are used as empirical data.
These raise the issue of taking into account the non-normality of observed income distributions. Results for the US states for a mixtures of normal distributions are shown to give substantive different results than for the case of standard normal income distributions.

Location: R42.2.113
Contact: Claude Adan, This e-mail address is being protected from spam bots, you need JavaScript enabled to view it