Erasure analyses in educational accountability testing programs usually investigate potential tampering by teachers or administrators on the base of between-group differences on the prevalence of wrong-to-right (WTR) erasures. Typically, this is done without controlling for the impact of legitimate covariates, such as examinee ability. Methodologies that do exist focus on person level detection and do not appear to provide a probabilistic statement that a test has been tamped. In this study, we extend the IRT-based detecting methods by Wollack, Cohen, and Eckerly (2013) and van der Linden and Jeon (2012) to a multilevel analysis. We include the use of covariates as explanatory variables. These methods are compared and demonstrated using a data from a large-scale high-stakes test.
