|Date: ||Wed, 4 Aug 1999 15:13:45 -0600|
|Sender: ||"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>|
|Subject: ||Re: Challenge|
|Content-type: ||text/plain; charset=US-ASCII|
Matthew Zack responded:
Kappa is usually calculated on the basis of categorical data. How
do you group your information (for example, V events;
duration/start or stop times) into categories? As the number of
categories increases, the theoretical maximum kappa decreases.
Does the ID code identify nursing home residents or raters? Does
the primary/rely code identify raters?
Our kappa calculations are based on the number of seconds of
agreement between the two raters for a specific key. So there
would be a kappa for every key involved.
The equation we use is a modification of the categorical one (I
believe) where kappa is the proportion of number of agreement and
nonagreement seconds divided by the total number of seconds of
the file. I'm not entirely clear where we found this modification, but
it is an acceptable form for the research we do.
In the sample dataset, I only included the code for
Primary/Rely. In the real dataset, we have the raters' initials.
Program Analyst, Sr.
Applied Gerontology Program
University of Alabama