Premium
Marginal and dynamic regression models for longitudinal data
Author(s) -
Schmid Christopher H.
Publication year - 2001
Publication title -
statistics in medicine
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.996
H-Index - 183
eISSN - 1097-0258
pISSN - 0277-6715
DOI - 10.1002/sim.950
Subject(s) - statistics , marginal model , regression , regression analysis , econometrics , random effects model , longitudinal data , covariance , residual , mathematics , cross sectional regression , computer science , medicine , data mining , polynomial regression , meta analysis , algorithm
Regression models for longitudinal data often employ random effects and serial correlation to account for residual variation between and within subjects. Most of these models are marginal models, separating the mean and covariance parameters. This paper discusses the use of dynamic models in which a lagged response serves as a predictor and compares these to marginal models. Regression parameters have a different interpretation in dynamic models as they describe changes in response levels, rather than the levels themselves. Lagged predictors are also useful with longitudinal data, explicitly quantifying the effect of previous levels of risk factors. These models are explored through analysis of data from the Childhood Respiratory Study, modelling lung function (FEV 1 ) levels as a function of age, height, sex and smoking status in children measured over a five‐year period. Copyright © 2001 John Wiley & Sons, Ltd.