Sunday, 8 February 2015

105 years of the Flynn effect: very fluid

The Flynn effect seems to be continuing, or at the very least the publications on the Flynn effect seem to be continuing, so it may be that even psychologists are becoming cleverer, or simply more adept at publishing papers. Taking the former possibility as the kinder hypothesis, if everyone is getting smarter, it is about time we sorted out the Flynn Effect.

Even if we follow the less kind hypothesis, it is worthwhile maintaining an interest in the phenomenon of the secular rise in ability, because the apparent rise in intelligence test scores, without a particularly evident commensurate rise in everyday intelligence, is a topic worthy of critical analysis. Neither IQ tests nor scholastic attainment tests were designed for cross-cohort comparisons, so they might be suitable for a year-by-year identification of the brightest students without being able to give valid generation-by-generation comparisons. However, there may be some signs of increasing intelligence if the rising number of publications and patents are to be believed. More impressively, chess grand masters are getting younger.

Into this torrent of Flynn-Effectism jump Jakob Pietschnig and Martin Voracek with a large raft of a paper which runs to 179 pages, which is what you get when you have the temerity and the Teutonic thoroughness to plough through 105 years of data and assemble 271 independent samples from 31 countries, totalling almost four million participants.

What sorts of childhood do German speakers have, which drives them to these immense labours? Perhaps the Viennese duo of Jakob and Martin sat through early childhood, like Einstein, in mute bemusement at life. It is said (and even though this is apocryphal, I am repeating it) that Einstein did not speak until he was four, and then muttered “The soup is cold”. His family clapped and cheered and cried because they had been so worried about his muteness, and finally begged him to explain why he had never spoken before. Einstein replied: “Because formerly everything was in order”.

Pietschnig, J., & Voracek, M. (2015). One Century of Global IQ Gains: A Formal Meta-Analysis of the Flynn Effect (1909-2013). Perspectives on Psychological Science, in press.

Anyway, our two heroes have ploughed through more psychology papers than is fit to mention, and I am here to save you. In this case a picture is worth several thousand words:



So, the Flynn Effect is primarily on Fluid IQ and least on crystallised abilities, though still positive for those. We seem to have got brighter, particularly at on-the-spot thinking, though knowledge has increased as well. The prospects for humanity look good.

I wish I had the courage to end on that note, but there are one or two other things to mention, so as to do justice to this important paper:

IQ gains vary according to domain: 0.41 for fluid, 0.30 spatial , 0.28 full scale,
and 0.21 IQ points for crystallised. Effects are stronger for adults than children, and have decreased in recent decades. The authors suggest that factors associated with life history speed seem mainly responsible for the Flynn effect’s general trajectory, whereas favourable social-multiplier effects and effects related to economic prosperity appear to be responsible for observed differences of the Flynn effect across intelligence domains.

This formal meta-analysis of the Flynn effect provides strong evidence for continuous global generational IQ test score gains in the general population over the past century.

Gains were stronger between World Wars I and II, but showed a marked decrease during the World War II years: 0.72 vs. 0.21 IQ points annually.

There were decreasing gains in more recent decades. The decreasing strength of the IQ gains over time was reflected by meaningful negative effects of timespan for fullscale, fluid, and crystallized IQ, as well as year of onset for fluid and spatial IQ. Supported by the observed IQ change trajectories, evidence for decreasing gains in recent decades can be considered to be robust. Regression slopes of joinpoint regressions significantly decreased in the last segment of all IQ domains

Joinpoint regression analyses for all IQ domains showed a significantly better fit for regression models assuming changes in the strength of regression slopes over time than for models without incorporating changes in the slopes. This result indicates that IQ test score gains have not been linear over the past century, but rather seem to have been alternately accelerating and decelerating and finally decreasing during more recent years. Storfer already has proposed such changes in the strength of gains over time, estimating gains of 3.75 IQ points per decade from 1900 to 1920, of 2.50 from 1920 to 1960, and slightly smaller gains after the 1960s. In contrast, the present evidence indicates that a decrease in the strength of gains only emerged in the mid-1970s, yielding moderate gains of 2.30 IQ points per decade. However, the pattern preceding this period appears to be considerably more differentiated, indicating that gains during the early 20th century have been relatively weak (0.80 IQ points per decade), then showed a sharp increase in the 1920s (7.20 IQ points per decade), decreased again from 1935 to 1947 (2.10 IQ points per decade), but later again recovered until 1976 (3.00 IQ points per decade).

