-----------------------------------------------------------

Statistics Seminar Series

Monday, January 29, 2007 @ 11:30AM
Cullimore Hall, Room 611
New Jersey Institute of Technology

Censored Median Regression Models


 

Sundar Subramanian, Ph.D

Department of Mathematics and Statistics
University of Maine

Median regression models provide a robust alternative to regression based on the mean. For right censored data, two methods that have been proposed for parameter estimation are the “inverse censoring weighted” (ICW) approach and the “missing information principle”. In this talk we describe an ICW estimator of the regression parameter vector when the data are both left and right censored, but when the left censoring variable is always observed. The solution of the ICW estimating equation specifies the estimator. The estimator is consistent and asymptotically normal, which allows the construction of confidence intervals for any desired regression sub-vector through the minimum dispersion statistic of Basawa and Koul. The efficacy of the proposed method is checked through a small simulation study. We also briefly discuss the main issue in censored median regression parameter estimation, namely the curse of dimensionality.