Questions on _g_ and intelligence

[nuenke]

I came across a recent study that looks at intelligence and specific areas or patches of gray matter. They did not do a correlation as such, but it does look like they are locating those areas related to g, as well as other factors like fatty tissue around axons, glucose uptake, etc. But at least the IQ vs. Brain size is getting narrowed down to IQ and specific brain reqions.

 

[Nereid]

The APA report addresses some issues quite well; it addresses others incompletely; and it misrepresents some issues. Consider the discussion about heritability. They discuss only MZA data and say nothing about path analysis. Why? The results are in agreement, but the literature claims are that path analysis is more robust. In this area, they had no way of knowing what would later be discovered by Dr. Paul Thompson at UCLA: "We were stunned to see that the amount of gray matter in frontal brain regions was strongly inherited, and also predicted an individual's IQ score..." His work was done with MRI. Their coverage of the Scarr-Weinberg findings was poor.

It seems that there may be more than one 'APA report'; could someone please tell me whether http://www.apa.org/releases/intell.html is the one we're discussing?

"APA Task Force Examines the Knowns and Unknowns of Intelligence

1996 Press Release
What is intelligence and can it be measured? These questions have fueled a continuing debate about whether intelligence is inherited, acquired, environmental, or a combination of these and other factors. In a field where so many issues are unresolved and so many questions unanswered, the confident tone that has characterized most of the debate on these topics is clearly out of place, according to a new report by the American Psychological Association (APA).

The report, entitled 'Intelligence: Knowns and Unknowns,' was written by APA's Task Force on Intelligence. The task force convened in January 1995 to prepare a dispassionate and authoritative report in response to the fall 1994 publication of Herrnstein and Murray's The Bell Curve. 'Their book sparked a new and vigorous round of debate about the meaning of intelligence test scores and the nature of intelligence itself, a debate in which little effort was made to distinguish scientific issues from political ones,' stated Ulric Neisser, PhD, chair of the task force.

Because there are many ways to be intelligent, there are also many conceptualizations of intelligence. Standardized intelligence test scores (IQs), which reflect a person's standing in relation to his or her generational peers, are based on tests that measure a number of different abilities. Psychometric testing, the use of standardized tests to assess specific abilities, has generated the most systematic research though many questions remain unanswered. According to the task force report:

* Intelligence test scores partially predict individual differences in school achievement, such as grade point average and number of years of education that individuals complete. In this context, the skills measured are important. Nevertheless, population levels of school achievement are not determined solely or even primarily by intelligence or any other individual-difference variable. The fact that children in Japan and Taiwan learn much more math than their peers in America, for example, can be attributed primarily to differences in culture and schooling rather than in abilities measured by intelligence tests.

* Test scores also correlate to some extent with measures of accomplishment outside of school, for example adult occupational status. This correlation is linked with school achievement because, in the United States today, high test scores and grades are prerequisites for entry into many careers and professions. However, a significant correlation between test scores and occupational status remains even when education and family background have been statistically controlled.

* Differences in genetic endowment contribute substantially to individual differences in (psychometric) intelligence, but the pathway by which genes produce their effects is still unknown. The impact of genetic differences appears to increase with age, but it is not known why.

* Environmental factors contribute substantially to the development of intelligence, but it is not clearly understood what those factors are or how they work. Attendance at school is certainly important, for example, but it is not known what aspects of schooling are critical.

* The role of nutrition in intelligence remains obscure. Severe childhood malnutrition has clear negative effects, but the hypothesis that certain 'micro- nutrients' may affect intelligence in otherwise adequately-fed populations has not been convincingly demonstrated.

* The differential between the mean intelligence test scores of Blacks and Whites does not result from any obvious biases in test construction and administration, nor does it simply reflect differences in socio-economic status. Explanations based on factors of caste and culture may be appropriate, but so far there is little direct empirical support for them. There is certainly no such support for a genetic interpretation. At this time, no one knows what is responsible for the differential.

* It is widely agreed that standardized tests do not sample all forms of intelligence. Obvious examples include creativity, wisdom, practical sense, and social sensitivity, among others. Despite the importance of these abilities, very little is known about them, how they develop, what factors influence their development, and how they are related to more traditional measures.

* Although there are no important sex differences in overall intelligence test scores, substantial differences do appear for specific abilities. Males typically score higher on visual-spatial and (beginning in middle childhood) mathematical skills; females excel on a number of verbal measures. Sex hormone levels are clearly related to some of these differences, but social factors presumably play a role as well.

The task force distinguishes sharply between scientific research and political rhetoric. 'The study of intelligence does not need politicized assertions and recriminations; it needs self-restraint, reflection, and a great deal more research.' According to the report, the questions that remain are socially as well as scientifically important and 'that there is no reason to think them unanswerable, but finding the answers will require a shared an sustained effort as well as the commitment of substantial scientific resources.'

The American Psychological Association (APA), in Washington, DC, is the largest scientific and professional organization representing psychology in the United States and is the world's largest association of psychologists. APA's membership includes more than 132,000 researchers, educators, clinicians, consultants and students. Through its divisions in 49 subfields of psychology and affiliations with 58 state and territorial and Canadian provincial associations, APA works to advance psychology as a science, as a profession and as a means of promoting human welfare.[/color]

 

[Nereid]

• High g Loading
  Matrix relations (.94)
  Generalizations (.89)
  Series completion (.87)
  Verbal analogies (.83)
  Likeness relations (.77)
  Problem arithmetic (.77)
  Paragraph comprehension (.73)
  Perceptual analogies (.70)
  
  [...]
  
  • The knowledge and skills tapped by mental test performance merely provide a vehicle for the measurement of g. Therefore, we cannot begin to fathom the causal underpinning of g merely by examining the most highly g-loaded psychometric tests.
  (Ibid. p74.)

• If 'matrix relations' tests have a _g_ loading of 0.94, what else is there is in these tests? In well constructed matrix relations tests, what is the typical individual variance (i.e. an individual takes the same (type of) test many times)?

 

[Nereid]

Contrarily, research has demonstrated that:

• There are plausible reasons ... for assuming that individual differences in g have an approximately normal, or Gaussian ("bellshaped"), distribution, at least within the range of ±2σ from the mean. That range is equivalent to IQs from 70 to 130 on the typical IQ scale (i.e., μ = 100, σ = 15).

(Arthur Jensen. The g Factor. p88.)

Does Jensen go on to say what those 'plausible reasons' are? Subsequent to when Jensen wrote this, what research has been done to establish what that distribution actually is? Why did Jensen add the caveat "within the range of ±2σ from the mean"?

 

[Nereid]

Based on the article I cited above, this would indicate we should be seriously looking into the extent of malnutrition among the black population in the US, especially in pre-school aged children, since it seems nutritional interventions for school-aged children may be too late to help.

Why so? Your comment implies that malnutrition affects most blacks in the US, but not whites. It doesn't add up, since the W-B IQ gap is largest at the highest SES level and lowest at the lowest SES level. Is malnutrition related to SES? If so, wouldn't it make sense that the higher SES levels would have better nutrition? Likewise, would you support seriously looking at the malnutrition in Hispanics, non-Hispanic whites, and Asians with respect to Ashkenazi Jews? Is it likely that Asians have malnutrition that causes them to have a mean IQ below that of Ashkenazi Jews?

