SPEARMAN'S "G" AND HIGH RANGE IQ TESTS

For you high IQ types, mensa may not be selective enough. But for mere mortals 132 is very respectable. On the subject of entrance requirements, I would like to refer you to this article by the late Grady Towers;http://www.eskimo.com/~miyaguch/grady/societies.html Please pay particular attention to the last 3 paragraphs. You will find his conclusions simply astonishing.

He is simply saying that for those who sat for just one test like the MEGA and assuming a minimum of 0.7 correlation between the MEGA and the mensa test, then 86% would not pass the mensa test based on his calculations.

In detail here:http://www.eskimo.com/~miyaguch/grady/followup.html

Personally I have a lot of doubt on the high range tests of various societies. First of all, they were not created by psychologists, hence the construct validity is in serious question. Is it measuring what it's suppose to be measuring? Is it measuring psychometric "G". Or is it measuring a specific ability that has very low correlation with "G" .

In this interview between Chris Langan and Arthur Jensen:http://eugen.leitl.org/tt/msg12769.html A. Jensen basically says that IQ tests (even today) become less "G" loaded the higher we go ( Q12). Also, there is the problem of getting sufficient sample size (obviously) to obtain statistically significant conclusions (remember some societies claim rarities of 1 in a billion!). In fact Chris Langan point blank questions him (Q6) about "power-tests" to which he politely replies that there are standard ones out there that could be used (Ravens + Concept mastery) - read : "the ones out there created by ultra high IQ societies dont count".

Unless the test createors can provide rigourous proof that it is a test of cognitive abilitives of some sort (and what it is exactly that it is measuring), we can only say that they are puzzles. Granted one requires a certain amount of intelligence to solve it. I am certainly not a trained psychologist nor did I take any psychology units at UNI and my grasp of basic statistics is at a basic level (Statistics 101) and simple regressesion analyses. However from what I have read so far on psychologists like John Raven (son I think of John C Raven - the guy who first created the Raven's Matrices) is that they usually start out with what they want to measure and then devise a test to measure it (kinda obvious isnt it?) and one of the main tools psychologists used in the past were statistics because things like reaction time measurement, MRI scans of the brain (more neurological investigative type techniques) were simply not available then.

In fact John Raven never stated that any of his tests were tests of "intelligence" (page 54 bottom left corner): http://www.eyeonsociety.co.uk/resources/CognitiveAbilityAndOccupationalPerformance.pdf

On page 52: We can now return to our conclusion that the IRT-based item analysis of the RPM really does show that there is a continuum in "cognitive (actually "conceptual") abilitiy" and that this continuum can be assessed using a range of items running from easy "perceptual" items to difficult "analytic" ones. It involves the ability to discriminate figure from ground; the ability to discern order and meaning in (or make meaning out of) confusion; the ability to form high-level, usually non-verbal, concepts which enable one to make sense of the evironment. Spearman used the Latin word educere - to draw out to characterise and discuss this componenet of general intellogence - g- and contrasted it with reproductive ability, the ability to reproduce already verbalised knowlege.

He goes on to say that one's performance on such tests may even comes down to motivation (page 55 top left hand corner), rather than just pure cognitive ability.

J. Raven isnt even sure his tests measure "intelligence" but knows that it is measuring some kind of "meaning-making" ability. Back to the Langan-Jensen interview; Jensen's answer to Q12, he mentions that once you approach the high ranges e.g. 4-Sigma, you are not measuring "g" anymore but a specialized ability.

You see "g" depends on a high correlation with the performance of testees to many other types of abilities (e.g. maths, abstract logic, verbal e.t.c validating the concept that an intelligent person will perform equally well on all types of tests of cognitive abilities). Once a testee approaches the higher ranges the correlation drops (think of it, there are less people scoring at this range for the same correlation compared the lower ranges, as you extrapolate upwards you will ultimately reach 0 correlation which would be the ceiling of "g").......which is counter to the concept of "g" and since "intelligence" in the psychometric sense is anchored to "g" , you cant call it intelligence anymore. Jensen logically concludes that you are then measuring a specialized ability but not intelligence. A real live example would be, that someone who says score very high on one of the spatial high range tests like LS36 may not score at the same level on the Mega (which have a range of problems and not just abstract-spatial type). But if both tests are measuring "g" accurately, the testee should score equally high on both tests.

My undestanding of G based on what I read on Spearman and what he tried to do: He carried out a series of tests on a sample group of students in a school on a varied ranges of cognitive tasks such as mathematics, literature, reading comprehension e.t.c.He then "discovered" that if a student does well in one domain he tends to do well in most of the others. Spearman then postulated that there is something general of about "intelligence" in that if one does well in one task, he "tends" to do well in other types of tasks.A test A is then said to be "g" loaded if a testee does well in test A and then does well in test B,C,D and so on (high correlation).

If Test A is "the" test, then test A will have "predictive" power for other types of tests/tasks of cognitive abilities.Also it is not sufficient to claim "correlation" unless there is proven correlation for many many testees (note the correleation is on the results of the tests taken by many testees).

Question : Based on the above; How does a test maker in one of these ultra high IQ societes prove that their test is "g" loaded (read high correlation on varied tests of other types, for many many testees)? Note : at the very high ranges there are so few results to correlate with (read Arthur Jensens reply to Q12 from Chris Langan):http://eugen.leitl.org/tt/msg12769.html

Also the higher you go the less "g" loaded the test becomes (for obvious reason).And since psychometric tests are dependent on "g" what are we then measuring?

However, I certainly cannot disagree that the high range tests is in fact measuring some form/component of intelligence.But is it "g"? Maybe a better question at this point is what is "g"? Spearman's concept of "g' is actually a statistical one that relies on correlates. The factor analyses he carried out is suggestive of "g" but he could never extract it? certainly words like "it points to" , "is suggestive of", but he never really defined exactly what it was. He did distill it down to a few common factors such as being able to discern similarities or differences, grasping new concepts, speed of processing, working memory, all evident in subjects that tend to do well in a diverse range of mental tasks (he found that even the ranking of the subjects in every type of mental tasks stays the same for each).

However other psychologists arrived at different conclusions about "g";

Raymond Cattell - posits two types of "g" namely fluid and crystallized.

John Carrol - suggests the Three Stratum theorem e.t.c

But still no one to this day that I know off have defined exactly what it is or rather at least agree to the range of of cognitive abilities that it is composed off?

I would like to share with you the recent finding of an American Psychologist Douglas K. Detterman.http://books.google.com/books?hl=en&lr=&id=2OjruFlEWukC&oi=fnd&pg=PA223&dq=Detterman,+D.+K.+(1982).+Does+%22g%22+exist%3F&ots=eBm-ilkIgH&sig=6IPJwp9jwWnH6QniXM55AWHzVZg#PPA235,M1page 235-236

They speculated that subjects who have damaged or inefficient central processes should perform similarly on all tasks (equally badly) because the damaged central process causes the whole system to perform badly. On the other hand , those with efficient central process will be more variable on all tasks because any limitation on any particular tasks will be dictated by peripheral processes that do not affect the entire system hence maintaining efficiency for the "whole".To investigate this, they divided up the distribution into 5 equal parts. Within each division of the distribution, they correlated subtests of IQ tests with each other. They did the same for basic cognitive tasks from a battery of basic cognitive tasks.

What they found was simply shocking. They found that "g" correlates twice as high among those with low "IQ" compared to subjects with high IQ, lending support to the above.

If true, it is certainly suggestive of no existence of a unitary "g". But many abilities especially at the high ranges.