The Great Awokening came very suddenly.

For me, like for many people, it started with Gamergate, in 2014. I grew up on South Park and other edgy 1990s-2000s stuff and had been aware of 4chan and Encyclopedia Dramatica pretty much since they were around. I liked to laugh at internet drama. Edgy humor was not about hatred or being mean, the world is a messy place and it’s better to laugh than to cry about it. At that time, following internet politics was a very fringe hobby. I was shocked when the mainstream media started writing obviously coordinated articles about how toxic and male video game journalism is and Zoe Quinn was given a UN hearing. I didn’t expect internet drama to show up on normie news. After this, the faucet has been turned on, there was all kinds of extreme leftist propaganda everywhere, feminists, race baiters, all kinds of weirdos. Almost overnight, the Tumblr-style fringe leftist narratives I had been laughing at for years suddenly showed up in Windows notifications right next to the time and the weather.

David Rozado’s famous quantative illustration of the Great Awokening. Starting suddenly around 2013-2014, fringe extreme left content went mainstream.

I had lots of reasons to care about this. First, it felt wrong – these were crazy topics and crazy people, it was scary to see serious people talking about systemic racism or toxic masculinity in mainstream media, as if NASA just started seriously discussing flat-earthism. This stuff belonged on Tumblr. Also, there was an element of suppression in woke propaganda – they were saying that people with the wrong opinions should be cancelled, or worse. Woke also had an obviously wrong idea of how society works but they were pushing politicians to act as if, for example, all disparities were really the consequence of discrimination. So the Great Awokening felt like an attack on the world I knew and it felt appropriate to become more politically active and try to find a way to resist it.

I bet that this is exactly what happened to a lot of people reading this, and that they, like me, found Donald Trump very likeable in 2015-16. Trump was not presidential and not even really believable as a conservative leader. He had affairs with prostitutes, gaudy gold decorations in his skyscraper, and a lifetime of business with liberal New York elites. But he seemed like a red-blooded old school guy whose heart was in the right place. You wanted him to be like this because his opposition was so terrible. This went on all the way until 2024. You saw the mass media censorship, BLM, “trans” children, the bald guy in the Biden government who was some kind of dog fetishist, de-banking, Johnny Depp’s fake rape accusation by an obvious BPD nutcase, and you wanted the guy on the other side, the one they were attacking, to be better. You were willing to look the other way about cringe boomer stuff, Truth Social, gold statues, Mad Dog Mattis, and the ineptitude of Trump’s first term. At the very least, a lot of Americans certainly were, so Trump was re-elected in 2020 2024. He arguably ended wokeness, but he turned out to be just another geriatric boomercon. He didn’t end the Ukraine conflict, he started ridiculous tariff wars with allies over Twitter and launched another forever war in the Middle East for Israel, this time likely with real economic consequences even for Americans due to the closure of the Strait of Hormuz. He did this for no other reason I can think of than to swing trade on his tweets and to please Benjamin Netanjahu. Liberals could have avoided this with a little self-control. You sow the wind, you reap the whirlwind, you push child trannies, you get ICE raids, because the people you alienate will forgive your opposition a lot of craziness for the simple reason that they are against you. Liberals get the right-wing they deserve.

Orbán was never as bad as Trump is now, but his story is actually very similar, and so is the message to Hungarian liberals.

Orbán, much like Trump, was a reaction to left-wing extremism and ineptitude. In Hungary, the left wing firmly held power in the 2000s. In many Mediterranean and Eastern European countries, even the left is patriotic (“we are such a great nation, even our workers/gays should live well”). Hungarian leftists are centered around a Budapest-based internationalist clique, and they are much more like modern urban American liberals. They think the country they live in is hopelessly primitive, its history should be regarded with collective shame and it should be reformed radically until it’s not even the same country anymore. When I was in high school, there was a Holocaust remembrance day every second week, pushing the narrative that the citizens of Hungary, occupied by Germany at the time, are just as guilty as the Germans themselves, and we inherited their sins. There was great pearl-clutching in the media anybody noticed Roma crimes, and Hungary was one of the first countries in Europe to recognize same-sex unions in 2009. Maybe the highlight of this era was when the Socialist government received a Romanian national delegation to celebrate their national holiday of unification (that is, stealing Transylvania from us). Imagine the American government celebrating the Pearl Harbor attack on its anniversary with the Japanese! All this was based on the cargo cult idea that by adopting progressive Western ideology the country will become like the West in other ways too (that is, rich) – this failed because the socialists only raised taxes and gave it to boomers and the inactive.

Orbán’s rule (2010-2026) coincided with the Great Awokening in the West, including the EU. All the way until 2024, his opposition was the same Budapest-based leftists, even if regularly re-rotated to new parties. Orbán cut taxes and brought about quite decent economic development, at least until COVID. He also did away with the shame-based national ideology of the leftists. He gave citizenship to the millions of Hungarians in neighboring countries, a huge national issue, almost too good to be true (after World War I, about a third of Hungarians found themselves living abroad, the border having been moved over their heads, and they are still there). In 2015, millions of military-aged male illegal immigrants started jumping the borders of Europe, encouraged and guided by liberal NGOs. Orbán had none of this: he built border fences to stop them. Meanwhile, the liberal press wrote that “The problem is that they keep going”, lamenting that the doctors and engineers who soon drove trucks of peace into Christmas market crowds in Germany, and drained the social benefit system everywhere else will not enrich Hungary instead.

Although in many ways he came closer to a truly popular and successful leader for Hungary than I ever though possible, Orbán was never perfect. The economy was only truly good 2014-2019. There have been stories about corruption from the start. Orbán’s hobby is soccer and he wasted a lot of public funds on useless stadiums, which was never popular with even the most right-wing people I talked to. The state media was pushing increasingly primitive propaganda, people didn’t like the style even when they broadly agreed with the message.

Orbán could have been defeated any time by yielding 80% of the ideological ground to him (and this is exactly how he was ultimately defeated). Nobody likes illegal immigrants, a shame-based national ideology, feminism, or aggressive gay rights campaigning, especially not in Hungary. But Orbán’s opposition, the Hungarian left, campaigned against him by:

- calling him a dictator

- calling his system an autocracy

- calling out his stopping of illegal immigrants as inhumane

- talking about how encouraging women to have children is sexist

- calling laws that let police deal with antisocials fouling up public streets “criminalizing homelessness”

- talking about how we need more LGBTQIA++ visibility or whatever

- finally, in the perennial tradition of the Hungarian left, selling out their country to enemies in the West, campaigning in Brussels for the EU to suspend our funds and cause an economic crisis, compelling the populace to replace the government. They even had them suspend our membership in Erasmus so university students would be particularly angry at the government.

This last point is no joke, and not widely remembered for some reason even by Hungarians. In a leaked 2020 video recording, opposition MP Anna Donáth openly bragged to her colleagues that she regularly consulted with EU commissioners Vera Jourová and Gwendoline Delbos responsible for the process against Hungary and “saved them from own goals”. When the EU suspended Hungary’s federal funding, it did so with the help and following the instructions of the Hungarian opposition. Hungarians hated these people so much that they were willing to put up a lot of bad stuff by Orbán.

The new government and the end of Hungarian history

Despite a decade of propaganda about the opposite, Hungary is and was a perfectly democratic country. You can’t rule here in a way that doesn’t suit most of the people. There is no deep state of the secret service or the military. There is no rule by EHC elites (journalists or academics). There is no oligarchic mafia choosing figureheads. The Hungarian government is traditionally unable to suppress popular dissent – dictatorship doesn’t work here, Hungarians are unruleable, we wore down the Turks, the Austrians, the Communists and the woke EHC liberals equally. For better or worse, our leaders are always compelled to please the modal Hungarian voter. (It is really unfortunate that the model Hungarian voter is not very smart or forward-thinking). Orbán ruled because he was, at the very least, more acceptable than the alternative to most people. The EU, Soros and other liberal movers and shakers tried astroturfing opposition parties all the time since 2010 to replace Orbán, but all attempts failed. On the other hand, as soon as Orbán started ruling poorly after 2022, a new grassroots opposition party sprung up within months and won the next election by a landslide simply by not campaigning with extreme leftist stuff. This is in stark contrast to the West, which I think works better than in Hungary because it is less democratic and more able to concentrate power in the hands of competent people, if there are any around. Recently, Western elites grew increasingly detached from common people and renamed democracy as populism. (See also Arctotherium’s great recent piece about the closure of the internet.) By the 2020s, the strong gods are back and every country reverts to the style of government that fits its traditions and the psyche of the locals: an oligarchy in the US, aristocratic rule in Western Europe where elites battle over luxury beliefs and peasants must shut up and pay taxes, and a tribal council in Hungary where the popular can leader can get away with anything until he falls off his horse, then he goes to the dogs. I don’t think Hungarians appreciate or even know how special the ideological flatness of our society is, how close the values of leaders are to those of the people. In a way this is very impressive for Hungary, but it also shows you how little democracy matters – we are not a very rich or successful country. Access to resources, the geopolitical luck of not having big neighbors treading you down, and good mechanisms for making smart people leaders (even if non-democratically) are what matter for a country’s success.

There have been some rumbling, including in my own comments, that the new Hungarian government will be one of woke EU plants who will invite masses of illegal immigrants and give all our money to Ukraine. Fortunately, there is not much sign of this happening. Orbán’s remaining media could dig up very little dirt on any new government members. The EU and the new government swiftly made a political deal about releasing our funds in exchange for very little – some cosmetic changes to institutions and maybe throwing a few Orbán oligarchs under the bus. They carefully avoided gay and immigrant topics which the EU would risk alienating voters with. It seems the EU really just wanted Orbán gone and otherwise we will be left to our own devices. Of course, only time will tell, but even if in the best-case scenario we just became an unremarkable, peripheric country in a declining EU.

A shattered right-wing revolution

This is sad because Orbán (and Trump) was not fully reactionary. At the time of the early Great Awokening, there was an idea in the air that this is the last struggle of this whole corrupt post-1968 system and something better, something not based on lies is coming. This was also present in science. Replication studies, data sharing, free R packages, sci-hub, alternative online media, the rise of outsider politicians all felt like the start of a new era where you can’t gatekeep progress – real progress, not handouts for Marxists – by paywalling knowledge, calling people Nazis, or only hiring NGO experts and propaganda journalists as the only source of truth. It felt like a revolution was under way and that Hungary, uniquely in its history and just about turning the corner economically around 2013, can be at the front of this revolution.

This didn’t come to be. I think woke has died for a generation, as it did after the 60s when it first emerged, but the old boomer system won and it is stronger than ever. The alternative right wing attracted low-quality people, grifters, and liars, and collapsed under its own incompetence. I still don’t like Anatoly Karlin’s EHC theory – smart people believing in false ideas is a tragedy, not an argument that these ideas are true after all – but he had the foresight the realize that the left wing attracts, and the right wing repels, the kind of people who can successfully run a complex system like a corporation or a country. The right wing populist revolution of Orbán and Trump failed shamefully – Orbán is out and I think Trump will be out soon. The most scandalous excesses of wokeness are probably also over, but I now think that woke was never the true nature of power, just an early experiment of totalizing it which proved too heavy-handed. The China Convergence will continue, the future belongs to top-down ideology, huge corporations and supranational entities with mass surveillance and soft and surgical, but still total power. The American elite now have precedents on just how much plutocracy, surveillance and social engineering it can get away with. The EU also found out that it can replace disobedient national governments by finding “rule of law” violations. China is China, richer and more powerful than ever. Orbán’s fall was the first case of the EU ousting an independent national government, a short-term benefit for Hungary but a dangerous precedent.

My previous post was about female nature, but in a wordcel way. After writing it, I started wondering if women really have a different psychological structure.

Few things are as interesting to as many people as sleep, sex, and women, so it boggles the mind that we know so little about these things. Everybody knows that there are sex differences in personality – women are usually more neurotic, conscientious, agreeable and extraverted but less open – but this is just the tip of the iceberg. When we calculate sex differences in psychological traits, we assume that the same traits exist in both sexes. But what if this is wrong? What if women have a fully separate psychology?

There is surprisingly little research into this. Some early theories, like Sigmund Freud’s of psychosexual development or Lawrence Kohlberg’s of moral development were criticized by feminists as male-centric theories, which I guess is fair enough. Highly simplified, Freud’s ideas was the desire for sex drives behavior, and Kohlberg’s that acting based on abstract universal principles is the highest stage of moral development, which does sound more true about males. Psychoanalyst Karen Horney published a volume called Feminine Psychology in 1967, in which – among other things – she said Freud is wrong to say that women have penis envy, it’s men who have womb envy: by bearing children, women are naturally creative, while men need ersatz behaviors like building civilization, which is also something I can’t seriously disagree with. Wikipedia has articles on feminine psychology and feminist psychology if you want to read more, neither is very good and didn’t tell me much more than I knew. Unfortunately, a lot of work was done by psychoanalysts with not much methodology, and the field is fused with feminist grievance studies.

As far as I know, beside feminists it was only PUA/manosphere/incel types who tried devising a specifically female psychology. There are many ideas by incels that ring true – for example, the female preference for celibacy over settling or alpha widowhood – which sound true. Incels also like to cite data, but they also like to cherry-pick and overgeneralize. Shoutout to the Nuance Pill for exposing many wrong things about incel social science. Politics consumes all interesting fields in psychology.

To the best of my knowledge nobody tried to come up with a serious, quantitative sex-specific psychological theory. So in this post, I try to do just this.

Data and methods

Although a female psychology can be many things, for now, I wanted to focus on personality psychology. Specifically, I tried to establish if there are personality traits which only exists in men or women. This is hard to do, as you will see! Fortunately, I had very good data to start with: a dataset from the SAPA personality project.

The SAPA dataset has responses from 23679 people on 696 personality items (=questions). However, not every person responded to every item. “SA” in SAPA stands for “synthetic aperture”, an analogy from radio astronomy, where apparently it is possible to combine the inputs of multiple small telescopes to simulate a much larger (impossibly large to actually build) telescope. Bill Revelle, the creator of the project, thinks outside the box. In the context of the current project, “synthetic aperture” means that while nobody sees all 696 items which would take hours to answer. Instead, everybody sees only a sample of them so the test is not unrealistically long and yet we have lots of observations from any selected pair of items. For the covariance matrix (a table of the similarity of all questions to all other questions), which is enough to create models of personality, this is enough. The effective sample size – number of people responding to any item – is on average 2930, 1022 male and 1908 female, which is more than enough for us. I am working with polychoric correlation matrices, important (but much longer to calculate) because these are ordinal items.

Male-female personality traits are the same… mostly

The first question is if the data yields the same overall structure of male and female personality. The dominant idea (the so-called lexical hypothesis) in personality psychology that personality consists of a finite number of traits (characteristics), and the everyday words and expressions we use to describe personality and behavioral preferences are more or less just synonyms of these traits. For example, “I prefer to be along”, “I find clubs to be annoying”, “I’m a loner” all reflect the personality trait “introversion”, a preference to be alone.

Mathematically, what happens is that 1) we collect data on many questions/items like this, 2) we calculate a correlation or covariance matrix, indicating the similarity between all possible pairs of items, 3) we use dimension reduction techniques like factor analysis to create a much lower number of factors than there were items which still capture most of the information originally available. Often it is possible to reduce hundreds of items to just 3-6 factors while still retaining ~50% of the variance (~information).

The problem is that I had no idea how to model SAPA data which pools items from multiple questionnaires on similar but not-quite-the-same personality factors, so I first ran parallel analysis on male-only data to find out how many factors to even extract. Parallel analysis compares the eigenvalues of factors (~how much variance a factor explains, 1 being a typical ordinary variable) of real data with random data, and finds how many real factors still work better than those extracted from noise. In the current case, this number was 6, neatly matching the maximum number of first-order personality factors in the questionnaires the items come from (HEXACO and QB6).

The 6-factor EFA solution (ran with oblimin rotation with psych::fa) to male data has quite OK fit for a model this simple (RMSEA=0.057, TLI=0.871). The six factors account for 24% of the variance which is not that great, but remember, we are not trying to figure out the structure of personality, we just want to say if the skeleton of such a structure is the same in males and females.

Here are the top five items (by absolute factor loadings) loading strongest on the six factors, separated by sex:

The first five factors are just the Big5 (I added the names, but of course these items emerged from the mathematical characteristics of the data, not because I was specifically looking for Big5 indicators). Factor 6 is not quite the same though. In females, we could call it “sensory sensation seeking” – recklessness, an appetite for adventure. In males, it has elements of it – taking things easy, being easy to satisfy. Of course this is the opposite of sensation seeking, but it’s a feature of the method that sometimes it finds the other extreme of the same trait.

The congruence coefficient is a correlation-like similarity index of factor loadings in male-only and female-only models. This is how similar the six factors extracted in the two sexes are to each other:

The factor congruence matrix of males (columns) and females (rows). I cheated a little before by moving the male and female neuroticism/extraversion to the same position, they actually emerge as either the first or the second factors, but this is a feature of the method. Note, however, that the last factor has substantial off-diagonal coefficients: as it emerges in one sex, it resembles other traits in the other sex almost as much as the same trait.

For the first five factors, congruence is excellent –coefficients go from 0.91 to 0.97. The Big5 is indeed the same in the two sexes. But for the sixth, this is only 0.65. Male easygoingness is similar to (lack of) female neuroticism in (factor 1, congruence coefficient 0.4) and agreeableness (factor 4, congruence coefficient 0.41). Female sensation seeking is also somewhat similar to the first four factors of males.

There is no difference between the sexes in how the Big5 emerges. The sixth personality factor might be somewhat more different, but even this isn’t quite what I would call a “sex-specific personality trait”.

So as crazy as it sounds, the big personality traits are indeed the same in men and women. If want a separate psychology of women, we must dig deeper.

Separate mini-traits

Reductionism works well in some areas of psychology and less well in others. In intelligence research, the primacy of the g-factor is well-known. If you extract the first principal component or highest-order factor score of cognitive tests, you have pretty much all the information such tests can tell you about real-life outcomes – there is only one way to be smart. But reductionism doesn’t work so well in personality research. Sure, we can extract a five or six personality traits and predict a lot of things with them, but there is a lot of evidence that what remains in the rest of the data is not just noise, and lower-order personality traits – facets, nuances – still contain a lot of information. One shouldn’t be surprised if people who agree more with the question “I am good at working with people” are rated as better colleagues than others over and beyond their overall “Extraversion” or “Agreeableness” scores. Maybe sex-differences are also in these mini-traits.

Unfortunately, there is a potentially endless number of these mini-traits, so even with a dataset as big as SAPA, you have to be careful how you look for them. My strategy was to look for items which:

1) have higher variance in one sex than in the other

2) have different correlations with other items in one sex than in the other

3) form a meaningful conceptual cluster (not stand-alones, that might be just an error)

Bear with me on this a little! Let’s assume that the mini-trait “competitiveness” exists in males, but not females. Men like to openly fight for status and resources and get satisfaction from victory: females don’t. What would it look like?

“Competitiveness” is not a major personality trait, it won’t stand out as the first five or six or ten personality traits, so we can’t just use the factor analytic approach we did above. It won’t have low factor congruence or measurement non-invariance across the sexes because there will be no such factor. The data trends still revealing a mini-trait like this will be the following:

1) Because this is a meaningful trait only in men, they will differ from each other more than females. Hence “higher variance”. Guys will get what the question “I like it when I defeat others” means and some will recognize themselves in it, others less so. Females won’t be able to relate so they will give more neutral responses.

2) Again, because it is a meaningful trait only in men, responses will correlate with responses to other items differently in the two sexes. One would assume that more competitive men are also less neurotic and less agreeable (and a thousand other relations to small nuances of personality): in females, how they respond to “I like it when I defeat others” will have a different pattern to other responses. Hence “different correlations”.

3) If competitiveness is a real trait, it will be indexed by multiple items. Males will also have higher variance and different correlations on items like “I feel shame when others best me”, or “I dream about ruling the world”, all related to competitiveness. Theoretically, a mini-trait could be a nuance (a single item), but looking for clusters of similar items protects us against flukes, items which for some weird reason, maybe some error, work differently in males than females. A conceptual cluster of similar items showing characteristics 1) and 2) is stronger evidence for a true mini-trait.

As a first step, let’s plot male-female differences against male/female item variance ratios (labels only included for the most interesting items >2 SD from the mean):

Bigger image because Substack has enshittified how these appear.

There is a lot to unpack here. First: higher male variance reigns supreme, most dots are above the thick black line. Self-reported personality is more different in men than women. Second, the V-shape shows that means are variances are correlated, also a common observation. Third, we can see that items in a conceptual cluster (criterion 3) with abnormally high male or female variance (criterion 1) indeed emerge. Actually, there are four such clusters, representing all 4 possible quadrants of outlying data and all the four ways mini-traits can emerge:

1) High male mean, high male variance: a male-typical trait where same males are truly extreme (upper right). “enjoy hurting people I love”, “it would not upset me to see a child or animal suffer” and “would not feel sorry for an animal in a trap” belong here. There is arguably a second cluster of items like this, concentrated around more conventional aspects of psychopathy: “cheat to go ahead”, “rarely worry”, “most things taste the same to me”, “would take drugs that may have strange or dangerous effects”. One could say that these items index the male-specific mini-trait “sadism/psychopathy”, something males are more prone to, but some males truly extremely so. Male variance is about 50% greater for these items, and other than crying during movies and loving dance and flowers these are some of the most sexually dimorphic items too.

2) High female mean, high female variance: the polar opposite, something women are prone to but a minority extremely so, lower left. “Panic easily”, “Begin to panic where there is danger”, and “Need protection” all belong here, indexing the female-specific mini-trait “fragility”. A minority of women collapse under stress, although all have tendencies in this direction.

3) Low female mean, high female variance: a minority of women exhibit an otherwise male-typical trait (lower right quadrant). “Interested in science”, “Enjoy intellectual games”, “Have a rich vocabulary”, “Enjoy wild flights of fantasy” are typical items here, forming a conceptual network around the female-specific mini-trait “nerdiness”. A minority of women resemble males in the otherwise typical abstract intellectual orientation of the latter (AKA “wife material”).

4) Low male mean, high male variance: a minority of men exhibiting an otherwise female personality trait. “Cheer people up”, “Concerned about others”, “Inquire about others” are examples of a cluster like this, a male-specific mini-trait “prosociality” (AKA being the gay friend of women). Most guys don’t like doing small-talk about niceties but some do. I can’t confirm that this has to do with sexual orientation and most people like this are presumably heterosexual, but I wanted a memorable label for this mini-trait

Here is a complete list of the items I thought index the four sex-specific mini-traits:

Along with the items and the mini-traits they index, you see here: the male-female difference in Cohen’s D and the male/female variance ratio. So far, this is criterion 1 (variance difference) and 3 (conceptual cluster of multiple items). How about criterion 2, different correlations? To test this, I went brute-force: I measured the sex difference in average item intercorrelations within each mini-trait, and compared it to 1000 random samples of items of equal size. I didn’t just look at whether the items are more similar than random items - obviously they will be, they are conceptually similar. I looked at if the sex difference in similarity is abnormally large. Basically: do these items resemble each other more in one sex than the other to a greater than expected degree? Results:

MaleObserved and FemaleObserved are the mean item intercorrelations within mini-traits. These are quite reasonable. MaleZ and FemaleZ tells us how much greater these intercorrelations are than the intercorrelations of random items. Z-scores are ~3-6, so extremely so. These items tap into something common, unsurprisingly given that they were selected based on conceptual similarity.

Next comes the big question: is the sex difference in the intercorrelations greater in one sex than in another – in other words, is the conceptual similarity greater in one sex than another, which we said was a criterion of a sex-specific personality trait? Kinda. The strongest evidence is for the “Gay friend” (Prosociality) mini-trait, where the extra male similarity exceeds the conventional significance threshold of z~1.96. There is also tentative evidence for Psychopathy-Sadism, which would have a p-value of 0.097 if we wanted to get one. We could even say that we had a directional hypothesis of higher male intercorrelations so we can hack out a one-sided p-value of 0.048.

The case for the female traits, Nerdiness and Fragility, is much weaker. Contrary to our hypothesis, male intercorrelations are higher, although not significantly so. Although females have greater variance in how much they confess to be nerds or panic in scary situations, there is not much evidence that these questions tap into something very different in males and females. Nerdiness and Fragility seem to be traits both males and females have, only the latter with higher variance (and, in the case of Fragility, a much higher mean).

All in all, if you dig deep enough, you can discover some sex-specific personality traits, but it’s not something that jumps at you. Males and females seem to have broadly similar personality structures.

The way forward

This is not definitive evidence yet. People often complain – correctly – that EFA is not a good way to analyze personality data, one needs to run confirmatory factor analysis (CFA). I started doing this, but ran into silly convergence issues which is hard to fix because these models run for several hours, so I decided to post the EFA results first. This is the plan going forward:

- Model 1: modelling personality as suggested by the EFA at the beginning, with six correlated factors, each item loading on one as suggested by the EFA. These models didn’t fit the data very well (nor did they have to, because nuances), but strict measurement invariance across the sexes was tenable according to some metrics.

- Model 2a-d: adding each of the four (hence a-d) mini-traits as an additional latent variable, with loadings only on the items identified above. This will test if the structure of personality is improved by adding the mini-traits to the six big traits. There is no sex effect here yet.

- Model 3a-d: this is where the sex effects begin. We will set the variance of the mini-trait to 0 in one of the sexes: in other words, we will assume that the mini-trait only exists in one sex. In that sex, the model will be like Model 1, in the other, like Model 2. A better fit in this model compared to Model 2 will confirm our hypotheses about a sex-specific mini-trait.

- Model 4a-d: these models will model a weaker version of the sex-specific mini-traits. In these models, the mini-traits will exist in both sexes, but their variance will be allowed to vary across the sexes. In other words, these models will assume that the mini-traits exist in both sexes, they are only more prominent in one sex. Comparing these models to Model 3 will allow us to say if the mini-traits are sex-specific or only sexually dimorphic.

Larsen et al: Toward individualistic reproduction: Solving the fertility crisis could require a further marginalization of men

The cross-national correlation between gender equality and lower fertility is exceptionally strong ( r ≈ 0.81). After the 1960s, a unique mating regime spread across parts of the world — with female emancipation, individual mate choice, and effective birth control — followed by a continuing rise in singlehood and declining fertility. Almost all women still want to reproduce, but many struggle to find a good-enough partner. This article argues from an evolutionary perspective that many men ’ s utility to “ free women ” has been so diminished that solving the fertility crisis by increasing pair-bonding rates seems unfeasible. A viable means for aiding the survival of low-fertility nations could be to provide women with the economic and social resources necessary for them to conclude that having children alone makes for a better life than remaining childless. Such policies would likely exacerbate male marginalization, but new technologies are on the horizon that could offer men reproductive equality.

The paper’s thesis goes something like this.

Most births happen in a pair-bonded relationship. Women pair-bond with men they consider above them – in a gender-equal society, this is by definition very few men. This is the main cause of declining birth rates. Because we cannot bring back pair bonds without undoing feminism, we can only fix the birth rate crisis by severing the link between pair-bonding and childbearing. In other words, we need to create a society of single mothers and artificial wombs, supported by the state.

I think this is a bit of a troll paper. But the first author is Scandinavian, so who knows. Either way, it’s a topic worth discussing! In this post, I will freely mix points from the paper with my own interpretation and often leave references to Larsen et al.

The dual mating strategy

When women choose their reproductive partners, they have to balance two contradictory goals. First, because they can’t have very many children in their lives (unlike men who can theoretically impregnate several women each day), they want to make sure that the few children they birth inherit great genes from their father. For this, they want a high-quality partner.

Second, they need more than just genes – pregnancy is long and the child needs to be supported until adulthood. A partner who is too high quality may not commit (because he has other options), so they need a man who sticks around.

Note that sperm donors and providers don’t have to be the same men at all! The best case scenario for a female, from a purely biological point of view, is to get pregnant from the highest quality man around, and have his child raised by an obedient male worker drone who pays for food, fixes the sink, and lifts heavy things. Manosphere veteran Heartiste called this phenomenon “alpha fucks, beta bucks”, and in evolutionary psychology it’s known as a dual mating strategy. (Wikipedia refers to it by another term, “strategic pluralism”.)

The problem is that being a “high-quality” partner, the kind of man the typical woman is attracted to, is by definition rare. High-quality is what’s better than the rest. Not all men can be the best hunters, top athletes, bosses, rock stars or whatever women are attracted to: these qualities all imply a large group of men beneath them, compared to whom they are high-quality. Simply put, it’s a natural consequence of the female reproductive strategy that the average man cannot be too attractive for the average woman. Women all want a piece of the best guy around and they don’t like average chumps very much. (Heartiste: “Five minutes of alpha beats a lifetime of beta.”) On the other hand, most men consider the average young woman quite attractive and they are willing to compromise on their standards. The classic dataset demonstrating this is OKCupid’s now deleted histograms of male and female ratings:

and number of messages sent, the OKCupid currency of mating effort, visible much higher for men:

Larsen et al cite some newer data supporting this phenomenon.

Topinkova & Diviak (2025) analyze real data from Czech dating site users. The analysis is based on a contact matrix mapping contacts between all possible combinations of men and women in the two largest cities. The distribution of messages received (“indegree”) by women and men looks like this:

20-30% of men have 0 indegree (no girls message them at all). The average female receives attention from 15-20 times as many men as the average male from women.

This journal article uses data from swipestats.io, a website people use to track their Tinder history, and finds that 1) males swipe left and right about 50% of the time, but females reject 95% of males they see, and, not unrelated to this 2) a man has a chance of about 2% to match with somebody they liked, while this is 36% for a woman.

Historically (pretty much everywhere, every time) this was not a problem because every society imposed limitations on mate choice. You couldn’t just fuck around any way you wanted. For a woman, limitations were twofold: economic and moral. On the economic side, a woman needed the physical and financial support of her husband to stay alive. On the moral side, a woman who had too many partners or who was unfaithful was shunned, often by the other women. It’s easy to see why you would want these limitations. Average men also want access to sex, children, and certainty of paternity, and they are quite useful for working and fighting. A few men monopolizing all the females destabilizes society, the dominant male always has to guard his mates against the horny incels. Even if he kills them all, his group of one guy and many women will be vulnerable to external enemies from groups who kept their males. Even if there is no mate-guarding or invasion, if horny average men cannot attract women with their resources, they are less motivated to produce them.

In the modern West, both limitations were removed.

Economically: women are free to study and work in any job they want. Many governments provide various forms of financial support for women (e.g. mandated governing board positions), and men also readily give resources to women even if they don’t commit to them. Western living standards are so high anyway that you don’t need to worry about starving.

Morally: female promiscuity is tolerated (and with dating apps and urban anonymity, often completely hidden). With the pill, antibiotics and legal abortions, promiscuity doesn’t carry health costs or the risk of an unwanted child either. Even if a single mother has a child, she doesn’t have to suffer stigma – in fact, she might get extra support from the state, the government effectively stepping in as the father.

With these changes, in the modern West there is not much reason an average woman would want to pair-bond with an average man. This is not a misogynistic statement and it doesn’t necessarily imply any moral fault of the individual female (although it does have catastrophic social consequences). She doesn’t need his work, she isn’t attracted to him, she also isn’t coerced to accept him lest she remains a spinster. Sure, she would like pair-bonding and children with a high-value male: but a woman cannot pair-bond with a man she doesn’t find attractive. Because woman are hardwired to be attracted to men firmly above them, in a gender-equal society the pool of attractive men is necessarily small. Emancipated women don’t see average men as suitable partners for family formation. The few guys women would start a family with would rather enjoy being the minority of men who get to have copious sex than found a family with any of the applicants. Without pair-bonding and stable relationships, birth rates crater. The authors’ own table:

In the paper, these changes are framed as inevitable, but I think this is only because the authors wanted to get published in a mainstream journal. Normally, you would read a brutal diagnosis like this in much more… alternative places. For example, here is essentially the same story told by Andrew Anglin of the Daily Stormer, with the unforgettable title “Belle Delphine’s Personal Issues are Not Worse Than Anyone Else’s”. (Anglin, originally a neo-Nazi, became a Hunter Thompson-style essayist around COVID, although he kept blaming everything on the Jews. There used to be many kooky, smart-crazy guys like him in America.)

Anglin:

Probably the biggest part of the development of civilization itself involves the creation of taboos around sex. [with the] removal of thousands of years of sexual taboos, you are now witnessing the animal nature of man laid bare. […]

The process of destroying these ancient taboos took time, of course. Even when I was in school, there was still at least a vague idea that a teenage couple that was involved sexually should be looking forward to marriage. That idea is now gone, and the female reigns supreme. She is able to choose whatever sexual partner she wishes, whenever she wishes. And, as in the gorilla video above, this typically involves teasing lesser males, to create interest by the targeted alpha male, who will then rush the bitch and have his way with her. […]

You’ve created a paradise for women, and a kind of hell for men, where the woman is able to strive (without much special effort) to get sex from the top level males, while the average male (and increasingly, the above average male) is left masturbating on the internet. This is all the epitome of barbarism, as being a core animal drive, sexuality is naturally barbaric (which is why putting restraints on it was at the core of the concept of civilization).

Anglin includes a lexicon definition of barbarism as “the absence of culture and civilization”. The sexual revolution removed civilization – restraint – from sexuality, and what you see now is the natural manifestation of the biological nature of women. I think it’s wrong to describe this as a female paradise though, because women are also happiest in traditional gender roles, and getting their hearts broken by a series of hookups with Chads who won’t commit (Heartiste: “cock carousel”) is not ideal for them either.