Where to begin to untangle this?

For a start, sentences in English with comparatives ('most', 'higher') don't help IMHO; the reality of population groups is very rich, and teasing apart the influences of multiple factors calls for hard thought.

Take the Black and White SES: if early childhood malnutrition contributes, then you would need to know about the relative experiences of groups who are now in their 40s to 60s. Cet par it may very well be that 'the higher SES levels would have better nutrition', but my impression is that in 21st century US society things are most certainly not otherwise equal.

Take Asians: my impression is that a significant proportion of Asians in the US are migrants or children of migrants; further, with some notable exceptions, they are primarily 'economic migrants' - they made conscious choices and effort to migrate. In this sense, the 'control group' of whites would be those who arrived in the US several centuries ago (and there'd be no black control group; unless I have misread my US history, most folk who came from Africa didn't make get a choice).

is anything If you are unfamiliar with the relative brain size findings this may be helpful:

Is There a Biological Basis for Race and Racial Differences?
By J. Philippe Rushton
Insight, May 28, 2001

What I've found is that in brain size, intelligence, temperament, sexual behavior, fertility, growth rate, life span, crime, and family stability, Orientals fall at one end of the spectrum, blacks fall at the other end and whites fall in between. On average, Orientals are slower to mature, less fertile, and less sexually active, and have larger brains and higher IQ scores. Blacks are at the opposite end in each of these areas. Whites fall in the middle, often close to Orientals.

The relation between brain size and intelligence has been shown by dozens of studies, including state-of-the-art magnetic resonance imaging. Orientals average 1 cubic inch more brain matter than Whites, and Whites average a very large 5 cubic inches more than Blacks. Since one cubic inch of brain matter contains millions of brain cells and hundreds of millions of nerve connections, brain size differences help to explain why the races differ in IQ.

Racial differences in brain size show up early in life as well. The U.S. Collaborative Perinatal Project followed more than 50,000 children from birth to seven years. In the 1997 issue of the journal Intelligence, I showed that these Orientals had larger brains than Whites at birth, four months, one year, and seven years; the Whites had larger brains than Blacks at all ages. In the United States, Orientals are seen as a "model minority." They have fewer divorces, out-of-wedlock births, and fewer reports of child abuse than Whites. More Orientals graduate from college and fewer go to prison. Blacks, on the other hand, are 12% of the American population but make up 50% of the prison population.

Genes play a big part in athletic ability, brain size, IQ, and personality. Trans-racial adoption studies, where infants of one race are adopted and reared by parents of a different race, provide some of the strongest evidence. Oriental children, even if malnourished before being adopted by white parents, go on to have IQs above the white average. Black infants adopted into middle-class white families end up with IQs lower than the white average.

I trust this is a piece of journalism and not a scientific paper; I sincerely hope that Rushton has done good science to back each of the points he makes here. In particular, I would expect that he has done studies in other countries, to demonstrate (for example) that he's not just reporting on some unique, US, human condition.

 

[Mandrake]

Suppose I want to know my _g_ and how it may vary. I understand that I can take a test (e.g. an IQ test with high _g_ loading), some chronometric tests, or an EEG. From just one test – of any of these three kinds – what +/- number would my _g_ come with? What is the distribution (e.g. Gaussian)? How does each type of test vary wrt this +/-?

Since most observers do not have the laboratory devices and skills to measure intelligence via chronometric or electroencephalography techniques, the most common approach is to use an IQ test. As you probably know there are very many IQ tests, although only a few of them are used in most serious research. Of these the Raven's is most often cited in research programs. The WAIS versions are also widely used. For most tests, _g_ has to be extracted by weighting the subtest scores according to their _g_ loadings. Obviously, different tests will have different _g_ loadings, different subtest structures, and different associated errors. As with measuring physical phenomena (consider temperature) there are errors that can be identified in connection with many aspects of the measurement. Most of these errors are small. Ultimately the reliability coefficient is of central importance. Jensen: "The difference between the reliability coefficient and unity represents the proportion of the total variance of the measurements that is attributed to measurement error. ... In my laboratory we have been able to measure such variables as memory span, flicker-fusion frequency (a sensory threshold), and reaction time with reliability coefficients greater than .99. ... The reliability coefficients for multi-item tests of more complex mental processes, such as measured by typical IQ tests, are generally about .90 to .95. This is higher than the reliability of people's height and weight measured in a doctor's office! The reliability coefficients of blood pressure measurements, blood cholesterol level, and diagnosis based on chest X-rays are typically around .50." [The _g_ Factor, P. 50]

The concept of TRUE SCORE is related.
Regressed true score = [(reliability coefficient) x (test score - mean score for population)] + (mean score for population)
This is obviously a hypothetical score that attempts to factor out measurement error. For a very detailed discussion of all things related to measurement error, see Jensen, A.R. (1980). Bias in mental testing. New York: Free Press. Jensen says that tests with reliability coefficients of less than .90 should (generally) not be used.

Since _g_ has to do with my brain, and I know all kinds of things affect the performance of ‘brain tasks’, I’m sure psychometricians have done extensive research into the effects of the following on one’s _g_, as estimated by one of the three kinds of _g_ tests:
- alertness, e.g. taking the test mid-morning after a good night’s sleep vs one taken at 2am
- drugs, e.g. caffeine, alcohol, anti-histimines; especially those which are known to affect reaction times and medications for mental conditions
- pain, esp headaches
- general wellness/fitness, e.g. fever, top physical form, hunger,
- brain affecting illnesses or conditions, e.g. epilepsy, depression, Alzheimers, tumour, physical injury, PTSS
- mood
- age

How do estimates of _g_ vary for each of these classes of factors?

I don't have a source at hand with a ready answer. When an IQ test is given, it is the responsibility of the test administrator to determine that the person taking the test is fully alert and not encumbered by factors that would render the test inaccurate. Some tests, such as the WAIS are age adjusted.

 

[Mandrake]

Nereid:

What has psychometric research – not just in intelligence – shown wrt effects somewhat analogous to the placebo or white coat effects? How does the act of administering a psychometric test affect the subject (either consciously, or, more importantly subconsciously)?

There are various papers that have examined the conditions of test taking. The range of things considered includes, for example, stress. Jensen used pulse rate to measure stress, but found that it did not significantly affect scores. A placebo effect would imply that something causes the person taking the test to score artifically high. I am unaware of any such finding. There are also various reports that such things as stimulants or even music can temporarily boost test scores. Presumably these findings indicate an induced error in the positive direction, since no findings have reported permanent improvements in intelligence due to such factors.

 

[Mandrake]

I was not in the US when the book was published, but my reading seems to indicate it generated quite a deal of controversy.

There were a lot of people who were unaware of the findings reported in The Bell Curve. When I saw it, I was amazed to see that virtually every item in it had been previously reported in even greater detail. It was basically old science, but suddenly hit uninformed people in the face. Instead of trying to understand it, they reacted by writing ignorant missives in newspapers and magazines.

The Bell Curve can be divided into three parts:
1 - a detailed summary of history, research findings, and theories (including some of the unsound ones)
2 - an analysis of the National Longitudinal Study of Youth
3 - a discussion of the social and economic impacts that may relate to 1 and

If so (I’ve not read the book, so I can’t myself comment on how extensive, or impartial, the authors were on these topics), it would seem entirely appropriate for criticisms of the book to be made from any of these perspectives.

