Tuesday, May 3, 2016

P is for Polygraph

Polygraphs, also referred to as lie detectors, seem to be part of a social experiment that never ends. Though many of us have heard of polygraphs (especially given their widespread presence in old detective shows), it was not until recently that I looked into the experimental design behind the polygraphs, and the highly debated believability of their results.
First, a little background on polygraphs… The polygraph was invented in 1921 by John Augustus Larson, a 21 year-old physiology student at University of California, Berkeley who moonlighted as a police officer. The device, which Larson referred to as the cardio-pneumo psychogram, could measure blood pressure, pulse, respiration, and skin conductivity, all with the premise that these are measurable indicators that change when the subject of the tests is being deceptive.
Like any good experiment, polygraphs use controls – control questions, that is. A series of control questions, usually broad statements meant to inquire into the subject’s past character and truthfulness, are asked by the administer of the test, and then a series of relevant questions concerning the case at hand are asked (a repeated-measures design, if you will). If a subject is being deceptive, or perhaps rather, is being anxious or nervous, the polygraph should record a change in the physiological measures. Someone who is being deceptive concerning the case should theoretically be more anxious when asked to answer relevant questions, while someone who is not being deceptive concerning the case should be more anxious when answering control questions.
Though the polygraph has been regarded as one of the greatest inventions (in fact, the original polygraph device constructed by Larson is housed in the Smithsonian), its validity has been debated since its invention. Larson conducted many tests on his own, compiling a list of cases where his device helped solve murders, thefts, etc. (interestingly, Larson first tested his device on his wife…). However, many scientists today regard the basis of polygraphs as pseudoscience. Former US Attorney General John Ashcroft estimated the false-positive (or type I error, if you will) rate of polygraphs to be 15 percent. Yet, the Journal of General Pyschology published an analysis of 41 criminal cases where polygraph tests were used, concluding an accuracy over 90 percent.
If the test seems to be so accurate, then why the ongoing debate (as with all science, right)? Well, studies are conflicting in their results of accuracy, which could be a result of “p-hacking”, where researchers throw out inconclusive polygraph results to improve the accuracy rate of polygraphs. One also has to question the experimental design of the polygraph, where neither a subject or administer is blinded. One could potentially skew the polygraph results from either perspective by causing undue nervousness or anxiety during questioning, or by being aware of how the measured indicators change and resisting (common spy trick). The best summary of the flaws of the polygraph test come from William Iacono, a psychophysiologist at the University of Minnesota: "A big problem is that it's not really a test of anything,” highlighting the fact that little is really known about how the body behaves when lying, and therefore these physical measurements may not actually be measuring marks of deception.
Needless to say, given all of the debate, I would not want to be on the wrong side of a polygraph in a courtroom.
1.     “Telling the Truth About Lie Detectors.” USA Today. http://usatoday30.usatoday.com/news/nation/2002-09-09-lie_x.htm
2.     “The Truth About Lie Detectors.” American Physiological Association. http://www.apa.org/research/action/polygraph.aspx

1 comment:

  1. I appreciate how you tied the history of the polygraph to statistics. It is very interesting how the original polygraph that Larson developed did not change too much in the sense that current lie detectors still measure blood pressure, respiration, skin conductivity and pulse. Its use remains largely controversial when it comes to court cases, however I personally think it is too "orwellian" and that results should not be presented as conclusive forensic evidence since our understanding of the physiological events that transpire when lying are limited. As you mentioned the type 1 error is relatively high, and the inherent bias of the test is not rigorous enough to consider the results valuable in the court setting.