The 2016 Conference on Test Security will be held on October 18-20 in Cedar Rapids. Our Director of Research Innovation, Dr. Wim J. van der Linden, will be hosting a presentation on the topic of “Bayesian Detection of Cheating on Tests.”
What is Bayesian Detection and what are the advantages?
A Bayesian approach to the detection of cheating on tests has several advantages relative to classical statistical hypothesis testing. First, it is based on the correct probability distribution of the number of items on which the test taker has cheated given his observed number of aberrant responses. Second, whereas classical hypothesis testing only allows us to control its Type I error, the Bayesian approach does allow us to directly account for the incidence of cheating in the population of test takers. Third, the approach resolves the problem of whether or not to condition on the responses by the source in the detection of answer copying, which has plagued the literature since Frary et al. (1977). Fourth, it automatically accounts for the presence of estimation error in any of the parameters of the psychometric model (e.g., ability parameters). A natural Bayesian way of presenting evidence of cheating is through reporting of its posterior odds given the responses observed for the test taker. In this presentation we will show the odds for four different types of cheating: item pre-knowledge, item harvesting, answer copying, and fraudulent erasures on answer sheets. For each of these types of cheating, the odds can be calculated using a simple, extremely fast algorithm known as the Lord-Wingersky algorithm in test theory. The only difference exists in the parameters that need to be fed into the algorithm.
For more information about this presentation, please refer to the description of Presentation 23 in the event program. https://cete.ku.edu/2016-conference-test-security