@voxday gets it wrong on IQ (again)

@voxday gets it wrong on IQ (again)

As a headline that is a bit ‘cat licks its own bum’ level of non-news but on we must go…

The other day Vox was disparaging about the value of scientific evidence. I’m not entirely sure if he is clear himself about what he means but when it comes to IQ he is happy to post anything that he feels supports his case.

This time, it is a pair of studies that point to a 4 point decline in IQ in France in a 9-10 year period. Vox quotes a second study that was an analysis of the first. This second study was an attempt to discern the cause of the decline by looking at the magnitude of the changes at a subtest level. This second paper concluded that the decline ‘likely has a primarily biological cause’. Vox declares it was due to immigration.

This is a very good example of studies that, while not necessarily wrong, aren’t really saying much at all. To see why you have to track back from Vox’s claim (immigrants somehow making whole countries less intelligent), to what the actual paper he quoted said, to the original paper that the second paper analysed and from there to what the actual original study was.

The paper Vox quoted was  “In France, are secular IQ losses biologically caused? A comment on Dutton and Lynn (2015)” by Michael A. Woodley and Curtis S. Dunkel (
http://dx.doi.org/10.1016/j.intell.2015.08.009) . Note that they didn’t collect any data but are commenting on an earlier paper. That earlier paper was  “A negative Flynn Effect in France, 1999 to 2008–9” Edward Dutton and Richard Lynn (http://dx.doi.org/10.1016/j.intell.2015.05.005). Now they also didn’t collect any data either.

In fact, the actual data collected was by test publishers – specifically the producers of the Wechsler Adult Intelligence Scale. That test is updated on a regular basis and for a given country & language a new version of the test need calibrating against the population it is being used for. In this case version IV of the test (WAIS IV) was being standardised in France in 2008/9 from the previous version WAIS III that had been standardised in 1999. While the WAIS IV was being standardised using a sample of 876 people, they also had a smaller sample take the earlier version as well (i.e. some people sat both WAIS III and WAIS IV) so the test producer could compare any differences.

They key question then is how many people were in this smaller sample. The answer is SEVENTY-NINE. Put another way ‘not a lot’. The Lynn & Dutton paper is fairly open about this:

It might behoove us to be more cautious in reaching conclusions based on these results than based on the other studies cited for two reasons: the sample (N = 79) is a relatively small and the WAIS IV manual does not tell us the degree to which it is representative of the French population in terms of variables such as education or geographic region. Clearly, it cuts out those who are under the age of 30 years or over the age of 63 years, but its average age (45 years) is approximately similar to the median age of the French population, which is 42.4 years as of 2014 (Central Intelligence Agency, 2015). In addition, the Full Scale IQ on the WAIS IV sample of 79 subjects was calculated based on a comparison with the WAIS IV sample of 876 subjects, which was representative of the French population on key variables such as education and region. The scores of this sample of 876 subjects were set at 100 and a comparison made with the sample of 79 subjects. As can be seen in Table 4, on this basis the IQ of the sample of 79 subjects was 101.1 with an SD of 14.7, where the French norm would be 100 and the SD 15. As such, the smaller sample can be regarded as representative of the French population in terms of intelligence.

The issue is the sample size is a limit on how representative a sample can be. With 79 people spread around the mean, there can only be a few people outside of one standard deviation of the mean (about 25 altogether or roughly12/13 each for people below or above 1 SD). Likewise, the sample is unlikely to have many immigrants or people from ethnicities other than the majority.

Nor was this decline consistent across all the subtest. The WAIS tests estimate IQ by a series of different subtests which present quite different styles of tasks.

Secondly, the results for France, for the subtests given in Table 3, show substantial differences in the rates of the decline of different abilities. The largest declines were in Vocabulary (.43d), Comprehension (.32d) and Information (.34d) and the results in Table 4 confirm these by showing the largest decline of 4 IQ points in the Verbal Comprehension Index. Table 3 also shows that Symbol Search was the only subtest that did not show a decline but registered a small increase (.05d). In the Symbol Search test the examinee visually scans two groups of symbols, a target group (composed of two symbols) and a search group (com- posed of five symbols), and indicates whether any of the target symbols match any of the symbols in the search group. The score is the number of correct responses obtained in 2 min.
Thirdly, the results show no change in the Digit Span subtest. This confirms the conclusion of Gignac (2015) that there was no change in forward or backward digit span in the United States over the 85 years from 1923 to 2008. The present results also show that there was no change in the Working Memory Index of which digit span is a component.

So the Lynn & Dutton study already needs a hefty chunk of caveats and note that this is a study from researchers inclined to see IQ as a measure of intelligence, and to see IQ as having a genetic component and to see the notion of national comparisons of IQ as meaningful. In other words about as sympathetic to Vox’s take on IQ without being so nutty as to be incapable of getting published.

