Bayesian analysis of the logistic kink regression model using metropolis-hastings sampling

Abstract

Threshold effect manifests itself in many situations where the relationship between independent variables and dependent variable changes abruptly signifying the shift into another state or regime. In this paper, we propose a nonlinear logistic kink regression model to deal with this complicated and nonlinear effect of input factors on binary choice dependent variable. The Bayesian approach is suggested for estimating the unknown parameters in the models. The simulation study is conducted to demonstrate the performance and accuracy of our estimation in the proposed model. Also, we compare the performance of Bayesian and the Maximum Likelihood estimators. This simulation study demonstrates that the Bayesian method works viably better when sample size is less than 500. The application of our methods with a birthweight data and risk factors associated with low infant birth weight reveals interesting insights.