22 August 2009
Spurious Correlations and "Compelling Inconclusiveness"
Spurious correlations are very much on my mind. As I noted in an earlier blog, I read Risk by Dan Gardner over the summer and enjoyed it very much (about the way people incorrectly assess risk, guided by emotion, fanned by the media and organisations with their own agendas). Recently my colleagues at the Institute of Development Studies passed me an article by Steven Greenberg in the British Medical Journal which shows how "unfounded authority" can be generated by citation distortions, using evidence from the medicial journals. Then in Foreign Policy magazine this month there is a note by David Lehrer on how the Journal of Spurious Correlations is a refuge for "compellingly inconclusive" results in the public policy research sphere. I also read somewhere about Rejecta Mathematica, a new online journal that has similar goals in its field. This is a trend I welcome--too often the applied policy research community polarises conclusions and overstates claims to get into the journals or policymakers hype a particular paper because it confirms their own biases. For example in the very same issue of Foreign Policy a "Prime Numbers" feature claimed that low birthrates are needed to create wealth and are not driven by wealth. This strikes me as a very strong statement--I would expect causality to run in both directions ("compellingly inconclusive"). But no articles were cited to backup this claim. The final thread on this entry relates to a paper I am writing on "The M&E of M&E". One section of the paper reviews the impact that the participation of intended beneficiaries has on project outcomes. A paper I co-authored in 2001 on this issue claims “De facto participation lowers the ratio of project to local wages; increases the labor intensity of projects that provide community buildings, roads, or sewers; and lowers the cost of creating employment and of transferring funds to poor individuals.” In writing this M&E of M&E paper I am reviewing more recent reviews of whether participation affects project outcomes. My 2001 paper is cited in two papers, authored by highly respected economists from the US and Europe, as showing that participation has "no effect" on project outcomes. I suspect they are referring to the result we found that de jure participation has no effect--but the much more important result is that de facto participation does. Thanks to the Greenberg paper I know know what this is called: "citation diversion--citing content but claiming it has a different meaning". You could also call it sloppy research. Just don't call it spurious correlation.
Posted by Lawrence Haddad at 19:35