Non-normal Type-I Censored Samples in Logistic Regression

• Activity Number: 619
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Evrim Oral*+
• Companies: Louisiana State University
• Keywords: Type I censoring ; Level of detection ; Modified maximum likelihood ; Logistic regression

Abstract

In the framework of logistic regression, Type I and Type II censoring occur in many practical situations. Type I censoring occurs when a sample of size N is drawn and the observations which have values below or above certain pre-determined limits are censored. Specifically, left censored data are characteristic of many HIV studies due to the inherent limit of detection (LOD) in the assays. Most of the research dealing with censored samples is based on the normality assumption; however, skewed distributions are more common in biological sciences. For instance, left censored HIV biomarker measurements tend to be highly positively skewed. In practice, researchers often log-transform the data and apply ad-hoc methods which lead them to biased estimates. In this study we consider non-normal Type I censored samples in the framework of logistic regression and we present modified maximum likelihood estimation of the parameters.