Dear John,
as far as I am informed the (non-asymptotic) solution of this problem is
generally
based on some (unproved) proposals.
More details can be found in the following papers:
Santner &Snell: Small sample CIs for p1-p2 and p1/p2 cont. tables
JASA vol. 75 p. 368-394 (1980)
Chan: Exact tests of equivalence and efficacy ....
Statistics in Medicine vol. 17 p. 1403-1413 (1998)
J. Röhmel & U. MansmannConfidence Intervals for the difference of two
independent ...
(oral presentation on an IBS conference)
(you can get it from U. Mansmann: [log in to unmask])
Kind regards
Robert Németh
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
|