Do polygraph tests qualify for the prize?

[rwguinn]

Re: Re: Re: Re: Re: A serious question...

Because arrogant people in this thread are telling me it can't be done.

snipped >>>

Would it be shocking; world-view altering, to find that dependable, responsible, shy people write somewhat differently (but significantly differenty) than undependable, irresponsible outgoing people? I think it could almost be predicted based on common sense. But again it's an empirical question.

Snipped>>>
Plus, if it did work, I think I could get a nice publication which helps toward tenure.

I wouldn't be surprised if the above were true. I would be surprised and amazed it one could tell the difference (by handwriting method alone--the way letters are formed, not the writing content) between "dependable, responsible, shy people" and undependable, irresponsible, shy people

or between "undependable, irresponsible outgoing people" and dependable, responsible outgoing people

 

[Jekyll]

Re: Re: Re: Re: Re: A serious question...

Because arrogant people in this thread are telling me it can't be done.

Plus, if it did work, I think I could get a nice publication which helps toward tenure.

Good for you for testing it, but seriously, do you see this going anywhere? Finding that neat people tend write precisely (but not necessarily legibly) is like finding that anorexics tend to be thin.

 

[kk2796]

Each was to play the role of a manager who was accused of cooking the books by top management. 4 of the subjects were told they were in fact innocent of the charge. The fifth one was told he was in fact guilty, but if he could fool the interrogators that he would receive 500$ (in real money!). Thus giving him real-world incentive to lie.

Okay, here's exactly why polygraph tests fail. They measure anxiety. The problem is - any intelligent, innocent party is going to feel just as anxious answering the million dollar question (i.e. "did ya do it?") as a guilty party.

The above experiment would impress me if it went (note emphasized addition):

Each was to play the role of a manager who was accused of cooking the books by top management. 4 of the subjects were told they were in fact innocent of the charge, and if they could convince the interrogators of this, they would receive 500$ (in real money!). The fifth one was told he was in fact guilty, but if he could fool the interrogators that he would receive 500$ (in real money!). Thus giving him real-world incentive to lie.

In the real world innocent suspects and guilty suspects are faced with the same reward/punishment stakes: an innocent party unable to convince the tester of his/her innocence might wind up in prsion.

Thus, the actual task put to polygraphs is to discern the anxiety of "being found guilty falsely" from the anxiety of "being found guilty correctly" (something the original test does NOT attempt). Add into this the idea that a criminal is more likely to know a thing or two about fooling polygraphs by taking measures to subvert their anxiety, and you've got the real reason not to put credibility into polygraph testing.

 

[kk2796]

And I now see the conversation has moved well away from polygraphs! Okay, the graphology conversation is quite interesting as well. Folks should be aware that the field of data mining is slowly evolving into what might best be described as a whole new science (or at least, a new way of approaching many old problems). If you have 101 attributes of an entity such that no one in their right mind would think the first 100 attributes would correlate in any way to the 101st attribute, data mining has this exotic tendency to scoff at your "right mindedness", use the 101 attributes for 10 test subjects to build a model, and use the model to predict the 101st attribute for 100 new subjects with remarkable accuracy, using only the aforementioned 100 test subjects.

Newcomers to the field tend to get very excited by this - they tend to think that by just finding enough measurable attributes, they can predict almost anything! However, data mining is not magic. If the 100 factors are truly indepent of the 101st factor, data mining will not "magically" find a correlation. Another problem data mining runs into is that of attrbiute interdependence - if bpesta22 takes 100 different measures of a paragraph of writing, he might find that 90 of them have strong inter-correlations, reducing his actual field of influential attributes to 10 (if everyone who has a short upstroke on k's also has less space between words, there aren't two fully useful attributes here).

My personal take on this subject: I'd be extremely surprised if bpesta22 were able to develop a method that consistently beat chance (p < .05) in predicting a single personality characteristic (from a field of dozens) based on handwriting. Such a finding might merit publication in many professional journals. Such a finding would no doubt be useful to forensics teams, who would *love* to put writings of complete strangers through such a system. Part of the reason I'd be so shocked is that this finding would be so valuable, there are certainly numerous attempts already made to find such hidden correlations.

So yes, it would be a surprising discovery. It would not, however, be supernatural, paranormal, or any other such thing. As with any correlation, enough research and analysis would determine natural, and perhaps obvious-in-hindsight, causal agents.

Consider: Suppose people who lead very busy lifestyles tended to
1) write in a hurry and thus leave less room between letters and
2) experience higher divorce rates

If the above example held (unbeknownst to the anyone ahead of time), and bpesta22's regression analysis of handwriting predicted divorce rates with better-than-chance accuracy, would he be eligible to win $1 million?

I hope not.

 

[bpesta22]

Re: Re: Re: Re: Re: Re: A serious question...

Good for you for testing it, but seriously, do you see this going anywhere? Finding that neat people tend write precisely (but not necessarily legibly) is like finding that anorexics tend to be thin.

Maybe it's arrogance on my part, but I think it would be a neat discovery-- a sort of in your face to the professional graphologists. Yeah, handwriting correlates with personality, but throw out your graphology books based on "intuition" and centuries of confirmation bias, and instead look at what data tell you (of course then finding that although some aspects of personality are predictable from handwriting, the effect is small, and using it to predict personality is kinda stupid given that we have valid personality tests that are much cheaper to administer).

So, it could be a step ahead for skepticism. Or, it could backfire. Or, most likely, I won't find anything significant.

Can you tell from my writing I'm a bit neurotic.

 

[KRAMER]

No.