Behavior and interpretation of the κ statistic: Resolution of the two paradoxes

摘要: Two apparent paradoxes have been identified for the kappa (kappa) statistic: (1) high levels of observer agreement with low values; (2) lack predictability changes in changing marginals. The first paradox is a function prevalence trait sample, while second related to symmetry observations disagreement categories. While examining behavior as distribution responses contingency table, it was discovered that any measured level (Po) there are three characteristic values kappa: max, min, and nor, each which only Po. allow an observed (kappa o) be placed into perspective. By observing categories, readily understood predictable. We define expressions (SA) (SD) order represent quantify these effects. Kappa alone has little interpretive value we recommend studies reporting also report Po, SD, P++ (agreement on presence trait).