A reader (Bob Williams), contacted me as he had a rather lengthy set of comments he wished to make in response to the guest blog post by Scott Barry Kaufman on "The Flynn Effect and IQ Disparities Among Races, Ethnicities, and Nations: Are There Common Links." His comments would not fit in the small "comment" feature of the blog. So, reproduced below are Bob Williams comments "as is" (and as extracted from the body of an email sent to the blogmaster).
Bob Williams states:
I would like to offer some detailed comments:
Literacy involves the ability to write, read, and comprehend information of varying levels of complexity. It is estimated that there are 774 million illiterate adults in the world, 65% whom are women (UNESCO Intsistute for Statistics, 2007). In the United States alone, 5% of the adult population is completely nonliterate (Kirsch, Jungeblut, Jenkins, & Kolstad, 1993).
Literacy is usually referenced as a binary condition, not as degrees. But the discussion here suggests that it is to be both considered as binary and as a continuous variable. Literacy should follow the same path as education in its relationship to IQ, namely that educational achievement flows from IQ and not vice versa. I will discuss this further below.
One study showed that the IQ and literacy scores of Blacks increased in parallel from 1980 to 2000 (Dickens & Flynn, 2006). Murray showed essentially no change over the period 1980 to 1990. C. Murray/Intelligence, 35 (2007) 305–318
The importance of being able to read for performance on an IQ test cannot be understated. IQ tests must be used as intended. If the test involves reading, the testee must be able to read. Professionals are unlikely to violate this common sense requirement. There are, however, high quality IQ tests that do not require reading, either of instructions or test items. The Raven's set is the best known example, but other tests designed for children and the illiterate are also available (the Kohs and DAM are examples).
Instead of measuring ‘intelligence' in a nonliterate test-taker, the test is measuring that person's inability to read. This happens only when the test is administered by someone who cannot follow instructions or has an agenda and wishes to collect bad measurements. These findings have led some researchers to propose that such IQ gaps found across ethnicities, races, and nationalities suggests a difference in innate brain capacity (see Lynn & Vanhanen, 2006).
Besides the findings you mentioned, there is a wealth of data showing that intelligence is determined by biological, not social factors. Much of The g Factor was devoted to demonstrating this. The heritability of IQ consistently shows up above 80% in adults and can be determined by diverse methods (Falconer's formula, twin correlation, path analysis, and by subtracting the environmental effect from 1.0).
If increasing literacy were really explaining a number of seemingly different IQ trends, then you would expect to see a few things. First, within a population you should expect increased education of literacy skills to be associated with an increase in the average IQ of that population. One of the important studies of the Flynn Effect (FE) was conducted by Nettelbeck and Wilson [Intelligence 32 (2004) 85–93]. The controls in their study were extraordinary. With virtually every variable held constant, even to the people who did the measurements and their instrumentation, they found a FE, but no change in inspection times (IT). I asked Nettlebeck if there could have been any nutritional or social changes between his two study groups (separated by 20 years). He said emphatically that there was not.
There is also a problem with focusing on literacy because it requires time to develop or not develop, yet IQ can be measured in young children and even infants. Those early measurements are predictive of adult IQ and educational attainment. [See Fagan's work with infants.] And this:
"Within the United States, the mean Black–White group difference in IQ has not changed significantly over the past 100 years despite significant improvements in the conditions of Black Americans. The same magnitude of difference is observed as early as age 2 1/2 years." [Rushton, J.P. and Jensen, A.R. (2005). Thirty Years of Research on Race Differences in Cognitive Ability. Psychology, Public Policy, and Law, Vol. 11, No. 2, 235-294.]
Second, IQ gains should be most pronounced in the lower half of the IQ bell curve since this is the section of the population that prior to the education would have obtained relatively lower scores due to their inability to comprehend the intelligence test's instructions. As you know, this has been reported in some studies and not found in others. As Must and Must have shown, the FE is not invariant, at least in Estonia. There is little reason to expect that it is invariant elsewhere. The Nettelbeck study, for example, did not report this finding, nor would it have been expected in such an otherwise homogeneous group, but the FE was still present.
