A statistical cure for the voodoo

A recent issue of Perspectives in Psychological Science demonstrates the kind of full-throated give and take that makes for better science. An article by Ed Vul et al, questions the results of functional magnetic resonance imaging (fMRI) studies that relate imaged changes to personality traits, emotional states, and social interactions. The heart of the critique is how the correlations are calculated and how their statistical significance is reported. Their analysis is backed up with a very substantial survey of the investigators of the studies in question. In the same issue, the supporters and detractors of the critique are given their say. Most impressive is the statistical perspective of Lindquist and Gelman. It is even-handed and informative and should be required reading for any student about to engage in research of the increasingly familiar "high dimensionality" data sets (e.g. in genomics and imaging) that allow a multiplicity of questions to be asked or hypotheses to be tested.

Update: 7/7/2009: Nice piece on this topic from NPR.

No comments: