 |
 |
Hi Patrick, With 3 groups, the procedure of running an F-test (ANOVA) across all groups, then looking at the pairwise t-tests to see which group differences are significant and driving the overall F-test, can be done, and the error rate is controlled exactly. However, with 4 or more groups, this procedure no longer guarantees control over the error rate for all pairwise t-tests, and can expose you to increased number of false positives. The way to address this is, using the exact same data, design, and t-contrasts, use the option "-corrcon" available in PALM. The F-test can be dropped. The results will be saved in the file "cfwep" files. Use the option "-save1-p" if you'd like 1-p values as the output, or "-logp" to save as -log10(p). Hope this helps. All the best, Anderson -- Anderson M. Winkler FMRIB / Analysis Group [ Blog <http://brainder.org/> | Twitter <http://twitter.com/AndersonWinkler> ] On 7 August 2015 at 20:56, Subscribe FSL Patrick McCunn <[log in to unmask]> wrote: > Hi FSL experts > I'm attempting to run an ANOVA on my FA data thats been run through TBSS. > I have 8 subjects in each of 4 groups (32 total). I have run the GLM set up > for a 1-factor 4 level ANOVA, but I'm slightly confused as to what the > contrasts and F-tests actually tell me. The experiment is being run totally > blind so I'm simply looking to determine if there is a significant > difference in voxelwise FA between groups and furthermore which groups are > statistically different from each other. I apologize if this is an overly > simple question. > > Thank your time > Patrick >