Reverse Flynn Effect: U.S. Adult IQ Is Declining

As noted in Edward Dutton and Michael Woodley’s book, At Our Wits’ End, the Flynn effect, in which intelligence, as measured by IQ test scores, was observed to have risen rapidly during the 20th century among a broadly-based population around the world, may have begun to peak and reverse starting around the mid-1990s. This phenomenon has been reported in studies done in Scandinavian countries, France, the United Kingdom, Finland, and several German-speaking countries.

In a paper to be published in the May-June 2023 issue of Intelligence, “Looking for Flynn effects in a recent online U.S. adult sample: Examining shifts within the SAPA Project”, the authors examine 13 years of data on a sample of 394,378 U.S. adults over the period 2006–2018 from the Synthetic Aperture Personality Assessment Project (SAPA Project) to test whether the Flynn effect was still in evidence. Here is the abstract.

Compared to European countries, research is limited regarding if the Flynn effect, or its reversal, is a current phenomenon in the United States. Though recent research on the United States suggests that a Flynn effect could still be present, or partially present, among child and adolescent samples, few studies have explored differences of cognitive ability scores among US adults. Thirteen years of cross-sectional data from a subsample of adults (n = 394,378) were obtained from the Synthetic Aperture Personality Assessment Project (SAPA Project) to examine if cognitive ability scores changed within the United States from 2006 to 2018. Responses to an overlapping set of 35 (collected 2006–2018) and 60 (collected 2011–2018) items from the open-source multiple choice intelligence assessment International Cognitive Ability Resource (ICAR) were used to examine the trends in standardized average composite cognitive ability scores and domain scores of matrix reasoning, letter and number series, verbal reasoning, and three-dimensional rotation. Composite ability scores from 35 items and domain scores (matrix reasoning; letter and number series) showed a pattern consistent with a reversed Flynn effect from 2006 to 2018 when stratified across age, education, or gender. Slopes for verbal reasoning scores, however, failed to meet or exceed an annual threshold of |0.02| SD. A reversed Flynn effect was also present from 2011 to 2018 for composite ability scores from 60 items across age, education, and gender. Despite declining scores across age and demographics in other domains of cognitive ability, three-dimensional rotation scores showed evidence of a Flynn effect with the largest slopes occurring across age stratified regressions.

Full text of the paper is available at the article link above, or as a PDF from a button on that page. Here is a chart summarising the results:

The authors conclude:

This study set out to investigate if a Flynn effect or a reverse Flynn effect was a phenomenon in a large United States sample recruited between 2006 and 2018. Regardless of education, gender, and age, lower annual scores were observed for composite cognitive ability measured by 35 items, and the matrix reasoning and letter and number series scores measured across the 13 years of assessment. These differences were replicated across the 60-item composite ability scores from 2011 to 2018, however, three-dimensional rotation scores measured during this 8-year period showed evidence of a Flynn effect of varying magnitudes across 18- to 60-year-olds. The largest differences in mean ability scores were often observed for participants between the ages of 18 to 22. Beyond age, a reverse Flynn effect was also present across all levels of educational attainment, with the rate of decreasing scores being steeper for those with less than a 4-year college degree. While additional work needs to be done to further incorporate other demographic measures from the SAPA Project, the current study indicates that the Flynn effect and its associated reversal may no longer generalize across all ages or levels of education. It also underlines the need for further research using large adult samples to understand if the Flynn effect or if its reversal is a phenomenon in the United States during the twenty-first century.

The divergence between scores on three-dimensional shape rotation tests and all others is intriguing. I have written here earlier (2022-02-06) about the recent popular psychology distinction between “Wordcel” and “Shape Rotator” people in “ ‘Wordcel’ vs. ‘Shape Rotator’—An Introduction”. Could it be that what this research is revealing is that shape rotators, being more securely moored in the real world (many gravitate toward careers in engineering, the physical sciences, and computing), and, if graduates of higher education, more likely to have spent their time in departments less afflicted by the intellectual diseases of postmodernism, political correctness, and woke ideology, have not yet suffered the dulling of intelligence those doctrines engender? If this were the case, you would expect to see the Flynn effect still increasing scores on shape rotation tests among that cohort, while the decline in intelligence among wordcels, who already scored low on shape rotation tasks, would have little effect on the overall score.


