Thanks Todd! This is great. @Ben van Buren - Cohen's Kappa is used in many academic articles but it did not originate there. It's actually from the Cohen & Cohen book from 1960. I'm using a more recent version, the citation is: Cohen, J., Cohen, P., West, S. G., & Aiken, L. S. (2013). Applied multiple regression/correlation analysis for the behavioral sciences. Routledge.
Very informative. However, what do you do when: a) the Pe is 1 (and then the equation is divisible by 1)? Assume the Kappa is 1? b) have a very low Kappa when the raters agree on all but one of the ratings? Surely it should be higher? I have 2 raters,, 20 subjects. If they agree on 19, and differ on 1, the Kappa is nearly 0.
I have a very strange Kappa result: I have checked for a certain behavior in footage of animals, which I have assessed twice. For 28 animals, I have agreed 27 times that the behavior is present and have disagreed only once (the behavior was present in the first assessment, but not in the second). My data is organized as the following matrix: 0 1 0 27 And that gives me a Kappa value of zero which I find very strange because in only 1 of 28 assessments I disagree. How come it is considered these results as pure chance?
Hi. What about calculating sample size for Kappa? Do you think it is problematic to set the null hypothesis at K=0.0? I believe this would be the same at what others call to set K1=0.0, when many state the K1 should be the minimum K expected. Thanks
Dear Dr. Grande, i have maybe a simpe question. But Reseacher and RA are people, which give their responses to the survey f.a.,right?! So, this number can be very high then. And I´ve got 5 criteria like satisfactory etc . But I think i´ve understood how to do this. I should probably split people , giving responses into groups, in order to come up with the coefficient.
Dr. Grande: what would you do if the Kappa agreement turns to be too low? Should both coders recode the material in order to match and increase the value? Or what do you suggest? Thanks in advance.
It depends on the matter. In my field it is necessary, when the mean value turn out too low, a discussion is held to talk about troubleshooting. And then re-code the material.