Downloads & Free Reading Options - Results

Connectome-based predictive modeling (CPM), one among many methods of brain-wide association studies (BWAS), has gained substantial popularity in the neuroimaging community to predict behavioral phenotypes from individual’s brain connectome. Using CPM, many studies have reported significant success in the prediction of general intelligence, personality traits and symptom severity in diseases like social anxiety or depression. Recently Marek et al. (2022) examined the effect sizes in multivariate brain-wide associations and observed that variances of possible prediction accuracy increased with decreasing sample size. Therefore, for a study employing a small sample size, it is much more likely to observe spuriously high predictive accuracy by chance. We would like to test this theoretical hypothesis by looking at the actual CPM literature and see whether the found effect sizes are comparable to the bootstrapped distribution that Marek et al reported. If the reported effect sizes in the CPM-literature deviates significantly from the bootstrapped distribution of effect sizes, we can further evaluate if there exist a potential inflation of prediction accuracy by published CPM studies.