ABSTRACT
The earliest efforts for detecting preknowledge involve the use of person-fit statistics. A person-fit statistic detects item score patterns that are unlikely given the item response theory model that is employed. The effects of item compromise are probably the most severe in instances when an actual exam file along with the key is stolen. D. Maynes introduced a methodology which makes use of a set of miskeyed, unscored items called Trojan horse items. P. Obregon conducted a simulation study that illustrated the promise of using the log odds ratio index to identify items that may have been compromised. L. McLeod introduced the FLOR log odds ratio index, and J. Smith and S. Davis-Becker used differential person functioning statistics for person flagging. Detection methods have been developed in response to the emerging threats to identify potentially compromised items and individuals who may have benefitted from pre-knowledge.
