Comparison of On-the-Fly MST With Preassembled MST on PISA Data

• In Event: AERA Poster Session 2
  In Poster Session: Poster Session 1 (San Diego)

Abstract

Recently, Multistage Adaptive Testing (MST) has been adopted by numerous large-scale testing/assessment programs including GRE and PISA. Most current MST methods were developed as extensions of Luecht and Nungester (1998), in which the test booklets are optimized at a few fixed trait values (e.g., -1, 0, and 1), instead of any point in the continuum of the latent trait. As a result, MST may yield more biased estimators in comparison with computerized adaptive testing (CAT). The objective of the study is to incorporate an on-the-fly mechanism into MST to provide some unique advantages over both CAT and MST. A pilot simulation study based on PISA data demonstrated that the new method outperformed MST regarding estimation accuracy and item pool usage.

Authors

• Xiuxiu Tang, Purdue University

• Tong Wu, Indiana University - Purdue University

• Yi Zheng, Arizona State University

• Kit-Tai Hau, Chinese University of Hong Kong

• Hua-Hua Chang, Purdue University