We have recently had more opportunities of examining the reliability of investigation and measurement in the studies of physical therapy, and subsequently have problems in the application. The purpose of the present study is to examine the characteristics of the four kinds of coefficient of reliability; Intraclass Correlation Coefficient (ICC), Cronbach's coefficient alpha (α coefficient), Cohen's coefficient of agreement (κ coefficient) and Kendall's coefficient of concordance (W coefficient). The purpose is also to clarify appropriate application of them by performing simulations. ICC can be widely applied, but depends on the variance of examiners, examinees, and repetition of measurements, thereby needing a description of SEM. Fleiss's κ coefficient is applied in the case of three and more examiners, but in comparison with normal κ coefficient, Fleiss's κ coefficient is used in the case of two examiners. It is desirable to use both W coefficient and κ coefficient for making up each disadvantage. The studies of physical therapy often deal with the data of degree or name measures, and κ coefficient and Kendall's coefficient of concordance are applied more often, compared with ICC. Each coefficient has both advantages and disadvantages, and it should be applied under the considerations of several precautions. The important things are to observe the data carefully and not to insist on the coefficient or significance itself.
View full abstract