It is fair to argue item 3 forever. People have different views. Items 1 and 2 are matters of science and, if they are to be disputed, must be disputed on scientific grounds, not emotional ones. The basic rant that came from journalists was something to the effect that god made all men equal and, therefore, they must all be equally intelligent. Some people even cited the Declaration of Independence to demonstrate that blacks do not have a mean IQ of 85.

How clearly did the book present the scientific results?

Very clearly and very carefully. Most of items 1 and 2 were understated and supported by massive parallel findings. That is to say that to make a simple point, the authors cited findings from many independent studies, different countries, and different time frames, all reaching the same conclusion. In instances where there were conflicting findings, they clearly stated so. The book was written in a form that made it easy to read, by comparison to the typical Jensen textbooks.

E.g. how well did the tentative, provisional nature of all science come across?

The book was not intended to address all science and it did not. It was also not a general discussion of philosophy.

How was the study of intelligence (a branch of psychology) contextualised wrt neuroscience? Biology? Genetics? Other parts of psychology?

These issues were discussed to the degree that was possible one decade ago. There have been important findings since that time, especially in connection with laboratory measurements. Jensen, A. R. (1998). The g factor: The science of mental ability. Westport, CT: Praeger is a much better present day reference book.

Around the time of the publication of the book, some 50 people signed their names to an ad in a leading US newspaper, on race and intelligence (if memory serves).

The newspaper was the Wall Street Journal, Tuesday, December 13, 1994. The letter it printed was written by Linda Gottfredson, although it was not attributed to her. It was signed by, as I recall, 52 scholars.

IIRC, all but four were at US academic institutions.

Close. I just counted six. Detterman reported (in Intelligence) the details of how, why, and under what conditions that letter was written. There are very good reasons behind its final form and the people who signed it.

How many were active psychometricians? In the psychometric field of intelligence studies? At the time, how many professional psychometricians were there in US academic institutions?

To the best of my knowledge all of them, but I haven't looked up the lesser known people.

 

[Mandrake]

Does Jensen go on to say what those 'plausible reasons' are? Subsequent to when Jensen wrote this, what research has been done to establish what that distribution actually is? Why did Jensen add the caveat "within the range of ±2σ from the mean"?

This subject is discussed in much more detail in Bias in Mental Testing, Chapter 4. For example, Jensen wrote: "The simple fact is that a test unavoidably yields a near normal distribution when it is made up of (1) a large number of items, (2) a wide range of item difficulties, (3) no marked gaps in item difficulties, (4) a variety of content or forms, and (5) items that have a significant correlation with the sum of all other item scores, so as to ensure that each item in the test measures whatever the test as a whole measures." He goes on to point out that it would take a lot of effort to produce a test that is so screwed up that it would not produce a distribution that "departs at all radically from the normal."

 

[hitssquad]

Linda Gottfredson is a 'psychometrician'?

The newspaper was the Wall Street Journal, Tuesday, December 13, 1994. The letter it printed was written by Linda Gottfredson,

How many were active psychometricians? In the psychometric field of intelligence studies? At the time, how many professional psychometricians were there in US academic institutions?

To the best of my knowledge all of them, but I haven't looked up the lesser known people.

Gottfredson is a sociologist.

 

[selfAdjoint]

This subject is discussed in much more detail in Bias in Mental Testing, Chapter 4. For example, Jensen wrote: "The simple fact is that a test unavoidably yields a near normal distribution when it is made up of (1) a large number of items, (2) a wide range of item difficulties, (3) no marked gaps in item difficulties, (4) a variety of content or forms, and (5) items that have a significant correlation with the sum of all other item scores, so as to ensure that each item in the test measures whatever the test as a whole measures." He goes on to point out that it would take a lot of effort to produce a test that is so screwed up that it would not produce a distribution that "departs at all radically from the normal."

THis fact is behind the expectation of geneticists that the genetic component of g is produced by many genes. If there were just one, or a few, the different phenotypes would line up in discrete bunches, rather than a continuous distribution such as is found with sufficiently large populations.

 

[Mandrake]

Where to begin to untangle this?

My suggestion is a LOT of reading of top quality textbooks and peer reviewed papers.

For a start, sentences in English with comparatives ('most', 'higher') don't help IMHO; the reality of population groups is very rich, and teasing apart the influences of multiple factors calls for hard thought.

In a discussion forum, such as this one, people can and do provide quotes and links, but cannot reasonably be expected to produce the necessary volume of information that would be required to detail research procedures, error analysis, multiple confirmations, etc. Unfortunately there are people who jump to the conclusion that all of the research that has been done was either done with malice (to discredit stupid people) or with incompetence. Neither is true. In the case of SES comparisons, there have been a lot of studies reported by very capable scientists. The details of their comparisons are not secret. Virtually all comparisons are made with an attempt to remove as many variables as possible. For example, you will find that in The Bell Curve many of the W-B comparisons are done for cohorts with identical IQs. If you examine the past few issues of Intelligence you will find data comparing job status by IQ deciles (I previously quoted it and do not wish to look it up again, but it favors blacks). In the US blacks above the 40th percentile earn more than whites of equal IQ.

In the case of SES, the issue is complex primarily because IQ causes SES and not vice versa. This point may not be a happy finding for some people, but it is well documented in The _g_ Factor. A discussion of SES as it pertains to IQ may also be found in The Bell Curve, starting on page 286.

The issue of malnutrition is obviously a canard in the US. The IQ gap exists for all SES groups and we all know that we do not have a nation of malnourished blacks and properly nourished whites. The cause if the intelligence gap is largely genetic. The entire environmental contribution to IQ is in the range of 20-30%, but that is an overstatement that assumes no error. The strong evidence of IQ and more importantly of _g_ heritability continues to pile up even more now that we have brain imaging. We can actually see the huge overlap between identical twin brains and the sharp reduction when brains of DZ twins are imaged. We have path analysis and MZA studies, both of which show nearly identical values in the 70-75% range. We have detailed inbreeding depression studies that are based on heritability, with measurements matching predictions. We have adoption studies showing virtually no family influence, no correlation between adoptees and their adoptive siblings, and all of these kids turn out to have IQs in the range of their not-adopted peers and in the range predicted by knowing the IQs of the biological parents. We have nutrition studies, showing that even famine does not affect the IQs of children born to malnourished mothers. This is not Ethiopia.

Take Asians: my impression is that a significant proportion of Asians in the US are migrants or children of migrants; further, with some notable exceptions, they are primarily 'economic migrants' - they made conscious choices and effort to migrate.

So, just look at the IQ measurements of Asians who were born in Asia and who still live there. I don't see your point. That data is not missing. I previously commented that Asians in Asia test the same as Asians in the US. Asians adopted in the US and Europe turn out to have IQs higher (this is all statistical, as I hope you know) that their adoptive families.

Is There a Biological Basis for Race and Racial Differences?
By J. Philippe Rushton
Insight, May 28, 2001

I trust this is a piece of journalism and not a scientific paper; I sincerely hope that Rushton has done good science to back each of the points he makes here. In particular, I would expect that he has done studies in other countries, to demonstrate (for example) that he's not just reporting on some unique, US, human condition.

