topic Re: I have two variables actual and test, i want ROC curve for test variable ? in Statistical Procedures

I have two variables actual and test, i want ROC curve for test variable ?

Dinesh_ — Fri, 27 Oct 2017 07:12:39 GMT

ods graphics on;
proc logistic data=change desc plots(only)=roc;
   class actual;
   units age=10;
    model test(event="Yes")= actual age/outroc=x scale=none clparm=wald clodds=pl rsquare;
run;
quit;
ods graphics off;

/*I have written this code, is it a correct approach?*/

Re: I have two variables actual and test, i want ROC curve for test variable ?

StatDave — Fri, 27 Oct 2017 14:06:36 GMT

The actual response (which I assume is binary like your test variable) should be the response variable in the MODEL, not the test variable. Note that without age, your test and actual variables would just produce a 2x2 table if they are both binary. As such, an ROC curve would only have one point on it, so it wouldn't be useful. With age in the model, you'll get as many points on the ROC curve as age levels, but the curve evaluates your test and age variables together, not just the test variable alone.

Note you don't need the DESCENDING option if you are using EVENT=.

Re: I have two variables actual and test, i want ROC curve for test variable ?

Dinesh_ — Mon, 30 Oct 2017 03:52:55 GMT

Thank You