The following article is an updated revision of an article I published in August 2019 about how 21st century Northern Americans score on an IQ test normed before the second World War. The reason for the update is that in December 2019, the sample size of my study increased by 13% (from n = 15 to n = 17). I had originally hoped to collect more data before publishing an update but with the uncertainty surrounding the coronavirus crisis, it’s unclear when that will be.
The Flynn effect, popularized by James Flynn, refers to the fact that IQ tests supposedly get easier with time. Although by definition the average IQ of American or British (white) people is always 100, the older the IQ test, the easier it is to score 100. Thus to keep the average at 100, tests like the Wechsler must be renormed every 10 years or so, otherwise the average IQ would increase by about 3 points per decade.
Although scholars continue to debate whether the Flynn effect reflects a genuine increase in intelligence (perhaps caused by prenatal nutrition or mental stimulation) or just greater test sophistication caused by modernity, there’s been remarkably little skepticism about the existence of the Flynn effect itself.
Malcolm Gladwell writes:
If an American born in the nineteen-thirties has an I.Q. of 100, the Flynn effect says that his children will have I.Q.s of 108, and his grandchildren I.Q.s of close to 120—more than a standard deviation higher. If we work in the opposite direction, the typical teen-ager of today, with an I.Q. of 100, would have had grandparents with average I.Q.s of 82—seemingly below the threshold necessary to graduate from high school. And, if we go back even farther, the Flynn effect puts the average I.Q.s of the schoolchildren of 1900 at around 70, which is to suggest, bizarrely, that a century ago the United States was populated largely by people who today would be considered mentally retarded.
While few people believe our grandparents were genuinely mentally retarded, it’s taken for granted that they would have scored in the mentally retarded range by today’s standards.
But is this true? I began having doubts over a decade ago when I examined the items on the first Wechsler intelligence scale ever made: the ancient WBI (Wechsler Bellevue intelligence scale). Meticulously normed on New Yorkers in the 1930s, this test remains far and away the most comprehensive look we have at early 20th century white Northern American intelligence, and while some of the subtests looked easy by today’s standards, others, especially vocabulary, looked harder.
The Kaufman effect
What also struck me was how little instruction, probing or coaching people got when taking the ancient WBI, compared to its modern descendant the WAIS-IV. This matters a lot because the way the Flynn effect is calculated on the Wechsler is by giving a new sample of people both the newest Wechsler and its immediate predecessor, in random order to cancel out practice effects, and then seeing which version they score higher on. If they average 3 points lower on the WAIS-IV normed in 2006 than on the WAIS-III normed in 1995, it’s assumed IQ increased by 3 points in 11 years.
The problem with this method (as Alan Kaufman may have discovered before me) is that the subset of the sample that took the newer version first has a huge advantage on the older version compared to the norming sample of the older test (over and above the practice effect which is controlled for), because the norming sample of the older test was never given coaching and probing.
A Promethean once said maybe the Flynn effect is just a statistical artifact of some kind. He never told me what he meant, but it got me thinking:
One problem with how the Flynn Effect is calculated on the Wechsler is that it’s assumed that gains over time can be added. For example it’s assumed that you can add the supposed 7.8 IQ gain from WAIS normings 1953.5 -1978 to the 4.2 IQ gain from normings 1978 – 1995 to the 3.7 IQ gain from normings 1995-2006, for a grand total of 15.7 IQ points from normings 1953.5 – 2006.
This would make sense if he were talking about an absolute scale like height, but is problematic when talking about a sliding scale like IQ. For example, suppose the raw number of questions correctly answered in 1953.5 was 20 with an SD of 2. By 1953.5 standards, 20 = IQ 100 and every 2 points = 15 IQ points above or below 100. Now suppose in 1978, people averaged 22 with an SD of 1. That’s a gain of 15 IQ points by 1953.5 standards. Now suppose in 1995 people average 23 with an SD of 2. That’s a gain of 15 IQ points by 1978 standards. Adding the two gains together implies a 30 point gain from 1953.5 to 1995, but by both 1953 and 1993 standards, the difference is only 23 points.
Another problem with studying the Flynn effect is the content of tests like the Wechsler is constantly changing. This is especially problematic when studying long-term trends in general knowledge and vocabulary. If words that are obscure in the 1950s become popular in the 1970s, then people in the 1970s will score high on the 1950s vocabulary test. Meanwhile the 1970s vocabulary test may contain words that don’t become popular until the 1990s, Thus adding the vocabulary gains from the 1950s to the 1970s to the gains from the 1970s to the 1990s, might give the false impression that people in the 1990s will do especially well on a 1950s vocabulary test, when in reality, many words from the 1950s may have peaked in the 1970s and are even more obscure in the 1990s than they were in the 1950s.
An ambitious study
Given the Kaufman effect, the statistical artifact, and changing content, I realized the only way to truly understand the Flynn effect is to take the oldest quality IQ test I could find and replicate its original norming on a modern sample.
