This event has already occurred.
Salmon Thoughts, "Voodoo Correlations," and Severe Tests: How the Error-Statistical Philosophy Can Help the Science of fMRI
Ph.D. candidate at Virginia Tech
2 Riverside Circle, Roanoke, VA 24016
Research abstract: Recently the science of fMRI has been subjected to a number of criticisms. Some of these criticisms take on a particularly methodological approach questioning the validity and reliability of certain data collection and analysis practices in fMRI research. The most noticeable and controversial one of these criticisms was a recent manuscript entitled “Voodoo Correlations in Social Neuroscience” by Vul, Harris, Winkielman, and Pashler. Vul et al. develop their criticism by citing the results of studies that investigated the reliability of psychological scales and fMRI measures. On the basis of these results, they calculate the upper bound for any meaningful correlations that could be obtained between results of psychological measures and fMRI data. Yet, correlation coefficients that exceed this upper bound are reported in many published articles. In order to make sense of this puzzling fact, Vul et al. appraise specific procedures of data analysis that were employed in these studies and conclude that these high correlations result from what they call non-independence errors in statistical analyses. They conclude that the validity of any inferences that come from studies that employ these procedures is highly questionable. Several statisticians and fMRI practitioners have published commentaries to Vul et al.’s conclusion, which vary considerably. The commentators disagree about the terms and arguments of the debate, offer different diagnoses of the problem, and suggest different solutions. This debate demonstrates the significance of the methodological difficulties that arise in fMRI research. Instead of taking sides, I propose that Deborah Mayo’s error-statistical philosophy of science can help with the issues of the debate on Vul et al.’s puzzle. I formulate the problem(s) in terms of the notions of error probabilities and severe tests, and on the basis of this formulation, I suggest solutions to the Vul et al.’s puzzle and other similar difficulties that arise in different contexts of fMRI research.
About the Speaker:
Emrah Aktunc is a Ph.D. candidate at Virginia Tech with work in philosophy of science. His specific interests are in general philosophy of science, scientific confirmation and evidence, philosophy of statistics, and methodological and epistemological problems in the crossroads of behavioral science, biopsychology, and philosophy of mind. His dissertation research is on issues and problems of evidence in fMRI experiments, especially as they are used in cognitive neuroscience.