Stronger gains were observed for adults than children, showing large effects for fluid and spatial IQ (.28 and .66 respectively). Past research has attributed increasing gains with age mainly to effects of better education. If so, then educational effects would be expected to affect crystallized IQ most (e.g., Flynn, 2010). However, effects of age on crystallized and full scale IQ were negligible in the present study, although the signs of the change coefficients were directionally as expected.

GDP growth per capita was substantially positively associated with fullscale .09 , crystallized .18, and spatial IQ  .50, but showed negligible effects for fluid IQ. This finding is consistent with previous reports of links of economic prosperity with IQ in several nations (Lynn & Vanhanen, 2002). Associations between IQ gains
and GDP have been linked to educational improvements (Rindermann, 2008), thus conceivably reflecting effects of better educational infrastructure.

Comment: In my book only spatial shows a substantial correlation with GDP, and crystallised is merely indicative. Absence of an effect on fluid IQ is puzzling, suggests knowledge rather than wit is being boosted.

Stronger gains on low-g tests. IQ gains appeared not to be taking place on psychometric g. Findings of meaningful negative effects of medium and high g-ness of tests on IQ gains (.12 and .02, respectively) are supported by the overall lower gains observed in crystallized IQ (i.e., the domain with highest g-ness). These findings are consistent with previous evidence showing negative associations between g-ness and IQ gains (Te Nijenhuis & van der Flier, 2013) and corroborate the importance of environmental influences on generational IQ test score changes (Rushton, 1999)

The authors then go on to consider causes for the Flynn effect, and summarise them in Table 2, which I find too big on my screeen to constitute any sort of summary. It looks like a large traffic light. So, in linguistic, rather than tabular form, here is what they say:

At least three aspects of the present evidence support the role of education as an important contributing factor to the explanation of IQ gains: First, we observed substantial global increases in crystallized IQ. This may reflect, at least to a certain extent, effects of more and better schooling. Positive associations between crystallized IQ task performance and highest educational qualification are widely accepted and have previously been shown to be associated with gains on crystallized IQ measures. Nonetheless, although education has been shown to account for portions of crystallized IQ gains, gains have been reported to remain substantial after controlling for education (Pietschnig, Tran, & Voracek, 2013).

Second, larger IQ gains were observed for adults than for adolescents and children in fluid and spatial IQ domains. Surprisingly, no meaningful effect of age on crystallized IQ was found. However, this may be due to the effect of growing GDP which could mask age effects. Increasing numbers of average educational years may therefore explain stronger gains for adults than children.
However, although IQ test performance of children and adolescents has been observed to increase to a lesser extent than that of adults, gains for young samples still were substantial in our data. Although increases in formal educational years may not play a crucial role among children and adolescent IQ gains, increasing exposure to early childhood education programs, as witnessed in more recent years, might so. Although most such programs are aimed at more mature children, some of these programs are aimed at infants (e.g., the US-based ABCDerian project, which had average enrollment ages of 4.4 months; for an overview, see Hunt, 2011, pp. 288-291). Thus, even IQ gains in infants may be suitably explained by educational improvements. However, it remains difficult to decide the explanatory potential of early education programs for IQ gains in children and adolescents, because such programs differ considerably in coverage and availability between (and even within) investigated countries and timespans.

Third, IQ gains were predicted by average increases in GDP per capita across all domains, with the exception of fluid IQ. Positive associations of GDP with IQ gains have been observed in several studies and countries. In particular, the substantial effect of GDP on crystallized IQ may be linked to educational effects. It has been proposed that investments in better education lead to economic growth and vice versa, thus leading to a positive feedback loop of economic prosperity, education, and intelligence. Of note, it has been shown more recently that increases in GDP may be better described as a function of education rather than the other way round which in turn would reverse the causality assumption of the regression model applied by us. Regardless of the causality of the observed association, the positive sign of the association is consistent with this theory.

