Janes C L
J Nerv Ment Dis. 1979 Jun;167(6):343-7. doi: 10.1097/00005053-197906000-00003.
This paper reviews the rationale for correcting for chance agreement beween two raters. It is suggested that under certain conditions correction for chance agreement is both unnecessary and inappropriate. When a chance correction is indicated, the chosen measure should be one that can be shown to be logically in accord with the two judges' method of determining each case's disposition. Two agreement measures, kappa (K) and Maxwell's Random Error Coefficient of Agreement (RE), are described in common terms and are then compared in terms of their assumptions about the judgment process. The plausibility of K's assumptions are challenged, and K's use is discouraged on the grounds that judgment table data are not treated in a way compatible with the process by which the table was constructed during the judgment process itself. Use of Maxwell's RE is recommended under conditions of equal proportions in the disagreement cells.
本文回顾了对两位评估者之间的偶然一致性进行校正的基本原理。研究表明，在某些情况下，对偶然一致性进行校正既无必要也不合适。当需要进行偶然校正时，所选用的指标应能在逻辑上与两位评判者确定每个案例处置方式的方法相一致。文中以通用术语描述了两种一致性指标，即卡帕系数（K）和麦克斯韦随机误差一致性系数（RE），然后根据它们对判断过程的假设进行了比较。对K的假设的合理性提出了质疑，并因判断表数据的处理方式与判断过程中构建该表的过程不兼容而不鼓励使用K。在不一致单元格比例相等的情况下，建议使用麦克斯韦的RE。
