>
Dangerous for people taking the test or administering the test? Your comment seems ambivalent about the efficacy of tests.Dangerous to the organizations that rely upon the test, in that lack of the belief in the efficacy of polygraphs decreases their usefulness as a screening tool.
>But is it accurate when used on people that have extensive training about how to beat the test?
No, its not particularly useful for any query if one has specific training in how to control one's physiological responses. The Soviets used general coaching, and I believe these modern firms are using biofeedback.
>So why would telling people that it doesn't work be dangerous if it is "pretty accurate for specific questions?"
The polygraph measures physiological indicators such as skin conductivity, heart rate, respiration, and muscle tension that are metrics of stress, stress being considered a proxy for deception.
The greatest effect of the polygraph coaching is that it reduces the subject's belief in the efficacy of the test, thereby reducing the stress response during the testing, which in turn makes untruthful responses appear truthful.
>Good investigators have other skills use other things such as body language, speech patterns (e.g sudden shifts in tense when describing events), and facial micro-expressions (using systems such as FACS). So a good investigator wouldn't need a polygraph test.
All of those are useful techniques, as is voice stress analysis which is supplanting polygraph in corporate security, but all can be fooled.
The biggest added advantage of all "objective" deception tests is that they increase stress on the subject, causing more indications of deception to leak through the facade. When people believe the tests aren't very effective, their efficacy decreases.
The future is in functional MRI technology. Memory and fabrication are two very different mental tasks, and I would imagine the difference should be detectable with current technology, although I'm certainly no expert.
The two firms I've seen in the space are Cephos and No Lie MRI, which are claiming ~97% accuracy (a useless number I realize, but the best I can find from their public information.)