 |
 |
Dear SPM Community I am having trouble. The correlation of a behavioral measure, social responsiveness scores (SRS), with activation throughout the whole brain leads to some strikingly odd results. Essentially the whole brain (k > 400,000) is significantly correlated with SRS scores. I have been trying to troubleshoot how this could occur for the last week and cannot resolve the problem. I get reasonable results for other behavioral measures but I cannot trust my analysis at this point because of this result with SRS. Have you ever seen anything like this? My Data: fMRI from a serial response time task where you press buttons according to the stimuli. The stimuli are either 'Random' or they have a 'Sequence' pattern. I have two groups: cases and controls. I have several behavioral measures on these subjects including SRS. SRS scores were significantly correlated across the whole brain during both random (simple task activation for pushing buttons and viewing stimuli) and sequence. I set up my design matrix as: case, control, case covariate, control covariate. [0 0 0 1]. I am using SPM8. Troubleshooting so far (all on the contrast of 'Random' regressors): - Could it be my new code? Or the contrast images produced from the 1st level analysis be awry? No. I ran my new script without the behavioral covariates looking solely at activation in controls during and found the same activation I had in previous analysis: case, control. [0 1]. They were strong visuomotor regions like one would expect. - Could it be that I should not have both cases and controls in the same model? No. I changed model to: case, case covariate. [0 1]. Similar results to case, control, case covariate, control covariate. [0 0 0 1]. - Could there be something wrong with software? No. I ran the scripts on another machine with the same results - What happens if I shuffle my SRS scores so they no longer match their associated control? Smaller whole brain encompassing significant cluster results. - Is my raw data off? No. SRS scores for both groups have no outliers and behave as expected. Correlating the mean activation across the whole brain from the contrast image of each subject with their SRS scores results in a pretty strong correlation of r = .65. I am not sure what else to explore. Do you have any suggestions? Thank you!