MISSPECIFICATION IN LONGITUDINAL MODELS

描述：（改编自研究者摘要）本次更新

DESCRIPTION: (Adapted from Investigator's Abstract) This renewal application is requesting support to extend the findings of the first funding cycle where the investigators described the impact of model misspecification in linear regression to the second cycle where they will focus on analyses with ordinal and categorical outcomes. Specifically, the investigators propose to assess the impact of omitted (confounding) covariates, the impact of measurement error and selection bias on parameters estimated by logistic regression, GEE, and other categorical approaches. With respect to the assessment of confounding, the investigators propose to generalize the confounding structure of their earlier models and to extend the linear results to GEE estimation. Topics to be addressed in measurement error include extending the latent variables approaches used in the social sciences to biostatistics, to assess the impact of misclassification in binary and ordinal outcome measures, and to examine the impact of distributional assumptions concerning confounders. Topics associated with selection bias to be addressed include assessing the impact of informative censoring on selected models and adjustments of modeling to sampling approaches (such as oversampling). Finally, the procedures for detecting model lack of fit will be developed. The application implicitly suggests that the techniques developed will be applied to existing data sets.