Different Somers'D values from Freq and Logistic

ftsr — Sat, 24 Jan 2009 01:41:57 GMT

I was running SAS on a set of 10 observations each given a rating between 1-3, and another variable A, B or C is assigned to each of them.

e.g.
Obs F1 F2
Ob1 A 1
Ob2 A 3
Ob3 B 3
Ob4 C 2
Ob5 C 3
Ob6 B 1
Ob7 A 3
Ob8 B 3

I used the following to obtain Somers'D
"Proc freq data=xxx;
tables F1*F2 / measures;
run;"

But it comes out differently from the Somers'D in the output of the logistic function
Proc logistic data=xxx;
class F1;
model F1=F2 / rsquare stb;
run;

Then I tried "model F2=F1", the Somers'D still doesn't agree with Freq.

Could anyone help me to understand this please?

Re: Different Somers'D values from Freq and Logistic

Doc_Duke — Sun, 25 Jan 2009 15:25:01 GMT

Based on the documentation, it looks like FREQ applies a correction for ties and LOGISTIC does not. Your example has at least one tie. Try new sample data without ties.

Doc Muhlbaier
Duke

topic Re: Different Somers'D values from Freq and Logistic in SAS Procedures

Different Somers'D values from Freq and Logistic

Re: Different Somers'D values from Freq and Logistic