Editor’s note: This article was initially published in The Daily Gazette, Swarthmore’s online, daily newspaper founded in Fall 1996. As of Fall 2018, the DG has merged with The Phoenix. See the about page to read more about the DG.
Stephen E. Fienberg, professor of Statistics and Social Science at Carnegie Mellon University, presented a lecture titled “In Search of the Magic Lasso: The Truth about the Polygraph,” yesterday in Science 101. The lecture addressed his collaboration with the National Research Council (NRC), and their conduction of a thorough study of the accuracy of polygraph testing. Inspired by the controversy over the alleged espionage by Wen Ho Lee, a nuclear scientist at the Department of Energy’s Los Alamos National Laboratory, Stephen and the Council presented Congress with a report of the insufficiency of polygraph graph testing in regards to accurately pinpointing spies in government agencies.
As the audience snacked on refreshments before the lecture, they viewed scenes from “Le Polygraphe”, a French-Canadian film that, according to Fienberg, accurately represents the polygraph experience. A woman was recently murdered, and, as a suspect, her boyfriend, the protagonist, was given a polygraph test. He repeatedly claimed his innocence; however, the examiner insisted he had failed. Soon he began to doubt his own memories, a common characteristic of those in his situation, which often leads to a false confession.
It was a situation all too familiar to Wen Ho Lee, a nuclear scientist picked up at the airport in Albuquerque, New Mexico, accused of passing secrets to the Chinese, and forced to take a polygraph test. Despite the ambiguity of the results, Lee was forced to resign from the Los Alamos National Laboratory because of his “unauthorized exit with a computer.” It became apparent to Congress that perhaps it was the polygraph test itself, and not Wen Ho Lee, that should be discharged, resulting in their request for further investigation.
The Council examined thousands of papers and reports, finally narrowing down the pool to fifty-nine “quality” studies. Using different types of statistical analyses, the Council concluded that the accuracy level of the polygraph varies. Through the use of different ROC Curves, which concentrate on sensitivity and specificity, they were able to see that specific incident polygraph tests were able to detect deception at a rate higher than chance (0.5) but lower than perfect (1.0). They suggested that for this reason, polygraphs were not ideal for spy screenings. Despite the Council’s findings, the Department of Energy responded by declaring they would not change their policy and would continue to polygraph everyone.
Fienberg ended his lecture on the promising note that there are several alternatives to the polygraph, though they are “not ready for prime time.”
A more detailed description of Fienberg’s article is accessible in Statistical Science” or online at Project Euclid. Fienberg also mentions that if one is having difficulty in accessing the article, to e-mail him at fienberg@stat.cmu.edu.