Dear all,
I was wondering what the best way is to check the assumptions (and what
are they?) for a linear mixed model (LMM). I'm sorry if my questions
sound a bit stupid, but I'm really a newbie in the field of mixed modeling.
I started out using normal multiple linear regressions and checked the
assumptions (linear relationship, normality of residuals, homogeneity of
variances, absence of multicollinearity) both visually (via the plot(lm)
command) and with tests (normality, heterogeneity of variances,vif)
Because of the way my samples were collected, it was advised to me to
use linear mixed effect models with a random variable to account for
spatial pseudo-replication (I did this by including the following term
random=~1|"region", with region being the grouping factor). So the fit
looks like this
lme.1<-lme(y~a+b,random=~1|
region)
Now, can I just proceed with the terms as specified in the significant
regular linear model as fixed terms and add the random effect? What I
mean with the "terms as specified in the regular linear model" is that
in order to comply with the assumptions of multiple linear regression, I
had to transform one of the variables or add a quadratic term. Can I
just you these transformed variables in the LMM or do I need to start
again with the raw variables? And linearity, is it not an assumption of
LMMs?
According to the book of Pinheiro and Bates (2000), there are
assumptions associated with the fixed (within group errors are
independent and normally distributed) and the random terms (random
effects are normally distributed and independent for the different
groups). Using the fit specified above, what is "group" and what is
"random effect". I thought group was determined by the grouping factor
(region), but random effect is ~1|region which is the same? Do you need
to test the random effects for each fixed effect?
I hope I'm being clear
Thanks for any help or advise!
--
*/Ellen Pape (PhD student)/*
Ghent University
Marine Biology Research Group
Krijgslaan 281, S8
9000 Ghent (Belgium)
Email: Ellen.Pape@Ugent.be
Tel: ++ 32 (0)9 264 85 23
Fax: ++ 32 (0)9 264 85 98
[[alternative HTML version deleted]]