topic Penalized likelihood estimation in Statistical Procedures

Penalized likelihood estimation

MMadi — Thu, 15 Sep 2016 10:27:55 GMT

I found that Firth’s penalized likelihood approach can be used insted of binary logistic regression in the prediction . However, I couldn’t find it in SAS university addition

So could you kindly please tell me how can I find it in this SAS addition

thanks

Re: Penalized likelihood estimation

Rick_SAS — Thu, 15 Sep 2016 12:14:28 GMT

SAS procedures have hundreds of options. Only the most common options are surfaced through the tasks in SAS Studio and SAS UE. However, the tasks merely generate SAS code (shown in the Code tab) and you can always add options to the generated code. In this case:

1. Consult the PROC LOGISTIC documentation to learn that the FIRTH option is specified on the MODEL statement.

2. Use the Binary Logistic Regression task to set up the model, but don't run it yet.

3. Click on the Code tab and click the Edit button.

4. The code will be copied to a new tab called something like Program 2. You can edit this program. Click at the end of the MODEL statement and type FIRTH before the semicolon.

5. Click the "running man" icon to run the SAS code.

In the program output, you should verify that the FIRTH method was used by looking at the Model Information table. It will have a row that says: Likelihood penalty: Firth's bias correction

Re: Penalized likelihood estimation

JacobSimonsen — Thu, 15 Sep 2016 13:12:43 GMT

Using Firths penalized likelihood instead of the ordinary likelihood is an option in the model statement in proc logistic. It is still binary logistic regression so it is not right to say that you instead use Firths likelihood. Here is a simple example


data mydata;
  input n x exposure;
  cards;
100 10 0
100 5 1
;
run;
proc logistic data=mydata;
  class exposure(ref="0")/param=glm;
  model x/n=exposure/firth;
run;