[R-sig-ME] mixed effect model query
Emmanuel Charpentier
emm.charpentier at free.fr
Wed Aug 5 18:02:07 CEST 2009
Fresh out of my limbic system : isn't searching to adjust odel with
fifth-order interaction begging for trouble ? Even with a $#!+load of
data ? Especially wit very few levels of the random effects ?
Also, I do not think that the "tube" effect for females has any reason
to be corelated to the "same tube" effect for males (in fact, your
"tube" effect should be sex:tube).
Other than that, lmer should be able to fit such a model. I'd probably
start with a first-level interaction model, wtching out for evidence of
overdispersion (and reverting to a quasi- model if necessary) :
model1<-lmer(successful~(maleage+relweight+malemobtrue+femmobtrue
+timetoattemptsecs)^2+(1|maletub)+(1[femtub),family=binomial)
HTH,
Emmanuel Charpentier
Le mercredi 05 août 2009 à 14:36 +0100, Mcdonald, Grant a écrit :
> Dear sir/Madam,
>
> I have a designed experiment where females insects were either allocted an old or young male insect and allowed 30 minutes in which to mate.
>
> My response is there for a 1(success) or 0(failure to mate),
> and my explanotory variables are therefore, age of male, the ratio between male and female size, and index of male and female activity level, and the time it took within the 30monute mating trial to start a mating attempt in the first place.
>
> I also have though information on which mass rearing tubs the insects were grown, two tubs for females and two tubs for young males and two tubs for old males (the tubs for female and youg males are the same). I wanted to add these in as random effects. All combinations between tubs were represented almost equally in a 2x4 experimental design (2 female tubs and 4 male tubs)
>
>
> my model would look like this,
>
> model1<-lmer(successful~maleage*relweight*malemobtrue*femmobtrue*timetoattemptsecs+(1|maletub/femtub),family=binomial)
>
> does this seem correct as i feel i cant used aov due to the non normal error structure and glm will not allow me to use tub as a random effect
>
>
>
> Kind Regards
> Grant McDonald
> Imperial college London
>
> P.S. I apologise if this message is in poor format as I have not emailed before and i am unsure as to how to go about it.
>
More information about the R-sig-mixed-models
mailing list