posted 07-26-2006 11:26 PM            

1. I
2. SR
3. CQ
4. RQ
5. CQ
6. RQ
7. CQ
8. RQ
9. CQ
10. RQ

There must be at least a +3 in all spot totals for NDI.

A -3 in any spot total, regardless of other spot totals is DI.

END.....

posted 07-27-2006 08:51 AM            

If you have a distortion in the CQ before the last RQ you're pretty much S.O.L.

Anyone have any thoughts?

posted 07-27-2006 11:18 AM            

1. N
2. SR
3. C1
4. R1
5. R2
6. C2
7. R3
8. R4
9. C3

All questions are bracketed by controls before and after, and there are three controls to set.

Though, as Barry pointed out previously, it's not on the list.

r

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

posted 07-27-2006 02:21 PM            

You can eliminate the last 2 RQ's as long as your 1st RQ is secondary in format and your 2nd RQ is primary in format. You cannot have two primary RQ's like a ZCT "You Phase," as I have seen many times when I do QC for AAPP. It just makes the test invalid.
You can also cover two two issues with one primary RQ on each issue and 1 secondary RQ on each issue; however, this is highly unrecommended since the more issues, the less accuracy - particularly on truthful people.

As for the situation of a distortion on CQ9 - after asking RQ10, simply repeat CQ9 to end the examination. If you have informed the examinee prior to running the test, that some questions may be repeated as part of the routine process (but not that they have done anything wrong), they will not know that repeating CQ9 was because they caused a distortion, and it will not affect the validity of the examination.

Elmer

posted 07-27-2006 05:03 PM            

I was unclear at first, but I think your statements pertain to the USAF-MGQT.

It would be great to make a document available to the members here who could not attend that AAPP conference.

Taylor,

I haven't heard that anywhere, but have no basis to disagree. I first saw the USSS-MGQT format from a former government examiner who I believe uses it exclusively. He's a very skilled examiner, and he calls it an MGQT. I was a little skeptical, and found the technique described in an AAPP manual.

I'll post a link when I get a moment.

Do you have document on the LEPT? I'm sure others would be interested.

I would point out that the USSS-MGQT's paired RQs, bracketed by CQs is similar in structure/sequence to the TES

N1
N2
SR
1C1
1R1
1R2
1C2
2R1
2R2
2C1
3R1
3R2
3C2

which, according to the 1995 document, allows scoring each RQ either left or right to the stronger CQ.

I have no idea if this is current.

On the issue of validity, I've many thoughts so I'll start another thread.

r

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

posted 07-27-2006 08:39 PM            

You are right. I was referring to the AFMGQT. I apologize.

I have never heard of the LEPT. I have heard of the LEPET which is an acronym for the Law Enforcement Pre-Employment Test as designated by DODPI. Perhaps that is what to are referring to. The LEPET test is an outgrowth of the AFMGQT which allows as many as 5 RQ's although most recommend no more than 4 RQ's for the same reason as my earlier post - the more issues - the less accuracy. Unless there is new research that I have not been made aware of, the AFMGQT and any of its outgrowths have no studies to validate them. The DODPI MGQT - Army version is the only MGQT (Reid-based) format that has validation studies to support it.
Since the AFMGQT is a better version (when sticking to one issue with multi-facets)as a result of the addition of the Sacrifice RQ and the two extra CQ's, it may have a higher accuracy for truthful examinees than the Army version which has a horrible accuracy level for truthful. The AFMGQT should also have as great accuracy level for deceptive examinees just as the Army one does.

I have never had training, and very little exposure to the Secret Service MGQT so I have no standing to comment on it.

I have documents on both the LEPET (if that is what you meant) and the updated DODPI AFMGQT. As far as making them available - it would depend on the number of requests.

[This message has been edited by lietestec (edited 07-27-2006).]

posted 07-27-2006 08:39 PM            

I've mentioned this before, but the way around some of those rules is to go with a Utah MGQT, which oddly enough, looks very much like the Navy CIS MGQT above. (One version adds an intro question at the beginning and a neutral at the end.)

posted 07-27-2006 10:21 PM            

quote:

---

The LEPET test is an outgrowth of the AFMGQT which allows as many as 5 RQ's although most recommend no more than 4 RQ's for the same reason as my earlier post - the more issues - the less accuracy.

---

I can't find anything published on this topic, but here are my own mathematical musings on the subject of too many RQs on mixed issue or multi-facet tests which are subject to vertical or spot scoring rules.

We don't often think of it this way but the statistical methods underlying why the polygraph works involve the same old distribution curves, means, standard deviations and standard errors that we beat into the heads of every graduate student in the country. In polygraph, our scoring models can be understood as based upon the standard error of mean differences across a bi-modal distribution of truthful and deceptive subjects. All this means is that we expect to observe one mean and deviation range for truthful subjects and another mean and deviation range for deceptive subjects. (Matte provides and illustration of this somewhere in his book). We could use common z-score methods to estimate the proportion of truthful or deceptive subjects which an individual's scores meet or exceed. In the social and biological sciences (which polygraph is a part of) we accept as the "normal range" all persons whose scores occur within two standard deviations from a sample or estimated population mean. 95 percent of all person fall in this range. Regarding the other 5 percent, we don't say "abnormal" or "weird," though we are sometimes tempted to do so. Instead we say those 5 percent are "outside of normal limits." Scores more that three standard deviations from the mean are sometimes termed "outliers," as they are few and far between, not representative of the normal population and have the effect of skewing data (distribution curves). In polygraph, with a bi-modal distribution we expect to see test scores that are within the normal range for one group (truthful or deceptive), and outside the normal range for the other group; this allows us to assign that score to that group with some statisical precision or significance.

