Mixed effect analysis show reasonable Group and Condition effect, but Interaction effect all over the brain even after FDR corr

Dear Dr Yan,

     I run the Mixed effect analysis using DPABI, with  the between-subject factor as Group, and the within-subject factor as Condtion. The Group effect and the Condition effect F value nii image was corrected using GRF, and several brain regions showed significant effects. However, the Interaction effect nii image showed extremely large F value (up to 6.47787e+11). After FDR correction with a q value of 0.0001, the significant voxels were still all over the brain (as Seen in the following pic). 

     To check whether the Interaction effect is correct, I used Image Calculator to obtain a difference map of Condition 1-Condition2 for each subject in each group, and then Run a two-sample t test to examine the group difference. I think this procedure is equivelent to Mixed effect analysis's interaction effect or repeated measure ANOVA. I obtained significant differnce (after GRF correction) only at restricted brain regions, instead of broadly distributed brain regions in the Interaction effect of Mixed effect analysis. This makes me doubt whether the Mixed effect analysis is correct ? OR what's the reason behind the above discrepancies?  What should I do to get the correct Interaction effect?




I guess there might be something wrong with your model specification.

Dear Dr Yan,

     Thanks for you reply.

     I check the model specification, the image file folders are  arranged as Group1Condition1, Group1Condition2, Group2Conditon1, and Group2Condtion2 as shown in your course PPT. 

    I re-run the Mixed effect analysis, and obtain the same Interaction result.

    Weiredly, when I deleted all covariates, the Interaction effect seems reasonable, and only several brain regions shows significance after GRF correction (though FDR correction results no significant voxel). Are there other explanations for it?



Then check your covariates. There might be some problems.