Hello, this is my first post although I have been a subscriber to the list
for a couple years now, and this is more of a general question on data
presentation than a specific R problem. I am currently in a weekly group
that has been working mainly through examples and data with lmer, and as of
late we have been discussing the issue of how to present data. We have been
facing considerable debate on how to graphically display models that contain
multiple factors and often significant interactions.
Sometimes a final model contains several linear terms, factors and
interactions. However when presenting the data as a figure it might be
desirable only to plot the effect of one interaction between a linear
variable and a factor, whilst controlling for the effect of other terms in
the model (i.e. one should plot the model means, as given by other
software). The question is how should one correctly do this?
One approach would be to use the predict() function to obtain predicted
values from the final model after setting the covariates to some nominal
level. However, it is not clear how to correctly calculate the standard
error for groups using this approach. Plus it is so far not possible to use
predict() with lmer.
Additionally, is it best to present the observed data in the figure with the
model estimation? And how should model variation be presented?
Sorry, if this post is confusing, but it is a synthesis of questions from
multiple people.
Cheers,
Mikey