null model fit statistics using PROC LOGISTIC and FIRTH option

deleted_user — Wed, 30 Sep 2009 15:31:37 GMT

Dear all,

I want to get -2LogL of the null model using proc logistic. I thought there are two ways to do that.

1) Run the model without specifying any independent variable, i.e.

proc logistic data = mydata descending;model youtcome = /firth;run;

In the output, I found:

Model Convergence Status

Convergence criterion (GCONV=1E-8) satisfied.

-2 Log L = 38.858

2) Run the model by specifying independent variables, and use the model fit
statistics in the Intercept Only column. For example:

proc logistic data = mydata descending; model youtcome = weight/firth; run;

In the output, I found:
Model Convergence Status
Convergence criterion (GCONV=1E-8) satisfied.
Model Fit Statistics

Criterion Intercept Only Intercept and Covariates
AIC 31.570 31.805
SC 33.790 36.244
-2 Log L 29.570 27.805

Why the -2LogL in the Intercept Only column (29.570) is different from what I got from the first method (38.858)? And if I specify different independent variables, the model fit statistics in the Intercept Only column also vary. So what does "Intercept Only" mean? And which one is the correct null model fit statistics?

If I turn off the "Firth" option, the two null model fit statistics (-2logL) are the same. But the problem is that my data has quasi-complete problem and "Firth" should give me more reliable results.

Thanks!

koko

topic null model fit statistics using PROC LOGISTIC and FIRTH option in Statistical Procedures

null model fit statistics using PROC LOGISTIC and FIRTH option