0:37
Intro. [Recording date: December 22, 2025.]
Russ Roberts: Today is December 22nd, 2025. Before introducing today’s guest, I want to remind listeners to go to econtalk.org and vote on your favorite episodes of 2025.
And now for today’s guest, mathematician and author, David Bessis. He was last here in October of 2025 discussing his marvelous book, Mathematica.
Our topic for today is a provocative essay from his Substack that we will link to. Title of that essay, “Twins Reared Apart Do Not Exist: The Shaky Science of Genetic Determinism.” David, welcome back to EconTalk.
David Bessis: Hello Russ, delighted to be back.
1:18
Russ Roberts: Now, your book Mathematica takes what is for some people a controversial view of mathematics. Unlike Poincaré’s quip that, quote, “Mathematicians are born, not made,” you argue that almost anyone can do math if they approach it correctly or, more importantly. The great mathematicians are not uniquely genius-like. They just have an approach that’s very powerful that can be learned by normal human beings. And, it is that approach, that method, and their dogged focus and hard work that yields greatness. And, you’re talking in your book about Descartes, and Einstein, and Grothendieck. So, you do not believe that everything is predetermined by genetics.
So, you were taken aback. Your essay was provoked by a claim that in studies of twins who were raised in different households who were, quote, “separated at birth,” those studies find that genetics are destiny. And, here’s what you wrote as the takeaway. This is your summary of what is one view of this literature on twins separated at birth. Quote: ‘When you separate two identical twins at birth, raise them in two random families’–I’m going to say that again. We’ll edit that slip out.
… when you separate two identical twins at birth, raise them in two random families, and test their IQs in adulthood, the two results are barely more divergent than two different tests of the same person. [Italics in original.]
So, your IQ [intelligence quotient] tested as a child, obviously you’re the same person. You’re going to go through life and get older; your IQ is going to get tested. Twins raised in totally separate households with, it seems, totally different nurture, but their nature overwhelms it and they are as if they were a single person. You’re hardwired at birth to be smart or not smart or very smart.
So talk about why this was disturbing to you. You confessed in the piece, which I love, that you had a horse in this race that bothered you.
David Bessis: Yeah. So, I think it’s really important to frame the debate in the correct term, because one thing I noticed since I started publicly discussing about math talent and its origins is people try to pigeonhole into one of the two extremes. One extreme would be everything is genetically determined and there is nothing you can do to change your destiny. And, the other supposed position would be the blank slate: like, everything is open and you can become whatever you want. I do not believe in any of these extremes.
And, I think it’s really important because I think that many of your listeners, hearing the first sentences of your introduction, will think, ‘Oh no, not another guy who will pretend that genes don’t matter at all.’ Of course, they matter. I mean, it’s obvious. We know that there are some genetic defects that prevent you from having normally functioning cognition, and that’s just the extreme of a spectrum. And, it’s clear that there is some variability[?heritability?], and we’re not equal. We know, for example, that brain size, the stupidest metric you can imagine, is somehow correlated with outcomes. So, it’s likely correlated also with math[?], but the correlation is so weak that it may not be meaningful.
And, this is this thing about having genes being part of a determination versus being the entire determination that is the issue at stake.
Now, the studies you mentioned are studies that were performed in the 1950s, in the 1960s, in the 1970s, and in the 1980s, and who have been widely publicized, especially on social media. And, the specific post you were mentioning is a visual that was reposted by someone I respect as an intellectual who is Paul Graham. He is not just an intellectual. He is a VC [Venture Capitalist] and entrepreneur. But I respect his writings. I respect his faults[?].
And I saw him posting–that was two years ago–I saw him posting that visual saying, ‘Yeah, you take twins, you separate them at birth, you study the IQ, no difference from two different tests of the same person.’ That looks like a very compelling argument, and it’s sufficiently compelling for millions and millions of people to view that thing, to like it, to enjoy, and to believe that it’s correct.
Now, if this is true–if you really perform that experiment and it really yields that result that there is no measurable difference between people that are raised completely separately–then of course it means something very deep about genetic determination.
And, this thing seemed to contradict my own perception or my own talent as a mathematician developed, and also what I understand about the way it works and what I’ve written about my books. So, it was very annoying. I saw that post when I was making the final corrections on my book before the publication of the American edition, and I was really disturbed. I mean, if this post is correct, then I am delusional. But not me: because I’m quoting people like Einstein, like Descartes, like Grothendieck, and they all say, ‘I was not born with something special. I just developed it by a certain practice.’ So, are they delusional? Are we all delusional? So, I had to have a look at the science.
