Polygraph Testing - Science or BS?

[Eos of the Eons]

I can see how a good polygraph operator can ignore the polygraph altogether, and just use the questions to trip up someone into confessing. Any contradictions can be brought up, causing the person being questioned some grief that might show up on the polygraph, etc.

As for inventing a good lie detector machine without having a good operator to gauge answers? Bad idea, relying on a stupid machine for anything definitive.

 

[drkitten]

As for inventing a good lie detector machine without having a good operator to gauge answers? Bad idea, relying on a stupid machine for anything definitive.

I'm not sure I agree here. Part of the motivation behind the development of the lie detector is the same as the motivation for a lot of police science -- to take the (biased) human as far out of the loop as practical and to substitute objective evidence. Measurements of blood pressure and respiration rate are about as objective as they get, compared with statements like "well, he looked shifty to me" or the somewhat more sophisticated "his denials lacked credibility."

There's a lot of case law on the draft, conscientious objection, and exactly how much leeway the local draft board has in evaluating the "sincerity" or "credibility" of an applicant for CO status, and one of the problems -- one of the reasons local draft boards get slapped down by the actual court system -- is because many draft boards had a tendency to disregard statements that they didn't themselves believe, in many cases simply saying "the applicant lied." The current standard is, frankly, a mess -- but one thing that is NOT acceptable is a simple blanket denial on the grounds that the applicant was not credible enough.

There have been similar and equally severe problems with biased security officers (anyone of THAT TYPE is automatically untrustworthy, where "that type" could be a homosexual, a minority, an atheist, or simply someone who votes the wrong way); the examples of both Oppenheimer and Turing should be cautionary. In theory, a lie detector should be able to produce a collection of evidence that all skilled operators agree upon how to interpret, in the same way that radiologists agree on what a particular X-ray means.

We're not there yet. Indeed, we may not ever be there, although (pace ID) we're at a point where we're substantially better than chance. I think a polygraph is a classic example of a technology that has been rushed into widespread use before it's mature enough to be used, and I think it's unfortuante that the flaws in the technology have such serious consequences.

But, frankly, I'd rather have a good, objective lie detector than none at all. I'd simply rather have none at all than a bad one,.... which is what we have now.

 

[Eos of the Eons]

Yeah, I'd prefer to have none at all. In my view, the polygraph is a prop. If used as a prop, then you can use contradictions against someone. If there are no contradictions, then maybe the person is just telling the truth.

 

[Valen]

Previous removal of polygraphs in the military

I'm really surprised (and disappointed). I know that Prof Ray Hyman was involved in a review of polygraphs in the military and that their results (polygraphs don't reveal lies) had removed the use in most parts of the military about 10-15 years ago. I'd had hoped it would have squashed that issue.

The scary part is the belief that using it clears spies and terrorists.

 

[zigaretten]

Somewhat related:

http://www.iht.com/articles/2008/09/15/asia/15brainscan.php

"India's use of brain scans in courts dismays critics"

"The new technology is, to its critics, Orwellian. Others view it as a silver bullet against terrorism that could render waterboarding and other harsh interrogation methods obsolete. Some scientists predict the end of lying as we know it."

"....But the technique received its strongest official endorsement, forensic investigators here say, on June 12, when a judge convicted a woman of murder based on evidence that included polygraph and BEOS tests.

The woman, Aditi Sharma, was accused of killing her former fiancé, Udit Bharati. They were living in Pune when Sharma met another man and eloped with him to Delhi. Later Sharma returned to Pune and, according to prosecutors, asked Bharati to meet her at a McDonald's. She was accused of poisoning him with arsenic-laced food."

"After placing 32 electrodes on Sharma's head, investigators said, they read aloud their version of events, speaking in the first person ("I bought arsenic;" "I met Udit at McDonald's"), along with neutral statements like "The sky is blue," which help the software distinguish memories from normal cognition.

For an hour, Sharma said nothing. But the relevant nooks of her brain where memories are thought to be stored buzzed when the crime was recounted...."

What I'm getting from this article is that this technology is neither proven nor developed enough yet, but that it could be a real lie detection method in the future. But this is all very new to me, so.........

 

[JoeyDonuts]

I'm really surprised (and disappointed). I know that Prof Ray Hyman was involved in a review of polygraphs in the military and that their results (polygraphs don't reveal lies) had removed the use in most parts of the military about 10-15 years ago. I'd had hoped it would have squashed that issue.

The scary part is the belief that using it clears spies and terrorists.

No sir. Still very much alive and well as far as the Navy is concerned. Anyone who has any hope of getting a sensitive posting will have to go through it.

I wonder what percentage of NCIS' operations and funding concern polygraphs? Might explain why they are so steadfastly using them in a screening capacity.

When all else fails, look to money as a motivator above all else.

