05-15-2014 10:17 AM
Can using multiple comparison procedures that accommodate for correlation between linear combinations (such as MaxT, MinP, Tukey etc) be used to control the FWE for any set of non-orthogonal contrasts? That is to say, could I construct a contrast-matrix involving any set of non-orthogonal vectors with overlapping information and still be sure that FWE is controlled and that the interpretation is valid? (Given that the contrast had a valid interpretation to begin with...)
05-19-2014 02:01 PM
I believe the MinP method would cover this (see page 285--288 in Westfall et al., Mutiple Comparisons and Multiple Tests Using SAS, 2nd ed.).