Dear Prof. Andrew,
I am very interested in applying NBS to my DTI data. But I got a problem: how can I know the decreased/increased connectivities between two groups when my data consists of three groups? I used the F-test to find a general difference between any of the three groups (using contrast [1,1,1]) and the NBSview gave a good result. However, I want results between any two of the three groups not just one. And if there were several covariates like age, education, is that I just need to add two columns in the design matrix and set the corresponding contrasts with 0?What can I do with my problems? Would you help me?

qin

Dear Qin,

Regarding your 2nd question about nuisance co-variates: in short, yes, add a separate column modeling each co-variate to the design matrix and set the corresponding contrasts to zero. Click on the Help button on the Statistical Design panel to see an example of nuisance co-variates (ANCOVA).

Regarding your 1st question about the F-test: to determine which groups are driving the effect identified with the F-test, I suggest printing to the screen the connectivity strengths for each group and then either studying the pattern of connectivity strengths between groups, or more formally, performing an appropriate post-hoc test. This should be repeated for each connection comprising a significant network. See page 30 of the reference manual for details about how to print out the connectivity strengths to the screen.

Andrew



Originally posted by qin xu:

Dear Prof. Andrew,

Thank you very much for your help!

Qin

Hi Andrew and Qin,

I have come across the same question of Qin's first one. If I am going to perform a post-hoc test after doing the F-test in the NBS (of course, the premise is that I get a significant result in this step), what do I need to do? I guess two alternative choices. (1) Perform pair-wise comparisons between the three groups using t-test in the NBS (i.e., suppose having A, B and C groups. Possible comparisons include A vs. B, A VS. C, B VS. C). (2) Extract some variables from the nbs variable which is gotten by typing the command global nbs in the matlab, and then use a post hoc test to compare.
I am not sure whether the guesses are appropriate. Would you like to give me some suggestions? Thanks so much!
Additionally, anyone who sees this post can also give me suggestions and I will highly appreciate. Thanks a lot!

Jiaolong

Hi Jiaolong,

Both options that you have suggested seem very reasonable to me. Remember that if you use three separate t-tests (this is probably what I would do), the significance level should probably be reduced to p = 0.05/3 = 0.02 to account for multiple comparisons, although some might argue that p=0.05 is ok.

If you extract the variables using "global nbs", you can use something like Tukey's test to do post hoc analysis.

Andrew
Originally posted by jiaolong qin:

Hi Andrew,

Thanks a million for your prompt reply.

Jiaolong