07-23-2013 03:07 PM
I am looking at the documentation and some examples using exact logistic regression. The official SAS documentation has examples such as http://support.sas.com/rnd/app/da/new/daexactlogistic.html where the p-values match in all 4 cases (conditional score, conditional probability, parameter, and odds ratio). However, I have come across other examples, such as this one from UCLA (http://www.ats.ucla.edu/stat/sas/dae/exlogit.htm), where the p-values differ. The parameter and OR are of course the same, and in this case the score and probability are equal (though I understand how they could be different, being two different tests), but there is no match between parameter tests and conditional tests. For example, the variable FEMALE has score and probability conditional exact p-values of 0.3401; however, the parameter/OR p-value is 0.4557.
Which p-value should be reported? Is one set of p-values taking the second factor (apcalc) into account whereas the other is not? This would only be a true issue when the differing p-values resulted in one rejecting and the other failing to reject H0.