Second year review

This second year of my blogging career I managed to pen 24 posts, which comes close to the fortnightly schedule I vaguely aimed for.

Five of these post alone were on the topic of Covid19. Somehow I managed to not say anything really stupid in any of them. In fact my very initial, March 9th, rough estimate of 75.000 to 300.000 deaths in Germany still looks quite good … unfortunately.

My post on the imminence of synthetic porn also came out just in time. Now there are generative models like DALL-E [1] and on reddit there was an amateur project that was quite advanced.

I actually managed to do some data analyses with “Ashkenazi grit in chess” and the post on demographic change in France and Germany. My “random thoughts” kind of articles have dropped that moniker and just become less data heavy explorations of different interesting topics.

I am still not sure about the future of the blog. At the moment it seems like other more technical interests are making inroads on my time. Not just my time to write and analyse, but also my time to think about topics and to come up with new ideas what to write about. So it seems likely that my publishing frequency will drop again. On the other hand, there is still a steady flow of new ideas to write about, so I hope the drop will not take me too much from one post per two or three weeks.

[1] https://openai.com/blog/dall-e/

Covid again – the UK mutation

So, there is a new strain, originating in the UK, which is „70% more contagious“. Don’t even ask what that’s supposed to mean. The salient fact is that cases in the UK are on course to double every ten days now, even under lockdown conditions.

Source: Very extensive lesswrong-post

Basically, in this figure we see how in October and November the percentage of the new strain amongst sequenced strains rose super-exponentially to reach 10% in mid November. Then the figure mercifully ends.

Source: Same

But here, in number-of-cases figure for a bunch of random countries we see that the exponential kept going up until it burst out of the chest cavity of the UK like an alien. Meaning, that while other strains, aka „normal Covid“, were nicely going down, this strain took over and started dominating the case numbers. Because it is still not even close to 90% of all cases, and because „normal Covid“ is likely still going down, the number of all Covid cases is still not rising with the speed with which „new Covid“ is spreading. In the last three days the number of cases went from 400 to 500 which corresponds roughly to doubling every ten days.

If „new Covid“ is doubling every 6 or 7 days under lockdown conditions … well, let’s calculate some example numbers. Let’s say 300 of the 500 cases per million in the UK on December, 23rd are of the new strain. Of course, Christmas and new lockdown rules will make everything worse or better, but if we just extrapolate, 10 doublings will take just 60-70 days and take the number of cases to 30% of the population.

As of now, in the UK a record number of patients is already being treated for Covid. So frankly, this looks very bad.

Given that the new strain has been spreading for 2,3 month in the UK, it is obvious that it is already everywhere else in Europe and in the US. And in fact is has been detected everywhere people bothered looking for it. How much are we behind? 5 doublings aka a month? This will become clear when local outbreaks just start exploding. The first local exponential can than be used to extrapolate for the whole country. Given the way exponentials work the numbers won’t be horribly off even if you ignore all the other local seeds of the new strain.

Of course Christmas is going to spoil all the data for at least another week probably two. But within a month we will see whether the whole Covid saga will end with a big bang, where just when the vaccines are starting to be doled out, one last big wave gets most of the population anyway.

Antisemitism is the sodium-glutamate of conspiracy theories

The whole defining feature of conspiracy theories is that there is a cabal somewhere that pulls the strings. A smallish group of people with nefarious intentions. For Alex Jones these are „the globalists“, for David Icke more daringly „the lizard people“. The whole empirical evidence therefore often amounts to a smallish group of people popping up in different contexts, that can be pulled together into a story of purposeful action.

Because of the verbal IQ advantage jews are overrepresented at basically all higher echelons of decision making. Because the reason for the over representation cannot be mentioned in polite society, even the over representation cannot be mentioned in polite society. That makes them the perfect ingredient to conspiracy theories: „Another jew! That cannot possibly be a coincidence!“ Well, because it isn’t.

