How does selective breeding and 'return to the median' work

I don’t know if return to the median is the correct term for what I mean. I have heard of it before regarding selective breeding, how even if you select for certain traits there is a high chance that children who do not possess them will outnumber those who do and you go back to where you started.

Example: (These stats may not be 100% correct, this is based on memory) the chances of 2 people with an IQ of 160 having a kid with an IQ of 160 is only about 1 in 44. That is far higher than the odds of people with IQs of 100 having a kid that high (I assume that is 50000:1 or higher, I don’t recall).

Anyway, there is a very good chance that people whose IQ is 4 standard deviations to the right will not have a kid smarter than themselves.

I’ve asked in the past about experiments where humans were selectively bred for certain traits. One is bred for emotional health/well being, one is bred for IQ, one is bred for aggression and physical resilience, etc. I assume you just start with a large pool of people and select the top 1%, or 5% (I don’t know what % you use) and interbreed them, then do the same with their kids and the same with their kids, etc. Pick the top % who have the trait and breed them generation after generation.

At first you can have people whose traits are maybe 2-3 SD to the right of the median in the selected traits (the top 1% of the starting group being highly over-represented in the bred groups), but after that doesn’t ‘return to the median’ (or whatever it is called) make it very hard to keep pushing the limits? If you create a human race where the average IQ is 160 in 3 generations of selective breeding, then you have less than a 2% chance that any kid born will even match that existing level, let alone exceed it. I’d assume the odds keep getting lower and lower. What are the odds that 2 people whose IQs are 180 will have a kid whose IQ is higher than 180? Several hundred to one?

Which brings me to my question about the updated Cosmos series with Neil Degrasse Tyson. He talked about how many breeds of dogs only came into existence in the last hundred years or so. Since a dog’s generation is maybe 2 years, that gives 50+ generations.

How did the dog breeders manage to create such a wide diversity of dogs without the dogs just becoming more and more ‘normal’ and returning back to their original nature (the same way that less than 2% of people with IQs of 160 will have a kid whose IQ is that high)?

The genetic variance between 2 humans who are at extremes of various traits (emotional health, IQ, etc) doesn’t seem that big. You still have 99.9% similarity between them. So after a few generations does the DNA change, maybe it is only 99.8% similar between the divergent groups and that is how the changes take hold and remain?

Like, are there 2 stages to selective breeding:

  1. Selecting those who are very genetically prone to the traits in question, but whose DNA is not noticeably different from others who do not possess those traits
  2. DNA changes to cement those new changes, creating a new ‘median’

So if you created a race of happy, emotionally healthy humans via selective breeding would you eventually have a situation where the new median is maybe 2 standard deviations to the right of non-bred humans? How many generations would that take?

The term is “Regression to the mean” and it’s not just a biology thing, it’s a statistic thing in general (but note the misconceptions/ fallacies section)

By picking the ones that conformed to their wishes and only breeding those.

Regression to the mean applies to a single generation in a “randomized” population and to complex traits. If, in a population with mostly brown eyed people you bred two with blue eyes, all the kids would have blue eyes. And if vice versa you’d still have 75 % offspring with brown. No regression towards the mean. (If we pretend there are only blue and brown eyes and ignore all the genes we don’t mention in high school biology.)

If you pick only those with 160 IQ to breed, the first generation will only have a small number of the super-intelligent, but they will have more super-intelligent than the population as a whole, and more of the highly intelligent, and intelligent, and more of the “average”. Pick only the best from this new population and the third generation will be even “better”.

Of course, for this to be successful, we have to prevent those *not *selected from breeding. Before you know it we are in Boys-From-Brazil territory.

That depends on what you mean by successful. If all I want to do is create a sub-group of humanity with unusual traits, it doesn’t matter to me that the rejects are breeding somewhere outside my empire of intelligence.

Probably the most famous study of selective breeding for mammalian behavior is the domestication of the silver fox.

About 5% of males and 20% of females were allowed to breed.

The regression to the mean refers to the mean of the breeding population. If a new splinter population is created through selective breeding, reproductive isolation, runaway sexual selection, etc. then a new mean will be created (e.g. founder effect).

As an aside, and I don’t know if this still has much pull in modern thinking, but one idea is that humans basically self-domesticated. This partially explains human neoteny and the increased willingness to cooperate with others, lessened aggressiveness compared to primate ancestors, etc.

One can only guess at what selective breeding could do for modern humans. There are physical limits. To the great consternation of mothers, we’re about maxed out for skull size vs. birth canal circumference. People who are 7-8 feet tall have a lot of problems with their feet, knees, hearts, etc. It’s interesting to wonder if you could breed passive cow-like humans (with us being the auroch). A big hit for futuristic totalitarian dystopias, I’m sure.

Selective breeding really does not depend on new genes coming into existence.

I think the focus on a trait like intelligence is maybe making the discussion more complicated than it needs to be, because we don’t even know the extent to which genetics controls intelligence.

Go back to basic genetics. A gene codes for something like a protein product that produces a result in the final organism (the phenotype). Genes in a population exist in multiple versions - alleles. The paired nature of DNA means that an individual actually has two genes in most locations and it’s possible to have two different alleles in the same individual.

