Dear Loreen,
> A recent discussion about how best to set up individual differences analyses
Was this on the mailing list? Maybe you can provide a link to get an impression of the arguments.
Concerning 1.): Yes, why not. If you want to test whether the differential activation is related to some parametric varibale then you have to set up a contrast like A - B at some point.
Concerning 2.): If the variance is large you might well get no sig. findings for A, even if all subjects respond with activation changes going into the same direction. Activation changes don't have to be small in that case. In practice beta values usually fall into a certain limited range, thus also limiting the variance, so it might be more likely that some subjects show a change and others no change, resulting in an overall null finding. If some subjects respond with increased activations and others with decreases, you will neither get sig. findings for A.
For illustrative purpose it can be useful to extract the beta/con estimates, percent signal change, first eigenvariate for clusters showing a sig. relationship, which should facilitate the interpretation (e.g. positive activations but to a different extent, positive and negative activations, ...). As you focus on sig. voxels/clusters you should not provide any additional statistics though (e.g. detecting some sig. clusters correlating with variable X on whole-brain level, extracting some estimates, reporting that the extracted estimates correlate highly and significantly with variable X, this is double-dipping).
Best,
Helmut
|