[R-sig-ME] randomized block design model specification

jos matejus matejus106 at googlemail.com
Thu Dec 3 13:10:31 CET 2009

Dear list members,

Could anyone shed some illumination on the best way to specify a model
with the following experimental design?

Briefly, I have 75 individual rats, 47 of which are lactating and  18
are non reproductive. I measured the gene expression of 6 genes for
each rat. The underlying research question is whether the gene
expression of rats is different depending on reproductive status and
whether this varies between genes. Many of the 7 genes are strongly

My first approach was to use a simple linear model with gene
expression as the response variable and reproductive status and geneID
as explanatory variables. Something like:

M1.lm <- lm(expression ~ reprostatus*geneID)

However, as the expression levels of the 6 genes were measured from
each rat, there is potentially an issue with non independence. I
therefore thought I could use the rat ID as a random effect. Something

M2.lme <- lmer(expression~reprostatus*geneID+(1|ratID)

My question is whether this seems like a sensible approach. I guess I
am confusing myself a little as I cant visualize the consequence of
having rat ID as a random effect as I only have one observation for
each rat gene combination. Also, does the above lmer model assume that
the reprostatus*geneID interaction in the fixed effects is the same
for each rat? If so, does it make sense to include this interaction in
the random effects term aswell? (although my intuition tells me I dont
have the replication for this)

Many thanks in advance for any advice offered

More information about the R-sig-mixed-models mailing list