Home > Statistics Every Writer Should Know > The Stats Board > Discusssion
Can you perform a Kappa analysis with missing values
Message posted by chris on August 2, 2000 at 12:00 AM (ET)
I'm using SAS %agree macro to perform an inter-rater reliability analysis and get Kappa coefficients. However, I have a couple of raters with missing values. Does anybody know if Kappa can or cannot be computer when there are missing values? Thanks, friends.
READERS RESPOND:
(In chronological order. Most recent at the bottom.)
Re: Can you perform a Kappa analysis with missing values
Message posted by Doug Mahoney on August 14, 2000 at 12:00 AM (ET)
A little more information would be helpful. However, I'll run through a couple of scenarios that you maybe facing. (1) One rater responded, the other didn't. In this case, i'd impute a discordant result to be conservative. If you are not doing a weighted Kappa, any discordant pairing will do. If you are doing a weighted Kappa, things are a little more difficult. I'd use the conservative approach and impute the most discordant result for your table. That is, if the responses are from 1 to 4 and reviewer A reported a 2 and reviewer B was missing, impute a value of 4. (2) Both missing responses. This is the most difficult situation to handle. You could try something along the lines of the EM , but i've found that it is just easier to report the number of missing observations and use the complete pairs of data. At least than the reader will be able to evaluate the usefulness of the results.
Your $5 contribution helps cover part the $500 annual cost of keeping this site online.
