A Score Test for Latent Class in Left-Censored Data Due to Detection Limit

• Activity Number: 77 - Hypothesis Testing: Bayesian, Nonparametric and Likelihood Methods
• Type: Contributed
• Date/Time: Sunday, July 29, 2018 : 4:00 PM to 5:50 PM
• Sponsor: International Chinese Statistical Association
• Author(s): Hua He* and Wan Tang
• Companies: and School of Public Health and Tropic Medicine, Tulane Univeristy
• Keywords: Left-censored; metabolites; mixture model; score test; Tobit model

Abstract

Left-censored data due to detection limits such as metabolites are common in medical research. Assuming that the data are from a single population a normal distribution, the Tobit model, or censored normal regression, is a standard method for analyzing such data. However, in practice it is often the case that there are more censored observations than what would be expected under such a Tobit model. In such cases, a mixture model consisting of a censored normal distribution and a point distribution with values below the detection limit would be appropriate. For analysis of such data, a fundamental question is to test the existence of such sub-population with a point distribution. In this talk, we will develop a score test for the latent subgroup. Simulation studies and real data examples are presented to illustrate the proposed score test.