Enhance Classification Accuracy in Computerized Adaptive Testing From the Test Development Perspective

• In Event: Poster Session 14
  In Poster Session: Conceptual and Technical Studies on Assessment Design and Development Topics in K-12 Education

Abstract

For most assessment programs using computer adaptive testing, it is important to provide not only accurate ability estimates, but also correct classification decisions. Classification errors happen most often to the “threshold” examinees – whose thetas/abilities are close to the cut scores. This study aims to illustrate the possible improvement in ability estimation and classification accuracy by anchoring the items around the cut scores during the item development stage. The match between the ability distribution and the item pool was also a factor examined. The preliminary results suggest that conditions with anchored item parameter around the cut scores tend to yield somewhat better classification accuracy, compared with the condition when the ability and item parameter distributions are both from N~(0,1).

Authors

• Che-Ming Allen Lau, Pearson

• Jie Lin, Pearson