Non-R question.

Hola!

I have a problem which is not strictly R, although R will be used for the
analysis.

We have data from a large investigation of drug abuse, initially
analyzed by logistic regression. But the pupils are selected by first
sampling schools, and as it happens the prevalence of use varies sharply
from school to school, so there is over-dispersion.

Now we are interested in comparing the prevalences estimated from this
study with prevalences from 3 earlier studies (93, 96, 99), which were
analysed assuming no overdispersion, withouyt investigating that issue.
For various reasons we can not get hold of the original databases from
those studies, only the published summary statistics. But we know the
structure of sampling were the same as in the 2002 study, so probaby the
overdispersion were the same (at least assuming that is the best I can
do).

So assuming equal overdispersion I can get four confidence interval s
of "binomial" p's, but people will want an hypothesis test of the overall
null: p_1=p_2=p_3=p_4. Is there some way I can construct an hypothesis test
for that null, from the four confidence intervals?

Hoping this is enough information of the background, as the mail is
already to long!

Thanks,

Kjetil Halvorsen