Acknowledgements:We thank Pierre Azoulay, Scott Stern, Nico Lacetera, Dietmar Harhoff, and participants in numerous seminars for comments and suggestions. Lisa Bassett, Anne-Marie Crain, Michaël Bikard, Devin Fensterheim, Robyn Fialkow, Jacob Magid, and Lexie Somers provided exceptional research assistance. All errors are our own. Financial support for this research was provided by the National Science Foundation, under grant #0738394.Electronic copy available at: http://ssrn.com/abstract=2014481Governing knowledge in the scientific community:Exploring the role of retractions in biomedicine
ABSTRACTAlthough the validity of knowledge is critical to scientific progress, substantial concerns exist regarding the governance of knowledge production. While as or more important to the knowledge economy as defects are in the manufacturing economy, mechanisms to identify and signal "defective" or false knowledge are poorly understood. In this paper, we investigate one such institution -the system of scientific retractions. By analyzing the universe of peer-reviewed scientific articles retracted from the biomedical literature between 1972-2006 and comparing with a matched control sample, we identify the correlates, timing, and causal impact of scientific retractions, thus providing insight into the workings of a distributed, peer-based system for the governance of validity in scientific knowledge. Our findings suggest that attention is a key predictor of retraction -retracted articles arise most frequently among highly-cited articles. The retraction system is expeditious in uncovering knowledge that is ever determined to be false (the mean time to retraction is less than two years) and democratic (retraction is not systematically affected by author prominence). Lastly, retraction causes an immediate, severe, and long-lived decline in future citations. Conditional on the obvious limitation that we cannot measure the absolute amount of false science in circulation, these results support the view that distributed governance systems can be designed to relatively swiftly to uncover false knowledge and to mitigate the costs that false knowledge for future generations of producers.