« Web host recs | Gene Expression Front Page | "Myth," "fact," and other such things.... »
March 11, 2004

Loose ends

Not sure this is worth posting, but here it is anyway...

Shortly before Christmas I discussed the Flynn Effect: the long-term increase in average IQ scores observed in many countries. I have been thinking about one of the points made in comments, which I didn’t do justice to at the time.

Roughly, it was argued that:

(a) the differences in IQ levels between different populations are constant over long periods of time; and

(b) this constancy supports the view that the differences have a genetic basis.

This raises a question of fact - are the differences between different populations really constant over time? - and a question of inference: assuming that the factual claim is correct, what can we legitimately infer from it?

I will come back to the question of fact, but first I want to pursue the question of inference. (Warning: continuation includes tedious methodological discussion.)


Suppose that in two or more populations the mean IQ of each population is observed to rise substantially over a period of decades, but the differences between them remain constant.

Strictly, the constancy of a difference implies nothing about its genetic or environmental origin. At the same time, one may feel that if the difference is
constant a genetic explanation is more plausible. I share this feeling, but find it difficult to pin down exactly why. One possible argument might run as follows.

If the difference between two populations is genetically determined, we would not expect it to change rapidly over time. Differential reproduction w.r.t. IQ within populations, at the rates usually observed, would produce a change in mean IQ of less than 1 IQ point per generation, so the maximum expected change in the difference between two populations would be less than 2 points per generation. If, on the other hand, the difference is environmentally determined, we might well expect it to change rapidly (on a time scale of a few decades), since environmental changes during the 20th century were large, but probably at different rates in different countries. So if we observe a large widening or narrowing of differences in IQ, the genetic hypothesis is prima facie refuted, and the environmental hypothesis is corroborated (in Popper’s sense). But if no substantial change in differences is observed, the environmental hypothesis is weakened if not refuted, and the genetic hypothesis is corroborated.

Alternatively, we might view the matter from a Bayesian perspective. We have two mutually exclusive and exhaustive hypotheses: either the difference has a genetic cause, or it has an environmental one. We assume that each hypothesis has a prior probability greater than 0 and less than 1. We also assume that the probability of observing a constant IQ gap, if the genetic hypothesis is true, is greater than its probability if the environmental hypothesis is true. Given these assumptions, it is easy to prove from Bayes’ Theorem that if in fact we observe a constant IQ gap, then the probability of the genetic hypothesis is increased, while that of the environmental hypothesis is reduced. As this conclusion holds whatever the prior probability of the hypotheses (provided it is not 0 or 1), we may conclude that the genetic hypothesis is strengthened by the observation of a constant IQ gap, without relying on the more controversial parts of Bayesian theory.

I therefore think that in principle the argument is sound, as far as it goes. I hope so, as I recently used a similar argument myself in relation to social mobility! However, the weakness of all such arguments is that they rely heavily on assessments of prior probabilities. We form expectations based on general knowledge or theoretical principles, and judge the evidence in the light of those expectations. But a priori expectations must always give way to observed facts. Improbable events do sometimes happen. If in fact IQ levels do change substantially over a period of decades, and we have good reason to believe that the changes are environmentally determined, then observed constancy of differences gives no support to a genetic hypothesis. Whether the initial differences were genetic or environmental in origin, the observations would be the same. A difference in IQ levels is merely a by-product of the actual levels, determined separately, and if these change by equal amounts, for environmental reasons, then we can infer nothing either way about the origins of the difference. (A + k) - (B + k) = A - B, whatever the reasons for the difference between A and B.

I conclude that the (assumed) constancy of an IQ difference does not in itself provide strong support for a genetic origin of that diference, given the facts about the Flynn Effect.

Some comments on my post however made a more specific point. This is that the increase in IQ due to environmental improvements must eventually reach some limit, determined by the genes of the population in question. Eventually all populations will reach their full genetic potential, and any differences between them will then be purely genetic in origin. If there are no genetic differences, then as the populations approach their full genetic potential, the observed IQ differences between them will narrow and eventually vanish.

This is a sound argument. However, there is little evidence that a limit to the Flynn Effect has yet been reached, even in advanced countries. Herrnstein & Murray (Bell Curve, p. 309) suggest that ‘the rising trend in test scores may already be leveling off in some countries’, and cite a 1986 paper by Lynn and Hampson in support. But the more recent paper by Lynn and Pagliari (1994) specifically set out to test this hypothesis for the USA, and found no slackening of the rate of increase (see earlier post for references). The only evidence for a slowing of the increase that I know of is Teasdale and Owen’s study in Denmark. (I think I have seen a similar report w.r.t. Sweden, but I cannot find the reference.) When the rest of the world has achieved Scandinavian levels of living standards and social equality, any remaining differences in mean IQ will be worth looking at. It is obvious that third-world countries are far from reaching this position.

The above discussion has all been based on the assumption that in fact IQ differences between populations are constant. On reviewing the evidence in Neisser and Storfer I think this is an oversimplification. All advanced countries have experienced a Flynn Effect, but its size has varied somewhat from country to country. It seems to have been particularly rapid in post-war Japan and some West European countries, but significantly slower in the UK and the USA, with the result that other countries have caught up with, or in some cases, overtaken them. In these cases IQ seems to mirror trends in economic development, which gives some support to an environmental explanation of the IQ differences concerned.

Posted by David B at 04:28 AM