Scott Alexander at Astralstarcodex just posted his monthly dump of website links. These are always a grand hoot. The below caught my attention:

The GSS doesn’t have an IQ test. What is has is the “wordsum” item, which is basically a vocabulary test. Here’s how the GSS describe’s the item:

“We would like to know something about how people go about guessing words they do not know. On this card are listed some words--you may know some of them, and you may not know quite a few of them. On each line the first word is in capital letters -- like BEAST. Then there are five other words. Tell me the number of the word that comes closest to the meaning of the word in capital letters. For example, if the word in capital letters is BEAST, you would say "4" since "animal" come closer to BEAST than any of the other words. If you wish, I will read the words to you. These words are difficult for almost everyone -- just give me your best guess if you are not sure of the answer. CIRCLE ONE CODE NUMBER FOR EACH ITEM BELOW.”

Here’s an example of what the items look like.

Here’s a nice paper on the topic. Seems like wordsum has been used to refer to constructs such as: verbal ability, intelligence, vocabulary, cognitive sophistication, knowledge of standard English words, linguistic complexity, and knowledge and receptivity to knowledge.

So maybe this measures a person’s general intelligence. Or maybe it measures a person’s verbal complexity, vocabulary, knowledge of standard English, etc.

Also … hasn’t IQ increased pretty significantly across cohorts, via the Flynn effect? So why are we seeing the opposite here?

I was under the impression that IQ stayed relatively constant, or maybe declined a bit, as you age. I did some very shoddy googling and found a not-so-sketchy website that had the following figure:

So fluid intelligence, things like reasoning skills, decline with age, and crystalized intelligence, like…knowledge of words…increases over age? (I’m glad the declining fluid intelligence doesn’t apply to 30-something sociology professors. That’d be dreadful).

So let’s see what wordsum does across age. I’m just using the entire GSS 1972-2021, using survey weights, and computing local polynomial smooth plots to eyeball trends.

General Social Survey, 1972-2021, survey weights used to compute local polynomial. All respondents age 70 and younger.

Well, well well, this doesn’t look like the trajectory of fluid intelligence, does it!? Looks more like crystallized intelligence. That is, wordsum is probably more a vocabulary test, for my money.

Let’s also take a look at this graph. Wordsum scores increase by about a point, or ~10% of the total amount of change possible in the variable, between ages 20 and 40. So the linked tweet might well be measuring age differences and claiming them to be IQ differences.

Let’s look at the wordsum scores by age, by generation:

General Social Survey, 1972-2021, survey weights used to compute local polynomial. All respondents age 70 and younger.
1901/1926= Greatest Gen, 1927/1945= Silent Gen , 1946/1964= Baby Boomers , 1965/1980= Gen X , 1980/1995= Millennial , 1996/2003= Gen Z

What is there to write home about? It seems like there is not much difference at all among greatest, silent, and baby boomer cohorts. Gen X’ers converge by ~ 50, while Millennials seem to have fallen behind in their 30s (what a rude graph!). But the biggest gap we see is in, like, one decade of about 0.5 points. That’s hardly remarkable.

But here’s another thing: aren’t younger cohorts more diverse? And have more non-native folks? I’d assume that a growing proportion of ESL folks, as well as marginalized racial minorities, might introduce contrasting trends among these results. So let’s look at them separately by race. Sadly, I only see a racial variable that's “White, Black, and Other.” Let’s admit the major shortcomings of “other” and look at these groups separately.

Heck yes, younger cohorts!! There’s barely any difference across white cohorts. It looks, to me, like black and other groups are increasing wordsum scores substantially across cohorts. In fact, it looks like there’s a smidge of convergence among Gen Z’ers. Woohoo! Go young people!! So I definitely think that there’s a lot of confounding going on, not just by age, but also by the issues that may arise with administering a 10-item vocabulary test to an increasingly diverse population.

There’s also been an increase in educational attainment over time. Let’s check that out. First, let’s look at the proportion of the GSS with a college degree across cohorts. We’re looking at folks aged 25+, so no Gen Z’ers.

General Social Survey, 1972-2021, survey weights used to compute local polynomial. All respondents between 25 and 70.

Big time educational expansion. College degree holders grew from 10% to 35% for white folks, 10% to just under 30% for “other” folks, and 5% to 20% for black folks. So massive expansion of education. What’s happened for folks with less than a college degree?

General Social Survey, 1972-2021, survey weights used to compute local polynomial. All respondents under 70

I see no change among white respondents, increasing wordsum scores among black respondents, and noisy growth-ish among “other” respondents, for folks without a college degree. So, improvement compared to what we saw in the original motivating figure.

What about college+ groups?

General Social Survey, 1972-2021, survey weights used to compute local polynomial. All respondents aged 22-70

For black and other groups, there’s no change across cohorts. We see a sizable decline of around one point among white cohorts, from around 8ish for Silent Gen folks to maybe 7ish for Gen Xers and Millennials. Not nothing. But considering the massive expansion of college education, a 1 point decline or no change is a pretty marginal shift compared to the 3x expansion of the college group. That is, much more selection in older cohorts. Who wants to bet that the top performing 1/3 of younger college cohorts would have better wordsum scores as the Silent generation groups?

So here are my big takeaway conclusions:

The graph Scott Alexander linked to probably reflected mostly the point that older people are older than younger people.
The graph probably represented the fact that a much larger share of folks go to college now than in the past.
Younger people are probably, at minimum, not dumber than their older cohort counterparts (I count people younger than myself as well).
Nevertheless, young people still need to improve on completing their readings before class and reading the syllabus.