Kappa Mult Variables on Stata 14

[Elijah Sard]

Hello,

I am using Stata 14 and conducting a research project in which I'm trying to assess the kappa between two raters on multiple different variables as well as an over-all kappa (I'm not sure if "pooled" is the correct terminology).

The data is as follows: 425 patients; each with varying numbers of Chest X rays (anywhere from 1-7 representing consecutive days); on each chest x ray TWO different raters have assessed several variables including: is there an abnormality, yes/no; is this consistent with pneumonia, yes/no; is the infiltrate upper lobe, 1,2,3,4.

I have searched within stata help and various articles online and have found one method for finding an overall kappa that involves converting all kappa's using the fischer's z transformation, then converting back. I found this somewhat confusing and was hoping for another suggestion or some clarification.

Thank you so much.
Eli

[Nick Cox]

I'd check out Daniel Klein's kappaetc (SSC) to see if it does this while noting that "fischer" here should be "Fisher". I've seen this mutant twice in 2 hours on two different forums.

http://www.statalist.org/forums/foru...lable-from-ssc

[daniel klein]

While kappaetc does not handle this problem directly (as the literature I used does not discuss it), here is a suggestion that pools the observed and expected agreement across variables instead of pooling the individual kappa statistics.

The z-transformation can be done with the tanh() function. Among other programs, this is also implemented in Nick's corrci (SJ). The help file describes the procedure and the code is open source, as usual. Applying this to kappa statistics is probably trivial. The much more important question is how your dataset looks like. kap, kappa (and kappaetc, as well as other related commands) expect the data to be set up in a given way. Usually observations are expected to correspond to subjects (patients) while variables record ratings.

Best
Daniel