Using the Binomial Model with Unequal P To Help Detect Cheating in Multiple-Choice Exams
Clunie Bryan Gregor, Harry J. Khamis · Journal of Geological Education · 1992
Where copying is suspected on multiple-choice exams, the significance (as corroborative evidence) of matching incorrect choices shared by the suspect and the supposed source needs careful evaluation. The binomial distribution evaluates the probability of a number of chance duplications only when the probability (p) of a particular incorrect choice being made is equal for all of the incorrect alternatives offered in all of the questions implicated, which is generally not the case. This difficulty can be circumvented by estimating, for each incorrect response shared by suspect and source, the probability for the whole class of matching the source's choice. The P-value, sometimes called the significance probability, is the probability that at least as many matching incorrect choices as are observed between suspect and source could have occurred by chance alone. A P-value of less than .05 is considered to be statistically significant evidence of cheating.