Skip to content
Prev 7420 / 20628 Next

lmer on dataset with missing values 'not at random'

Another approach to the problem of interactions with missing cells would be to design specific contrasts which assess interactions in those cases, but which exclude the missing cells. To do that, you would need to define the entire experiment as a single factor situation, and write the appropriate contrasts.


-----Original Message-----
From: r-sig-mixed-models-bounces at r-project.org [mailto:r-sig-mixed-models-bounces at r-project.org] On Behalf Of Kevin Wright
Sent: Monday, January 30, 2012 1:31 PM
To: Charles Determan Jr
Cc: r-sig-mixed-models at r-project.org
Subject: Re: [R-sig-ME] lmer on dataset with missing values 'not at random'

Charles,

You are trying to fit fixed effects for two-way and three-way interactions
when, as you have shown, there are no data for certain combinations of
those factors.  Generally, to estimate fixed-effect interactions you need
data in every cell of the interaction xtabs table.

You can start by removing the interactions from your model.  If the
interactions are of interest, you could try fitting the interactions as
random effects, but you need to understand this leads to a different model
with different interpretations and inferences...

Kevin
On Mon, Jan 30, 2012 at 12:44 PM, Charles Determan Jr <deter088 at umn.edu>wrote: