General Departmental Seminar Series
A unified approach to model misspecification due to informative missingness, measurement error and confounding
Mari Palta, Department of Biostatistics & Medical Informatics and Preventive Medicine, University of Wisconsin
Wednesday, Oct 18, 2000, 4:00 pm
1221 CSSC, 1210 W. Dayton St.
Several types of model misspecification can be reformulated as problems of omitted covariates. We consider models for longitudinal data where the relationship between the outcome and the covariates is known as long as all covariates are correctly measured, and included . However, when there are unknown confounders or measurement error in covariates, this amounts to covariates being omitted from the correct model. So called random coefficient models for describing informative missingness can also be viewed in this context. Longitudinal measurements present special opportunities for detecting and dealing with omitted covariates. Our previous work (Palta and Yao, 1991; Chao, Palta and Young, 1997; and Shen, Shao, Palta and Park, 2000) introduced models for the joint distribution of measured and unmeasured covariates, and allowed the derivation of marginal models with correct mean structure We discuss the consequences of fitting models which do not take into account the possibility of omitted covariates and give examples of constraints, which allow estimation of the original parameters of interest. Finally, we address approaches to detecting the misspecification. The above results are illustrated in a data set on sleepiness in a cohort of Wisconsin state employees, who were surveyed at two separate time points (Young, Palta., et al., 1993).
Back to General Departmental Seminar Series