posted 06-30-2004 10:33 AM            

I made a similar post, but I buried it within another and received no responses. I am re-posting here with a few more details.

My goal is to come up with cut-off scores for the Objective Scoring System (OSS) for Bi-Zone tests. As many of you probably know, Don Krapohl has crunched some numbers and established cut-offs and a probability table for the standard DoDPI / Federal ZCT and other formats containing three relevant question formats. As of yet, nobody has decided to do the same for Bi-Zone tests.

(For those of you who don’t know, OSS is a method for scoring (manually or computer) polygraph tracings for evidentiary purposes. It is “objective” in that scoring decisions are based on measurements of tracings, so every scorer should arrive at the same decision, i.e., the examiner’s subjective opinion as to whether the tracing get s a +1, or +2, etc., is not an issue. The system allows scorers to estimate the probability of error (based on data derived from exams in which ground-truth is known). In other words, one could say – with evidence to back it up – there is only a three percent chance that an innocent examinee would achieve a score of -14, for example, using the OSS.)

I have the DoDPI confirmed tests database, but there is not enough data in the field tests (i.e., too few tests) to do what I would like to do, so I need your help.

Who would be willing to send me CONFIRMED Bi-Zone tests (Axciton or Lafayette computerized charts)? By “confirmed” I mean the subject admitted (or confessed) to lying to the relevant issue, or some other evidence, e.g., DNA, confirms to a near certainty the results. (Eyewitness evidence is not sufficient as it is not all that reliable.) For NDI tests, a confession by the guilty party is sufficient.

What I need are your tests and how you know they are confirmed. I don’t care if you run three charts or four, if you use outside issue / symptomatic questions or not. As a matter of fact, I invite a variety. I will divide the tests up and see if any of that stuff matters in the end (providing there is sufficient data to do so). I can’t get too many tests – the more data the better!

When I say “Bi-Zone,” I mean the following format:

1 Neutral / Irrelevant
2 Sacrifice Relevant
3 Outside Issue / Symptomatic
4 PLC
5 Relevant
6 PLC
7 Relevant
8 PLC
9 Outside Issue / Symptomatic

I realize some of you remove the outside issue / symptomatic questions, which means your tests are a little shorter. I have created an Excel spreadsheet to automatically make the computations for either version. (The measurements are automatic too as I will be using Raskin’s “Extract” program to make the measurements, which means everything will be as objective as possible.)

When all is said and done, I will see that the results are published (somewhere) so you’ll have them as well.

You can send tests to me via mail or email, if you know how. (Lafayette lets you strip the identifying information if that is a concern. I suspect Axciton can do the same, but I’ve never tried it.) Let me know if you’re willing and how many of each type (NDI and DI) tests you can send.

If you’d like to know something the data might reveal – that I haven’t thought of – let me know, and I’ll see if I can work that into the research. (I have a few more ideas, but I won’t bore you with them now.)

As an aside, does anybody know of any validity studies on the Bi-Zone? I thought I learned at polygraph school it is one of the most “validated” techniques we use, but it appears that is not the case. I spoke to Don Krapohl of DoDPI, and he is not sure if there have been any validation studies or not.

Thanks in advance,

Barry

I forgot to mention I don't care what your results were. If you scored a person NDI, for example, and your test was conducted properly (meanining, among other things, the examinee was lying to the RQs), but you later learned through other evidence (e.g., DNA, a confession, etc.) the person was lying, then include those. (It's important to get a full picture of where scores fall to create a probability table.)