Quantitative Bias Analysis

Matthew P. Fox, Timothy L. Lash

Random error is nearly always quantified in epidemiologic research results, while systematic error is rarely quantified. This disparate treatment exists despite the fact that systematic error often dominates the uncertainty about an estimate, and the fact that methods for quantitative bias analysis have been described for decades. Shifting point estimates to account for bias, as well as widening study intervals to account for the uncertainty due to systematic error, provides a more complete assessment of total study error and reduces the chances of inferential errors. Below we provide links to some of the key papers on the subject as well as our textbook, which describes how to apply the methods to some of the most common problems experienced in epidemiologic research: selection bias, unmeasured or uncontrolled confounding, and misclassification.

Leave a Reply