Both Alex Jones and David Icke run into the conspiracy theorists dilemma: If they do not want to increase the likelihood of getting cancelled considerably, they cannot connect the cabals to the number one choice of evil cabals – the jews. Of course their followers have no such compunction, as a look into youtube comments will make abundantly clear. Antisemitism is the sodium-glutamate of conspiracy theories – every conspiracy theory just works better if you add a little.

Demographic Change in Germany

While the French statistical institute INSEE publishes a very comprehensive list of names given to kids each year, the same data for other countries is hard to come by. In Germany, we have the strange situation that there is no official statistic, because there „is no legal basis“ for the collection of this data, but some guy collects an almost complete list of given names each year as some kind of private project.

Now, this guy doesn’t provide anything like a complete dataset – though he seems to fuel the yearly „most popular baby name“-articles that pop up everywhere. But since 2015 every year not only the 500 most popular baby names in Germany are published, but also the 10 most popular Turkish-Arabic names in Germany and their position on the overall list.

From the position of these names on the overall list, it is not immediately possible to derive an estimate of the percentage of kids with Turkish or Arabic names. Consider:
If the distribution of names is such that neighboring names on the list have almost the same frequency, the first name of a – say – 20% minority will pop up very far down the list, when this gentle slope has reached 0.2 of the frequency of the first name (assuming the same distribution for both majority and minority).
If, however, the distribution drops very quickly, for example as a power law, 1, 0.5, 0.33, 0.25, 0.20, … the first minority name would already appear at position 5 or 6.

In this post I will use the distribution of the French names to estimate the percentage of Turkish or Arabic names in Germany. This is sloppy because the French names do not comprise a single ethnic group any more, so I should probably remove non-French names for a better estimate. But also in that case the distributions of French names, of German names and of Turkish and Arabic names in Germany probably differ quite a bit, so let’s bite the bullet and keep in mind that these estimates are very rough.

Basically, what we will do, is look up the frequency of the names in the French distribution at the positions in the frequency ranking that are occupied by Arab/Turkish names in the German list. The sum of the frequencies will be compared to the sum of the frequencies of the first 10 (German) names. This gives us a straight-forward comparison of the frequency of Arab/Turkish names with the frequency of German names in Germany.

If we assume that both the German names and the Turkish-Arabic names are distributed just like the French names, we get the following estimates for the percentage of boys with Turkish-Arabic names:
2015 – 19.0%
2016 – 20.9%
2017 – 21.7%
2018 – 21.8%
2019 – 23.8%

If I assume the different French distributions of different recent years I get a standard error of around 0.7%, so which French distribution I choose doesn’t seem to impact the result too much.

However, the results for the girls differ significantly.
2015 – 19.3%
2016 – 18.9%
2017 – 19.2%
2018 – 18.6%
2019 – 20.0%

It might be the case that the overlap of Turkish and Arabic names is larger among boys and therefore the female names do not in the same way reflect the large influx of Arabs after 2015. This also opens up the possibility that even the estimates for the boys are underestimates, because Arabic names and Turkish names might be two largely separate distributions.

But given that the last official statistic gave slightly less than 10% muslim babies in 2004 and given that in France the percentage roughly doubled within 15 years, an estimate of 20-25% doesn’t seem absurd.

Update on Covid19

Truth to be told, I have gotten pretty bored with Covid. But things are picking up again, so maybe it makes sense to write down some idle speculation. Interestingly, we have mostly learned so far that we don’t have much of a clue of what’s going on. One of the reasons seems to be, that there isn’t really a binary distinction between „still vulnerable“ and „immune“. Instead there seems to be a spectrum of immunity.

If you just had a full blown infection, you are probably 99.9% immune. Reinfection might happen occasionally, but seems to be quite rare. But if you did not have a full blown infection, you might still have been exposed to the virus in small doses. This exposure already leads to some protection and when the day comes and you get enough viral particles up your nose to start the illness, you will have a milder case than without earlier exposure. If you have never been exposed to Covid19, you might have some protection because of earlier infections with other Corona-viruses. Karl Friston calls these phenomena „immunological dark matter“.

