[Factor analysis, a measure of interrater reliability in single cases. Application to the AMDP Psychopathology Scale]

The reliability of a measure is a prerequisite for its validity. The measurement of the interrater reliability (IRR) of the 115-item AMDP System, filled out by 68 raters from several Centers for a single videotaped patient, raises two unusual aspects in IRR. First of all, a matrix of correlation coefficients between raters for such a number of individuals and items does not give a global visualization of differences in rating patterns. The authors apply the Q-technique of factor analysis (FA) or inverse FA in order to solve this problem. Second, the usual measures of IRR do not apply to single cases. In testing the adequacy of FA multivariate indexes as a measure of IRR (factor scores, Mahalanobis distances), the authors demonstrate a bias consisting of the influence of the mean item score on these indexes and they put forward an original modification of the classical FA program. They report a simulation of scores and the real patient CRO in order to demonstrate the adequacy of their modified FA as a measure of IRR in single cases.