America has already differentiated into castes

India: 95% within-caste marriage - America: 93%

I’m writing this because we’ve obviously been differentiating into castes in the USA - much like all good things (height, attractiveness, reaction time, life satisfaction, etc) end up being positively correlated with “g,” there is a cluster of people who are the exception to the “80% of Americans are overweight or obese” rule because they exercise and watch what they eat, AND have good careers and are educated and are more conscientious and have better time preference and own real estate, and so on.

And just so we’re not playing with subtext here, I’m going to use a definition of castes that’s defined by 1) distinct cultural mores, 2) strict intermarriage, and 3) ritual hierarchy.

What are the cultural mores? When and what you eat, whether you smoke, whether you have a college degree or not, your occupation, how many hours you work, how you spend your leisure time, whether you get married or not, whether you own a home or not, and more.

But those things aren’t independent!

They are not independent variables, of course. They have a correlation matrix, probably a very high “r” correlation matrix - and that’s my entire point here!

If you look at things, and there’s this massive cluster of “all the good things” that are all strongly correlated, and the population medians and means are significantly different across those things, and people only mate and have kids within those clusters, what you have is a caste.

The primary caste division that I’m pointing to is often called “the Professional Managerial Class (PMC),” and that’s a good enough label for the central division.

Essentially, we see roughly ~70-80% of the population that largely don’t have undergrad degrees, are fatter, shorter, have fewer marriages, work fewer hours per week, and own less real estate.

And then we see the PMC, which has undergrad and grad degrees, high marriage rates, high home ownership, taller and fitter men, longer work weeks, and skinnier and more educated women.

Really, you could probably leave it there - at two castes, a Pareto split in the US population. But we know that doesn’t tell the whole story - people with grad degrees, *really are* more likely to marry other people with grad degrees. People at higher incomes assort with people from similar backgrounds.

So you do get some lift from finer divisions, and even those are fairly well separated in terms of our definition.

But where’s the evidence, smart guy?

Why, when *I* look around, everyone looks pretty much like me, so I don’t see this huge differentiation.

And yes, I agree! When I consider my friends, my family, the people I see when I walk the dogs or train at my gym, I largely only see people like myself, too.

But that’s a function of two things:

First, selection effects - as Scott has written about, the degree to which our workplaces, social circles, and families are selected is extreme - in his chosen example on a single metric, creationism, the odds against a “random draw” are 1 / 10^45 for him knowing zero creationists, an absolutely bonkers number. But, as all of us know, this is actually extremely common and the least of the extreme correlations we see in our social circles.

And then think of all the OTHER things your social circle is extraordinary on:

In our ACX commentariat circles, a lot of us have post-graduate STEM degrees (<4% of the population), make more than 6 figures (<15% of individuals), own homes in Tier 1 MSA’s (<10%), have friends that exercise regularly (<10%), don’t primarily eat fast and ultra-processed food (<30%), and so on.

The second reason “everyone looks pretty much like me” in our lives is salience. Much like when you buy a certain marque and model of car, suddenly you see it everywhere, you only really notice people you consider relevant to your own social domain. Even if you’re shopping or buying gas or spending time somewhere you usually don’t, so less filtered and selected, you largely only notice other people in your own caste.

This is because even if you see people unlike you driving around, shopping, buying gas, spending time at the water park with their family, you don’t notice them, or use them as a basis of comparison with yourself.

Just to use a fairly recondite example, I don’t even notice or look at cars that don’t have at least 500hp when I’m driving, because my fun car puts down 4 digits of power. This means I ignore probably 99.99% of cars on the road as completely irrelevant to me, and only really sit up when I see / hear an exotic, a seriously modified car, or a Hayabusa or something like that.

Well, similarly you just don’t notice people outside of your caste, even if you happen to (uncharacteristically) be somewhere that they exist.

Show me some numbers!

Right, so let’s just take it from the top:

Broadly, I’ve looked at men ages 25-65 in aggregate at a few household income thresholds.

This comes from a lot of data sources - Census ACS, NHANES, PSID, Bureau of Labor Statistics, and more. If you’re interested in those data sets yourself, or in double checking me, see the footnote, where I’ve put a link to my jupyter notebook.¹

But what we see is a *very* clear differentiation, where all good things (income, height, BMI, spousal BMI, educational attainment, number of confidantes, being generally in good health, real estate ownership, wealth, etc) cluster together, and tend to cluster harder the more you go up the ladder.

There are obvious reasons why income would be correlated with wealth and real estate ownership, but good health? Spousal BMI? Good diet? Number of close confidantes? Marriage and divorce rates? Recreational time spent on screens? I guess in retrospect, most of these ARE obvious.

I really tried to find a source for “was a high school / college athlete” by these, and couldn’t find it, but I’d bet even THAT is highly sorted among the castes.

As we know with “g,” all good things are correlated.

Please lord, give me temperance and caste-ity, but not yet

What are the most caste-separation-like markers here?

• How about once you’re past $170k in HHI, you’re between 70% and 92% likely to have a spouse with a college degree?

• That a top 1% man works nearly 1k more hours per year than a “below median” man, and 800 more hours per year than a median earner.