Because we are dealing with normed data, there are important concerns here - one of which is the representativeness of the sample regarding the intended population and individual test subject. So it's no big surprise that the polygraph method, developed on mostly psychologically normal persons, does not work well with mentally retarded persons or serious psychiatric cases. (There are theoretical rationale for these limitations as well, and the combination of data models and theory makes for robust science.) Because we are dealing with normed data, some individual scores will [u]always[u] fall outside of the normal range - for either mode (truthful or deceptive). These scores will be inconclusive. Any examiner who expects to never see inconclusive scores is fooling himself. So, how many of those inconclusive scores should we expect to see. Probability science tells us to minimally expect somewhere between 2.5 to 5 percent of these scores to be inconclusive. Its unlikely things will get much better than that (unless some brilliant mathematician invents some new kind of predictive statistics). This, of course, assumes the most robust test we can construct - and that would be the single-issue Zone test. Real-world complications caution us to expect less than perfect performance, so we might easily double our expected optimal inconclusive rate to 5 to 10 percent - still not too bad. However, this gets more complicated with mixed-issue and multi-facet tests.

Think about this the estimated inconclusive rate is the inverse of the resolution rate (and the resolution rate is the inverse of the inconclusive rate). So, by this estimation our resolution rate for single-issue zone tests is in the area of 90 to 97.5 percent (.9 to .975). With mixed issue/multi-facet tests we can estimate inconclusive rates as the inverse of the resolution rate raised to the exponent of the number of distinct RQs. (whew).

So, best case 2.5 percent

two distinct questions = 1 - (.975 x .975) = .05
three distinct questions = 1 - (.975 x .975 x .975) = .07
four distinct questions = 1 - (.975 x .975 x .975 x .975) = .1
five distinct questions = 1 - (.975 x .975 x .975 x .975 x .975) = .12

and worst case scenario - inc at 10 percent (.10)

two distinct questions = 1 - (.9 x .9) = .19
three distinct questions = 1 - (.9 x .9 x .9) = .27
four distinct questions = 1 - (.9 x .9 x .9 x .9) = .34
five distinct questions = 1 - (.9 x .9 x .9 x .9 x .9) = .41

As you can see, with a four (mixed issues / multi-facet) question test, we can estimate inconclusive rates as ranging from .1 to .34

With five distinct questions, we can estimate inconclusive rates ranging from .12 to .41

I suppose its a matter of policy and what level of inclusive results is tollerable to the consumer. Personally I think it gets excessive with four or five questions.

I've asked several people to comment on this model, and no one has yet pointed out a serious error. Keep in mind though that it is only a hypothetical model. It really should be investigated through data.

An anecdotal survey of polygraph examiners, treatment providers and supervising officers indicates they do tend to seen inconclusive rates in these ballparks. However, some supervising officers and therapists, have been taught that inclusive results means a bad examiner. The result of this is that some examiners have marketed themselves to those supervising offices and therapists as never having inconclusive results. I think this is asking for trouble in the long run, as we are now tempted to sell "confidence" and not "science."

As for the statement about more questions = less accuracy - I'm not sure this is the most correct way to state this concern. More questions improves the sensitivity of the test, which is desirable in screening situations even though we have to anticipate a corresponding increase in false-positives (that will be rectified upon further investigation, in a properly formulated testing and screening program). Screening sensitivity is a form of validity. As Don Krapohl said at APA, "there is no such thing as free accuracy." I think it gives the wrong message to interpret these models and expected phenomena as somehow invalid.

OK, now for a new thread.

t

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

posted 07-28-2006 10:05 AM            

There is some research (out of Utah) that shows that accuracy is reduced when a person is lying to some and truthful to other RQs, but you're right, the tests are still "valid."

Much of the screening research shows accuracy rates at chance levels with the exception of the TES which only tests two RQs. There are a few that put them at 70 to 80%. The problem is the data is all over the place when it comes to identifying what questions are the problem. In other words, if a guy lies to one RQ in a multi RQ question test, you can't say with much confidence what question is the problem. (The test may be sensitive to deception, but specificity is a problem. Thus, the successive hurdles solution, which really requires a screening exam, breakout of the seemingly problematic question(s), and a new run of the screening exam, minus the problem issue if resolved in the breakout test(s). Too many issues could keep you going for days of testing!)

posted 07-28-2006 02:46 PM            

Has anyone here seen the material on the Jack Ruby polygraph?

70+ relevant questions, and he passed.

that was 1964 and things have changed since then

what fun.

r

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

posted 07-28-2006 03:53 PM            

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

posted 07-28-2006 04:39 PM            

Some of those guys are still around.

Arlen Spector was apparently present in the room during the test.

r

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