The Impact of Statistical Constraints on Classification Accuracy for Multistage Tests

• In Event: Roundtable Session 18
  In Roundtable Session: Factors Influencing the Accuracy of Multistage Testing Results

Abstract

This study investigated the impact of statistical constraints on classification accuracy for the multistage testing (MST) approach. In the context of the licensure and certification exam, the various levels of the test information functions, cutoff scores, and the population’s ability distributions were considered using a 1-2-2 MST design and simulation components. Automated test assemblies were performed using a linear programming model to ensure the desired test information functions. The results indicated that a more than 50% drop in the test information function decreased the correct classification rates by more than 2%. The interaction between cutoff scores and population’s ability distributions were also examined and discussed.

Authors

• Ryoungsun Park, Wayne State University

• Jaehoon Seol, American Nurses Credentialing Center

• Jiseon Kim, University of Washington - Seattle

• Barbara G. Dodd, The University of Texas - Austin