posted 09-25-2006 08:36 PM            

I came across this article while reading the news on MSNBC and thought it might be of interest for review and discussion.

quote:

---

The Supreme Court shook the foundation of forensic science in Daubert v. Merrell Dow Pharmaceuticals, when it ruled in 1993 that evidence should meet standards of peer review, replicability, documentation and stated rates of probable error--all of which DNA analysis can claim. Traditional methods, in comparison, fall far short. "These techniques are not quite junk science," says Michael Saks, a law professor at Arizona State University, "but they are junky. They make claims of accuracy or even infallibility that they can't back up. The fact is," he says, "we don't know how accurate these methods are because no one's ever done extensive studies to find out."

The few small studies that have been conducted suggest that error rates run from the low single digits (for fingerprints) to the teens (hair matching) and the 60s (voice recognition). The accuracy of a match depends not just on the sample quality but on the experience, judgment and integrity of the forensic examiner.

---

posted 09-26-2006 08:53 AM            

this is from another (long) thread

quote:

---

It is my understanding that there are different schools of thought on this (excuse the cliché) and I am not certain that any particular way has proven to be considerably better than the other. If memory serves me, there is another idea that one should not score the response until after the verbal answer of the subject.

---

I'm not aware, of the top of my memory, that it has been thoroughly investigated regarding scoring a response before or after the answer, but I do recall that some evidence has suggested that answering the question is not necessary. I think the logic of scoring after the answer rest on the assumption that the answer is either a "lie" or the "truth." However, conditioned response theory tell us - in concealed information testing, for example - that the question stimulus itself can evoke a response from guilty subjects.

In terms of

1) peer review
2) replicability
3) documentation
4) and stated rates of probable error

... we may be functioning with some blind spots around some of our tools and techniques, and some of the things we're taught in polygraph school. OSS and other information developed by some of the more prominent and recognizable researchers have met some of these conditions, but we've still got to get the word out - and we have more work to do.

All of this ammounts to a need for good theory and good measurement. Ultimately, all tests are math tests.

For me, the science stuff keeps the ugly trench work interesting.

r

------------------
"Gentlemen, you can't fight in here, this is the war room."
--(from Dr. Strangelove, 1964)