Two group analyses with BayesianGroupAna.py possible?

AFNI Message Board
1

Hi,

I’ve been playing with BayesianGroupAna.py and would like to perform the equivalent of a between group t-test. Is this possible? Having read the paper, it would seem that it should be theoretically possible.

The example analysis in the paper describing the rational for BayesianGroupAna.py, operationally defined brain activity as the Z scored correlation between a seed and the other pre-specified ROIs with the effect of interest being the relationship between these z-scores and the theory of mind inventory score. However, I have a more classic task (regressors and contrasts thereof) and am wondering if I can even analyze such a task using BayesianGroupAna.py? To be more specific, I’ve created regressors of interest from my task and computed GLM contrasts using an afni_proc.py pipeline. Does it make even sense to feed the results from such contrasts (averaged over pre-specified ROIs typically activated by the class of task as indicated by Neurosynth) into BayesianGroupAna.py assuming it can perform the Bayesian equivalent of a two group t-test?

(Also (tangentially), reading the MBA help text Example 3, is SA the product of sex and age? If so, shouldn’t the last two rows in the example then have SA as positive since sex and SA are both negative?)

Thanks in advance,

Colm.

2

Colm,

Use RBA instead of BayesianGroupAna.py for region-based analysis since I’ve been maintaining and adding more functionalities into RBA.

I’ve been playing with BayesianGroupAna.py and would like to perform the equivalent of a between group t-test. Is this possible?

Currently RBA allows for between-subject (categorical and quantitative) variables as predictors. I’ll try to add the functionality of modeling within-subject variables soon.

Does it make even sense to feed the results from such contrasts (averaged over pre-specified ROIs typically activated by the class
of task as indicated by Neurosynth) into BayesianGroupAna.py assuming it can perform the Bayesian equivalent of a two group t-test?

Sure. Basically if you have a GLM under the conventional framework, you can use RBA. I’ll try to expand its modeling capability to the case with repeated-measures variables.

reading the MBA help text Example 3, is SA the product of sex and age? If so, shouldn’t the last two rows in the example then have
SA as positive since sex and SA are both negative?)

Thanks for catching the mistakes there!

3

Excellent, Gang, thanks. I’d not noticed RBA.

If I may put the fox back in the hen house: I have more than one contrast, thus there will be multiple models suggesting a multiple comparisons issue. How does one deal with this in the context of the Bayesian approach.

I just skimmed the RBA help text, example 3 interaction suffers from teh same sign issue as MBA.

4

If I may put the fox back in the hen house: I have more than one contrast, thus there will be multiple models suggesting
a multiple comparisons issue. How does one deal with this in the context of the Bayesian approach.

Are those contrasts the comparisons among the levels of a within-subject factor? If so, when I implement an approach to handling a within-subject factor in RBA, would that be able to chase away the fox from the hen house?

I just skimmed the RBA help text, example 3 interaction suffers from teh same sign issue as MBA.

Thanks again!

5

Gang,

I’ve been running RBA as follows:


	if [[ ${laterality} == "unilateral" ]] ; then
	    graphLayout="3 6"
	elif [[ ${laterality} == "unilateral" ]] ; then
	    graphLayout="4 5"	
	fi
   RBA \
		 -prefix ${laterality}.${glt}.rba \
		 -chains 4 -iterations 1000 \
		 -PDP ${graphLayout} \
		 -Subj Subj -ROI ROI -Y contrast \
		 -cVars "group" \
		 -model "1+group" \
		 -EOI 'Intercept,group' \
		 -dataTable ../${dataTableFilename} > rba_results.txt 2>&1

The group factor has two levels C and M. One thing I noticed is that in the Intercept.pdf file the graphs are all labelled with the ROI name. The same can’t be said for the group and C-M pdfs they are both absent the ROI labels. is this a bug?

6

One thing I noticed is that in the Intercept.pdf file the graphs are all labelled with the ROI name. The same can’t be
said for the group and C-M pdfs they are both absent the ROI labels. is this a bug?

Definitely a bug! Will look into the issue later.