I wonder about the following and whether they are able to compensate in the analysis for it.

I would expect that the gaussian distribution of the population at every level of education would be changing as more people access that educational level (and can “earn” the degree due to grade inflation and reduction in the standards for a degree). It may be best to give a simple example even if not accurate to describe what I mean.

When 10 percent of the high school graduates go on to college there is a good chance that these are on average a higher IQ and tighter distribution than if 50% of the graduates to on to earn a college degree. This shift would negatively impact the average and impact the std deviation (probably increasing it) of the college degree folks and it would also lower the average and maybe reduce the standard deviation of the high school level graduate as a portion of the population that raised the average have now been put into a new bucket.

Fifty years ago it was not uncommon to have a high school graduate that was very intelligent. They just did not have opportunity or it wasn’t a must do to go to college. My experience is that this is much less common today.

The following illustrates it based on a quick LibreOffice Calc spreadsheet where I created 1148 samples from a random distribution of mean 100 and std dev 10 sorted them and assumed the highest 10 percent goes to college and then the highest 50% go to college.

10% Go To College 50% go to college
College IQ Sum 11798 69272
College Count 100 648
College IQ Ave 118 107
HS IQ Sum 102798 45324
HS Count 1048 500
HS IQ Ave 98 91
Total Count 1148 1148

There are a number of potential sources of bias in the study, of which that is certainly one. First of all, this is based upon self-administered tests collected over the Internet, so one relies upon self-reporting and the honesty of the participants. Participants were obviously filtered by having Internet access, and it is probable that between 2006 and 2018, as Internet access has become more widely available, that it will include those with lower cognitive capacity who were later adopters of the Internet (and excludes entirely those not yet online). The sample was 65.03% female, which is obviously disproportionate to the population as a whole (and you have to ask why, which may imply other selection effects that may have shifted over time). Finally 32.02% of respondents reported they were currently attending college, which is wildly disproportionate to the total population, and suggests taking the test may have been a class assignment, which would have selected by the curriculum being studied.

These data are very “soft” compared to, say, those from Finland, Norway, and Sweden, which have nearly universal male military service and test all conscripts for IQ, with yearly databases going back to 1950. It’s interesting, nonetheless, that a similar effect was seen in the U.S. data as in the more representative data which first suggested a reversal of the Flynn effect.


That is a large sample in absolute terms, but represents only about 1 in 800 of the US population. The potential for a biased sample is high, as suggested by the gross over-representation of females.

Trying to imagine what the “Trends of 35-Item” graph would look like if those straight line Excel trendlines had not been superimposed – it looks like the scores were fairly level 2006 - 2010, and then fairly level again (at a lower level) 2014-2018. One wonders if there were some significant changes in sampling techniques in the intervening years?


As regards the average intelligence of the college student population (which made up 32.02% of respondents in the SAPA sample, as self-reported), Audacious Epigone (AE) has two posts from 2017 on the topic:

These results were obtained by taking scores on the General Social Survey’s Wordsum vocabulary test and rescaling them to have the statistics of IQ scores (mean 100, standard deviation 15). Here is the summary of results from the first post.

Graduated in IQ
1960s 112.3
1970s 109.1
1980s 106.0
1990s 103.9
2000s 102.9
2010s 100.0

In the second post, AE plots the percentage of college graduates who scored 9 or a perfect 10 on the Wordsum test from 1974 through 2018.


The questions in the test did not change over this period.

AE notes:

Four decades ago, 12% of the population had degrees. Today, 33% does. If, in the early seventies, that 12% roughly corresponded with the top 12% of the IQ distribution, then the 6% of the population that aced the Wordsum test would comprise 1 in 2 of those grads. If today that 33% roughly corresponds with the top 33% of the IQ distribution, then the 6% of the population acing the Wordsum test would be a bit more than 1 in 6 of today’s grads. QED*.


Some interesting data about depression with an inflection point in 2012:

1 Like