But selective breeding looks at a whole population. There are many different alleles in the population - it could be dozens or hundreds. A natural population has a certain ratio of those alleles that mix around and generally reach some equilibrium. If you want allele #23 in your in your final breed, you can find two individuals in the population who have that and breed them. At least some of their offspring end up with 23. Eventually, you can get them to “breed true” where their DNA contains only allele 23 and nothing else.

For a single gene, you could accomplish this in a single generation. Two parents that are 23-? and 23-? will produce 1/4 of their offspring that are 23-23. If you can tell those apart, then you’re done. Sometimes 23-? and 23-23 look identical (have the same phenotype) and that complicates things, because it means 1/4 of the offspring will be ?-?. That’s one reason why establishing a stable breed is more difficult in real life than a single generation.

Of course, doing anything complicated also involves multiple genes. You don’t go from a Chihuahua to a Great Dane with a single gene. The more genes you need to combine, the more generations you need until all the random combinations come out the way you want. Eventually, you reach your own equilibrium of alleles based on artificial selection rather than the equilibrium from nature that came from natural selection.

As far as returning to the median goes… most of the time, you haven’t completely eliminate all of the old genes. Plus, the old genes can be put back into a population by breeding with wild species. Once you stop artificially selecting, then natural selection takes over again and you tend to wind up with the same equilibrium that the wild species had originally.

Whenever the subject of religion comes up and how we are here on earth because we are supposed to love each other or be decent people etc etc etc comes up I wonder what would happen if you take a pool of random humans and select one group to make them loving, kind, emotionally healthy, resilient to emotional harm, etc. and create a second group to make them enraged, bitter, emotionally fragile, etc. It seems to cut to the chase pretty well, if you want loving humans then selectively breed them. Breed one species for high quality of life, subjective well being, love, compassion etc and breed the second species for low quality of life, rage, cruelty, hate, etc.

What would you notice in generation 1, 5, 10, 20, 50?

I’m assuming by generation 3 you would have an average that is 1-2 SD to the right (or left depending on your goal) of the starting point. But beyond that, what happens? Do you reach a point where you are 7 or 8 SD to the right or left of the starting point, or would that require such radical biological/genetic changes to be impossible on such a short timeline of 20-50 generations?

So using IQ (since that is easier to measure) What happens if you have a large pool in the 3rd generation but only let those with an IQ of 160+ breed? What happens to the average? Does the new IQ fall into a bell distribution around that point or due to physiological limitations (trying to pack an IQ 160 into a brain that is arguably designed for an IQ 100 could cause problems) is the distribution more uneven? It is just a series of bell curves moving further and further to the right (or left) with each generation, or do you run the risk of pushing physiological limitations so far that distributions stop being in a bell curve format and are more uneven with a higher % to the left of the new mean, something like this? If so, don’t the physiological limitations get pushed past, but wouldn’t that take potentially hundreds of generations?

There are several issues.

A trait (“intelligence”, a difficult to define concept) may involve a large number of genes and also the interaction of those genes. There’s no guarantee the distribution is bell curve… it depends on how those genes interact. the problem may be that the chromosome with that trait (or that helps express it best) is linked to other traits that may not be desirable.

I recall a program on CBC radio years ago about heritage breeds. They mentioned that original pig breeds were often as smart as dogs. But, they were bred in the last century or more to select for fast and large meat production. As a result, characteristics not selected for, like intelligence, were lost. So the modern, hefty and productive sow is so dumb, they have to be constrained in cages so they won’t roll over and squash their young. Similarly, you hear about dogs bred for show often being “high-strung”, stupid, and plagued with other problems (such as bad hips). Same with race-horses - concentrating on speed has bred animals prone to injuries like broken bones, from pushing the envelope. Breeding for one thing means breeding out other, possibly useful traits.

As for creating breeds of dogs that conform to a certain standard, often the standard is created by a single allele, the presence or absence of which determines whether the animal conforms to the standard or doesn’t. And that allele will always breed true (barring mutation) if the parents are homozygous for the allele.

So if you wanted to create a human breed, and one standard for the breed is that the breed must have blue eyes, it would be extremely easy to select for that trait, and there would be no problem with regression to the mean for eye color.

But if you’re trying to affect a multigene trait, it’s not that simple. Forget intelligence, let’s focus on something a lot easier to understand, like height. Suppose we want to breed for height. We take tall individuals, mate them to other tall individuals. And we find sometimes the offspring are taller than the parents, sometimes the same height, and sometimes shorter. If we keep eliminating the shortest half of our breeding population every generation, eventually we’re going to have tall people, but we won’t be able to get them any taller. This is because our population is now homozygous for every allele that significantly affects height. But one problem might be that having a heterozygous gene might lead to more height than a homozygous allele. Image a protein where AA is taller than aa, but Aa is taller than AA. Well, selective breeding means you eliminate the aa individuals from the population, but you don’t eliminate the Aa invididuals, and thus the aa phenotype persists no matter how hard you select against it.

Plus your tall individuals are going to show problems for several reasons. Suppose they have a gene for increase bone growth. Well, that could cause increased height, but could also cause a whole host of health problems. And having a small founder population could lead to a high frequency of unrelated genetic problems. If a particular founder had or was a carrier for a particular genetic disorder and that founder was the ancestor of most of the future breed, that genetic disorder would end up being extremely common in the future breed.