09-10-2014 08:14 PM
I'm doing a conditional logistic regression on a categorical variable with 3 levels (0, 1 and 2). The highest level vs the reference level gave me a p-value of 0.0285 which is significant but the Odds ratio estimates is (0.295, 1.064) that covers 1.
Any idea why this is the case?
p.s. this is a case/control study with 195 matched pair.
Thanks a lot!
03-28-2015 05:28 AM
IMO the way the confidence interval is estimated depends on the procedure and options invoked, but in some cases the size of the confidence interval may depend somewhat on how many other exclusively different levels there are in that categorical effect, whereas the significance of a particular level does not (so much). If my understanding is correct, then the significant p-value evidence is the stronger of the two.
E.g. the CI around a trinomial proportion estimate is about 1.33 times that of an otherwise equivalent binomial estimate. That scale factor increases to about 1.66 for 4 outcomes or higher.
Or...maybe you inadvertently performed a one-sided test for the level p-vaalue? Is that possible?
Hope that helps.