Civil Liberties Dilemma Medical Analogy Natl Acad of Sci Study

The value, and scientific validity, of polygraph testing is widely debated.   The Federation of American Scientists has an instructive summary:

Six prior reviews of field studies:
  • average accuracy ranged from 64 to 98 percent.
    • Ten individual field studies:
  • correct guilty detections ranged from 70.6 to 98.6 percent and averaged 86.3 percent;
  • correct innocent detections ranged from 12.5 to 94.1 percent and averaged 76 percent;
  • false positive rate (innocent persons found deceptive) ranged from 0 to 75 percent and averaged 19.1 percent; and
  • false negative rate (guilty persons found nondeceptive) ranged from 0 to 29.4 percent and averaged 10.2 percent.
    • Fourteen individual analog studies:
  • correct guilty detections ranged from 35.4 to 100 percent and averaged 63.7 percent;
  • correct innocent detections ranged from 32 to 91 percent and averaged 57.9 percent;
  • false positives ranged from 2 to 50.7 percent and averaged 14.1 percent; and false negatives ranged from 0 to 28.7 percent and averaged 10.4 percent.
    •  
    So there's some doubt as to the accuracy of polygraphs!   The National Academy of Sciences recently (Oct 2002) published a study which casts even more doubt.

    If you assume that polygraphs are 90% accurate, this means 10% of the time you test someone, they can lie and get away with it, or (even worse) they can tell the truth and be accused of lying.   And even if you re-test someone, one percent of the time they will pass both tests when lying or fail both when telling the truth.   Considering that thousands of polygraph tests are given each year, by the government and by industry, it means at least tens of people could be tested twice and be wrongly accused.   And "accused" often means the loss of a job or career!   In medicine, there are procedures to reduce the "false positive" rate, but they aren't used with polygraphs.

    And if re-testing is routine, with a 90% accuracy rate, 10% of the people will have two different test results — they will pass one, and fail one.   And such mixed results are often cause for suspicion by themselves.   It's also easy to show that an "accuracy rate" of 95% may actually be more like two percent.

    There is also considerable debate about the role of the polygraph examiner.   Apparently, drastically different results can be obtained, depending on who's doing the examining.   One article states:     A DOE security contractor who administered a 1998 polygraph test to Los Alamos scientist Wen Ho Lee found him to be innocent of committing espionage, providing classified information to an unauthorized person, or intending to harm the United States.   Unusually strong readings indicated an absence of deception.   The test was reviewed by two additional polygraphers, who concurred with the finding.   But then the Federal Bureau of Investigation (FBI) examined the very same data and concluded that Lee had failed the test. As you know, after being held in solitary confinement for nine months, and hearing Attorney General Janet Reno rant about his "serious crimes" and "danger to the U.S.", Wen Ho Lee was eventually released, with all but minor charges being dropped.   The Washington Post has a series of articles on the Wen Ho Lee case.   According to a 29 June 2001 article: FBI Assistant Director Neil J. Gallagher told a Senate committee in June 1999 that he had "full confidence" in the initial investigation by the Department of Energy.   But in a report sent to Congress yesterday [which was 28 June 2001], the GAO said Gallagher "had ample opportunity to know and should have known" about an FBI memo expressing "serious concerns" about the investigation. That document was sent from the FBI's Albuquerque field office to headquarters on Jan. 22, 1999 — more than a month before Lee was fired from Los Alamos.

    It is very common for polygraph examiners to state or imply that the subject under interrogation "has a problem", and then to ask if there is any information that can clear it up.   In effect, the subject is being interrogated, with no evidence to suggest they have committed any wrong, and without an attorney present.   So the classic constitutional protections against self-incrimination are ignored here.   The common excuse "if you're innocent you have nothing to hide" seems rather presumptuous — would you allow the police to search your home, car, or person without a warrant, even if you're innocent?

    Another question — does the 90% accuracy (or whatever the figure is) apply to the entire exam, or to each question?   If there are twenty or thirty questions asked, even a 99% accuracy rate on each question is unacceptably low.   If it applies to the entire exam, it would seem that long exams, with many questions, would have a lower accuracy rate than short exams.   Most of the tests of polygraph exams do not look into this issue, if any of them do.

    Also, most tests of polygraphs use innocuous questions — if a person lies and gets caught, there is no penalty.   If they lie and get away with it, they may win a small amount of money (a hundred dollars, for example).   If their lives or careers were on the line, results would probably be different.   Basically honest people would probably be more likely to be caught; pathological liars might find it more exciting to lie and win and thereby be more successful.

    I am personally aware of a corporate executive who had to pass a polygraph exam to obtain a security clearance.   The married executive had had a long-running affair with another executive at the same company.   This was common knowledge in that company.   But he had to answer "no" to questions about extramarital affairs in order to gain that particular level of clearance.   He was able to pass the exam.

    And he's not the only one.   Some of the most successful spies have lied and not been caught.   Aldrich Ames is one example.   He passed numerous polygraph exams without any problems.   A close investigation of his spending patterns (he spent far more than he could afford on his salary) would have suggested a problem.   After his conviction, he even wrote a letter about polygraph testing.
     
    The fundamental problem with the polygraph is the risk of accusing an innocent person.   If you have some evidence against an accused, and the polygraph further confirms he's lying, then go ahead and punish him (fire him, convinct him of spying, whatever) — that's the usual argument.   The problem is, what if the polygraph is wrong (and the studies above prove that it often is)?   If the evidence is already strong, why give a polygraph?   The evidence alone is enough.   If the evidence is weak, why give a polygraph?

    Those who advocate polygraph testing, by saying "it's not a panacea, but is a powerful tool in assisting law enforcement" (or some similar platitudes) — always assume that a polygraph can only confirm guilt.   What is you suspect someone, but he passes the polygraph?   Is he exonerated?   Usually not.   And are people given polygraphs only if there is already substantial evidence against them?   Again, no. The polygraph seems to be just another way to punish people.   If you have "suspicions" about someone — vague as they may be — you can give them a polygraph and hope they fail.   Shades of McCarthyism.   The only way the polygraph is useful in these situations, is to punish anyone who might be guilty.   If you don't care about punishing the innocent, polygraphs can be useful.

    The only way, it seems to me, that polygraph testing could be used — effectively and fairly — is under the follow conditions:
    1. No blanket polygraph testing.   If a crime has been committed, people are not polygraphed en masse.   Only if there is already evidence pointing to someone, that indicates they probably committed the crime, are they given a polygraph.   This also means people are never to be polygraphed just to qualify for a job.

    2. A polygraph exam, by itself, is never considered proof, or even evidence, of guilt.   If the first condition is followed this will never be an issue.   But if, somehow, a person is given a polygraph, and their questions indicate they might be guilty, this is never considered unless there is already evidence pointing to their guilt.

    3. A polygraph is never considered evidence of guilt unless there is already substantial evidence pointing to the person's guilt.   If may not be enough evidence for reasonable doubt, but should be at least a preponderance of evidence.   In the U.S. legal system, criminal convinctions cannot be obtained unless evidence shows the accused is guilty beyond a reasonable doubt.   This usually means 90% or 99% certainty.   In civil cases, a preponderance of evidence (usually meaning slightly more than 50% certainty) is sufficient.

      A preponderance of evidence is usually enough to take away a person's job (in accusations of theft, or unauthorized release of company secrets), or a security clearance.   The polygraph could be used to add an element of certainty.   If the person is about to be fired, give him a polygraph and confirm that he's guilty.

      Of course, if he passes the polygraph, then what?   Or, if he refuses to take the exam?   The only way people could be encouraged to take polygraph exams, if they have been accused and the evidence suggests they are guilty (but it's a preponderance, not beyond a reasonable doubt), is if they have something to gain.   In that case,

    4. Use the polygraph to exonerate someone.   When the evidence is otherwise against him, and he takes a polygraph, and he passes, he's exonerated.
    Given all these caveats, it's hard to see how the polygraph can be used at all. Of course, those who advocate the polygraph are often those who sell their services as examiners — certainly a conflict of interest.   And polygraph advocates usually don't support scientific studies of polygraph trustworthiness.   They confidently quote statistics like "it's 90% accurate."   Even if that were true, that's not good enough.
    REFERENCES
    The Testing Dilemma ("95%" accuracy is really much lower)
    The Lie Behind the Lie Detector
    FAS Polygraph Testing Info

    Antipolygraph
    Letter from Aldrich Ames on Polygraph Testing (FAS)
    Steven Aftergood on Polygraph Testing (Science, 3 Nov 2000)

    The Truth About the Polygraph Salon 3 Mar 2000
    Beating the Polygraph Salon 3 Mar 2000
    Scientific Validity of Polygraph Testing A Review (FAS) Nov 1983

    Wen Ho Lee Case (Washington Post)
    FBI Official Misled Congress About Lee Case, Says GAO

    Top Civil Liberties Dilemma Medical Analogy NAS Study