If all these predictions hold up, there would be support for the notion that secular IQ gains and race differences are not different phenomena but have a common origin in literacy. Not likely. The differences between racial groups are g loaded. Here is the short section 5 (conclusions) from a recent paper by Rushton and Jensen:
"Heritable g is at the core of the debate over how much the mean Black–White gap in IQ and school achievement is due to the genes rather than to the environment, and therefore, how much it can be expected to narrow. While g and genetic estimates correlate significantly positively with Black–White differences 0.61 and 0.48 (P < 0.001), they correlate significantly negatively (or not at all) with the secular gains (r = -0.33; P < 0.001) and 0.13 (ns). Similarly, g loadings and heritabilities from the items of the Raven Matrices correlate significantly positively with each other and with Black–White differences (mean r = 0.74, P < 0.01). Although the secular gains are on g-loaded tests (such as the Wechsler), they are negatively correlated with the most g-loaded components of those tests. Tests lose their g loadedness over time as the result of training, retesting, and familiarity (te Nijenhuis et al., 2007).
Some issues, however, remain to be resolved. For example, Lynn (2009) found a secular rise in the Developmental Quotients of infants in the first two years of life, which he suggested was due to improved pre-natal and early post-natal nutrition. He supported his conjecture by pointing to equivalent gains in birth weight, stature, and brain size, and the correlation of these variables with later IQ. If it becomes possible to disentangle environmental factors that do affect g, from the environmental factors that do not affect g, the negative correlation between g and secular gains may increase from -0.33 to nearer - 1.00.
Predictions about the Black–White IQ gap narrowing due to the secular rise is based on faith rather than evidence. There is no more reason to expect Black–White differences in IQ to narrow as a result of the secular rise in IQ than to expect male–female differences in height to narrow as a result of the secular rise in height. The (mostly heritable) cause of the one is not the (mostly environmental) cause of the other. From the present perspective, the Flynn Effect (the secular rise in IQ) is not a Jensen Effect (because it does not occur on g).."
[Rushton and Jensen, Intelligence, Volume 38, Issue 2, March-April 2010, Pages 213-219 ]
To test these predictions, Marks looked at samples representative of whole populations (rather than individuals), and used ecological methods to calculate statistical associations between IQ and literacy rates across different countries.
The problem is that he is looking at a consequence of intelligence that has little, if anything, to do with causation. But Marks made it clear that he thinks he is looking at the cause and not the effect. Here is a quote from the paper you referenced:
"Secondly, the differences in IQ scores that exist today between different populations are artifacts of large, confounding literacy differences that exist between these populations. Thirdly, white-black differences in IQ scores are caused by literacy differences between these racial groups. Comparing the average IQ test scores of racial groups or populations without controlling for literacy is illegitimate."
None of the above is correct. The population group differences have been shown to be highly heritable biological differences and are not the result of social factors. The race difference shows up in testing at age 2-1/2 and in the robust correlate, head size, at birth. Literacy is not a birth nor an early childhood parameter. Marks is off the mark.
It should also be noted that Mark's findings only speak to populations (not individuals) and do not say much about causation. Marks made a very clear comment (see above) about his view of causation. It seems to me that anyone wishing to show that literacy is a factor in IQ test scores would use structural equation modeling to evaluate the obvious alternatives, with the goal of showing the one that best fits available data.
variables that affect both literacy and IQ. Still, the result that population level literacy changes with population IQ is suggestive that increased literacy is causing increased IQ.
No, it is not. It instead suggests that IQ is accompanied by literacy. It is well known that years of education correlates positively with IQ and to such an extent that educational attainment can be used as a rough proxy for IQ. But people do not become more intelligent as they spend more years in school. Education is not intelligence, but rather acts as a tool set for the application of intelligence. IQ tends to be stable over most of our lives, after about age 6, but even before that age, it is highly predictive of later measurements and is predictive of educational achievement. For a second time let me suggest Fagan's paper: The prediction, from infancy, of adult IQ and achievement Intelligence, Volume 35, Issue 3, May-June 2007, Pages 225-231
Joseph F. Fagan, Cynthia R. Holland, Karyn Wheeler
Marks did just that by scanning the literature for datasets containing test estimates for populations of groups taking both the Armed Forces Qualifications Test and tests of literacy. One study on nine groups of soldiers differing in job and reading ability found a correlation of .96 between the Armed Forces Qualifications Test and reading achievement (Sticht, Caylor, Kern, & Fox, 1972).
