9 Comments

Is the 8th grade exam supposed to be impressive? It looks quite easy, and at least the math (arithmetic) sections looks to be at a lower level than what current 8th graders are taught in most Western countries.

Expand full comment

Thanks for the interesting article Peter.

I'm curious, your conclusion appears to be that the Flynn effect is in essence not "real" (i.e., not measuring what we really care about w.r.t increasing IQ) at a population level.

Then does that suggest that at the individual level one could improve their IQ? If so, isn't this a highly controversial claim in your field?

Further, wouldn't it suggest that IQ is becoming a worse rather than better at approximating somebody's level of intelligence?

Thanks again for the interesting piece.

Expand full comment
author
Aug 27·edited Aug 27Author

Your first question: It depends on how much experience one has had with standardized written tests. If one has had no experienced at all, then practice and coaching would certainly improve one's IQ score.

"In the present research two studies are used to investigate the relation between g loading of tests and practice (test-retest) and coaching (active teaching) effects. The data on practice do not support the hypothesis that the higher a test’s g loading, the less susceptible it is to preparation, but the data on coaching support the hypothesis. There is evidence that practice and coaching reduce the g-loadedness of a collection of tests." https://onlinelibrary.wiley.com/doi/abs/10.1111/1468-2389.00182

Your second question: I would say "better." People are spending more years in school (secondary + post-secondary), so most people have become familiar with standardized written tests. This is why the Flynn effect is leveling off. Fewer and fewer people are completely unfamiliar with them.

Expand full comment

There's a cottage industry around people taking IQ tests over and over again with more and more test prep so they can brag about the one time out of ten they got 120

Expand full comment

Peter Frost,

I give you proper credit for engaging with my comments and actually winning the argument with me after I left comments on this same article when it was first published weeks ago in the Aporia Substack. You argued that it is ambiguous whether or not skull volumes have been getting bigger, and, though the single dimensions of the skull (skull shapes) have most clearly changed in singular dimensions (narrower, longer, taller), they are not necessarily in the direction expected by increasing intelligence. The "taller" dimension would be expected for increasing intelligence, but if anything skulls would be expected to be getting shorter and wider, not longer and taller as the true trend shows. So, now i just don't know what to make of those skull trends. I hand it to you.

I will take a different argument now, one that is more directly relevant. If a large battery of tests show an increase in scores, then the "default" (perhaps the Bayesian prior) should be that this represents an increase in intelligence. Small exceptions exist, like the practice effect. Very large exceptions, as a hallow Flynn effect of 30 points would be, seemingly do not exist.

For almost everyone else, such an exception at hand would be: "Oh, but what about the phony IQ difference between Whites and Black Africans?" But, racial hereditarians don't have that exception at hand, and neither would I count it as an exception. It is a real intelligence difference, much like the intelligence difference between generations, because it is not just IQ scores but also educational attainment, literacy, numeracy, GDP per capita, longevity, fertility, height, myopia, brain size (though with uncertainty as you highlighted), and more such relevant variables almost all going in the same direction.

The adult height trend is an interesting one: an increase in up to two standard deviations in some populations in spite of being 75 to 90 percent heritable. It would be a puzzle if true intelligence were left behind among such trends.

I will stop there before this comment balloons into an article in its own right. Thanks.

Yours,

Abel

Expand full comment

Thanks, Peter—very good article and an excellent summary of some of your prior writings as well on the evoandproud blog.

There are some linguistic studies such as Reali, Chater, and Christiansen 2018 that suggest language to decrease in complexity according to population size; complex linguistic traditions can only be sustained and passed on to succeeding generations among smaller populations. There are a few other studies that reach similar conclusions. I wonder how this might tie back into verbal IQ and cognitive development as a whole.

As an explanation, it seems simple and sensible enough to me for something I found confounding when I first realised it studying ancient languages as a teenager, which was that they are far more complex and powerful than any modern languages I've seen.

An interesting exercise is to compare English common in literature of the early 20th and 19th centuries to the English found in literature of today. The grammar and vocabulary are far more complex, enough that I've seen teachers say that young people now often have tremendous difficulty understanding it. Translations of ancient languages such as Latin and Attic Greek are far simpler now than in the 19th century, and comparing to the increasingly hard-to-find originals, far less accurate I would say due to the loss of English complexity and features considered archaic.

I wonder often how this loss of linguistic complexity in English, and all widely spoken European languages I'm familiar with, is affecting our verbal intelligence and cognitive development? My long held suspicion has been that English is in the process of beginning a split into multiple languages like Latin once did, as it already has several dialects and forms that aren't very intelligible among native speakers. It seems this would actually be good for cognitive development in the long term. However, I think communications technology might prevent this from occurring, as younger speakers lose native dialects and adopt to the 'prestige' dialects.

Supposedly, screen usage is bad for neurocognitive development regardless of content (!), or so a few studies I've read from a scientist called Stoyan R Vezenkov suggest. I would be very curious to see some of these intelligence studies repeated on younger adult populations that may have been most impacted by this. As always, there's a need for much more research!

Expand full comment
author

Yes, most languages have shown a decline over time in the morpheme/word ratio, i.e., they have shifted from synthetic to analytic. That's a long-term evolution that may be linked to increase in population size.

But the simplification of English looks like something else. It's much more recent, has largely occurred over the past century and seems, if anything, to be accelerating. Whenever I publish an article, the editors often replace the present perfect and the past perfect with the simple past, apparently because many readers have trouble understanding those verb tenses. I see similar changes occurring in French.

Expand full comment
deletedAug 27·edited Aug 27
Comment deleted
Expand full comment
author

It's a bit strange to see opposing results for forward digit span and backward digit span. Backward digit span is more challenging and has a higher g loading, so it should be a better measure of cognitive ability. But forward digit span has some g loading as well.

Your second point confirms my own impression (and thanks for the reference!). There has been a qualitative decline in new patents.

Expand full comment

Some time ago, I wrote an article for Ideas Sleep Furiously (which became Aporia) arguing that the Flynn effect was on both g and non-g, but it was on non-g MORE than on g. My argument was that Jensen's method of correlated vectors can be corrected and fleshed out by inferring from the extensions of the regression line to where g-loading=0 and g-loading=1. With that method, both intersections are positive (albeit with uncertainty for g-loading=1). Jensen's original primary method was to merely infer from the slope of the regression line, which is a faulty argument that can lead to a wrong result when the slope is negative. So, this thing with forward and backward digit spans fits with my argument.

Expand full comment