Item-Wise Interindividual Brain-Behavior Correlation in Task Neuroimaging Analysis

Brain-behavior correlations are commonly used to explore the associations between the brain and human behavior in cognitive neuroscience studies. There are many critics of the correlation approach, however. Most problems associated with correlation approaches originate in the weak statistical power of traditional correlation procedures (i.e., the mean-wise interindividual brain-behavior correlation). This paper proposes a new correlation procedure, the item-wise interindividual brain-behavior correlation, which enhances statistical power via testing the significance of small correlation coefficients from trials against zero rather than simply pursuing the highest correlation coefficient. The item-wise and mean-wise correlation were compared in simulations and an fMRI experiment on mathematical problem-solving. Simulations show that the item-wise correlation relative to the mean-wise correlation results in higher t-values when signal-to-noise ratio is equal to or larger than 6%. Item-wise correlation displayed more voxels with significant brain-behavior correlation than did mean-wise correlation. Analyses with item-wise (rather than mean-wise) correlation showed significant brain-behavior correlation at the threshold of p < 0.05 corrected. Cross validation showed that odd- and even-ordered trials have greater stability in terms of the item-wise correlation (r = 0.918) than the mean-wise correlation (r = 0.686). The simulations and example analyses altogether demonstrate the effectiveness of the proposed correlation procedure for task neuroimaging studies.