ABSTRACT
Using statistical analysis to detect test score irregularities is an important component of a broader test security policy intended to support the accurate measurement of examinees' performance and ensure that test results are meaningful and valid. Statistical methods that focus on irregularities at the level of total test scores rather than the level of individual items often utilize longitudinal data to detect unexpected changes over time. Two categories of statistical methods are item-level methods, and score-level methods. In any description of statistical test fraud detection measures implemented in a statewide assessment program, it is important to emphasize that statistical analyses should be situated in a broader test security policy. The chapter assesses two statistical approaches for conducting pass rate analyses: the two-proportion z-score and multilevel logistic regression. The two-proportion z-score approach examines differences in pass rates across two time points for a group of students relative to aggregate pass rate differences across those same time points for all students.
