||Regression Methods for Data with Incomplete Covariates.
Whittemore, A. S. ;
Grosser., S. ;
||Stanford Univ., CA. Dept. of Family, Community and Preventive Medicine.;Health Effects Research Lab., Research Triangle Park, NC.;National Institutes of Health, Bethesda, MD.;National Science Foundation, Washington, DC.;Alfred P. Sloan Foundation, New York.
||EPA-R-813495-01 ;NIH-CA-23214; EPA/600/J-86/530;
Regression analysis ;
Statistical analysis ;
Chronic disease ;
Maximum likelihood estimates
||Most EPA libraries have a fiche copy filed under the call number shown. Check with individual libraries about paper copy.
Modern statistical methods in chronic disease epidemiology allow simultaneous regression of disease status on several covariates. These methods permit examination of the effects of one covariate while controlling for those of others that may be causally related to the disease. However, they do not accommodate data in which one or more covariates are incomplete, e.g. missing or measured with error. The paper uses assumptions about the probability laws governing covariate incompleteness to obtain estimates of regression coefficients relating disease to the unobserved complete covariates. The estimates are obtained by maximizing the likelihood of the observed incomplete data via the EM algorithm.