In another piece, blogger Arctotherium presents societal solutions to mating in the context of the prisoner’s dilemma. Monogamous marriage is a cooperate-cooperate solution: the female gives up her hypergamic drive in exchange for security and social standing, the male gives up resources in exchange for access to sex and reproduction. Feminism a defect/cooperate strategy: female interests are maximized and males are expected to cooperate in exchange for nothing or very little, which is unlikely to work in the long run.

What the future will bring

Note that both female strategies – even the Chad cock carousel one – are reproductive strategies. They are meant for – and capable of – producing children. Female serial monogamy doesn’t produce children because of advances in contraceptive technology, but technically this is not necessary. By separating sex and reproduction even further, we can theoretically solve the fertility crisis. This is what the Larsen’s “further marginalization” means: in the future, women would have serial flings with Chad when they are young, and when they are ready to settle down, they would have children independently from a sexual relationship. Larsen et al specifically suggests government-supported communities for single women with free 24/7 childcare and increased maternity benefits. Also,

To make such experiments more appealing to men, nations could present them as sex/gender neutral. Of course, biological males do not have a uterus, so they would be greatly disadvantaged. But theoretically, women could choose to have a child and let the father be the single parent. […] In the longer run, artificial wombs could allow men greater reproductive equality. […] Billions of humanoid robots could be providing us with goods and services in the 2040s (Gavin, 2024 ; Neumann, 2025 ), bringing us toward a future of “ sustainable abundance ” (Diamandis & Kotler, 2023 ; Goldstein, 2025 ).

Women get more free money & license, men put up with it until artificial wombs and Star Trek Communism makes everybody happy, how could this fail! These solutions are so absurd that I can’t read it as any other way that “either we limit female mate choice or we die out”. This might be the sentence they snuck in about their real views:

“If our communities fail to solve the fertility crisis and disappear, it seems unlikely that the fertility-belt nations will ever repeat our unique experiment with individual mate choice. In such a future, there may not be many free women, at least as we understand the concept in the West.”

“Nice feminism you got there, shame if something happened to it”:

By the way, this “fertility belt” issues is not to be understated. Global fertility rates are pretty much a direct function of gender inequality (r=0.81):

On a map, countries with a positive fertility rate in dark blue:

You shouldn’t underestimate how fertile this fertility belt is. More Births recently published this analysis, which you should read in full. Did you know that Ethiopia or the Democratic Republic of the Congo both have more births, individually, than either the US or the EU? Did you know that Nigeria – a single country – has more births than the US and the EU put together?

A few decades from now, humanity will overwhelmingly consist of people from countries with high gender inequality (many of them having migrated to the West, with many countries already receiving migrants than they have births), so unless you believe that the Congo or Nigeria will soon become feminist, feminism will have extirpated itself in the strictest evolutionary sense of having caused the people who endorsed it to be outbred by those who didn’t.

I don’t think it matters whether you consider female liberation a key victory of the Enlightenment (Larsen et al), a dangerous byproduct of modernity (Arctotherium) or a Jewish plot to destroy the West (Anglin). Getting hung up on what you can call people you disagree with is pointless. All analysts of the phenomenon agree that fertility rates are collapsing due to reduced pair-bonding, and that there are three possible future solutions to this:

1) Restoring pair-bonding through the disenfranchisement of women, specifically limits to economic accomplishment

2) Severing the link between pair-bonding and reproduction by adopting a drastically different reproductive model based on uniparental child-rearing

3) Doing nothing, collapsing Western fertility and facing replacement by non-Western populations.

This is not a matter of being a Jew or a Nazi, these are simply the three possible ways things can go.

Solution 1 (the disenfranchisement of women) is the most scandalous, so much so that noticing this whole issue draws accusations of being a misogynistic far-right incel. I understand why this solution – artificially increasing the sex difference in material and reproductive resources – upsets people, but one must notice that this is the only solution that has been historically tried successfully. Not that I know how one would go about it now. Women are not stupid or incompetent, they are pretty good at the types of office jobs most people do in developed countries. They are usually diligent, obedient and precise, which is good enough for most mid-level workers. Gender diversity quotas only affect elite jobs (board directors, MPs) which are a fraction of a percent of all jobs. Although there is a lot of ideological push in this direction too, the market keeps hiring women because they are decent workers. Would you ban a smart 18-year-old girl from to college? Would you prohibit her taking a prestigious job once she graduates, like the Communists did with the descendants of old elites, so she stays poor enough to have to marry some chump? Would you chaperone her, ISIS-style, so she doesn’t sneak out on a date with Chad? With feminism, we have made fish soup out of the aquarium of old social norms and you can’t bring the fish back with incel fantasies.

Solution 2 (post-pair bonding reproduction) would involve the most drastic changes, and many people would see it as dystopian. With this solution, the typical person would be raising at least one child alone (as long as both men and women do this, 1 is enough for a total fertility rate of 2), with the child often being conceived via assisted reproductive techniques or surrogacy. Basically, women would pay for Chad’s sperm and men would pay a surrogate mother or an artificial womb to produce a child. Both sexes would raise the children of strangers they are attracted to but can’t secure commitment from (women) or don’t have sexual access to (men). The advantage of this solution would be that women’s rights wouldn’t be infringed, but – as it is often the case when you want to accommodate women – you would run into practical problems. Will single men really pay for a surrogate mother to have kids? If they do, isn’t that also some infringement of women’s rights? If they don’t, will the average woman really raise two kids of some stranger alone with additional financial incentives? Who will pay for these incentives – and the whole circus? This would presumable be for the men, but will they do it with explicitly no access to mates or a family?

Solution 3 (replacement by a non-feminist society) is probably the default scenario, and the one I think is the most likely to happen. I don’t think we will have the guts to do any form of Solution 1 and Solution 2 is not practical. Humans are genetically and memetically diverse, and evolution ruthlessly prunes organisms carrying genetic and memetic variants that make them fail to reproduce themselves. Africa and Afghanistan are not very nice places by modern Western tastes, but for the tastes of evolution – the only taste that really matters – the people living there are more suited for long-term survival. Despite neo-Protestant beliefs to the contrary, we are not above the material world, and biology will have the last word about this too.

The new paper

One prediction of minority stress theory is a dose-response relationship between intolerance and mental health. Even if no country is perfectly tolerant towards non-heterosexuals, they should be better off in relatively more tolerant ones. This is the phenomenon Johnson et al 2026 “Exploring the global prevalence of mood and anxiety disorders in LGBTIQ+ people: A systematic review“ investigates.

The paper is a systematic review of all the studies the authors could find about the rate of depression, anxiety disorders and bipolar disorders in non-heterosexuals. The fundamental question is: are these rates higher in countries which are less tolerant? “Tolerance” is measured by the Global Acceptance Index (GAI), a score the Williams Institute publishes for each country.

The paper finds a null effect. There is no correlation between acceptance and psychiatric disorder rates: non-heterosexuals are not better off in more tolerant countries, which is the opposite of what the minority stress hypothesis would predict. Not only is there no relationship globally, there is none on the regional level either. You could think that maybe within Europe, where all countries are reasonably tolerant, additional tolerance has no benefit: but there is no relationship in Africa or Asia either (not a lot of data though). The only significant relationship is within North America, where increased tolerance predicts higher rates of depression, at p<0.001 no less – at face value, this would suggest that increased tolerance makes non-heterosexuals worse off.

Amazingly, the authors completely ignore their main result that the level of societal tolerance does not matter for non-heterosexual mental health. Their “Implication and future research” section reads:

“This systematic review highlights the continuing need for sexual and/or gender-identity affirmative mental healthcare worldwide to tackle the higher prevalence rates of affective disorders we identified in LGBTIQ+ people. Affirmative and culturally competent mental healthcare could be key in the improvement and protection of mental health in LGBTIQ+ people, possibly serving as protective factors against both environmental and internal experiences of homophobia, heteronormativity and stigma.”

But you just wrote that homophobia and stigma – a low GAI – do not correlate with “LGBTIQ+” mental health so they cannot cause it! Actually, you reported a positive correlation within North America1 between GAI and depression rates, which could even mean that “affirmative and culturally competent” healthcare harms non-heterosexuals!

Sometimes authors just write stories reviewers will like for the discussion, so I like to focus on the data in papers like this. On that account, Johnson et al is very good, but it contains lots of amateur mistakes, so let’s re-analyze it.

Problems with the paper

The authors posted their data, but for an inexplicable reason it (Supplementary table 3) is posted as a series of pictures, not a proper spreadsheet. With AI, this is easier to convert back to a proper table, but it took me an hour or so to 1) realize that Word images copy-pasted to LLMs appear as a weird PNG with every second row black, so you need to take screenshots, and 2) figure out how to get around the unkillable Excel “feature” that converts numbers to dates (both 9.2 and the gene name SEPT2 auto-convert to September 2 irreversibly and no way to opt-out). Also, even though they run meta-analysis, so they had proper data created somewhere, all the authors shared were a text summary of findings, which I had to convert into proper diagnosis rates. The final data is here in case you want to run your analyses or check for coding errors, this one can be used for re-analyses unlike what the authors posted with their actual paper.

There are lots of errors, missing details and mismatches between the posted data and the paper:

- The dataset contains 119 effect sizes for depression, 81 for anxiety, and 23 for depression. But the paper (Table 1) reports 106 depression, 45 anxiety, and 22 bipolar studies.

- The paper reports results by geographical region (e.g. “Asia”) but the dataset contains countries, with no clear indication which country is in which region. Although I guess with the exception of Turkey it’s obvious which country belongs to which group.

- I only checked a few single studies but there are many errors. For example, Hanna et al 2019 is highlighted even in the abstract as having a sample size of “over 254,462,596”. Of course it was not „over” this very specific number but exactly this many, and it was not participants but patient encounters in the US hospital system! Likely many people went back to the doctor multiple times and were double-counted. Hershner et al contains both an American and a Canadian sample, but the sample size is reported only based on the American one.

- All the studies the authors use have both heterosexual and non-heterosexual participants, but the former are only used to fluff up the sample sizes and only diagnosis rates from non-heterosexuals are used. It would be interesting to see the rate of heterosexual diagnoses too, and maybe see non-heterosexual rates relative to these instead of on their own. What if psychiatric problems are discovered at a higher rate in more tolerant societies, and this obscures the fact that the relative rates of non-heterosexuals compared to others are lower?

- The main problem: The authors used a single GAI for each country, the most recent one published. This is a shockingly lazy choice because the GAI is not static, the Williams Institute publishes it for 4-year periods going back to 2000, and it’s literally one click away from the main website. The authors basically assumed that Americans in the 1994 study by Cochrane & Mays lived in a society exactly as tolerant as those in the 2021 study by Atteberry-Ash et al, so they used the same GAI score (7.42). This is not only wrong from a methodological point of view, but it also makes it impossible to analyze within-country temporal trends. Minority stress theory doesn’t just predict that non-heterosexuals in more tolerant countries are better off: it also predicts that they are better off within the same country if it becomes more tolerant. Johnson et al could and should have analyzed this but couldn’t. In my re-analyses I added the most recent GAI score available when each study was published.

- I’m not sure how the diagnosis rates were calculated, because the data table doesn’t contain these as a single number but often for separate groups (e.g. depression rates for lesbian women and gay men separately). A weighted mean would be the best way to go, but because sample sizes per group were not posted, all I could do was an arithmetic mean.

Re-analysis

In the re-analysis, I used the complete dataset as uploaded by the authors. I assumed that the reported rates are averaged across all non-heterosexual subgroups. I also added the time-appropriate GAI scores, but kept the original ones too for replication purposes.

Symptom frequencies:

The original paper reports 35.3% for depression, 34.3%, and 5.6% for bipolar disorder. I can’t replicate these numbers from the raw data, but mine are close enough I guess.

Johnson et al. report truly very low correlations between GAI and psychiatric disease: 0.031 for depression, 0.043 for anxiety, and 0.11 for bipolar, none statistically significant. I cannot replicate these Spearman correlations either, my values are a bit higher, but still not statistically significant.

But why would you insist on Spearman correlations? Both the GAI score and illness rates are continuous variables, at least an interval but arguably a ratio scale and a bit skewed but not too bad:

Note how much nicer the time-appropriate GAI score (dynamic_GAI) is compared to the single-year score the authors originally used (GAI_Flores)! We get a much better distribution of tolerance ratings by looking at the historical tolerance levels of when each study was published.

So let’s try Pearson correlations and proper scatterplots, with a LOESS fit line to see if the relationship is very different from linear:

Not perfectly linear and skewed, but good enough. On the right, you can see the results with the authors’ own GAI scores. There are slight, non-quite-significant positive correlations between GAI scores and psychiatric illness rates. On the left, you can see what happens if you use the GAI score which was current when the study was published. Because tolerance levels may change within the same country, in my opinion, this is an obviously better way to ballpark the actual level of (in)tolerance participants were exposed to. You can see that with this better data, the correlations about double and now reach significance for both measures with decent sample sizes, anxiety and depression. The correlation is positive, so more tolerant countries have bigger problems! For example, the study with the lowest anxiety disorder rate is Khorashad et al 2021 from Iran, which is not exactly the Palm Springs of the Eastern hemisphere.

This is still a very crude way to run this analysis because it ignores that the same country is sampled by multiple studies. A better way to analyze this would be to run a mixed-effects model, with a random intercept by country:

rate ~ 1 + GAI + (1|country)

That is, rates of psychiatric illness are expressed as a function of intercept and the GAI score (which can vary both between countries at the same time, and within the same country across time) and a random intercept (=mean) by country to get rid of the problem of multiple sampling from the same country. This method handles multiple observations coming from the same country, and estimates the difference GAI makes, whether it varies within or between countries.

When I run this, I get the same result as with ordinary correlations: with the authors’ static GAI there is a trend for a positive effect, but with my time-appropriate GAI scores it becomes significant for depression (p=0.021) and anxiety (p=0.038). P-values are borderline because there is not much data.

I was a bit worried because almost half the effect sizes (55 out of 127) came from the US, and only Australia (N=10) and Canada (N=12) had a meaningful number otherwise. What if the positive correlation is a US-only trend, depression and anxiety rates increasing with increased gay acceptance, and big enough to influence global analyses? Nope:

The five vertical dot stacks in the US represents that GAI scores were updated five times within this country. Each study is assigned the score current when it was published. As you can see, the trend for positive correlations between increased tolerance and increased, not decreased anxiety and depression rates is present both within and outside the US. Non-heterosexuals are worse off in more tolerance places (and, within the US, at times when tolerance was higher)! The trend is not huge and not always significant, but it’s there and its presence shows that we didn’t make a mistake when we just pooled all studies in a single analysis.

Interpretation

Even in their initial form, the results of Johnson et al contradicted minority stress theory, because they didn’t find what the theory predicts: a dose-response relationship between societal tolerance towards non-heterosexuals and rates of mental illness. The authors hushed this up. In the re-analysis of the dataset, I found two major errors: 1) the analyses, as reported, didn’t seem to have used all the data, 2) they used a static tolerance score for each country, ignoring that tolerance levels may have been different when data for older studies was collected. Fixing these, I found a result even less compatible with minority stress theory: higher rates of mental problems among non-heterosexuals in more tolerant countries. Much like the gender equality paradox (women endorsing stereotypical female roles more, not less, in more gender-equal societies), here we have a gay tolerance paradox: for some reason, non-heterosexuals have it worse in more tolerant countries.

The explanation of this paradox is most likely that minority stress theory is false, and poor mental health in non-heterosexuals is not caused by intolerance. It can be caused by common genetic factors: overlapping genes cause both non-heterosexuality and neuroticism/anxiety/depression, which makes non-heterosexuals have problems in the same way taller people also have a bigger head. This is why, once we take these common genes out of the picture by comparing pairs of genetically identical monozygotic twin pairs where one member is heterosexual and the other isn’t, the difference in mental health is greatly reduced or disappears completely.

Another possible explanation is that non-heterosexuality causes poor mental health, except not by exposing you to intolerance, but by exposing you to some other bad thing. A recent paper I also reviewed considers two such things: “different lifeways” and “narrative possibilities”. “Different lifeways”: if you are non-heterosexual, you are automatically a statistical minority and can’t have a biological child with a person you love, plus there are statistical realities like that non-heterosexuals live farther from their families and that LGBT subcultures can be particularly toxic (“intraminority stress”). All of these can generate depression and anxiety but none of them can be fixed by more tolerance. “Narrative possibilities”: your identity is based on narratives about who you are. If this narrative is that you are a frail, sickly victim, then you may identify with this story and feel this way yourself. Because so many stories about non-heterosexual identity are about illness and discrimination, it’s easy to imagine that members of this community identify with these and instead of seeing themselves as normal people with a statistically rare sexual preference they start to believe that they are eternal victims of a cruel society always just one step away from AIDS or suicide. Paradoxically, both different lifeways and narrative possibilities can be worse in more tolerant countries where you are more free to openly live as a gay person and more exposed to narratives of victimhood.

It is true that during his last term (2022-2026), his rule became increasingly erratic and Hungary in general deteriorated. This is why we voted him out of power, like we always could have. I’m afraid that these four years, when he really was the corrupt and incompetent ruler of a backsliding country will become his legacy, although it is no more characteristic of his first 12 years than him being an autocratic dictator. In this post, I want to commemorate the last great European statesman whose loss – first to insanity, then, starting today, likely to obscurity – is a loss for the whole continent. The decline of the EU will continue, and it will be in no small part because they ignored Orbán’s ideas.

The corruption hoax

The main accusation against Orbán was that he was corrupt, stealing public funds and giving it to his friends, especially infamous construction magnate Lőrinc Mészáros. This excellent piece of investigative journalism is as close as it gets to a single source evaluating this claim. (Any AI or maybe your browser itself can get it translated for you in a second.) The author of this article (published in G7, now moved to Telex, both Soros-funded leftist outlets, so this is hostile media as it should be in this case) went over every public tender won by Mészáros. Because he is by far the richest of Orbán’s oligarchs, the money lost on his corruption is most of the money lost in total. The piece is not new, published in 2019, but by then the corruption narrative was well-established so it’s a good evaluation of what it was based on.

The article finds that Mészáros won a large and increasing number of public tenders over time:

Number of tenders won by Lőrinc Mészáros, alone (black) or in a consortium (red)

There is of course no doubt that Mészáros, who was a small-scale entrepreneur when Orbán was elected, became rich due to his government connections. But he didn’t just shovel money out of the state treasury: Mészáros operated a legitimate construction business and he built the projects he was contracted for, so the real question is how much money would have been saved without his involvement.

The article finds that Mészáros and his partners won HUF 498 billion (about 1.5 billion dollars) worth of public contracts. Almost all of it (426 billion) was for EU-funded projects – Mészáros won the construction tenders of 11% of all public contracts and 20% of all EU-funded ones. There was evidence for corruption: 38% of his tenders had prices rounded to the nearest large number (an indicator of corruption because this shows that it is not a real price estimate), compared to 21% of other tenders, and his prices were about 15% higher than those offered by other Hungarian companies:

Tender prices offered by Lőrinc Mészáros (red) or other Hungarian companies (black). Zero is the estimated price, the numbers show deviations from this.

One should note that many of the “other Hungarian companies” offering lower than estimated prices are also run by what the author would like consider Orbán’s oligarchs, and it’s fishy that he compares Mészáros to these and not all the tenders. Did the foreign companies overcharge too?

There is no doubt that Lőrinc Mészáros got rich due to his political connections and his tenders were overpriced. However, the impact of this is vastly overstated. Despite the image that Lőrinc Mészáros basically owns the country and certainly gets all the public funds, Mészáros won a relatively small minority of public contracts, almost all of them EU-funded, and the amount of overcharging was about 15%. Running the math on the ~500 billion forints he won in tenders, about 75 billion forints, or 300 million dollars were lost to the corruption of the greatest Orbán oligarch between 2010 and 2018, which is not great but it’s the fraction of a percent of Hungary’s GDP. It was easy to create the impression in the hostile press of a country-size corruption empire because a lot of money moves about even in the legitimate businesses of Mészáros & Co, so Telex, 444 and the other Soros-funded media outlets could write a story almost every day about them spending billions on something, insinuating that all of it is freshly stolen public funds.

300 million dollars is the upper estimate of what could have been saved, assuming that giving contracts to him has no positive externalities which re-appear in the economy. Mészáros’s company is massive, and often the choice was him or a foreign company building a project. For example, in one case it was either Mészáros building a bridge or a Slovak-Italian company for 20% less. Certainly there are benefits to having domestic companies building projects, this way the profits of the owner will at least be spent on Hungarian chateaus and fine dining and not Slovak or Italian ones. The true problem the EU had with Orbán’s oligarchs was that the money they got was money Western European multinational companies couldn’t get.

I guess the best way to estimate the real effect of corruption is to look at the economy as a whole. If public funds are stolen by government oligarchs and the only way to get money is through the right connections, then you would expect the economy to languish. This is certainly not what happened in Hungary. In the first 12 “normal” years of Orbán until 2022, the Hungarian economy grew reasonably well, about as much as the Eastern European region, certainly more than the EU as a whole, or Greece, a country with a similar development level to Hungary but economically micromanaged by the EU.

Simply put: Orbán’s oligarchs’ corruption was obviously real, but the scale of the problem was too small to be economically meaningful.

Orbán until 2022

To understand Orbán’s success and downfall, you need some context. Americans laughed when Putin started explaining his war in Ukraine on Tucker Carlson’s interview in the context of events that happened centuries ago, but that’s Europe for you: everything is the way it is because of something that happened a very long time ago.

If you imagine Hungary as a poorer New York City surrounded by a poorer Arkansas, you are not very far from the truth. Budapest is a big, dense, cosmopolitan, artsy, left-leaning city with pockets of serious wealth. But it is an anomaly and the rest of the country is rural, sparsely populated and conservative, basically a surviving medieval peasant society with a veneer of modernity like high-speed internet and department stores. Unlike in most of Eastern Europe where my impression is that most of local politics is just flavors of “Romanian nationalism” or “Estonian nationalism” with ideological debates recently imported from the West, Hungary has political sectarianism based on this divide going back at least to the 19^th century. At the time, this was called the urban-rural opposition (népi-urbánus ellentét), and it resulted in a two-party system after the fall of Communism, which again is not unlike the US Democrat vs. Republican opposition around 2015.

The leftists, based in Budapest, ruled before Orbán for eight years (2002-2010), they wanted economic redistribution and proto-woke identity politics. For example, in 2005, we had our own Trayvon Martin case in 2005 with a “racist” public attack which turned out to have been committed by a gypsy, and in 2009, the police chief of Miskolc was dismissed for saying that mostly gypsies commit street robberies. Even worse: despite their educated urban background, the leftists were economically incompetent, they raised taxes and spent the money on boomers and NEETs Tony Blair-style, so the economy tanked even before the 2008 crisis. Hungarians are chiefly economic voters – this is why the EU also decided to oust Orbán by stealing our funds and creating an economic crisis – so they voted them out and voted Orbán in with a 2/3 majority.

Orbán essentially did all the right things, which is why he was voted back three times. In 2010, the 2008 financial crisis was still raging. To solve this, the EU wanted austerity (cutting back on public spending to balance finances), which doesn’t work, you can only grow out of crises like this, if you cut spending it will just lead to recession and even more crisis. This is what happened in Greece, the country where the EU got to test its own idea of crisis management. Orbán knew this wouldn’t work so he cut taxes instead – for example, a 16% flat income tax starting in 2011 – attracted foreign investment in factories and other businesses, and reduced unemployment benefits. Leftists were seething because the flat tax was unfair to the poor, new factories were just “assembly lines” (összeszerelő-üzem) and Orbán should have founded the new Microsoft instead, and there was some other bullshit about all the other policies too. But Orbán’s policies worked, as I showed above, the economy grew decently and above EU average, life expectancy grew by 2 years until 2019 (per OurWorldInData, and admittedly this is below the European average). Hungary was never particularly dangerous but crimes dropped precipitously, although this is also a global trend which partially evaded the US:

A lot of nice things were being built constantly. Most of the town centers and other landmarks were renovated, a lot of new houses, convenience stores and other things were built. Nice cafés and restaurants opened, people visibly spent more. In 2010, lots of people were unemployed – by the last peace year in 2019, nobody was. My own personal situation and everybody’s around me reflected this statistical reality, things were getting better. I had my quarrels: Orbán built highways but ignored other roads which I think are now the worst in Europe, and he didn’t spend much on Budapest after 2014 which he saw as a liberal lair and rich enough anyway, so the city remains potentially beautiful but quite gritty. But the opposition didn’t campaign with issues like this, they wanted to remove Orbán for not supporting women’s rights or because building a fence to keep out illegal immigrants is evil.

Everything about Orbán being an autocrat is a lie. Unlike in the UK or Germany, where you can get arrested for being critical towards migrants, sexual minorities, and even politicians, in Hungary you could say or do whatever you wanted, it’s just the government didn’t support woke policies. They were very tolerant towards gays and light drugs until the 2022-2026 term, with basically no anti-gay policies and slap-on-the-wrist policies for weed. Hungary was, ironically, maybe the most democratic country in the Western world. There were serious competing visions in politics and a very free and colorful media environment everyone was able to consume, despite leftists imagining that people in rural areas don’t have broadband internet or cell phones to consume whatever propaganda they like. The reason for this was a bit perverse – one side was funded by George Soros and other Western woke-globalist think thanks, and the other by public money – but it worked reasonably well. Elections were free and – as you saw on April 12 2026 – winnable by anybody, but of course if you run a coalition of communists, woke antifa-sympathizers and CIA agents which is what happened in 2014, 2018 and 2022, you were defeated. Orbán’s Hungary 2010-2022 was ultimately a fine place to be.

Orbán after 2026

I wish I could tell you that Orbán had to be removed because despite his best efforts, the foreign political and economic pressure was too much for the country to bear. Unfortunately, however, in his last term between 2022-2026, Orbán probably lost his mind and genuinely governed the country terribly. This destroyed the relative prosperity of his first 12 years and put the country on the back foot. In his last four years, Orbán ran a Maduro-style left-wing populist government: the narrative was that the nation was under constant foreign attack, this is why everything sucks, and only the government can protect you from them.

Orbán’s practical philosophy was building things. You would think that after the EU cut our funds (under the pretext of corruption, but in reality for not sufficiently celebrating migrants and gays and generally being disobedient to Brussels), he would find other sources of funding for construction projects, if for no other reason than because his oligarchs made a living from this business. No: he appointed a uniquely destructive person, János Lázár as minister of construction, whose entire 4-year work was cancelling all construction projects in the country. He didn’t just delay them for a temporary lack of funds: he cancelled them all, including the most important one which would have replaced all the outdated railway cars. It’s not that there was simply no more money: he ordered a completely unnecessary Danube bridge in Mohács and two very expensive highway projects (M6, M49) in the middle of nowhere for almost HUF 1 trillion, while the government also bought Budapest airport and the telecommunication company Vodafone for about HUF 2 trillion more. The EU froze about HUF 8 trillion in funding, so this waste is not trivial. Without construction, the economy stalled.

Orbán’s economic philosophy was the free market. But during the 2022 inflation crisis, he did completely insane things like capping prices in stores and at gas stations. Hungary is a major consumer of natural gas, the price of which skyrocketed after EU sanctions against Russia. There were always government subsidizes for gas, but they were not very expensive. With the new prices, the government still maintained the subsidy – heating a house can be as low as 30-40 dollars per month – and there was zero encouragement of saving like mass building renovation programs. He introduced increasingly insane redistribution policies the country simply didn’t have the money for, like government-subsidized loans for anybody who wanted to buy apartments, eliminating the income tax for mothers with at least 2 children, and introducing a 14^th annual monthly pension after the 13^th which was introduced during COVID. (This latter is especially egregious because pensions are easily the biggest state expenditure, about an order magnitude more each year than what Lőrinc Mészáros ever made). Meanwhile, he cut KATA, a very attractive taxation scheme a lot of freelancers used, where you could pay a fixed monthly contribution of about 150 dollars as income tax and health insurance, regardless of how much money you made. Hungarians were spooked by the government’s non-spending so they didn’t spend either, consumption and GDP didn’t grow.

Most egregiously, instead of developing cities, Orbán taxed them. Hungarian cities collect a local industry tax, which by law is theirs. Orbán took most of this away as a “solidarity tax”, first as a temporary COVID measure, then an increasing yearly contribution. So now even well-to-do cities have no money for anything. Hungarians, as I said, are economic voters, so they were having none of this anymore.

Orbán’s ideological legacy

Despite his faults and the graceless last four years of this rule, Orbán was a great politician, maybe the last true European statesman, out of place in the modern EU leadership of Blackrock plants, weird grandma fetish people and cocaine-snorting Instagram whores. He was broadly read and well-educated, always firm, intelligent and lucid in interviews up to the very end. In a better world, he would have become the leader of the EU, not the pariah of it. Orbán as prime minister of Hungary was ultimately like a race car driver stuck in a Pinto, he was destined for more and pined for a geopolitical presence so much that in his last term he seemed to have lost all interest in day-to-day life in Hungary.

Orbán had a vision for the EU which would have been better for everyone than what came to be. For a while, he believed that he can lead an EU-wide movement supporting this: it’s a tragedy that everybody chickened out after the Ukraine war. Orbán basically believed that people need two things: prosperity and freedom of association, the latter meaning that we (both people and nations) should be free to claim some space we can run on our terms with like-minded people. For the sake of freedom of association, Orbán wanted to strengthen the nations of Europe instead of having the bureaucratic, ideologically homogeneous superstate the EU is developing into. His resistance to illegal immigration was also part of this policy, and so well vindicated that his opponents had to find a different attack angle. For the sake of prosperity, he wanted reasonable and flexible regulation, a realistic energy policy and pragmatic collaboration with countries like Russia and China, where we might not like the political system but we can still make money with them. The EU, in contrast, believes in micromanagement and regulation, replacing economic goals with ideological ones like reducing carbon emissions or female quotas, and an ideology-based foreign policy, which resulted in economic stagnation and geopolitical isolation. Orbán was aware that Russia is not a rival to the EU, but the US is, and without Russian raw materials and energy the EU will be sidelined globally by the US and China, a process well underway, especially if it picks internal fights over no-brainers and local issues like immigration or gay rights. The EU, on the other hand, completely adopted the Victoria Nuland-esque ideological interventionism a certain wing of the Democratic party believed in during the late Obama era, and they were completely blindsided when Trump left them holding the bag, sanctioning themselves and funding America’s proxy war against their own main energy provider.

Bad times are ahead for the EU which I guess will become a new languishing Latin America, but I hope the crosshairs come off Hungary for a while and the wife-beating Fidesz oligarch we just elected will not do anything too crazy. Maybe we can stay an unremarkable and by global standards not too poor country at peace on the periphery of the EU, like Portugal or Greece.

- Humans have a unique capacity to perceive second-hand information as equally real to their senses (simulation). We use the simulation to immerse ourselves in fiction and understand complex systems (like a society) by creating a simplified version of them in our head.

- Normally, we understand the simulation is not real. Hyperreality is when this mechanism fails. Hyperreal information is wrong, but we believe it because anecdotes support it. Creating hyperreality (a fake story supported by anecdotes) is an essential tool in political propaganda and a cornerstone in the very successful reality-making of cultural Marxists in the past 50 years.

- At other times, people sincerely believe facts that only exist in fiction (simulacra) because their contact with the topic is superficial and they never check the truth. Female serial killers are my example for this.

In the movie The Matrix, this is the book Neo hollows out to hide his hacker tools inside.

Because I’m interested in the social construction of reality, I decided to read and review this book (actually it’s just an essay). This book is bad. I still decided to post a review of it. Also, because the topic is so important, after the review I re-wrote Simulacra and Simulation in a normal way.

Fashionable nonsense

I knew nothing about Baudrillard before reading this book, so I did some background reading first. Baudrillard is a French postmodern philosopher, which is a huge red flag from the start. Some of his colleagues are better known: he studied under Pierre Bourdieu, a Communist who invented the term “symbolic violence” (meaning that anything that leftists don’t like can now be called violence and can be treated as such), and was part of the generation of Michel Foucault, a sadistic homosexual pederast, and Jacques Lacan, a psychotherapist and narcissistic fraud who beat his patients.

There are apparently many people who think these French postmodernists were great thinkers and the incomprehensible gibberish they wrote is just too deep for most readers, an idea dissected by Alain Sokal in his book Fashionable Nonsense. (I read this book a long time ago but forgot that it explicitly deals with Baudrillard.) Sokal also demonstrated that postmodern philosophy is a fraud by successfully publishing a deliberately bullshit article, “The Hermeneutics of Quantum Gravitation” in one of their leading journals, Social Text.