In 2008 I made it my mission to replicate Wechsler’s 1935-1938 norming of the very first Wechsler scale. Ideally I should have flown to New York where Wechsler had normed his original scale, but if Wechsler could use white New Yorkers as representative of all of white America (WWI IQ tests showed white New Yorkers matched the national white average), I could use white Ontarians as representative of all of white Northern America (indeed white Americans and white Canadians have virtually the same IQs). The target age group was 20-34 because this was the reference age group Wechsler had used to norm his subtests.
It took over a decade but I was gradually able to arrange for 17 randomly selected white young adults to take the one hour test. They were non-staff recruited from about half a dozen fast food/ coffeehouse locations in lower to upper middle class urban and suburban Ontario. The final sample ranged in education from 9.5 years (early high school dropout) to 18 years (Masters Degree in Engineering from one of Canada’s top universities). The mean self-reported education level was 12.9 years (SD = 2.12) suggesting that despite the lack of female participants, the sample was fairly representative (the average Canadian over 25 has about 13 years of schooling); in cases where those below the age of 25 were in the process of finishing a degree, they were credited as having it.
Testing conditions were not optimum (environments were sometimes noisy, at least one person had a few beers before testing; another was literally falling asleep during the test) and 17 people is way to small a sample to draw statistically significant conclusions about 11 different subtests. One man with a conspicuously low score was removed from the sample after he stated that he had years ago suffered a stroke.
Nonetheless, the below table shows how whites tested in 2008 to 2019 compared to Wechsler’s 1935-1938 sample, with the last column showing the expected scores of the 21st century sample, extrapolating gains James Flynn calculated from 1953.5 to 2006 (see page 240 of his book Are We Getting SMARTER?) to the current study: circa 1937 to circa 2013.5.
Note: the 11 subtests were scaled to have a mean of 10 and an SD of 3 in the original young adult norming sample, while the verbal, performance and full-scale IQs were scaled to have a mean of 100 and an SD of 15. Note also that vocabulary is alternate test, not used to calculate either verbal or full-scale IQ on the WBI. One third of my sample did not take Digit Symbol so for these, Performance and full-scale IQs were calculated via prorating.
|Test:||Nationally representative sample of young white adults (NY, 1935 to 1938)||Randomish sample of young white adults (2008 to 2019, ON, Canada)||Expected WBI scores in 2008-2019 based on Flynn’s calculated rate of increase|
|Information (general knowledge test)||10 (SD 3)||8.41 ( SD 2.55)||12.3|
|Similarites (verbal abstract reasoning)||10 (SD 3)||13.35 (SD 2.91)||15.54|
|Arithmetic (mental math)||10 (SD 3)||9.18 (SD 4.34)(this subtest contained a unit conversion item that seemed biased against Canadians so for those who advanced far enough to fail this item, scores were prorated; had they not been the mean would have been 7.53 (SD 3.54))||11.02|
|Vocabulary||10 (SD 3)||9 (SD 2.5)||14.95|
|Comprehension (Common sense & social judgement)||10 (SD 3)||9.47(SD 2.93)||13.93|
|Digit Span (attention & rote memory)||10 (SD 3)||9.71 (SD 2.63)||11.46|
|Picture Completion (visual alertness)||10 (SD 3)||10.71 (SD 3.1)||14.52|
|Picture Arrangement (social interpretation)||10 (SD 3)||10.24 (SD 2.73)||13.35|
|Block Design (spatial organization)||10 (SD 3)||13.12 (SD 3.31)||12.91|
|Object Assembly (spatial integration)||10 (SD 3)||11.82 (SD 1.89)||14.06|
|Digit Symbol (Rapid eye-hand coordination)||10 (SD 3)||11.12 (SD 2.82)(note: only 12 of the 17 subjects took this subtest)||14.66|
|Verbal IQ||100 (SD 15)||103.8 (SD 14.73)|
|Performance IQ||100 (SD 15)||109.3 (SD 12.11)|
|Full-scale IQ||100 (SD 15)||106.9 (SD 13.63)||122|
The Flynn effect is dramatically smaller than we’ve been led to believe at least on tests of specific information that may become obscure over generations. By contrast Similarities (abstract reasoning) and Block Design (spatial analysis) have indeed increased by amounts comparable with Flynn’s research. These two abilities may conspire to explain why some of the largest Flynn effects have been claimed on the Raven Progress Matrices, an abstract reasoning test using a spatial medium.
It’s unclear if these are nutritional gains caused by increasing brain size, neuroplastic gains caused by cultural stimulation, or mere teaching to the test caused by schooling, computers and brain games.
Lynn (1990) argued the Flynn effect was caused by nutrition, citing a twin study proving nutrition gains are more pronounced on Performance IQ (consistent with the Flynn effect). Research on identical twins (where one twin gets better prenatal nutrition than the other) has shown that by age 13, the well nourished twin exceeds his less nourished counterpart by about 0.5 SD on both head circumference and Performance IQ, but not at all on verbal IQ. Thus it’s interesting that 21st century young Northern American men today exceed their WWII counterparts by about 0.5 SD on both head circumference (22.61″ vs 22.3″) and Performance IQ (109 vs about 100).
One possibility is that Performance IQ gains are entirely caused by improvements in the biological environment (prenatal health and nutrition), while verbal IQ gains are entirely caused by cultural advances (i.e. education); though somewhat negated by knowledge obsolescence.