09/28/2017 Guess the publication (1st ed.)
Solution: In their 2009 paper, originally called “Voodoo correlations in social neuroscience” and later renamed “Puzzlingly high correlations in fMRI studies of emotion, personality, and social cognition”, Vul and colleagues critically question the presence of impossibly high correlations in social neuroscience fMRI studies. While followed by many critical discussions about the correctness of assumptions and statistical arguments in their own paper (see for example 1, 2), one important contribution of this article to the field is bringing a focus on the problem that non-independence errors can lead to significant correlations out of noise. The most prominent form of a non-independence error is for example selection-bias. This happens when findings do result from correlations between voxels of certain brain areas and behaviour, when the behavioral variable was influencing the null hypothesis in the first places. The manuscript was considered problematic due to (1) the use of the original title and general language in the manuscript (e.g. “voodoo correlations” immediately implies “fraudulence”) and (2) the inclusion of likewise wrong statistical arguments. Especially the strong negative connotations attributed to the whole field of social Neuroscience were reason to spark a lot of outcry post publication.
Take away message: When investigating a statistical relationship between a behavioural measure and voxels in the brain, you cannot apply a secondary non-independent test to that same set of voxels using the variable that you have already shown to distinguish the data. The resulting conclusion is biased.
Original manuscript: http://journals.sagepub.com/doi/abs/10.1111/j.1745-6924.2009.01125.x
Some critical reading related to this topic: