[R-sig-ME] modelling proportions, with aggregated data, and the new/old lme4

[Ben Bolker]

Joerg Luedicke <joerg.luedicke at ...> writes:

> I would certainly check out a Poisson model with the number of
> successes as outcome and successes+failures as an offset.

  That seems odd to me; the Poisson+offset model
should be appropriate when p<<1 (i.e. for very small p,
the Poisson variance mu=n*p is approximately the same as
the binomial variance n*p*(1-p); in this case p is not small.

  Of course, I could be wrong.

> 
> On Sun, Mar 18, 2012 at 11:51 AM, Rolf Turner <r.turner at ...> wrote:
> > On 19/03/12 07:10, Ben Bolker wrote:
> >
> > <SNIP>
> >
> >
> >>> head(dat)
> >>>   successes failures id sex subdist
> >>> 1       560      726  1   F       4
> >>> 2       844      510  1   M       4

  [snip]

> >>>
> >>> In community #1, which is in subdistrict #4, there are 560 women
> >>
> >>   you mean 510, right?
> >
> > <SNIP>
> >
> > Sure looks like 560 to me.  Time for a trek to the optometrist, Ben?

  Looked at the wrong line (failures in men rather than successes in women).