Exact Bayesian Inference for Assessing the Accuracy of Polygraph Testing
The polygraph test is often used in law enforcement as an investigative tool, and in the courtroom where it is used to bolster support for the innocence of a defendant. The amount of available data to evaluate polygraph accuracy that is taken under realistic conditions is severely limited. With a fully Bayesian approach we analyze the largest such data set that exists. We derive exact results for the posterior distribution of the negative and positive predictive values, which can be evaluated with a computer algebra system. We show that the uncertainties, even given the largest and most realistic data set currently available, are great, casting doubt on the use of polygraph testing in criminal trials.
KeywordsPositive predictive value Negative predictive value Sensitivity Specificity Dirichlet distribution
The author would like to thank the editor and the reviewer for their careful review of the manuscript. Their comments have certainly led to a better paper.
- Kelly Jack (2004) The truth about the lie detector. Am Herit Invent Technol 19(3):14–21Google Scholar
- Krapohl DJ (1996) A taxonomy of polygraph countermeasures. Polygraph 25(1):35–56Google Scholar
- Wolfram Research Inc., (2010) Mathematica, Version 11.0, Champaign, ILGoogle Scholar
- Zelicoff Alan, Rigdon Steven E (2017) Bayesian inference for interpretation of polygraph results in the courtroom. Law Probab Risk 16(2–3):91–109Google Scholar