Dear FSL,
I have a question with regard to the results I get from randomise. The results for the same subjects change a lot depending on how many groups I use as input in randomise and I wonder if this can be correct.
I did the following analyses: I compared two resting state scans with each other. I compared a resting state scan after an intervention with a resting state scan after a control condition. The first time we only analyzed one group of subjects. I performed a paired t-test to compare the two resting state scans (of the same subject) with each other in the following way: I performed dual regression to get the stage 2 output. For each subject I made a difference map (of the two resting state scans) based on the stage 2 output. Then I used randomise to perform a one sample t-test on the difference map and get the results for my networks of interest. For this purpose I used the following design matrix and contrast:
matrix:
/NumWaves 1
/NumPoints 168
/PPheights 1.000000e+00
/Matrix
1.000000e+00
1.000000e+00
1.000000e+00
.. (with 168 ones below each other)
contrast:
/ContrastName1 R2minR1
/NumWaves 1
/NumContrasts 1
/PPheights 1.000000e+00
/RequiredEffect 0.301
/Matrix
1.000000e+00
This resulted in a lot of significant results. Now as data collection was ongoing, we got the opportunity to add a second group to our analysis and see if there were differences between these groups. So we changed the design matrix and contrast to the following:
matrix:
/NumWaves 2
/NumPoints 214
/PPheights 1.000000e+00 1.000000e+00
/Matrix
1.000000e+00 0.000000e+00
1.000000e+00 0.000000e+00
..
0.000000e+00 1.000000e+00
0.000000e+00 1.000000e+00
.. (with 46 ones below each other in the first column and 168 in the second; so in the second column we put all the subjects that we also analyzed in the first analysis stated above)
contrast:
/ContrastName1 main effect co
/ContrastName2 main effect pt
/ContrastName3 pt minus co
/ContrastName4 co minus pt
/NumWaves 2
/NumContrasts 4
/PPheights 1.000000e+00 1.000000e+00 1.000000e+00 1.000000e+00
/RequiredEffect 0.534 0.279 0.603 0.603
/Matrix
1.000000e+00 0.000000e+00
0.000000e+00 1.000000e+00
-1.000000e+00 1.000000e+00
1.000000e+00 -1.000000e+00
Now we expected to find more or less the same results for the first analysis with the 168 subjects as for the main effect (contrast 2) in this second analysis (since we use the same subjects in both analyses). The results do indeed look really similar in fsleyes, but the threshold has changed drastically. In the first analysis mentioned above we had a lot of significant results at a threshold of 0.9875. In the second analyses we do not find any results for the same subjects at this threshold. To get the same results in fsleyes we have to use a threshold of 0.6. Is it possible that the threshold changes this dramatically from adding 1 more group in the design matrix or can there be another explanation? I hope you can help me with this!
Best, Jasper
########################################################################
To unsubscribe from the FSL list, click the following link:
https://www.jiscmail.ac.uk/cgi-bin/webadmin?SUBED1=FSL&A=1
|