Testing the unidimensionality assumption of the Rasch model

摘要: Statistical tests especially designed to test the unidimensionality axiom of Rasch model are scarce. For two them, Martin-Lof (ML-test) and splitteritem-technique, an extensive power analysis has been carried out, showing clearly superiority ML-test. The disadvantage ML-test, however, is that its null distribution deviates strongly from asymptotic chi-square unless one huge samples. A new with degree freedom proposed. Its superior converges rapidly chi-square.