Rushton is a serious and respected scientist. He has published extensively in scientific journals and has conducted research internationally. He is a member of ISIR. Here is a list of his publications:
http://www.ssc.uwo.ca/psychology/faculty/rushton_pubs.htm

 

[selfAdjoint]

The issue of malnutrition is obviously a canard in the US.

But other physiological conditions have not been completely ruled out. Birth weight for example. Low birth weight is correlated with lower IQ scores in later life, and US blacks have systematically lower birth weights than whites. This gap persists into the more middle class communities too, so it isn't just a diet thing. Of course birth weight could have a large genetic component too, but it is at least ameliorable.

 

[Mandrake]

Gottfredson is a sociologist.

I assume you are familiar with her significant contributions to the literature of psychometrics. I discussed the academic requirements that I believe are appropriate for a psychometrician in a recent message here. She meets all of them. We have people who are outstanding psychometircians who hold degrees in various fields. Her work in recent years has been almost exclusively with respect to intelligence:
http://www.udel.edu/educ/gottfredson/reprints/

Her work with respect to education and teaching is focused on intelligence testing. She and Robert Gordon have produced the best papers available on the subject of the importance of intelligence in everyday life.

 

[hitssquad]

The shapes of the g distributions in human populations

Does Jensen go on to say what those 'plausible reasons' are?

• 16. Nothing of fundamental empirical or theoretical importance is revealed by the frequency distribution per se of the scores on any psychometric test composed of items. This is true regardless of whether we are dealing with raw scores or standardized scores or any otherwise transformed scores. Therefore, it would be trivial and pointless to review the empirical test literature regarding the form of the distribution of mental test scores.
  
  In a given population, the form of the distribution of raw scores (i.e., number of items passed) is entirely a function of three interrelated item characteristics: (1) the average probability of getting the correct answer by chance, i.e., by pure guessing, (2) the average level of difficulty of the items (as indexed by the percentage of the population that fails them), and (3) the average correlation between items. Item difficulty is completely under the test constructor's control. Score increments due to chance guessing are a function of the number and quality of the alternatives in multiple-choice items and the nature of the instructions to subjects regarding the penalty for guessing at the answer instead of omitting response when uncertain (e.g., total score based on number of right minus number of wrong answers). The item intercorrelations can be controlled to a considerable degree (but never completely) through item selection. Hence, in constructing a test it is possible, within broad limits, to produce almost any desired form of frequency distribution of the raw scores in a given population.
  
  If we have no basis for arguing that the obtained scores have true measurement properties in addition to merely having a rank-order correlation with the latent trait that they measure--and this seems to be typically the case for psychometric test scores--the precise form of the obtained score distribution is essentially arbitrary. The very most that we can say in this case is that (within the limits of measurement error) our test scores have some monotonic relation to whatever the test really "measures." If we could truly measure whatever latent variable, such as g, accounts for the variation in the obtained scores on an absolute scale (i.e., one having a true zero and additivity of scale intervals), the form of its population distribution could turn out to be quite different from that of the test scores we have actually obtained.
  
  Certain forms of distribution are simply more useful than others, psychometrically and statistically, and it is this consideration that mainly determines the form of the distribution test constructors decide to adopt. The aims of maximizing the statistical discriminability of scores throughout a fairly wide range of talent and of obtaining a fair degree of internal consistency reliability (i.e., interitem correlation) are what largely dictate item selection. The test scores that result under these conditions of item selection typically (and necessarily) have a symmetrical and more-or-less "bell-shaped" frequency distribution. It is not truly the normal (or Gaussian) curve, although it usually resembles it closely. By juggling item characteristics the test constructor can get a distribution that reasonably approximates the normal curve. Or the scores can be transformed mathematically to approximate a normal distribution. (Such "normalized" scores are obtained by converting the raw scores to ranks, then converting these to percentile ranks, and then, by reference to a table of the areas under the normal curve, converting these to normal deviates, i.e., normalized z scores.) The reason for thus normalizing a score distribution is not mainly theoretical, but statistical. The normal curve has certain mathematical properties that make it extremely useful in statistical analysis and interpretation.
  
  The argument is often made on theoretical grounds, however, that the main latent trait reflected by most complex cognitive tests--namely g--should be normally distributed in the general population. This argument, if accepted, justifies and indeed demands that IQs (or any other type of scores on any highly g-loaded tests) should be purposely scaled so that the form of their population distribution closely approximates the normal distribution. What can be said for this argument? There are three main facets:
  
  First, there is the argument by default: Unless there is some compelling reason to suppose that the form of the distribution of g is something other than normal, we might as well assume that it is normal, which is at least statistically convenient.
  
  Second, there is the argument from the Central-Limit Theorem in mathematical statistics, which essentially states that the distribution of a composite variable representing the additive effects of a number of independent elements (components, causes, or influences) rapidly approaches the normal distribution as the number of elements increases. This should be the case for g, to the extent that we can argue on various theoretical and empirical grounds that individual differences in g are the result of a great many different additive effects: for example, individual differences in the efficiency of a number of different cognitive processes, each of which is somewhat independently conditioned by polygenic inheritance interacting with a multitude of different environmental influences encountered throughout the course of development since the moment of conception. The population distribution of any variable with such multiple additive determinants, theoretically, should approximate the normal curve.
  
  Third, there is the argument by analogy with human characteristics that actually can be measured on an absolute scale, such as height, brain weight, neural conduction velocity, sensory acuity, choice reaction time, and digit span memory (i.e., the number of digits that can be recalled entirely correctly after one presentation on 50 percent of the trials). We may reasonably presume that individual differences in each of these variables has multiple determinants, just as in the case of g. Indeed, we find that in very large samples of the general population the distribution of each of these variables (measured on an absolute scale) approximates the normal curve. Marked deviations from the normal curve usually occur in the regions beyond ±2σ from the mean of the distribution. These deviations from normality can usually be explained in terms of certain rare genetic or environmental effects that override the multiple normal determinants of variation. This line of argument by analogy makes it quite plausible that g (or any other complexly determined trait) is normally distributed, but it cannot prove it. Also, the argument by analogy is weakened by the fact that not all complexly determined biological variables that can be measured on an absolute scale necessarily conform to the normal distribution. Age at death (beyond five years), for example, has a very negatively skewed distribution, because the mode is close to 75 years and the highest known limit of human longevity is about 113 years. (Below age five, the age of death is distributed as a so-called J curve, with the mode immediately after birth.)
  
  Fourth, the assumption of a normal distribution of g reveals a remarkable consistency between various population groups that show a given mean difference (in σ units) on highly g-loaded tests, such as IQ tests. By knowing the means and standard deviations of two population groups on such a measure, and by assuming that the latent trait, g, reflected by the measurements has a normal distribution in each group, one can make fairly accurate estimates of the percentages of each group that fall above or below some criterion that is not measured by any psychometric technique but is known to be correlated with g to some extent, such as number of years of education, occupational level, or as being judged by nonpsychometric criteria as mentally retarded or as intellectually gifted. Even though these percentages may vary widely from one criterion to another, when the percentages are transformed to normal deviates (obtained from tables of the normal curve), the differences between the groups' normal deviates on various g-related criteria show a considerable degree of constancy. This could not happen if the distribution of g were not approximately normal.
  
