ISSN : 1229-067X
A person-fit analysis (PFA) has been developed to identify individuals whose latent traits are not measured accurately by a test. This study investigated the distribution and effectiveness of a person-fit index: the standardized log-likelihood index (lz) for Graded Response Model(GRM). Findings in simulation studies employing various manipulated variables demonstrated that the empirical distribution of the lz were close to the standard normal distribution and the lz was effective in detecting individuals showing aberrant response patterns given the IRT model (i.e., graded response models). The application of the lz to the empirical data of the Rosenberg Self-Esteem scale (1965, 1979) illustrated that GRM was suitable to the scale, producing relatively few individuals of a significantly large value.