The social construction of reality is an unusually interesting topic for a French postmodern philosopher and Baudrillard is somewhat famous even in right-wing circles, so I thought this book/essay will be different. I was wrong, Baudrillard is no better than Lacan or Foucault, and Simulacra and Simulation is just rambling gibberish. Most of Simulacra and Simulation is just a series of grammatically correct but meaningless non sequitur sentences, just like what Postmodern Generator produced 30 years ago:

“This was the approach of the Jesuits, who founded their politics on the virtual disappearance of God and on the worldly and spectacular manipulation of consciences-the evanescence of God in the epiphany of power-the end of transcendence, which now only serves as an alibi for a strategy altogether free of influences and signs. Behind the baroqueness of images hides the eminence grise of politics. In a sense, if subconstructive nationalism holds, we have to choose between neocapitalist narrative and textual submaterial theory. Lacan suggests the use
of Sontagist camp to read and analyse reality. However, any number of materialisms concerning semanticist pretextual theory may be revealed. Bataille uses the term ‘dialectic deconstruction’ to denote the role of the writer as observer. The form produced by confinement is followed by an innumerable, diffracted, slowed-down mechanism. As ethnology collapses in its classical institution, it survives in an antiethnology whose task it is to reinject the difference fiction, the savage fiction everywhere, to conceal that it is this world, ours, which has again become savage in its way, that is to say, which is devastated by difference and by death.”

I parsed together two real paragraphs from Baudrillard with a Postmodern Generator paragraph. Can you tell them apart? I just copied them and I already kinda can’t. Neither sentence means anything and there is no flow between them. There is no message in this text.

Sometimes Baudrillard is clear enough to be objectively wrong:

“All the referentials combine their discourses in a circular, Möbian compulsion. Not so long ago, sex and work were fiercely opposed terms; today both are dissolved in the same type of demand. Formerly the discourse on history derived its power from violently opposing itself to that of nature, the discourse of desire to that of power-today they exchange their signifiers and their scenarios.”

How were sex and work ever “fiercely opposed terms” – isn’t prostitution the oldest profession? When was this “not so long ago”? Why would the discourse of history not just oppose, but “violently” oppose the discourse of nature, and especially the discourse of desire the discourse of power? Didn’t all historians going back all the way to Herodotus notice that conquerors were motivated by natural desires?

“Atomic war, like the Trojan War, will not take place.”

Get out of here. If you disagree with Schliemann and Cline that the Trojan War was real, cite evidence and preferably write the whole book about it.

A lot of the clearer paragraphs read like schizophrenic ramblings:

“Because what, ultimately, is the function of the space program, of the conquest of the moon, of the launching of satellites if not the institution of a model of universal gravitation, of satellization of which the lunar module is the perfect embryo? Programmed microcosm, where nothing can be left to chance. Trajectory, energy, calculation, physiology, psychology, environment-nothing can be left to contingencies, this is the total universe of the norm-the Law no longer exists, it is the operational immanence of every detail that is law. […] Through the orbital instantiation of a system of control like peaceful coexistence, all the terrestrial microsystems are satellized and lose their autonomy. All energy, all events are absorbed by this eccentric gravitation, everything condenses and implodes toward the only micromodel of control (the orbital satellite), as conversely, in the other, biological, dimension, everything converges and implodes on the molecular micromodel of the genetic code.”

Schizophrenia is (among other symptoms) a thought disorder and a lot of things in this paragraph resemble the thinking of schizophrenics. Baudrillard is saying that the space program was basically an act of sympathetic magic (like a voodoo doll). By launching objects into orbit, political elites symbolically tied all people to the gravitational pull of their power. Belief in supernatural abilities like this (known as “magical thinking”) is common in schizophrenia. Schizophrenics are also prone to see links between unrelated events and to seamlessly jump between real and symbolic interpretations of the same event (like a government launching a real satellite being trivially the same as making people its slave “satellites”, or programming society like a rocket engineer designs an orbital launch). Both are examples for what we call “loosening associations”. Two other things we often see in the thinking of schizophrenics are defining characteristics of the postmoderns: tangentiality (jumping from topic to topic and never getting to a point), and neologisms (inventing new words which only make sense to the patient). I think there was a lot of fraud in postmodern philosophy, but on some level these were very intelligent people on the schizophrenia/schizotypy spectrum. These rambling texts were their attempt to write down the torrent of semi-coherent ideas they had about the world.

“This scenario: the extremely harsh bombardments of Hanoi. Their untenable character must not conceal the fact that they were nothing but a simulacrum to enable the Vietnamese to seem to countenance a compromise and for Nixon to make the Americans swallow the withdrawal of their troops. The game was already won, nothing was objectively at stake but the verisimilitude of the final montage. […] This objective is always fulfilled, just like that of the charting of territories and of disciplinary sociality. What no longer exists is the adversity of the adversaries, the reality of antagonistic causes, the ideological seriousness of war. And also the reality of victory or defeat, war being a process that triumphs well beyond these appearances.”

Not only did the Trojan war never happen, Baudrillard also implies that the Vietnam War was also just theater, the Americans and the North Vietnamese were in cahoots, conspiring to control their respective societies. This is QAnon territory, my guess is that he wasn’t 100% serious about this, but if half of your essay is a joke one wonders how much of the rest can we take seriously.

If you read carefully, Baudrillard does have some coherent and vaguely interesting paragraphs, although not very original:

“Disneyland is presented as imaginary in order to make us believe that the rest is real, whereas all of Los Angeles and the America that surrounds it are no longer real, but belong to the hyperreal order and to the order of simulation. It is no longer a question of a false representation of reality (ideology) but of concealing the fact that the real is no longer real, and thus of saving the reality principle. […] Disneyland is not the only one, however. Enchanted Village, Magic Mountain, Marine World: Los Angeles is surrounded by these imaginary stations that feed reality, the energy of the real to a city whose mystery is precisely that of no longer being anything but a network of incessant, unreal circulation-a city of incredible proportions but without space, without dimension. As much as electrical and atomic power stations, as much as cinema studios, this city, which is no longer anything but an immense scenario and a perpetual pan shot, needs this old imaginary like a sympathetic nervous system made up of childhood signals and faked phantasms.”

I.e. in the modern world, maybe especially in the US, the entertainment industry produces a fake reality people (obviously knowing it’s fake) prefer to their mundane lives. Sure, but by 1981 every punk band had a song about how fake and bland American consumerism is. Baudrillard writes a lot about American phenomena which surprises me about an old Frenchman in the 1980s.

I really like calling theme parks “stations” of an alternative reality. Just like stop at the gas station to buy gas, you stop at the reality station to stock up on experiences of an imaginary world.

“Is any given bombing in Italy the work of leftist extremists, or extreme-right provocation, or a centrist mise-en-scene to discredit all extreme terrorists and to shore up its own failing power, or again, is it a police-inspired scenario and a form of blackmail to public security? All of this is simultaneously true, and the search for proof, indeed the objectivity of the facts does not put an end to this vertigo of interpretation.

What a political event means is defined not just by what happened, but also how it was interpreted. It can be very influential even if it was fake. COVID, for example, was a nothingburger as a real epidemic but a civilizational milestone in how it gave politicians license to ignore the law about personal liberties. This is an interesting point.

This is how all the holdups, airplane hijackings, etc. are now in some sense simulation holdups in that they are already inscribed in the decoding and orchestration rituals of the media, anticipated in their presentation and their possible consequences. In short, where they function as a group of signs dedicated exclusively to their recurrence as signs, and no longer at all to their “real” end. But this does not make them harmless. On the contrary, it is as hyperreal events, no longer with a specific content or end, but indefinitely refracted by each other.”

Also, political action is performative: when people took over the Capitol on January 6, 2021, they didn’t literally want to kick out the real politicians and establish their alternative government inside the building. It was just a protest, organized at a high-profile location for dramatic effect. The point about plane hijackings is quite original. Normally, you would hijack a plane to force it to go to another place. For example, in 1971 some guys hijacked a plane in Communist Romania to flee to the West – a commercial plane, taking off close to the border no less, is a very practical way to get through an otherwise hermetically closed border. In most plane hijackings, however, the hijackers just want to make a point or kill people like any other terrorist.

These aside, there is very little substance to Baudrillard’s essay. Like all postmoderns, he uses big words and incoherent prose to make midwits believe that he is saying something important. In fact, his message is shallow and unoriginal, not really anything beyond “people believe in a fake reality”. He invents potentially useful terms – specifically “simulation”, “hyperreal” and “simulacrum” – but uses them sloppily. Sometimes the three terms are used as synonyms, but sometimes even the same word means a different thing a different context. For example, in the essay and some secondary sources I looked at the following, wildly different things are described as “hyperreal”:

- A sex doll simulating a real partner, Disneyland (in both I knowingly choose to immerse myself in a fake reality)

- Identification with fictional heroes (also immersion, but less conscious)

- “Heidiland”, a Swiss alpine village invoking an idyllic novel (here, seeing a Potemkin imitation of fiction in real life makes the fiction feel more real)

- Reality TV (cherry-picked and curated story presented as real)

- Political propaganda (knowingly misleading people)

- The inability to distinguish reality from a simulation of reality (the complete opposite of Disneyland).

That said, the social construction of reality is an important topic and Baudrillard’s terms are catchy. To use all this for postmodern bullshit is such a waste that I decided to write my own version of Simulacra and Simulation, in just a few, hopefully lucid pages. I don’t want to explain Baudrillard (“people believe in a fake reality, man” comes close by itself) but write an alternative essay about the same topic. I will salvage Baudrillard’s three neologisms: simulation, hyperreality, and simulacrum. I will not bother trying to use them in Baudrillard’s way, which is sloppy and inconsistent: instead, I will attach them to precisely defined concepts as I see fit.

Simulacra and Simulations – the East Hunter version

The representation of reality in the human mind comes from two sources: the senses (including memories of what we sensed and things we logically deduced from it), and second-hand information (including gossip, the news, and fiction).

Simulation

This second part is what Baudrillard may have called the simulation, a parallel reality running in our head. In our evolutionary history, most of the information in the simulation was probably correct and a meaningful addition to sensed reality. For example, it must have been great to hear from another caveman that that the next valley had a good hunting ground, even if we never saw it ourselves, or to be able to track deer based on signs using trail markers we only heard about and never directly saw the deer leave. However, in more modern times, life became too complex to reliably cross-check the simulation with the senses. Therefore, our tendency to perceive sensed and simulated (~based on second-hand information) reality as equally real led to some interesting or dangerous phenomena.

One interesting thing the simulation does is that it enables us to suspend disbelief and enjoy a reality we know to be fictional. It’s a very interesting byproduct of human cognition that we entertain ourselves by switching off sensed and switching on simulated reality. Baudrillard uses Disneyland, the story of Santa Claus, and reality TV as examples, but novels, ordinary TV shows, sports fandoms, porn, and increasingly even politics works this way. The modern person moves on autopilot in his real, sensed, physical life – eat, work, sleep, repeat – and most of his mental energy is dedicated to the escapism of fandoms, hobbies or politics. This is not a very original observation but I think we don’t appreciate the strangeness of there being things in our head that feel real even though we never personally experienced them. Could there be an intelligent alien species incapable of enjoying fiction, or is the ability to take imaginary realities seriously necessary for having ambition and planning for the future?

Another interesting fact about the simulation is that it is the only way we understand the modern world. This is because a modern society, unlike a tribal group of a few hundred people, is too complex for you to know everything about it. Any understanding of it has to cut corners and leave out details so it is necessarily incomplete. For example, let’s look how different people explain social inequality, a phenomenon you can still sense personally e.g. by seeing homeless people and others in expensive cars side by side. An ordinary sociologist will tell you that your rearing family – essentially unearned privilege – determines social success, so we must tax the rich and give more benefits to the poor. A HBD-aware one will tell you that genetic factors are much more important, compensatory education is a waste of time, but something like embryo selection can work. Because it’s more closely based on sensed reality (objective data, e.g. twin models or educational experiments), the latter is obviously more correct, but HBD is still an imperfect model of society which leaves out many individual cases and special considerations. Some people are impacted by trauma or benefit from compensatory education, nothing is 100% heritable, shared environmental effects are not 0. HBD is science which provides a good enough understanding of society to choose better or worse social policy, but it’s not a complete explanation of everything that happens to people.

And this is when we try to use quantitative methods and the best of our logic to make sense of a complex reality! In the mind of the ordinary person, the reality of the world outside their immediate surroundings is a flurry of anecdotes, slogans and impressions. Depending on which specific anecdotes, slogans and impressions – usually through some media – you were exposed to, you basically live in one or another alternate reality.

Maybe the first time the existence of these alternative realities became apparent was the 2016 US elections, where voting maps often looked like this:

In this example of Beaumont, Texas, 95% of the northern city district voted for Hillary Clinton, while 91% of those in rural Lumberton, immediately next door, voted for Donald Trump. In the sensed reality, these people lived extremely close, but their simulated reality was very different and this mattered much more. They probably drove down the same roads and saw the same buildings each day, but depending on what they saw on a screen they carried in their pocket they were upset either about rape culture, the gender wage gap, Black poverty or immigrants, elite pedophiles and the loss of American industrial power, all more or less completely fictional issues in the sense that they either don’t exist or don’t directly affect the modal voter in a way he would realize unless the media told him it does.

Donald Trump winning the 2016 elections (and a few months earlier, the successful Brexit vote) shocked left-leaning Western elites because it made them realize that they don’t control the simulation anymore. With social media, blogs and independent media outlets, it was not enough to have CNN and the New York Times push your message: for better or worse, people had easy access to alternative narratives. The subsequent push for censorship and against “misinformation”, peaking during Covid, was an admission about the importance of the simulation: people vote not just based on what they personally experience, but also what they hear, so if we have the resources to broadcast a well-designed message to many people, we can hijack democracy and get anybody elected. Actually, this is a big reveal about the nature of democracy itself. If misinformation can be a problem, then every election ever is rigged because all you need to sway them is saturate the airwaves with your propaganda. The only way it could be otherwise is if you believe that not only does the mainstream media present a complete and unbiased image of reality, but the modal voter is willing and able to accurately process this image. Of course, Western elites loved this arrangement until they had the only megaphone, but once others were doing it, they were worried that they would lose the script.

Another effect of the power of the simulation is that we often don’t believe reality until we see it in fiction. This is the “word magic” I wrote two posts about. In these, I showed that there are many instances where the now-trivial idea can be traced back to a single work of fiction. My favorite example of this was “bodybuilding”, which went mainstream after Arnold Schwarzenegger’s movie “Pumping Iron”:

Google Ngram word frequencies of “bodybuilding” in print media before and after “Pumping Iron” came out. See the original post for more methods.

In the sensed reality, it was always possible to lift weights and become muscular and, on some level, people had been doing it since antiquity. But it took the charisma of Arnold to make it part of the simulation for many people. Today just about every young guy works out.

Hyperreality

We mentioned immersion in fiction as one of the byproducts of having the simulation, a mental image of reality which is based on second-hand sources and which we still perceive to be equally real to the reality we sense. However, when we use fiction for entertainment, even though we suspend disbelief we are usually 100% aware of it being fictional and we can switch it off any time. Conversely, when we use the simulation to understand a reality too complex for the senses, which was our second use case, we sincerely believe that the simplified model we have is a correct depiction of reality – social success depends on how your parents raised you, the Chinese cannot innovate, global warming will kill us all.

But sometimes we get an interesting in-between case, when we know something is oversimplified or downright fictional but we still choose to believe it as a model of reality. This is what I decided to use Baudrillard’s cool term hyperreality [1] for.

It comes to us naturally to identify with fictional scenarios. The first thing small children do is pretend-play as characters in their favorite cartoons. Even adults sometimes base their entire personality on something they saw as a simulation, especially like when they are young. This is what people did when they saw the Sex Pistols and dressed up as punks themselves. Baudrillard implies that hyperreality is a modern phenomenon, but I disagree. Alexander the Great basically LARPed as Achilles. Both Nero and Commodus are known to have styled themselves after the mythological Hercules. We don’t really have accounts of the thoughts of normal people from these times, but looking at how easily immersion in fiction comes to us, I bet that every village drunk in the Roman Empire half-thought he was Bacchus or at least Diogenes, and every legionnaire imagined himself to be some mythological hero. This is hyperreal thinking, although a mild one, because for most people it doesn’t take much to switch the LARP off.

A nice modern high-profile example this blurring of fiction and reality is an official NATO tweet framing the real-life Ukraine war in terms of young adult fiction:

By my definition, hyperreality starts when people choose to ignore the fictional nature of the contents of the simulation in their evaluation of sensed reality. Under my definition, Disneyland would be a simulation (we enjoy it but know it’s fake the whole time) but seriously believing that the Manichean black-white universe of a Marvel movie is a good template to understand the Russia-Ukraine war is hyperreal [2].

Note that while it’s easy to define hyperreality (simulation=I know and act like it’s fake, hyperreality=I know it’s fake but act like it isn’t), it’s not easy to say if it was present in a specific case. Hyperreality is a spectrum, non-insane people eventually snap out of the simulation with a big enough push but it’s hard to say beforehand how much is needed. Where is the line between enjoying a Western-themed Disneyland ride and dressing up as a cowboy everyday and acting like you’re John Wayne? Did the NATO intern really believe that Putin is Voldemort, or was this tweet designed by a committee for maximum virality? The simulation becoming hyperreality happens in degrees, but there clearly are edge cases when we consciously immerse ourselves in fiction for fun, and others where we act wholeheartedly as if fiction was fake.

Hyperreality works particularly well when 1) there is a partial confirmation of the simulated reality from the senses but 2) realities alternative to the hyperreal fiction only exist in the simulation (that is, you would need to do research to realize that the anecdote you just saw is an outlier). This is often how propaganda works: it creates a story and finds cherry-picked real-life anecdotes to support it. Maybe statistical reality contradicts the story, but the combination of a fake story (a simulated reality) and the anecdotes (sensed reality) is more powerful than the real story (just another simulated reality). For example “America is full of homeless junkies”, supported by videos of Kensington Avenue is a hyperreal message. “Most the US is not like Kensington Avenue at all” is the statistical reality, but as a message it feels weak and defensive. It would work better if it was supported by a series of Norman Rockwell-esque images and anecdotes. The anecdotes are key: if you are a good propagandist and you want to bring your message about a dysfunctional America home, you want to repeat the same message many times, each time supported by a different image of junkies, criminals or a different cherry-picked statistic. It takes a rare and special kind of person to look beyond impressions and anecdotes and immediately want to see the numbers.

Ironically, given that Baudrillard doesn’t seem to like hyperreality or simulations very much, it was his colleagues, European leftist philosophers, who mastered using hyperreality as a propaganda technique. The idea of this was fully developed by Herbert Marcuse, who discovered that the marketplace of ideas was just like a real marketplace: when there is little liquidity, prices are low so you can buy up everything with little cash. (If nobody bought stocks for some reason, after a while you could have NVidia for 1$.) In the same way, if you are the only one writing about the topic, your word is the only one there is. Marcuse’s idea was to organize a “long march through the institutions”, to hire fellow Marxists (and have them hire more) and write about society in a way charged by their politics, including both theory and slanted news reporting in support of it. At the time (this is the 60s, not the 2010s), there were not many people writing about race relations or feminism so with a handful of faithful and productive propagandists he could corner the market and give the impression that every serious person who writes about race considers racism to be a very serious problem that underlies all aspects of Black dysfunction, or that women are oppressed and need to reach outcome equality with men. Fifty years later when this paradigm fully matured, the 2020 BLM riots were the perfect example of Marcusian, hyperreality-based political action. (See this recent article for many stories about a Marcusian ideological takeover of American institutions in the 2010s). A simulated narrative was constructed by the press and by academics (“Black people are unfairly targeted and killed by racist police”), and partial confirmation was provided by massively reporting the real-life killings of Ahmaud Arbery and George Floyd. Although statistical data shows no evidence for police using more force against Blacks, and these police killings are orders of magnitude more rare than most people, especially liberals, believe, stories and anecdotes are sufficient to trigger hyperreality. Memetically successful alternatives (4chan style online racism) also used statistics and anecdotes like viral videos of Black violence side by side.

Simulacra

Another amusing byproduct of our ability to equate sensed and simulated reality is that sometimes we believe in something completely fake because it looks good in the simulated reality and not challenged enough by the senses. To find an exact function for one more term used by Baudrillard, this is what we could call a simulacrum. In my framework, simulation is a parallel reality in our head, based on second-hand information, and we are normally realistic enough about what is – we can immerse ourselves in it (Disneyland, fiction), but we know it is a different thing than the world we sense around us. Hyperreality is when simulated reality makes us interpret sensed reality incorrectly (seriously LARPing as fictional characters, BLM). A simulacrum, in turn, is when we only have second-hand, simulated information about something which happens to be false but we don’t realize this.

Female serial killers are my favorite example of a simulacrum. There were many male serial killers who killed for thrill-seeking, sadism or sexual motivations, like Ted Bundy or Andrei Chikatilo. There are fictional stories where females do this too (e.g. Eye of the Beholder, Killing Eve), but in reality, this is almost completely unknown. Practically all the “female serial killers” killed people under their care with poison or similar methods, usually for financial gain [3]. For a long time, I wasn’t aware of this. I obviously knew that Eye of the Beholder was just a movie when I saw it, but I knew that 1) serial killers randomly preying on people existed and 2) some serial killers were female, so I implicitly took the movie as confirmation that the intersection of 1) and 2) was not empty. Another example of a simulacrum is the fake news that the Burj Khalifa (in some versions of the story, all of Dubai) is not connected to a sewer system, which fits the stereotype of dumb rich Arabs but isn’t actually true. Simulacra exist when we care about something just enough to pick up the first piece of superficial information in our way. Short-form social media content is a fertile breeding ground for simulacra: a lot of the information people have about the world comes from clickbait images, short videos, infographics and cherry-picked text anecdotes, all of which may or may not be AI-generated.

[1] This is not Baudrillard’s definition. Baudrillard is unclear if “simulation” and “hyperreality” are one and the same thing. We need to distinguish them. I propose that all internal reality (whether or not we are aware of its inaccuracy) should be called “simulation. Hyperreality would be a narrower and fully contained category: all hyperreality is simulation”, but not all simulation is hyperreality. Hyperreality is when we base our understanding our reality on a simulation we know to be fictional. Disneyland is not hyperreal because it’s only “hyper” (part of the simulated, not the sensed reality), but not “real” (nobody thinks it’s real).

[2] Maybe some clarification is needed for this point. False knowledge (the simulation being wrong) is not hyperreality by itself. For example, if a company lies to you about its performance (simulation, second-hand information) and you decide the invest (reality), your real-life decision is based on the imaginary reality of this company being successful. But this is not an example of hyperreality because you sincerely believe the fake performance report is the truth. For hyperreality, there must be some knowledge of the truth and still at least a subconscious decision to ignore it. For the same reason, it’s also not automatically hyperreal if you believe Russia is ontologically evil and Ukraine is ontologically good (as long as you possess the concepts of good and evil without having seen them in fiction). Having a strong opinion about the conflict doesn’t make your reaction hyperreal. What does is if you borrow your description from a source of information you know to be fictional. If you used the Bible as your fictional reference, it would still be hyperreal. But if you believed the Avengers to be a documentary, Thanos to be a real person, the Russians to be his army, and the Avengers to be embedded with the Ukrainians, your interpretation of the conflict would not be hyperreal. Choosing (at least subconsciously) to mix fiction with reality is what constitutes the hyperreal.

[3] As far as I know there was only a single female serial killer somewhat fitting the stereotype of violently hunting strangers for pleasure without a male partner, Joanna Dennehy who stabbed five random men for fun in 2013, three of whom died. Even she didn’t kidnap or torture them and she did her crimes over a few days unlike the classical male serial killers who often operate for years. Other rarities include Aileen Wuornos (a prostitute who shot and robbed six or seven clients), Tamara Samsonova (a schizophrenic grandma who killed her tenants), Nicola Edgington (schizo who killed her mother and later stabbed two strangers on a spree, one died), Louise Peete (an adventuress who shot, rather than poisoned, her lovers for money), and Sharon Kinne (a good-looking sociopath and fugitive for 50 years who opportunistically killed three people over her career). But I went through the whole Wikipedia list for this sentence which should show you how rare it is even for a female schizo to go on a spree. Women are not very violent.

But is this all true?

I decided to check this empirically[1] in a large public dataset of Americans. Participants self-reported if they were non-Hispanic White, Asian, Black or Hispanic/Latino. All four races were reasonably well represented.

In the amazing Mike Judge series Silicon Valley, some startup founders try to build a “Shazam for food”, an app that lets you can take a picture of your meals and it uses AI to automatically code it and calculate nutrients. There are apps like this now, which are OK but far from perfect. In the absence of a perfect AI, there is no very good practical way to record what people eat. Ideally, they would measure with a scale and accurately log everything they eat, but you can’t realistically recruit people conscientious enough to go through with this. Another way is to do a 24-hour dietary recall, that is, once a day ask people what they ate. Another is the Food Frequency Questionnaire (FFQ), where people report what they usually eat. It looks like this:

My dataset has this FFQ, recording whether and how frequently participants eat a large number of food items.

First question: The consumption of which food items varies the most across races?

For this, I built a regression model for each food item with consumption frequency as the dependent and race as the single predictor variable. R² from these models tell us which food items have the most race-dependent consumption patterns. Here is the list:

As you can see, the top items are typical ethnic food, probably preferred either by Asians (stir-fried items, Asian noodles, dumplings) or Hispanics (Mexican rice, tortillas, beans). Eating the skin of potatoes seems to vary a lot across racial groups, this is unexpected.

Second question: Which food items are the most preferred by a single specific race?

You can get a high R² value by having two races prefer a food item and the other two almost never eat it at all. We are going one step farther now: which are the food items which are only preferred by a single specific race and not the others? In other words, which are the most White, Black, Asian and Hispanic foods?

For each food-ethnic group combination, I calculated a race preference score:

(frequency_race-frequency_{other_races})/frequency_mean*100

In plain language, I calculate how much the preference of a certain race for a food item differs from the mean preference of the other 3 races. Dividing by the grand mean frequency and multiplying it by 100 makes it approximately a % difference. For example, the White race preference score for “Wine” of 170 means that White people consume about 170% more wine (not just 70% more!) than the other three races. Ranking the food items by this race preference score highlights the “Whitest”, “Blackest” etc. foods.

Asian and Hispanic foods are unsurprising, these ethnicities eat a lot more of the food items you’d stereotypically expect them to like:

Asians eat Asian food and Hispanics eat South-Central American foods, no surprises here. What is the Blackest food then?

Of course it’s fried chicken. Watermelon is not in the FFQ, but unusual fruit juices and non-diet beverages among the top items somewhat confirms that American Blacks like very sweet, fruity beverages. In line with the Bennett review I linked above, Blacks don’t seem to have a very healthy diet, many top items are deep-fried or fatty meals and sweets.

What about White people?

Whites stand apart from other ethnicities mainly by drinking more alcohol. All alcohol items (wine, liquor and beer) appear among the top 20. The other things that stands out are dairy items: yoghurt, cottage cheese, cheese, butter, milk etc. This is most likely because there are very big race differences in the rates of lactose intolerance: it is rare among Europeans but very common outside of Europe. The basal human phenotype is that you can’t drink milk anymore when you stop breastfeeding. In regions where people had animals they could milk (goats, sheep, cows) it was very advantageous during famines to be able to drink this milk, even when you were an adult. Therefore, multiple hotspots across the globe independently evolved lactase persistence (the ability to drink milk past infancy), but it didn’t happen everywhere. Among Americans, the rate of lactose intolerance is several times higher if they ancestors came from outside of Europe. Interestingly, Asians have practiced agriculture for a long time but they still have the highest rates of lactose intolerance.

Third question: How different overall is the diet of American racial groups?

Other than Asians eating Far Eastern dishes and Hispanics South-Central American foods, at first glance the race differences in diet don’t look that big. But of course we must be aware of the univariate fallacy: there are many human characteristics, and a small difference in each of them can balloon in a very different overall phenotype. For example, people will tell you that men are women are not that different in any single characteristic (unless it’s upper-body strength), or that there are only small race differences in the frequency of any genetic variant (unless it’s Duffy negativity). But this is what we call the univariate fallacy: objects differ from each other along multiple characteristics and if we use all of them like we should, we can classify them much better. In reality you can classify sex or race with high accuracy if you look at a larger pool of psychological or genetic characteristics. For example, having a different English accent (an immediately noticeable difference) is the sum of many small differences in how we create phonemes.

For this reason, it is important that we go beyond single food items and look at the big picture: how well can we predict race from self-reported food preferences?

For this, I used a simple random forest classifier in R. All food preferences were used as predictors and the four-category race variable was the outcome. In a holdout sample, 82% of people were correctly identified as members of their race based on food preferences alone, over twice the random guessing rate of ~40%. Cohen’s Kappa for this is 0.74 which is a pretty good value.

Confusion matrix in the holdout sample (food preference-predicted and actual race)

Note that most of the misclassification happens between Blacks and Whites. This shows what we saw before, that Asians and Hispanics are the ones with particularly distinctive diets. This is also apparent when we plot the most important variables: these are almost all ethnic food items, showing that whether you eat something very ethnic like “tortilla as side dish” or “stir-fried vegetables” provides the model with the most information about your probable race.

Fourth question: How accurate are stereotypes about ethnic food preferences?

Some people may find stereotypes about ethnic food preferences offensive. I don’t understand this - what’s wrong about liking fried chicken? – but even if you do, it doesn’t matter very much because the stereotype is either correct or not, which is what matters. Stereotypes are usually quite correct – Lee Jussim famously wrote that this is the best replicated finding in social psychology – which is unsurprising because humans evolved to observe each other accurately.

To establish if ethnic food stereotypes are accurate, I created a poll which a friend with a large following posted on Twitter. In this poll, I copied the name of each food item I had and asked participants to rate if, in their view, “White people”, “Black people”, “Asian people”, or “Hispanic people” prefer this food the most. If they were unsure, they were prompted to give a random guess. I received 25 answers, which was a positive surprise given close to 200 quite weird questions and no compensation. Most participants were from the US or some other English-speaking country, with a college degree.

The first thing I noticed is that people believe food preferences are much more concentrated within races than they really are. To quantify this, I calculated a pseudo-R² of respondent-rated food preferences. The most extreme case would be if everybody said the same race likes a food item the most: for example, if all 25 respondents agreed that Black people like fried chicken the most. I wanted to be R²=1: race explains all the food preference. The other extreme (R²=0) would be a flat histogram: an exactly equal number of people saying that either Whites, Blacks, Asians or Hispanics like fried chicken the most.

The actual R² value was defined as:

In plain words, for each food item, I summed up the deviation of preference ratings from a flat histogram (=what % of people say each race prefers that food, and how different is this from 25%?) as the numerator and divided this with the most extreme case (everybody agrees Blacks like fried chicken the most) too see how close it is to the truth.

For the current case with K=4 races:

Survey respondents guessed a bit better than random (r=0.15, p=0.09) which food items showed the most race-specific preference patterns. As I said, they usually believed preference for food items is much more concentrated across the races than they really were.

For most food items, the concentration of preference within a specific race was actually quite low (Actual R²). However, external respondents often – falsely – agreed with each other (Predicted R²) that a single race prefers a food item the most.

To really evaluate stereotype accuracy, for each race I correlated self-reported food preferences (% extra preference compared to the other races, as above) with community-predicted preference. This community-predicted preference was the proportion of respondents saying a certain race prefers that food, minus the mean share of the others. This was to reproduce the logic of self-report data, which also looked at the difference between the preference of each race and the mean preference of the others. For example, if out of four respondents three said Black people like fried chicken the most and fourth said it was Asians, Black preference was 0.75-(0.25+0+0)/3=0.67, Asian preference 0.25-(0.75+0+0)/3=0, while White and Hispanic each 0-(0+0.25+0.75)/3=-0.33. Thus, the scale of self-reported and community-predicted preferences are not comparable but they follow a similar logic. If stereotypes are accurate, we would expect that Black people really said they like this food more than the other three races.

For Hispanics and Asians, stereotype accuracy is very strong:

These correlations at r~0.8 between community-predicted and actual food preferences are about as strong as you will ever see in the social sciences. People correctly guess which food items are preferred by Asian and Hispanic Americans. On the other hand, this is not that impressive, because both ethnic groups have a distinctive cuisine with well-known items. In the case of Hispanics, even the language of the food names is often different. Maybe it’s surprising to some that Asians and Hispanics really eat their ethnic food, but it’s not that big a shock.

Even for Blacks, there is strong stereotype accuracy, with r=0.53. Fried chicken is correctly predicted to be the most stereotypical Black food, but this is not the end of the story, other reasonably Black items are also seen as such by respondents.

I find the White chart the most impressive:

Unlike Asians and Hispanics, American Whites don’t have a plainly obvious ethnic cuisine. The Asian and Hispanic correlations arguably just reflected that these groups have an easily identifiable ethnic cousine, and the Black correlation was boosted by the Asian/Hispanic food items in the bottom left quartile, correctly guessed to be rarely eaten by Blacks. But here we see a nice linear relationship between predicted and actual preference, and a correlation of 0.6. It is interesting to note that the White preference for alcohol consumption is not widely known – look at how far wine, liquor and beer are above the regression line.

In summary, diet is one of the many things that shows ethnic/racial differences in America. People overestimate how much the races differ in their consumption of single food items, but they get it right which food items are typically consumed by which race, and a machine learning approach can classify the race of respondents with over 80% accuracy.

[1] You can also look up the review by Bennett et al (2022).

