Hi So-Wan,
I'm not a stats freak, just preparing for an exam in multivariate
statistics and multiple regression methods, trying to test my progress
by answering your questions. No guarantee taken ...
> it's an analysis with 2IVs and 3 possible moderators, i
reckon that it may be better that i test each moderator
separately,
If you are sure there can't be higher order interactions then that
should be fine. In that case I would run three separate multiple
regressions each including the IVs and their combinations with the
actual moderator.
If you do assume a second order interaction, I would take the products
of each IV with all possible combinations of two moderators into the
analysis. Then run separate analysis for all combinations with each
moderator.
You probably don't assume a third order interaction, if so just create
variables with all the moderators for each IV.
but i'm still not very sure what i should put
into each 'block' in the analysis.
Do you mean in sequential regression? I think standard regression is
better for moderation analysis, because you don't need a test Rēchange.
You don't want to know if the regression model is improved by taking the
interaction term into the equation. In fact you don't really care about
the outcome for main effects of the IVs as long as you get a significant
b-weight of the interaction term, which means the moderation actually
takes place. But may be I got it wrong.
> i know i have to create interaction terms for each IV with
each moderator; however, do i need to create interaction term
by multiplying the 2 IVs, and another interaction term with
the 2 IVs and the moderator?
Only if you think that one IV could be moderating the other - which
probably means - no.
I hope that's helpful.
cheers
Nina
|