And now, the thing is, when you look at the science, it’s much more complicated than that. And, actually that experiment, that perfect experiment that looks like it is going to be decisive, is just doesn’t exist in real life. And, that was a big surprise for me.
7:17
Russ Roberts: I have to make the observation that there is the possibility that Einstein, Descartes, and Grothendieck are falsely modest. They’re just trying to make a humble-brag. And yourself as well. You do know yourself pretty well, I suspect, but maybe you’re quite gifted.
It’s interesting to me that you wouldn’t go the other way. Most people want to say, ‘I’m incredibly gifted; and I’ve done great things with it.’ That I’m not so gifted–even if you’ve achieved a lot, somehow that seems a little bit demeaning. So, you’ve embraced a very interesting summary of your own life.
David Bessis: Yeah. So, that’s interesting because it raised many different questions about the motivations of people. So, when my book was published, I was interviewed by Quantum Magazine, who specialized in mathematics. And, I was saying, I don’t think genius is innate. I think it’s a state that you develop. I’m not saying that everybody can get there. I’m just saying that the people who got there had to experience a very special journey, and without that journey, they would not be there.
So, let me maybe try to quantify that, because I can actually phrase that into a quantified conjecture about the odds of becoming a one-in-a-million genius. Let’s put it like that.
It’s clear that someone like Einstein or Descartes or Grothendieck are one in a million–if not less, actually, maybe closer to one in a billion. But still, let’s take one in a million. That’s something you can sample: you take the Fields Medalist, you get the winners of the gold medal at the International Math Olympiad, you get the Grandmaster at chess–of whatever subpopulation that you characterize as cognitively one in a million. I’m not claiming that these guys are just random people with random genomes. Maybe they’re gifted. Actually, it’s very likely that they’re gifted from an innate perspective. What I’m saying is maybe they’re just in the top 1%, and to get into the one in a million, they need to experience a very unique journey.
In October, when we were discussing my book on mathematics, we were discussing this example of Bill Thurston. Bill Thurston was born with a squint, and he started his life not being able to view the world in 3D [3 dimensions] because his eyes were looking at different regions of the space. But, he experienced a very unique childhood with a very loving mother who spent hours and hours with him training his perception of depth. And, through persistent effort, he developed the ability to assemble two-dimensional images to reconstruct 3D vision.
And, he continued playing that game, and he learned to view the world, to think in 4D, four dimensions, and then in five dimensions by using the same technique. And, when he entered primary school, he decided that he would practice visualization every day. That guy–and he’s documented–maybe he was in the top 1%, or one in 1,000, from a cognitive standpoint. Maybe; that’s quite possible. But, what happened to him, this very unique journey of trying to play a very special game with his imagination, led him to develop a very unique geometric ability. And he became one of the most famous geometers of the 20th Century. He was instrumental to solving the Poincaré Conjecture through his Geometrization Conjecture that generalizes it. And, he got the Fields Medal. He’s an obvious genius in geometry; and it’s quite obvious that something that’s not innate happened to him.
Now, what would be humble-bragging in that situation? I don’t know. I don’t know if it’s better for him to say, ‘I was born in a genius,’ or, ‘I had a crazy unique life and it’s hard to replicate.’ In both cases, it’s really hard to replicate; but it’s just maybe more honest to say that you went there through a special journey.
And, this journey is interesting, because maybe people will not be able to replicate that, but maybe there’s some lessons to learn from that that could apply. Maybe not at the same scale, but still, maybe something to learn for[?from?] everyone.
Russ Roberts: When I was in graduate school, I felt that many of my classmates liked to pretend that they never did any homework or cracked a book, that somehow there was something more impressive about just having this innate ability. But, maybe that’s an economist’s problem. I don’t know.
11:58
Russ Roberts: But, let’s go to the–let’s talk a little bit before we get to the actual studies about numbers. So, you give the example of 30%, 50%, 80% as the possible correlation between some innate trait and its manifestation in real life. And, what I think is fascinating–and these numbers get thrown around like they’re scientific–you know, that IQ is 80% genetic. As if that means something really exact. It doesn’t in so many ways. It is not a precise statement. Because IQ is not a really precise measure of intelligence. And, for a hundred reasons, that might be wrong.
But, if we pretend for a minute that we could measure this with some precision, you make the case that 30 and 50 actually still leave a great deal of room for human–for choosing your own destiny, for making your own way in the world, despite your genetic gifts. And, 80% starts to get a little closer to determinism about–and less of a blank slate of any kind. You want to just talk about that for a minute? I think it’s interesting. We can’t–you have some nice pictures in the essay. I encourage people to look at them. But, without the pictures, go ahead.
