posted 04-18-2008 10:30 PM            

They have created a slick user interface, and have faithfully recreated the mathematical transformations and decision rules of our laboratory model - all in a package that feels as familiar as the LX Software.

OSS-3 includes specialized decision rules for single-issue (Zone) exams, multi-facet (MGQT exams, and mixed-issues (screening) exams. There are specialized algorithms included in the model to perform quality assurance tasks such as evaluating whether there are, or are not, significant differences between the RQs themselves, and other tasks.

Following a discussion in this forum, sometime last year, and the suggestions of Paul Menges, we have included a Test of Proportions (additional algorithm) that will provide a statisical estimate regarding whether marked artifacts have occured randomly, or non-randomly (i.e., strategically) in attempt to alter the test outcome. (This feature is also available in the Limestone implementation of OSS-3 - more on that in a bit.)

There are a number of other options, such as alpha decision thresholds for error tolerance, Bonferonni corrections, and a nice selection of powerful and important usability features. Other features include scoring window options (what better thing to argue and arm-wrestle over), and which decision rules to use. Decision rules have been optimized for balanced sensitivity and specificity, and are consistent with the published scientific literature (primarily Stuart Senter and Andrew Dollins) on two-stage scoring and decision rules. There are specialized decision rules for mixed-issues screening exams, that are optimized in a different way than the two-stage (Senter) rule, to provide maximum sensitivity to deception while retaining specficity to truthfulness and constraining inconclusives as much as possible. Keep in mind, however, that validation of these screening rules is limited to three screening samples (PCSOT maintenance, PCSOT disclosure, and LEPET), and there may be much still to learn about screening exams and screening algorithms.

I particularly like the chart display and requirement to review the questions prior to scoring, because the old adage "garbage in is garbage out" should be a mantra for chart interpretation. Despite its formidable statistical power and advanced user features, the tool is quite easy to use. The interface designer did a great job (I also suspect Mark Handler had something to do with it).

Here is a link to an example of the Lafayette's OSS-3 report.
http://www.oss3.info/lafayetteOSS3sample.html

A a screenshot of part of the report.

The OSS-3 model is a massive extension of the original OSS model of Krapohl and McManus (1999) and Krapohl (2002), and uses a decision model suggested by Gordon Barland (1985).

OSS-3 uses Kircher features (respiration line-length (Timm, 1982), electrodermal phasic amplitude increase, and increase in cardiovascular diastolic baseline). These featuers are repeatedly described as the most useful and robust physiological indicators (Kircher and Raskin, 1988, 2002; Kircher Kristjiansson, Gardner & Webb, 2005; Harris, Horner & McQuarrie, 2000, and Raskin, Kircher, Honts, & Horowits, 1988). One of the major under-discussed advantages of a computer algorithm based on humanly-recognizable features is that it become possible to study and resolve differences in cases when we find our hand scores disagree with the results of the algorithm. That is not possible to nearly the same degree with algorithms based on idiosyncratic or unknown features.

OSS-3 is not necessarily new. Its an evolved scoring model that now uses advanced mathematical transformations. It is based on known signal detection and statistical methods, and builds on decades of existing polygraph research. We are simply the construction crew that assembled the various pieces of polygraph technology into a project for which we look forward to the opportunity to show our development and validation efforts to others.

We have invested untold hours into this project, and we hope you enjoy it and let us know what you think.

Also, be sure and thank Chris Faucett for his commitment to making the tool available on the Lafayette system.

You can download the Release Candidate from my .ftp server at:
http://www.raymondnelson.us/ftp/OSS3-RC1.zip

If you are running a beta version of the software, you should uninstall any previous betas before installing the Release Candidate.

Next up: look for some new algorithms from Jamie, Tyler and the folks from Limestone - including specialized tools for statistical analysis of CIT and SPOT exams, an update to their implementation of Miritello Rank order system, and a tool for calculating the relative magnitudes described by Raskin, Kircher, Honts & Horowits (1988).

enjoy.

r

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

posted 04-19-2008 09:47 PM         

Is OSS-3 compatible with WIN98? The download 'appears' to have installed correctly- yet a I get a message "OSS3.exe expects a newer version of Windows" when trying to launch the program.

posted 04-20-2008 09:30 AM         

You wrote "Have you tried to run anything through the tool.." The program will notlaunch to access any "tools." When trying to start the OSS3, the program gives a message to upgrade to a newer version of windows (I'm running on Win98 presently).

I'm not sure I understand the "tool" which you are referring to. When uninstalling the program, a "Modify", "Repair" or Uninstall" option box will appear- if this is the tool you are talking about- yes, tried that- and still no go.

Any suggestions?

posted 04-20-2008 11:57 AM         

Yeah- I know, Win98 has seen its days. I've got two machines with Vista- but neither are set-up with Lafayette's software.

