Varieties of Bias

In reporting the results of a comparative study, investigators often mention the possibility of bias, even if statistical adjustments have been made. A few specific types of bias may be highlighted and discussed briefly. Most commonly, a rationale is offered to justify a low level of concern about such residual bias. Less frequently, some actual evidence is provided to shed light on whether certain types of bias are plausible. Even less frequently, a sophisticated mathematical analysis may be performed to account for the uncertainty about residual bias. In the end, the reader may feel vaguely dissatisfied. It may be unclear whether all potential sources of bias have been considered. Are there other varieties of bias that could have threatened the validity of the study’s findings? Equally important, the nature and implications of possible biases are unclear.

Discussions of bias have been hampered by the lack of an explicit model for causation. Without such a model, it may be difficult to specify the circumstances in which bias seriously threatens the study’s conclusions. In this chapter, we describe the conventional ways of thinking systematically about varieties of bias. The fundamental problems faced in the biomedical and social sciences are quite similar, but the language used to describe these biases tends to differ. The difference is partly a function of the kinds of research that characterize these fields.

Epidemiologists and clinical trialists ...

Get Bias and Causation: Models and Judgment for Valid Comparisons now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.