posted 09-14-2007 09:08 PM         

The PolyScore (scoring) windows are unchanged from 5.5. PolyScore uses a variety of respiration features, including some related to a generalization of line-length.

PolyScore 6.0 is incrementally better than version 5.5. It relies on the same underlying features, but evaluates them differently. In particular, PolyScore 6.0 actually has two separate algorithms that it votes for a final decision (sort of a built-in quality control). If these two algorithms disagree, then the result is an inconclusive.

Pulse: after 1.5 sec latency from QuestOnset, scoring window 8 sec

EDR: after 1 sec latency from Quest Onset, scoring window 13 sec

Respiration: after 1.5 sec latency from Quest Onset, scoring window 16 sec

posted 09-15-2007 12:19 AM            

I appreciate the update. Polyscore, as with other algorithms, play an important role in research and the "techy" side of polygraph. They also provide a great "prop" for younger examinee's who rely heavily on the computer for information and socialization.

Now my problem. Many examiners are (seemingly) starting to rely upon the algorithms more and more. I have two examiners around my area that apparently rely upon them solely to make their decisions (for them).

I see this as an emerging deficiency for the profession. ANYONE who must rely upon the computer to give them the result of an examination should pack up their shingle, close their shop and go fishing (permanently!).

If you are not comfortable with your own abilities, relying on the computer is going to cause you a lot of problems at some point in the future.

Buck-up folks! If you're not sure how to score your polygraph charts; ask someone! Go to training! Or, at worse, read george's site; he'll fill you in!

But, for goodness sake, please learn your profession and stop relying on technology to read your mind and charts. R2D2 doesn't know the difference between a DB, SW, Bp or SI.

This (so called) technology is going to cause us trouble if we keep thinking that a computer can do better than a trained, attentive and professionally trained examiner.

Best to all,

Jim

posted 09-15-2007 11:13 AM            

I am with Sackett, it is a nice tool but you must learn to score charts!

posted 09-16-2007 11:54 AM            

quote:

---

PolyScore 6.0 is incrementally better than version 5.5. It relies on the same underlying features, but evaluates them differently.

---

The read-between-the-lines on "incrementally better" is that the difference in performance is either: 1) theoretical, 2) observed only in a single sample, 3) untested, or 4) not statistically significant. That's OK, because when you are working with an algorithm that is problems very accurate already (.8s to .9s can be considered very accurate in the social sciences) even non-significant improvements are hard to achieve, and the aggregation of several non-significant improvement may ultimately prove to be significant.

quote:

---

In particular, PolyScore 6.0 actually has two separate algorithms that it votes for a final decision (sort of a built-in quality control). If these two algorithms disagree, then the result is an inconclusive.

---

Calling this a form of built-in quality control is a piece of marketing hype that I wouldn't take too far. Its an oversimplification, offered most likely for the benefit of non-propeller-headed (non-statistician) folks who want to understand the concepts better.

The OSS-3 screening algorithm also uses too algorithms to improve decision accuracy - in the form of a Kruskal-Wallis nonparametric comparison of multiple mean values (a form of ANOVA) before allowing an NSR result. In the Limestone implementation of OSS-3 you can switch this extra feature "on" for MGQT (multi-facet) and single issue Zone exams (when using two-stage decision rules). It is off by default because that is how we trained the algorithm. Turning it on can be expected to force a few more INC results, with a possible increase in decision accuracy.

You can see the formula for the KW test here
http://www.oss3.info/OSS-3_formulae.pdf

and a flow-chart of the same operation here
http://www.oss3.info/Flow_Chart_OSS-3.pdf

enjoy.

I'm back to the I-40 super-slab/race-track/time-warp.

Later.

r

------------------
"Gentlemen, you can't fight in here. This is the war room."
--(Stanley Kubrick/Peter Sellers - Dr. Strangelove, 1964)

posted 09-16-2007 04:40 PM            

------------------------------

sackett:

quote:

---

R2D2 doesn't know the difference between a DB, SW, Bp or SI.

---

I once heard computer scoring algorithms described as "for discussion purposes only."

I some ways I would not disagree with that, and I agree with the spirit of Sackett's statements. Examiners need to know how to score their own charts.

Gordon Barland has said more than once that chart data analysis should begin with numerical scoring, but should not end with numerical scoring (or something like that.) He has also emphasized that there is a difference between scoring charts and analyzing charts - which I presume him to mean analyzing all of the events that have contributed to the data/tracing features.

Now the other side of the argument is that computers don't recognize the difference between DB, SW, Bp or SI (or BN) because they were never trained to do so.

If some smart person can authenticate human signatures at better than 70% using a Komogorov-Smirnov algorithm, then it would certainly seem possible to identify bad polygraph data using some form of math.

Surprisingly little has been done in the area of polygraph signal processing since Kircher and Raskin 1988. John Harris has some proprietary/patented features that are of no use to the profession unless using his algorithm - and are unlikely to be further developed by any subsequent researchers. Bruce White has some interesting ideas, but he's working in an undocumented cloistered vacuum for which we aren't permitted to understand the current limitations nor contribute to any progress (other than sending him data). The Kircher features are simple and understandable in their intended form, but are perhaps rather blunt against artifacted or CM data.

Taylor:

quote:

---

I do like the scoring algorithms when you have a DI subject and the computer reports 100% DECEPTION (IN BOLD RED LETTERS) and you can turn the computer around to show the examinee.... sometimes you get a confession right there.

---

I think a lot of scientific minded folks would cringe at the thought of these theatrics. It doesn't take a genius to see through the ruse. Still, it could be fun with some block-heads.

quote:

---

One time Polyscore came up with .01% possibility of deception. I had the guy completely 100% DI with an admission.

---

I think we would all be very curious to see those charts if you still have them. How messy were they? Did you mark artifacts before running the algorithm? Do you have any sense of why that occurred? These things are frustrating and concerning, but also represent important opportunities for potential learning. Any chance you could send the data for us all to see?

OK, I'm heading back to the relativity machine - home ways is bestways so best I be offways. (I know the Burgess allusion is a violent one, but I like it anyway.)

Peace,

r

------------------
"Gentlemen, you can't fight in here. This is the war room."
--(Stanley Kubrick/Peter Sellers - Dr. Strangelove, 1964)