« Not quite Coase | Main | Sex and the single woman »

Bayesian lie detectors

Here is a new way to get at the truth:

Drazen Prelec, a psychologist at the Massachusetts Institute of Technology in Cambridge, US, has devised a scoring system, or “Bayesian truth serum” to encourage people to divulge their honest opinions...

Prelec says if people truly hold a particular opinion, they tend to give higher estimates that other people share it. So if someone did have more than 20 recent sexual partners - but lied about it - that person would probably assume a higher rate of such behaviour in general than someone who had not had so many partners...

For example, he describes a situation where two paintings are viewed by a group of 10 people who are then asked, privately, to pick their favourite. Seven people say they prefer painting A, while three vote for painting B. If, on the second question, all 10 people said they thought everyone else would prefer painting A, then those three people expressing a personal preference for painting B might be thought of as a safer bet for having told the truth. That is because, argues Prelec, despite what they thought was more popular, those individuals still chose the other painting.

In other words, thinking that others are disagreeing with you predicts truthtelling. Prelec claims that the formula works best on groups of ten or greater.

I can, of course, think of caveats. The group must be your peers rather than hated rivals; conformity often brings social benefits but not always. But nonetheless I expect there is a grain of truth here.

Do you recall the old saying? "The surest way to get good information from a person is to ask for advice."

Here is the full story.

Posted by Tyler Cowen on October 26, 2004 at 07:10 AM in Science | Permalink

TrackBack

TrackBack URL for this entry:
http://www.typepad.com/services/trackback/6a00d8341c66b253ef00e55083a3ff8833

Listed below are links to weblogs that reference Bayesian lie detectors:

» A Bayesian Truth Serum for Subjective Data from Mahalanobis
A closer look at the scoring function. [Read More]

Tracked on Oct 26, 2004 11:59:17 PM