engleski

Quantitative treatment of performance-based items and the metric quality of composite test scores

The study presents an empirical demonstration of a model for quantitative treatment of performance-based test scores composed of multiple choice items. The model was proposed by Kulenović (2004) with a basic assumption that the choice of an incorrect option is not a random function of an implied score on an underlying construct or a latent variable measured. Once distracters are made so that their choice is dependent on scores on the measured construct, individual items potentially turn into better measures, and the composite scores necessarily become more valid estimates compared to those based on a binary model. The performance of the formulated model of differential weighting was tested and exemplified using several sets of empirical data on different forms of a nonverbal abstract reasoning test and a verbal ability-based emotional intelligence test. The results showed an improvement in performance of both items and derived composite scores when the proposed differential weighting model was used, even with the tests initially created under standard, binary-scoring paradigm.

composite test scores; differential weighting model

nije evidentirano