 

[ponderingturtle]

Most of all, if they say you have failed it (which I hope for your sake doesn't happen) DO NOT ADMIT TO ANYTHING. Do not fill out a statement, do not carry on any more conversation with the administrator other than "I would like to speak to an attorney." Your situation may be different, and God I hope so. But if you get even the slightest feeling that something isn't right, TERMINATE THE SESSION IMMEDIATELY.

I fail to see how this could possibly help you get the job. Keep a job sure but it seems likely that any thing odd like this would prevent you getting the job.

 

[ponderingturtle]

I can see how a good polygraph operator can ignore the polygraph altogether, and just use the questions to trip up someone into confessing. Any contradictions can be brought up, causing the person being questioned some grief that might show up on the polygraph, etc.

As for inventing a good lie detector machine without having a good operator to gauge answers? Bad idea, relying on a stupid machine for anything definitive.

That would depend on how good the machine was. Personaly I would be more likely to trust a machine than something that required a person to make value judgements. Bais is enherint in the system with a person in it, and that can lower the reliability considerably.

 

[ponderingturtle]

Yeah, I'd prefer to have none at all. In my view, the polygraph is a prop. If used as a prop, then you can use contradictions against someone. If there are no contradictions, then maybe the person is just telling the truth.

Are you kidding look at his sifty eyes, I know he is lying, he is just good at it, that is why his story has not contradictions.

 

[I Ratant]

That would depend on how good the machine was. Personaly I would be more likely to trust a machine than something that required a person to make value judgements. Bais is enherint in the system with a person in it, and that can lower the reliability considerably.

.
"...how good the machine was"?????
It's a paper recorder, that puts ink on paper, from pens that are driven by instrument amplifiers which put out a signal based on some changes in physiological responses.
Whether any of these responses have anything at all to do with fact or fiction has never been established.
The inventor promoted the idea that because -he- had measureable responses to lies, everyone then does respond in the identical manner.
(But how important, to the inventor, were those lies? That's where the problem is.)
Much like the Rorshachs, if you see what Rorshach saw in those smears, you're as "normal" as Rorshach... and how normal was he?
The examiner makes value judgements based on the responses seen on the paper, and how the subject handles himself during the interview.
fMRI may, that is, may be a more reliable indicator of fact/fiction but the process hasn't been refined sufficiently. After all, it requires genuine liars to provide the data to be judged as relevant.
When the examiner runs the calibration question past you.. "Is today the fumth of Nielk?", you are -instructed to lie about it.
Big deal!

 

[I Ratant]

Are you kidding look at his sifty eyes, I know he is lying, he is just good at it, that is why his story has not contradictions.

.
The interviews are videotaped, so personal value judgements can be used to reinforce or contradict the paper results.
The examiner sits just outside your peripheral vision, with the recorder humming away while he makes notes on the paper.

 

[drkitten]

.
"...how good the machine was"?????
It's a paper recorder, that puts ink on paper, from pens that are driven by instrument amplifiers which put out a signal based on some changes in physiological responses.
Whether any of these responses have anything at all to do with fact or fiction has never been established.

Yet another over-the-top, easily-refutable rant.

Yes, it has been established that these physiological responses have something to do with fact or fiction, which is why the false-positive rate seems to average around one in seven, instead of one in two. But one in seven is still better than 80% accuracy, and Randi will run the MDC with applicants who claim less than 80% accuracy.

The problem isn't that lie detectors don't work -- it's that they don't work well. At least, not well enough for the purposes to which The Man wants to put them.

 

[Sherman Bay]

JoeyDonuts, with much practice, maybe you can control the polygraph output like this (near the end). Imagine the examiner's surprise!

http://www.youtube.com/watch?v=CnL0GiznkyM

 

[George Maschke]

Yet another over-the-top, easily-refutable rant.

Yes, it has been established that these physiological responses have something to do with fact or fiction, which is why the false-positive rate seems to average around one in seven, instead of one in two. But one in seven is still better than 80% accuracy, and Randi will run the MDC with applicants who claim less than 80% accuracy.

The problem isn't that lie detectors don't work -- it's that they don't work well. At least, not well enough for the purposes to which The Man wants to put them.

Actually, polygraph testing has not been proven through peer-reviewed research to reliably differentiate between truth and deception at better-than-chance levels under field conditions. A statistical analysis by Dr. Alan Zelicoff of the polygraph community's best field studies suggests that an innocent percent taking a polygaph "test" has only about a 50% chance of passing. See, "Positive and Negative Predictive Values of Polygraphs: Results from published 'field' studies":

https://antipolygraph.org/articles/article-051.pdf

But it gets worse: polygraph "testing" has no grounding in the scientific method. It lacks standardization and meaningful control. For this reason, like the art of cold reading, its sensitivity and specificity cannot truly be established. For more in this regard, see Dr. William G. Iacono's article, "Forensic 'Lie Detection': Procedures Without Scientific Basis":

https://antipolygraph.org/articles/article-018.shtml

 

[marting]

Yet another over-the-top, easily-refutable rant.