In the town I’m from, there was an accident some 50 years ago when a bus drove in front of a train and a dozen or so people died. There is a memorial at the site, often with fresh flowers and a commemoration each year. In contrast, there are barely any memorials[1] anywhere in the world to the victims of the COVID pandemic, which by official figures killed some seven million people. The pandemic, definitely the craziest event of my life, is being buried by a collective cultural amnesia. Contrarianism is mainstream, people love to roast their governments, but they do it for whatever tier things like the Epstein list, minor corruption scandals or putting fluoride in the water, but their insane response to COVID just six years ago seems to be forgiven or at least forgotten. I sometimes think that people want to forget what happened: the news media hyped up a relatively harmless respiratory virus into a world-ending event for clicks, people believed them and freaked out, and governments – advised by and with the full support of an either brainless or profoundly corrupt medical establishment – threw 300 years of personal freedoms in the garbage and implemented ridiculous non-pharmacological interventions to stop it and then doubled down when they didn’t work. Along the way, they ended the prosperous 2010s by throwing a monkey wrench into our fine-tuned just-in-time economies and printing money to paper over the damage.

You should be foaming at the mouth with rage at your fellow citizens for their cowardice, the medical establishment for its stupidity and corruption, and your government – almost no difference what kind! – for exposing you to lockdowns and mask mandates. This post is my way of doing it.

COVID was a mild virus

I read about medieval epidemics when I was a teenager. I saw Outbreak and all the epidemic disaster movies, read stories like The Plague or Year of Wonders and although never The Stand by Stephen King, I read Night Surf which it was expanded from. So when I heard that there was a deadly epidemic in Wuhan and the Chinese were implementing an insanely strict quarantine I was worried. There have always been two key questions about COVID, and all policy decisions were based on what people thought about them: 1) how dangerous is this virus, 2) how effectively can spread or mortality prevented by various methods. In January 2020, based on what was happening in Wuhan I thought it was a very dangerous virus which can be stopped by drastic non-pharmacological measures like quarantines and masks.

I was wrong on both counts, and so was everybody else. But while the COVID hysteria lasted until it was replaced by the new Current Thing in Ukraine, I was already banging my head against the wall in April 2020.

Google Trends showing a shift between the Current Things. (Ukraine started being a hot topic weeks before the war broke out, coinciding well with dropping interest in the virus.) When will Vladimir Putin get the Nobel Prize for Physiology and Medicine for curing COVID?

I stopped being worried about the severity of COVID around February 2020 when I saw an interview with a young couple aboard the Diamond Princess who laughingly said that the COVID they caught is nothing compared to the flu they had had before. I can’t find the exact interview but here is a similar one with a young woman in a Yokohama hospital, who, when interviewed by a worried CBC host literally starts describing her situation as “I feel pretty well, I never really had any symptoms like a cough.” This interview has been on Youtube since February 11, 2020, a month before European countries really started doing lockdowns. The Diamond Princess natural experiment proved to be a pretty good laboratory of COVID in general. 2% of the 712 confirmed cases – mean age: 68 years old – died, all in their 60s or, usually, much older. Wikipedia gives the case fatality rate – confirmed death by confirmed case – as 1.02% globally. Of course, this is an overestimate of the deadliness of COVID. Many cases were not noticed because they were asymptomatic, there were not tests available, or people (like me, who never had a single COVID swab) didn’t bother to test. When you add these in, you get the infection fatality rate – deaths per all cases. IFR is tricky to calculate: you have to sample a random and relatively representative sample of people who didn’t plan to be tested (like everybody who passes through a train station over an hour) and check the percentage with a positive test – you can extrapolate that the population to see how many were infected and compare fatality rates that that. For the sake of simplicity I just show the IFRs reported on Wikipedia again, based on this study:

These numbers are quite similar to the EU-wide annual population mortality rates for these age groups – slightly lower before age 65 and slightly higher above. In other words, you were about as likely to die from COVID as from living another year. Are you worried you will die within 12 months? You dying from COVID had about the same probability of this happening once you caught it.

COVID was reasonably dangerous for people in their 70s and 80s but not a big deal for anybody younger. Total IFRs are somewhere around 0.5-1% which is not great, not terrible. In comparison, the Spanish Flu had a population fatality rate of at least close to 1%, but in some cases several percent and mostly killed the young and healthy. The milder Hong Kong flu had a case fatality rate about a quarter to a half of COVID and hardly anybody heard about until COVID itself struck. Tuberculosis, traditionally the deadliest modern infectious disease, kills over 1 million people per year, compared to about 7 million over 4 years from COVID, and you never hear about it. I’m not even getting started about the historically famous pandemics like the Black Death or smallpox epidemics. I’m very sorry for those who lost loved ones in COVID, this obviously was a real and potentially very dangerous illness, but a pandemic with only COVID-level lethality, especially one basically only affecting anyone the elderly is historically not a big deal and should have been on the back pages of newspapers.

Non-pharmacological interventions didn’t work at all

Of course me saying that COVID was not a big deal is just my opinion. There was a virus and there was a disease and many people certainly died from it, and even those that didn’t often got sick. It was not a world-ending event, but we would have certainly been better off without it, so it was understandable to act against it. Initially, this was done with non-pharmacological interventions: masks, closing businesses, curfews, and sometimes stay-at-home orders. I think that with the possible exception of masks these have such a high societal and economic cost that they are almost not worth it by default, but this is also subjective. But what if non-pharmacological interventions don’t work at all? What if they are all cost, no benefit, and we are not saving a single life by collapsing society for months? I think this is the most likely description of what happened.

I first suspected that lockdowns don’t work in March 2020 when I looked at Italian data. This was before Hungary even had a lockdown. Italy did its first local lockdowns on February 21 in Lombardy. I wasn’t very surprised that China did a lockdown, but I wasn’t expecting a country in Europe to quarantine entire cities with armed cops. (Of course in the years to follow European society showed its true face, going all-in first on COVID and then on the Ukraine hysteria.) I was sure that even though such a barbaric measure was not worth it – after all, from the Diamond Princess data I knew it was a relatively mild disease only seriously affecting the elderly – it would work, because it worked in China and after all, how would an infectious disease not stop from a quarantine?

So I was surprised that cases and deaths went like this:

Ironically, Wikipedia captions this picture as “showing the effects of lockdown and lifting”. I guess it does show the effect: absolutely nothing, neither the lockdowns nor their lifting did anything to either cases or deaths at all.

You may dig up some paper showing some effect of masks or lockdowns, but a drastic overreach by the government like what happened during COVID needs much stronger evidence for efficacy than p<0.05. It needs to pass the most stringent statistical significance criterion, the ocular impact test (it needs to poke your eye out). There are some examples for drastic measures with an obvious immediate effect:

Left: the number of illegal immigrants crossing the Hungarian border in October 2015 before and after the completion of a border fence. Right: shots and 911 calls before and after George Floyd’s death. Building a fence keeps illegals out and stoking a fake outrage over racism kills Black people, no fancy statistics needed to prove this.

Non-pharmacological measures against COVID, however, were not like this at all. I invite you to try Tim Scott’s COVID Charts Quiz game too where you need to guess when and where lockdowns were initiated by eyeballing graphs. You won’t be able to because, like in Italy, there isn’t much of a correlation. But COVID lockdowns were like feminism or affirmative action: if the data showed they don’t work, it was all the more evidence that we are not doing enough of them.

Anecdotes like this are useful, but we want more objective data. So I fired up the tinycovid19 package by Joachim Gassen which can load COVID time-series data directly into R, from all countries and all the way from 2020 to the end of 2023. The most straightforward metric to calculate is the cross-correlation between lockdown severity (indexed by the Stringency Index calculated by this team, also used by Our World in Data) and COVID cases/deaths. The cross-correlation is just like the ordinary correlation, except temporal shifts in the dataset are explored. A cross-correlation with 0 lag is literally just the correlation. By introducing positive and negative lags in the data and seeing if the correlation changes we can check if one variable tracks the other. In the current case, we would expect that trends in COVID infections and deaths will lag lockdowns: if the severity index increases (lockdown is on), fewer people get infected and eventually fewer die. Because the disease has an incubation period and the severely ill take time to die, one would expect at least a few days of lag between the severity index and cases and some more in the case of deaths.

This is not what the data shows at all. Here are the cross-correlations from a few countries (Hungary, the US, Germany and Great Britain):

These charts show how the correlation of the two time series (stringency index and cases/deaths) changes if we introduce time lags. The dots show the maximum correlation: our best guess of the temporal relationship between the time series is the lag value below. German cases are weird for some reason, but otherwise we see that trends in cases and deaths often precede, rather than follow, the stringency of lockdowns, especially if you consider the ~1 week incubation period and the time needed for a lethal infection to run its course after that. Lockdowns followed rather than caused trends in the disease: in other words, they didn’t work.

Here is the distribution of the correlations and maximum lags across all countries (after deleting -30 and 30 lags, indicating that the 30-day window is not enough to capture the temporal relationship):

Deaths track stringency measures more accurately (higher maximum correlation, left chart) than cases. I guess the number of detected cases depends on available tests, the willingness of people to get tested, and many cases are asymptomatic, but deaths really do increase whenever there are more viruses out there. Either way, across all countries on average both cases and cases correlate with stringency measures with about 0 lag, which shows that stringency measures didn’t change cases and subsequent deaths. What happened is that whenever cases and deaths increased due to the natural course of the pandemic, stringency measures were increased, like when some island tribe starts throwing virgins into a volcano during a drought to appease the gods. Note that the data could show both this panic effect and a genuine effect of the lockdowns in the form of a double cross-correlation peak: but this is not what we see.

When I showed them this data, some people told me that maybe this is the wrong thing to look at, maybe lockdowns didn’t immediately lower COVID deaths but freed up hospital capacity which still saved lives or something like that [1]. So let’s look at another metric: mean stringency index over the entire 2020-2023 COVID period vs. COVID mortality. Did countries with tendentiously stricter lockdowns end up with fewer deaths? No, the raw correlation is 0.06, so in the opposite direction, p=0.43. Let’s control for median age, GDP and vaccination rates: then, the correlation is -0.03, p=0.76. Let’s use total cumulative excess deaths instead of confirmed COVID deaths: r=0.08, p=0.48. One could crunch this data in other ways but in practice there is not much to talk about if the ocular impact test is not passed and we need to quibble about technical specifics to see a non-zero effect. There is not much in this data, non-pharmacological interventions did nothing obvious to save lives.

Let’s recap this again: everything you did between 2020 and 2022 was completely in vain. It’s not even that masks, lockdowns and all the other insanity were not worth it, that much is obvious even as a thought experiment. No amount of 75 year olds allowed to live a few more years is worth a 10% drop in GDP and giving the government the precedent that it can micromanage what I wear on my face or if I can keep my business open. It’s that there is no obvious evidence that non-pharmacological interventions did anything at all. Also, this is not some secret knowledge that only became clear in retrospect: anybody with a brain could have realized based on Italian data in March 2020 that lockdowns don’t work, and it should have been blatantly obvious by October 2020 when the real global insanity only just began. Your head should be spinning because your country did this to you too and if not you or your prime minister then some egghead in the health department should have looked at the Lodi case numbers around March 10 2020 like I did and said that if the numbers are not going down after 20 days of lockdown for a disease with a 10 day incubation period at best then we should not do this whole clown show.

Vaccine hysteria

It’s really sad that the only lingering trace of COVID remembrance in the contrarian community is anti-vaxxerism. Vaccines worked (at least to prevent death in those few at risk), they had few side effects and forced mass vaccinations would have been a drastic and maybe morally wrong but at least functional social engineering measure.

There is little serious doubt that vaccines worked. The Hungarian study I linked by Vokó et al 2022 is probably the most extensive study about this, because Hungary had the only sensible vaccination policy in the EU, using Western (Pfizer, AstraZeneca and Moderna), Russian (Sputnik-V) and Chinese (Sinopharm) vaccines at the same time which enabled studying their efficacy side by side. They all worked great, despite liberal screeching about non-Western vaccines. The typical efficacy against death was 100% for young people: not a single vaccinated person under the age of 44 died, compared to hundreds of unvaccinated. On the other hand, there is not much evidence that vaccines were harmful, or at least that vaccine damage is more common than severe COVID (both are very rare). For an old or infirm person, a COVID vaccine could be life-saving, for a young one, it was a low-cost, low-benefit choice which I don’t have a strong opinion about other than it shouldn’t have been mandated. You certainly can’t stop the pandemic in its tracks with vaccines like many people hoped: even after massive vaccinations, cases remained widespread. The rational choice would have been to tell old people to self-isolate and wear masks, otherwise let COVID rip, fast-track vaccines and then let everybody get them whenever possible. We would have ended up exactly as we are, with many mostly elderly people dead, COVID one of the many circulating viruses around us but mostly harmless due to herd immunity and natural selection for less pathogenic variants which spread better, except without two years of mass hysteria.

Aftermath to a tragedy

Rather than being forgotten, COVID should be remembered every day. It should loom over us as World War 2 or the Holocaust does. There should be a torrent of memes, novels, movies, essays, newspaper articles analyzing every aspect of it. The worst part of COVID was not the seven million people (mostly old or infirm) who died from it – that is very sad, but as far as an epidemic goes, COVID was not a big deal. It was, however, a totally unprecedented global mass hysteria, fully supported by pretty much every government and, sadly, the medical-scientific establishment. The bad part was not the vaccines, which worked and were not without precedent: it was the masks and especially the lockdowns, which not even Stalin or Pol Pot did and which were rolled out when it was already clear that they didn’t work at all.

When I became first interested in politics, the biggest cliché I kept hearing was that the powerful love to use crises to expand their reach and this is how tyranny emerges. Everybody knows about Caesar and Hitler and saw the Star Wars prequels. COVID gave Western governments the opportunity to give a shot at this and it worked perfectly. After COVID normalized sidelining scientific evidence and ignoring laws and rights at will in policy decisions, no leader in the West is worried about drastic ad-hoc measures anymore. Trump randomly announcing and cancelling world-transforming tariffs, the EU sanctioning itself against its main energy provider or threatening nuclear war over the sanctity of the almost 80-year-old holy borders of Donetsk Oblast are all a direct consequence of the COVID precedent. All you need to do is a media campaign to sufficiently spook the semi-sentient modal voter and they will beg you to save them from the 0.05% IFR respiratory virus, the weather geting too hot in the Sahel in 100 years, Russian hackers or whatever else you make up. To some degree this was always true, but in an atomized, feminized, smartphone-addicted society the strategy works much better than with TVs and newspapers. I guess there are still good automotive engineers or cheese makers in the West but the traditional elite of 30–70-year-old, technically educated family men with common sense and a stake in the future is gone or lost power and without them there is nothing to stop the Chinese-style Western total state. Some people believe (I was one of them) that woke is the basic nature of this global total state, but it is not necessarily girlbosses, ugly dykes in videogames and Black Tolkien dwarves that the people in power want: they want more power. In retrospect, I think that MeToo, Black Lives Matter or Greta Thunberg hysterics were early attempts at creating a moral crisis that makes people call for a dictator: they did a lot of damage but mostly failed because they were too polarizing and galvanized a solid almost-majority against them, mainly along pre-existing ideological lines. In contrast, the COVID response had majority support and cut across political divides (for example, Hungary had typical lockdowns and Sweden did not) which prevented organized resistance. This power-grabbing strategy was immediately rolled onto the Ukraine hysteria which destroyed that country and wiped the EU off the map as a serious economic and geopolitical player (not to mention the immense damage it did to the economic and political life of Hungary, which is of course what I care about the most). Because we chose to forget COVID instead of facing how it normalized the state using mass hysteria to encroach on rights and freedoms, I’m afraid that very bad times are ahead.

[1] I’m not sure how this works with no effect on cases, or even deaths, because at the height of the pandemic a non-trivial proportion of all deaths were due to COVID. Also, nobody explained to me what capacity restraint hospitals had: there is no medication available, mechanical ventilators had a mortality rate approaching 50%, so beyond creating an illusion of professional care there is not much an ICU can do that a military bed in a giant tent cannot.

[1] Using AI I was able to dig up some. Apparently there were also some novels written about COVID.

I didn’t read any fiction in well over a decade, but recently I decided to give it a try again. One reason is that although I don’t doubt the primacy of science I increasingly feel that there is a subjective element of the human condition which is not easily captured by scientific methods. I’m interested in everything about people, not just a specific discipline. The other reason is that I feel that human science has a creative crisis and it is in desperate need for new ideas. There used to be lots of ideas, models and theories in psychology, especially social psychology, but many turned out to be politically motivated garbage and either bled out in the replication crisis (like priming) or live on as zombie theories (like stereotype threat or recovered memories). Since the replication crisis the idea seems to go hypothesis-free, throw at lot of data at a learning algorithm and try to make sense of the (super-certainly replicable) model it creates. The best example for this approach is GWAS, a true 2010s phenomenon, which, as amazing as it is, didn’t tell us very much beyond things like that “genes are important for traits” and that “intelligence genes are expressed in the brain”. Another favorite example I have for impressive but ultimately futile data-churning is this study about geographical effects of personality which tells us – with 100% certainty, huge database, no p-hacking for sure – that there is a correlation of -0.01 between altitude and neuroticism. I understand the enthusiasm for these methods, they are better than p-hacking N=10 experiments like social psychologists or writing up your weird opinions as science like psychoanalysts, but something is amiss. It turns out that the causes of human behavior are complex and multivariate… which is exactly what a novelist or poet would have told you. After a decade of non-fiction, I wondered if they have any other interesting things to say.

Below is my impression of some fiction books I read over the summer, specifically with the eyes of a social scientist who normally works with quantitative data.

Oscar Wilde: The Picture of Dorian Gray

This was one of my favorite books as a teenager and a motivation for learning proper English. I expected the book to still be fun, which it was, but I was surprised to find that it is even more relevant for my newer interests in human diversity. In many ways, this is a story about how to deal with unpleasant truths about human nature.

Protagonist Lord Henry is a dispassionate observer of people who wants to get the truth instead of polite noble lies. If living today, would probably be writing a HBD blog:

“He had been always enthralled by the methods of science, but the ordinary subject-matter of science had seemed to him trivial and of no import. And so he had begun by vivisecting himself, as he had ended by vivisecting others. Human life – that appeared to him the one thing worth investigating.”

Lord Henry, like Arthur Jensen, is careful not to be bogged down by theory and instead go where the data takes him:

“But he never fell into the error of arresting his intellectual development by any formal acceptance of creed or system, or of mistaking, for a house in which to live, an inn that is but suitable for the sojourn of a night, or for a few hours of a night in which there are no stars and the moon is in travail.”

He does go to places following his data. Lord Henry knows about hedonic adaptation (happiness coming not from having things, but successfully passing challenges):

“sick with that ennui, that taedium vitae, that comes on those to whom life denies nothing”

as well as alpha widows:

“Do you think this girl will ever be really contented now with any other of her own rank? I suppose she will be married some day to a rough carter or a grinning ploughman. Well having met you [handsome, high class Dorian Gray], and loved you, will teach her to despise her husband, and she will be wretched.”

Lord Henry wants the truth about how people are, not white lies. Today, he would probably find it interesting that despite polite opinion to the contrary finding your mate is ultimately a market transaction, that society is largely a genetic meritocracy, or that the great causes of the modern West like anti-racism, climate change, COVID, feminism or support for Ukraine are some combination of scam, mass hysteria, and self-destructive religious frenzy. But he wouldn’t do anything drastic, beyond making sure he is not personally affected: he knows how people are, and he knows that they are not changing any time soon.

“To become the spectator of one’s own life […] is to escape the suffering of life”.

Dorian Gray, on the other hand, is like the guy who reads about “the Jews” on 4chan and goes on to shoot up a synagogue. When he hears Lord Henry talk about how love is fleeting and superficial, he leaves the actress he just proposed to, breaking her heart and driving her to kill herself. His cruel and self-destructive lifestyle is what causes his portrait (but not his real face) to turn ugly, the main theme of the novel. This is the polar opposite of the chill, cynical stoicism of Lord Henry, and should remind everyone of taking red pills too seriously.

A minor thing I only now noticed about this book is just how gay it is. Nominally, Lord Henry is married and Dorian Gray dates only women but the story is full of descriptions of male beauty and one can’t help feeling like some of the characters are more than friends. Of course, Oscar Wilde was not just homosexual, but a pederast and even went to prison for it.

Elizabeth Marshall Thomas: Reindeer Moon

Thomas is an anthropologist who lived with African hunter-gatherers for years in her 20s. She is, however, also a 20^th century American woman, and both of these things are apparent in her book about Pleistocene hunter-gatherers in Siberia.

The story itself is silly. A caveman girl becomes a strong independent womyn after her parents die on a journey and she has to spend a winter in the woods with her sister. She then becomes the town (cave?) bicycle, has sex with every caveman she meets, breaks up her engagement and her sister’s because of her feelings, and everybody else is at fault for this. Fortunately, between a bear hunt and babies starving to death in the winter the cavemen are dedicated feminists, beating your wife is scandalous and women are free to divorce and even leave the tribe, so they put up with her behavior.

The world-building in Reindeer Moon is, however, seriously good. Thomas is great at describing how it is to live in a world where humans are wildlife, there is absolutely no civilization and life is largely about getting the minimum amount of calories not to starve. Fortunately, this takes up much more space than the drama. I read a lot of non-fiction about hunter-gatherers but never fully realized how it is to live like this (which is the majority of human history and likely the source of many of our behavioral adaptations).

One reason fiction is important is because it simulates immersion in personal experiences we can otherwise not have. Personal experience is not the same as scientific knowledge but it can give motivation or perspective for it. For example, I became interested in hereditarianism because when I lived in a bad area of Budapest, I noticed that the underclass people on public transport didn’t just lack money, they were different in other ways. They were often short and fragile and almost always ugly. They were restless like children, they fidgeted, vocalized and moved about while waiting for the tram, often chewing something or smoking. They were almost always sad and angry, every interaction even between friends and family members sounded like a fight. They were often injured, wearing bandages and casts or having scars. Coming from an upper middle class bubble you can believe that poverty is the consequence of generational trauma or capitalist exploitation, but with this experience I wasn’t surprised to learn that intelligence, health and wealth are genetically correlated and rare genetic variants (mutations) have a large effect on all. Some people get a bad draw of the genetic lottery so they end up sick, mentally ill, ugly and poor. A fiction author with a good eye for people can simulate such an experience like mine – or Thomas’s – for you in a novel [i].

Lev Tolstoy: War and Peace

In Hungary, you have to read these classic novels in high school and there are tests about book content and the author’s biography. Back in the day I read many long books other kids found boring but I found Tolstoy insufferable, so I did like everyone else and bought the abbreviated version (“Kötelezők Röviden”). After so many years I decided to give War and Peace a try as an adult.

Tolstoy is a very modern artist: his book is torture but you’re supposed to fawn over it. I really wanted to read this book but couldn’t. War and Peace is in desperate need for an editor, the text is overwritten, rambling, never getting anywhere. About 100 pages in, I gave up when a letter written by some countess to another was reproduced verbatim, multiple pages of meaningless teenage girl rambling including pointing out when she reached the end of a page writing. I wanted to read this book so bad that even then I skipped about 50 pages to where the Napoleonic war was already raging, but gave up again after a few dozen more because despite some good battle scenes and discussions of war strategy the text was just not going anywhere.

Not all books are good just because they are classics.

George R. Stewart: Earth Abides

This is easily the best novel I read now and one of the best in my entire life. Stewart was an English professor writing this in his 50s (published in 1949) so despite being sci-fi it is a serious, realistic, drama-free story written in crisp, precise English.

Isherwood Williams (Ish) is a PhD student in post-WW2 California when a virus wipes out ~99.9999% of humanity. This number is important: he is not left completely alone, but only a few dozen people survive even in a place like the San Francisco Bay Area. The survivors are too few for any of society to remain: there is no government, but there are also no bandit gangs, mutants, zombies or any other post-apocalyptic tropes. The story is about Ish’s life until his death as an old man. The survivors live a pretty comfortable life (by post-apocalyptic standards), mostly scavenging the remains of the old world. The power stays on for months and the water runs for decades. They keep cars serviceable for 20-30 years, and the roads and bridges largely last until the end of Ish’s life. They eat canned goods from grocery stores and hunt with rifles and ammo they loot from department stores. They have many children who mostly survive – there is no medical care beyond looting pharmacies but also not much infectious disease as all human carriers died.

There are many great themes in the book which are too realistic and un-dramatic to be greenlit for a modern Hollywood show (there is a 2024 adaptation but I haven’t seen it):

- Malthusianism. When the novel was written, virtually all countries on Earth were growing and it was common to worry that this will end civilization. (Ehrlich’s The Population Bomb was published 20 years later but the less-known Road to Survival by Vogt came out one year before.) Stewart hints that the disaster in the book is the natural consequence of having so many people on Earth, and population explosions and collapses happen to many wild animals after competition from humans disappears.

- Smart fractions and network theory. The 10 or so survivors are all urban people and (except for Ish, an academic) not very bright. They happen to have a carpenter who can maintain houses, but they have no farmer, no mechanic, nobody who can fix power plants or aquifers. Knowledge in a complex society is specialized, if you lose the few dozen people capable of maintaining a system, you lose that system. Learning these skills from books is not easy and may not even be feasible (see the next point). Intellectual activity requires a network of specialists. Ish is the only survivor interested in books or abstract knowledge, except for one son. When he dies young, science practically dies with him.

- People only make as much effort as they have to. This is probably the most surprising idea Stewart has which sounds right. Life is comfortable after the apocalypse, there are cars, free products from stores, for a while even power. Ish tries to convince his fellow survivors to learn to set up a gas-powered freezer, grow plants, prepare to fix failing infrastructure, build a government, maintain the city library, but they see no immediate need so they don’t do it. Civilization fails one small step at a time and neither is a wake-up call. When there is no more water, they dig wells and latrines. The next generation has never seen a modern city, only ruins, so they are happy to live as scavengers. The next one after this are hunter-gatherers as canned foods and ammunition perish. Ish’s greatest civilizational contribution is teaching kids how to make bows and arrows: this is technology they can use and improve. This sounds like third world behavior but remember: we have less than a dozen people who saw civilization, none of them are experts at building it, and there is no government organizing projects. Without a sufficiently large network of organized experts, civilization dies.

- Pro-natalism and eugenics. The one kind of rebuilding everybody is interested in is repopulation. Without much planning, everybody takes one of the surviving women as a wife (one man has two) and they have a baby each year. However, nobody is allowed to touch one intellectually deficient girl as they fear her condition is hereditary. (When a stranger tries this, they kill him.) Ish worries a lot that because only average or below-average people survived, there will be biological limits to rebuilding civilization.

- Workable Rousseauism. Violence is almost completely absent from Stewart’s post-apocalyptic world. This is because there is not much to fight about: goods are free and abundant, and access to women is tightly regulated: everybody gets a wife and that’s it. Plus, at least the original survivors happen to be reasonably nice people and not very young, and there is not much contact with outsiders. All of this is probably needed to eliminate violence at least for a generation.

Overall a very thoughtful and enjoyable book.

Cormac McCarthy: Blood Meridian

I always wanted to read this book because of its reputation as one of the great American novels. It’s a very gritty Western. I wasn’t very impressed by it, it feels long and repetitive. It circles through the following elements:

- Descriptions of desert scenes and astronomical/weather phenomena (these are great)

- Pointless, cartoonish violence

- Some semblance of a not very interesting story about a military expedition to Mexico.

McCarthy apparently researched the story well, so he uses vocabulary and includes details which are genuine for the mid-19^th century American West. This is fun but not easy to read. He makes an effort to make the story as hard to follow as possible, key events are only hinted at or hidden between descriptions and violence which makes the novel dreamlike. The novel is best known for its violence which McCarthy describes in a distant, dispassionate way. It stops being shocking very fast and becomes ridiculous. Any time something weak or innocent appears in the story you can bet it will be gruesomely slaughtered in just a few pages. I chuckled at the end where they shot a dancing bear to death for fun because as soon as there was an animal on scene not actively killing people I knew what was going to happen.

The most interesting thing about this book is its depiction of the American culture of violence. Many of the characters (based on real people) are cold-blooded killers but otherwise very smart and calculating, even intelligent and well-mannered. This is strange for me as a European reader, violence here is strictly the domain of brutes and governments and there is no pop culture glorifying it (other than the one we import from the US). The takeaway is not that the US is a barbarous violent country, it’s that it’s so great that even some its psychos are gentlemen.

Frigyes Karinthy: A Journey Round My Skull (Utazás a Koponyám Körül)

Frigyes Karinthy is my favorite Hungarian author, a real Budapestian working mostly during the interwar period. You may know his idea of six degrees of separation, the realization that all people on Earth are connected through just a few others. For example, I’m connected to Stalin through just two handshakes (through my grandmother who once shook the hand of Communist dictator Rákosi in the 1950s) and to Muammar Gaddafi through just one (through an old uncle who worked as a contractor in Libya in the 1980s). My other favorite work by him is The Thermometer (page 65), a great illustration of measurement invariance – you can cheat the indicator, but not the construct, by breathing on the thermometer in the cold.

Karinthy had a very interesting inner world. In this book, detailing his 1930s brain surgery in Sweden due to a benign tumor he writes (about a long distance call from Hungary):

“The telephone wire runs, and within the wires the words, to and fro, within a fraction of the second, but what does the wire care about human voice? The poles don’t care either, they are made of wood, mutilated, cut-down invalids of the global society of trees. They only care about other trees, maimed and moaning as they are; if they ask anything, it’s if the telephone wire has a message for them. Are you coming from far away, Brother Wire? – the Stockholm telephone pole asks. What’s up in the South? Is it true that the cherries are already blossoming, even though the trees here only will next month, and me, never?” (Translation mine.)

Treating inanimate objects as having thoughts and feelings like this is called animism and it is common in kindergarten-age children. One of the reasons I like Karinthy is because I also have these animistic daydreams when my mind is wondering. Sometimes I catch myself thinking how buses feel about having to carry that many people or if small buildings envy big ones. Are some adults more prone to this childlike thinking than others? Is this related to intelligence, creativity or humor? This is what I mean when I say that fiction can be hypothesis-generating.

Jorge Luis Borges: The Aleph and Other Stories

I always wanted to like Borges more than I eventually could. I was especially impressed by his 1940 Tlön, Uqbar, Orbis Tertius (not in this collection) a story ahead of its time about some people tracking down facts about the mysterious, hidden lands of Tlön and Uqbar only available in a certain edition of a certain encyclopedia. It reads like an SCP Wiki story. It must have been very cool to live in a time when you could discover serious new knowledge in a library of rare books instead of immediately getting all the knowledge about everything via Wikipedia and ChatGPT [i].

That said, most of Borges’ stories circle this style and content but never really get there. You will recognize early versions of lots of well-known sci-fi and horror tropes though. Sometimes in fiction it takes generations of authors to really nail down some idea and while the latest are the best, you must give credit to the original inventor of the idea, which Borges often is. I always felt this way about Shakespeare too, he is not that great before you realize how many people he inspired and that his ideas are getting copied and reshuffled by Hollywood to this day.

Overall impression

Reading fiction is great and you should try it again if you ever stopped. It’s not an alternative to science, it’s entertainment, not work or learning, but it beats movies and games (not to mention short-form brainrot which should be treated like a drug). There is something psychologically right about imagining a story you read with your mind’s eye, the immersion in fiction creates perspective and there are lots of ideas in the works of good authors that are either valid on their own or even inform scientific work.

Many modern social scientists get all their ideas about people from R-based analyses of huge datasets and online conversations with like-minded people. Many also live in not just first-world, but upper-middle-class first-world bubbles and have little personal contact with raw data or the experiences of people unlike them. This carries the risk of becoming what we call a “barbarian professional” (szakbarbár): a person highly skilled in a narrow discipline but missing perspective, somebody with a good hammer who sees every problem as a nail. I understand the benefits of quantitative, data-driven science and I’d pick that one too if I had only one way to learn about humans. But being good at statistical genetics or psychometrics or anything else doesn’t let you know everything about the human condition because all these disciplines have limitations, and reading fiction is one way to fill in some of the gaps.

[i] Of course, fiction can create a false narrative too, like the anti-psychiatry literature which made people lobby for the closure of asylums and landed thousands of deeply ill people unfit for society on the streets, or Hollywood fantasies about ass-kicking tough women and dindu nuffin gentle giants.

[i] Sometimes this still happens though: a few years ago a mention of Markland, the Viking name given to Labrador, was discovered in a 14^th century Genoese codex. This shows that knowledge of the Vikings’ discovery of America survived almost to the time of Colombus. Incidentally, Borges also wrote about how the Scandinavians invented much of the modern world, including discovering America, before everybody else.

One more thing to keep in mind is that sibling/MZ controls can falsify the minority stress hypothesis, but cannot prove it. If the mental health disparity is gone within sibling/MZ pairs, then there is no more discussion, minority stress is just false. If the disparity remains, minority stress is one possible explanation for it: gay people really are worse off mentally, it’s not in the genes/family environment, it may be because you are discriminated against, but also e.g. because you can’t have children with someone you love or participate in a toxic subculture (distinct lifeways), or because you internalize the progressive image of gays as perpetual fragile victims (narrative possibilities). I wrote about these recently.

I finally did a formal literature search of all genetically informed studies investigating the link between mental health and sexual orientation. Below are the new ones I found – I think this is all. I always tried to convert the statistics in the papers to Cohen’s d to make them roughly comparable.

Rothblum & Factor 2001

This is the earliest study I found and claims to be the first of its kind. It compares 184 lesbians with their heterosexual sisters. Lesbians were mode educated, much less likely to be married and more likely to live with a female partner (duh) or with children, more likely to live in a large city farther from their parents (distinct lifeways!), heavier and more likely to have attended therapy, but actually no worse off mentally based on the Brief Symptom Inventory and with significantly higher scores on a self-esteem scale.

