Dear SPM experts,
I have a question about conducting a paired t test with GIFT component images and adding covariates.
In my study, each participant underwent 2 resting state scans (and these make up the two conditions of the study that I want to compare). Using GIFT, 20 components were identified and I want to now compare the activity of these components across the two conditions (with a number of subject-specific covariates added).
You can do a paired t test using the 'SPM stats' tool in GIFT, but this does not allow you to add any covariates. In addition, a paired t test conducted directly in SPM does not work with subject specific covariates (they do not account for any additional variance). Thus, I have read that the best approach here is a 'summary statistics approach' i.e., make a contrast image for each subject and then enter it into a one sample t test (with covariates).
I have done this for my first component but I am slightly concerned as I carried out a quick check. Completely discarding my covariates, I just compared the following three approaches: 1) a standard paired t test in SPM 2) a paired t test conducted in GIFT 'using the SPM stats' tool, and finally 3) the summary statistics approach outlined above. I noticed that all of these approaches give slightly different results. The overall regions activated look similar but there are some differences in the effect sizes/p-values.
Does this sound normal? I am particularly surprised that there is a difference between the results of 1) and 2) as I would have assumed these approaches were identical.
Thanks for any help you can give,
Gerard
|