| Last Updated:
Creating DOI. Please wait...
In observational studies of exposure effects confounding adjustment is challenging. It is generally advised not to control for variables that are measured after exposure has started. The reason for this is that such variables may actually be mediators of the causal relation between exposure and outcome, and conditioning on such variables may introduce bias. In our manuscript, however, we argue that conditioning on such variables may actually do more good than harm, in case they are proxies for unmeasured confounding variables. We derive expressions for the bias under different data analytical models, to show when adjustment for a mediator variable might actually be preferred.
CC-By Attribution 4.0 International