Μελέτη της συμφωνίας δύο ή περισσοτέρων αξιολογητών
Ζήση, Ελένη Γ.
In many sciences, as for example in medicine, psychology or in social sciences, raters (doctors, psychologists and others) often classify subjects according to a characteristic into I categories of classification. In order to study the agreement between two raters, one has to study consequently a two-way IxI contingency table. In this dissertation, the most known measures of agreement between two raters are reviewed, such as the kappa of Cohen, the intraclass kappa coefficient, and the tetrahoric correlation coefficient. We also discuss about some generalizations and extensions of kappa statistics. Furthermore, we are interested in modeling the agreement between two raters. So, the model of independence, the symmetry model, as well as the quasi-independence and the quasi-symmetry model are presented. Also, the agreement model of Agresti, that of Tanner and Young and a new log-linear model are described. When the subjects are classified by R raters, R>2, then a R-way contingency table is constructed and has to be studied. Measures and models of agreement are described also for this case. Complementary and explanatory in some points to the study of agreement among raters is the study and modeling of their disagreement. Some of the topics in the field of interater agreement, which have been in the center of research during the last years are the generalizations and extensions of models, the repeated measurements, the scales of classification, the size of sample, etc. Since the scales of classification are very important in the study of interater agreement, we briefly describe in the Appendix some of their characteristics and we present some scales of classification for clarifying related issues.