Using Person-Fit Statistics to Detect Cheating Due to Item Preknowledge

• In Event: 65.079 - Roundtable Session 23
  In Roundtable Session: 65.079-18 - Methods of Detecting Response Collusion and Preknowledge

Abstract

In K-12 state assessment, cheating may take place when teachers share previously administered items with students, or when students post items on the social media website. In both cases, exposed items may be known to the test agency, but the extent to which a test is compromised is uncertain. Given this scenario, the effectiveness of two person-fit statistics, a residual-based statistic (Z_C) and a likelihood-based statistic (l_Z), to detect cheaters were investigated through a simulation study in two stages. In stage 1, true item parameters were known. In stage 2, item parameters were calibrated. Preliminary results on Z_C suggest it has high power with known parameters, but its effectiveness may be limited by bias in estimated item parameters.

Authors

• Xi Wang, University of Massachusetts Amherst

• Feifei Li, Educational Testing Service

• Lixiong Gu, ETS