[R-sig-ME] Help with some nearly-binomial data
Mike.Lawrence at dal.ca
Tue Apr 27 17:06:40 CEST 2010
If the purpose of the control condition is to to look at the relative
effects, why not simply run a binomial model on all the data, test and
control, add group as a variable and look for group*generation
On Tuesday, April 27, 2010, Hywel Lloyd <hapdlloyd at googlemail.com> wrote:
> Hi, I've got stuck trying to model some data. Sorry for the trouble, but if
> anyone's got any help or suggestions I'd be really grateful.
> I'm trying to use mixed models to look at the rate that the hatching success
> of butterflies recovers after inbreeding over a number of generations. The
> data I have is the proportion of eggs that hatch for test and control
> populations so I would use a binomial distribution.
> But the main value I am interested in is the inbreeding depression, *s*,
> calculated as 1 - (test hatch rate/control mean hatch rate). This has two
> important roles - to give the relative decline in fitness and, by using the
> contemporaneous outbred hatching in each generation, it helps to control for
> other (lab) influences that may affect hatching from one generation to the
> next. The problem is though that occasionally the inbreds outperform the
> control, giving negative values for *s* which can't then be compared to the
> binomial distribution. Originally I simply used the test line mean hatch
> rates and assumed a normal distribution under the central limit theorem but
> this cut my number of data points from ~450 to 15 which seemed a bit
> Does anyone have any suggestions for how I can proceed?
> I had considered simply changing the equation for *s* to 2 - (test/control),
> but I'm not sure if I could then convert my estimates for the fixed effect
> back into terms of the original inbreeding depression?
> Thanks for any help you can give me.
> [[alternative HTML version deleted]]
> R-sig-mixed-models at r-project.org mailing list
Department of Psychology
Looking to arrange a meeting? Check my public calendar:
~ Certainty is folly... I think. ~
More information about the R-sig-mixed-models