Yes, it has been established that these physiological responses have something to do with fact or fiction, which is why the false-positive rate seems to average around one in seven, instead of one in two. But one in seven is still better than 80% accuracy, and Randi will run the MDC with applicants who claim less than 80% accuracy.

The problem isn't that lie detectors don't work -- it's that they don't work well. At least, not well enough for the purposes to which The Man wants to put them.

Bad comparison. The MDC can only be run with an 80% accuracy when a test can be repeated and is stastically independent enough to meet the MDC's level of proof. Usually these tests are completely independent. Repeated polygraphs tend to be highly correlated and, unfortunately, can't be significantly improved by repetition.

 

[drkitten]

Actually, polygraph testing has not been proven through peer-reviewed research to reliably differentiate between truth and deception at better-than-chance levels under field conditions.

Ah, yes, the old "where are the randomized double-blind studies of parachutes" argument.

Have polygraphs been proven through peer-reviewed research to reliably differentiate between truth and deception in laboratory conditions? Yes, quite well.

Is it practical to do research under field conditions? Not at all -- you can't get the control group, since you don't have known ground truths.

Have other fields of science faced this problem? Yes, of course. Do we have methods available that we can adapt from, say, medicine to polygraph testing? Of course.

It's statements like the one you made above that make me completely dismiss antipolygraph.org as a useful source of information, I'm afraid.

 

[drkitten]

Bad comparison. The MDC can only be run with an 80% accuracy when a test can be repeated and is stastically independent enough to meet the MDC's level of proof. Usually these tests are completely independent. Repeated polygraphs tend to be highly correlated and, unfortunately, can't be significantly improved by repetition.

The hell they can't. Give me fifty polygraphers and fifty subjects -- we'll split the million fifty ways and each walk out $20,000 richer. That's several months salary for an afternoon's work.

Of course, Randi probably wouldn't approve a protocol that complex -- and justifiably so. I'm not seriously suggesting an MDC. My point is simply that a false positive rate of 14%, while appallingly bad in practical application, is in fact extremely good from a statistical standpoint. And, frankly, wouldn't even be that bad as an investigative tool. If there are eight suspects, and I can cut the likely field down to two (one of them guilty, the other a false positive), I've just saved myself three quarters of the effort in investigation.

 

[balrog666]

If the determination of pass/fail was based on the polygraph output, then the examiner could be in the next room or a thousand miles away.

And subsequent reviewers would be able to reach the same conclusion(s) at some rate significantly above chance.

Seems simple enough to test that.

 

[marting]

The hell they can't. Give me fifty polygraphers and fifty subjects -- we'll split the million fifty ways and each walk out $20,000 richer. That's several months salary for an afternoon's work.

Of course, Randi probably wouldn't approve a protocol that complex -- and justifiably so. I'm not seriously suggesting an MDC. My point is simply that a false positive rate of 14%, while appallingly bad in practical application, is in fact extremely good from a statistical standpoint. And, frankly, wouldn't even be that bad as an investigative tool. If there are eight suspects, and I can cut the likely field down to two (one of them guilty, the other a false positive), I've just saved myself three quarters of the effort in investigation.

If the question is polygraphy woo and not can a polygraph determine truth for a given subject to MDC levels then I agree. However, the wide range of "accuracy" reported by poorly controlled tests means we don't really know the real accuracy rate is for any given polygrapher and population of subjects. So we might need more subjects/polygraphers for the MDC. For forensic applications, the concealed knowledge test can actually put bounds on the accuracy but the generally used probable lie test is questionable, particularly for fishing (screening) applications.

Polygraphy isn't woo and I don't claim it is. But it is extremely poorly researched with abysmal controls. And the practical (lack of) results are demonstated by the 60 mins. program. Further, much of the success of polygraphy depends on people believing it works (hence the 98% numbers polygraphers claim) and being unaware of how it is scored. Most exams assume the innocent subject can be made to focus attention on a comparison question while actually believing it to be a relevant question they have to lie to. This is incresingly questionable given wide availability of accurate information. And then there is the question of the impact of countermeasures on accuracy, most probably reducing it.

The NAS report I liked earlier discusses this in great detail.

 

[I Ratant]

For verification of the basic concept; machines detect falsehoods to a verifiable and worthwhile percentage, 99% of the time, say, how does one get reliable sources of genuine "my ass is grass if they catch me lying" liars" to use to test the limits of the machine?
The current "standard" of 1 or so errors in 7 isn't what one would accept in whether one might get flattened by a bus when standing on a street corner. Were the odds that poor, everyone would stay home!
Almost all the subjects of polys given in national defense ARE perfectly good reliable patriotic citizens, and, we feel quite dirty after being tested, because we've put our livelihoods on the line for woo masquerading as real science.