Dear Experts,
I have startede experimenting with GraphVar, it is very interesting but I have a few questions.
1.) In my study I perform behavioral testing as well as resting state scans before and after a training session. I am using "group comparison" to run a paired t-test on multiple graph metrics. I define a collumn in the variable spread sheet called "group" with value either "pre" or "post", with all of the pre scans listed followed by all of the post scans listed in the same order. This seems to be working fine.
My question is if I can include a behavioral covariate in this paired T-test analysis using graph var. For each subject I also have behavioral measures of performance before and after training.
2.) I actually have 2 consecutive resting state scans for each subject pre and post training (a total of four scans per subject). I cannot find a way to set up the contrasts for performing the nescessary statistics properly with graphVar. I simply calculated correlation matrices for all scans, transformed the correlations to Z-values, and averaged each pair of consecutive scans for each subject before converting back. I then carried on the analysis using 2 correlation matrices per subject (one pre and one post, each an 'average' of two consecutive scans). Is this valid?
3.) If I test against random groups and generate random networks, and select "random networks/groups" in the corrections menu of the results viewer, am I in fact performing a correction for multiple comparisons? I am having difficulty understanding the manual in this respect. I am confused as to what graphVar is actually doing to correct.
Thanks so much for the help,
Dimitri
Behavioral Regressor with Paired T-test
Hi Dimitri,
to your questions:
1. this seems correct
2. the current version of the programm does not support adding covariates for paired t-tests. This would have to be implemented by using a GLM (we will try to include this in one of the next versions). With respect to the averaging: you probably could do something like this (although the question would be why you want to average both pre and post scans respectively - to get a more adequate measure? However, I am not an expert for this question). For performing the set up you want to do, you could use the toolbox Metalab_GTG "https://www.nitrc.org/projects/metalab_gtg/"; there you can specify a GLM.
3. If you tests agains random groups (i.e., non-parametric permutation testing), you do not have to create subject specific random networks (you could if you would like these for normalizing your graph topological measures -> checkbox "normalize graph metric with random networks" in the "Network Calculations" panel) as the non-parametric p-values are derived by permutation testing. If you select "random networks/groups" in the results viewer, the displayed p-values will be the permutations testing derived p-vaules (there is still no correction for multiple comparisons involved). To do multiple comparison correction with your non-parametric p-values you would have to select "random networks/groups"/Bonferroni" or "random networks/groups/FDR".
I hope this helpls,
Johann
Johann,
Johann,
Thank you so much. I will try the GLM. The averaging is to get a more adequate measure, as you said. With a GLM I suppose it wouldn't be nescessary.
Thanks again,
Dimitrios