  Probably the best answer at present concerning the distribution of g is that, although we cannot determine it directly by any currently available means, it is a reasonable inference that it approximates the normal curve and there is no good reason for assuming that the distribution of g is not approximately normal, at least within the middle range of about four standard deviations. Most psychometricians implicitly work on the statistically advantageous assumption of normality, and no argument has yet come forth that it is theoretically implausible or adversely affects any practical uses of g-loaded tests. But the question is mainly of scientific interest, and a really satisfactory answer to it cannot come about through improved measurement techniques per se, but will become possible only as part and parcel of a comprehensive theory of the nature of g. If we have some theoretical conception of what the form of the distribution should be in a population with certain specified characteristics, we can use random samples from such a population to validate the scale we have devised to measure g. The distribution of obtained measurements should conform to the characteristics of the distribution dictated by theoretical considerations.

(Arthur Jensen. The g Factor. pp101-103.)

 

[Moonbear]

I just wanted to share that the work done by Haier and co-workers and mentioned by Mandrake has just been published in this month's NeuroImage journal. What struck me as I read the article was the lack of similarity between the younger college-aged subjects and the older middle-aged subjects. Though, as they note, comparisons between the two groups are difficult because different machines were used for the two groups (two different institutions were involved in recruiting the volunteers and performing the imaging). However, if that much variation could be due to just the machines being used, it doesn't seem like a very promising technology. I think they were doing a bit of hand-waving with that explanation. I don't see any comments section with this journal, but I'd be curious to see if any are published in the next issue from people with expertise in imaging. They did break down regions pretty specifically...this certainly is interesting technology if nothing else.

NeuroImage 2004, 23: 425-433
Structural brain variation and general intelligence
Haier RJ, Jung RE, Yeo RA, Head K, Alkire MT

Total brain volume accounts for about 16% of the variance in general
intelligence scores (IQ), but how volumes of specific regions-of-interest
(ROIs) relate to IQ is not known. We used voxel-based morphometry
(VBM) in two independent samples to identify substantial gray matter
(GM) correlates of IQ. Based on statistical conjunction of both samples
(N = 47; P < 0.05 corrected for multiple comparisons), more gray
matter is associated with higher IQ in discrete Brodmann areas (BA)
including frontal (BA 10, 46, 9), temporal (BA 21, 37, 22, 42), parietal
(BA 43 and 3), and occipital (BA 19) lobes and near BA 39 for white
matter (WM). These results underscore the distributed neural basis of
intelligence and suggest a developmental course for volume– IQ
relationships in adulthood.

However, in the very same issue, this article was also published raising questions about the validity of the way this technology is being used. This article, though, focuses primarily on flaws in group comparisons, not regression analyses, though does talk a bit about sample sizes. I'm not familiar enough with this technology to know if these same concerns would also relate to Haier's use of voxel-based morphometry, but it's always good to keep in mind potential caveats to any scientific study, especially ones utilizing new methods and applications.

NeuroImage 2004, 23: 17-20
Why voxel-based morphometric analysis should be used with great caution when characterizing group differences
C Davatzikos

A variety of voxel-based morphometric analysis methods have been
adopted by the neuroimaging community in the recent years. In this
commentary we describe why voxel-based statistics, which are
commonly used to construct statistical parametric maps, are very
limited in characterizing morphological differences between groups,
and why the effectiveness of voxel-based statistics is significantly biased
toward group differences that are highly localized in space and of
linear nature, whereas it is significantly reduced in cases with group
differences of similar or even higher magnitude, when these differences
are spatially complex and subtle. The complex and often subtle and
nonlinear ways in which various factors, such as age, sex, genotype and
disease, can affect brain morphology, suggest that alternative, unbiased
methods based on statistical learning theory might be able to better
quantify brain changes that are due to a variety of factors, especially
when relationships between brain networks, rather than individual
structures, and disease are examined.

 

[Moonbear]

But other physiological conditions have not been completely ruled out. Birth weight for example. Low birth weight is correlated with lower IQ scores in later life, and US blacks have systematically lower birth weights than whites. This gap persists into the more middle class communities too, so it isn't just a diet thing. Of course birth weight could have a large genetic component too, but it is at least ameliorable.

Age of the mother factors into this as well. Lower birth weights are associated with teen pregnancies, and teen mothers also are not likely to get adequate prenatal care.

Am J Prev Med. 2003 Oct;25(3):255-8.**
Correlates of unplanned and unwanted pregnancy among African-American female teens.
Crosby RA, DiClemente RJ, Wingood GM, Rose E, Lang D.

BACKGROUND: Evidence suggests that unplanned/unwanted pregnancy may be an important antecedent of negative birth outcomes, such as low birth weight. This study identified correlates of perceiving a current pregnancy as both unplanned and unwanted among unmarried African-American adolescents aged 14-20 years. METHODS: One hundred seventy pregnant adolescents were recruited during their first prenatal visit. Adolescents completed a face-to-face interview administered in private examination rooms. Adolescents also completed an in-depth self-administered survey. Measures were selected based on two potential influences: (1) relationships with boyfriends and (2) parent/family involvement. Age and parity were also assessed. Contingency table analyses were used to identify significant bivariate associations. Correlates achieving bivariate significance were entered into a forward stepwise logistic regression model. RESULTS: Pregnancy was reported as unplanned and unwanted by 51.2% of the study population. In a multivariate analysis, adolescents indicating lower levels of parental involvement were about twice as likely (adjusted odds ratio [AOR]=2.05; 95% confidence interval [CI], 1.1-3.9, p<0.03) to report that their pregnancy was unplanned and unwanted. Adolescents who already had a child (AOR=2.3; 95% CI, 1.3-5.7, p<0.009) and those younger than 18 years old (AOR=2.3; 95% CI, 1.1-4.5, p<0.02) were more than twice as likely to report that their pregnancy was unplanned and unwanted. A variable assessing whether each adolescent's current boyfriend conceived the pregnancy approached significance (AOR=2.33; 95% CI, 0.99-5.46, p=0.052). CONCLUSIONS: Findings provide initial evidence for specifically targeting intensified prenatal care programs to teens perceiving their pregnancy as unplanned and unwanted.

From introduction of above cited:

African-American adolescents were studied because the birth rate among African-American adolescent females (aged 15–19 years) is higher than that among all other ethnic/racial groups of U.S. females of the same age (85.3 per 1000 v 51.1 per 1000).[*8 ]

Their source for these statistics is:
Ventura SJ, Mathews TJ, Curtain SC. Declines in teenage birth rates, 1991–1998: update of national and state trends. Hyattsville MD: National Center for Health Statistics, 1999 [National Vital Statistics Report 47(26)]

 

[Evo]

I assume you are familiar with her significant contributions to the literature of psychometrics. I discussed the academic requirements that I believe are appropriate for a psychometrician in a recent message here.

No, according to you, sociologists know nothing about psychometrics. Go back and read your own posts. It doesn't matter what their areas of expertise are, according to you.

We have people who are outstanding psychometircians who hold degrees in various fields.

Not according to you, go back and read your own posts!

Amazing how if you agree with them, they meet "your" conditions and if you don't agree with them, they don't.

