Dissecting the OA Citation Advantage: Bayesian approaches will be superior to randomised control trials
The idea that open access articles are more highly cited is attractive, particularly for open access advocates. While there have been many retrospective cohort analyses that suggest citations advantages of various sizes, randomized control trials have shown no effect. The problem with randomised control trials is that they require a tight definition of the posited effect and sufficient statistical power, which can be challenging to determine in advance. RCTs reported have for instance focussed on making a a subset of articles in a closed journal free to read. While this addresses the literal question of an access related citation advantage, it controlls away various potential effects of interest such as closed vs open journals, differences in indexing or promotion etc. There is also the risk of disciplinary bias based on the journals studied.
The question we should be asking is "how can we best ensure that research reaches the right place for its use". This is better addressed using very large and disciplinarily diverse datasets. A time domain signal processing approach could provide a statistical model of pathways from article publication to downstream citation. This can provide a statistical estimate of whether pathways for closed vs open access are different.