Algorithm for Computerized Adaptive Test Based on Latent Rank Theory

Propositions of the Algorithm and its Validations by Simulation

Abstract

The purpose of this paper is to propose new algorithms for computer adaptive test based on latent rank theory and to evaluate them by Monte Carlo simulation.

Result of the evaluation, the estimation error of the LRT-CAT using the new algorithms was 1.7%, 2.8%, 5.5%, 8.9% in the case of item bank of 50 items, 40 items, 30 items, 20 items, respectively. This indicates that computer adaptive test based on the latent rank theory using the item bank constructed with a small sample data where a computer adaptive test based on item response theory can not be applied can perform on a practical level.