These findings show that there is little doubt that education plays a role in explaining the Flynn effect. Nonetheless, schooling is unlikely to account for the full extent of the IQ gains, and particularly the large gains for fluid IQ cannot be attributed to better education.

They say that there is little evidence that technology has a boosting effect. Neither has the decrease in family size, though the data are ambiguous about whether increased family size is a factor. Guessing in multiple choice settings is probably only a minor contributor at best, though consistent with the fluid IQ gains where guessing is most likely, because the items are difficult. Nutrition is discussed as a cause, but no clear confirmatory pattern emerges.

Although we did not directly test for decreasing IQ variability, findings suggest that at least nutrition and education may play an important role in this context and thus, consistent with prior research, may well have led to decreases in IQ variability.

Slower life history has been observed to be associated with a decline of the strength of g over time and promotes ability differentiation (Woodley, 2012a; Woodley & Madison, 2013). Consistent with the life history model, the Flynn effect in the present meta-analysis is apparently not on g. Indeed, the observed effects of test type suggest a negative association between IQ gains and psychometric g.
Predictions of the life history model appear to fit well to the observed patterns of IQ gains in the present meta-analysis. As this model does not warrant uniformity of changes across countries or strength of changes across time, life history speed would be suitable to explain the erratic pattern of IQ changes in our data. Different causes associated with life history speed could thus be either present or absent in single countries, but still would yield overall IQ gains due to compensatory effects of other factors being present. In other words, not all related causes need to be present in order to decelerate life history speed and consequently lead to gains; rather, causes may be effective one at a time.


I think this paper is a substantial step forwards, and gives a very well established broad based view of rising IQ scores. It will serve as an excellent source for many researchers, and considerably assist them in scoping putative causes. I have already stated my preference for raw data, individual items which serve as trace elements, particularly raw digit span scores and maths scores, where we can be more sure about the fundamental elements, as opposed to those crystallised accomplishments which are more subject to standardisation effects. However, the landscape of Flynn effect data have now been mapped out in great detail, and others will jump in to dig deeper into the rich archaeology which Pietschnig and Voracek have laid out for us in such detail.

I will give the authors the last word:

The totality of retrievable empirical evidence on this phenomenon, as quantitatively summarized here, points towards components of life history speed, such as improvements of education and nutritional factors as well as a reduction of pathogen-related factors, as the prime candidate causes of the Flynn effect, whilst differences in the strength of gains between intelligence domains may be accounted for by social multipliers and economic prosperity. Future research will show whether the now observed global decrease of IQ test score gains will ultimately lead to an end of these gains or even to a reversal.


  1. I wonder whether there's an analogy to Moore's Law with computer chips: the underlying causes of the doubling of transistors per square inch every year or two seem to be manifold, but they unfold at a pretty steady rate over time.

  2. Dear Steve, Good analogy. In the case of Moore's Law we know exactly what happened. Very different labs took different approaches to packing more power into each circuit. Things improved by leaps and bounds if looked at closely. I remember the excitement of the Pentium chip, which seemed to be a major jump. We can see the blueprints, pore over the detail. From a distance, however, Silicon valley was a force which came up with fairly steady advances. It the case of the Flynn effect, whyever it is, we are necessarily more vague. It is a bit like increasing life span: living conditions are getting better, but whether it is nutrition, central heating or blogging which has made the difference is unclear. Blogging could be lifting human intelligence, couldn't it?

  3. This is an extraordinary paper and I commend the authors on their hard work (whatever that's worth).

    A few thoughts:
    1) Spatial IQ is much more closely linked to rising GDP than the other broad domains of IQ. Perhaps this corroborates the suggestion I've seen in a few places that spatial IQ is more closely linked to "practical intelligence" or real-world savviness? On the other hand, then spatial IQ measures would be better predictors of global job performance, which doesn't seem to be the case. Maybe the relation arises from higher spatial IQ producing better technology.

    2) The paper pretty much debunks the annoying meme that crystallized gains have been small in any absolute sense. Hopefully we'll see less of that myth in future.

  4. Dear Elijah,
    indeed, we observed strong gains in crystallized IQ and this seemed to hold for most countries in our examination. Conceivably, these gains may have been the first ones to hit the ceiling and this might have led to this notion that crystallized IQ has not been increasing.
    Best regards,