In the current second wave the case fatality rate is significantly lower than in spring. There are several possible reasons: One is the „dry tinder“ effect, which postulates that there is a certain number of old and frail people who are ready to go and are just waiting for the next severe flu. If the flu season has been mild, dry tinder accumulates and when Covid comes along it burns up. Now, we have burned through the tinder and death rates will stay low. In the summer lull this was a popular argument by anti-lockdown activist. They combined it with the claim that we had basically already reached herd immunity.

Unfortunately full herd immunity is unlikely for most places, as even cities like Madrid that have been hit hard in the first wave have strong second waves now. And without herd immunity it seems very unlikely that we have actually burned through more the 20-30% of the dry tinder. More likely reasons for the lower case fatality rate is increased testing, partial immunity by minor exposure and better treatment protocols.

The increased testing part is a bit scary, because it implies that currently we are slowing the growth rate by testing, diagnosing, isolating and tracking. This seems to be a benefit that will run out if the number of cases becomes too high. So at a certain point we might see an acceleration when testing and isolating happens for a much smaller fraction of cases than right now.

Karl Friston predicts that the second wave will kill far fewer people. Given that he is some 30-50 IQ points smarter than the average epidemiologist, he is well worth listening to. Unfortunately this might well hold for his native UK, but it obviously doesn’t hold for Eastern Europe and I doubt it automatically holds for Germany. Incompetence in the spring might be the best protection in the fall.

Developmental feedback disruption in modern environments

It was always clear to me that shortsightedness cannot be genetic despite being highly heritable. It is simply too disadvantageous in a pre-glasses environment. In recent years is has become clear that lack of exposure to sunlight is one of the culprits, though near-work such as reading is also likely to blame. Myopia is maybe the most common example of our modern lifestyle disrupting developmental feedback loops and leading to developmental disorders and malformations, but there are many others.

Allergies and asthma is another example. It has long been known, that these disorders are a result of lack of exposure to … dirt, basically.

Another example is flat feet. These are due to shoes preventing foot muscles from developing the strength to support a normal arch.

Now scientist seem to start looking into the development of jaws and teeth. Crooked teeth are a uniquely modern trend. Apparently in skulls older than 500 years, teeth almost always fit perfectly together. Our modern diet leads to an underdevelopment of especially the upper jaw, which means that teeth crowd into too little space. As a result teeth grow crooked, wisdom teeth need to be removed, teeth that just wont fit need to be pulled and the rest straightened by braces.

These stories set me thinking. What more developmental feedback loops might be disrupted by our modern environment? In which ways is our mental development disrupted, which role does developmental disruption play in anorexia, depression, ADHD, autism, etc?

It also gives me a different perspective on mutational load. Many of our developmental feedback loops are probably much more resistant to random mutations than to massive disruptions of the lifestyle we evolved to lead. Hopefully, a healthcare of the future incorporates these insights into preventive care that stops this kind of damage from occurring in childhood and teenage years.

Urbanisation and assortative mating

There are two dysgenic trends currently operating:
Populations with lower IQ have higher fertility rates. This is the main reason, why the average IQ is lower whenever a new version of the IQ-database is published [1].
Within almost all populations the fertility patterns are dysgenic because smarter people (mostly educated women) are having fewer kids.

Over time this should lead to exponentially fewer very high IQ people. However, this doesn’t seem to be happening just yet. Mathematics for example is enjoying a golden age, with century old problems being solved. The main reason for this is likely increased assortative mating.

Assortative mating is currently increasing mostly due to urbanization. The more ambitious and more capable people are drawn to the cities, often to study and ultimately to work, marry and have (too few) kids. The same mechanism also holds between countries with high IQ immigration, i.e the more ambitious and capable people from the global south are drawn to the economic powerhouses in the global north.
Additionally social segregation ensures that people marry within the same class. Via gentrification and similar dynamics, more affluent people take over or maintain a grip on the nicer quarters.
This is also ultimately a result of urbanization. If you are rich in a village, you live in the biggest house, but if you are rich in a city you live among your peers, spatially segregated from the lower classes.