David Bessis: Yeah. I do think that your being a mathematician helps you see through the noise. Because, it’s a very noisy conversation with many subliminal messages sent through the channel and making everybody confused. When you say, for example, maybe IQ could be, let’s say, 50% determined by your genes, what does it mean? What does it mean for you? Because you’re not the whole population. It’s just 1%. So, 1% in that population.
So, in my article, I included three visualizations of what it would look like. So, it’s a very simple graphic where you have the X axis is your polygenic score, and the Y axis is the actual phenotype–your outcome, your IQ, in that example.
So, the 50%, 30%, 80% is what is called the heritability. That’s basically the regression coefficient where deciding at 100% all the points are on the line. It’s just a direct causation between your polygenic score and your IQ.
Now, 50%, when you look at it, yeah, there’s a trend. That’s clear. But, 50% correlation coefficient is not a very strong correlation. It’s still very weak.
And if you have–which is the default assumption for everyone–if you have an average polygenic score, then you still have a chance to be eventually maybe in the top 1%. It’s not bad. It’s not a high chance. It’s sure lower than if the correlation was weaker. But you still have a chance.
Now, the question is: What is the remaining factor?
And, I think it’s also an important confusion that many people have noticed in their life: that you get two kids from the same family who have widely divergent outcomes. That’s one of the things that Steven Pinker has mentioned in his book: Every parent of two kids starts to think that things are genetic because they’re so different.
And, I have two kids, and I actually realized they’re very different. But, everything is very different about them, not just the genes. But, they were born at a different moment in a different family because there was an initial member when the second one was born. And the parents were a bit older and maybe at a different point in their career and maybe having learned from the mistakes they made with the first child and trying to do things a bit differently. And then, the dynamics of the interaction–it’s different dynamics and everybody’s different.
One thing that is true is that the non-genetic parameter is not that easily pigeonholed into very simple, basic social economic criteria. That is, there is some social economic dimension. It’s clear that it’s better to grow up in a family that is better off and better educated with more books; but it doesn’t decide everything. And, there’s still a massive amount of idiosyncratic things that happen to you and only you, which may be decisive.
And, the example of first one is actually very idiosyncratic. And, I do believe that it’s super-important. There is–it’s actually the notion that Steven Pinker used at the beginning of Chapter 19 in The Blank Slate. The chapter is called ‘The Children.’ It’s about genetic determinism and causation. It starts with the three laws of Turkheimer, and I think they’re very important. Are you familiar with these three laws?
Russ Roberts: No. I saw one of them in your essay. Who is Turkheimer? And then, tell us his rules.
David Bessis: So, Eric Turkheimer is a behavioral geneticist studying the influence of genome on behavioral traits. And, I think it was in maybe 2000 that he published that very famous paper about the Three Laws of Behavioral Genetics and what they mean. And, it’s the first thing that Steven Pinker cites at the beginning of his chapter on that.
And, I think it’s good to go back to the Three Laws because they help clarify a lot. The Law Number One is that everything is heritable[?] to a certain degree.
And, you know, maybe it was bold and new when it was written and it was great to write that, but I think everybody agrees on that now.
It’s clear that there is some genetic–it’s very hard to find someone who truly believes in the blank slate here. Certainly not me, certainly not you.
So, there is a genetic determination; that’s obvious.
The Second Law says that the family you’re born in usually matters less than your genome. So, this is kind of sobering for a blank slatest viewpoint, saying, yes, there is social determinism, but it may not be as important as genetic determinism.
But, the really beautiful one is the Third Law. The Third Law says there is a huge amount of a variability that is neither coming from your genome nor your family. And, this is your life. Who you are–what happens in your brain, what happens in your life, your friends, your chance encounters, the ideas, your things, the games you play, and everything like that.
Now, when you read the paper–and so after I got stuck in that rabbit hole; I spent three months reading everything I could on the subject. It’s very interesting that Turkheimer makes an interesting comment about this Third Law. He says, maybe what matters from your family is something that is unique to you. It’s like Heraclitus. Nobody swims twice in the same river. Nobody grows up twice in the same family. It doesn’t happen like that.
So, maybe that’s an interesting perspective to say that it’s not because it doesn’t apply equally to all children in the family, that parents do not matter. And, this nuance that sounds like a bit like nitpicking is actually, I think, very important, and people underestimate that.
Russ Roberts: Say it again.
David Bessis: So, what is proven–when you study siblings, the correlation between siblings is not that high. It’s a fact. It’s an experimental fact. The correlation between identical twins raised together is super-high. The correlation between fraternal twins raised together is much lower, and that’s a fact. This is undisputable. And, the correlation between siblings who are not twins is even weaker than that.
