Research

The Flynn effect, and what happened to it

Raw scores climbed roughly three points a decade for most of the twentieth century. Nobody fully agrees why, and in several countries the climb has stopped.

What the effect is

Through the twentieth century, people kept getting better at IQ tests. Not slightly better — substantially, consistently, across every country with long-running data.

The gains ran at roughly three points per decade on the standardised scale, and they were largest on the tests supposedly least dependent on schooling. Raven's Progressive Matrices, the flagship culture-fair instrument, showed some of the steepest rises of all.

The reason this is not obvious from published scores is that test publishers restandardise. Every time a test is renormed, the average is reset to 100. The rise is visible only when you give an old test to a modern sample, or a modern test to an old one — which is exactly what James Flynn did across dozens of datasets, giving the phenomenon its name.

The scale of it is easy to understate. Score a modern sample against 1930s norms and the average lands somewhere around 120. Score the 1930s sample against modern norms and it lands near 80. Taken literally, that would mean the average person of the 1930s was near the bottom of the modern distribution, which is plainly absurd and is itself the strongest clue about what the effect is not.

What might have caused it

No single explanation covers the pattern, and the honest position is that several causes probably operate together with different weights in different places and periods.

Schooling. Both the length and the character of education changed enormously. Time in school rose, and the content shifted towards abstraction and classification — precisely the habits of mind that matrix items reward.

Nutrition and health. Better childhood nutrition, fewer infectious diseases and less early-life illness all plausibly raise cognitive development. This fits the observation that gains were largest at the bottom of the distribution in early decades.

Reduced neurotoxin exposure. The removal of lead from petrol and paint is associated with measurable cognitive gains in exposed cohorts.

Familiarity with abstraction. Flynn's own preferred explanation. Modern life demands constant use of hypothetical and categorical reasoning — what he called putting on “scientific spectacles”. A test asking what a dog and a rabbit have in common gets “both mammals” from a modern respondent and “you use dogs to hunt rabbits” from a pre-industrial one. Only the first scores.

Smaller families and more parental attention. Plausible, harder to isolate from everything else that changed alongside it.

The reversal

Since roughly the 1990s, several countries with good long-running data have shown the gains flattening and then going into reverse.

The Norwegian conscript data are the clearest case, because the sample is close to a full male birth cohort tested on the same instrument for decades. Analyses of those data found the rise peaking with cohorts born around 1975 and declining afterwards. Similar reversals have been reported in Denmark, Finland, the United Kingdom and France.

The strongest piece of evidence in the Norwegian work is that the decline appears within families — between brothers born years apart. That rules out the explanation people reach for first, which is that the population composition changed through migration or differential birth rates. Something in the shared environment shifted between cohorts.

What that something is remains open. Candidates include changes in school curricula and how reading is taught, changes in media consumption and attention, and the possibility that the gains from better nutrition and health simply ran out because there was no further deficiency to correct.

What the effect tells us about the tests

The Flynn effect is at least as informative about IQ tests as it is about people, and this is the part that usually gets skipped.

If a century of gains represented a genuine rise in general intelligence of the size the numbers imply, the consequences would be visible everywhere and they are not. The more defensible reading is that tests measure a mix of general reasoning and specific culturally-trained habits, and that the trained component rose enormously while the general component moved much less.

That has a direct practical consequence. Comparing a score measured against 1980 norms with one measured against 2020 norms is not valid, and the difference can exceed the clinical thresholds used for educational placement or disability determination. Which version of a test was used, and when it was last normed, is a real question with real consequences, not a technicality.

The takeaway

IQ scores rose for a century, by an amount too large to represent a straightforward increase in intelligence. They have stopped rising, and in some places started falling, for reasons nobody has pinned down.

Both facts point the same way. The number an IQ test produces is a rank against a particular group at a particular moment, and it is sensitive to the environment in which that group grew up. That is a reason to take scores seriously as measurements and lightly as verdicts.