How to fit a set of models with all possible variable combinations

[Andrew Xavier]

Hi all,

I have the following code

webuse nhefs, clear
stset age_lung_cancer, fail(lung_cancer)
stcox former_smoker smoker male urban1 rural
estat ic

I want to find the model with the smallest IC.

I want to fit all possible models that can be fitted with 1 variable at a time, 2 variables at a time, 3 variables at a time and so on.

How do I do this? This seems a simple thing to do but I'm struggling to figure this out.

Thanks

Andre

[Nick Cox]

ssc install tuples

will allow you to go

tuples smoker male urban1 rural

and given those 4 variable names, you get 15 tuples (= 2^4 - 1) as accessible local macros, allowing you to go

forval j = 1/15  {

   
 stcox former_smoker `tuple`j''  estat ic  } 

where you may want to go further. If you loop

forval j = 0/15 { 

} 

then you invoke tuple0 which does not exist (= is empty), but that's fine, and gives you a way to include the benchmark case of no covariates.

You may want the IC of concern to be put somewhere convenient each time around the loop.

[Andrew Xavier]

Thank you a lot for your help!

This is the final code

webuse nhefs, clear
stset age_lung_cancer, fail(lung_cancer)
stcox i.former_smoker i.smoker i.male urban1 rural interview_age
estat ic

tuples i.former_smoker i.smoker i.male urban1 rural interview_age, nopython display 

forval j = 1/63 { 
    stcox former_smoker `tuple`j''  
    estat ic
}