Now, you have many ways to look at that. One is to say family don’t matter that much because two siblings can have widely divergent outcomes. Another way to look at that is to say there is an influence from your background, but it’s not deterministic. It depends on the dynamics of who you are, in which situation–you know, everybody knows that parents behave differently from one parent to another.
Russ Roberts: One child to another. Yeah.
David Bessis: Yeah, one child to another. And, if you think about it from a developmental perspective, when you think that the brain is a very complex thing, very non-linear machine that develops over decades, you have to realize that this is a very unstable, dynamic, multidimensional problem. So, it’s the thing that tends to be stochastic. So, maybe actually your outcome is not that deterministic, in fact.
21:23
Russ Roberts: Yeah. I think it also–I don’t really like Rule Number Two–Law Number Two–the quantification. It might be true for things that are easily quantified, like height. Height is genetic, and it’s also determined by your environment, your nutrition, both in the womb and after birth. And, it is not destiny. There’s regression to the mean. There’s a random component that is at play. You want to say something?
David Bessis: Yeah, but height is the one example of a very socially important trait that is actually massively determined by your genes.
Russ Roberts: Right. That’s what I’m saying.
David Bessis: And, the case is, even though you still have to get decent nutrition, but if you get good nutrition and good healthcare, then you’re going to probably grow to your potential height, the one your genes determined. And, this results in modern societies in a very strong correlation between your genome and your height.
Russ Roberts: Exactly. So, I didn’t mean to–
David Bessis: Intelligence is not like that.
Russ Roberts: I didn’t mean to suggest that because there’s regression to the mean it’s not genetic. There’s a random component, but there’s obviously a very strong genetic component in how tall you are, the color of your hair, a thousand traits that we have as human beings. How happy you are, how friendly you are, how shy you are, how smart you are–these are words that we use to capture certain aspects of the human experience, but they’re very difficult to quantify. So, I just wanted to make that point.
22:59
Russ Roberts: Let’s move on to the–I’ll call it the science, but I will say that I’m not an expert on this field. I have some interest in it, as listeners will know. I did ask Claude [the chatbot–Econlib Ed.] to summarize–excuse me, yeah, it was Claude–to summarize the studies on this question of twins raised in different homes, especially with respect to IQ. And then, I asked it for critique. And, I can say safely that it is pretty clear to me that you know as much as Claude does about this; and maybe even more, which is not a small thing. So, you’ve gone very deeply into it. And it is, unfortunately, just one more example of the phrase ‘studies show’ not being that reliable.
So, there are five studies that were looked at in the post that you were talking about, but it turns out that really there’s problems with a number of them that are kind of obvious. Like, one of them has apparently fraud; and others have very tiny sample sizes–populations–in their study.
But, two are relatively large. But, it is worth saying right at the start, as you do, that they’re all small because there aren’t that many twins reared in separate homes just by–and I never thought about that. Even if that’s the only thing listeners get out of this because people say, ‘Oh, well, they’ve studied twins in different houses.’ And, it turns out, and it’s, like, ‘Well, how many?’ Not many, it turns out. So, the most famous study, which is Bouchard–I think it was 81 pairs of identical twins? Is that right?
David Bessis: So, I think it’s 56 or 58. In the 1990 Science paper, it’s like 56 or 58. I’m not sure. In the 50s, pairs of identical twins.
Russ Roberts: It’s not tiny. It’s not tiny–
David Bessis: It’s tiny–
Russ Roberts: but it’s not big.
David Bessis: And, it’s global. It’s not a local study. So, there were big articles in the New York Times, and some of the twins who were reunited for this study were coming from different continents. So, you don’t find even 50 in the United States. It’s a very unusual circumstance to have an identical twin, first, although it’s not the normal situation to have an identical twin. Then, to get adopted is another rare event. So, if you correlate the two, you get a very small sample. Now, to get adopted by different families is very unique. And, actually, in many regions and in many periods, you would not do that. You would not separate identical twins because it’s often considered–I’m not an expert on that–but it’s often considered that it’s not a good idea to do that.
So, you end up with a very special situation. And, the fact that it’s special is actually very important in the story because you try to extrapolate general laws about human cognition that apply to billions of people, and you try to learn that from 50-something pairs of people in a very unusual circumstance. And, just from that, there’s an issue.
Now, it’s a very seducing idea. And, I admit that the first time you’re exposed to that, you find it beautiful. It’s what people dream of. Economists, they love that–like, a natural experiment. You don’t have to do anything–just measure it–because nature has done the experiment for you.
