Ein pragmatischer Ansatz in der statistischen Theorie psychologischer Testscores

Abstract

The present paper discusses the applicability of C1assical Test Theory and of the Rasch-Model under different paradigmas of test administration. As is demonstrated, the confidence interval provided by Classical Test Theory for the true-score, is highly inaccurate in most cases. While application of the Rasch-Model allows for the construction of a superior confidence interval in these cases, there exist other cases, however, where the RaschModel does not apply, while Classical Test Theory produces reasonable results. The same holds for the statistical significance of score-differences. If tests are (as usually) defined as fixed selections of test items, however, then both models overestimate the score differences. Comparison of true-scores of different Ss on the same test should, therefore, be based on the well known Test of McNemar, instead.