The Providence Journal
Detecting lies in lie detectors
by Howard Fienberg
July 11, 2001

The parents of missing Federal Bureau of Prisons intern Chandra Levy do not believe that U.S. Rep. Gary Condit, D-Calif., has revealed all he knows of her whereabouts. Although police investigators are satisfied with Condit's answers, the Levys want him to take a polygraph (popularly known as a lie-detector) test.

Polygraphs are such familiar instruments, whether in criminal investigations or job-background checks, that few question the reliability or validity of polygraph testing. But how well do these devices really detect lies?

The U.S. Department of Energy administered it to American physicist Wen Ho Lee when first accusing him of leaking nuclear secrets to the Chinese. He passed the test, but the FBI subsequently reviewed the same test results and declared Lee a liar. Convicted spy Aldrich Ames passed numerous polygraph examinations over the years while passing CIA secrets to Russia.

The American Polygraph Association last year asserted that 12 studies demonstrated an average accuracy of 98 percent. On the other hand, University of Minnesota psychologist William Iacono accuses polygraph examiners of "wishful thinking." He said last year that "almost no published peer-reviewed scientific papers exist that bear out the accuracy of current polygraph techniques." That may be why the American Medical Association and the American Psychological Association contend that polygraphs yield little more than a 50/50 chance of success.

Data are hard to come by. The false positive rate could be anywhere from 1 percent to 20 percent. The National Research Council recently convened a panel on polygraphs to try to settle the matter, but its report is not expected until the end of this year at best.

So what is the precise problem with polygraph testing? According to critics, it lies not in the hardware but in the interpretation of the test results. Using a polygraph as a lie detector involves studying a subject's physical response to specific questions. The series of questions asked usually starts with queries unrelated to whatever matter is being investigated to establish a baseline of physical reading, including blood pressure, sweat-gland activity and breathing. Then other questions more directly address the areas of suspicion.

But interpreting the results of the exam is a subjective task and confounding factors are numerous. A fair number of fearful but innocent subjects could produce false positives and psychologically savvy or unstable guilty subjects could yield false negatives. And there is a burgeoning industry of books and Web pages promising to teach how to fool polygraphs.

With an error rate that defies calculation, polygraph tests are much more useful as tools of intimidation than as instruments of truth. Experts agree that, faults aside, the tests can persuade guilty parties to confess who would otherwise have remained stoic. But few polygraph examiners speculate on what effect this intimidation has on the innocent.

I can sympathize with the Levys in their demands. Their daughter is missing and there is little concrete evidence of what happened. But the emotional pleas of parents must not be used to supersede the inherent weaknesses of lie-detection devices. The old "string 'em up"-style lynch mob has given way to a modern "hook 'em up" mob of polygraph proponents. Unfortunately, neither is very good at uncovering the truth.

Howard Fienberg is research analyst with the Statistical Assessment Service (STATS), a think tank in Washington, D.C., aimed at improving public understanding of scientific, statistical and social research.

[also printed in the Houston Chronicle]

return to Howard Fienberg's page