Modeling Repeated Measures with Monotonic Ordinal Responses and Misclassification, with Applications to Studying Maturation

• Published in: Journal of the American Statistical Association Vol. 92; no. 440; pp. 1304 - 1211
• Main Authors: Albert, Paul S., Hunsberger, Sally A., Biro, Frank M.
• Format: Journal Article
• Language: English
• Published: Alexandria, VA Taylor & Francis Group 01.12.1997; American Statistical Association; Taylor & Francis Ltd
• Subjects: Adolescents; Algorithms; Applications; Applications and Case Studies; Biology, psychology, social sciences; Black people; Children; Diagnostic error; EM algorithm; Estimation bias; Exact sciences and technology; Insurance, economics, finance; Longitudinal data; Markov chains; Markov mixture models; Markov models; Mathematical models; Mathematics; Medical sciences; Missing data; Modeling; Modelling; Multivariate analysis; P values; Parametric models; Probabilities; Probability and statistics; Reliability, life testing, quality control; Sciences and techniques of general use; Statistical analysis; Statistics; Studies; Tanneries; Transition probabilities; Markov chain; Mixed distribution; Missing data; Data analysis; Error diagnostic; Risk factor; Repeated measurement; Maximum likelihood; Markov model; Covariate; EM algorithm; Medical application
• DOI: 10.1080/01621459.1997.10473651
• ISSN: 0162-1459

Many longitudinal studies of children are concerned with the modeling of monotonic responses such as growth and sexual maturation. The National Heart, Lung, and Blood Institute Growth and Health Study (NGHS) is a longitudinal study designed to examine the effect of growth and maturation on the development of obesity and related cardiovascular risk factors among black and white adolescent girls. Sexual maturation is measured with an ordinal outcome and is known to be measured with sizable diagnostic error. Of interest is examining the effects of race and age on the sexual maturation process. Here we propose a class of models for analyzing repeated monotonic ordinal responses with diagnostic misclassification in which we separately model the underlying monotonic response and misclassification processes. We develop an EM algorithm for maximum likelihood estimation that incorporates covariates and randomly missing data. We use the method to analyze the NGHS sexual maturation data.