Standard Error of Kappa

The standard error(s) of the Kappa coefficient were obtained by Fleiss (1969). Different standard errors are required depending on whether the null hypothesis is that κ = 0, or is equal to some specified value.

We give below the appropriate standard error of the Kappa coefficient for testing whether Kappa = 0:

Algebraically speaking - SE (0) (κ)    =     √ 1 [pe2 + pe − ΣRiCi(Ri + Ci)/N3] (1 − pe)2N where: pe is the expected proportion that agree, Ri and Ci are the row and column totals for the contingency table, N is the number of observations.

Standard errors for where the null is not equal to zero, and for the weighted kappa coefficient can be found in Fleiss (1981)