Dear Ronald, as far as I understand, the preference (dominance) of Fisher's test has a historical/computational background. It was quite cumbersome to calculate the probabilities for different margins by hand (huge number of tables) and, therefore, as long as the computers were not able to do this within a feasible time Fisher's test was the best what one could do. In the meanwhile it is no more necessary, but people get accustomed with this test. However, it is especially problematic if you are interested in shifted hypothesis (e.g. so called non-inferiority problems, where H_0: p_1>=p_2+delta should be tested). You can perorm the tests (and you get some p-values, which are, however not really informative, as it was stressed several times on this list), but as far as I know, there is no way to calculate "exact" confidence intervals. There is a paper from Yates (J. Royal. Statist. Soc. vol. A147 p.426 + comments) one of Habermann (JASA, vol. 53 p.555) commenting on conditioning. This topic is also shortly discussed by Agresti (categorical data analysis, wiley). Concerning the likelihood, your estimate of the odds ratio is NOT the usual n_11*n_22/n_12*n_21, which belongs to the unconditional estimate. The differences are thought not to be large... I hope it helps kind regards Robert -------------------------------------------------- Focus Clinical Drug Development GmbH, Neuss ================================================================= Instructions for joining and leaving this list and remarks about the problem of INAPPROPRIATE MESSAGES are available at http://jse.stat.ncsu.edu/ =================================================================