topic Re: P-values from Firth’s Penalized Likelihood in Statistical Procedures

P-values from Firth’s Penalized Likelihood

Miracle — Sat, 15 Dec 2012 10:54:37 GMT

Dear all SAS users,

I'm wondering if there's a way to get the p-values for the parameter estimates and odds ratio estimates and as well as the profile-likelihood confidence intervals produced from proc logistic with Firth's estimation?

Any input is greatly appreciated. Thanks in advance.

Re: P-values from Firth’s Penalized Likelihood

Reeza — Sat, 15 Dec 2012 19:41:43 GMT

Isn't it included in the standard output with the firth option?

proc logistic data=have;

model y=x1 x2 x3/firth;

run;

Re: P-values from Firth’s Penalized Likelihood

Miracle — Sun, 16 Dec 2012 06:05:40 GMT

Hi Reeza. Thanks for your reply.

Is "Pr > ChiSq" under "Analysis of Maximum Likelihood Estimates" the one?

I was reading Logistic Regression Using SAS: Theory and Application, Second Edition by D Allison Paul. On page 58 it says "we should always specify CLPARM=PL option when using Firth method to deal with quasi-complete separation due to inaccurate conventional Wald confidence interval in such situations. Wald p-values can also be inaccurate for the problematic parameters, so inference should be focused instead on the confidence intervals." On page 59, it says "You should never trust Wald chi-squares for parameters with non-existent maximum likelihood estimates".

So I'm wondering if I can report the p-values under "Analysis of Maximum Likelihood Estimates".

Thanks in advance.