 |
 |
Quoting Matthew Brett <[log in to unmask]>: > It has some small discussion of the topic. You also may have seen the > recent Vul et al paper complaining about _plotting_ peak voxels from > SVC corrections. The argument would be that it is valid to do small > volume correction to test significance, but you should not use the > peak t value or the plots from the peak voxel to add to your argument, > because they will now give a biased impression of the strength of your > effect. Outside neuroimaging, if you do a canonical correlation and draw the plot, you'll have the same effect. Virtually every plot in multivariate statistic gives a 'biased' impression of the strength of the correlation, in this reading. This is because plotting the peak statistic isn't much different in this sense than plotting the correlation for the best linear combination of the variables. Same holds for plot of a posteriori contrast after Sheffe's post hoc comparison in ANOVA. The effect is ubiquitous. Best, Roberto Viviani Dept. of Psychiatry III University of Ulm, Germany ----- End forwarded message -----