Time-Varying Transformation Models for Estimation of Conditional Distributions and Quantiles with Longitudinal Data

• Activity Number: 619
• Type: Topic Contributed
• Date/Time: Thursday, August 5, 2010 : 8:30 AM to 10:20 AM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Xin Tian*+
• Companies: National Institutes of Health
• Keywords: Conditional Distribution ; Linear Transformation Models ; Longitudinal Study ; Quantile ; Smoothing estimate ; Time-Varying Coefficient

Abstract

We propose a class of time-varying transformation models for analyzing the conditional distribution with longitudinal data, and present a rank-based smoothing method for estimating covariate effects, conditional distributions and quantiles. Using a structural nonparametric approach, these models are flexible and useful for longitudinal analyses with ordinal or non-Gaussian time-dependent outcomes where the traditional conditional-mean based longitudinal models are inappropriate. An example is the evaluation of hypertension in children, which is defined by the blood pressure percentiles conditioning on gender, age and height. We demonstrate through simulations and a large epidemiological study of childhood cardiovascular risks that our models give meaningful interpretations and the estimation procedures can be easily implemented.