This paper considers some appropriate and inappropriate uses of coefficient kappa and alternative kappa-like statistics. Discussion is restricted to the descriptive characteristics of these statistics for measuring agreement with categorical data in studies of reliability and validity. Special consideration is given to assumptions about whether marginals are fixed a priori, or free to vary. In reliability studies, when marginals are fixed, coefficient kappa is found to be appropriate. When either or both of the marginals are free to vary, however, it is suggested that the "chance" term in kappa be replaced by 1/ n, where n is the number of categories. In validity studies, we suggest considering whether one wants an index of improvement beyond "chance" or beyond the best a priori strategy employing base rates. In the former case, considerations are similar to those in reliability studies with the marginals for the criterion measure considered as fixed. In the latter case, it is suggested that the largest marginal proportion for the criterion measure be used in place of the "chance" term in kappa. Similarities and differences among these statistics are discussed and illustrated with synthetic data.
Stored only when both papers are in the corpus — 1,575 citations exist upstream.
Imported from OpenAlex (CC0) — nothing here is generated.