Type of statistical test in GraphVar Turbo

Submitted by hpsig86 on

Dear GraphVar team.

I just wanted to ask what sort of statistical tests are used while doing the following comparisons using GraphVar GLM Turbo (for reference I am using structural connectivity matrices from DTI tractography as inputs):

1) Having "Group" (patients vs controls) as a between factor and three variables as nuisance covariates. I.e. assessing differences between the two groups on graph metrics. Are the nuisance covariates regressed out prior to t-tests?

2) Having clinical measures as between covariates and three variables as nuisance covariates? I.e. assessing significant correlation between graph metrics and clinical measures in a patient population. I assume that this does a multiple regression as the outcome shows beta regression weights?

Thank you,

Hilmar

Dear Hilmar, Yes, you're correct. The patients vs. control model with nuisance covariates is actually something we run a lot for our own data. The nuisance covariates are always regressed out prior to the t-test/regression. GraphVar then shows the group difference in case of the t-test, and the standardized regression weights (scaled to be like correlations, from -1 to 1) in case of regression, and the T-values and p-values for both. You can also run an ANCOVA if you add both the between covariates and the between factor in the same model. Under the hood, GraphVar runs a GLM and then calculates F-tests for speed, but this is mathematically equivalent to running t-tests, regressions and AVCOVAs. Thank you for using GraphVar! Best, Lea Waller

Dear Lea.

Thank you for the prompt reply. This really helps - always good to know better what does on under the hood.

I have one more question if that is OK. I hope its alright that I post this question in the thread I created, although the two questions don't relate directly.

I want to recreate a plot of group means on global efficiency from the results viewer in better quality (i.e. when pressing show/hide group means). Apparently pressing "Export" does only export the difference in means. So I tried to export these values using "Calculate variables and export" option with same set up, but when I plot means of groups the graph looks completely different than in the results viewer. Is there something I am missing? I hope my question is clear.

Kind regards and thanks,

Hilmar

Hi Hilmar,

I totally forgot to mention that you also can export these corrected group means from the results viewer via "export" if you select the respective group in the drop-down window (here you have several choices: e.g. 2 groups: difference in means =  sex_female - sex_male; mean group 1  = sex_female = 0; mean group 2 = sex_male = 1). 

Best,
Johann

Johann Kruschwitz

Fri, 02/10/2017 - 13:55

Permalink

Hi Hilmar,

with respect to your previous post about nuisance covariates and ANCOVA I belief that the discrepancy you observe in the exported versus GLM results occurs as the exported graph measures (calculate and export) are the "raw" graph metrics and the difference in means (results viewer) is a result where the nuisance covariates are already regressed out from the graph measures prior to the t-test/regression.

Best,

Johann

Forums
GraphVar