Hello Niklas,

Near the end of your enumeration of design variables, you state:

"three independent variables, scale: interval"

Did you mean to say, three dependent variables?

Moderation simply means that there is an interaction between a given IV and the moderator variable. Interaction simply means that differences in levels of one factor (e.g., communication type) are inconsistent across levels of another factor (e.g., incongruity). It doesn't mean that IV levels track identically or in a complementary fashion.

Could this data set be handled via mancova? Yes, most likely.

Could it be handled via multivariate regression? Yes, most likely.

Could it be handled by running three univariate regressions or ancovas? Yes, most likely.

The choice of which is best would hinge on your specific research question(s), and how you intend to use the results.

A lot of folks find man(c)ova troublesome since the first thing that users often do upon finding a significant effect is to run univariate analyses and report on those. Hence, the "reality" of a multivariate effect is essentially ignored.

The concept of an arbitrary linear combination of scores on a set of DVs that maximizes the distance of group centroids in space is kind of tough to chew on, conceptually. As well, the "arbitrary" part might well imply that the specific combination holds no special meaning in the context of the variables for that population. At least have a look at the parameter estimates which can help you to understand which variables contribute to the difference, in a relative sense. Discriminant analysis subprograms typically offer this type of information as a matter of course.

Good luck with your work.

what are your outcome variables(measurments or proportions) and how many groups? This will tell you what mode of analysis is required.