One impressive example for the effect of these mechanisms is the intellectual output of Budapest’s jews in the first half of the 20th century. Before the Holocaust roughly a quarter of 800,000 Hungarian Jews lived in Budapest. These 200,000 people produced a handful of super geniuses that together among many other things build the first atomic bomb [2]. It is certainly plausible that stronger assortative mating and a certain amount of self-selection of city dwellers raised the number of geniuses in Budapest even beyond what one could expect from 200.000 people from a group with an average IQ of probably around 110.

Of course in the medium term covering our need for technological and scientific geniuses via increased assortative mating is not sustainably. Let’s hope it lasts until we engineer our way out of the dysgenic dilemma.

[1] https://viewoniq.org
[2] https://slatestarcodex.com/2017/05/26/the-atomic-bomb-considered-as-hungarian-high-school-science-fair-project/

Cad-Dad-Theory

One of the basic HBD-theories is the Cad-Dad-theory. The general idea is that in societies where women are dependent on a male provider to raise children successfully to adulthood, people evolve over time to have stable longterm relationships. Generally it is evolutionarily advantageous for men to just move on to the next fertile woman once a lover is pregnant, but this is not a behavior favored by evolution if it means your children will die. So men evolve to be dads.

This is supposed to have happened in European peoples because for a couple of thousand years plowing was work that required a man’s strength, so women and their children generally depended on a male provider. In Africa however, horticulture was and is the norm, i.e. cultivating gardens with a hoe, not a plow. This was and still is to a large degree women’s work. So in Africa women had the ability to provide for their children on their own. According to the theory, African men consequently stayed „cads“. Random internet dictionary: „an ill-bred man, especially one who behaves in a dishonorable or irresponsible way towards women.

This theory is then used to explain the high rates of single motherhood in African Americans, which is somewhere in the vicinity of 75% and the low rate of marriage. Now for all I know, this theory might well be completely correct. But in this blogpost I want to point out, that African Americans are certainly not a good choice to prove the correctness of the theory or gauge the size of this trait difference, for two simple reasons:

As we know, African Americans average 15 IQ points below the white mean and have the correspondingly lower socio-economic status. If you would control for these factors, the difference between whites and blacks in single motherhood or stable relationships would certainly shrink significantly. The difference attributable to IQ and SES differences cannot at the same time be attributed to some evolved character trait.

Additionally, and maybe even more importantly, the sex ratio of the African American community is severely skewed. This is mostly due to the fact the young African American men are quite likely to get shot or to go to jail (for example for shooting another African American man). In the fertile age range the ratio of women to men in the African American community is something like 130:100. This oversupply of women leads to significantly worse treatment of women and less stable relationships as has been shown for example by using the local variation in sex ratio in Chinese municipalities.

If you control for these two factors I wonder how much of the discrepancy would still remain.

Higher variance in men

In this blog post I want to discuss the observation that got Harvard president Larry Summers into trouble. Summers mentioned in a 2005 talk that men generally have a higher variance in cognitive ability and opined that this might explain the under representation of women in top level science. This led to intense negative press and was likely a factor in his resignation in the following year.

In mathematical ability this ratio of standard deviations for boys and girls is generally around 1.2. This can be observed in the math SAT (where you have to correct for the fact that more girls take the test) and in almost all countries that participate in PISA. This seems like a small difference but it quickly has a very noticeable tail effect. For example, if we put the mean at 100 and the female standard deviation at 15 to make the numbers IQ-like, that going two standard deviations beyond the mean puts girls at 130 and boys at 136. Because an IQ of 130 is roughly 1:50 but an IQ of 136 is roughly 1:100, we can expect to see twice as many boys as girls at the ability level around 136.