Yes, because reading is highly g loaded. Jensen devoted a long discussion to this in The g Factor (pages 279-282). Reading does not boost g; it is g that makes reading comprehension high or low as a function of individual intelligence.
Another study obtained reading scores for 17-year olds for those same ethnic groups and dates and found a correlation of .997 between reading scores and Armed Forces Qualifications Test scores (Campbell et al., 2000). This nearly perfect correlation was based on six pairs of data points from six independent population samples evaluated by two separate groups of investigators.
These "nearly perfect correlations" are the subject of a very well done paper in Intelligence: The issue of power in the identification of “g” with lower-order factors, Pages 336-344, Dora Matzke, Conor V. Dolan, Dylan Molenaar
They commented in the conclusions section, "Our examination of published studies revealed that most of our case studies, which reported a perfect correlation between g and a lower-order factor, were underpowered, with power coefficient rarely exceeding 0.3." The comment above strikes me as precisely the kind of underpowered case Matzke et al. were addressing.
"On the basis of the studies summarized here, there can be little doubt that the Armed Forces Qualifications Test is a measure of literacy."
There is plenty of evidence presented in The Bell Curve that the AFQT is heavily g loaded. There is one g and it doesn't matter how it is measured, the g is still the same thing. IQ tests get virtually all of their validity by acting as a proxy for g. If g is measured by the Raven's, the WJ-III, or a battery of reaction time tests, the thing at the root is the same g. The entire argument of literacy should be immediately questioned when it is used in connection with the FE, since it has been clearly shown that the largest secular gains have been observed in abstract test items. The Raven's shows this quite well. I once had about a one hour chat with John Raven and asked him if he thought the gains in the Raven's tests were g loaded. He only replied that the scores were increasing. When I asked Jim Flynn if the FE gains were g loaded, he gave me his usual historical (Rabbit and Hound story) answer, then said "I don't know." Most studies that have attempted to determine the g loading of the FE have found no loading. Rushton and Jensen have made this point as have Must and Must and others.
In The g Factor, Jensen showed that a positive FE could be shown for abstract test items, while a negative FE could be shown for scholastic items (see page 322). It is more than odd that the latter (literacy) would contribute in the opposite direction along with abstract items.
Potential research avenues to be explored.
The Marks study suggests a crucial environmental factor is literacy. Marks has tried to make that point, but he has skipped the necessary details and has ignored the strong biological effects as well as the early age factor that I have previously pointed out. If this is so, then interventions that increase literacy will also narrow the IQ gap found between different races and nationalities. To the best of my knowledge, no social or training factor has ever been shown to cause a real and permanent increase in g. Training increases the Spearman s loading and lowers the g loading. It does not improve real intelligence (g).
This latest research on the environmental effects of nutrition (Colom et al., 2005, but see Flynn, 2009), disease, literacy, and more on both the rise in IQ and ethnic, racial, and national disparities in IQ point to the importance of the environment for developing intelligence as well as the importance for researchers to be very careful when they use intelligence test performance (especially verbal tests) to make inferences about hereditary differences between different ethnic groups and nationalities.
Environmental factors account for about 17% of the variance in intelligence in developed nations. It must be more in undeveloped nations, but environmental factors act through biological mechanisms (toxins, disease, etc., not social factors) and, so far, these have all been negative. Ethnic group IQs are highly heritable as is evident by such displays as regression to the mean (for the group), inbreeding depression, twin studies, and adoption studies. Marks failed to explain the finding that adopted children reach adult IQs that are as predicted by their biological peers and those IQs are uncorrelated with their adoptive siblings. You would expect similar literacy within a family, but these studies show that only heritability was able to explain the outcomes.
