Dear Jan,
>I recently started to use spm_ancova for analyses of behavioral data and I
>have come to really like the flexibility of it. That is, the use of F/T
>contrast to test for main effects, interactions and other specific
>hypotheses in the data. However, I do have a couple of questions about the
>function.
>
>1. Degrees of Freedom
>---------------------
>spm_ancova compute the (effective) dfs for an F-contrast using the
>Satterthwaite approximation
>
>[trRV trRVRV] = spm_SpUtil('trRV',xX,V);
>[trMV trMVMV] = spm_SpUtil('trMV',X1o,V);
>df = [trMV^2/trMVMV trRV^2/trRVRV];
>
>Here is my question: is df(2) the corresponding degrees of freedom for a
>T-contrast? (This seems to be the case (at least in spm_contrasts), but I
>just want to be sure on this.
Yes, the df(1) should always be 1 for t-contrasts - a t-contrast is
just special
case of an F-contrast (ignoring the sign)
>And consequently, can compute a p-value for the F/T contrasts using
>p = 1 - spm_Fcdf(F,df), % for F-contrasts and
>p = 1 - spm_Tcdf(F,df(2)); % for T-contrasts (T-value also stored in F)
Absolutely.
>2. Error Covariance Constraints
>-------------------------------
>What is the format of the specs for the error covariance constraints (V)?
>Is this the same as the covariance components usually stored in xVi.Vi (a
>cell array of different components)? Or do all these different covariance
>components have to be collapsed into a single matrix in order to work for
>spm_ancova?
They have to be collapsed into a single matrix V. This would normally involve
optimizing the hyperparameters of each component using ReML. For interest,
look at spm_reml_ancova; however this is a more general routine with
a slightly
more complicated model specification that allows for hierarchical models.
With very best wishes,
Karl
