Evaluating Effectiveness of Standard Error of Score Estimation as a Computerized Adaptive Testing Termination Criterion

• In Event: 29.082 - Roundtable Session 9
  In Roundtable Session: 29.082-12 - Computer Adaptive Testing Designs and Analyses

Abstract

In computerized adaptive testing (CAT), it is not uncommon to use the standard error of θ estimation (SEE) value as a test termination criterion because test forms with the same level of SEE are expected to have the same measurement precision even if they differ in the number of items. This study evaluated the effectiveness of SEE as an approximation of actual conditional standard of measurement under various conditions with dramatically different test lengths and investigated its consequences in final score estimation in CAT. The finding of the study suggested that using the SEE as a sole criterion to terminate CAT might lead to less than ideal level of control in measurement precision when the test length is extremely short.

Authors

• Chansoon (Danielle) Lee, National Council of State Boards of Nursing

• Kyung (Chris) T. Han, The Graduate Management Admission Council

• Doyoung Kim, National Council of State Boards of Nursing