• That the only ones with even a CHANCE of being non-obese themselves, or having a non-obese spouse, are the top 5% or better men by earnings?

• That PMC men have ~4x lower smoking rates than non-PMC men?

• That PMC men spend at least a half hour less staring at their phones per day than non-PMC men?

• That ~50% of PMC men vigorously exercise (while spending 30-50% more time sedentary on average than non-PMC men, likely due to blue / white collar divides).

• That PMC men are roughly twice as likely to be married?

• One bummer? Being PMC doesn’t actually lower your chances of divorce that much - you’re about 60-80% the rate of non-PMC divorce rates (which as anyone knows, are crazily higher in America than they are in Europe or Asia).

Caste-ing call

Broadly, these map to six archetypes. Just for fun, I’ll make up visual archetypes for each of them.

The <$38k bottom quintile have the lowest incidence of six-footers, the most smokers, their education stopped in or capped out at high school, they’re un-or-under-employed (working only ~20 hours a week), and are (unsurprisingly) the least married, least exercising, and least home owning segment.

The $38 - $88k Median guys are on average obese (35%), their education largely capped out at high school, they are marginally under-employed averaging 34h per week, and spend a little more time per day on their phones, have a few more friends, see friends the most on a weekly basis, and have better life satisfaction than the bottom quintile.

The $88 - $170k Above Median gents are still obese (32%), about a third of them have college degrees, they work a full work week, they moved up pretty noticeably on life satisfaction and number of friends, and are probably a mix of physical jobs, service jobs, and entry level white collar jobs.

The $170k - $340k Mid PMC’s are STILL obese (30% obesity rate, this actually genuinely surprised me), they have undergrad degrees (60%), they work PMC white collar jobs for 42 hours a week, and spend a half hour less per day on their phones.

The $340k - $500k top 5% High PMC’s are taller, skinnier, have skinny spouses who also work, grad degrees, work FAANG and finance and executive PMC jobs for 45+ hours a week.

The top 1% - Probably a mix of FAANG and finance people, executives, business owners and entrepreneurs, with your odd media personality or elite athlete thrown in. They’re essentially all married and homeowners, are the tallest, most fit, and the most educated segment, they have the skinniest, most educated spouses, the longest work weeks, the least recreational phone time, more close friends but less time to spend with them, and slightly worse divorce and life satisfaction stats than the top 5%.

To find a maiden pure and caste

One truth we’re told is apparently universally acknowledged: an eligible bachelor is in need of a wife.

Now remember, marriage itself is already a differentiator - PMC men get married at 80-90% rates, whereas median-and-below men get married at 20-50% rates.

So the degree of separation you’re going to see here is already skewed, by carving out the best of those median-and-below men to even be in the “married” comparison.

How can I so confidently assert that?

One of the more interesting posts on this front was Cremieux’s deep dive into whether the “marriage premium” was driven by marriage itself causing men to up their game, or by the best men being selected into marriage - spoiler alert, it was selection effects (it’s always selection effects!):

Are All the Good Men Married?

So remember, “intermarrying within the castes” is a key part of the caste definition - so do we see that in the data? And how!

From Horowitz & Keller (2022), A comprehensive meta-analysis of human assortative mating in 22 complex traits:

“Assortative mating (AM) is the phenomenon whereby individuals with similar trait values mate with one another at levels higher than expected by chance¹. Contrary to the maxim “opposites attract,” nonzero phenotypic correlations between human^2–21 and nonhuman¹ mates are overwhelmingly in the positive direction, with only a handful of examples of disassortative mating, or negative mate correlations, reported in the literature.”

With their headline graphic:

But we can do better than this - we already saw that the educated end up with the educated, that the lower BMI and higher income men end up with lower BMI spouses, but we can clearly see that the degree of assortation / differentiation has been steadily increasing year on year for decades:

It’s largely not secular trends

And just as a point of comparison, I can show you that this is largely not a result of underlying secular trends, because look at these at the individual level for the higher income green and red bands - look at these baselines, then look back up at the spousal graphs above. You see they are differentiating more from the lower income bands versus the secular trend baselines:

Do you see similar demographic separation if you cut just by education? Say, HS / College / Grad? But of course!

I did look at that, too. But you lose about half the differentiation / effect size in terms of clearly separated population characteristics, and you only get a tiny bit more assortation (on education, for instance, a 53% incidence of a spouse with grad degree if you have a grad degree, vs 44% if you make more than $250k HHI, or 49% at $500k+).

The die is caste - ritual hierarchy

So far, this just sounds like a “class” argument, not a “caste” argument. It’s purely money driven if we’re cutting by income buckets! But my point is that it is NOT purely money driven, money was just the most granular-but-still-separating dividing line in the data. You see the exact same clustering if you divide by coarser buckets like HS / College / Grad degree.

Moreover, think of yourself! I would bet that your friend circle is dominated by people with similar educational attainment as yourself - either “college / college” or “grad / college or grad,” and that if you exercise, your social circle exercises, and that you and your circle don’t smoke, and only drink moderately, and so on.

