Programming the statistical procedures from SAS

PHREG with frailty

Reply
Regular Learner
Posts: 1

PHREG with frailty

I am running PHREG models with frailty (SAS 9.4), and I have two questions concerning them: 1. how do I get more model fit metrics and 2. how do I get it to work for stepwise regressions. 

 

 

1. My PHREG model with frailty works fine with this code:

 

ods select ParameterEstimates fitstatistics NObs;

proc phreg data=dataname;

class id;

model time*event(0) =

        many_variables

        / risklimits TIES = EFRON details;

random selfid;

run;

 

However, it only gives me one model fit score that I find little documentation on -- marginal log likelihood. (i) Is there a way to tweak the code to get more model fit output so as to say one model is better than another? (ii) And what is a non-technical description of the marginal log likelihood? I assume it's much like a -2 log likelihood... 

 

 

2. I have many variables and would like to run a stepwise search for significant variables. Unfortunately, the stepwise process does not run when I include a random statement. Is there a way to tweak the code so as to get it to work and not lose the random statement?

 

I am running it like this...

 

ods select ParameterEstimates fitstatistics NObs;

proc phreg data=dataname;

class id;

model time*event(0) =

     many_variables

     / selection=stepwise slentry=0.25 slstay=0.15 details;

random id;

run;

 

The outout is only the final model with ALL my variables listed, and then these notes:

NOTE: The SELECTION= option (MODEL statement) is ignored for a frailty model analysis.

NOTE: SLENTRY= is ignored since SELECTION=NONE.

NOTE: SLSTAY= is ignored since SELECTION=NONE.

NOTE: DETAILS is ignored since SELECTION=NONE.

 

(i) So how do I tweak the code so I can run stepwise and keep in the random statement, and (ii) is there a way to do it as a forward and backward stepwise process?  

 

 

Thanks for any help you can offer!

-DM

 

Ask a Question
Discussion stats
  • 0 replies
  • 110 views
  • 0 likes
  • 1 in conversation