The Accuracy and Efficiency of a Reference-Based Adaptive Selection Algorithm for Comparative Judgment

Several studies have proven that comparative judgment (CJ) is a reliable and valid assessment method for a variety of competences, expert assessment, and peer assessment, and CJ is emerging as a possible approach to help maintain standards over time. For consecutive pairs of student works (representations) assessors are asked to judge which representation is better. It has been shown that random construction of pairs leads to very inefficient assessments, requiring a lot of pairwise comparisons to reach reliable results. Some adaptive selection algorithms using information from previous comparisons were proposed to increase the efficiency of CJ. These adaptive algorithms appear however to artificially inflate the reliability of CJ results through increasing the spread of the results. The current article proposes a new adaptive selection algorithm using a previously calibrated reference set. Using a reference set should eliminate the reliability inflation. In a real assessment, using reference sets of different reliability, and in a simulation study, it is proven that this adaptive selection algorithm is more efficient without reducing the accuracy of the results and without increasing the standard deviation of the assessment results. As a consequence, a reference-based adaptive selection algorithm produces high and correct reliability values in an efficient manner.