Even so in a broader discussion Lynn & Dutton are dismissive of immigration being an explanation for findings of IQ decline more generally:

These immigrants are likely to have had some impact on reducing the average IQ of the populations, but it is doubtful whether the increase in the number of immigrants with lower IQs has been sufficiently great to have had a major effect. For instance, in Norway it was shown by Sundet, Barlaug, and Torjussen (2004) that immigrants comprised approximately 2–3% of their conscript samples and that these would have reduced the IQ by only around 0.1–0.2 IQ points (correspondence with Sundet, quoted in Dutton, 2014). In addition, Dutton and Lynn (2013) have observed a decline in IQ scores among Finnish military conscripts from 1997, despite a negligible number of non-Europeans in Finland of the appropriate age at that time. Furthermore, increasing numbers of immigrants with lower IQs than the host population has apparently had no effect in reversing the Flynn Effect in the United States

In other words, even assuming the dubious premise, immigration just isn’t big enough to cause the observed effect. This is reiterated in the second paper by Woodley and Dunkel:

Replacement migration in France involving populations exhibiting lower means of IQ and higher rates of total fertility, such as Algerians, Moroccans, Tunisians and Roma (Čvorić, 2014; Lynn & Vanhanen, 2012) may be increasing the rate of secular losses at the level of g, consistent with speculations advanced in Dutton and Lynn (2015), however the additional loss in g due to this process is anticipated to be very small.

Again, researcher’s sympathetic to biological and national explanations of IQ and IQ changes find that immigration is not sufficient to explain the finding.

So even granting all of these…

  • The overall dubious premise of IQ as a measure of innate intelligence.
  • The additionally dubious premise of national IQs and comparison of national IQs.
  • The inconsistent ‘decline’ in different mental skills tested.
  • The small sample size.
  • The lack of any details about the sample itself (to either the readers or the writers of either paper).
  • That the sample and the testing done wasn’t devised for the purpose it was then used (i.e. the researchers have to assume that the data makes sense for the analysis they performed, that the sample was appropriate etc etc)
  • That the methods used by either set of researchers were sound (I assume so but remember that Vox is hyper-sceptical of peer review and I doubt he is acquainted with the technique used in the paper he quoted).

…neither paper ends up agreeing with Vox’s conclusion.

What is frightening about this decline in French IQ is how rapidly it has taken place…Immigration isn’t just bad for a nation’s economy, it is horrifically damaging to a nation’s prospects for the future. https://voxday.blogspot.com.au/2016/07/the-enstupidation-of-france.html

Nope. What is frightening about this research is how it is so easily exploited by the far right to demonise immigration, even when researchers draw quite a different conclusion.


7 responses to “@voxday gets it wrong on IQ (again)”

  1. Here’s an increasingly notorious quote from Aaron Banks, one of the significant people who helped to bankroll UKIP and the Leave campaign in the UK referendum:
    “It was taking an American-style media approach,” said Banks. “What they said early on was ‘facts don’t work’ and that’s it. The remain campaign featured fact, fact, fact, fact, fact. It just doesn’t work. You have got to connect with people emotionally. It’s the Trump success.”
    I think that pretty much sums up the entire problem in a nutshell. Find something that sounds as though it fits your emotional argument and don’t worry if the supporting material doesn’t actually support it, because no-one is going to look at that.


  2. Is the penultimate quote saying that even _if_ you accept the idea that these immigrants to France are x amount of IQ lower than the French population, the amount of x and the % of the population they represent isn’t capable of creating the observed drop?


    • Yes and the authors are accepting that [personally I think the notion of national iQs is weak]- and similar drops have occurred in other countries with low immigration and not occurred to the same extent in countries with high immigration. The ‘biological’ line being suggested here is a ‘dysgenic’ one – i.e. the claim that people with lower IQs tend to have more kids than people with high IQs, so countries should get less smart over time. The effect does not match experience of course, hence people hunting around for it where they can.


    • Ironically, if that group of 79 was disproportionate in the number of recent immigrants then various things make more sense. In particular one of the subtests that showed the biggest difference was Vocabulary. Now that is one that is g-loaded for native speakers of the target language but (obviously) not so for people for who the test is not their first language. Arithmetic on the other hand (less language sensitive) showed little change.

      Liked by 1 person

      • I was reading the vocabulary part & thinking the numbers have little meaning if they are using French to test immigrants where French might be their 2nd or 3rd language.

        One if my complaints with IQ tests is dominant cultural bias and class are built into the questions of all the ones I’ve seen.


Blog at WordPress.com.

%d bloggers like this: