by CKR
A few years back, the National Academy of Sciences was asked to evaluate the usefulness of polygraph tests. They duly appointed a committee, which gathered the evidence and came to some conclusions. Those conclusions, basically, were that the studies that had been done on polygraphing were poor, so in some senses the jury was out, but what they could say was that for investigating individual incidents (like a specific instance of embezzlement in a bank), a polygraph could give results that were better than flipping a coin, but far from perfect.
For screening, like evaluating whether someone should be granted a security clearance, polygraph tests appear to be much less accurate.
The various agencies that had installed requirements for polygraph tests in response to someone's intuition or political pressures saw no reason not to continue what they had been doing.
Now the Washington Post confirms some of what's in that NAS report: different agencies get different results from their polygraphs. In a bureaucratic addendum, those agencies all insist that their results are the ones that should be followed. One of the instances appears to be harassment, something that has to be easy for a polygraph examiner to slip into or to justify.
The government wants to fix the problem, the article says, by streamlining the process and making it more uniform across the agencies, but the subjectivity inherent in polygraphing that the NAS reports makes any such thing unlikely. Besides, this is what bureaucracies always say when confronted with evidence of the sort in the WaPo article.
The article does mention that Aldrich Ames, who passed CIA information to the Soviets for years, managed to pass all his polygraph tests. The NAS report also considers countermeasures against polygraphs, which it finds can be effective.
The article doesn't mention the NAS report, though. It's fairly easy to find by googling (not the first in my search, but on the first page), and it seems to me that it's indispensable background to a story on polygraphing. Facts are always useful, along with the anecdotes.