Voodoo Correlations Revisited: Biased Analysis in Neuroimaging and Behavioral Data

Nearly a decade ago, Vul and colleagues (2009) published a paper criticizing then-current methods of region of interest (ROI) analysis. Their criticism focused on what is called biased or circular analysis, in which data is extracted from voxels that are already defined as statistically significant, leading to inflated effect sizes. In this talk I will outline the history of the debate, explain why it is still relevant in both neuroimaging and behavioral studies, and discuss ways you can prevent inadvertently doing biased analyses.