hi steve.
well, I was thinking of performing several GLM with different covariates, as in this preliminary part, we actually don't know how they will behave. For what i know, i cannot insert as many regressor as i want in a GLM. is it correct ? is there a "rule of thumb" dependent from the number available subjects??
concerning the need for start correcting for multiple comparisons across the several models. I wanted to start with several one-covariate glm, and when individuated those more related to vbm results, i would have done a single multiple covariate glm.
>>> I wrote:
otherwise, could I create a further mat/con with just group composition, without any covariate. and use this one to select the best smoothing. Is this decision so sensitive to possible covariates values differences?, or it should be more presumably sensitive to data quality, subject number and group composition?
@@@@@ you answered:
Both really - but in general we would not recommend an unfocussed fishing expedition, and would recommend choosing one smoothing and leaving it fixed.
what do you exactly mean with unfocused fish expedition??, of course i was thinking of defining one smoothing and then using for all the GLMs. (that's why i was thinking about a preliminary rendomize with no covariates)
when
thanks in advance
regards
Alberto
|