ddSimilar: a joint statistical cluster analysis of test collusion

• In Event: Electronic Board Session IV

Abstract

To identify test collusion, previous studies are either model-based or machine learning-based. Model-based methods require model
assumptions, and all those methods may not work well when the test data are small. We proposed ddSimilar, which uses a
Bayesian non-parametric prior to enhance the similarity index to detect group-level test collusions.

Authors

• Luyao Peng, UNIVERSITY OF CALIFORNIA, RIVE

• Sandip Sinharay, EDUCATIONAL TESTING SERVICE