An introduction to instrumental variables analysis: part 1.
Bennett DA.
There are several examples in the medical literature where the associations of treatment effects predicted by observational studies have been refuted by evidence from subsequent large-scale randomised trials. This is because of the fact that non-experimental studies are subject to confounding - and confounding cannot be entirely eliminated even if all known confounders have been measured in the study as there may be unknown confounders. The aim of this 2-part methodological primer is to introduce an emerging methodology for estimating treatment effects using observational data in the absence of good randomised evidence known as the method of instrumental variables.