L2E Estimation for Finite Mixture of Regression Models with Applications

• Activity Number: 274
• Type: Invited
• Date/Time: Tuesday, August 2, 2011 : 8:30 AM to 10:20 AM
• Sponsor: International Indian Statistical Association
• Author(s): T. N. Sriram*+
• Companies: University of Georgia
• Keywords: Asymptotic normality ; Consistency ; Extreme values ; L2E functional ; Robustness

Abstract

For count response, we propose a robust estimation method for finite mixture of regression models based on minimum integrated L2 distance between parametric conditional and true conditional mass functions. The estimator, called the L2E, is shown to be consistent and asymptotically normal. We demonstrate the efficiency and robustness of the L2E estimator through a variety of simulation studies. Finally, the L2E in conjunction with a two-component Poisson mixture regression model is shown to provide a competitive fit to a hospital length of stay data, which contains extreme values. Our L2E approach is also shown to readily extend to the continuous response case, thereby expanding its scope of use and application.