Dear SPMers,
I have a question about the most straightforward way to conduct covariate analyses in SPM12, and would greatly appreciate any help.
Here’s what I have done:
18 participants (one group) were scanned pre- and post intervention. I have conducted a paired samples t-test to investigate activity changes between the two timepoints. Specifically, per participant and for both sessions, I generate a contrast image comparing two conditions of a task (let’s say condition A -1 and condition B 1) on the first level. The two con images per participant are then entered into the paired samples t-test on the 2nd level. However, I also want to investigate if a behavioural measure is statistically associated with the change in activity from pre to post-intervention. Normally, I would subtract the pre from the post image in ImCalc and run a one-sample t-test with the behavioural measure as covariate. However, I have also been wondering about alternatives that will leave me with just one con map after the 1st level analysis. So, a t-test using the difference score, instead of a paired samples t-test using both time points if you will.
I am thinking about modelling both sessions per participant on the 1st level in one go, thereby generating one con image instead of two. However, this means I would have to implicitly specify the baseline (A) condition. In order to get the same map as the subtraction map from ImCalc, I would then have to contrast the pre session (-1) with the post session (1). The resulting image I could use to conduct a one-sample t-test (contrast: ‘1’) to assess activity differences and, secondly, use in the covariate analysis.
My first question is this: is this approach desirable, and would there be any obvious drawbacks associated with this approach? I know that there are differences between implicit and explicit modelling, namely related to onset of the BOLD response, but I am wondering what the magnitude of the difference would be and if there would be discrepancies between the two approaches (i.e. the paired samples and one-sample t-test). Secondly, I was wondering which of both options you would consider?
Many thanks for you answers,
Ing. Stijn Michielse MScRes
School for Mental Health and Neuroscience – Division Mental Health
Department of Psychiatry & Neuropsychology
[log in to unmask]
www.maastrichtuniversity.nl
|