Bias in methods for deriving standardized morbidity ratio and attributable fraction estimates

It is shown that some of the proposed methods for deriving standardized morbidity ratios and attributable fraction estimates will, in general, produce biased estimators, although the low variance of certain estimators sometimes compensates for their bias.

Abstract

This paper examines several methods for deriving standardized morbidity ratios (SMR) and attributable fraction (attributable risk percentage) estimates. We show that some of the proposed methods will, in general, produce biased estimators, although the low variance of certain estimators sometimes compensates for their bias. A few methods are based on statistical fallacies and should be avoided, especially the method of deriving the expected number of cases from a regression equation that does not include the exposure under study.