CFE 2019: Start Registration
View Submission - CMStatistics
B1786
Title: Improved estimators for zero-inflated count data models in the presence of multicollinearity Authors:  Talha Omer - Jönköping University, Sweden (Sweden) [presenting]
Par Sjolander - Jonkoping University, Jonkoping International Business School (Sweden)
Kristofer Mansson - JU JIBS (Sweden)
BM Golam Kibria - Department of Mathematics and Statistics Florida International University USA (United States)
Abstract: Zero inflated count-data models are used when the data is in the form of non-negative integers. A surplus of zeros induces overdispersion in the dependent variable of the count regression model. Under these circumstances, zero-inflated models can be used effectively. There is a clear empirical relevance for these types of models, for instance when modelling the demand for health services when most patients have zero visits, or when counting the number of insurance claims within population, etc. However, multicollinearity is a frequently observed, but usually disregarded, empirical problem for these types of data sets. Multicollinearity increases the variance of the estimated coefficients and make the estimates very sensitive. Therefore, we address this relevant problem by considering some improved estimators such as Ridge and Liu estimators for non-negative count models. The performance of these estimators has been evaluated by Monte Carlo simulations, and based on the MSE and the MAE performance criteria, the simulations illustrate that our improved estimators better than the usual maximum likelihood estimator and some other Liu estimators in the presence of multicollinearity. At the end, an empirical application is conducted for the improved Liu and ridge estimators and its results support the simulation study.