[R-sig-ME] comparing groups of proportions

David Duffy davidD at qimr.edu.au
Fri Feb 11 03:47:31 CET 2011

On Thu, 10 Feb 2011, array chip wrote:

> Actually, for model 4 below, the syntax should be (change 1|group to 
> 1|subject):
> 4. glmer(prop.success~group+(1|subject), weights=no.trials, data=data,
> family=binomial)
> But I still have certain confusion on how can a variable with only 1 
> observation per level (like subject in this case) can be a grouping 
> variable? Any education on this would also be appreciated.

Because it can either be seen as Bernouilli with Subject and Group, or 
Binomial with Group.  The former is what allows you to see dispersion 
problems (extra-binomial variance).  One test for dispersion problems is 
simply the contingency chi-square by Subject (which I guess in this case 
would be within each group) with a large number of degrees of freedom 
(number of Ss).  One is testing whether members of the same group differ 
significantly from one another in terms of response: another name for this 
test is the "variance chi-square", and divided by the df is the measure of 
dispersion used in the most usual quasilikelihood model. The Tarone score 
test is a one degree of freedom dispersion test against one specific type 
of alternative (beta-binomial).

Just 2c, David Duffy.

| David Duffy (MBBS PhD)                                         ,-_|\
| email: davidD at qimr.edu.au  ph: INT+61+7+3362-0217 fax: -0101  /     *
| Epidemiology Unit, Queensland Institute of Medical Research   \_,-._/
| 300 Herston Rd, Brisbane, Queensland 4029, Australia  GPG 4D0B994A v

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