There is a lot of interesting data in this paper (note the extremely low p-values the asterisks mean, these are credible findings!):

But the problem is that no SDs are reported, only means and if the results were significant. This means we could not convert these findings for use in a meta-analysis, but the results do not support the minority stress hypothesis.

Balsam et al 2005

This paper uses an analytical strategy which is common in other fields but very unusual here: a multilevel model where family membership was indicated with a random intercept, estimating a between- and within-family effect within the same model.

If I understand the (quite opaquely written) results well, there was no within-family difference in distress, self-esteem, and life satisfaction, but non-heterosexuals went to therapists and had more suicidal ideations/actions than their siblings. This is the typical finding in studies like this (much of the mental health disparity goes away within families) but the reported statistics are not enough to quantify how much of this effect there is and the findings cannot be integrated in a meta-analysis. Because simply comparing first-order relatives leaves 50% of the genetic variance free, some but not complete attenuation is expected even if genetic common causes are the true explanation.

Another problem with this study is that they don’t report non-significant findings in numerical detail. Don’t do this, every finding is important.

Rosario et al 2014a, Rosario et al 2014b

I completely missed these two highly relevant papers from the Growing Up Together and Nurses’ Health Study. The dataset is the same in both papers, but the survey wave is different so I think it’s somewhat fair to treat them as two separate studies. Depressive distress is reported by the participants as a function of sexual orientation.

The data reported here lets me run quite a few stats. I decided to calculate population (ignoring who belongs to which family) and within-family Cohen d differences, pooling numbers across all types of sexual minorities. One paper reports the within-family Cohen d directly, another reports an odds ratio which can be converted to it.

These papers do not show meaningfully smaller sibling effects than what we see in the population, so they are technically in line with the minority stress hypothesis.

Post & Veling 2021

This study is about rates of non-heterosexuality in a cohort of psychotic patients, siblings, and healthy controls. In most studies, sexual orientation is the independent and mental illness the dependent variable – this is reversed here, and the study investigates the rate of non-heterosexuality (compared to healthy controls) in psychotic patients and siblings. Elevated rates of non-heterosexuality among healthy siblings of psychotic patients support a familial link between the two phenotypes.

The odds ratio non-heterosexuality in patients (population effect) and siblings (within-family effect) compared to healthy controls was calculated based on rates reported in Table 1.

We only care about the raw counts now. Of the 582 psychotic patients, 40 are non-heterosexual (“LBG”, 6.8%), but out of the 361 controls, only 10 (2.7%). Among healthy siblings of psychiatric patients, non-heterosexuality rate is somewhere in-between at 26 out of 604 siblings (4.3%). This gives us a population odds ratio of 2.6 and a within-family odds ratio of 1.6. By getting rid of 50% of the genetic variance, a big chunk of the mental health disparity went away, in line with genetic confounding. Again, with just siblings, there is still 50% free genetic variance so we wouldn’t expect complete attenuation. This study at least partially falsifies the minority stress hypothesis, and shows a familial link.

Xu et al 2023

This a very ambitious paper which still made me scratch my head a little. It is the largest in the literature: it uses Swedish population registries to compare mental health (medical records for depression, substance abuse, or suicide) as a function of sexual orientation which is measured by entering a same-sex marriage or an opposite-sex marriage. It’s actually two papers rolled up in one with this methodology.

The first analysis is the more straightforward one. It is a biometric analysis using full siblings and adoptees, which works just like a twin study. Is suicide more common in biological than adoptive sibling pairs? Then it is heritable. Is your risk of suicide higher if your biological sibling is in a same-sex marriage, compared to an adoptive sibling? Then the two are genetically correlated.

The path coefficients are on Figure 2:

It drove me crazy how and why they fitted this model which include both a genetic correlation (A1-A2 curved arrow) and a causal path from one trait to the other. Then I realized the causal path (the straight one going right to A1/E1) is just for illustration. In Cholesky decomposition (a method for estimating genetic correlations), which is what the straight arrows show, you can recover the genetic correlation from heritabilities (a11, a22) and cross-trait genetic paths (a12) with this formula:

The formula recovers the genetic correlations perfectly, so the models were not fitted as depicted, this is just an illustration and the Cholesky path can be ignored. From here, we can use the method to convert these path coefficients to basically estimate how much the MZ and population correlations between same-sex marriage and the three mental health problems would be (even though we don’t have any twins):

MZ=0, POP=0.223 for depression

MZ=0.154, POP=0.254 for substance abuse

MZ=0.08, POP=0.27 for suicide.

By getting rid of the genetic path, most but not all (except for depression, note the zero E correlation) of the correlation goes away. This mostly, but not completely, falsifies the minority stress hypothesis.

The other study reported in this paper is trickier. It is a straight sibling control analysis, and it shows a much smaller drop between population and within-family estimates:

But I have many problems with this.

First, we have the problem with the definition of sexual orientation. Presumably everybody in a same-sex marriage is gay, but there might be lots of unmarried gay people (especially older ones who couldn’t marry yet) or even gay people in an opposite-sex marriage. This issue came up in commentaries to the paper and the authors claim it’s not a very big problem, but I’m totally convinced. For example, what if the gays who get to marry are systematically the psychologically healthier ones?

Second, the paper reports hazard ratios (p/(p+q)), not odds ratios (p/q) which cannot be converted to any other metric unless you know cell counts, which here you don’t. My Cohen’s d conversions are based on pretending these are odds ratios. HRs and ORs are approximately equal if the rate of hazard is constant and the occurring events are relatively rare. The first assumption cannot be tested, but the second is questionable in the current paper with depression rates >13%, substance abuse rates >4% and suicidality rates >2.3%. There are actually ORs and HRs reported in some cases and they are quite different:

Third, the authors prefer an age-corrected model and don’t report enough data for others. Because at least depression and suicide are more common in older people and gay marriage is presumably less common, I’m worried this might be an issue.

This study is huge (bigger than all the others combined) but I’m worried how much its results are affected by these issues, especially because the biometric method shows something else and I’m not sure how to weigh it about the rest of the literature.

Either way, these studies almost all show that the non-heterosexual vs heterosexual mental health gap shrinks substantially if we take genetic (OK, “familial”) out of the picture. An overlap between the causes – probably genetic causes – of poor mental health and non-heterosexuality must be a big reason for their correlation and minority stress at the very most a minor one.

What happened to Genny Savastano with the cartels is what a behavior geneticist would call a large, generalist nonshared environmental effect: some event outside your family that drastically changes multiple aspects of who you are. Your whole personality, your mental health, maybe even your looks changing after some watershed event is a common feature in fiction. Does this happen in real life? This is what this post is about.

(See also the previous one about the shared environment).

Genny Savastano before and after a large non-shared environmental effect

The search for the nonshared environment

In a 1981 paper, Robert Plomin and David Rowe coined the term “nonshared environment”. “Nonshared”, as in not shared by siblings, this referred to everything which makes people different, but isn’t genes and isn’t the environment we experience together with our siblings, such as family socioeconomic status in childhood or the systematic rearing strategies of parents. Everything that makes MZ twins different is nonshared environment.

Much of modern Western society operates on the premise of large, generalist nonshared environmental effects being real. When a politician wants to just give poor people money or their kids compensatory education so that they turn out as more functional or when a therapist wants to cure somebody’s personality disorder, this is all a gamble on something like whatever happened to Genny Savastano. You get pushed a little and you move a lot.

Of course you read East Hunter so you know how the correlations you see in social science are genetically confounded. Kids who went to elite schools or those whose parents smoked while being pregnant with them are different from others in important biological ways: they inherited the genes which made their parents willing and able to afford good schools or stupid enough not to care about the health of their offspring. Genetic confounding is a big problem for the shared environment – the apparent effects of parenting, growing up in poverty, childhood trauma etc. are most just due to genes clustering in the same family. However, the non-shared environment is a different story, at least at first sight. In a typical twin study heritability is about 50%, with the rest being mostly nonshared environment. In other words, at first sight, the nonshared environment is about as important as genes, except potentially easier to change.

The gloomy prospect

However, this impression is misleading. The problem is that what a behavior genetic study sees as nonshared environment are not necessarily Genny Savastano-style life-changing experiences. They are not necessarily experiences. Not necessarily even environments, actually. The following things all look like nonshared environmental effects under the statistical microscope of the behavioral geneticist:

- Measurement error, like researchers miscoding your responses

- Luck and accidents, like guessing an IQ test correctly or being sick when you sit it

- Random but biological processes like somatic mutations, epigenetic changes, developmental noise etc. I don’t mean being poisoned or working out, that would definitely be an “event”, these processes can be endogenous (emerge randomly from within your body).

- Birth order

- Differential treatment of siblings by parents, teachers, or each other

- Differential life experiences of siblings.

Arguably, only the last two are really “experiences” in the sense that we could design an intervention based on them. On the other hand, if the ~50% nonshared environment is the first four things, then tough luck, we have to accept that while not everything is genetics, we can’t do much about the rest.

About this latter idea Robert Plomin and Denise Daniels wrote in a classic 1987 paper on the nonshared environment that:

“one gloomy prospect is that the salient environment might be unsystematic, idiosyncratic, or serendipitous events such as accidents, illnesses, and other traumas, as biographies often attest.”

Since then, “gloomy prospect” became the standard term for the idea that the nonshared environment is mostly just intractable randomness and not identifiable life events we can use to build interventions. In a huge 2000 review on the nonshared environment, Eric Turkheimer and Mary Waldron concluded that the gloomy prospect is largely correct. This was based on very small correlations (r~0.01-0.02) between environmental measures of siblings once genetic relatedness is controlled for – we’ll soon see how such a study works, the first wave of them is described in this book about the NEAD project, a genetically informed study about the effect of family dynamics. Plomin & Daniels were originally not enthusiastic about this idea, and Robert Plomin in his great 2011 commentary still calls the gloomy prospect “pessimistic” and calls for more studies to figure out how the nonshared environment works. He seems to have changed his mind, however, by the time he recently wrote this review, where he says that “it is time to take the gloomy prospect seriously”.

From what I can tell, the gloomy prospect is now the mainstream view of the nonshared environment. Most researchers believe that the nonshared environment is the first four items on my list above: random, idiosyncratic or at least uncontrollable events that behave more as noise than anything else. However, I side with 2011 Plomin much more than 2024 Plomin and I think the gloomy prospect actually has quite a bit of light shining through.

Nonshared environments are (a little) generalist

Now it’s time to talk methods. How can we test the gloomy prospect hypothesis?

If nonshared environmental effects consist of random measurement error or accidents and luck, then they should be specific to time points and traits. Guessing the IQ test right (measurement error) doesn’t change your IQ on a second measurement, and winning the lottery (luck/accident) should make you richer, but likely not smarter. This covers roughly the first two items on my list above about what a nonshared environmental effect can be [1].

Testing if an effect is specific to traits and times can be done by calculating genetic and nonshared environmental correlations: the overlaps not in the traits themselves, but in the genetic and nonshared environmental effects that cause them. Two traits can be modestly heritable but their genetic correlation can be 1 if that little heritability is caused by the same genes. “Same genes” is easy to understand in a molecular genetic study: in a study like that, you know which specific genetic variants relate to trait 1 so you can check how much the same variants also relate to trait 2. In a pedigree study, which is most studies like this, this happens indirectly.

The core idea of this indirect method is that if two traits are genetically correlated, they correlate more if measured among more genetically similar people. If, say, height and IQ are genetically correlated, you can make a better guess of somebody’s height based on his twin’s IQ if this is an MZ and not a DZ twin. Conversely, if the correlation is nonshared environmental, genetic relatedness won’t matter, a bunch of MZ twins will correlate just as much as a sample of random pairs of people. This is also what happens if one trait causes the other. If I shoot you in the head you will be deader than your control regardless of whether this is a random person or an MZ twin because bullets to the head cause death. We are using the inverse of this logic in twin control studies trying to falsify such causal hypotheses[2].

If the gloomy prospect is real and all nonshared environmental effects are just random perturbations, nonshared environmental correlations should be zero. In this case, the nonshared environment is just random noise, its effect will not ripple through either many traits or longer periods of time. This is demonstrably false: nonshared environmental correlations are not zero.

In a famous paper about the generalist effect of genes, Kovas & Plomin write that:

Nonshared environmental correlations are low. For example, in the same two studies, the nonshared environmental correlation was .39 between reading and mathematics at 7 years (Kovas et al., 2005), and the average nonshared environmental correlation was .24 between five components of mathematics at 10 years (Kovas et al., 2007).

Yes, low, but not zero. Let’s whip out the figure from my shared environment post:

In this figure, I showed the distribution of phenotypic correlations (what you normally measure, no twins or any fancy stuff) and genetic, shared and nonshared environmental correlations (what you need twins for) of a sample of traits from three papers reporting many. Genes and shared environments are generalist, a genetic effect or the effect of the rearing family normally changes you not in one, but several ways. Nonshared environments are less like this. Nonshared environmental correlations are lower than trait correlations (purple histogram shifted to the left of the blue one), which in laymen’s terms means that stuff that happens to you outside of your rearing family makes you deviate from the blueprint your genes and upbringing created for you. Genes are generalist: if you have the IQ genes, you also have the height genes and the health genes. The shared environment’s effect, while weak, is non-zero and highly generalist. Your parents being rich and well-behaved may not be as much an influence on who you turn out to be as the typical sociologist imagines, but it does have an effect and this reverberates across many traits. On the other hand, the nonshared environment make you deviate from this track. Under a strong gloomy prospect hypothesis, all nonshared environmental influences would influence exactly one trait, making between-trait nonshared environmental correlations a hard 0. In other words, the nonshared environment would be measurement error or something indistinguishable from it.

But these numbers are not zero. The mean nonshared environmental correlation across all trait pairs is 0.11 across cognitive tasks, 0.1 between the combination of GCSE grades and psychosocial characteristics, and 0.15 between school tests. These are not random perturbations or measurement error, nonshared environmental effects do ripple across multiple traits (a little).

We see the same if we look across development. The nonshared environment is a major influence on, for example, intelligence at all ages. The strongest version of the gloomy prospect hypothesis says that this is because of random effects: measurement error, you being sick or distracted during one test but not the other and similar non-replicable effects affect your performance on any test, but by the next testing the same influences will not be there anymore.

The developmental twin study of Bishop et al (2003) shows almost this. In a sample of twins, siblings and adoptees repeatedly tested from 1 to 12 years of age. The rank order of kids was quite similar at all ages, but this was mostly due to genetic effects: the mean nonshared environmental correlation was 0.02. In the unique study of Lyons et al (2009), however, a different pattern was found. This is one of the best papers on the stability of intelligence, comparing intelligence at age 20 and age 55 in the Vietnam Era Twin study. IQ stability across these ages is 0.74 with a genetic correlation of 1, but the nonshared environmental correlation is also 0.19 – much lower, but not zero.

IQ is not the only trait where the effect of the nonshared environment is stable across the lifespan. Similar results were found with various methodologies for depression and anxiety (stable within-MZ pair differences across time), attention problems (non-zero nonshared environmental correlations), and personality (substantial environmental stability).

Whatever the nonshared environment is, it is not something for the moment. Although it does this much less than genes or the shared environment, the nonshared environment has effects which ripple across traits and time.

Zeroing in on the nonshared environment

These non-zero nonshared environmental correlations already falsify the first item (measurement error) and maybe also the second (luck and accidents) on my list of what the nonshared environment can be. Because of these correlations, the nonshared environment cannot be all random: if it was, its effect wouldn’t generalize across time and traits. However, there are other things which could show this pattern but still be gloomy (useless for interventions). For example, it is plausible that random biological noise can create lasting, multi-trait individual differences. Although nobody knows how it got there, you do have that somatic mutation, epigenetic change or weird developmental noise your MZ twin brother doesn’t, it changes multiple things about you, and it stays with you as you grow up – a stable, generalist nonshared environmental influence we still can’t do much with. To truly falsify the gloomy prospect, we need to find causal influences of specific life events on traits.

This will be harder than it sounds and ultimately we will fail to get there all the way, but we’ll try.

The best attempt at this is a study like Burt et al 2009. This study wanted to know if hanging out with dysfunctional peers as a teenager makes you more mentally ill in the “externalizing” (aggressive, neglecting duties, using drugs etc.) way. They took several steps to break it down to causal effects: 1) they compared MZ twin pairs to get rid of genetic confounding, 2) they measured both deviant peers and externalizing problems at age 14 and age 17 to get the direction of causality (effects don’t flow backwards in time), 3) deviant peers were rated by external interviewers, not only by the participants themselves (so we don’t have the problem of reporting biases, such as more problematic kids trying to look cool and saying their friends are all rule-breaking badasses). The model and the results look like this:

As you can see, the hypothesis turned out to be false. Kids who have more deviant peers at age 14 (Peer_14) don’t have more externalizing problems at age 17 (b21 path). But the opposite pattern holds: if you are a troublesome kid at 14, you’ll find like-minded peers to hang out with by the time you are 17. Because these are all MZ twin differences – “more deviant peers” or “more externalizing behaviors” are relative to a genetically identical person – none of this can be genetics. Causality is not 100% nailed down – more on that later – 1, but because of the nature of the variables and the careful controls I’d bet this is a causal effect.

In the table below I summarized a bunch of other similar studies:

Sometimes the best science is the least surprising. Asbury et al 2008 and Oliver et al 2008, for example, find that kids who rate school more positively than their MZ twins do better in class and they are better rated by their teachers as well. Some kids like school and do well and it’s not all genetic.

In Viding et al (2009) we are reminded of the limitation of this design. This study found that kids who were getting more parental discipline at age 7 than their MZ twins had more conduct problems at age 12. It’s not that parents became convinced by age 7 that one of their twins is more mischievous and held on to this belief: conduct problems were rated by teachers. On the other hand, two different things can still explain this data: 1) parental discipline causes poor behavior, 2) kids have pre-existing, non-genetic tendencies to bad behavior, which parents notice at age 7 and teachers at 12. The former would be the kind of causal non-shared environmental influence we are looking for. The latter would be in line with “biological noise”, some non-genetic force causing genuine, stable trait differences even within MZ pairs. Biological noise is gloomy because we can’t do any more about a somatic mutation or an epigenetic change than about the DNA sequence. In some cases, like Burt et al 2009, the cross-lagged model (similar traits measured twice) makes a strong case for a causal effect. In other cases, like the Viding study, who knows.

We can get rid of this biological noise and other pre-existing differences by getting even more specific than MZ twin pairs and studying the same person before and after potentially life-changing events, just like in the Genny Savastano example in my opener. I already reviewed one such study: in that one, using huge international samples we saw that in just about every country, men are happiest when they work more and their spouses work less, while women are happiest when they work less and their spouses more. Feminism is easily the most harmful and empirically false political idea which makes both men and women unhappy, implodes birth rates and doesn’t even extract a net positive fiscal contribution from women to grow the economy. In the context of this post, these findings mean that work duration is a nonshared environmental effect on life satisfaction.

Recently, Haehner et al used a sample of seven panel studies with about 200k people to check if personality changes after critical life events. There were multiple personality assessments of each person: the question was if traits tend to change after something important (like having a child or getting divorced) happens. For example, here are their results about “loss-related” life events:

Some of these results make sense. People are miserable before a divorce (-0.2-0.3 SD in life satisfaction) but get significantly better by 2 years after. Emotional stability is worse before widowhood: it is stressful to care for a dying spouse, and you are relieved when the suffering is over. Other findings are strange: after a separation, life satisfaction tanks, completely the opposite to a divorce despite the similar event. Widows are the most satisfied in the year of their spouse’s death, and much worse both before and after. None of this is Genny Savastano territory, but the results are certainly not 0, big life events change personality.

Ruggeri et al 2023 is a meta-analysis of similar studies, with somewhat fewer participants than Haehner et al 2025. It reports similar findings of small but believable effects: for example, entering a new relationship increases conscientiousness and life satisfaction, marriage reduces openness, while childbirth reduces extraversion.

Poly-environmental scores

You probably noticed that the nonshared environmental effects I’m beating the drum about are rather small. For example, in Haehner et al personality changes after critical life events were on average 0.05 SD, so maybe the gloomy prospect is still real if even critical life events have such a small influence.

The problem with this reasoning is that the nonshared environment is not a single event. Single genetic variants also have a small effect, but it would be a fallacy to say that because of this heritability is low: high heritability arises from many small effects.

In his 2011 commentary, Robert Plomin suggested using “poly-environmental scores”: studying the effect of not a single event, but multiple events simultaneously. Much like for genetic effects, there is a “missing nonshared environmentality” gap between the small effects that can be assigned to single specific events and the large effect suggested by twin studies, and we have to come up with ways to fill it. This is not a hopeless endeavor. For example, in Turkheimer & Waldron 2000, the authors concluded that the gloomy prospect is mostly true, but even they produced a poly-environmental measure that accounted for 13% of MZ variance (Table 6).

On the other hand, Gidziela et al (2022) from the Plomin lab tried to do something similar in the TEDS twin study. It didn’t go very well: specific nonshared environmental influences accounted from somewhere between 0% and 5% of various behavioral problems (depending on what model they ran). Of course the performance of a poly-environmental score depends on how well we sample the causal environments. In genetics, there is the finite DNA sequence we can sample, but here it is possible that we just didn’t measure the environmental influences that matter.

Not all is gloomy

The nonshared environment looks very important in twin studies. Some of it is just measurement error, and goes away with better measures of traits and zygosity. Some of it is somewhere between error and accidents with transient effects: nonshared environmental correlations between traits and times are low. Some of it is biological noise, and not much more malleable than a genetic effect. Also, it is hard to detect the truly causal effect of life events, because we need special data (longitudinal, multi-rater panels) and when we do this, the effects we find are small and sporadic. The gloomy prospect remains true in the sense that we can hardly base interventions on these effects. Too many interventions have been tried and failed and within-MZ studies didn’t give us any ideas about new ones. Just like genetic effects, nonshared environmental effects consist of a great number of tiny influences so just like we cannot chemically mimic the effect of the effect of a few large-effect SNPs to boost intelligence, we can also not just copy the environment that makes people smart or emotionally stable.

But we still have multiple studies that show a genuine effect of life experiences on traits, maybe even persistent effects on multiple traits. We are genuinely changed by the things that happen to us, even if these effects are too chaotic to be deliberately controlled.

Sometimes, however, these small effects can become correlated and create a long vector of change. This must be what happens in between-era, between-country environments which as I previously wrote are environmental effects which may trump even genetics. When you live in a certain era or a certain country, all of these small environmental effects have a shared component.

[1] If you want to sperg out about causality in longitudinal studies which don’t have other frills, your paper is Rohrer & Murayama 2023. Simply put, we don’t know if it is externalizing problems that caused seeking out bad peers, or some cause or consequence of it. For example, it is possible that drug use causes externalizing problems and drug using kids meet up eventually to smash bus stops: it’s the cause of externalizing problems, not the problems themselves, that makes you seek out a bad peer group. (This is actually not a very good example – why do we see a clear temporal order if EXT and DBI have a common cause? Not being able to think of an alternative hypothesis is why I’m prone to believe that being dysfunctional really is what makes you seek out other troublemakers.) These limitations ultimately make it difficult to pin down any of these causal nonshared environmental effects, they can always be confounded by something else. But we can be sure of some things: 1) the confounder must be something the design allows to vary (so it cannot e.g. be genetic in this case), 2) we can establish the direction of causality, even if it works with confounding: peer groups cannot cause problems directly or indirectly, because there is no lagged path that way, 3) we can falsify causal hypotheses because if there is no effect, we have nothing to talk about.

[1] Although the evidence is mostly to the contrary, luck and accidents can theoretically be generalist, winning the lottery or losing a leg can make you e.g. depressed, even in the long run.

[2] We are using the inverse of this logic in twin control studies trying to falsify causal hypotheses, looking at not the correlation (similarity) but the difference of pairs of people from which only one is exposed to something. If exposure causes a change – for example, if smoking weed causes schizophrenia, or see my bullet example – the difference of the exposed and unexposed shouldn’t diminish just because they are genetically related, but this is just what happens if there is no causation but a genetic confounding.

Gusev is a statistical geneticist so the Sun Tzu move from him in a debate like this is to make everything about the technical details of the models used in specific papers, pointing out flaws and limitations in the ones he doesn’t like and pulling out new ones he does like. For example, in this paper MZ twins reared apart correlate 0.78 for general cognitive ability, MZs reared together 0.8, DZs reared apart 0.32, and DZs reared together 0.22. My reading of this is that growing up in the same family doesn’t make either MZs or DZs more similar and MZs are over twice as similar, so intelligence looks quite heritable. This is also what the authors conclude with formal modeling (Table 3, heritability=81%). But Gusev says the real story is that DZAs are more similar than DZTs (the difference is actually not statistically significant) and an analysis explicitly done to show that assuming non-additive genetic effects doesn’t add much to the model (Table A1) produces nonsense estimates so the study is “a complete mess”. On the other hand, he didn’t engage very much with the IMO very plausible argument that sib-regression and RDR miss some kind of important genetic variation, and the fact that the lack of similarity in adoptees makes strong shared environmental effects rather unlikely and this is hard to explain by some issue of the design.

Normally it’s great when people argue with numbers instead of calling each others racists and Marxist wokes but in this case it’s starting to look like arguing about social psychology or brain imaging or candidate genes with numbers: technically, there are numbers in molecular genetics but there is something wrong with them. For example, this recent paper found a 74.6% heritability of cognitive performance using sib-regression, one of Gusev’s preferred methods. Some hereditarians celebrated this but I’m not really buying it. The study used six cohorts, but the results didn’t really replicate across them (Table E.2), all we have is a messy average. For educational attainment, heritability was 0.08, low even by the standards of this method. Here are some sib-regression and RDR findings from this paper and the original RDR paper by Young et al 2018:

For the multi-cohort Markel et al 2025 paper I show both the population average and the range across cohorts. The SEs should give you a sense of how different these estimates are - random error can produce about 2 SE difference between estimates.

These molecular methods don’t just fail to replicate pedigree studies: they fail to replicate each other too, other than the fact that they usually find lower heritability than pedigree methods. One reason I’m reluctant to believe them over pedigree studies is that the latter – MZAs, twins, extended pedigrees, adoptees – tended to replicate now-classic findings like h²~0.8, c²~0 for IQ quite well.

An interesting thing I realized in my exchange with Gusev is that he treats within-family molecular genetic heritability estimates (RDR, sib-regression) as exactly analogous to quantitative studies. In a molecular design like sib-regression we exploit very small fluctuations in the actual genetic relatedness of people with the same degree of legal and pedigree-based relatedness. On average, siblings are 50% genetically identical, but because of random processes, some of them end up slightly more or less related genetically. We can not only check if those who are slightly more genetically related end up slightly more phenotypically similar (they do), but we can quantify the importance of, say, 1% of extra genetic relatedness. If we extrapolate this, we can estimate how much extra similarity going from 0% to 100% would cause: the effect of all genes, i.e. heritability.

The logic of sib-regression

There are plenty of cases in the hard sciences where these extrapolations work just fine. If you measure how much energy it would take to heat a gas 1 by degree you have a pretty good idea how much it would take to heat it 1000 degrees. I think Gusev has a hard science background so I see how he sees this as just another way to do a quantitative genetic study. But when we get a smaller slope in a design like this than expected from a twin study, can we really discard real-life examples about the similarity of MZAs and the dissimilarity of adoptees, along with all kind of other pedigree studies, and conclude that the substantial similarity of family members is due to the shared environment after all?

I made this chart based on Bingley et al, see also a similar one here. It shows the phenotypic similarity in education (asterisks) of various family members, together with what would be expected from a purely genetic model for simplicity (red lines, expected correlation=male MZ correlation * genetic relatedness). For education, there are clear shared environmental effects, maybe from within-generation effects – siblings and cousins are more similar than expected, although parent-child and avuncular pairs are not – but the effect of genes shines through. Would you have thought that people are more similar to their uncle or his kids if he is an MZ, not a DZ twin of their father?

In this post, I’ll reflect on some other ideas commenters raised and I’ll also present the ideas I think are the best arguments against hardcore hereditarian claims.

Gene-environment correlations

One argument that keeps coming up against hereditarianism is gene-environment correlations: the idea that genes merely index the environment truly causing our traits. I wrote about the passive, reactive, and active forms in my previous piece. For us, the active form is the most interesting: the idea that genes don’t directly code your intelligence or your personality, only what behaviors you are likely to engage in, and your gene-typical behaviors will shape your traits over time. If your behaviors resulted in different experiences, or if these experiences were artificially provided to people who don’t normally seek them out, the effect of genes would be different. Active gene-environment correlations are supposed to explain the Wilson effect and genetic amplification over development. The importance of genes grows as you get older (Wilson effect) and it’s not because new genes enter the game, it’s because the effect of the same genes becomes stronger (amplification) – maybe because the experiences the genes expose you to indirectly accumulates over time.

One way to think about this is that

"it matters not one whit whether the effects of the genes are mediated through the external environment or directly through, say, the ribosomes".

Some genes make us synthesize different proteins, some make us behave differently, both make us different, the distinction is academic, we are seeing gene effects on phenotypes. I never liked this argument very much because the biological and behavioral pathways imply a very different landscape for the potential for interventions. If, for example, IQ genes cause true biological changes then maybe we can design nootropics to make people smarter, but not much else. If, however, some people just inherit the tendency to engage in behaviors that make them smart, then we can phenocopy them, we can make everybody do whatever some genetically predisposed people do naturally, and make everybody smarter.

In his book “Making sense of heritability”, Croatian philosopher Neven Sesardic makes an argument for the biological pathway I rarely see. He describes just how the causal paths between genes and environments would run assuming biological (direct genetic) and behavioral (gene-environment correlation) pathways. Using a typical MZ IQ correlation of 0.75, the numbers could go something like this:

In both cases, we are assuming that genetic variants first cause an endophenotype (something like “bigger brain” or “better axon myelination” or “more efficient serotonin synthesis”, it doesn’t have to be one specific thing). The endophenotype, in turn, either causes IQ directly (biological pathway), or it first makes you engage in a behavior or seek out an environment, which in turn causes IQ (behavioral pathway). In the latter case, we are dealing with active gene-environment correlation: your genes don’t immediately make you smart, but they make you behave in a way that ultimately does. The argument also works for evocative correlations, except your genes don’t make you do something, they invite other people to do the same.

You can immediately see that because there are more steps between the genes and IQ in the behavioral pathway, we need higher effects to still end up with the same IQ correlation at the end of the paths. In both cases, we generously but realistically assume 100% penetrance (a perfect correlation between genotype and endophenotype). In the biological effects case, we have r=0.86 between the endophenotype and IQ (75% heritability), which is high but not crazy, genes can sometimes reliably code phenotypes or evolution would be impossible. However, if we assume gene-environment correlations, not only would the endophenotype need to have a very strong effect (r=0.93) on IQ-increasing behavior, but this IQ-increasing behavior should be incredibly effective at raising IQ!

Just what are some people with the right genetic predisposition doing that makes them smart? Whatever it is, even if we assume a perfect correlation between endophenotype and IQ-increasing behavior, it would need to have an r=0.86 effect on IQ. It cannot be better schooling: curriculum reforms don’t even really improve subject-specific grades in the short term. It cannot be more schooling: going to school longer has at most a minor, non-g-loaded influence on IQ. It cannot be puzzles, brain teasers, being mentally active: these just don’t work after publication bias and placebo effects are accounted for. It must be accessible even outside of a rich first world environment because high IQ heritability was found even in Nigerian and Indian samples (although not a Sudanese one and this paper argues using Russian data that heritability was quite low during the chaotic, highly unequal 90s). If a high-quality environment would be needed to max out your genetic potential, heritability would be universally low in the third world because even people with a high genetic predisposition for intelligence could not engage in the behaviors that actually raise intelligence.

For these reasons, whatever behavior the people genetically so predisposed seek out to make them smart must be something 1) quite different from what families and schools do to make you smart, 2) different from brain teasers or cognitive training because these don’t work, 3) available even in poor countries.

I’m not writing this to totally discount active gene-environment correlations. Maybe it’s many small, frequent, self-paced behaviors that make you smart, and you can’t emulate these by forcing high-intensity interventions on people for a few months. I remember being 6-7 years old in our small town commieblock apartment in early 1990s Hungary. We had an old i386 computer and CDs with demo games I would play with a lot, although we had no money for full games or an audio card. I borrowed and kept an English dictionary by the computer and whenever I encountered a new English word, I would pause the game and look it up. (I had already taught myself how to read in kindergarten.) I would steal literature from my grandparents and when I had a little money, I bought myself a geographical atlas which I spent hours browsing, fantasizing about places like Nauru and Monte Cinto and Monte Rotondo. I clearly had a deeply ingrained, strong natural tendency to teach myself things and I like to believe that this contributed to who I am today, not just biological processes coded by my genes. If I’m correct, it would still be hard to narrow down the specifics of my IQ-increasing behavior because it was so many small things. Even if you could somehow faithfully design the East Hunter Kid Smartening Intervention, most 6–7-year-olds wouldn’t like doing these things as much as I did so they would skimp out, cut corners and do everything half-heartedly and it likely still wouldn’t work. If genes cause trait-changing behaviors this fine, then maybe it is better to treat their effect as just gene expression because we can’t emulate them any easier than a thousand minuscule changes in protein synthesis.