OSS3 appears to install correctly on Win98- as it appears to create the necessary folders\files. And in reading the setup.ini file of OSS3, it also appears that it is suppose to function with Win95, Win98, Win2000 etc.

But when I try to 'start the program' it asks for a newer version of windows. I'm just curious if the OSS3 program 'must' have a certain Win Version or newer to function.

posted 05-20-2008 01:17 AM         

I too, would like to take this opportunity to say 'thank-you to all of you' for your devoted time in the development of OSS-3 to advance test data analysis.

I confess my knowledge of statistics however is null (to which I apply the meaning of none, and not the making of null hypothesis statement). In hopes of coming to a rudimentary understanding of statistical terms and applications in an effort to put your work to good use, I continue with the struggle of reading a few books- unforntunately for me it serves as a very good 'sleeping aid'.

My lack of knowledge causes me issues in understanding the test results in terms of the P-values being rendered; as well as understanding other 'tests' available in the software for artifact distribution and spot score differential analysis.

As an example, on a recent mixed-issue screening exam, the Test Result was: Inconclusive- Not able to render a decision. Yet the Spot Scores Rendered were:
R4 .227 Result: No Significant Reaction
R6 .541 Result: No Significant Reaction
R8 1.00 Result: No Significant Reaction
R10 .981 Result: No Significant Reation

I'm curious as to 'why' the Test Result rendered was Inconclusive, when all the RQs spot score results state: No Signfificant Reactions.

I've been using the OSS-3 since April after Ray was kind enough to provide the readers of this forum with Layfayette's OSS-3 release candidate. I find the software is much 'user friendlier' than Polyscore, and I like the fact the measurements are on the report as well. I would have like to have seen on the report the selected 'time frame window' being evaluated.

posted 05-20-2008 08:11 AM            

If you are still using the Release Candidate, you should uninstall it and install the Release version 1.0.

OSS-3 is capable of scoring very aggressively, or very conservatively, depending on how you set it. The default settings will provide the best balanced accuracy and low inconclusive rates.

Inconclusives can occur because of a number of conditions, most of which have to do we data for which we become suspicious.

Your case seems to be providing p-values from the normal range of the truthful distribution, which should be NSR. However, the p-values are quite different, so I'd guess the case is INC due to a significant result in the KW-ANOVA. The KW is like an algorithm within the algorithm, and tests all the RQs against each other, to determine if any of them is very different (like the old Sesame Street song - one of these things is not the same...).

We had about 7 different beta versions, and one Release Candidate that you seem to have. Before the final version last week, and we fixed a small number of decision parsing and report content errors with RC1.

Can you strip the ID from the case and send it to me. I'd like to evaluate it with the final release version.

Here is a link:
http://www.oss3.info/ftp/OSS-3.zip

As will all pre-release software, you should uninstall any previous version before installing a new version.

more later.

r

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

posted 05-20-2008 03:04 PM            

Thanks for the case. The algorithm appears to me to be scoring the case as intended. The spots themselves score truthful. However, there is a significant difference between the spot scores, with RQ8 giving a much stronger score than the others. The algorithm is intended to alert you when one of the questions looks very different from the others.

The test itself is a mixed issue (screening) exam that looks to me like a PCSOT maintenance or monitoring exam. It’s a good looking test.

By default, screening cases are scored with the KW-ANOVA. Multi-facet (MGQT) cases are not. Multi-facet exams are known-allegation tests. Screening exams are unknown-incident and mixed issues. You can turn on or off the KW algorithm for screening cases, multi-facet exams, and even single issue exams. You can also turn on and off the Test of Proportions, which will test whether the location of artifacts is signficantly different from what would be expected if they occurred randomly.

The algorithm is just a tool. If you don't mind the way the test looks (I don't), you can turn of the KW for this exam. That is not incorrect, as long as you are confident the data . Just turn it back on for the next test.

How were your hand scores? I get generally strong scores, but not at RQ6. You'll see differences in computer and handscores because hand-scoring involves a single CQ (generally left or stronger, but Backster still says weaker), while the computer uses the average. The difference will be most evident when you have a couple of very strong CQ reactions that stand out from others. Handscores next to those CQs will be strong, while others may not. However, those strong CQs will affect all computer scores. This may be one reason why computers have tended to provide better specificity to truthfulness than handscorers.

Did you change the default scoring windows? It looks like you use 15 secs for pneumos.

The subject caved in on C7 and answered 'yes.' Honts' research indicates there is no problem scoring the question normally.

You may not see much difference in the results, but the screening rules are capable of offering slightly better sensitivity to deception, compared with the MGQT rules. That doesn't affect this case. The 2-stage rules will provide slightly better balanced sensitivity and specificity, but are not optimal for screening exams involving mixed issues.

more later.

can I post the charts for others to see?

r

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