Skip to content
Prev 10629 / 20628 Next

Presenting results of a mixed model containing factors

Sarah Dryhurst <s.dryhurst at ...> writes:
Note here that your among-block variance is effectively zero ...
As is generally true in R modeling (with the default treatment
contrasts), the main effects represent the difference between the
treatment and the control at the baseline level of the other effect.
The interaction represents the difference between the double-treatment
effect and the combined (additive) effect of the two treatments.
The "main effects" are the effects you see, but they have to
be interpreted carefully.
Most of these questions aren't specific to lme4, it's just that
the standard suite of answers get more complicated in _any_ situation
(glm(), lmer(), etc.) where there isn't a simple, unique additive 
decomposition of effects.  There is a fair amount of controversy
even about how to handle main effects in the presence of interactions
(the "SAS type III SSQ/marginality" argument): e.g. google for
"venables exegeses linear models".

  There is a strong effect of the interaction in your case.
If you want to get 'average' effects of the main effects, which
may or may not make sense, you can use sum-to-zero contrasts.

  Ben Bolker