Bob Williams
Technorati Tags: psychology, educational psychology, intelligence, Flynn effect, race and IQ, ethnicity and IQ, national IQs, guest blog post, Bob Williams, Scott Barry Kaufman
Bob Williams states:
I would like to offer some detailed comments:
Literacy involves the ability to write, read, and comprehend information of varying levels of complexity. It is estimated that there are 774 million illiterate adults in the world, 65% whom are women (UNESCO Intsistute for Statistics, 2007). In the United States alone, 5% of the adult population is completely nonliterate (Kirsch, Jungeblut, Jenkins, & Kolstad, 1993).
Literacy is usually referenced as a binary condition, not as degrees. But the discussion here suggests that it is to be both considered as binary and as a continuous variable. Literacy should follow the same path as education in its relationship to IQ, namely that educational achievement flows from IQ and not vice versa. I will discuss this further below.
One study showed that the IQ and literacy scores of Blacks increased in parallel from 1980 to 2000 (Dickens & Flynn, 2006). Murray showed essentially no change over the period 1980 to 1990. C. Murray/Intelligence, 35 (2007) 305–318
The importance of being able to read for performance on an IQ test cannot be understated. IQ tests must be used as intended. If the test involves reading, the testee must be able to read. Professionals are unlikely to violate this common sense requirement. There are, however, high quality IQ tests that do not require reading, either of instructions or test items. The Raven's set is the best known example, but other tests designed for children and the illiterate are also available (the Kohs and DAM are examples).
Instead of measuring ‘intelligence' in a nonliterate test-taker, the test is measuring that person's inability to read. This happens only when the test is administered by someone who cannot follow instructions or has an agenda and wishes to collect bad measurements. These findings have led some researchers to propose that such IQ gaps found across ethnicities, races, and nationalities suggests a difference in innate brain capacity (see Lynn & Vanhanen, 2006).
Besides the findings you mentioned, there is a wealth of data showing that intelligence is determined by biological, not social factors. Much of The g Factor was devoted to demonstrating this. The heritability of IQ consistently shows up above 80% in adults and can be determined by diverse methods (Falconer's formula, twin correlation, path analysis, and by subtracting the environmental effect from 1.0).
If increasing literacy were really explaining a number of seemingly different IQ trends, then you would expect to see a few things. First, within a population you should expect increased education of literacy skills to be associated with an increase in the average IQ of that population. One of the important studies of the Flynn Effect (FE) was conducted by Nettelbeck and Wilson [Intelligence 32 (2004) 85–93]. The controls in their study were extraordinary. With virtually every variable held constant, even to the people who did the measurements and their instrumentation, they found a FE, but no change in inspection times (IT). I asked Nettlebeck if there could have been any nutritional or social changes between his two study groups (separated by 20 years). He said emphatically that there was not.
There is also a problem with focusing on literacy because it requires time to develop or not develop, yet IQ can be measured in young children and even infants. Those early measurements are predictive of adult IQ and educational attainment. [See Fagan's work with infants.] And this:
"Within the United States, the mean Black–White group difference in IQ has not changed significantly over the past 100 years despite significant improvements in the conditions of Black Americans. The same magnitude of difference is observed as early as age 2 1/2 years." [Rushton, J.P. and Jensen, A.R. (2005). Thirty Years of Research on Race Differences in Cognitive Ability. Psychology, Public Policy, and Law, Vol. 11, No. 2, 235-294.]
Second, IQ gains should be most pronounced in the lower half of the IQ bell curve since this is the section of the population that prior to the education would have obtained relatively lower scores due to their inability to comprehend the intelligence test's instructions. As you know, this has been reported in some studies and not found in others. As Must and Must have shown, the FE is not invariant, at least in Estonia. There is little reason to expect that it is invariant elsewhere. The Nettelbeck study, for example, did not report this finding, nor would it have been expected in such an otherwise homogeneous group, but the FE was still present.
