Rounding up the usual suspects: confirmation bias in epidemiological research

Investigators performing epidemiological research frequently form hypotheses based on data availability. One might ask how it could be otherwise. After all, what is the point of forming hypotheses if they can’t be tested? But when questions are identified to suit available data rather than data being identified to suit important questions, commonalities in measured and unmeasured variables extend across multiple studies and lead to a confirmation bias. Expected relationships are confirmed, and unexpected relationships remain undiscovered, even when their unveiling would have important informational value. We argue that this confirmation bias results from a structural cause, in particular misalignment of epidemiological research priorities with the social utility of research.