Abstract

One of the most pernicious myths perpetrated by standard methodologic training is that of unbiased or objective inference. Unbiased inference exists only within highly simplified models that do not begin to reflect the complexities of genuine public-health and risk-assessment settings. In those settings, neither unbiased estimators nor unbiased judgements can be identified.

All inferences depend on biased judgements, albeit some judgements are hidden within social conventions (eg, 0.05 significance levels, 95% confidence levels). Those conventions are often transmuted into claims of objectivity, thus obscuring the fact that all judgements (and hence all inferences) are biased—and not only by recognised vested interests. These conventions and claims incorporate biases prevalent in the environment in which we are educated, work and communicate. They limit our ability to see, question, and deviate from these conventions, which become priors embedding our thoughts. Because conventions provide a sense of orientation and security, they may feel compelling and be taken for granted even when they have little or no basis in fact or utility. While this problem has long been recognised, it is rarely accounted for in research and review. Transparency is the first step in such accounting: Methodology must lay bare conventions, preferences, and special interests that affect judgements so that the sources of inference can be critically evaluated.