ABSTRACT
This chapter aims to provide logical and analytic evidence to encourage Bayesian reasoning for cheating detection, regardless of the method employed. It argues that these frequentist inferences are often misleading, and that this is especially true with regard to establishing the probability of very-low-incidence events, such as cheating. The chapter considers the traditional frequentist p-value is really the wrong P. It suggests that most cheating detecting investigations are not really interested in the probability of observing data, given the null "not a cheater" condition; rather, the inference of interest is the probability that a condition has been met, given what has been observed in the data. This is precisely what the Bayesian paradigm provides. The benefits of calculating the Posterior Probability of Cheating, as opposed to relying on traditional p-values, is demonstrated by means of a series of analytic examples. These analytic examples demonstrate the problem with null hypothesis testing and "statistical significance" for very low-incidence events.