 

[Nereid]

_g_ and the INDIVIDUAL (not group)

Suppose I want to know my _g_ and how it may vary. I understand that I can take a test (e.g. an IQ test with high _g_ loading), some chronometric tests, or an EEG. From just one test – of any of these three kinds – what +/- number would my _g_ come with? What is the distribution (e.g. Gaussian)? How does each type of test vary wrt this +/-?

Since most observers do not have the laboratory devices and skills to measure intelligence via chronometric or electroencephalography techniques, the most common approach is to use an IQ test. As you probably know there are very many IQ tests, although only a few of them are used in most serious research. Of these the Raven's is most often cited in research programs. The WAIS versions are also widely used. For most tests, _g_ has to be extracted by weighting the subtest scores according to their _g_ loadings. Obviously, different tests will have different _g_ loadings, different subtest structures, and different associated errors. As with measuring physical phenomena (consider temperature) there are errors that can be identified in connection with many aspects of the measurement. Most of these errors are small. Ultimately the reliability coefficient is of central importance. Jensen: "The difference between the reliability coefficient and unity represents the proportion of the total variance of the measurements that is attributed to measurement error. ... In my laboratory we have been able to measure such variables as memory span, flicker-fusion frequency (a sensory threshold), and reaction time with reliability coefficients greater than .99. ... The reliability coefficients for multi-item tests of more complex mental processes, such as measured by typical IQ tests, are generally about .90 to .95. This is higher than the reliability of people's height and weight measured in a doctor's office! The reliability coefficients of blood pressure measurements, blood cholesterol level, and diagnosis based on chest X-rays are typically around .50." [The _g_ Factor, P. 50]

Thanks Mandrake.

However, I am still somewhat in the dark about _g_ and the individual; specifically, the +/- which psychometricians (in the field as well as in well appointed labs) assign to the results from just one test, of each kind (IQ, chronometric, EEG). Your comparisons with some tests done in doctors' offices is quite apt - and I'd like to explore this some more.

For now, I merely note that Jensen mentions 'measurement error'; my question certainly includes that, but is considerably broader.

The concept of TRUE SCORE is related.
Regressed true score = [(reliability coefficient) x (test score - mean score for population)] + (mean score for population)
This is obviously a hypothetical score that attempts to factor out measurement error. For a very detailed discussion of all things related to measurement error, see Jensen, A.R. (1980). Bias in mental testing. New York: Free Press. Jensen says that tests with reliability coefficients of less than .90 should (generally) not be used.

You're here talking about estimates of some kind of population mean (or other group statistic); I intend to get to this, but want to start with a good understanding of _g_ and the individual.

Since _g_ has to do with my brain, and I know all kinds of things affect the performance of ‘brain tasks’, I’m sure psychometricians have done extensive research into the effects of the following on one’s _g_, as estimated by one of the three kinds of _g_ tests:
- alertness, e.g. taking the test mid-morning after a good night’s sleep vs one taken at 2am
- drugs, e.g. caffeine, alcohol, anti-histimines; especially those which are known to affect reaction times and medications for mental conditions
- pain, esp headaches
- general wellness/fitness, e.g. fever, top physical form, hunger,
- brain affecting illnesses or conditions, e.g. epilepsy, depression, Alzheimers, tumour, physical injury, PTSS
- mood
- age

How do estimates of _g_ vary for each of these classes of factors?

I don't have a source at hand with a ready answer. When an IQ test is given, it is the responsibility of the test administrator to determine that the person taking the test is fully alert and not encumbered by factors that would render the test inaccurate. Some tests, such as the WAIS are age adjusted.

In doing good science, of course we expect that the administration of tests be done so as to eliminate or control for extraneous factors which may influence the effects we seek to collect data on. My question here is only with research results on the measured size and nature of each of the above effects (and any others which psychometricians have discovered) on an estimate of _g_, for each class of test (IQ, e.g. WAIS; chronometrics, EEG); preferably expressed as a range that could be expected in the result of just a single test.

 

[Mandrake]

No, according to you, sociologists know nothing about psychometrics. Go back and read your own posts. It doesn't matter what their areas of expertise are, according to you.

Please quote me instead of commenting incorrectly. I was critical of the sociologist you referenced because she has not conducted psychometric research and she has not published in psychometric journals. She is not a participant in the peer review of psychometric papers. If you examine the list of publications you presented you will see that she is interested in other subjects.

Not according to you, go back and read your own posts!

I suggest that you read them before making such comments. On 8-25-2004 I wrote:

There is no requirement that a psychometrician hold a particular university degree. This is especially so because the field of psychometrics is quite removed from much of psychology and makes particularly heavy demands on statistical knowledge and laboratory research. The thing that distinguishes a psychometrician (or other specialist) is his devotion to the subject at hand, years of study, years of research, and participation in the publication of peer reviewed research. The women you listed are not qualified to peer review psychometric research.

 

[Nereid]

Nereid:

There are various papers that have examined the conditions of test taking. The range of things considered includes, for example, stress. Jensen used pulse rate to measure stress, but found that it did not significantly affect scores. A placebo effect would imply that something causes the person taking the test to score artifically high. I am unaware of any such finding. There are also various reports that such things as stimulants or even music can temporarily boost test scores. Presumably these findings indicate an induced error in the positive direction, since no findings have reported permanent improvements in intelligence due to such factors.

Thanks.

Perhaps I haven't been clear enough; the effects of test conditions, pre-test expectations, etc on estimates of _g_ wouldn't necessarily be to boost an estimate, they could depress it (e.g. the 'white coat effect' makes some people appear less healthy than they 'really' are).

The music and stress factors look interesting; how much research has been done on such factors when tests are done in the field (vs in well appointed labs)? How large was the 'music' effect?

Assuming that psychometrics does cover things such as personality, aptitudes, and interests, and that psychometricians active in these fields have been as successful as Jensen appears to have been in intelligence, to what extent have _g_ psychometricians employed well constructed personality, aptitudes, and interests tests in conjunction with their _g_ tests?

 

[Nereid]

My suggestion is a LOT of reading of top quality textbooks and peer reviewed papers.

For a start, sentences in English with comparatives ('most', 'higher') don't help IMHO; the reality of population groups is very rich, and teasing apart the influences of multiple factors calls for hard thought.

In a discussion forum, such as this one, people can and do provide quotes and links, but cannot reasonably be expected to produce the necessary volume of information that would be required to detail research procedures, error analysis, multiple confirmations, etc. Unfortunately there are people who jump to the conclusion that all of the research that has been done was either done with malice (to discredit stupid people) or with incompetence. Neither is true. In the case of SES comparisons, there have been a lot of studies reported by very capable scientists. The details of their comparisons are not secret. Virtually all comparisons are made with an attempt to remove as many variables as possible. For example, you will find that in The Bell Curve many of the W-B comparisons are done for cohorts with identical IQs. If you examine the past few issues of Intelligence you will find data comparing job status by IQ deciles (I previously quoted it and do not wish to look it up again, but it favors blacks). In the US blacks above the 40th percentile earn more than whites of equal IQ.

In the case of SES, the issue is complex primarily because IQ causes SES and not vice versa. This point may not be a happy finding for some people, but it is well documented in The _g_ Factor. A discussion of SES as it pertains to IQ may also be found in The Bell Curve, starting on page 286.