Gene-environment interactions

Another form of gene-environment interplay is a gene-environment interaction (G*E). This means that depending on the environment, the same genes can result in different phenotypes.

In the same book I mentioned above, Neven Sesardic introduces the concept of “common sense” G*E. You can have genius genes, if you get a head injury or some brain-scrambling perinatal infection, you will never be anything but an imbecile. If you grow up in the jungle without books, you will never learn how to read. This is one type of G*E that is definitely real but probably doesn’t do much to undermine hereditarianism.

Sesardic calls the other kind of G*E “biometric”: depending on the level of some environmental variable, the effect of genes differs. Because the “environmental variable” can be anything, the law of large numbers guarantees that some version of this is also true. For example, in the Bingley paper, the environment was being exposed to a school reform: do genes matter more in those exposed to more schooling? (Not really.) Is the heritability of IQ changing over time? Probably not. Education? Maybe decreasing. Do mothers with red hair have a larger shared environmental influence on your personality? OK, I don’t have a study for this one. Because there are infinitely many interactions to be tested, some must be real but we will never be able to reliably tell which ones because the possibility for p-hacking is infinite.

The most serious story about G*E is the Scarr-Rowe hypothesis: genes are more important in the upper social classes because the environments are already maxed out. The Scarr-Rowe hypothesis treats the environment as a source of trouble: if you are poor or your parents are stupid and violent, you have a disadvantage which masks your genes so they matter less and the shared environment more. The best you can hope is to have none of these problems – indicated by a higher social class of origin – so your genetic potential can shine through, evident from higher heritability and lower shared environmental influences. The Scarr-Rowe hypothesis blew up in popularity after one 2003 paper by Eric Turkheimer which found evidence for it. The paper was cited almost 2000 times but it was never really replicated. A meta-analysis published in 2015 found such an effect in the United States, but not outside of it – maybe strong European social nets max out environments even for the poor? But then a very large study – bigger than the meta-analysis – found no Scarr-Rowe effect in America either. Some other studies found some evidence for the Scarr-Rowe hypothesis, some none at all, one of the largest in the UK Biobank even the opposite pattern: lower heritability in the upper social classes. Maybe there is something to the Scarr-Rowe hypothesis, but I’m not sure. Heritability seems mostly independent from social class (and also race).

However, I still think G*E is not only real, but it makes genes the less important force if we zoom out far enough.

Heritability is the proportion of individual differences that arise due to differences in people’s genes. There is a lot to unpack in this sentence. Between MZ twins, every difference is automatically nonshared environmental: no difference in genes, no heritability. If they were real, Star Wars clone troopers probably had somewhat different personalities and intelligence: these are would be all environmental.

In the same way we eliminated genetic differences in these examples to reduce heritability to zero, we can also constrain functionally important environmental differences to zero. This is arguably what happens in a typical twin study, which compares families who may differ in their income levels, education, violence at home and many other traits, but still generally live in the same era in the same country.

It’s kind of a big deal that growing up in a richer, better educated or less violent family is not a “functionally important environmental difference”, as suggested by low shared environmental effects in the quantitative genetic literature. It falsifies much of sociology, psychoanalysis, and popular Marxist style ideas about what causes inequality in society. But just as no heritability of personality in cohorts of clone troopers didn’t mean that genes cannot influence personality, this here also doesn’t imply the environment can’t: it’s just that most of the meaningful environmental differences are not there between families living in the same country in the same era.

It is called the ecological fallacy when we assume that an explanatory mechanism that works on one level also automatically works on another level – for example, that if genes explain most individual differences in a contemporaneous twin/molecular genetic sample, then genes will also explain the same in every context. Yes, between adult males, most height differences are due to having different genes – heritability. But brothers in the same family tend to be taller than their sisters even though heigh-related genes are segregated randomly between siblings: women just end up shorter even with the same genes because increasing estrogen levels during puberty fuse their growth plates early. Within an elementary school, age is the best predictor of height.

A purely genetic explanation also breaks down when we pool different countries and different eras, which is the kind of G*E that I thinks challenges hardcore hereditarianism most effectively.

The most conspicuous example of a gene-environment interaction can be seen for height. (Forget about IQ and the Flynn effect: we are not sure if people actually got smarter in the past century or only better at writing tests, probably the latter. Height is objective.) In typical family studies, height is highly – almost perfectly – heritable. Yet, there has been over a standard deviation of an increase in height in the past century or so: given the same height genes, you ended up much shorter in 1825 than in 2025.

Nobody lives like it’s 1825 anymore so one could argue that these are unimportant environmental differences[1]. But we have a natural experiment about something similar: North vs. South Korea. The Koreans are just one people with no plausible segregation of height genes. Yet, after the artificial division of the country in the 1950s, South Korea ended up a modern capitalist country and North Korea stayed poor and communist. Contemporary South Koreans are over a standard deviation taller than North Koreans.

Mean heights of North and South Korean males. Before you come up with some galaxy brain story about how South Koreans were always taller, look how the height difference disappears in the oldest cohort. A long time ago, all Koreans were poor and short but now only North Koreans are. Within the same country and era, it’s genetics that mainly causes height differences, but big enough differences in the environment can be more important than genes.

If we could jointly analyze extended Korean families, we would likely find that the influence of the shared environment – which side of the border you ended up on – is not only real, even for something like height, but possibly trumps genetics. Given the same genes, it may not do much for your height if you end up in an American family with a better or a worse diet, but North Korea vs. South Korea-level environmental differences are big enough to matter.

Another striking evidence for a gene-environment interaction are the differential income and development levels within Europe. The map of my preferred development indicator, nominal gross wages (no unrealistic oil or financial offshoring effects, no weird PPP deflators, no discrepancy in what counts as “net”) looks like this:

For Lithuania and Romania, so called “super-gross” wages are displayed, which include taxes paid by the employer, not by the employee. Most countries do not include these, so much for my preferred indicator.

In the HBD-sphere, it is generally assumed that international differences in development are due to differences in the genetic potential of populations, mostly captured by IQ scores. I do believe that there is likely a difference in the genetic potential for intelligence (and maybe other traits) of major races, so on a global scale there is much merit to the HBD argument. But within Europe, genetic distances are minimal, migration and intermarriage was frequent, so it is not plausible that genetic differences account for the quite large differences in development. Actually, there is no difference in phenotypic IQ either: Russia, Hungary or Poland scores just as high as England or France. Yet, there are easily twofold or threefold differences in development between genetically highly similar populations (often legally part of the same country just a century or so ago), most notably when comparing Austria and Germany to their Eastern neighbors who were exposed to Communism, and countries that made it in the EU after the fall of Communism to those that did not. For example, up to World War 1, Czechia was just one of many areas in the Habsburg empire, not much more different from Austria than two US states, same laws, same taxes, same architecture, the same or even higher level of development, in many ways the same people – something like a third of the population of contemporary Czechia was ethnically German until World War 2. Yet, after just 40 years of Communism, despite the 30 years that passed since then, average Czech incomes are still less than half of Austria. I’m in the bad company of far-left activist authors when I say that interventions by powerful external actors, “structural forces”, and setbacks in the distant past can sometimes explain important differences today better than genes, but I think there are cases when this is true, and I’m acutely aware of the Eastern European example.

These between-era, between-country differences in environments are a grossly understudied field and likely a strong source of potential environmental effects. This is the drum a good anti-hereditarian should be beating, not arcane issues with genetic modelling methods.

[1] It is also true that height may have been just as heritable 200 years ago as today, the people with the most IQ genes ended up the tallest, it’s just that the absolute value of “tallest” changed. (One could compare the correlation between a height PGS extracted from archeological samples with femur length or skeleton length and compare its accuracy with a modern sample to check if this is true.)

Lavoisier was more right than wrong. He started with a solid observation – that heat causes objects to expand – and gave a mechanistic explanation of why that happens. He was wrong about the specifics – heat is not a fluid – but he was right about the big picture that objects expand and melt because their particles stay together to a different degree under different conditions. The heat fluid was just a small cog in a greater machinery.

Behavior genetics is one of the very few fields in the social sciences which, just like thermodynamics, has a paradigm, a mechanistic theory about quantifiable elements and testable rules that govern reality. It proposes that the resemblance of family members (a core observation, just like objects expanding over heat) is caused by a combination of genetics and family environment. Behavior genetic studies tend to find a much stronger effect of genetic than environmental forces. Therefore, you can think of hereditarianism as going one step further and saying that it is not the resemblance per se, but the resemblance due to genetic reasons which is the “core observation” in need of further study, such as exploring what biological mechanisms cause family resemblance and to what extent they explain sociological phenomena.

Lately, some researchers, most notably geneticist Sasha Gusev and veteran researcher Eric Turkheimer have been attacking hereditarianism. They are saying that because recent molecular genetic studies were unable to pin down the genetic mechanisms that affect traits, these genetic mechanisms must not exist and the resemblance of family members must be environmental after all. Some of their criticism is well-founded. GWAS has turned out to be a disappointing method, and even after studying millions of people with millions of genetic variants we still cannot explain much about, for instance, hereditary disease.

However, these molecular genetic studies are not a strong test of the basic tenet of hereditarianism, the idea that the resemblance of family members is due to genetic and not environmental forces. They are not analogous to testing whether objects expand over heat: they are analogous of whether there is a “heat fluid”. Just like objects still expand even if there is no heat fluid, genetic forces are still the best candidates for explaining the resemblance of family members even if a very specific idea about what these genetic forces are – the relatively large, linear, additive effects of SNPs picked up by commonly used arrays – turned out to be wrong.

Below is the longer story of why this is.

How we look for genetic effects

There are two kinds of behavior genetics: quantitative and molecular.

It’s easier to do quantitative genetics. Twin studies are like this. You don’t even need real genetic information. Actually, you don’t even need to know what a gene is – people did twin studies starting in the 1920s, three decades before Crick & Watson. In a quantitative genetic study, you study members of the same family – often twins, who have the same age and grew in the same womb which gets rid of important confounders – and check if their phenotypic similarity scales with the degree of relatedness (genetic similarity). The idea is that while both genes and family culture could make members of the same family more similar, we can zero in on just the genes by only looking at pairs of people from within the same family, ideally those living together.

Quantitative genetics at work (source). Pairs of Swedish conscripts from the same families are increasingly more similar if they are more genealogically related. For some traits (height, BMI) this is more true than others (schooling, socioemotional skills), suggesting different degrees of heritability.

They all live in the same house in the same neighborhood, have the same books at home, same income, same parental education, so if half siblings are more similar than adoptees, full siblings than half siblings, MZ twins than full siblings or DZs, then we have evidence that genes matter. By analyzing the degree of similarity – with Falconer’s formula or with fancier path model approaches – we can be more precise and assign a specific number to the relative importance of genetic effects (heritability), of the rearing family which is considered very important by sociologists and psychoanalysts (shared environment), and everything else, including measurement error (nonshared environment). If within-family phenotypic similarity scales perfectly with genetic similarity (MZs are twice as similar as DZs), shared environment is zero. This is often the case. If, additionally, MZ twins are exactly alike, the nonshared environment is also zero and heritability is 100%. This rarely happens, maybe the closest case is height with a heritability over 90%. This logic of quantitative genetics is not flawless, it is not without assumptions (more on whether these hold later), but it is a solid framework to look at the causes of individual differences. The observations in quantitative genetics are the “objects expand over heat” of hereditarianism. You cannot sweep this off the table just because you found a new method.

Molecular genetics relies on knowing the actual DNA sequence. It is newer, much more high-tech, and in a way more serious. Actual genes instead of some vague info about relatedness! Proponents are correct that molecular genetics has the potential of replacing quantitative genetics entirely, because all the information available for the latter – and much more – is theoretically available for the former. But this is true only in theory because you need both the data and a good model of how genetic data leads to traits.

There are basically two ways of using molecular genetics in behavioral research:

- Quantitative genetics with extra frills. We can use the DNA sequence instead of genealogical information to estimate the degree of relatedness between people, and use the quantitative genetic logic from there: are more genetically similar people also phenotypically similar? You do this either with strangers of family members, because everybody is related to everybody else to some degree (do you know your fourth cousins?) and because even family members of the same degree randomly jitter a bit around the degree of genetic similarity expected from genealogy. Using strangers is the way of GREML-GCTA and similar methods, and family members are used in sib-regression and RDR.

- Looking for the genes. This is the real deal: genome-wide association studies (GWAS). This time, we are actually looking for genetic variants causing a trait: the molecular background of intelligence, personality, disease etc. We do this by getting molecular genetic and phenotypic information from very many people and simply looking for genetic variants which are more common in those with a higher level on a trait. Because we have very many genetic variants, we must set the evidentiary bar very high to avoid false positives. When we are done, we can create a formula for a polygenic score: a weighted sum of all the variants we found, a genetic predictor of the trait. We can test in new samples how well this works, we can philosophize about the nature of the variants we found (“height genes are express in the bones, IQ genes are expressed in the brain”) and we can use the score to make predictions about people or groups with known genes but unknown trait levels, such as an embryo.

However, while molecular genetics looks like a more mature method, it still has assumptions. If these are violated, the results can be off – false negatives and false positives are both possible. The assumptions are:

1. Gene effects are additive. One SNP=one unit trait increase, two SNPs: two units. This is reasonable because of what R. A. Fisher realized in his theory of geometric evolution way before DNA was even known. There is no such thing as a “gene for intelligence”: pleiotropy is widespread, genes do many things, if you have a new mutation that raises your IQ you will have other changes as well and they are more likely to be negative than positive. If you want to evolve high intelligence, you are best off with many genetic variants causing this, each with a small effect (so the pleiotropic side-effects cancel out) and each independent of the others (so your descendants are guaranteed to get the IQ raising effect regardless of what their other genes are like).

2. We know all the genes. This is iffier. Most molecular genetic studies see a few hundred thousand common SNPs, specific genetic markers with a lot of variability between people. But there are rare SNPs only a few people have, non-SNP genetic variation (CNVs, indels etc.), and if we go crazy, mitochondrial DNA and maybe some non-DNA mechanism of inheritance (microbiome, epigenetics, some totally undiscovered mechanism?). Both the GCTA and GWAS approaches only see a part of all genetic information: the additive effect of common SNPs. If there are others, there are false negatives. There are also false negatives if we don’t have enough power to detect small effects. Given that even for height four times as many SNPs and almost twice as much variance was found by going from 700k to 5M people, this is not a small issue.

3. There is no stratification. This is the assumption which can result in false positives if violated. Just because a genetic variant is associated with a trait it doesn’t mean it causes it. Asian people usually have a special version of the EDAR gene and can eat with chopsticks – but EDAR is not the “chopsticks gene”, it’s just that there is an accidental overlap between people with the EDAR gene and chopsticks-eating culture.

Ancestral (left) and novel East Asian (right) versions of the EDAR gene. Sorry Gregory Cochran, after your blog’s name I’m also stealing your meme.

A less on the nose (but more realistically problematic) version of stratification is if genetic effects index a shared environmental effect. One version of this is a dynastic effect. Certain genetic variants are probably more common in the Rothschild family, but they don’t make you rich, they just show where you are from, and that does. In a naïve population-wide study, they can still end up being associated with income as proxies of Rothschildness. Another version of genetic nurture, where the genes change parental behavior and parental behavior changes you. You also inherit the genes, but this is not the important part, it was your mother having had them when you were little. Both are cases of passive gene-environment correlation: the genes being stand-ins for environmental causes.

These assumptions are different from those in quantitative studies, so if the two methods result in similar findings, we can be happy that these are robust. But if they do not, we don’t know if the quantitative or the molecular findings were wrong. There is no natural hierarchy between these methods: molecular genetics is not superior to quantitative genetics, it is not the next step in the evolution of the field, merely a new method. They both have strengths and weaknesses. Quantitative genetics uses more complete information so would expect it to see more genetic effects, at the cost of less biological information about the precise nature of them. Molecular genetics is much better for discovering mechanisms at the cost of likely missing the larger picture. You can think of it as looking for a specific kind of “heat fluid”. If it’s not there, thermodynamics is still real, it’s just that you need to find a different proximal mechanism.

If the two methods have matching results, it’s great. If we had a polygenic score that accounted for 80% of intelligence variance, we would have shown that not only is intelligence highly heritable, but caused by common SNPs with a linear effect large enough to be seen in our current sample. The fact that we don’t simply doesn’t say anything about the heritability of intelligence, merely that a very lucky outcome didn’t come to be.

The GWAS embarrassment

Modern molecular genetics turned out to be something of an embarrassment. In the early 21^st century people believed that just knowing what genes there are and how they work – the Human Genome Project – would be enough to understand all apparently hereditary disease. For example, Francis Collins, director of the Human Genome Project said in a 2000 interview that:

“In the next five to seven years, we should identify the genetic susceptibility factors for virtually all common diseases — cancer, diabetes, heart disease, the major mental illnesses — on down that list. Out of that will come the ability to make predictions about who’s at risk and, in many instances, that itself can be extremely helpful by allowing you to take advantage of surveillance, so that you are diagnosed early before a real problem arises. In the longer term — perhaps ten or 15 years — this will lead to a new generation of therapies that is going to be much more successful and effectively targeted towards the basic problem than what we have now.”

, and the next year confirmed this opinion in a testimony in front of the US Senate.

Needless to say, no such thing has happened. Forget about intelligence and other based stuff for now: molecular genetics just doesn’t work well, even in medicine. In a typical modern GWAS, researchers look at data from hundreds of thousands, sometimes millions of people to figure out which genetic variants – or a combination of them, a polygenic risk score – are associated with certain traits. You would expect that with so much information and such high tech approaches – SNP arrays which digitize molecules to a TXT file, arcane statistical approaches, the known biological function of thousands of genes – they would be able to learn a lot. But they don’t.

I looked at some recent medical GWASes which looked state of the art, and they were unimpressive. In this study, using genetic information from 13790 bladder cancer cases and 343502 controls the researchers could construct a genetic score which predicted about 50% of extra cancer risk per standard deviation. There are similar figures for heart disease. A trick I often see in these studies is that the authors focus on “new loci”, specific genetic variants associated with the trait. Their discovery is claimed to be a big deal, and they write lengthy paragraphs with a lot of biology and chemistry about what the genes close to these variants may do to cause the disease. This makes the paper science-y and keeps the grant money flowing. But each of those variants confers maybe 0.001% extra risk! All these speculations about mechanisms are an exercise in futility because the causal link from gene to trait is just not there. The observation that we cannot really find the genetic variants underlying the heritability suggested by quantitative genetic studies is called the missing heritability problem.

Based traits like educational attainment or intelligence are actually among those with better GWAS results. The most recent educational attainment GWAS, based on over 3 million people, could build a genetic predictor which correlated about ~0.37 with actual educational attainment, which is not bad given that the number we are trying to replicate is ~0.65 (the square root of a meta-analytic twin heritability of 0.43). This predictor also had a decent correlation with cognitive ability. Not bad, but 20 years ago we would have expected much better with 3 million genomes.

Also, for these traits, there is a particular embarrassment. In 2018, a group of Icelandic researchers discovered that your parents’ genetic variants explained about half as much variance of educational attainment than your own, even if you did not inherit these. This led to the recognition of indirect genetic effects: the idea that genetic variants can be proxies for environments instead of true biological causes. These are the “genetic nurture” and “dynastic” effects I wrote about, and they mean that even of the small effect of discovered genes a big part is an illusion. To address this, modern studies use within-family designs: for example, in a GWAS they would compare sibling pairs, not a group of random people, to discover which genetic variants truly cause traits. They usually keep finding these indirect effects, which make GWAS findings even less impressive and missing heritability an even bigger problem. This suggests that, ironically, genetic variants may pick up the admittedly non-zero shared environmental effects on socially important traits better than the actual genetic ones. (Keep in mind that within-family studies are always less powered so they are biased towards null findings.)

Apples to oranges

Turkheimer’s and Gusev’s whole shtick is acting as if molecular genetic studies are a test of “heat expansion” and not just a specific “heat fluid” – that if we are not finding all the genetic effects using the current molecular genetic methods, they must not be there. They play switcheroo with terms, jumping between “heritability” meaning heritability by quantitative methods or how much variance molecular methods explain.

For example, in his debut piece “No, intelligence is not like height”, Gusev claims that “IQ is much less heritable” than height, based on molecular genetic studies with vastly different statistical power to detect genetic effects on the two traits. (See Joseph Bronski’s response to that particular post.)

Turkheimer makes me scratch my head even more, claiming things like that:

“[Hans Eysenck] estimated that the heritability of neuroticism was .81. […] This week, a preprint describing a massive meta-analysis of sibling GWAS dropped. The estimate of the direct genetic effect heritability of neuroticism was .081, down from Eysenck’s estimate by an order of magnitude.”

, confounding a twin result with an underpowered within-family GCTA-style study. (David Hugh-Jones calls him out on this in comments on this piece.) They especially like to pile on the fact that within-family estimates are lower than population estimates, hinting that this means that shared environmental effects are very important and heritability is overstated. But this is a minor thing: the majority of quantitative heritability is not found even in simple between-family molecular genetic studies. Compared to this, the fact that a third of the remaining effect is also gone once we look within families is just salt on a headshot.

Gusev and Turkheimer treat quantitative genetic studies as outdated, messy and based on arbitrary methodological choices. Turkheimer hardly engages with them anymore, semi-joking that a still very small within-family molecular genetic study is “the end of social science genomics” and that not taking molecular genetic results as the true estimates of heritability would be a “Plomin-style” ignoring of “null results”. Gusev tries to be more serious and actually engages with quantitative genetic studies in two posts. In the first, he keeps making the argument that if a molecular method (RDR) finds lower heritability estimates then the quantitative estimate then the latter is wrong, which, as we saw, is a fallacy. In the second, he writes that “twin heritability models can tell you whatever you want to hear” and dazzles the reader with complicated equations and path models used in twin modelling. He singles out one galaxy brain paper, Bingley et al 2023, as evidence that heritability can be made very low and shared environmental influence very high. This paper gets this result by assuming equal avuncular shared environments, in other words, by making the clearly wrong assumption that all differences between the families of cousins are environmental. [Sasha Gusev and Unboxing Politics both pointed out in the comments that this is a wrong reading of the paper’s assumption. I’ll have to get back to this paper - let it be known that everything I write is potentially wrong and I acknowledge the errors I make.]

At this point, let’s circle back to the original analogy, that our undeniable core observation is that “family members are similar”, much like how Lavoisier’s observation was that heated objects expand. Lots of different things could be pushing heated molecules apart, so falsifying the heat fluid theory still leaves the chemist with a lot of different ideas to explore. However, by definition there are only two things that can make family members similar: genes or the shared environment. By denouncing genetic effects, Gusev and Turkheimer are implicitly saying that shared environmental effects account for almost all of the similarity of family members.

Now I’m not a bigoted hereditarian and I wrote a piece about the importance of the shared environment myself, but this is a very tall order. Because of quantitative genetic findings – the only kind which can find shared environmental effects – it is extremely unlikely that shared environmental effects predominate.

The missing shared environment

Sasha Gusev is right that quantitative genetic studies have their own assumptions. But in some designs, the assumptions are minimal. Plus, different designs have different assumptions so it would be a big coincidence if they all produced the same fallacious result.

One design is monozygotic twins reared apart (MZAs). These are people with the same genes raised by different families: a perfect test of genetic effects without shared environmental confounding. MZA studies are very hard to do and there are roughly two samples with a reasonable size: the America MISTRA and the Swedish SATSA. Both concur that, for example for IQ, being reared together doesn’t matter much. (The MISTRA paper discusses lots of other phenotypes as well.) I made this mashup of results from the SATSA paper and this one:

On the upper left you can see the intelligence twin correlations in SATSA. Note that MZAs are just as similar as MZTs. Hello, Jay Joseph, this has DZA data too! In MISTRA, the MZA-based heritability estimate was 0.81 so almost exactly the same. One could quibble (see Kaplan 2012) that 1) MZAs were not raised completely apart, they had some contact, and 2) members of the same pair were adopted by similar families (selective placement) so in a way they did have some shared environments. On the chart I’m showing on the right, Bouchard et al show how 1) is wrong: there is no relationship between contact duration and similarity (and also note that MZA pairs were not much less similar than the same person tested twice!). The table on the bottom shows that 2) is wrong too. Bouchard et al show that selective placement is real (socioeconomic characteristics of the adoptive families are correlated) but controlling for it doesn’t change the twin correlations. Simply put, MZAs ending up in more similar families don’t end up more similar. If identical twins have similar IQs because their parents raised them a certain way – as implicitly claimed by Turkheimer and Gusev – then why does the shared environment appear so inconsequential?

We can of course look at the shared environment directly by studying adoptees. Do children growing up in the same family end up similar? Yes, as long as they actually live in that family. Bouchard & McGue’s classic chart about adoptee IQ correlations:

There are only a few studies which were published since then, some in Nancy Segal’s virtual twin samples (similarly aged adoptees) and this one in adults. These replicate Bouchard’s initial results very well. The virtual twins have somewhat similar IQs (r~0.25), but this is unsurprising: they were still children when they were studies and families do influence the IQs of small children. But in the adult study this is only at 0.06, not even statistically significant. For the sake of simplicity I focused on IQ, but see my previous post about attachment and Inquisitive Bird’s great shared environment post for other traits. If growing up in a certain family is so important and genetic effects are an illusion, why do adoptees end up barely or not at all similar?

Next up: twin studies. Gusev releases a lot of intellectual squid ink when he dazzles you with the many ways one can analyze twin data: in truth, most twin studies are fairly simple, they just assume additive genetic, shared environmental, and nonshared environmental effects. Basically, they assume that twice the excess similarity of MZs over DZs is heritability. These studies tend to show high heritability and low shared environmentality in basically whatever trait you use them for, something Turkheimer himself famously noticed 25 years ago. But there is a twist: this depends on age. In children, heritability tends to be lower and the shared environment is still important, whereas adults grow out of whatever their parents did to them and regress to their genetic tendencies, so heritability grows. This is known as the Wilson effect, which shows up for intelligence, psychiatric problems, attitudes, body mass index, and even attachment.

The IQ Wilson effect

Note that this is exactly the same we saw in the adoption literature.

Let’s assume that twin studies are done wrong, it’s not the genes but the parents that matter, MZs have uniquely similar parenting experiences, and this, not the genes, makes them more similar than DZs. Why is this most evident in adults who are long done being raised by their parents, one way or another? Why isn’t this mirage heritability strongest when parents are the most active, and why don’t children from the same family regress back to the relatedness-independent effects of the parenting they experienced, erasing MZ-DZ differences? The Wilson effect is evidence that the twin method can very much produce high shared environmental effects and low heritability: it totally does this for children, just not for adults. A hereditarian can easily explain this with Warne’s first law of behavioral interventions: the shared environment’s effect fades when its gone, but genes stay with us lifelong so ultimately they get the upper hand. An anti-hereditarian cannot.

We can also test some twin study assumptions directly. Quantitative genetic studies check if the similarity of pairs of family members scales with genetic similarity. Normally, there are two things that can make family members similar: family culture (shared environment) and genetics. By only studying pairs of people from within the same family, we are holding the first constant and we are only looking at genetics. If a trait is heritable, we expect full siblings to be more similar than half siblings and MZ twins more similar than either DZs or siblings. Twin studies are the best kind of quantitative study in many ways because twins are the same age and gestated in the same womb, so we don’t have to worry about these confounders.

An assumption of this method is that family culture is distributed evenly and genetically more similar family members don’t, for example, get more similar parenting. This is called the equal environment assumption. As it happens, this is wrong. MZ twins are reared more similarly: for example, parents often dress up MZ twin kids in the same clothes, but this rarely happens for DZs.

Felson (2014) is perhaps the nicest illustration of why the equal environments assumption doesn’t matter. MZs really have (and had, as children) more similar environments – we know this from self-report scales (in this study, from the American MIDUS sample). But these self-reports can be used as covariates in twin analyses: regardless of zygosity, do twin pairs with more similar environments turn out more similar? Felson says no:

We see here the twin heritability of many traits, measured in the ordinary way (circles) and after controlling for environmental similarity, which gets rid of the problem of unequal environments (at least as long as the 15 measures of these capture whatever is relevant). The heritabilities barely move: the only statistically significant drop is for neuroticism, which becomes slightly less heritable after adjusting for environmental similarity[1].

An even more straightforward way to test the equal environments assumption is to use misclassified twins. As it turns out, parents of twins are frequently wrong about the zygosity of their kids. Misclassified twins – MZs believed to be DZs or vice versa – should get a mismatched genetic and environmental package so looking at the degree of their similarity is a good way to gauge which one matters more. If the degree of genetic similarity is the key, misclassified twins should be as similar as their genes predict, and if parenting matters more, they should match the “social zygosity” their parents assign to them. The most comprehensive study I know about this is Conley et al 2013, who checked the parents’ beliefs about zygosity using genetic information and found that genes matter much more:

I parsed together three of their figures. The top two show the similarity of misclassified twins (DZ incorrect: DZs raised as MZs, MZ incorrect: MZs raised as DZs) compared to those correctly classified. It is much more common for MZs to be raised as DZs (almost a quarter of all pairs!) and they are just as similar as ordinary MZs. DZs raised as MZs, on the other hand, don’t catch up to real MZs in terms of similarity, but it’s hard to say anything precise about this due to low sample sizes (12-16 pairs). On the bottom chart, the authors show that this misclassification problem is not a trivial problem in twin research: Heritability is lower if you go by parental reports of zygosity (“perceived”), or even a standardized questionnaire (“classified”) and higher if you use real genetic information (“genetic”). Not only are equal environments not an issue, true heritability is even higher than seen in most studies which don’t have genetic data!

In one comment I found very interesting, Sasha Gusev wonders “how parents (and society) treat kids when they look identical”. This is important for us because maybe MZs are so similar because other people – not just your parents – react to their appearance in a certain way. It’s harder to make it in life if you are an ugly person. As it happens, there is direct research about lookalikes, people who look very similar despite being strangers: I’m aware of at least four such studies.

To me, lookalikes look like MZ twins (top left) but based on three algorithms they are actually between random people (non-LAL) and MZ twins and they are slightly, but significantly more genetically related than random people (top right). The Cell Reports paper I linked found that these look-alikes are slightly more similar in some traits – mainly genetics, height, and education – than random pairs of people (bottom left), but unfortunately didn’t directly compare them to twins. This was done in all the other linked studies by Nancy Segal, however. One example is shown on the bottom right: MZ twins, as we know, all have substantial personality similarity but lookalikes do not. There is not much evidence that how parents and society treats us when we look a certain way has a strong influence on us: it certainly isn’t the main reason MZs are so similar.

You might say that hereditarianism is wrong, but you cannot deny that family members are similar. If you don’t think this is genetic, you are saying it is shared environmental: there is nothing else that makes family members similar. I fail to see how the results with MZAs, adoptees, the Wilson effect, misclassified twins, or lookalikes are compatible with genetic effects being an illusion and the shared environment being dominant.

[1] To be fair, the trend is downward in all the other cases, but I’m autistic enough to notice that this method overcontrols for environmental similarity. This is because the environmental similarity measures might themselves be heritable and genetically correlated with the traits, which means that by controlling them we are getting rid of genuine genetic effects. Maybe saying “my twin cares about me” is not a totally objective measure of what this question means, but taps into a heritable tendency to trust others which is also captured by the neuroticism test. In this case, controlling for the former gets rid of some of the bona fide genetic effects on neuroticism, not just the effect of really having a caring twin.

Elite Human Capital (EHC) theory is essentially the idea that left-liberal, globalist people and institutions are right after all. Although its name sounds like something which may have been around for decades, it was coined by Anatoly Karlin, an interesting Russian cosmopolitan, and picked up by relatively well-known American blogger Richard Hanania. Both Karlin and Hanania used to be very right-wing so endorsing this idea is a drastic change in their views. EHC is big enough now to deserve a response (see another recent one here), especially because although both Karlin and Hanania – both very smart people - ostensibly came to change their views based on new data and careful deliberation, they are still wrong.

EHC doesn’t have a formal definition and Karlin and Hanania seem to have a slightly different idea what that is so I chose to write about the two versions separately.

Karlin’s EHC: anger and heuristics above data

Karlin’s endorsement of EHC stems from his disappointment with right-wing political circles and Russia’s performance in the early phases of the Ukrainian invasion:

“the world of the triumphant Right is the world of the local baron and his cronies, who keeps things ticking in his corner of the backwoods through intimidation and murky backroom deals”

and

“it is a “based” hyper-masculine world where loyalty is the paramount virtue, and where failure and incompetence is overlooked at best, actively promoted at worst. Treason is the only cardinal sin”

and an angry hope that somehow the polar opposite of this – the left-liberal globalist order of gay rights, climate change protests and Black Lives Matter – is better and there’s only ”perhaps a 10% penalty to effectiveness”.

You have to be from here to know this but this is a quintessentially Eastern European idea. I met lots of people here in Hungary who were not enthusiastic liberals at all but said things along the lines of “those in the West have more money and better roads, so if they say gays need to marry and we need to take more immigrants they are obviously right”. It's a type of cargo cult thinking.

Karlin seems to mainly want to hurt the right out of disappointment and leftist ideas are just a cudgel to beat them with:

(His piece is also called “Why Jail is Programmed for All Rightoids”, a turn of phrase he uses a lot).

