Estimating equations for kappa statistics.

Kappa statistics are often used for measuring agreement or association. The original measure suggested by Cohen for use with nominal data has since been generalized in a number of ways, including the incorporation of weights, adaptation for use with more than one rater and the application to bivariate data. Each modification requires a new approximation to the standard error and these have usually been derived by the delta method. This paper illustrates the use of estimating equations for kappa which offer a unified approach to many previously considered situations and which is also applicable to more complex designs. Copyright 2001 John Wiley & Sons, Ltd.