And in my article, I have included this very interesting interview of Thomas Bouchard, who conducted the study, and he described how he fell in love with the idea. And, honestly, I completely understand him. The first time I heard it, I fell in love with it. It’s a beautiful idea. It looks like it’s going to give you a complete, precise, and definite answer to the question of nature versus nurture. And, how can you resist that? It’s so tempting.
27:32
Russ Roberts: And, it’s not just the beauty of the experiment. We human beings, as I think listeners know, we like patterns. We love patterns. And, some of them are not representing what we think they are. But we like them anyway. So, we’ll maybe talk about this later as well, but there is a documentary, which I recommend with a little bit of hesitation because it’s, I think, a little bit deceptive. But it’s very dramatic. It’s called Three Identical Strangers, and it’s about triplets who are separated.
And, of course, they all look exactly alike, which is already fascinating that three human beings raised apart–we know they’re going to look exactly alike because they are identical triplets. But then, it turns out they have all kinds of–when they were reunited later in life, they have all these things that they have in common. And, I’m not going to get these right because it doesn’t matter, but this is the thing we love. They all married a woman with the same name. They all smoke–or they don’t smoke. They all like basketball. They all had a pet dog–a small dog, too. And the list just gets longer and longer.
And, it gives me a little bit of goosebumps just telling this story I’m making up because there’s something magical and mysterious. And, even though we might have dramatic and strong feelings about nature versus nurture, this story is so juicy. It’s so delicious.
David Bessis: And, it really started like that. So, you blended two stories, the three identical triplets and the Jim twins. The Jim Twins are the seminal case of a Bouchard study, because these two twins–I think they were born in 1940 and reunited in 1979–and they have the same name, the first name, first name, both of them, Jim. Both of them having married a Linda and then a Betty, or if it’s maybe first Betty and then Linda, but actually they were married twice with women who have the same name. They named their son with the same name. And, the two of them grew up in families where they have another adopted brother whose name was Larry.
Now, this is off–the Larry thing. Because, everything at this point could be involving some of their own agency, but the name of their adoptive brother? Like, genes are not supposed to have a telepathic influence on your parents so they would name your brother with the same name.
Now, when you take a step back and realize what this means, it just means that the two of them were placed into distant families, both in suburban Ohio, and probably people who adopt kids. They’re not random people. They tend to be family-oriented. They tend to have certain values. And, it looks like they were living in a very conformist world in the 1940s where people would always pick the same name for their children.
And, I never dated any Linda, but it’s not because of my genes. It might be because I’m French. So, it’s a simple fact. And, so, when you read that story, and they as kids, the two of them had a dog named Toy. When you read that with a little bit of training in data science and spurious correlation, you start to realize that there’s something off with the experiment. It looks like it was not completely randomized. So that your natural experiment is just something about the social background of suburban Ohio in the 1940s–which is interesting in itself, but it’s not what you want to prove.
Now, that’s interesting because it’s–actually, on a more philosophical, epistemological perspective, the whole story about twins reared apart is very typical of the pre-replication crisis psychology, where people were moving into quantitative science, but coming from, I would say, a more literary perspective on things, where stories were more important. And, I think that the lack of advanced statistical training of most scientists over time–they accepted P-value 5% as a viable criterion for truth, which is not, as we know now, is probably very important in the way the story developed and became serious.
Because when Thomas Bouchard saw interviews of his Jim twins, he thought, ‘Okay, I’m going to do a big study on that.’ And then he started working for 10 years on that, mobilizing huge resources and having a very ambitious project. And, once you start running a project like that, you have to find something, whether or not it’s there.
Russ Roberts: Well, I think it’s interesting that you attribute the 1950s, 1960s, and 1970s as the dark ages of statistical sophistication. Obviously, in modern times, when people are very well-trained in the techniques, at least–maybe not the interpretation, but the techniques–they still manage to have challenges, which is what the replication crisis is representing.
33:00
Russ Roberts: So, let’s talk about the actual study, the Bouchard study, and some of the others, because the things that are wrong–often in these kind of stories, data challenges and generalization challenges–the issues are quite complex. Something is missed about the way that the sample was created. These are not complicated. These are kind of straightforward.
So, what’s wrong? What was–in the design of these studies, what were some of the problems?
David Bessis: So, on paper, it looks like the twin three other part is going to be a perfect natural experiment that tears nature apart. The thing is: the twins share nine months of life in the womb. And you know, if you’ve been around a pregnant woman, you know that pregnant women tend to be careful about what they’re doing because this is a critical period actually for many things. Especially for IQ. Everybody knows that.
Now, this is just one of the first bias. There is more than that. There is the fact that they were not really separated at birth.
I think on average, it was like five months. Which is not nothing. We can say family are not that important, but the first five months seem to matter at least a little bit, especially in families where you’re going to have a reason to get an adoption after that. So, maybe if this was especially dysfunctional families. And some joint event happened that caused them to get separated. What was it? Was it social services taking the kids out because the mother was putting booze in the baby formula? You don’t know. But it looks like the odds of something really bad happening to them is higher than normal.
So, you have a bunch of correlations between those twins that may or may not be that important, but you need to control for that. And, what’s interesting, and it’s something that troubled me when I read that, because I was thinking, ‘Okay, how do they control for that?’ That’s instinctive to any, I would say any modern scientist. You have some bias somewhere, that’s fine. You can live with it if you control it.
Now, there’s a technique for that, which actually I traced back to a textbook from 1960. So, everything was already known in 1960. There is that book by Falconer who was a geneticist and statistician who created the famous Falconer formula that is used in twins-reared-apart and twins-reared-together studies that says it looks like it’s a natural experiment, but it’s not. It’s biased. And, the way to debias it is to take a control group of fraternal twins, which is normal. So, you say you have identical twins, not separated at birth, but separated after five months. So, living through nine months together in the womb, five month together after that, being separated for the same reason, you do the same thing for fraternal twins who share not 100% but 50% of their DNA [Deoxyribonucleic acid].
And, from that, you cannot be 100% sure, but you can have a first-order approximation by looking at the correlation between the identical twins, the correlation between the fraternal twins; you subtract one from the other multiply by two[?]; that gives you a gross approximation of what you want, the variability[?heritability?]. So, this is what you’re supposed to do.
And, what’s interesting is that the study started with the idea that they were going to do that. So, in the same interview I mentioned where Thomas Bouchard explained how he’s falling in love or he fell in love with the study–this interview is from the beginning of the study. I think he was a couple of years into it, and he was mentioning that he was collecting a control group of fraternal twins–which is the normal thing to do, which is the wise thing to do. Now, in the 1990 paper in Science, they mentioned that they have this control group of fraternal twins, and they say, due to space constraints[?], we decided not to publish it. Which is really bizarre, honestly–really, really bizarre.
Russ Roberts: So, that, I’ve always loved that phenomenon. It does raise a red flag, of course. It’s not a good excuse. It’s a implausible excuse. It’s an excuse that no referee–good referee–of a paper in reviewing it would say, ‘Nah, I think I’d like to see that at least.’ It would be okay to tell us the results and then say, ‘But, we can’t put the full table up.’ Or, ‘It’s really complicated, so we didn’t have room to show you everything, but here’s what we found and the data are available on request from the authors.’ But to say, ‘We’re not going to share it because we don’t have room,’ is ‘the dog ate my homework.’ It’s not a good answer.
David Bessis: Yeah. So, they give two excuses, and none of them are believable. The first one is: there is no space.
And the second one is: the sample is too small. So, it is true that the sample is smaller. I think it’s, like, 30 against 56–thirty fraternal twins against 56 identical twins. But it’s not very compelling because the margin of error is one of the square root of the sample size. So, having close to 2X in sample size actually is, like, close to 40% difference in precision, which is not that big. I mean, sure, it’s not as good because it’s smaller, but you cannot decide: one is gold standard–I’m going to keep it–and the other one is not even good for publication. It’s something–it’s not believable.
Now, the other one about the space constraint–so, of course, I made that joke in the paper because being a mathematician, it looks like Pierre de Fermat mentioning he has a very beautiful proof, but the margin is too small for him to include it–which is, you know, in mathematics, it got everybody busy for 350 years to just try to recover the proof. Which probably was wrong, actually. Everybody thinks that Fermat did not really have a proof.
Russ Roberts: Of course not. But if he did, it’s almost certainly true that the margin was too small.
David Bessis: Oh, yes. But, I don’t think if Andrew Wiles struggled so much and got the first proof that was initially wrong, leveraging 20th-century mathematics, it’s quite probable that Fermat made a mistake somewhere. Nobody can prove it, but it’s kind of likely.
Now, in mathematics, at least you can do your homework: you can try to recover a proof. In experimental science, this is just unacceptable. They spent, like, 10 years collecting identical and fraternal twins from across the world, and they are the only ones sitting on the data, and they just don’t want to share it.
I mean, this is unacceptable. And it’s not believable that the one line that it would take you to just mention the correlation coefficient between these pairs takes too much. Because if you have that, this is the silver bullet. That’s the–the–argument that’s going to be more compelling than everything you have written in your paper. So you can throw all the arguments away and just publish that. So, there’s no reason not to publish it.
So, what they do instead is they go through what you’re not supposed to do in data-driven science. They try to talk themselves out of having to show the control group. Which is very awkward.
Like, they say, ‘Okay, yes, it’s true that they share the womb for nine months, and maybe there are some factors like fetal alcohol syndrome that could cause correlations between the IQs of the twins, but it’s not clear they are that important.’ Which is really–honestly, I think there’s something about the paper being published in 1990, because for our eyes[?} right now, it’s just egregious. It’s just impossible that the referee could accept that. You know, this is a very important debate with very important ramifications: You know, people understand society and education and all that. And, you just cannot have that kind of sloppy criteria in a paper in Science. And, I actually think that Science should ask the data actually even today.
I mean, I’m not saying the paper is fraudulent. I’m just saying that the way it’s written is just unacceptable. And we want to see the data. Is there a correlation or not? It’s fairly important. You cannot just leave that thing confidential. It’s not acceptable.
42:12
Russ Roberts: Now, I want you to restate the argument about why fraternal twins are a good control group for people who may have gotten lost in the path, the torturous path that we just went down. So, in a perfect world where you had thousands of identical twins reared apart, what would you do with the fraternal–if you then had thousands, again, forgetting you’ve got the small sample size. I just want listeners to understand the intuition. If you had these fraternal twins that were raised apart, what would you do with their data to make sure that what you found with the identical twins was reliable?
David Bessis: Yeah. So, it’s interesting because it bridges with another kind of studies that is the twins-reared-together that actually are exactly what you described. We have thousands of pairs of identical twins that are raised together and fraternal twins that are raised together. And, your question is about how do the studies of twins reared together work? Because, this is actually what they do.
So, they operate under a number of simplifying assumptions that are debatable and are probably not exactly true. So, that means that there is a debate about how you model the actual heritabily, but at least you have a functioning model that is a first order approximation. So, how does it work?
So, the idea is: identical twins share 100% of their DNA. They are literally clones. So, when you look at the correlation between, say, their IQ, there is the genetic correlation. And, if you look in terms of variance, the variance is good because if variables are not interacting with each other, you can sum the variance of a different component, which is a simplifying assumption that is not exactly correct, but at least you can start from that. So, they–say, you have the genetic variance and you have the family environment variance, which is: the second one is the same for fraternal twin, except that the first one–the genetic variance–is only 50% for fraternal twins.
So, what you do is you take the difference. You take the correlation coefficients for identical twins, you subtract the correlation coefficient from fraternal twins, and you multiply that by two to extrapolate what would be the full contribution of the genome, because with fraternal twins you have 50% of the genomic contribution.
So, let me give you an example, which is–actually the twins-reared-together studies give something like that. Let’s say you have 80% correlations between the IQs of 1,000 pairs of identical twins. So, you look at the within-pair correlation. X is Twin 1, Y is Twin 2; you make a scatterplot of that, and you take the regression line.
What is the correlation coefficient? It’s something like 80%. Okay?
If you do the same thing for fraternal twins, you may get maybe 50%. So, the difference is 30% and using that simplification formula that dates back to Falconer in 1960, you get 60% heritability.
So, this is an estimate, and it’s debatable because genome is not linear and there are some interactions between the different variables. They’re not completely separated. And, the fact that you have an identical twin kind of changes your life experience. So, there is an active conversation about whether or not this approximation is correct, but at least you have one.
And, this is what Bouchard set up–I think you’re going to edit because I just punched my microphone. Yeah. So, this is what Bouchard set out to study when he collected data for the fraternal twins and identical twins. So, based on the two correlations, you should have an estimate.
46:32
Russ Roberts: But, I just want to say it a little different. If genetics was important, there’s this high correlation between IQ, say, for two twins raised in different households. So, if we found that–if we found that twins raised in different households are frequently, their IQs are very similar–we’d want to say–we might want to conclude–well, therefore–again, forgetting all the things you already mentioned about the womb and they’re not assigned randomly.
We discover later, you mentioned this in your paper–I saw this from Claude–some of them were assigned to families–the same family, just not the same house. It’s, like, an uncle and two brothers: two different houses in nearby neighborhoods.
So, the actual nurture part of this is extremely similar. So, we would expect them–putting Turkheimer’s Third Law aside for a moment–we’d expect them to have very similar IQs, not just because of the genetics, but because their environment was very similar.
But the idea of having fraternal twins as the control would be that if they were raised in separate households, they don’t have the same genetics. So, how far apart are their IQs? They’re also probably going to be raised in similar homes maybe by assignment. It’s not random, not a random house in America.
