3dMVM contrasts vs 3dttest++


I have a question regarding linear contrasts output from 3dMVM in comparison to output from 3dttest++

I have a 2x3 within subject design.

factor1: with 2 levels= pre, post

factor2: with 3 levels= one, two, three

We ran 3dMVM and setup several glt tests. One of the tests involved collapsing across 3 levels for Factor2 for level pre of Factor1

-gltLabel 1 ‘pre_Factor123’ -gltCode 1 ‘factor2 : .333One.333two .333three factor1 : 1pre’ \

The second contrast involved collapsing across 3 Factor2 for level post of Factor1

-gltLabel 2 ‘post_Factor123’ -gltCode 2 ‘factor2 : .333One .333two .333three factor1 : 1post’ \

The third contrast tried to find the difference between pre-post factor1, collapsed across all 3 levels of factor2.

-gltLabel 3 ‘Factor123_pre-post’ -gltCode 3 ‘factor2 : .333One .333two .333three factor1 : 1pre -1*post’ \

Is this correct?

We collapsed across all 3 levels for factor 2 and ran a simple paired ttest with 3dttestt++ to cross-check. We get pretty much the same result for means and mean differences, but the t and p-values are different? Why is that the case?

I think it might be the degrees of freedom.

We have 36 subjects per each of the 3 levels for Factor2. The statpar from 3dttest++ is 107. The statpar for the bucket of each of the 3 contrasts described for 3dMVM is 35.

Why is this?




Simply based on the information about the degrees of freedom, it sounds like your 3dttest++ script is incorrect. If you post the 3dttest++ script here, I may be able to spot the problem.

A side note: for such a 2x3 within-subject ANOVA, you may use 3dANOVA3 -type 4, which would give the same results as 3dMVM, but runs much faster.