Math vs Writing: difficulty and threshold effects

I saw this post going viral by Charles Murray about the high attrition rate of college math courses compared to the humanities:

As I discuss in the post post “Why Good Writing is One of the Hardest Things Ever,” I argue the reputation of math being super-hard versus the humanities is undeserved, particularly for writing. When done at a high level, they are both hard, but in different ways. Math requires deciphering the rules and the logical flow of the underlying concept, whereas writing requires decoding the value system of the recipient. This means attention to things such as style and tone, whereas with math, this is less important, only that the results are correct and novel.

Readers will plod through a poorly-written math paper if it contains nuggets of insight that cannot be found anywhere else, having read such papers myself, but the same cannot be said for writing, in which readers will jump ship long before getting to that point. So the whole piece has to be one-hundred percent dialed-in from start to finish. This makes writing less forgiving of imperfection despite being conceptually easier than advanced math.

I’m sure most of us can readily identify individuals in our social circles who are successful in coding or other STEM professions. By contrast, how many people do we know who are successful at writing? I know no one in my immediate social circle ‘IRL’ who is successful in this sense–that is, who earns a living from writing, has received critical acclaim, or has even a modest readership, and I contend that this disparity cannot be explained solely by selection effects. Even if my social circle were expanded by an order of magnitude, I still doubt I would know anyone who meets this criterion. This underscores just how rare such success is.

If I’m talking online, obviously, I’m casting a much bigger net, but still, we’re talking a tiny percentage of people who can do it well. For every couple thousand people on Twitter, maybe only a handful or so are highly and consistently competent at long-form prose. This includes bloggers Matt Yglesias, Noah Smith, Curtis Yarvin etc. These are people whose verbal IQs are easily in excess of 150. Conversely, for STEM, online, it seems there are way more people who are decent at it and who make a living at it. Learning calculus in high school has become so common, it long stopped signifying anything special or being brag-worthy. Everyone is doing ‘vibe coding’ now.

So in practical terms, a quantitative IQ of around 110-125 is good enough to learn calculus in high school, write some code, get a decent STEM job, etc. But a similar verbal IQ means only being semi-competent or merely proficient. The threshold for true mastery of the craft is closer to 140-150 or so. And someone with a ‘math IQ’ of 110 can learn algebra 2 (e.g. quadratic formula) and maybe some plane geometry (conic sections), which is not bad. But with a verbal IQ of 110, they cannot converse well beyond short-form text. They struggle at writing anything longer than a few sentences, and what little writing that is produced is poorly done.

In regard to the above tweet, what can explain the high attrition rate of STEM? The ‘math IQ’ threshold to pass those courses is closer to 120-130 (e.g. multivariable calculus), which excludes most students. The verbal IQ is lower because students are being graded on the ability to write passable papers and regurgitate information, not produce high-quality prose for discriminating audiences as seen online. Rather, the writing need only be adequate to satisfy the course’s minimum standards. I recall this myself in college, in which my humanities papers amounted to workmanship writing to meet the requirement of the assignment. Provided you turned in words on a piece of paper, it was close to impossible to fail.

An obvious counterargument is that the humanities have a lower consumer demand compared to STEM, so fewer people can make a living at it. The market has a smaller ‘carrying capacity’ for top authors, playwrights, or bloggers versus competent coders. This requires a higher threshold of talent for success and hence more rarity. The IQ threshold is too high, so consequently the vast majority of people will be shunted to professions requiring verbal IQs around 100-110. This is good enough for writing memos or briefs. But when going beyond this, the result is the dreaded ‘LinkedIn slop’.

Most content is mediocre because most people by statistical certainty must be mediocre. This is like watching people play a pick-up game of hoops and wondering, “Why can’t they play at the NBA level? How mediocre of them.” LinkedIn is extremely popular, with 1.1 billion users worldwide. Thus it’s much more representative of the general population, compared to a platform like Substack, which has 50 million users and only 17,000 writers. People who are really good at writing and have incisive insights, like those writers who make a living on Substack, are way outliers. They are probably in the top 1/2-1% of verbal ability, compared to the median LinkedIn user.

LinkedIn writing is closer to a verbal IQ of 110-120, versus 140+ for competent Substack writing. The difference is stark.

Moreover, the difficulty of verbal ability is evident by how so many high schoolers and even college students struggle at writing and reading proficiency, and are dependent on AI and other tools for completing even the simplest of tasks, like writing emails. Teachers commonly wince at having to grade essays. Or just read the comments of any online community that isn’t self-selected for high-IQ (e.g. medical forums, because people of all IQs get sick) and you see the full range of competency at writing, which is to say, there isn’t much.