topic Re: PROC LOGISTIC in Statistical Procedures

PROC LOGISTIC

robertrao — Wed, 08 Nov 2017 12:46:02 GMT

Hello,

I have values like shown below and each row corresponds to the values of a single hospital.

and I am trying to make a ROC chart using Proc Logistic. Unfortunately I get the statistics table but not the ROC curve

Please help me to figure where the mistake was made.

Thank you

HAVE

expectdMortality CASE-MIX-INDEX Hospital

2.91 2.04 1

2.57 2.03 2

1.60 1.46 3

ods graphics on;

proc logistic data=HAVE plots=roc;

model expectdMortality = CASE-MIX-INDEX;

run;

ods graphics off;

Re: PROC LOGISTIC

Ksharp — Wed, 08 Nov 2017 13:08:27 GMT

proc logistic data=sashelp.class plots=roc;
model sex=age weight height/outroc=roc ;
run;

Re: PROC LOGISTIC

Reeza — Wed, 08 Nov 2017 16:00:24 GMT

You can't do logistic regression with a continuous dependent variable, it should be binary or ordinal/nominal but definitely not continuous. You can do a linear regression if that's appropriate.

Re: PROC LOGISTIC

Rick_SAS — Wed, 08 Nov 2017 19:25:28 GMT

To follow-up Reeza's comment, your code looks like you might have data in events-trials syntax.

If your data include the variables NUM_DIED and NUM_TREATED then the observed mortality rate for each unit (doctor or hospital) is OBS_MORT_RATE = NUM_DIED / NUM_TREATED.

PROC LOGISTIC supports a syntax that looks very similar: you specify the number or events AND the number of trials on the left side of the MODEL statement, separating those variables by using the '/' symbol. Thus the syntax is

proc logistic plots=roc;

model NUM_DIED / NUM_TREATED.= case_mix_index;

run;