How accurate are lie detectors?

Lie-detector tests aren't completely worthless. How's that for an endorsement?

The polygraph, the most common lie-detection instrument, works on the assumption that the body reacts involuntarily to the stress of lying. It measures reactions such as changes in skin conductance, pulse rate, blood pressure, and breathing while the subject is asked a series of questions. The questioning process can take several forms. One early version was the "relevant-irrelevant" technique, which mixed queries like "Did you murder [name of victim]?" in with stuff like "Is today Tuesday?" Lies in response to the relevant questions would supposedly make the needles jump. The problem with this approach was that in such a context even an unfounded accusatory question could be stressful, producing a false positive.

The "comparison question" technique tries to get around this problem by making all the queries accusatory. In a sex-crime investigation, for instance, a suspect might be asked embarrassing control questions such as "Have you ever committed a sexual act you were ashamed of?" along with questions pertaining more directly to the case. The idea, which has a certain devious ingenuity, is that the innocent will show a greater response to the control questions (either because they're lying or simply flustered), whereas the guilty will show a greater response to the pertinent questions (which for them are more consequential).

The "guilty knowledge" testing method tries to discover whether a subject is privy to inside info about a case — things that only someone involved would know about. For example, suspects might be shown assorted photos of guns to see how they respond to the one that happens to show the murder weapon.

Besides investigation of crimes and the like, the other big use for polygraphs is general screening by employers looking to weed out iffy job applicants or catch workers in otherwise undetected wrongdoing. Pre-employment screening is common in law enforcement: one study found nearly two-thirds of agencies administered polygraph exams to applicants and rejected about 25 percent based on polygraph results alone.

Do the tests work? Depends how you define work. Probably the most comprehensive look at polygraph accuracy is a 2003 report from the National Academy of Sciences. After examining 57 polygraph studies the NAS concluded: "In populations of examinees such as those represented in the polygraph research literature, untrained in countermeasures, specific-incident polygraph tests can discriminate lying from truth telling at rates well above chance, though well below perfection." Their analysis of the 30 most recent polygraph data sets showed an overall accuracy of 85 percent, and an analysis of seven field studies involving specific incidents showed a median accuracy of 89 percent.

For screening purposes, though, the NAS found polygraph tests had too high a margin of error to be genuinely informative. If you made your criteria loose enough to catch most of the bad guys, you were overwhelmed with false positives; if you raised the bar enough to thin out the false positives, you missed too many bad guys.

And what about those countermeasures the NAS mentioned? Yeah, that's a problem too. Because polygraph tests rely on physical reactions, if you can control or mask your reactions at key moments in the questioning, you may be able to throw off the readings enough to produce an inconclusive result. Countermeasure techniques are surprisingly simple: they include discreet physical motions like pressing your toes against the floor or biting your tongue and mental tasks like silently counting backwards from 1,000 by sevens. The goal is to increase your baseline stress level enough to hide any revealing spikes.

If polygraphs are so fallible, why use them at all? In part because testing can intimidate people into confessing, deter bad behavior, and create an impression (however misleading) of vigilance. In other words: security theater. Heeding the NAS report, in 2006 the U.S. Department of Energy stopped blanket screening of its existing and prospective employees. Polygraph tests are now saved for specified instances — say, if someone fails to report a relationship with a foreign power.

Advocates of lie-detector tests foresee the day when technological advances will improve accuracy to the point where test results could be admitted as evidence. Much attention has been paid in recent years to functional magnetic resonance imaging, or fMRI. In its simplest form, fMRI lie detection works by scanning your brain to find out which areas are most active while you're being grilled; supposedly lying and truth-telling cause different areas to light up. No doubt due to the impracticality of using multimillion-dollar machines for everyday criminal interrogations, there haven't been many large-scale studies of fMRI accuracy. But the ones I've found show an accuracy rate of 76 to 92 percent — to be generous, about the same as you get with old-fashioned equipment at a fraction of the cost.

Cecil Adams @'The Straight Dope'