Originally posted by Alfonso Nieto-Castanon:

Hi Irene,

An analysis that uses groupCONTROLS and groupPATIENTS (where each group variable is encoded with 0's and 1's indicating the subjects within each group) and evaluates a contrast [0 1] is very similar to an analysis that uses groupPATIENTS with a contrast [1]. Both analyses estimate the average connectivity in patients, and both evaluate whether that average connectivity strength is significantly different from zero. The only difference is that the "groupCONTROLS,groupPATIENTS [0 1]" analysis assumes that the residual intersubject variance is equal across the two groups and uses the control subjects to better estimate that variance, while the "groupPATIENTS [1]" variance does not make that assumption so it uses only patients to estimate that residual variance (well, and probably I should add that this variance is important because in both analyses "significance" is determined by contrasting the size of the average connectivity strength in patients vs. the size of this intersubject variance). In general the former approach is prefered in the context of ANOVA (as the ANOVA analyses will make the same equal-variance assumption), while the latter is prefered in other contexts (as it makes minimal assumptions).

Hope this helps

Alfonso

Originally posted by Irene Lozzi:

Hello, I'm a newbie in MRI and I'm running 2nd level ROI2ROI analyses in CONN on a sample of patients and matched controls. 

With difference group1 > group 2 or viceversa, contrast 1 -1 or -1 1, I get no significative results. 

If I set "Effect of groupPATIENTS" (Does the average connectivity within groupPATIENTS differ from 0, when estimated at the zero-level oh groupCONTROLS?), with contrast 0 1, I get several significant clusters. 

A colleague told me that it's basically because this way I'm just considering within-group connectivity, ignoring the control group. But if I run "Does the average connectivity within groupPATIENTS differ from zero", the results are completely different. 

So, what am I really doing with a 0 1 contrast? 

Thank you for your help.