Hello, 

I have a couple of questions regarding doing analysis using Randomise. 

For reference, the study design is a two group (experimental vs. control) and two timepoint (pre-intervention vs. post-intervention) design. I analyzed the data by creating a change score for each participant (time 2 minus time 1) and then doing a two-sample t-test as suggested here: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#ANOVA:_2-groups.2C_2-levels_per_subject_.282-way_Mixed_Effect_ANOVA.29 and here: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_.28Two-Sample_Unpaired_T-Test.29

I would now like to control for age and dementia status to see how the results might be affected. I controlled for these two variables by mean centering them and then adding them to the design matrix as separate EVs as suggested here: https://fsl.fmrib.ox.ac.uk/fsl/fslwiki/GLM#Two-Group_Difference_Adjusted_for_Covariate.
My contrasts were then: 
con1 = 1 -1 0 0
con2 = -1 1 0 0
with the zeros representing the EVs for age and dementia status. I interpreted the first contrast as experimental changes greater than control changes, controlling for age and dementia status, and the second contrast as control changes greater than experimental changes, controlling for age and dementia status. 

My first question is: is this the correct way to set up design and contrast files when controlling for nuisance variables and, if so, am I interpreting the contrasts correctly? 

I expected a large reduction in significance after controlling for those two variables, but in fact, the results look nearly identical. The only distinguishable difference that I can tell is that the t-value is slightly reduced. For example, t = 3.79 now, down from t = 3.97 before controlling for age and dementia status. 

This brings me to my second question, re: the interpretation of the t-values given in the tstats file following Randomise. Can these be interpreted as regular t-statistics? And is the program default to report only positive values? I have looked for negative values, but the files seem to contain no values less than zero. Since I am looking at change scores, the sign of the t-stats is important for interpretation regarding the directionality of the change, e.g., there is a significant difference between the change scores for experimental and control groups, but how can I tell whether FA for each group increased or decreased? I am unsure whether my particular results contain only positive t-stats (i.e., increases in FA), or whether the default is set to report only positive values and there is a way I can test for negative changes as well. 

Thanks in advance for your help with this rather long post.

Best, 
Catherine Mewborn