Semiparametric Modification of KHB Method

Professors K. B. Karlson, A. Holm and R. Breen in 2012 introduced the econometric method (KHB) which allows to solve the confounding problem in binary choice models. Based on simulations the authors showed that their approach outperforms other proposed methods. However, the assumptions used in the simulations were rather restrictive: random errors in the models have only logistic and normal distributions, which is not always true in practice and may lead to the low accuracy of method estimates in violation of the relevant assumptions. In this paper, we propose a modification of the KHB method using the semi-parametric Gallant-Nychka approach which allows to relax the assumptions about the random error in the binary choice model. It was shown on simulations that with complex distributions of random error, semiparametric modification provides more accurate results than parametric approach. The problem of confounding the effect of family income on worker status with and without modification of the KHB method was studied on RLMS HSE data from 2011 year. The results are consistent with the simulations and, according to the AIC, the approach with modification outperforms the classical version of the method.