|
Abstract:
|
Score differencing, which is an examination of whether the performance of an examinee is similar over two sets of test questions, is one of the six categories of statistical methods for detecting test fraud (Wollack and Schoenig, 2018). Score differencing can be applied to detect several types of cheating such as cheating on unproctored tests, fraudulent erasures, and benefitting from item preknowledge (Sinharay & Jensen, 2018). The existing methods for score differencing (e.g., Fischer, 2003; Guo & Drasgow, 2010; Sinharay & Jensen, 2018) are frequentist methods with a few exceptions such as Wang, Liu, and Hambleton (2017) and van der Linden and Lewis (2015). A lack of Bayesian statistical methods for score differencing is surprising given that researchers such as Skorupski and Wainer (2017) have called for more research on the application of Bayesian statistical methods for detecting test fraud. This aim of this project is to develop several new Bayesian statistical methods for score differencing. The new methods will be based on posterior probabilities and Bayes factors (e.g., Kass & Raftery, 1995). The methods will be applied to operational data that involve actual test fraud.
|