The issue of malnutrition is obviously a canard in the US. The IQ gap exists for all SES groups and we all know that we do not have a nation of malnourished blacks and properly nourished whites. The cause if the intelligence gap is largely genetic. The entire environmental contribution to IQ is in the range of 20-30%, but that is an overstatement that assumes no error. The strong evidence of IQ and more importantly of _g_ heritability continues to pile up even more now that we have brain imaging. We can actually see the huge overlap between identical twin brains and the sharp reduction when brains of DZ twins are imaged. We have path analysis and MZA studies, both of which show nearly identical values in the 70-75% range. We have detailed inbreeding depression studies that are based on heritability, with measurements matching predictions. We have adoption studies showing virtually no family influence, no correlation between adoptees and their adoptive siblings, and all of these kids turn out to have IQs in the range of their not-adopted peers and in the range predicted by knowing the IQs of the biological parents. We have nutrition studies, showing that even famine does not affect the IQs of children born to malnourished mothers. This is not Ethiopia.


So, just look at the IQ measurements of Asians who were born in Asia and who still live there. I don't see your point. That data is not missing. I previously commented that Asians in Asia test the same as Asians in the US. Asians adopted in the US and Europe turn out to have IQs higher (this is all statistical, as I hope you know) that their adoptive families.


Rushton is a serious and respected scientist. He has published extensively in scientific journals and has conducted research internationally. He is a member of ISIR. Here is a list of his publications:
http://www.ssc.uwo.ca/psychology/faculty/rushton_pubs.htm

I want to return to this, but only after I've understood much of the basics better first.

 

[Nereid]

Does Jensen go on to say what those 'plausible reasons' are? Subsequent to when Jensen wrote this, what research has been done to establish what that distribution actually is? Why did Jensen add the caveat "within the range of ±2σ from the mean"?

This subject is discussed in much more detail in Bias in Mental Testing, Chapter 4. For example, Jensen wrote: "The simple fact is that a test unavoidably yields a near normal distribution when it is made up of (1) a large number of items, (2) a wide range of item difficulties, (3) no marked gaps in item difficulties, (4) a variety of content or forms, and (5) items that have a significant correlation with the sum of all other item scores, so as to ensure that each item in the test measures whatever the test as a whole measures." He goes on to point out that it would take a lot of effort to produce a test that is so screwed up that it would not produce a distribution that "departs at all radically from the normal."

THis fact is behind the expectation of geneticists that the genetic component of g is produced by many genes. If there were just one, or a few, the different phenotypes would line up in discrete bunches, rather than a continuous distribution such as is found with sufficiently large populations.

This doesn't come as any surprise; unfortunately, it surely makes doing good science in this field very difficult. For example, non-gaussianity is usually a clear sign that there is at least one systematic effect in play, but the converse is most definitely not true (I'm sure we can all give boatloads of examples where researchers fell into the trap of - unconsciously? - equating gaussianity with absence of confounding systematic effects). Further, if tests invariably yield gaussian distributions, hypotheses which predict non-gaussian ones will have a very hard time getting their 'day in court'.

Any answer to the "within the range of ±2σ from the mean" question? Edit: never mind; hitssquad's post (Jensen quote) addresses this nicely.

 

[Nereid]

Probably the best answer at present concerning the distribution of g is that, although we cannot determine it directly by any currently available means, it is a reasonable inference that it approximates the normal curve and there is no good reason for assuming that the distribution of g is not approximately normal, at least within the middle range of about four standard deviations. Most psychometricians implicitly work on the statistically advantageous assumption of normality, and no argument has yet come forth that it is theoretically implausible or adversely affects any practical uses of g-loaded tests. But the question is mainly of scientific interest, and a really satisfactory answer to it cannot come about through improved measurement techniques per se, but will become possible only as part and parcel of a comprehensive theory of the nature of g. If we have some theoretical conception of what the form of the distribution should be in a population with certain specified characteristics, we can use random samples from such a population to validate the scale we have devised to measure g. The distribution of obtained measurements should conform to the characteristics of the distribution dictated by theoretical considerations.[/list](Arthur Jensen. The g Factor. pp101-103.)

Thanks hitssquad; the whole quote is very illuminating, but I found this last para to be of particular value; we'll surely be returning to it.

 

[Nereid]

birth order, birth spacing

Age of the mother factors into this as well. Lower birth weights are associated with teen pregnancies, and teen mothers also are not likely to get adequate prenatal care.

Some time ago I think I read that a long-standing problem in US education had moved a giant step forward towards resolution - falling average SAT scores (or something like that). IIRC, the resolution was birth order and/or birth spacing - first-borns do differently (better?) on SATs than second-borns, who in turn ... Apparently the childhood environment wrt siblings has a considerable effect.

Can someone tell me if my poor old memory is even approximately right?

As SAT scores correlate well with _g_ (or, as I'm beginning to learn to say, "SAT tests have a high _g_ loading"), does it follow that one's _g_ is partly determined by one's birth order and/or age difference of one's siblings? What have US intelligence psychometricians found here? What have non-intelligence US psychometricians found (e.g. wrt personality, aptitudes, interests)?

 

[Evo]

Please quote me instead of commenting incorrectly. I was critical of the sociologist you referenced because she has not conducted psychometric research and she has not published in psychometric journals. She is not a participant in the peer review of psychometric papers. If you examine the list of publications you presented you will see that she is interested in other subjects.

Why would she have to be a psychometrician when she was arguing genetics? She's an expert in genetics, which is what she was discussing. Neither Hernstein or Murray are geneticists, which is why they are not qualified to make the assumoptions they did about genetics.

I suggest that you read them before making such comments. On 8-25-2004 I wrote:

There is no requirement that a psychometrician hold a particular university degree. This is especially so because the field of psychometrics is quite removed from much of psychology and makes particularly heavy demands on statistical knowledge and laboratory research. The thing that distinguishes a psychometrician (or other specialist) is his devotion to the subject at hand, years of study, years of research, and participation in the publication of peer reviewed research. The women you listed are not qualified to peer review psychometric research.

Which is exactly why your argument made no sense, they were both experts in genetics, which is what they were discussing. According to you, a person cannot be qualified in anything but their degree.

 

[Nereid]

Thanks hitssquad.

Reaction time is composed of decision time and motor time. Caffeine may affect motor time, but AFAIK it has not been shown to affect decision time. Other drugs such as Bacopa Monniera (AKA the ayervedic herb Brahmi) may affect decision time, as Bacopa itself has been shown to affect the related variable inspection time and has been shown to increase g (as measured by standard psychometric batteries in a controlled study).

There are huge 'problems' in sports wrt 'performance-enhancing' drugs. I would expect that for some sports (e.g. fencing, pingpong; motorcar racing?), a drug which could improve either decision time or motor time (or both!) would be of considerable interest. At the least I would expect that the drug testers in sports would have a list of drugs known to improve RT; when I get time I'll do some googling.

What other detailed studies have been done into the effects of drugs on measured _g_?

IIRC, some elite athletes get an 'unfair' advantage over their competitors because they have rare genetic mutations, e.g. very high production of red blood cells; I also remember reading that elite fencers are 'lefties', because lefties have inherently faster RT. What is known about the incidence of genetic mutations which markedly affect RT (in either direction)?

g is known to smoothly drop in adults with age, as I have mentioned many times on Physics Forums and which I mentioned many times as a raison d'être for anti-senescence efforts.

