Post hoc for multivariate analysis with large number of dependent variables?

Hello All,

Basically, the study is looking at if forefoot striking (FFS) and rear foot striking (RFS) runners experience similar changes when switching from shod to barefoot running. We had a group FFS and a group of RFS runners both run barefoot and with shod. I used a GLM multivariate procedure with a full factorial model with shoe condition (barefoot & shod) and foot strike (FFS or RFS) as fixed factors. For the contrasts types I had foot strike as a deviation and shoe as repeated. The foot strike * shoe interaction is of interest as we want to see if the groups differ in terms of the changes seen when switching from shod to barefoot running. We are also interested in where the specific differences lie; specifically, does RFS shod differ from RFS barefoot, does FFS shod differ from FFS barefoot, does RFS shod differ from FFS shod and does RFS barefoot differ from FFS barefoot? However, the main effects of shoe and foot strike from the multivariate test collapses everything together. For example the shoe effect compares all barefoot (FFS & RFS combined) to all shod (FFS & RFS combined). I was under the impression that I could use post-hoc t-tests with a Bonferroni correction to evaluate the specific differences, but we have 21 DV’s (3 joints (ankle, knee and hip) x 3 planes x 2 peaks and value at contact, +3 ground reaction force components) so the level of significance becomes 0.002. At that level we’re unlikely to see anything.

Is there is there some way to get the specific group/shoe comparisons from a multivariate test? Or is there another type of correction that would take into consideration that the high number of DVs? Or am I completely off base and should look at another test?