Dear Elizabeth,
> I have one group design with three factors [race (three levels), gender (2 levels) and condition (3 levels)]. To model this I have created one EV for each cell for a total of 18 EVs. So far we have only considered two of the three factors at once, either gender & condition or race & condition.
>
> We have performed pairwise comparisons between levels of one factor (e.g. race) within one level of the other factor (e.g. condition). So for example we asked during condition 1 where is Race1 > Race2, Race2 > Race1, Race1 > Race3, Race3 > Race1, and Race2 > Race3, and Race3 > Race1.
>
> We now want to present a graph of extracted parameters for all three races side by side, but to do so we need clusters that reflect the main effect of race within a condition, that is anywhere there is a difference between any of the three races.
>
> I think I need an F-test to ask this question that I would enter at the first level (single subject, single run), but I am not sure about which level I should enter it at. At the first level, when I click on all 6 of the above contrasts in an F-test, I get the error that F-tests must contain contrasts that are not linear combinations of others. When I only click half the contrasts, so as not to include both directions, I get the same error. Any information on how to properly set up this F-test would be greatly appreciated. Thanks!
>
let us "pretend" you only have three conditions and want to know where there is any difference between them. The contrast [1 -1 0] would give you the difference between A and B, [1 0 -1] between A and C and [0 1 -1] between B and C. It would seem reasonable to assume that the F-test would then comprise these three contrasts. BUT, if you have checked for a difference between A and B, and for a difference between A and C you have implicitly also checked for a difference between B and C (if there is no difference between A and B and no difference between A and C then we know there is no difference between B and C, right?). That means that the [0 1 -1] contrast is superfluous, and in fact a linear combination of the other two.
So, just loose your third contrast from the F-test and you should be fine.
Good luck Jesper
|