Of course being right didn’t help Larry Summers any.

But higher variance for men in all traits that are relevant for fitness also makes evolutionary sense. To see that, it is easiest to instead look at women. Higher variance in this case means that dice with more pips are thrown for fitness relevant traits. So if women had high variance in fitness relevant traits they would risk losing out, but also have the chance to win big. Unfortunately for women that is a bad gamble, because the number of kids they can have in a lifetime is limited by nature. Losing would mean 0 kids, but winning might mean 12 kids instead of 8 (in an environment long ago with nasty childhood mortality on top).

For men this gamble makes more sense, because they can have up to 1000 kids! Which much more nicely balances the chance to have 0. So if there is a way for evolution to increase the standard deviation in men for fitness relevant traits, we should expect to see this.

And in fact there is such a way! The X chromosome only occurs once in a male human cell, but twice in female cells. To handle the superfluous X chromosome, female cells randomly inactivate one of the X chromosomes during embryogenesis. This happens early but not early enough that women are not mosaics of different activated X chromosomes. So the traits encoded on the X chromosomes are averaged over two Xs in women, and are not averaged at all in men!

This increases the standard deviation for men of all traits encoded on the X chromosome compared to women. Unsurprisingly, the X chromosome is found to be enriched for the most fitness relevant traits like height and intelligence.

So we do have the observation, an evolutionary reason and a biological mechanism!

But we are already used to even the most clear cut fact in IQ research being extremely controversial.

Lies by omission in IQ matters

When newspapers or generally mainstream media talk IQ, which occasionally they feel compelled to do, the reporting is often less than complete. You might term it desperate if you don’t want to call it dishonest. After all this is a topic where just mentioning an unequivocal fact might have very negative consequences on your career as a journalist.

In this post I want to point out several types of „lies by omission“ that one should look out for in these circumstances.

Trick number one: Compare childhood IQ, not adult IQ.

Childhood IQ is much more malleable than adult IQ, so this is a way to circumvent the fact that in the long term IQ has almost zero shared environment effect [1]. This trick can be used to show the efficacy of interventions (just don’t ever follow up on the kids) or to prove that disparities in abilities are influenced by some environmental difference. But it goes further than that. Some ethnic disparities in IQ grow with age. For Arab kids this is called the „Simber effect“ [2]. In some studies this effect also shows up for the black-white gap. So by looking only at children, you can effectively hide a big part of an IQ gap that explains disparities in income and other life outcomes.

Trick number two: Ignore the g-factor.

The g-factor is the part of IQ that is both the heritable part and the predictive part. The rest of IQ is basically a measurement problem, which among other things, leads to the Flynn effect. The Flynn effect, the rise of IQ scores observed in many countries in the last century, does not increase g. The increases for the different subtests are anti-correlated with how predictive they are of the g-factor. By omitting these little facts, the Flynn effect can be used to invalidate all kind of observations about heritability and predictiveness of IQ and of the relative permanence of IQ gaps between groups.

Trick number three: Don’t mention the sample size or other sample attributes.

Social science is full of small scale studies that do not replicate. Especially small scale studies with politically expedient results. So there is always something to cite, to prove your point of view. Here is the rub, the main IQ results are extremely robust and often have massive sample sizes. The gap between white and black Americans for example is estimated by a meta study with N=6.000.000! [3] So studies that „refute“ such a result with sample sizes of a few hundred or maybe even just a few dozens should sensibly dismissed out of hand. Especially because these small studies often suffer from egregious sampling problems.

[1] Wilson effect
https://pubmed.ncbi.nlm.nih.gov/23919982/

[2] Simber effect
https://www.researchgate.net/publication/320310317_Understanding_the_Simber_Effect_Why_is_the_age-dependent_increase_in_children’s_cognitive_ability_smaller_in_Arab_countries_than_in_Britain

[3] Ethnic differences
https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1744-6570.2001.tb00094.x