I’m ambiguous about personal ideologies motivated by hurt – sometimes an objectively good idea also needs to be personal for you to care enough about it – but Karlin’s accusation of “Rightoids” as uniquely corrupt, nepotistic and incompetent is lazy and easily falsifiable. There are plenty of instances of “Rightoid” vices in the globalist EHC world, too. Let’s go no farther than the European Union, one of the most EHC institutions in the world. (While I was drafting this, Vice President Vance made the same point in his speech at the Munich peace conference.) The EU, infiltrated by woke pressure groups, currently not only steals Hungary’s funding for nebulous violations of the “rule of law”, but even keeps Hungarian universities out of exchange programs and EU grants as a way to intimidate its government into compliance or the population to vote against it. Because this is happening in the framework of a new, opaque “rule of law process”, a mechanism the EU invented during COVID (the vote on it was coupled to the vote on the recovery fund to force states to accept it), they don’t have an open list of demands, the goalposts are always moving and only a murky backroom deal can end it, if ever. On the other hand, loyalty is the paramount virtue, and where failure and incompetence is overlooked at best, actively promoted at worst: Hungary’s widely popular government which delivered way above EU average economic growth for over a decade got the “rule of law process” for not being ideologically loyal, but there was no penalty to EU leaders for their disastrous policies on immigration, energy and foreign policy as long as they toed the line. It was also not against the “rule of law” to send a literal antifa terrorist to the EU parliament who was sprung from prison by being elected (MPs have immunity): treason is the only cardinal sin, and beating people with a hammer who look like rightoids is EHC. Karlin talks about how Gregory Cochran, a right wing blogger, squandered the grant he got from Ron Unz, but fails to mention the SPLC, a left-wing grift operation which runs an attack site on dissident scientists and solicited donations from well-meaning Americans to “stop the Nazis”, BLM, which was immensely profitable to its leaders, or the myriad leftist NGOs, foundations and think tanks financed by George Soros or USAID which provide sinecures for many grifting “journalists” and eternal “students”. Every political system has both enforcement of loyalty and corruption, it is disingenuous to identify these as uniquely right-wing. In fact, in a way the entire past decade was a story of how the Western EHC system of government is only practically, not morally superior than others (and even that is slipping away). When they need it, the US, the EU or the UK will just as readily arrest people for online comments or criticizing politicians, lock them in their apartments for years, cancel elections or not recognize democratic elections that people they don’t like win, fire academics for noticing facts, and many other things.

A problem more specific than my angry rant above is that Karlin’s EHC theory commits a fundamental epistemic error: it puts heuristics above data. Normally, if smart and motivated people with a strong moral code care about something you can assume it is a good idea. But this is just a heuristic, a rule of thumb, until you know more about those ideas. There is no guarantee that smart and moral people are always right, and we need to look at real-world data to check if EHC policies work. As it turns out, many EHC ideas are already known to be disastrous:

- feminism (which drives down birth rates and makes women unhappy),

- the support for mass immigration (which adds people to Western countries who are a net drain on the state through lower education/income, higher use of social transfers and increased criminality),

- support for the Ukrainian war (which destroys both Ukraine and the EU economy to avert a Russian threat which never really existed anyway),

- racial egalitarianism (which only works if handouts are given on a racial basis because there is a likely biological discrepancy in the abilities and motivations of ancestry groups)

- drastic measures to stop climate change (which never include things like nuclear energy which actually work, have poor cost-benefit ratio, won’t work unless implemented globally which is not happening, and even then only stops an event which is predicted to be the second-to-least likely world-ending scenario, with most predicting humanity will be fine)

- transgender ideology (which causes iatrogenic harm to a significant percentage of impressionable teenagers)

For the record, I’m ambiguous about a core EHC issue, gay rights. I think a non-trivial percentage of at least men is genuinely homosexual, not much can be done to increase or decrease it, and this doesn’t directly do more harm to society than any other fetish. I might be convinced to even support gay marriage if it was less politicized and remained something few people even hear about.

Karlin likes to beat “rightoids” over the head with how EHC supports these ideas so they must be correct. But this is thinking backwards. The fact that Western elites support ideas known to be wrong doesn’t mean that these ideas are somehow less wrong now: it means that there is an intellectual crisis in the West because even the elites operate on religious fervor, not rational thought.

Hanania’s EHC: re-branding the managerial elite

Although he sometimes resorts to Karlineque trolling, Hanania’s interpretation of EHC is much more serious. Hanania’s minimum EHC theory is so intuitive and dispassionate that it is hard to disagree with it:

“Elites are people who:

1. Participate in and have contact with long-standing, established institutions that have traditionally had cultural, political, and social power, even when the power of those institutions is waning.

2. Have status that is connected to their affiliations with such institutions or those involved with them, rather than a direct relationship with a mass audience, in terms of voters or consumers.

3. Have norms and professional incentives centered around raising one’s status via impressing other members of the elite, rather than a mass audience.

Once we have a definition of elites, we can define populism, which is a derivative concept. A good working definition of populism is that it is a worldview that:

1. Blames elites for societal problems and stands in opposition to them.

2. Champions values and aesthetics that have more currency among the general public than elites, namely xenophobia, traditional cultural values and forms of art, an openness towards mysticism and the supernatural, and a leveling informality in social relations.

3. Assigns status to individuals based on a direct connection to a mass audience or voters, rather than success within established institutions.”

However, this description of EHC is a little tautological: Hanania simply describes the way people with above-average intelligence and Western middle class manners tend to run things as EHC. Obviously, both smart/elite people and dumb/non-elite people both prefer society to be run in a way that they can be or at least like its leaders. If you have the manners and intelligence to become a member of high society, you prefer high society to run the world in its own way. If you don’t, you prefer the loud and rowdy pro wrestling style Donald Trump embodies, because this is how your kind exercises power.

It is an open question which style is better and under which circumstances. Preferring the EHC style Hanania basically reads N. S. Lyons’ piece “The China Convergence” and likes it. In that piece, and similar works like Auron McIntyre’s recent book “The Total State” the core argument is that even if a modern country has no king or emperor with total power, it is also not really democratic (as in run by its citizens who compete in a marketplace of ideas and an open political arena). This is because even in such a country eventually a managerial elite develops which seeks to expand its own power by uniformizing society, replacing the organic structures of society (families, churches, and ultimately even states) by its own ideology and institutions, and then harnessing the work of the resulting human paste for its goals. It does this not by force, but by saturating the public discourse by propaganda, by lobbying state power via NGOs, and ultimately overtaking organizations by sending loyalists to work there. Basically, because voters rely on mass media for information, and because government institutions can interpret law flexibly, you can do state capture by buying media space and training loyalists for the government. This managerial elite, which Curtis Yarvin called the Cathedral, does not have formal leadership. Its cohesion comes from ideological conformity: Cathedral officials are selected to be true believers, so they predictably act according to their ideology and training when a decision is to be made. This is a lot like how a tight religious community operates.

Hanania practically equates EHC with people who would make ideal members of the managerial elite:

“In addition to high IQ, EHC has two other distinguishing features: an interest in ideas, and a moral code that it lives by that goes beyond tribalism or a primitive form of machismo.

So take two people with the same IQ, say 120. One ends up owning a number of car dealerships in Ohio and makes a lot of money, but he never reads a serious book in his life or thinks about much other than the well being of himself and his family. If he votes, he’s likely conservative, just from instinct, not because he understands anything about the benefits of free markets. The other moves to NYC and becomes a freelance journalist, barely scraping by as she publishes articles on the topics she’s interested in. She spends a lot of her time reading nonfiction and everyone she knows has some kind of interesting job as a writer, artist, academic, or political operative.

Despite the two of them being equally smart, only the second individual is Elite Human Capital.”

That is, EHC are people who are not only smart, but in a way religious: they have ideas about improving the world and they are willing to compromise on their comfort to live up to them and to proselytize.

Of course managerial elites/EHC can have various ideologies and goals, some better than others. Hanania likes EHC because he likes the idea of society being led by smart and motivated people, as long as they have a good mission: maximizing economic growth, American world dominance, and peace (or at least the fewest possible causalities). Conversely, his worry is that the EHC or managerial elite adopts policies which don’t work, which includes both wokeness and right-wing idiocy like anti-vaxxerism.

Hanania’s EHC idea makes a lot more sense than Karlin’s and it is worth engaging with. There are two reasons I don’t like it: one ideological and weaker, another practical and stronger.

The ideological argument is that Hanania’s EHC, like all managerial elites according to Lyons, is utopian and meliorist, which you might not like. It is right wing progressive, as in right wing (not egalitarian) but still progressive (sees the world as fallen and wants to improve it). Hanania has little respect for traditions and the status quo: forcing people to take vaccines as long as this reduces infection rates, expelling millions of Palestinians as long as this minimizes deaths, or letting in millions of Indians as long as they make the GDP grow is a no-brainer because abstract individual rights and nation states are just obstacles in the way to progress. Hanania is enthusiastic about the world becoming a uniform economic zone dominated by corporations (“the nomos of the airport”) where everything, the people absolutely included, is designed to be maximally efficient and minimally disruptive. I’m making this sound bad, but I see the argument that of all the worlds we have tried so far this Wall-E universe has the lowest amount of suffering. Maybe we have access to too much information not to be tired of all the ideological bullshit that drove history and the best thing is to just get the biggest McMansion and the most streamlined product delivery experiences possible. Brave New World is the conclusion of utilitarianism.

The practical argument is that Hanania ignores the possibility that when it comes to human affairs, exposure to EHC-style expertise may have a net negative effect on your ability to make good decisions. I get the appeal of Hanania’s idea that you need to base policy on numbers and logic, not heuristics and feelings. In practice, however, this is not how EHC operates at all. When I wrote about Karlin’s EHC, I listed some bad ideas that EHC holds, but I’m actually struggling to think of any good ones it has. Respectable opinions about human affairs are virtually all wrong, even if you zoom out from woke insanity: education reforms don’t work, democracy doesn’t help countries grow, declining fertility is basically because we did away with the gender wage gap, prisoners cannot be rehabilitated, etc. Elite schools on human affairs such as psychology, sociology, economics or foreign affairs train what Nassim Taleb called “empty suits”, not real experts. Just using her natural instincts about human affairs a typical Eastern European commieblock babushka – definitely not EHC – may have a better idea about what to do with mental illness [1], sex roles, or criminals than a typical Harvard-trained sociologist. Smart people, while still working in the smart EHC ways Hanania describes, are just as susceptible to stupid ideas as dumb ones, they are just better at papering it over with nice prose and good manners and their competence keeps catastrophic failure at bay better. Outside of STEM I’m not sure EHC ideas are any better than populist ideas. Hanania could move the goalposts and say that “I want an EHC, just not the current woke, anti-capitalist, feminism-education-climate change-obsessed EHC” but this would be just saying “I want smart people running society in traditional smart people ways, only with actually correct ideas this time”. Hanania tries not to sound this naïve but this is what I’m reading when I try to understand his point.

The most egregious example for Hanania’s naiveté about the dumbness of smart people is his piece “Why the media is honest and good”. Hanania defends the mainstream media like the NYT by saying that 1) they do good-quality reporting on neutral events, which is what most content is, 2) they report even facts against their biases such as racial crime disparities, it’s just that they put an ideological spin on them, 3) they don’t make up fully fake news. Conversely, right-wing alternatives don’t have neutral stories at all, it’s all rage-bait about the liberals, reporting is selective, and they make up or re-publish fake news.

I agree with this completely, it’s just that this means the opposite of what Hanania says. When you read “rightoid” publications like VDare or the Daily Stormer, you immediately understand that this is a publication with an extreme political agenda, so you take everything with a grain of salt. Conversely, the NYT packages its lies and biases in great travel reports, movie reviews and accurate coverage of geopolitical events which make it more, not less, dangerous. A smart person reading on the Daily Stormer something like that “Sanna Marin was better than Annalena Baerbock because out of all the worthless sluts promoted as politicians people at least wanted to have sex with her” will immediately understand it as satire, something South Park did 20 years on prime-time television. On the other hand, after reading the NYT for real facts he will fall for their propaganda stories about things like police violence about Blacks, Zach Goldberg famous example, with liberals overestimating the rate of fatal Black police shootings by orders of magnitude.

Never explicitly writing false facts and packaging it in good reporting makes EHC press worse, not better. EHC propaganda has a hyperreal element to it: it never explicitly says something like “Donald Trump is a paid agent of Vladimir Putin who is basically Hitler” but by publishing endless word salad articles insinuating this this is how the truth assembles in reader’s heads.

[1] If you think I’m bluffing on this, read the chapter „Understanding Why Some Clinicians Use Pseudoscientific Methods” in this book about the accuracy of clinicians and laypeople evaluating mental illness.

Journal of Intelligence is going down a similar path. This journal recently published a special issue called “Assessment of Human Intelligence—State of the Art in the 2020s”. One could write a lot about state of the art intelligence measurement: for example, short intelligence proxies which can be used in very large samples, the open-source ICAR-test or adaptive intelligence testing. Unfortunately, there is very little of this technical state-of-the-art in this issue and much more about political state-of-the-art which suggests nothing good about the future of this journal either.

Right in the editorial they start by denouncing Lewis Terman and Henry Goddard, fathers of the field, as “despicable” and “devotees of Spearman’s g theory” in contrast to Binet, Simon and to some extent Wechsler who „believed neither in g nor fixed intelligence”. They also assert that “society and the scholars in the field have moved on from eugenics, g theory, and fixed intelligence, the landscape of modern assessment has moved toward a framework that is more equitable and socially just in concert with a society that increasingly emphasizes the importance of diverse viewpoints”.

This statement, by the way, is a lie. In the most recent anonymous intelligence expert survey, Rindermann et al asked researchers pretty direct questions about these issues. About the g-factor, responding to a question like this:

the mean response was 6.84, so the average expert believed the g-factor is a pretty good explanation of test performance. They asked indirectly about the ordinary, within-group type of heritability but I can’t find it in the results. However, since they even estimated the thornier Black-White differences to be ~50% genetic, the mean researcher is probably quite a hereditarian, which would imply both a belief in “fixed intelligence” and “eugenics” to some degree – if something is heritable, you could expect it to be relatively stable and breedable. They strongly disagree that IQ tests need to be more “equitable” or “socially just”, because they judge the tests to be relatively unbiased, especially against ethnic groups, and they disagree that separate norms are needed:

It’s bad news that the editorial is like this, but the issue is actually not that inflammatory, except for one paper by LaTasha R. Holden and Gabriel J. Tanenbaum, with the title “Modern Assessments of Intelligence Must Be Fair and Equitable”. This paper is worth some analysis, which was my main goal writing this post, because it’s full of ideas one can expect to be central in the new, woke intelligence research these journals plan on doing from now.

From the abstract:

“First, we highlight the array of diversity, equity, and inclusion concerns in assessment practices and discuss strategies for addressing them. Next, we define a modern, non-g, emergent view of intelligence using the process overlap theory and argue for its use in improving equitable practices. We then review the empirical evidence, focusing on sub-measures of g to highlight the utility of non-g, emergent models in promoting equity and fairness. We conclude with suggestions for researchers and practitioners.”

To call for DEI in assessment – that is, the measurement of differences – is odd to say the least. I guess one can claim that trans women of color either have some unique ability they can use on the board of Boeing, or they deserve the board position as a handout to compensate for past injustice. But psychometric tests by definition need to be standardized (that is, non-diverse), result in different scores in different people (that is, non-equitable) and inform some judgement based on these scores (that is, non-inclusive). A test which everybody can take in their own way, gives everyone the same score or doesn’t allow judgements about things like university or job admissions is useless.

The paper repeatedly calls for abandoning g theory and adopting alternative theories, such as Kovács & Conway’s Process Overlap Theory (POT), to achieve DEI:

“We argue that non-g POT is a better model due to it being more parsimonious (see McGrew et al. 2023), being better for equity, and because it shows us where to focus when we consider forms of cognitive optimization or enhancement through intervention.”

“non-g POT is also better for working toward equity and fairness goals.”

“Overall, developers of cognitive assessments should be updating their current practices to be fairer and more equitable for diverse groups. Part of this approach should include thoughtful examination of the theoretical bases of these assessments and shifting to non-g models of intelligence such as POT.”

It’s easy to see why this line of reasoning is attractive. Race differences in intelligence imply that there is such a thing as intelligence. If you can make the concept of intelligence disappear, maybe all the race IQ maps and Black test results will stop making sense too. IQ realism arch-nemesis Stephen J. Gould already wrote way back in his 1981 book The Mismeasure of Man that “g is the rotten core of Jensen’s edifice”.

This was a fallacy back then and it is a fallacy now. What happens here is confounding two separate questions: the validity and the structure of intelligence.

It is beyond any doubt that if you get the sum scores of IQ tests – especially if these are nice factor scores, that is, you weigh the points people get for specific exercises by their g-loading before summing – they predict important life outcomes such as education or income. This is validity. Actually, even things which are not nominally IQ tests will function like one, such as the SAT, the ACT, the PISA test, dyscalculia measurements, school grades, short intelligence proxies etc., so once you do any kind of cognitive measurement your baseline assumption should be that you are getting something pretty close to what an IQ test would tell you.

One problem with this can be bias. The test score is just an indicator of what you are trying to measure (cognitive ability). It is like the mercury in a thermometer, which is not temperature itself. Under most circumstances looking a thermometer is a good way to know how hot the weather is, but there can be special cases (e.g. breathing on the thermometer) where it fails. It is theoretically possible that e.g. uneducated people or Black people who have a slightly different culture are not less intelligent, only less familiar with the content of IQ tests, so given a certain level of intelligence they get a lower score than a more educated or White person. This would be bias. Sometimes there is bias against immigrants or African Blacks (but see this), but the tests are unbiased against American Blacks. This issue is so beaten to death that Arthur Jensen already wrote a whole book about it more than 40 years ago. To the extent a psychological test can be “equitable”, an unbiased test already is, and IQ tests are OK by this criterion.

So far, this is a purely statistical phenomenon, it is about prediction: you measure something to know something else. By showing low bias, you shouldn’t even be concerned that your measurement underestimates some people. When it comes to applied psychology – such as testing in schools or university admissions – validity is the end of the line. The SAT is not a scientific experiment, it merely tries to find the students best suited to elite universities.

The structure of intelligence is a completely different question. This is a question not really meant for psychometrics or social science anymore, but for basic laboratory cognitive psychology, neuroscience or even neurology.

g factor theory is based on the positive manifold, the observation that all test performances correlate. People who are better at remembering numbers are also, on average, better at completing matrices, they have faster reaction times, a better vocabulary etc. A g factor – a general intelligence you can use in every test so performance ends up correlating – is one possible explanation for this, but not the only one. Three alternative models from John Protzko’s great paper:

The network model (left) says that there are many independent human abilities that influence each other so they end up correlating. The bonds model in the middle says that not only are there many independent human abilities (bonds, in yellow), but they are even more particular than our language would suggest. An ability is not “reading” but “recognizing a tilted line that can be part of a letter”. Even cognitive tests we think of as very simple tap into many of these elemental “bonds”, so performance ends up correlating basically out of error, like being unable to push a single button on a tiny keyboard because our fingers are too big and clumsy. Process Overlap Theory (POT, on the right), which the DEI authors prefer, assumes that elemental abilities are organized into broad abilities like fluid or verbal intelligence. These are independent, and the only reason they correlate is because they all have to pass through the bottleneck of limited executive functions when we use them. This phenomenon gives rise to g. In all three models, there is no g, except as a statistical artifact.

Arguing that g is not real is like arguing that God exists. It’s not precisely defined just how real g or God has to be for our hypothesis to stand, so we can read whatever we want into the data. A very weak form of the hypothesis is obviously true: you can think of order in the universe as divine by itself, and there is certainly no organ in the human body which creates domain-general intelligence.

The weak non-g hypothesis is worth considering because on some level, human abilities are obviously modular so g theory in the strictest sense is wrong. If your occipital cortex is injured, you become blind. People with injuries to higher-order areas develop all kinds of strange specific deficits, such as the inability to learn new information (like in the movie Memento), recognize faces, pay attention to the left visual field, or name objects they can otherwise copy just fine.

Weird agnosias in some clinical cases. After brain injury/stroke, these people lost very specific abilities (source: Kandel-Schwarz: Principles of Neural Science).

Research with actual IQ tests and statistical models shows that people with brain injuries have specific, not general deficits. If g was some single ability in the brain, this is not what we would expect. At the most basic level, the brain must consist of networks to support specialized calculations and functions. General intelligence emerges from these modular abilities somehow.

But the stronger the claims of the non-g hypothesis get, the harder they are to defend. Just as there is no man with a white beard sitting on the clouds, there is no way to handwave away g-centric intelligence research with just neuroscience.

Maybe a very strong form of g-theory (that of almost a “g organ”) is untenable, but the alternative theories like network models or process overlap theory don’t stand up to data even in their weak forms. This is because of the very observation I brought up against g itself: injuries (and training effects) remain localized to abilities. This argument was recently made by John Protzko in two papers. Both cognitive training effects (improved abilities) and brain injury effects (impaired abilities) affect specific abilities, not intelligence. If you scroll up the models I showed up, you will see that in all of them the arrows go from lower to higher abilities: general intelligence is a consequence (or rather a “sum”), not a cause of lower-order abilities, so it should be affected by whatever changes to the latter. This is especially bad news for a network model in which training effects and injuries should spread all across the ability spectrum, but POT also struggles to explain localized deficits. (The bonds model can work as long as we believe that all of these injuries affect exactly one bond, but that is a tall order.)

While the brain is organized of specific functions at the most basic level, it’s questionable if we ever interact with the world with these. Maybe by the time we use our cognition for anything real, we are already using the interacting system of elemental functions known as g so that’s the right level of description. This would make non-g theories textbook cases of a Rylean category error: using the wrong level of complexity to describe what we are analyzing. One of Ryle’s examples of this phenomenon is a person being shown a cricket team. After hearing what each player does in the field, he asks which player is responsible for the “team spirit” he heard so much about in cricket. Of all of these brain areas, each performing “visual analysis” or “speech comprehension”, which makes this famous “intelligence”?

To return to a previous point I was making, problems with explaining lesion or training data is not the biggest reason you can’t just use POT or a network model to explain away g effects. The biggest reason is that the structure of intelligence has no relevance for assessment (validity). Whenever you can even think about DEI, the structure of intelligence is irrelevant. This is because assessment is only about information: if we got that information quick and dirty, so be it. Bear with me on this for a little.

In their foundational POT paper, Kovács & Conway contrast formative and reflective interpretations of factors. Look at Uli Schimmack’s illustration below:

In a reflective interpretation, shown on the bottom chart, there is a real latent trait like “intelligence” or “extraversion” floating somewhere in a world of Platonic ideals, unmeasured. In our fallen world, all the measurements we can make come with error, but they still reflect this trait to some degree. We can estimate the unmeasured perfect trait by getting the average estimate of many imperfect measurements: summing up scores, ideally weighted by factor loadings. The core assumption here is that the reason measurements correlate is because they tap into the same thing. G-factor theory makes this assumption, but POT doesn’t.

POT instead assumes the interpretation shown on the top chart, a formative g-factor. Cognitive tests correlate for whatever reason, not because they reflect a real g. We can sum up scores if we want and calculate IQ/g, but it doesn’t automatically mean that “intelligence” is real.

Kovács & Conway bring up socioeconomic status as one example of a formative factor. They say that just because income, education, place of residence etc. are correlated, they do not reflect some unmeasured, Platonic ideal “socioeconomic status” which causes them. Instead, “socioeconomic status” is simply a sum score of all of these things, a handy aggregate of all the social assets we have.

It is true that socioeconomic status is poorly and flexibly defined and its components correlate quite weaky. In this sense, a total SES measure is quick and dirty information. Still, if research showed that socioeconomic status causes things like criminality (validity), then we would have to seriously consider it an important variable in applied psychology. The only way structure mattered is if you claimed that because “total SES” is an arbitrary sum score, more information could be extracted by using its components, such as income or education, as predictors. (This is the very argument Gary Marks makes in the paper I linked above). The IQ analogy would be that because g is just a formative factor, we should use test scores on reasoning, vocabulary etc. tests instead of their aggregate, IQ/g. But this would be false, g contains almost all of the predictive validity of IQ tests. I wrote about this before: I think that whether something is best thought of as a formative or a reflective factor can be decided whether useful information remains in the specific variances (for example, if personality items residualized for the total trait score still predict behavior). By this standard, personality appears formative, but intelligence reflective. Even if you disagree with me, even if POT is the true model of abilities in the brain, even if g is just a quick and dirty sum score of a formative factor, it still contains almost all the information you can use in assessment, just as if g theory was real. We know this because we measured it. If one structural model (g-theory) maximizes validity, that is the end of the line in assessment, even if the actual structure is somehow different. Basically, the more we want non-g theories to inform real-life decisions, the more wrong they get: at some basic neurological level they can be true, but they certainly don’t mean anything like “intelligence doesn’t exist after all so we shouldn’t measure it”.

The fall of both intelligence journals in such a short time is sad, especially now that DEI might be on its way to be rolled back. Maybe process overlap theory to become the PC, cathedral-approved intelligence model for the next few decades, which would result in a lot of wasted time, money and energy.

The other take was that although there are many quasi-experimental studies claiming to show causality (e.g. that gay people suddenly get worse after a law banning gay marriage is passed), these are p-hacked by ideologically aligned researchers and a fair reading of them is that policy doesn’t affect minority mental health.

Some new papers about this topic came out which I want to review. One is a twin study, one is a quasi-experimental study, and one is a review which proposes alternative mechanisms that can harm non-heterosexual mental health.

The twin study is:

Marzecki et al (2024): Initiation and Stability of Self-Harm in Adolescence and Early Adulthood: Investigating Social and Aetiological Factors in Twins

This is not really a study about “how well off are the heterosexual siblings of gays” but more about “what causes self-harm in young people”, and non-heterosexuality is one of those things. Self-harm is measured three times in the British Twins Early Development study, and the most interesting result is this:

What you need to see is that the numbers going away from the AT circles are all much bigger than 0, while those going away from ET are only like that if they go vertical. This means that the same genetic effects (AT) affect both current and future self-harm, but environmental effects are not like this. Non-genetic causes of self-harm are always strong, but not the same across the three waves. This is what the abstract summarizes as “stability in genetic influences and innovation in non-shared environmental influences on self-harm”. This is a very common finding in genetics and one of the motivations of the “gloomy prospect” hypothesis (that while genes are not everything, everything else is just random short-lasting things which are not modifiable). I note that I think this is a pessimistic reading of that literature and will absolutely go out on a limb disagreeing with Robert Plomin in his field of specialty, but that’s for a later post.

Anyway, the co-twin analysis most interesting for us is an afterthought in this paper and it is hidden in the supplement. The paper tersely says:

“These [co-twin analyses] were tested for the variables that can conceptually be unshared between twins (sexual minority status and bullying victimisation). The association persisted between sexual minority status and lifetime self harm reported at 21 ( B 95% CI = 0.11 – 0.98), but not between victimisation and self-harm ( B 95% CI = 0.22 to 0.45). See Table S8 for details. Gender minority status may be unshared between MZ twins, but it was not tested due to a small number (N = 29) of MZ twins reporting to be transgender or nonbinary.”

So even within MZ pairs, non-heterosexuality (but not bullying victimization) is associated with significantly more common self-harm. The population effect is 0.89 (Table 2) so about 38% of the association goes away after genetic control.

This is interesting but there are things I’m scratching my head about.

TEDS is a longitudinal study. It measures not just self-harm, but also sexual orientation/preference at different waves. The age 21 measure sounded like this:

" It was assessed using a single question about the sex of people to whom participants are sexually attracted, with responses including ‘Always male’, ‘Mostly male but sometimes female’, ‘Equally males and females’, ‘Mostly female but sometimes male’, and ‘Always female’. "

This is what we call “sexual preference”, similar but not identical to “sexual orientation”. For example, you could identify as a bisexual woman but have only ever had boyfriends (a common observation) – in this case your orientation is bisexual but your preference is for men.

At age 26, TEDS asked about sexual orientation instead:

“At age 26, twins self-reported their sexual orientation, with the following options: heterosexual, homosexual, bisexual, pansexual, asexual, fluid, prefer to self-define, unsure/I don’t know, and prefer not to answer. These were used for the descriptive statistics, but for the inferential analyses, the scores were dichotomised into “heterosexual” and “sexual minority”.”

The paper we are looking at now uses this age 26 measure of “orientation”.

However, while self-harm was measured at age 16, 21 and 26, we are only seeing age <21 in the co-twin analyses. Sexual orientation is supposed to affect something that took place before it was measured! Because sexual orientation is supposed to be more or less stable this would be fine if there was no age 26 self-harm or age 21 sexual preference measure, but both are available. Why not use age <26 for self-harm? Why not use the age 21 sexual preference data as the predictor, at least in an alternate analysis? Actually, how is the SE only 0.05 but the CI 0.87 units wide? The co-twin analysis in this sample was either sloppily done, p-hacked or both.

The quasi-experimental paper is

Lee et al (2024): State-level anti-transgender laws increase past-year suicide attempts among transgender and non-binary young people in the USA

From 2018 to 2022, 48 anti-transgender laws (that is, laws that restrict the rights of transgender and non-binary people) were enacted in the USA across 19 different state governments. In this study, we estimated the causal impact of state-level anti-transgender laws on suicide risk among transgender and non-binary (TGNB) young people aged 13–17 (n = 35,196) and aged 13–24 (n = 61,240) using a difference-in-differences research design. We found minimal evidence of an anticipatory effect in the time periods leading up to the enactment of the laws. However, starting in the first year after anti-transgender laws were enacted, there were statistically significant increases in rates of past-year suicide attempts among TGNB young people ages 13–17 in states that enacted anti-transgender laws, relative to states that did not, and for all TGNB young people beginning in the second year. Enacting state-level anti-transgender laws increased incidents of past-year suicide attempts among TGNB young people by 7–72%. Our findings highlight the need to consider the mental health impact of recent anti-transgender laws and to advance protective policies.

This came out in Nature Human Behavior, a very prestigious journal, although explicitly a woke one. The paper is very similar to Hatzenbuehler et al 2009, Raifman et al 2017 and Raifman et al 2020: it uses data from online samples measuring mental health in young Americans, and checks if mental health suddenly got worse specifically for “transgender and non-binary” participants after an anti-transgender law (for example, banning transgenders from women’s sports) was passed in their state. The results are weird:

The dots show the estimated change from baseline in the self-reported number of suicide attempts for transgenders after such a law was passed (vertical line). A “time period” is basically one year, with the limitation that data took some months to collect. While there seems to be a highly significant effect on suicidal attempts, it takes place in the second time period! (They hack out a first period p=0.049 by arbitrarily limiting the sample to 13-17 year olds. No preregistration.)

I think it’s strange that the YRBS-style question they use asks about the number of suicidal attempts in the past year. This must be 0 for the overwhelming majority of respondents. The study uses a linear model to estimate the number of suicide attempts, but the results replicate nicely if suicide attempts are binarized.

I note that Raifman et al 2020, the only quasi-experimental paper that otherwise doesn’t have huge red flags, doesn’t do this binarization.

The problem, other than the strange 2-year wait period before a law starts impacting transgender suicide rates, is that the effect vanishes completely if a slightly different self-report measure of suicidality (“seriously having considered suicide”) is used:

By this measure, suicidality seems to drop significantly by the third year after anti-transgender laws! If we can deviate from how a quasi-experimental study is normally done and look at not the first, but the second measure after the intervention, surely we can also look at the third. We need more anti-transgender laws to protect the youth from suicide!

The authors commendably do placebo tests by using not suicide, but homelessness and full-time employment as the outcomes:

Ironically, both measures show the exact same effect as the target measure: a significant increase two time periods after anti-transgender laws. The authors use another statistical test (a “two-way fixed-effects difference in differences” model) which doesn’t separate the dots but compares all post-treatment waves to baseline. With this test, nothing is significant. Despite what you would think, the study did not compare transgenders to non-transgenders, despite having ample data on the latter. It is a “difference in differences” study because it compares transgenders in states introducing laws to those in other states. Ideally, you would use non-transgenders as controls because maybe you are missing some trends in suicidality correlated with the laws.

Simply put, Lee et al 2024 is a misleading study, like most quasi-experimental tests of the minority stress hypothesis I saw. A common pattern is that the authors show null findings and go on to discuss them as if they showed something else, including suggesting policy. An unbiased reading of the results would tell you that this is a null effect. The effect of anti-transgender laws on suicidality is only significant long after the laws are introduced, and even then only if a specific measure of suicidality and a specific type of statistical test is used. Placebo tests which shouldn’t work end up showing the same pattern as the main analysis, showing you how easy it is to get results like this by accident. This is totally hushed up in the discussion.

I increasingly think that the correct reading of the quasi-experimental literature is simply that changing legislation about sexual minorities has no effect on their mental health.

The review paper is Pachankis & Clark: The Mental Health of Sexual Minority Individuals: Five Explanatory Theories and Their Implications for Intervention and Future Research

Pachankis is a pretty woke author, but if you read between the lines, this paper is BASED. They immediately handwave away biological theories of poor mental health in sexual minorities, and go on to suggest minority stress and four alternatives as causal environmental hypotheses:

1) Distinct lifeways: being gay means you will lead a different kind of life

2) Narrative possibilities: the groups we identify with give us a story about who we are, and maybe sexual minorities have very negative stories

3) Social safety theory: sexual minority members have fewer affirmations of safety, like supporting parents or a place in society.

