[R-sig-ME] what to graph for publication from an lmer model and how

[Steven J. Pierce]

Margaret,

Take a look at the effects package. It has tools designed for graphing fixed
effects (including interactions), with support for mixed models. 

Steven J. Pierce, Ph.D. 
Associate Director 
Center for Statistical Training & Consulting (CSTAT) 
Michigan State University 
E-mail: pierces1 at msu.edu 
Web: http://www.cstat.msu.edu 

-----Original Message-----
From: Margaret Wardle [mailto:mwardle at uchicago.edu] 
Sent: Wednesday, December 21, 2011 5:56 PM
To: r-sig-mixed-models at r-project.org
Subject: [R-sig-ME] what to graph for publication from an lmer model and how

Hi, I'm working on a few lmer models that all test the effect of a
stimulant drug on responses to emotional faces.  I have a few results
that appear interesting, but I'm not sure what values are most
appropriate to graph for publication to represent results from an lmer
model.  I should note that drug dose (and all other variables) are
within-subjects, that is, each subject gets placebo, 10mg and 20mg of
the drug, each subject sees videos of happy, angry, sad and fearful
faces.

To get specific I have two main types of results that I would want to
illustrate graphically:

1. There is a linear effect of dose on the intensity (0-100%
intensity) at which people are able to identify emotions in faces.  In
a normal ANOVA paper, I would present a bar graph of the mean
intensity at identification at each dose, with the SEMs as error bars.
 Would this still be recommended in  the current situation, or should
I be graphing model estimates (or partial effects) at each dose?  If
so, how do I extract those, and what should the indicator of error
variance be?  I should note that I cannot use CIs from mcmcsamp
because my error structure is too complex (1+Dose|Subject).

2. I have a few interactions between categorical and continuous
predictors -- for example, as intensity of the emotional expression on
a display face increases (continuous predictor), activity in the
subject's frown muscle (DV) increases for all emotions (categorical
predictor) except happiness.  This is exactly what you'd expect if
people are mimicking the expression on the face they are looking at -
the frown more as the display face frowns more, but not as the display
face smiles more.  I can get plotLMER.fnc to produce a lovely line
graph illustrating the positive relationship between intensity and
frowning for all the negative emotions, and the negative relationship
between intensity and frowning for happiness with each emotion
represented by a different line color.  It's great, but once again I
will need some indication of error, and once again my error structure
is too complex for mcmcsamp and HPD intervals.  Suggestions?

I'm also open to other suggestions about how to present these results,
or just references to papers that do a good job of graphically
presenting fixed effects from lmer, if you don't have time to address
my design specifically. Thanks very much,

-Megin Wardle