The Effect of Distributional Misspecification on Estimating Percentiles with Longitudinal Data

• Activity Number: 659
• Type: Contributed
• Date/Time: Thursday, August 4, 2011 : 10:30 AM to 12:20 PM
• Sponsor: Biometrics Section
• Author(s): Kyeongmi Cheon*+ and Zhiwei Zhang and Albert Paul
• Companies: Eunice Kennedy Shriver National Institute of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development and Eunice Kennedy Shriver National Institute of Child Health and Human Development
• Keywords: percentile estimates ; misspecification ; mixture model ; mixture model

Abstract

Reference quartiles such as growth charts are widely used in medicine in order to detect unusual and potentially pathological measurements. Nonparametric estimates of extreme percentiles could be inaccurate since they tend to be associated with relatively large variability. To alleviate this problem, investigators have estimated the expected percentiles with model-based approaches. For example, in longitudinal studies, random effect models, which assume a Gaussian random effects and error distribution, have been used for percentile estimation. Through simulation studies, we find that percentile estimation is not robust to the misspecification of the Gaussian random effects or error structures. We propose an alternative mixed modeling approach with more flexible random effects and error distributions that are more robust to model misspecification for estimating extreme percentiles. Our methods will be illustrated with fetal growth data.