There’s a great blog written by Razib Khan, Gene Expression. Khan and I have pretty different political and theoretical views. But his blog is one of those beautiful opportunities that seem to be ever-dwindling these days of getting exposure to thoughtful arguments from intellectual spaces different than one’s own. He’s also a bookworm of almost Cowen-ian proportions, so it’s a great place to learn about great reads that, if you’re a milquetoast sociologist like myself, may fly past your radar.
Awhile back he had an interesting post: White Conservatives Are Falling Behind White Liberals On Intelligence, where he shows the following graph of mean scores on the wordsum scale (a ten-item vocabulary test that some argue is a decent proxy for intelligence).
What you’re seeing is that through the 1990s, Liberals and Conservatives had similar wordsum scores, but Liberal scores increased slightly in the 2000s and 2010s, while Conservative scores declined slightly.
This is interesting, but I’m a tad skeptical about the reality of these trends. I broadly comprehend the argument that ideological position can align with verbal competence (or broader ideas of ability, intelligence, knowledge, etc.). I’m just not totally convinced that political views specifically are particularly effective predictors of verbal competence. Some of this skepticism likely stems from my broader concern of partisanship driving analysis and argumentation within sociology (how many sociologists does it take to screw in a lightbulb? None, because they’re only able to spin to the left). While I’m personally quite sympathetic to left partisan beliefs, I worry that my discipline too frequently uses left beliefs and intelligence as interchangeable. I think that these overall mean differences run the risk of simply being accepted by academics in positions like mine. So based on these bundles of anxieties, I’m not totally ready to accept this conclusion.
Before moving forward, I want to make sure I can reasonably replicate his results. Here we go!
Yup. Just plotting a local polynomial smoothed line between wordsum and year (using survey weights provided by the GSS) separately by political views (polviews, with moderates as their own group and Very-through-slightly Liberal / Conservative combined into two groups) shows that liberals and conservatives were similar in their wordsum scores through about 1990. Conservative scores haven’t changed much, while liberal scores increased from about 6.5 to about 6.8, and moderates increased from about 5.9 to 6. We now see a gap of about 0.5 or so in the most recent wave.
My ability to precisely identify why I’m skeptical of these descriptive trends is a tad loose, as this question is a bit outside my wheelhouse. I decided to take a look across a few dimensions:
Is this just an age effect?
Is this just an education effect?
Do nonsensical comparisons create similarly sized gaps?
tl;dr. The partisan difference Khan identified is more robust than I assumed, especially among folks with a college degree. But after kicking the tires a bit, I wouldn’t expend too much mental bandwidth worrying about or using partisanship as a meaningful sorter of wordsum / verbal / intellectual ability. I am more concerned about what exactly wordsum is measuring, and the appropriateness of it being used across different eras of the GSS.
Is this just an age effect?
My first hunch: aren’t conservatives getting older, liberals younger, and isn’t age associated with cognitive functioning?
Let’s look at the relationship between age and mean wordsum scores using the whole dang sample of white GSS respondents, 1972-2018:
Yup, wordsum is about 1 point higher between ages 20 and 34 (5.5 to 6.5), and wordsum is about 0.75 points lower between 65 and 85 (6.5 to 5.75 or so).
OK, what does age look like for partisan groups over time?
Conservatives have consistently been older than Moderates and Liberals. Everyone’s getting older, myself included, but it looks like the age gap has grown a bit between the 1990s and today, when the wordsum gap grew as well. So perhaps a larger share of folks over 65 are conservatives and this is accounting for the change over time?
…or something totally different? The above figure shows the percent of partisan groups who are in three big age chunks—under 35, over 65, and in between—which roughly match up with the curve of wordsum scores. We see that Liberals used to be very young: about 1/2 of folks who identified as Liberal were under 35 in the 1970s. We see that Liberals really caught up with Conservatives in that middle age range from the 1990s onward, while a growing share of both Conservatives and Liberals are over 65. So perhaps some of the wordsum changes simply reflect changes in the age distribution of Liberals and Conservatives over time? Let’s look at partisan differences over time without accounting for age, and accounting for it.
In the above, I estimated simple regressions predicting wordsum with partisan differences separately by year. The black line just includes partisan differences among all GSS respondents. The red line restricts the sample to 35-65-year-olds, and includes age contrasts along with partisan affiliation. The y-axis measures the difference between Liberal and Conservative wordsum scores, with negative values indicating higher Liberal scores and positive values indicating higher Conservative scores. Shaded areas are 95% confidence intervals.
There’s a small difference between results that do and do not account for age. It looks like the partisan gap emerges in Obama’s second term, not Clinton’s second term, when you account for age. But the overall pattern remains. I wouldn’t necessarily hang my hat on either conclusion against the other.
In total, partisans are of different ages, and this kind of accounts for the observed trend over time. But not too much.
2. Is this just an education effect?
Perhaps folks who go through higher education are more likely to develop greater verbal skills. Presumably, we are in this higher education racket to help folks become a tad more intellectually developed than when they entered. Here are the differences in wordsum by degree status, by decade:
Not surprisingly, there are big differences in mean wordsum scores by degree status. In any decade, those with less than a high school degree average at around 5 out of 10. Those with more than a college degree average between 7.75 and 8.5.