If all these predictions hold up, there would be support for the notion that secular IQ gains and race differences are not different phenomena but have a common origin in literacy. Not likely. The differences between racial groups are g loaded. Here is the short section 5 (conclusions) from a recent paper by Rushton and Jensen:
"Heritable g is at the core of the debate over how much the mean Black–White gap in IQ and school achievement is due to the genes rather than to the environment, and therefore, how much it can be expected to narrow. While g and genetic estimates correlate significantly positively with Black–White differences 0.61 and 0.48 (P < 0.001), they correlate significantly negatively (or not at all) with the secular gains (r = -0.33; P < 0.001) and 0.13 (ns). Similarly, g loadings and heritabilities from the items of the Raven Matrices correlate significantly positively with each other and with Black–White differences (mean r = 0.74, P < 0.01). Although the secular gains are on g-loaded tests (such as the Wechsler), they are negatively correlated with the most g-loaded components of those tests. Tests lose their g loadedness over time as the result of training, retesting, and familiarity (te Nijenhuis et al., 2007).
Some issues, however, remain to be resolved. For example, Lynn (2009) found a secular rise in the Developmental Quotients of infants in the first two years of life, which he suggested was due to improved pre-natal and early post-natal nutrition. He supported his conjecture by pointing to equivalent gains in birth weight, stature, and brain size, and the correlation of these variables with later IQ. If it becomes possible to disentangle environmental factors that do affect g, from the environmental factors that do not affect g, the negative correlation between g and secular gains may increase from -0.33 to nearer - 1.00.
Predictions about the Black–White IQ gap narrowing due to the secular rise is based on faith rather than evidence. There is no more reason to expect Black–White differences in IQ to narrow as a result of the secular rise in IQ than to expect male–female differences in height to narrow as a result of the secular rise in height. The (mostly heritable) cause of the one is not the (mostly environmental) cause of the other. From the present perspective, the Flynn Effect (the secular rise in IQ) is not a Jensen Effect (because it does not occur on g).."
[Rushton and Jensen, Intelligence, Volume 38, Issue 2, March-April 2010, Pages 213-219 ]
To test these predictions, Marks looked at samples representative of whole populations (rather than individuals), and used ecological methods to calculate statistical associations between IQ and literacy rates across different countries.
The problem is that he is looking at a consequence of intelligence that has little, if anything, to do with causation. But Marks made it clear that he thinks he is looking at the cause and not the effect. Here is a quote from the paper you referenced:
"Secondly, the differences in IQ scores that exist today between different populations are artifacts of large, confounding literacy differences that exist between these populations. Thirdly, white-black differences in IQ scores are caused by literacy differences between these racial groups. Comparing the average IQ test scores of racial groups or populations without controlling for literacy is illegitimate."
None of the above is correct. The population group differences have been shown to be highly heritable biological differences and are not the result of social factors. The race difference shows up in testing at age 2-1/2 and in the robust correlate, head size, at birth. Literacy is not a birth nor an early childhood parameter. Marks is off the mark.
It should also be noted that Mark's findings only speak to populations (not individuals) and do not say much about causation. Marks made a very clear comment (see above) about his view of causation. It seems to me that anyone wishing to show that literacy is a factor in IQ test scores would use structural equation modeling to evaluate the obvious alternatives, with the goal of showing the one that best fits available data.
variables that affect both literacy and IQ. Still, the result that population level literacy changes with population IQ is suggestive that increased literacy is causing increased IQ.
No, it is not. It instead suggests that IQ is accompanied by literacy. It is well known that years of education correlates positively with IQ and to such an extent that educational attainment can be used as a rough proxy for IQ. But people do not become more intelligent as they spend more years in school. Education is not intelligence, but rather acts as a tool set for the application of intelligence. IQ tends to be stable over most of our lives, after about age 6, but even before that age, it is highly predictive of later measurements and is predictive of educational achievement. For a second time let me suggest Fagan's paper: The prediction, from infancy, of adult IQ and achievement Intelligence, Volume 35, Issue 3, May-June 2007, Pages 225-231
Joseph F. Fagan, Cynthia R. Holland, Karyn Wheeler
Marks did just that by scanning the literature for datasets containing test estimates for populations of groups taking both the Armed Forces Qualifications Test and tests of literacy. One study on nine groups of soldiers differing in job and reading ability found a correlation of .96 between the Armed Forces Qualifications Test and reading achievement (Sticht, Caylor, Kern, & Fox, 1972).
