Abstract
Brain imaging studies have traditionally struggled to break into 3-digit sample sizes: e.g., a recent Functional Magnetic Resonance Imaging (fMRI) meta-analysis of emotion found a median sample size of n=13. However, we now have a growing collection studies with sample sizes with 4-, 5- and even 6-digits. Many of these "population neuroimaging" studies are epidemiological in nature, trying to characterise typical variation in the population to help predict health outcomes across the life span. I will discuss some of the challenges these studies present, in terms of massive computational burden but also in ways that they expose shortcomings of existing mass univariate techniques. I will also discuss how these datasets present intriguing methodological problems heretofore absent from neuroimaging statistics. For example, the "null hypothesis fallacy" is how H0 is never strictly true, and yet with 100,000 subjects you'll eventually find some effect even if it is meaningless. This motivates work spatial confidence sets on meaningful effect sizes (instead of thresholding test statistic images), providing intuitive measures of spatial uncertainty.