Wow … what is going on with college and more than college !? If you think that wordsum represents something conceptually meaningful then:
(i) college graduates are not as capable today as 50 years ago (I doubt this)
(ii) expansion of college is lowering standards and degrees are worse signals of competence (perhaps, but I’m a tad skeptical of this
(iii) college educators are doing a worse job educating young people today than 5o years ago (IMPOSSIBLE!!! THE WORST AND DUMBEST OPTION!!)
(iv) wordsum doesn’t do a great job capturing whatever it’s meant to capture today than in earlier decades (my candidate for the best explanation).
I’d assume that there is a higher amount of differentiation in training among higher education today than 50 years ago. Perhaps a single global 10-item vocabulary test is a bit more applicable to an English major than to a Computer Science major?
So do partisans differ in educational attainment, and do differences vary over time?
Yup. There was a convergence in partisan educational attainment through the early 1990s. Since then, we see that Liberals with a college degree increased from about 0.3 to over 0.4, while Conservatives with a college degree increased from about 0.275 to about 0.325. Assuming that we in the biz aren’t totally incompetent in our jobs and that at least some verbal development happens during college, perhaps we see stratification between degrees, not within?
Are there partisan differences in educational attainment? For a variety of reasons, I’m collapsing time into decades in the below graph. It’s mostly to preserve sample sizes for meaningful comparisons.
There is so much weirdness going on here. Let’s review.
There is not much of a difference in wordsum scores among those with less than a college degree. This has been pretty consistent over time.
There is a much larger gap among those with a college degree or more. This gap has been observable since the 1980s, or pushing the partisan gap backward one decade, compared to pushing it forward one decade when accounting for age.
The wordsum gap is wholly among those with a college degree or more.
Here, we see a pretty large gap that has emerged over time. There wasn’t much of a difference between Liberal, moderate, and Conservative college-educated folks in the 1970s.
We see wordsum scores plummet among moderates in the 1970s and among Conservatives in the 1990s.
Let’s be mindful of the large decline in wordsum scores among college-educated folks over time (this is also found among college only / graduate groups separately). These declines are massive. If you think that the emergence of an aggregate 0.4 partisan gap is meaningful, then it’s probably an even more important question to ask why today’s college graduates have scores 0.5 to 1 point lower wordsum scores than college graduates in the 1970s.
Much of the emergence of the wordsum partisan gap likely happens through more folks entering into the higher education category, which has a larger partisan split.
My guess is that we are seeing some oddities among the wordsum scale as an accurate measurement of, well, whatever we are hoping it to measure. I’m not too willing to believe that today’s college graduates are a half standard deviation below those from a half-century ago.
4. Do nonsensical comparisons create similarly sized gaps?
So it seems that there is some difference that is based on age, some that are based on education, and some I can’t explain. But how exactly does the magnitude of the partisan gap compare to real important differences in wordsum scores across major social categories, and how does the partisan gap compare to completely nonsensical comparisons?
I think that education is probably the most meaningful group distinction for wordsum scores. Let’s use a college degree for a comparison to something that is really important for differentiating wordsum.
I would guess that astrological signs are the least meaningful group differences. Fortunately, the GSS includes respondent zodiac signs.
So we see the big descriptive wordsum gap emerge in the 1990s. Let’s look among 35-65 year olds. I ran a simple regression predicting wordsum scores with age, education, partisanship, year contrasts, and contrasts for the respondent’s zodiac sign. Below are the magnitudes of coefficients, with 95% confidence intervals.
I think this is really important to highlight. Liberals are predicted to have on average about 0.3 higher wordsum scores compared to Conservatives, after adjusting for age and education, among the era noted of growing partisan wordsum divides. The difference is statistically significant, and I suppose that 0.3 is kind of a big effect at the population level!
BUT … there’s a statistically significant and similarly sized difference in predicted wordsum scores between Geminis and Cancers. Now, this is produced through a decent bit of p-hacking and motivated reasoning. But you can easily find as big a partisan effect among total nonsense. That makes me more skeptical of the partisan effect as something terribly meaningful.
Additionally, look at that education effect. Folks with a college degree are predicted to have about 1.5 points higher wordsum scores. That seems to be a big and important effect. Partisanship and astrology sign … not so much. I think if we want to focus on partisan differences, we should say that it’s just slightly more important than meaningless zodiac differences. Might be there. Might not. Not the place where you want to be focusing to really understand who scores what on wordsum.
Summary
Are there partisan differences in wordsum? Probably, because it consistently pops up across analyses. I can’t really make it go away. The big questions are the timing and location of the partisan difference. It’s something to be mindful of. But I wouldn’t really write home about it. The differences seem pretty minor. If these trends continue for another 1.5 to 2 decades, then we might start to see serious gaps worth addressing.
I don’t think wordsum is doing what we want it to. I’m skeptical that groups plummeted in their verbal ability in a ten year period. I’m skeptical that today’s college graduates are significantly less verbally competent than those in the early 1970s. I' would instead wager that the wordsum vocabulary test is not doing as good of a job picking up what we want it to pick up in a more highly differentiated intellectual environment of a more educated populace. Maybe there are tricky ways to cook the items to make it hold up better. But I came out of this more skeptical of wordsum than anything else.
I wonder how much partisan splitting there is among college-educated folks. Are liberals more likely to select into fields of study which develop vocabulary knowledge, conservatives practical / mathematical fields? I don’t know. That could potentially explain a lot of what’s going on.