Re: A few questions on applied linear mixed models

I'm interested in mixed models as well but not an expert. Some
comments, if you are interested in "between subject" differences in
measurement technique, not sure you would test that with a random
covariate, which models the variance between subjects. I'm not sure
what you mean by "between subject differences". I think "error term"
or "residual" when I hear that. Wouldn't you expect zero variance, ie.
independence, between subjects? The AR covariance structure is
typically used to model the covariance between measurements across
time, not sure why you are using it to model variance between subjects.
It seems like the natural test is the difference in means between
measurement techniques. Seems like you could simplify things by
modeling the within subject difference (ie. TcPC02 minus PaCO2) over
the five measurements.

In mixed models, usually the variance terms are regarded as "nuisance"
parameters and are not of interest for testing, usually hypotheses of
interest are the fixed effect means.

.