This hasn’t even touched politics, which is insanely sorted:

But I don’t even need to bring in politics to demonstrate a 90%+ matching! I can do it with just two variables - education and income.

This is certainly a broader cultural phenomenon, too.

After all, who did the infamous Epstein associate with? Rich guys AND prominent academics - why those two clusters? What do academics, who generally cap out at around $250k income, have to say to rich guys, who make many times that, if this was purely a class division thing?

Why do rich guys, as we just saw, have more educated wives the richer they are?

If I were to sketch the following profiles, would you be able to very clearly put them in the appropriate “who will marry into whom’s family” buckets?

• Larry is a successful electrician who makes $800k a year as owner of his own electrician business. He drives a lifted and heavily modified pickup truck with big tires and a loud muffler, he listens to country music, his favorite hobbies are fishing and barbecuing with friends and family.

• Ted got a Phd in economics and works at a FAANG as vice president, and makes $700k a year. He drives a Prius and listens to classical music and audiobooks. His favorite hobbies are triathlon and reading.

• Geoff got his undergrad degree in Russian Literature. He doesn’t own a car or TV, and bikes to his job as a manager of a Starbucks, where he makes $70k a year. He listens to filk and Radiohead, and his favorite hobbies are making craft beer and mountain biking.

Okay - does a Ted-son ever marry a Larry’s daughter, despite the much closer match in income? No, he marries a Ted’s-daughter-equivalent or one income tier down roughly ~70% of the time, and sometimes a Geoff’s-daughter-equivalent another ~15%.

This is very much a “ritual hierarchy” division, not a purely class / economic division, which you can clearly see in the data.

You can clearly see 3 homogamous income clusters there in the chart.

And in the cases where there’s a mismatch between husband / father income tiers? The wife either comes from more wealth or has the same or higher educational attainment than the husband between 69 - 83% of the time that happens (and presumably the ~20% of the time neither is true, she was super hot, which is regrettably not investigable in the data).

Homogamous matching is by far the norm:

And in fact, this notably undersells it! Sure, the middle income buckets here have the most homogamous matching - but that’s because they’re easier to match! There’s a full ~47% of the population in those two buckets!

Because higher incomes are intrinsically rarer (ie 5% or 1% incidence), if you control by the base rates, the amount of effort being exerted to see the given degree of homogamy in each bucket is higher and higher the more you go up the “income” ladder.

“Below Medians” being so selected is largely a matter of the higher buckets pushing them down there via exclusion, coupled with lower overall marriage rates (ie you need a more-matched-than-usual couple to want to get married on a ~20% baseline)

Once again, and as has routinely amazed me any time I dive into Greg Clark-related idea spaces,² “spousal quality” is just insanely optimized, undoubtedly THE single most optimized decision most people make in their entire lives.

Only ~20% of the time will there be a mismatch on both education and income, and in those cases, between 67% and 87% of them (depending on where you cut) are mismatched by only ONE income tier!

To put that a different way, a full 90 - 98% of the time, with a weighted total average of 93%, our castes are essentially matching on income or education, or are at most only one income bucket down. That’s really high assortation!

The literal Indian caste system, which alone among the world has achieved Greg Clarkian “persistence rates” in child-parent statuses above 75%, was recently measured at ~95% within-caste matching in marriages (NFHS-5 conducted between 2019–2021).

American caste system confirmed!

What stood out to me most here?

That amazingly, you have to climb past 95%+ of people in the US to reach a tranche of people with reasonable, non-obese (but still overweight) BMI’s, non-overweight spouses, and to reach ~40% of people having grad degrees and ~80% having college degrees overall.

I feel like this is literally everyone I know, and everyone *they* know, out to the third degree at least. And apparently that’s <5% of the total population, even though it’s, oh, approximately all of Manhattan or SF, or most of NW DC and the DMV suburbs, and most of any rich suburb in a Tier 1 metro area. So I’m not sure quite how to bridge that disconnect, beyond handwaving at the selection and salience points I was making in the beginning of the post. And ~5% of adults nationwide is ~13.5M people, so you know - if you were sprinkling 13.5 mega-people around the country, primarily concentrated in Tier 1 metro areas, I guess that’s pretty much the distributions you’d expect to end up with.

It’s only here at the <5% threshold that the aggregate statistics even start representing what I think the ACX / SSC commentariat centroids might show up as, which is pretty extraordinary,³ and a great reminder to be grateful for the deliciously concentrated places like ACX that sort people so well and give them high similarity and large volumes of confreres, strained from a meager <5% slice of the larger population.

1. Jupyter link: https://drive.google.com/file/d/1z19L3h-MgqbjADrvAuCfiD-p7rR4QOZz/view?usp=sharing

2. I didn’t even have to dip into occupational status or wealth to hit ~90% caste intermarriage rates here, either, which are the other 2 Clarkian endpoints in the “persistence rate” endpoint he measures.

3. Although if we take the mean ACX IQ as measured in Scott’s surveys at face value, we’re starting at an already <=2% slice of the population. And as we just saw, approximately all good things cluster together, so if you’re at the 98th percentile already on one measure, you’re probably pretty high up on a lot of others, too.