can you compare two kappa estimates in proc freq?

[pau13rown]

i have two subgroups in my dataset, and i estimated kappa and obtained confidence intervals for each as per this code:

proc freq data=.... order=data ;
  tables var1*var2 / agree noprint;
  weight frequency / zeros;
run;

is it possible to test whether the agreement (as measured by kappa) is the same within these subgroups ie Ho: kappa1 = kappa2?

[Rick_SAS · Accepted solution]

I assume you have a categorical variable (gender, race, political affiliation) that you want to use to see whether the kappa for one group (females) is equivalent to the kappa for the other group (males).

I'm not an expert on this topic, but run the following example to see if it gives you some helpful information. You can put the categorical variable FIRST on the TABLES statement to get analyses that control for each strata. You also get an analysis for the overall kappa:

data Have;
input Sex $ var1 $ var2 $ Frequency ;
datalines;
F A X 20
F A Y 30
F B X 20
F B Y 20
M A X 40
M A Y 60
M B X 20
M B Y 10
;

proc freq data=Have order=data ;
  tables sex*var1*var2 / agree ;
  weight frequency / zeros;
  test kappa;
run;

See the PROC FREQ doc, especially the sections "Overall Kappa Coefficient" and "Tests for Equal Kappa Coefficients."

[pau13rown]

thanks!