[R] assessing the fit of a LME model
Thomas W Blackwell
tblackw at umich.edu
Tue May 13 18:19:23 CEST 2003
I'm coming to this from outside the field, and I hope Doug Bates
will reply ... but I would question the use of normal theory linear
models with counted data. Why not logistic regression (binomial)
or log-linear models (Poisson) ?
You see, coming to the question from outside, I'm still back at
the point of needing to answer *that* question for myself, before
I can think about the random-effects aspects of the problem.
I don't yet understand "vials within cages". Are the vials open
or closed ? What determines the total number per vial ? (Thus,
is a binomial or a Poisson error model appropriate ?) And, what
is "temperature" in the Bartlett test, and why does it make sense
to *cross* "cage" with "temperature" ? Was every cage in fact
tested at multiple temperatures ? (I would guess not, but I don't
fully understand the experimental setup.)
So ... in the end ... I offer you many questions and *zero* answers !
- tom blackwell - u michigan medical school - ann arbor -
On Tue, 13 May 2003, Federico Calboli wrote:
> I would like to ask a couple of questions on a LME model.
> I tested 4 selection lines at 4 food concentrations against a standard
> competitor stock. I had 3 replicate cages per selection line. In each cage
> I have 10 vials. I counted the number of wild type flies and competitor
> stock emerging in each vial. My main question is: is there any difference
> between selection lines?
> I did fit the following model:
> mod1<-lme(wt~selection*food, random=~1|c1/food, competition)
> The quantile plot is straight, the plot of residuals looks good, the
> standardized residual plot seems ok... BUT if I do a bartlett test for
> homogeneity of variance (done by cage*temperature, 48 cages in all) the
> variance is NOT homogeneous. Is my model still accettable? if not, what
> could I do?
> The same model on asin(sqrt(wild type/total)) does not fit anywhere as well.
> my dataset (NB *wt* is the number of wild type flies in each vial; *spa* is
> the number of comepetitor stock flies in each vial, *p1* is
> asin(sqrt(competition$prop)), *c1* is the cage nested in selection using
> selection cage food wt spa tot prop p1 c1
> (475 lines of data deleted)
> Federico C.F. Calboli
> Department of Biology
> University College London
> Room 327
> Darwin Building
> Gower Street
> WClE 6BT
> Tel: (+44) 020 7679 4395
> Fax (+44) 020 7679 7096
> f.calboli at ucl.ac.uk
More information about the R-help