A pragmatic approach to Rasch-modeling : The loss of information index

Abstract

Although attitude questionnaires only rarely satisfy the Rasch Model, the sum score is routinely used in measuring attitudes. This can lead to a considerable loss of diagnostically usable statistical information. As well in Latent Class Analysis, in the Mixed Rasch Model and/or in the Hybrid Model the use of the BIC which favours a smaller number of classes than the AIC, leads to a certain loss of information.Starting from a general test-theoretical model that includes not only Classical Test Theory but also Item Response Models, the present paper introduces indices for evaluating the relevance of the respective information loss.An application to the measurement of anti-Semitic attitudes shows that the rejection of the Rasch Model in favor of Latent Class Analysis does imply turning away from the concept of a quantitative attitude dimension. On the other side, however, this provides more accurate information on the structure and dynamics of the respective attitude. Precisely what Classical Test Theory neglects as measuring errors can in the given case contain crucial diagnostic information.