I am confused about a simple point. Are the 95% CI's of a log odds ratios only symmetric when computed from a logistic regression? What if I got an odds ratio from a fishers exact test and converted it into a log odds ratio, does that guarantee symmetry? Sorry for the silly question.
5
-
1$\begingroup$ It depends: exactly what confidence interval procedure are you using?? $\endgroup$– whuber ♦Commented May 15, 2023 at 21:15
-
$\begingroup$ Just a two-sided, fishers exact test for a 95% CI. I was told if I converted my odds ratio and CI's into log space, that this would be symmetric but found that wasn't the case. $\endgroup$– WorkhorseCommented May 15, 2023 at 21:20
-
$\begingroup$ That still doesn't pin it down. See the abstract at ncbi.nlm.nih.gov/pmc/articles/PMC2852239 for an explanation of why the particular CI procedure matters. $\endgroup$– whuber ♦Commented May 15, 2023 at 21:21
-
$\begingroup$ Yeah not sure I follow. But my test is pretty significant but maybe im not grasping the abstract well enough. $\endgroup$– WorkhorseCommented May 15, 2023 at 21:49
-
1$\begingroup$ I would not anticipate symmetric intervals in general. An interval for the log of an odds ratio based on the hypergeometric will not in general be symmetric (in either of the usual senses). An asymptotic interval for log-odds might be symmetric depending on how its obtained. $\endgroup$– Glen_bCommented May 16, 2023 at 1:44
Add a comment
|