Yes, because reading is highly g loaded. Jensen devoted a long discussion to this in The g Factor (pages 279-282). Reading does not boost g; it is g that makes reading comprehension high or low as a function of individual intelligence.
Another study obtained reading scores for 17-year olds for those same ethnic groups and dates and found a correlation of .997 between reading scores and Armed Forces Qualifications Test scores (Campbell et al., 2000). This nearly perfect correlation was based on six pairs of data points from six independent population samples evaluated by two separate groups of investigators.
These "nearly perfect correlations" are the subject of a very well done paper in Intelligence: The issue of power in the identification of “g” with lower-order factors, Pages 336-344, Dora Matzke, Conor V. Dolan, Dylan Molenaar
They commented in the conclusions section, "Our examination of published studies revealed that most of our case studies, which reported a perfect correlation between g and a lower-order factor, were underpowered, with power coefficient rarely exceeding 0.3." The comment above strikes me as precisely the kind of underpowered case Matzke et al. were addressing.
"On the basis of the studies summarized here, there can be little doubt that the Armed Forces Qualifications Test is a measure of literacy."
There is plenty of evidence presented in The Bell Curve that the AFQT is heavily g loaded. There is one g and it doesn't matter how it is measured, the g is still the same thing. IQ tests get virtually all of their validity by acting as a proxy for g. If g is measured by the Raven's, the WJ-III, or a battery of reaction time tests, the thing at the root is the same g. The entire argument of literacy should be immediately questioned when it is used in connection with the FE, since it has been clearly shown that the largest secular gains have been observed in abstract test items. The Raven's shows this quite well. I once had about a one hour chat with John Raven and asked him if he thought the gains in the Raven's tests were g loaded. He only replied that the scores were increasing. When I asked Jim Flynn if the FE gains were g loaded, he gave me his usual historical (Rabbit and Hound story) answer, then said "I don't know." Most studies that have attempted to determine the g loading of the FE have found no loading. Rushton and Jensen have made this point as have Must and Must and others.
In The g Factor, Jensen showed that a positive FE could be shown for abstract test items, while a negative FE could be shown for scholastic items (see page 322). It is more than odd that the latter (literacy) would contribute in the opposite direction along with abstract items.
Potential research avenues to be explored.
The Marks study suggests a crucial environmental factor is literacy. Marks has tried to make that point, but he has skipped the necessary details and has ignored the strong biological effects as well as the early age factor that I have previously pointed out. If this is so, then interventions that increase literacy will also narrow the IQ gap found between different races and nationalities. To the best of my knowledge, no social or training factor has ever been shown to cause a real and permanent increase in g. Training increases the Spearman s loading and lowers the g loading. It does not improve real intelligence (g).
This latest research on the environmental effects of nutrition (Colom et al., 2005, but see Flynn, 2009), disease, literacy, and more on both the rise in IQ and ethnic, racial, and national disparities in IQ point to the importance of the environment for developing intelligence as well as the importance for researchers to be very careful when they use intelligence test performance (especially verbal tests) to make inferences about hereditary differences between different ethnic groups and nationalities.
Environmental factors account for about 17% of the variance in intelligence in developed nations. It must be more in undeveloped nations, but environmental factors act through biological mechanisms (toxins, disease, etc., not social factors) and, so far, these have all been negative. Ethnic group IQs are highly heritable as is evident by such displays as regression to the mean (for the group), inbreeding depression, twin studies, and adoption studies. Marks failed to explain the finding that adopted children reach adult IQs that are as predicted by their biological peers and those IQs are uncorrelated with their adoptive siblings. You would expect similar literacy within a family, but these studies show that only heritability was able to explain the outcomes.
Bob Williams
Technorati Tags: psychology, educational psychology, intelligence, Flynn effect, race and IQ, ethnicity and IQ, national IQs, guest blog post, Bob Williams, Scott Barry Kaufman