This page has a nice graph of what is likely happening to you as you read this:
http://hiqnews.megafoundation.org/Definition_of_IQ.html

Age-related cognitive decline is also being discussed over at the Children of Millennium forums:

• As you age beyond the age of 18, your physical brain-decay (glycoxidation; amyloid beta build-up; mitochondrial damage; DNA damage) can be clearly watched in slow motion in the form of your raw scores predictably dropping point by point, year by year.
  
  IQ scores on IQ tests correct for this post-18 brain rot, and you are given a same-age-peers curve-graded IQ score (in addition to being allowed to cheat with a massively-larger vocabulary than your g would otherwise imply -- boosted vocabulary subtest scores via vocab cheating by oldsters on the Weschler averages +.80 S.D. {and that's not counting the other free full-scale IQ points they get because their peers have physical brain rot}, according to a brand new study {see at the bottom of this post Verhaeghen; see also MacLullich, et al}).

Do all adults age at the same rate (wrt _g_)? What does research show wrt variations in the decline of _g_ with age, e.g. men vs women, menopause, those who use their intelligence vs those who don't, the 'old oldies' (those who remain in good physical and mental health well into their 70s, 80s, and 90s) vs everyone else?

 

[Nereid]

brains, IQ, etc

{extract; my emphasis}
Spearman's g is this number for the first principal component of just about every IQ test and surrogate ever invented. It is enormously stable and correlated with things like the SAT, the Armed Forces tests, and so on. It also has physical correlates like measured reaction time and volume of gray matter in the prefrontal cortex.

I came across a recent study that looks at intelligence and specific areas or patches of gray matter. They did not do a correlation as such, but it does look like they are locating those areas related to g, as well as other factors like fatty tissue around axons, glucose uptake, etc. But at least the IQ vs. Brain size is getting narrowed down to IQ and specific brain reqions.

" While gray matter amounts are vital to intelligence levels, the researchers were surprised to find that only about 6 percent of all the gray matter in the brain appears related to IQ.[/color]" {From Human Intelligence Determined By Volume And Location Of Gray Matter Tissue In Brain Source: University Of California - Irvine Date: 2004-07-20}

[...]

Some brain and head size related factoids:

The average female brain is smaller than the average male brain. This is true, even after the size difference is corrected for relative differences in body size. The average male brain is about 12.5% heavier.

The average female brain has more neurons per unit volume than the average male brain (about 11%).

The average brain and head size is smaller for blacks than for whites.

The number of neurons in the brain is fixed by age 4, but the brain size to intelligence correlation is weak at age 4. By age 7 there is a significant within family effect. Miller argues that this is consistent with his myelination hypothesis because myelination of the brain is not significant at age 4, but is much more so at 7 and continues through adolescence.

The correlation between body size and brain size in adults is between .20 and .25.

The correlation between head size and IQ ranges from .10 to .25 (various studies), with a mean of .15.

The correlation (one study only) between head size and _g_ is .30.

The correlation between brain size, as measured by MRI, and IQ is .40 (corrected for body size).

Moonbear has two highly relevant posts too, they are too large to copy here, so
link1, link2 .

" The relation between brain size and intelligence has been shown by dozens of studies, including state-of-the-art magnetic resonance imaging. Orientals average 1 cubic inch more brain matter than Whites, and Whites average a very large 5 cubic inches more than Blacks. Since one cubic inch of brain matter contains millions of brain cells and hundreds of millions of nerve connections, brain size differences help to explain why the races differ in IQ.[/color]" {from Is There a Biological Basis for Race and Racial Differences? By J. Philippe Rushton Insight, May 28, 2001}

It would seem that this topic isn't quite as clear-cut as SelfAdjoint's post would imply. For example, it seems that Rushton's work has been seriously questioned, and that he himself has found much smaller differences in his later work than he reported originally (http://www.mugu.com/cgi-bin/Upstream/People/Rushton/rushton-peters.html ).

The research quoted by both Moonbear and Mandrake seem to show that 'intelligence' isn't particularly well localised in the brain. Further, the sex differences would seem to suggest that brain volume, in whole or in part, should not correlate with intelligence.

But let's assume that they do. Then we find some really interesting results, for example:
- the variation in brain volume within a population group is far larger than that between population groups (so, naively, you would expect there to be little IQ difference between population group averages)
- just as skin colour is an adaptation to UV, so aspects of head size and shape are adaptations to local climates - e.g. arctic vs tropical (so, naively, you might expect that any IQ differences would correlate with climate adaptation, if only weakly)
- the prefrontal cortex comprises ~12.5% of human brains, and ~10.6% of baboon brains (source). If the brain volume variations claimed by Rushton are due purely to IQ, which is found only in the prefrontal cortex, then variations in human prefrontal cortex volume (as a % of total brain volume) should be far in excess of the difference between the average human and the average baboon (5 cubic inches (~82 cm³) is approx 6% of the total human brain volume) - so, naively you would expect that brain volume should have nothing to do with IQ (the regions which are responsible for intelligence are far too small - as a % of the total brain - for anything but huge variations in these to be responsible for the observed brain volume differences among humans).

 

[Mandrake]

It seems that there may be more than one 'APA report'; could someone please tell me whether http://www.apa.org/releases/intell.html is the one we're discussing?

The material you quoted was a press release. The comments I made were directed at the following:
Report of a Task Force established by the Board of Scientific Affairs of the American Psychological Association
Released August 7, 1995
A slighted edited version was published in the American Psychologist, Feb 1996.

 

[Waterdog]

The newspaper was the Wall Street Journal, Tuesday, December 13, 1994. The letter it printed was written by Linda Gottfredson, although it was not attributed to her. It was signed by, as I recall, 52 scholars.

Close. I just counted six. Detterman reported (in Intelligence) the details of how, why, and under what conditions that letter was written. There are very good reasons behind its final form and the people who signed it.

This was the advert in which Jensen, Rushton, and Lynn all acknowledged that the IQ gap between blacks and whites in the US was more likely the result of environmental influences than genetics. Quote:

"There is no definitive answer to why IQ bell curves differ acrossracial-ethnic groups. The reasons for these IQ differences between groups may be markedly different from the reasons for why individuals differ among themselves within any particular group (whites or blacks orAsians). In fact, it is wrong to assume, as many do, that the reason whysome individuals in a population have high IQs but others have low IQs must be the same reason why some populations contain more such high (or low) IQ individuals than others. Most experts believe that environment is important in pushing the bell curves apart, but that genetics could be involved too."

<http://www.lrainc.com/swtaboo/taboos/wsj_main.html>

That is the statement that Lynn, Rushton and Jensen all signed: "Most experts believe that environment is important in pushing the bell curves apart, but that genetics could be involved too." Genetics could be involved too. It is hilarious that most of the race-and-IQ enthusiasts seem not even to have read all of the documents that they cite in support of their position. If it has really been scientifically proven that IQ is mostly a result of genetics, as the highly vocal crowd of internet-message-board enthusiasts never tires of claiming, then why did all of those scientists sign a public statement to the contrary? Note that this statement also warns against procedures such as applying the results of twin studies conducted within a country to international IQ differences, another favorite ploy of the ever-inventive race-and-IQ crowd.