4) Fast life history theory: sexual minority members are exposed to early adversity which leads them to develop a fast life history (impulsivity, risk taking, hypersexuality etc.)

The last two theories are not good. Social safety theory is the same is minority stress if you think “lack of safety” is the same as non-discrimination, and a special case of “distinct lifeways” if you think it is not. “Fast life history” is strange, and I’m not even getting into criticizing the long-debunked Freudian assumption that mental problems are caused by “early adversity”. Pachankis & Clark are unclear what they mean by this theory but there is no way to see it favorably. Does early adversity cause both non-heterosexuality and fast life history? If so, why call it “fast life history” at all? We don’t normally name causal theories by another consequence of the same cause – we don’t have a “low white blood cell count theory” of AIDS, both that and the illness are caused by the HIV virus. Or is it that the fast life history normally also present in non-heterosexuals causes mental problems? Then it’s a special case of “distinct lifeways”.

The first two, however, are bangers and I never saw them elaborated like this anywhere else.

“Distinct lifeways” says that if you are non-heterosexual, you will have different experiences, including some which plausibly impact your mental health. This is obviously true. If you are gay, for example, you will be in the minority and by definition you will not have biological children with people you are sexually attracted to (yet), which is otherwise a pretty basic human experience. These alone could make you depressed. However, especially gays and bisexuals live farther from their families, have fewer people in their lives they can rely on, and use drugs more frequently, among other things. Risk behaviors are strongly correlated with same-sex attraction. We would not expect these patterns for other minorities, though, and unlike discrimination, they don’t decrease as a society becomes more tolerant. Minority stress can’t explain why non-sexual minorities don’t have mental health crises, and why sexual minorities are not better even in more tolerant places and eras, but distinct lifeways can. Non-heterosexuality is a gateway to a lifestyle which is not good for the psyche.

“Narrative possibilities” is even more damning. If you think about it, what topics do we mention when sexual minorities come up? There is an empty reference to “pride”, but other than that the whole discourse is about overcoming oppression, fighting for mental and physical health, the various ways modern society is bad for non-heterosexuals. Every story about sexual minorities is a sad one. This victim narrative is similar to what Jonathan Haidt calls “reverse cognitive behavior therapy”: society is telling non-heterosexuals that although they should be proud of themselves for some reason and blame others for their problems, the core of their essence is that they are problematic. This cannot be good for mental health. Just like distinct lifeways, but unlike minority stress theory, narrative possibilities can explain why other minorities don’t have a mental health crisis, or why non-discrimination doesn’t make it go away. Ironically, by becoming more tolerant and woke, modern Western societies may have become worse places to be non-heterosexual by amplifying this victim narrative.

I think a combination of a shared genetic predisposition, distinct lifeways and narrative possibilities is the true reason for poor sexual minority mental health. On the one hand, non-heterosexuality is one manifestation of a heritable character which also contains impulsive and neurotic traits. On the other hand, once you are non-heterosexual, you are exposed to an unconventional lifestyle which takes its own toll on mental health. The third punch is that in modern societies you are fed a victim narrative which tells you that you are not just a normal person with a rare sexual preference but a fragile, sickly being in a hostile world.

In this post, I replicate this finding using another large representative sample of Hungarians, (N=8000, although only ~2000 participants were asked the necessary questions).

In this dataset, participants indicated how afraid they are of some disasters on a scale of 1-5 („Absolutely not afraid” to „Very afraid”). The disasters are „climate change”, „migration to Europe”, „an upcoming World War”, „an upcoming economic crisis”, „a pandemic”, „AI”, „robots taking people’s jobs” and „politicians governing the country poorly”. Mean scores:

Nothing too surprising. In the middle of COVID, pandemics seems the scariest. This is followed up by climate change and two very down-to-earth problems (economic crisis and poor governance. A world war seemed less threatening in 2020, and AI/robots the least. Item correlations:

This looks like a decent positive manifold, but a simple single-factor model doesn’t fit it very well. The AI and robots items don’t fit the others, not even as a factor of their own, so I decided to drop them. This way, a single common-factor model has borderline fit (TLI/CFI~0.99, SRMR=0.042, RMSEA=0.095). A residual correlation of 0.09 remains between a fear of migration and a pandemic: adding this to the model this reduces RMSEA to 0.065 and improves the other indices as well. I didn’t want to torture the model further: with some quirks, modelling a fear of these disasters as originating from a single “dooming” factor fits the data reasonably well.

Because only one item is about climate anxiety, it’s hard to tell how much of it is just dooming. Normally one would take several items, fit a latent factor to them, fit another latent factor to the dooming items, and calculate the correlation of the two latent factors: the error-free similarity of “climate anxiety” and “dooming”. I estimated this to be 0.85 in my previous post. Here, the best I can do is a model that looks like this:

Here, “climate anxiety” is measured by just a single item with a fixed loading of 1, and “dooming” (“rsk” for “risk”) by five others. Item 2 and 5 are “migration” and “epidemics”: fear of these is more similar than implied by a single factor so a residual correlation is necessary. This model gives me a latent correlation of 0.74[1], still quite high and biased downward by the fact that that “latent” climate anxiety factor is equal to the item score due to the fixed loading. Fit is pretty good: TLI=0.997, CFI=0.995, RMSEA=0.065, SRMR=0.028.

It’s fair to say that even in this dataset, a fear of climate change is mostly just a reflection of a broader fear of mutually exclusive societal disasters. “Climate anxiety” is very similar to just “worrying about societal problems”. Only a fear of economic crises has a higher correlation with the dooming factor.

The next step is to compare this “dooming” factor to other psychological measures. My dataset has a questionnaire about symptoms of depression and anxiety, which I wrote about before. A single factor model has excellent fit to both the depression (TLI/CFI=1, RMSEA=0.028, SRMR=0.01) and the anxiety (TLI/CFI~1, RMSEA=0.05, SRMR=0.02) items, but depression is unfortunately only available to a subset of the sample with no overlap with dooming data, so we can’t use it. The dataset also has items about fear in a more specific sense which I think is interesting to compare to dooming. The questions were a bit weird, because participants were asked to say how safe they feel (1-5 Likert scale) in environments increasingly close to their immediate surroundings, specifically: 1) non-EU countries, 2) other EU countries, 3) countries neighboring Hungary, 4) in Hungary, 5) in their town, 6) in their neighborhood, 7) at home, 8) on the Internet. I don’t think the modal Hungarian teenager has a strong opinion about his personal safety in non-EU countries, but maybe the scale is a good indicator of a type of anxiety more similar to dooming and less to conventional anxiety scales.

Very simple, the farther away, the less safe. Item correlations are weird:

The reds fading out as we go away from the diagonal shows that there is an autoregressive relationship between the items. With increasing distance between the environments, the similarity of fear decreases: “non-EU” is more similar to “other EU” (both quite distant) than to “own home” (very close). This is what we typically see in longitudinal studies where more temporally close measurements tend to be more similar. It’s interesting to see that the internet behaves like a distant place! Together with the means decreasing away from “Own home” it tells me that participants don’t really have strong opinions about how safe they feel in these different environments, the scale just measures a general feeling of safety. This is bad news for the survey makers, but good news for us, because we want to measure a general feeling of safety.

I tried to solve the autocorrelation problem by allowing a residual correlation between each adjacent pair of items, something like this:

Note that subsequent items are always more similar than predicted by the single factor. The last item is the internet, for that one I didn’t include a residual correlation. Even this way (and by adding one more residual correlation between “non-EU” and “home”, the big red arrow) the model doesn’t fit very well (CFI/TLI=0.99, RMSEA=0.125, SRMR=0.06) so take results about this fear factor with a grain of salt.

When I combine these items with dooming and the life satisfaction scale in a single model, the latent fear, dooming and life satisfaction factors end up correlating like this:

The model has borderline fit (CFI/TLI=0.99, RMSEA=0.07, SRMR=0.06), mostly because of the poor fear model. Dooming is quite poorly correlated with either a feeling of personal safety or life satisfaction. Worrying about climate change is not the same as depression or anxiety, but it is not a separate, well-founded fear either. Worrying about climate change is part of “dooming”, a specific tendency to worry about catastrophic, sometimes mutually exclusive future events.

[1] This is the same as if the climate item was just included as an indicator of the dooming factor: 0.74*1=0.74. I tried fitting the model that way and it returns this exact number. I should be able to freely estimate the climate item loading instead of fixing it to 1 as long as the variance of the latent climate anxiety factor is fixed to 1: however, when I do that, the model is not identified. This is indirect evidence that lavaan has problems telling the “dooming” and “climate anxiety” factors apart, but I admit that I hoped for a properly estimated latent correlation.

Climate anxiety is, as the NYT once put it, “a term that includes anger, worry and insecurity stemming from an awareness of a warming planet”. In other words, as strange as it may sound like, some people are stressed because of the weather in future. The terms appears to be gaining popularity (Google Trends data for the full data range):

This increase can mean two things:

1) people are starting to be genuinely more afraid of climate change, or

2) climate anxiety is increasingly part of the psychiatric symptom pool, in other words, the depressed or anxious people who always existed are increasingly shifting their worries from criminals, infections, social rejection, or the other things you can worry about to climate change. The general increase in negative affect in the West can amplify this effect by increasing the anxious/depressed population.

In line with this second possibility, lots of studies tend to find that climate anxiety is not specific but correlates with other psychiatric problems:

- Wullenkord et al 2021: r=0.25 with general anxiety and depressiveness, -0.1 with “relatedness satisfaction”, 0.27 with “relatedness frustration”, 0.29 with “competence frustration”

- Cosh et al 2024 (systematic review): r=0.17-0.7 with psychological distress, r=0.14-0.49 with depression, r=0.17-0.6 with clinical anxiety, r=0.22-0.46 with stress, r=0.17-0.3 with pathological worry

- Gago et al 2024 (meta-analysis): r=-0.296 with wellbeing

But these correlations are not nearly high enough to be able to say that being afraid of climate change is just a new way of being crazy. They are actually somewhat in line with a causal link even. “Depression” or “anxiety” is not the same as being afraid of climate change, but maybe one reason some people are more depressed or anxious is that they realize the looming climate catastrophe.

What bugs me about these studies is that even if people who are afraid of climate change are just crazy in a way, “depression” or “anxiety” doesn’t quite capture the expected nature of their craziness. Climate anxiety is a fear of a monumental negative change that affects all society, not just a feeling of personal inadequacy (anxiety), that the world is bad (depression), or that you are not feeling OK for whatever reason (wellbeing). To be able to say if it is a separate fear or just a general tendency for dooming about a future we would need a psychological test of dooming, not just a depression or anxiety inventory.

It is very important if climate anxiety is a separate fear or just dooming: this tells us how reasonable it is. A reasonable fear should be statistically independent from others. I’m pretty chill in general but I would be terrified in a Latin American prison: my fear would not be a reflection of my neuroticism, but my very rational evaluation of the dangers of being around hundreds of tattooed thugs. In the same vein, if a fear of climate change is well-founded, it should be roughly statistically independent of the fear of other catastrophic scenarios. If your evaluation of the facts drives you to be specifically afraid of climate change, it’s a genuine fear, even if you end up wrong. If, on the other hand, people who are afraid of climate change are also afraid of a Nazi takeover, epidemics, rogue AIs and nuclear war, this sounds more like a neurotic pathology. The world can only end in one way and very different roads lead to different catastrophes, so a rational doomer shouldn’t be equally afraid of all of them.

Fortunately, I could look at a large nationally representative dataset of Hungarians with excellent measures of dooming. Among other things, this dataset (N~7000, <10% missingness for any variable here) asks people to rate on a scale of 1 to 10 sixteen potential dangers facing the country. I omitted a redundant one („political radicals” and „radicals” were both on the list). As heterogeneous as these dangers are, ratings are strongly correlated with a mean value of close to 0.7:

Some people are just doomers. A single factor explains 70% of the total variance, already suggesting that being afraid of one particular abstract future danger is more indicative of someone’s character than a rational evaluation of threats. There are some patches of stronger correlations though. Some make sense (bankruptcy-unemployment, or pollution-habitat loss), but it’s curious to see that a fear of immigration and terrorism is somewhat antagonistic to a fear of climate change and pollution. My dataset doesn’t ask about political orientation but these sound like separate right-wing and left-wing worries.

A single-factor model has a borderline fit to the data (RMSEA=0.06, SRMR=0.055, CFI=0.99, TLI=0.99, using lavaan, the sem() function and a DWLS estimator as this is Likert data). This improves only a little (RMSEA=0.058, SRMR=0.052, others unchanged) if we allow two common-sense residual correlation between depopulation and aging and bankruptcy and unemployment. Little additional improvement (RMSEA=0.055, SRMR=0.049) is seen if pollution and habitat loss is allowed to correlate. There are nuances to individual fears, but it’s not a very bad model to say that climate anxiety is mostly just “being afraid of bad things in general”.

A more straightforward way to test whether climate anxiety is a separate construct is by modelling it as a separate factor. For this, I separated the four items about what I call “ecological anxiety” – climate change, natural catastrophes, pollution, and habitat loss – from the others (we can call those indicators of “dooming”) and modelled two latent variables which were allowed to correlate. This looks like this:

As you can see, “dooming” is modelled as one latent variable on which a fear of every non-ecological catastrophe loads, while “ecological anxiety” (ecR) contains a fear of climate change and its consequences. This model fits quite well (RMSEA=0.048, SRMR=0.043, CFI/TLI=0.99). The “Ecological anxiety” factor correlates at 0.89 with “Dooming”, showing that there is little evidence that when people say in a survey that they are afraid of the consequences of climate change, they are expressing a specific fear rather than general dooming about the future.

How is this “dooming” similar to other bad psychological issues like depression? The dataset has questionnaires about life satisfaction (WHO5), depression (BDI9) and stress (PSS10). A single-factor model fit the life satisfaction and depression questionnaires well, but the stress questionnaire is more complex, with at least two separate factors necessary, one for positively worded items (“I feel competent to solve my problems”) and one for the negatively worded ones (“I feel overwhelmed by stress”) which I called “competence” and “stress”. I put all of them in a single model to extract the latent correlations between the constructs each questionnaire measures. The model fits pretty well (RMSEA=0.039, SRMR=0.041, CFI=0.994, TLI=0.993). This is what the latent correlations look like:

“Life satisfaction, “depression” and “stress” seem to be somewhat iffy constructs, they are not measured in a pure way by these questionnaires, they have a lot of overlap. However, as in the previous literature ecological anxiety or general dooming about the future is NOT just these three things. The strongest correlation is with stress, but even that is rather weak. Statistically, climate anxiety is not the same as depression, general anxiety or stress. However, it is almost exactly the same as a fear of non-ecological catastrophes. People who have climate anxiety, at least in my dataset, are just doomers, not people who decided based on their best judgement that a fearsome climate disaster is coming.

The most straightforward way to test shared genetic causes is to test the association within pairs of monozygotic twins. Are richer twins less criminal than their own brothers? Are heterosexual members of the same MZ twin pair better off mentally than their homosexual siblings? These people are genetically identical so the genetic background causes are controlled for. If these were the causes, the association will go away. If, however, there is a causal relationship – poverty causes crime or non-heterosexuality causes poor mental health – the association should persist even within pairs.

There is a less straightforward way to do this though, which is actually the better one. This is the calculation of genetic and nonshared environmental correlations.

When two traits correlate and we have twin data, we can infer if the correlation is because of a genetic or an environmental cause. (Note that we are not talking about the similarity of twins, as in ordinary twin designs, but the correlation of two different traits, measured in samples of people with different degrees of relatedness). If the traits correlate because of their shared genetic backgrounds, the correlation is higher among MZ than DZ twins: if you share more genes, you share more of the traits. If it is shared environmental, the MZ and DZ trait correlations are equal but higher than the population correlation: it’s living in the same family that makes the traits similar. If it is nonshared environmental, the MZ and DZ trait correlations are equal, and also equal to the population correlation. In this latter case, but only then, we have excluded genetic confounding. If one trait is causal for another, their correlation should be equal within pairs of MZ twins and look nonshared environmental in a multivariate twin design. Robert Plomin on this observation in this awesome commentary to one of his own papers:

“However, it was not until the mid-1990s that it was realized that multivariate genetic analysis could be applied to the question of the causality of non-shared environmental effects controlling for genetics. It was not mentioned in the 1987 paper or its commentaries. It took three steps to get to this realization. The first step was to recognize that environmental measures could be treated as phenotypes in univariate quantitative genetic analyses, as discussed in an earlier section on the ‘nature of nurture’. The second step was to incorporate environmental measures in multivariate genetic analyses to investigate the aetiology of covariance between environmental measures and outcomes. As in most quantitative genetic research, this research highlighted the genetic aspect of these results, often finding substantial genetic contributions to the links between environmental measures and outcomes. The third step was to realise that these same multivariate genetic analyses of environmental measures and outcomes estimate the importance of non-shared environment on an outcome independent of genetics and shared environment, using all of the variance within and between identical and fraternal twins. Unlike other analyses of non-shared environment, structural equation model-fitting estimates non-shared environment as a latent variable free of measurement error if the standard assumption is made that error of measurement does not correlate across measures.”

Although Plomin says the two methods are equivalent, he doesn’t elaborate how one can turn one kind of estimate into the other although it’s important pretty straightforward. One can imagine the correlation of two traits – here, sexual orientation and mental distress – to originate from a pathway like this:

Both traits are caused by both genetic and environmental influences. As it happens, both the genetic and the environmental causes overlap somewhat which causes a correlation of the traits. (For simplicity, we model no shared environmental influences.) The total correlation is a sum of these paths:

We take the square root of heritabilities and environmentalities because we the path is the correlation, not the variance explained, between genes/environments and the traits.

Within pairs of MZ twins, the genetic path doesn’t operate, so the estimated correlation is just the environmental path:

If the correlation is all genetic, none of it remains within pairs of MZs. This would falsify any causal claims between the traits. One could convert the resulting rPOP and rMZ estimates to any other effect size to get estimates directly comparable to a normal twin control design. An rDZ estimate would be simply the mean of the MZ and the population estimate.

To turn the specific case of the minority stress hypothesis, I’m aware of four papers – Zietsch 2011, Zietsch 2012, and analyses of the UK TEDS and a Finnish twin study by Oginni et al, of which I linked the latest two – which use multivariate twin designs to estimate genetic confounding in the relationship between non-heterosexuality and mental health. The two Zietsch papers neatly provide everything to estimate these equations, and one even derives them. In the two Oginni papers, all we get is this information:

The diagonals are the heritabilities – 57% and 54% - and nonshared environmental determinations – 43% and 46% of same-sex attraction and psychological distress, respectively. The genetic and environmental covariances are reported in numbers that sum to 1: 0.79/0.71 and 0.21/0.29 – in other words, they tell us how much of the full correlation between the traits comes from genetic and environmental sources, but not how much that correlation is. Fortunately, this is reported elsewhere in the papers as 0.25 and 0.3. For example, plugging in the data for the right table we can get:

This is 0.69 but it’s on a weird scale because we didn’t know the proper genetic and environmental correlations, only covariance components that summed to 1. We can make sense of this number if we calculate the estimated MZ correlation from just the second path:

This is 0.26, again on an arbitrary scale. However, the ratio of the two correlations is informative: we can expect that 0.26 / 0.69 = 38% of the total population correlation will persist within MZ twin pairs, and 62% will go away because it was due to genetic confounding. This would mean a within-MZ correlation of 0.11, compared to a population correlation of 0.3. Most of the relationship between non-heterosexual orientation and mental distress is genetically confounded.

This is in line with the fact that MZ correlations are almost twice as high than DZ correlations (0.2 vs 0.13). The authors estimate something similar with a fancy MRDoC path model and get an estimate of the within-MZ effect which is slightly higher (0.19) but 0.11 is within the confidence interval. In the Discussion, the authors bank on the fact that these numbers are not 0 and have the gall to describe their findings like this:

“The causal influence of same-sex attraction on psychological distress demonstrated in the present study is consistent with prior evidence indicating that the phenotypic associations between same-sex attraction and adverse health outcomes are not confounded by correlated genetic and environmental influences”

which is the polar opposite of what their numbers show. Most of the association is in fact genetically confounded.

For the record, in the other Oginni paper, the population correlation is 0.25, of which 0.085 is expected to remain within MZs. In Zietsch 2011, the population correlation between non-heterosexuality and depression is 0.248, rMZ=0.02, in Zietsch 2012, rPOP=0.29, rMZ=0.14.

It is unusual to write a war analysis 1) without partisanship and 2) without being a war nerd, that is, focusing on which technical military capabilities each party has. In other words, it is unusual to write war analysis with common sense. In war, as anywhere else, we don’t know what will happen but there is a range of possible outcomes, each with some level of desirability and probability. It’s pointless to pursue outcomes which are very unlikely even if we want them very much.

Partisanship clouds our judgement and makes unlikely but desirable outcomes seem less out of touch. Common sense, on the other hand, dictates that we must sometimes be content with a less than perfect solution to a problem because with every attempt to get a better one we would be digging ourselves into a deeper hole. I think that – contrary to popular wisdom on all sides of the conflict – the least bad outcome for all parties in the Ukrainian war is immediate ceasefire. Here is my case for it.

Victory conditions and the escalation lottery

Great powers sometimes have ideological, grand strategy goals with their wars. It is certainly the case with this war. We will soon look at the geostrategic consequences of the ceasefire idea, but as the citizen of a small country in the neighborhood, my principal concern if this war is winnable.

What is “winning”? Early in the conflict (just before he decided to become a globalist) Russian blogger Anatoly Karlin posted this image (the Chatham House has similar analyses even now):

This is a nice summary of what each conclusion of the conflict would be perceived as, with the qualification that Ukrainian goals are maximalist: they call not for zero territorial cessation at a minimum. For Ukraine, even Karlin’s “decisive Ukrainian victory” is a pyrrhic Russian victory because it leaves Crimea with the enemy. Russian victory conditions are murkier but they don’t include completely occupying Ukraine. Russia annexed five Ukrainian provinces so keeping those is probably a minimum wish for them. These correspond to the “pyrrhic Russian victory” scenario on the map, although the Russians don’t control the provinces outside Crimea completely.

I looked at Metaculus (this, this and this question) and created a colormap of Ukraine showing the probability of community-predicted Russian control for 2027-2030 (depending on the question but mostly 2027):

To make the extrapolations nicer, I copied some predicted probabilities to adjacent cities (for example, 0.01 predicted for Kiev to Zhytomir or 0.04 predicted for Mikolaiv to Kryvyi Rih) and added 1 to Russian-occupied cities firmly behind the frontline such as Mariupol. (There is no Metaculus prediction for Mariupol but the Russians spent enough of their money on reconstruction to take this as a strong bet.) The thick isobar line runs at 50% predicted Russian control and very close to the current front. If you look at the raw numbers on Metaculus, they drop off rather sharply on the sides of this line: the community predicts a frozen conflict. The worst outcome would be to keep fighting roughly along these lines for several more years. This way, both Russian and Ukrainian soldiers would die and EU/US money would be wasted for nothing. Note that even if you don’t mind Russians dying you are still sacrificing Ukrainians for this (and vice versa). It only makes sense to keep fighting if you expect a better outcome for your side. There are ways to achieve this, but they come with significant risks.

For Ukraine, the only reasonable way to victory would be to get NATO troops involved. I think this explains the strategically irrational choices they make by invading Russian territory and provoking the Russians with humiliating but strategically irrelevant rocket attacks on their home turf. You can think of it as an escalation lottery, if they poke the Russians hard enough they might strike back with something (like a tactical nuke) that provokes NATO to send troops. Just like the real lottery, it’s a very stupid game though because they must continuously swallow costs for an uncertain payout. Also, unlike in the real lottery, they can win too hard: getting nuked (or hit hard by any kind of Russian retaliation) is not great, and NATO involvement can mean nuclear Armageddon. The Ukrainian side is fighting for an unlikely reward with a strategy which can mean catastrophe just as much as success. Continued fighting is a bad deal for the Ukrainians (not to mention their Western allies who shore up the costs).

The Russians are slowly gaining land, and the Metaculus map actually gives them about 50-60% to conquer Sloviansk and Kramatorsk, the last two major cities in Donetsk oblast which they claim. But the same map gives them almost no chance to conquer the capitals of two other claimed oblasts (Zaporizhzhia and Kherson). A continued war may be slightly more beneficial for the Russians, but even they would have to absorb heavy losses for very limited gains and risk NATO involvement. Because they already have the juicy parts of the provinces they annexed (the Black Sea coast and the main cities Mariupol, Donetsk and Luhansk, including reasonable buffer zones), a status quo ceasefire should be an acceptable outcome for them.

From a practical point of view, neither side has much to gain from fighting, but heavy losses are guaranteed. It is pointless to wage war without a reasonable expectation of victory. A few weeks ago there was a huge scandal in Hungary when a government official said that we would have been smart enough not to stand up against the Russians like the Ukrainians did. He was of course completely correct, and not just in theory. When the Russians last invaded us in 1956, we stood down after minimal fighting, which kept the country intact for a relatively successful Communist period and ultimately a democratic transition when the climate was right and the Russians left on their own. Destroying the country in a futile attempt to resist and get the same occupation would have been stupid.

But what about consequences? What about the sanctity of borders? What about normalizing gaining land by invasion? Won’t the Russians steamroll the Baltics, Poland and Western Europe after winning in Ukraine?

Invasions and frozen conflicts

We have now gotten to the geopolitics of the war, which is normally beyond the pay grade of small state citizens like me. Arguments against the ceasefire are that it 1) sets a bad precedent, normalizing changing borders by invasions and 2) fails to resolve underlying issues (such as Russian militarism or the eastward expansion of NATO). I think these are mostly moot, but let’s take a look at them.

One argument why the war must continue is that otherwise we are setting a precedent for taking territory by invasion. If Russia wins, no matter the cost, any country will be emboldened to launch invasions against neighbors because the precedent has been set that the international community will be OK with the status quo you establish by arms. This argument effectively makes Ukrainian victory more valuable and thus worth pursuing even at lower odds of victory.

This argument needs to be taken seriously. Invasions used to be common but became rare, in part because they have a flatter cost-benefit ratio (wars became expensive and simply capturing territory became less lucrative) and in part because of the dedication of other nations to condemn them. (Read Richard Hanania’s essay on this.) Preventing warfare is important so anything that does it successfully needs to be given strong consideration.

However, the argument assumes that, at least after World War 2, there is a line drawn in the sand: these are the final borders, nobody gets to invade to change them, and Russian success in Ukraine would invalidate a 80-year precedent on this question. My problem is that no such precedent exists: lots of nations, including the United States itself, did thing similar to what Russia is doing, even after World War 2.

The most successful invasion would be if the invader could get the international community to legally recognize the territory he occupied. Even this has almost happened: in 2019 the US (although not “the international community”) did recognize the Golan Heights, a Syrian territory occupied by Israel in 1967. A second close example is Kosovo, which broke away unilaterally from Serbia after a war and NATO intervention in 1998-1999, supported by the US. Kosovo is widely recognized as a country internationally and the original independence movement would have been clearly suppressed by Serbia without NATO intervention, so the independence of Kosovo can be considered to be a second case of territorial change resulting from military invasion. Interestingly, the EU countries not recognizing Kosovo are those with large ethnic blocks who could break away if we took the Kosovo precedent too seriously: Spain (with the Catalans and Basques), Slovakia and Romania (both with large Hungarian ethnic blocks).

I don’t think Russia hopes to gain international recognition for its Crimea and Donbass acquisitions. It would be happy with de facto control with other countries tacitly accepting its control. If we lower the bar to this level, we find even more examples of the international community being basically OK with invasions:

- Northern Cyprus: perhaps the most egregious example. Turkey invaded Cyprus in 1974 and created a breakaway state in the north of the island which nobody else recognizes. Not much happened to Turkey because of this. Although it was already a NATO member, it started EU (technically EEC, the predecessor organization) accession talks in 1987. In 1974, the US needed strong allies in the Eastern Mediterranean and this was more important than the borders of Cyprus.

- Nagorno-Karabakh (Artsakh): this was an Armenian-majority autonomous province within the Azerbaijan SSR of the Soviet Union. During the fall of the Soviet Union, the population voted for independence and joining Armenia, which the Azeris opposed. War broke out and the Armenians won, so Artsakh remained a de-facto independent (although not internationally recognized) second little Armenia until Azerbaijan recently re-took the whole area in an invasion. The international community did not have strong opinions about either the Armenian or the Azeri invasions. During its independence, Artsakh had talks and some diplomatic relations with the US and the international reaction to the 2020 was mostly just to call for ceasefire and as few causalities as possible.

- Rojava: officially the Autonomous Administration of North an East Syria, a Kurdish quasi-state which emerged during the Syrian Civil War. Kurds mostly live in Turkey, Syria, Iraq and Iran, they have long wanted to have an independent state and they have some history of self-administration in all four countries. When the Syrian Civil War broke out, the Syrian government under Assad (internationally recognized, but opposed by the US) had its hands full fighting jihadists and other rebels so they largely left the Kurds to their own devices. The United States, on the other hand, supported them militarily. Although the war is mostly over (or so it seemed when I started writing this), Syria is still divided, and the Kurdish territories are de facto independent. The US keeps troops stationed in Syria, pillaging the oil fields there while keeping the official Syrian government under heavy sanctions. The international community was also OK with Turkey invading Kurdish-controlled Syrian territory in 2016-2019, keeping troops and running civilian life there.

My goal with these example is not to say that Russia is no worse than Armenia, Serbia, Turkey, or the United States, or any other country. I’m also not trying to compare, relativize or normalize these invasions or wars. My point is that there is no red line that Russia crossed in 2022, let alone 2014: there are numerous late 20^th century examples of border changes resulting from military action with little reaction from the international community. The “no invasion” argument is selectively enforced depending on whether the United States cares about a certain incident. Regardless of what we think about territory-gaining invasions, strong countries will probably forever keep invading weaker ones to create buffer zones and achieve other geopolitical goals.

A version of the “stop the invasion” argument is that if Russia is allowed to take Ukraine, it won’t stop there: just like when Hitler invaded Poland and eventually the Soviet Union after getting away with annexing Austria and the Sudetenland, a successful Ukrainian adventure will embolden Putin to take the Baltics, maybe the former Warsaw Pact countries, and eventually maybe even move on to the Atlantic. This is certainly possible, in the way a world-ending asteroid strike or a Black Death-style new pandemic is possible: there is a future universe where Russia gets strong enough to steamroll all of Europe. But it is irrational to overprepare for low-probability events. Current, stated Russian goals call for the “demilitarization” of Ukraine and limited occupation. Betting markets and the battlefield situation say they are not capable for much more. If we are afraid that the Russians will keep expanding west in the future for some reason, we should build serious militaries in Europe (now that the normal way of preventing wars, economic codependence, is gone for a long time). That doesn’t cost lives, and the monetary costs are spread across many years. Feeding more money and Ukrainian recruits to the war machine is like selling all your investments to build an asteroid-proof shelter.

What about the second argument against a frozen conflict, that it doesn’t solve the underlying issue and only lets the parties rearm for round 2? This argument always baffled me: how is a possible war in the future not obviously better than a real war right now? Besides, frozen conflicts have a good track record for preserving peace. Of the 9 frozen conflicts I can think of (China/Taiwan, the Koreas, Artsakh, Northern Cyprus, Transnitria, Abkhazia, Artsakh, Kashmir, Western Sahara) only two (Artsakh and Abkhazia) resulted in renewed wars, and in the former case there was a generation of peace. When the conditions were right, sometimes both sides of the armistice line developed into reasonable nice countries, as in China/Taiwan and in Cyprus. After a frozen conflict, Ukraine would still have most of its territory intact, it would have a port, and access to the EU markets. Russia would have the territories it can claim to be ethnically Russian, a defensible land bridge to Crimea, and maybe some economic re-normalization with the EU which would also be very important for the latter. Absent Ukrainian irredentism, there would be little immediate motivation to restart the conflict for a long time.

War profiteers

At the end of the post, it’s important to say that there are a few entities who would indeed benefit from a continued conflict. Unfortunately I think they have much more influence than those whom the war actually affects.

The most obvious of these is the military-industrial complex. The war is a great demonstration for all militaries in the world, they get to see in real life what the modern battlefield looks like and how their own equipment would perform. Weapon systems that work well generate sales. Military stock sent to Ukraine by Western militaries must be replenished. Politicians, career soldiers and industrialists in both countries make a lot of money on the conflict so they have little motivation to stop.

A less obvious one is the United States. The war is very cheap for the US: it is not affected by the loss of the Russian market and Russian resources, it sends old military stock and the money spent on replacements are a stimulus to the economy. There are no American soldiers dying in the war. On the other hand, the benefits from the war are huge. Weakening Russia is an obvious win, but not the only one. Maybe this is not obvious to Americans but beside Russia and China the EU is also a geopolitical rival for the US. The US has both raw materials and human capital, but the EU only has the latter: many raw materials like energy used to come from Russia which is now gone. The war caused an economic crisis in Europe, European industries are not profitable anymore compared to the Americans and the Chinese. The EU has little hope to become a competitor for the US either economically or militarily. Cynically, the US weakened two of its rivals at once by pretending to be allied with one of them. If you are American, you want your country to be geopolitically dominant and don’t care about the cost of this on Europe, Ukraine or Russia, you actually have a good case against ceasefire in Ukraine.