Dear SPMers,
I'd like to get confirmation on my understanding of a few things and then ask a couple of questions:
I have z-transformed functional connectivity maps to a seed region of interest by individuals and wish to see what clusters are significant to a variable A after adjusting for covariates B, C, and D in one model (Model 1), and then covariates B, C, D, and E in another model (Model 2).
1. It seems from my reading of the SPM listserve that the second level "one-sample t-test" and "multiple regression" options would give the same results for this. Is this correct? Are there circumstances where these are different?
2. It seems from my reading of the SPM listserve that the first model would have A,B,C, and D entered as separate "covariates", and the second model would have A,B,C,D, and E entered as separate "covariates" and in each case I would code the A "covariate" as 1 and all the rest 0's to get the separate results for the 2 models. Is this correct?
3. When I do this, the first model gives 3 significant clusters and the second model gives 3 significant clusters. 2 of the 3 clusters are in the same regions across the two models after multiple comparisons correction, but the 2 clusters are bigger in the second model. I would have expected that modeling an additional covariate would have resulted in smaller clusters since more of the variance presumably would have been accounted for. Does it seem feasible that I could get bigger clusters with including an additional covariate in the model?
4. The third significant cluster was in a completely different region from Model 1 to Model 2. I understand how including an additional covariate in Model 2 could result in a significant cluster dropping out of the Model 1 results, but it is puzzling to me that a new cluster in a different region of the brain could be significant after including an additional covariate. The third cluster in both models was borderline just significant, so maybe that could explain the results, but it is feasiable that a new significant cluster could become significant with the inclusion of an additional covariate?
Many thanks for your help with this,
Duke
|