So you’d want to compare–if genetics is really important, you’d want to find that the correlation for the identical twins was much bigger than it is for the fraternal twins.
And Bouchard, in theory, had that data and he did not share it. Not only did he just have the data, he did the finding.
Now, I don’t know if there may be something that Claude knows that you don’t know. According to Claude, Jay Joseph, a person who tends toward blank slatism, evidently, was very critical of the Bouchard study. He claimed to have access to their data. Is that correct? Do you know?
David Bessis: So, I mention that in a footnote. Yeah. So, he claims to have located the data in a subsequent paper by the team, by the same team, but basically show that there is no meaningful signal, which is, to be honest–
Russ Roberts: No statistically significant difference between the correlation of the fraternal twins raised apart and the identical twins raised apart. Which means that the identical part can’t be decisive.
David Bessis: Yeah. Which is not surprising when you look at the sample size. You have, like, 56 in one hand–
Russ Roberts: A lot of noise–
David Bessis: But, in the other hand, so the error bouts[?bounds?] will be pretty wide.
So, it’s kind of normal that the two confidence interval between fraternal twins and identical twins overlap. It’s not surprising. It’s a fact of life. You don’t get that many twins. The other part, I’m not saying that there’s no genetic correlation. Of course there is one. The question is, how much can you quantify it? And, this protocol probably was enabled to compute a decisive confident value for that share of genetic influence. And, that’s a fact of life. It’s a normal fact of statistics. The study is too small to prove anything. But, this is obfuscated in the paper.
So, after my article was published, it triggered many, many reactions. And of course, some hardcore hereditarians criticized me for requesting that control group data. And, they mentioned that there was some subsequent samples from the same study published years after that, which is already a bit off because when you do a control group study, it’s basically having a placebo arm. You’re supposed to get the same vintage from the data. You cannot say the control group in 1990 was not good, but I’m going to use the one from 1995 because I like it better. It’s not the way it works.
But, they were mentioning that if you take that later data and you put it in the model, then you get a confidence interval for the heritability that includes negative values. And they were mentioning–as if it was a good argument–look, they could not publish something where the heritability could be negative because that’s meaningless. Of course, that’s meaningless: let’s just say that the study doesn’t work. But that’s not that IQ is negatively correlated with–it’s not that. It just doesn’t work. That’s a fact. Why do you hide it?
51:37
Russ Roberts: So, let’s talk about some of the–readers can go to your essay. We’ll put a link-up to it. We’ve done our best in this conversation to make it accessible without visuals and statistics. It’s not a hard essay to read. It’s pretty straightforward. And, the visuals about the importance of a 30% or 50% or 80% number are quite helpful. I will just mention this an aside: it’s one of my frustrations in economics that people can actually publish papers where there’s a scatterplot of the data points and it’s a cloud, and it’s a circle. You can look at it and you go, ‘That looks like a circle of randomly done points.’ But, if you put a line through it, sometimes you can get a significant result; and they trumpet that as if it’s an important finding. It’s depressing to me. It’s the nature of our game.
But, I want to talk about the significance of this–not the statistical significance, the human significance. So, parenting is really important to me. I have four kids. I have four data points. And, of course they have some similarities because of genetics and they all grew up with the same parents. But they did grow up at different times. We learned things. We also probably made mistakes trying to correct things that we thought we’d done wrong with the earlier kids.
And, the idea that it’s irrelevant, that I had no impact on how my children’s curiosity, happiness, maturity, kindness, etc., the idea that the times I tormented my children by making them say thank you, or I’m sorry, were actually irrelevant–because parenting is important. And listeners can go back and listen to Bryan Caplan’s episode on parenting here at EconTalk from a long time ago. And, Bryan argues that parenting is not so important. And, there’s a definite literature in parenting that says, ‘You don’t have much control over how your kids turn out. It’s an illusion.’ So, you wonder–if you’re an analytical person at all–‘Did I make any difference? Even negative, did I make any difference?’
And so, that’s one of the reasons I think this debate is important. Why else is it important?
David Bessis: Oh, this first point is already quite important. I mean, I have two young kids; and just this debate, what is under your control and what is not, I think is completely biased when you start from saying you have no control at all. Of course, you control it very, very diffused, very not direct, but still, you don’t drink during pregnancy when you’re a pregnant woman. You should not do that. And, this is parenting. There are these labels on the wine bottles and they’re there for a reason. And, actually it’s proven this beyond doubt that drinking during pregnancy cause damage to your kid and especially to cognition. This is proven. This is parenting.
Russ Roberts: Keep going, David. I’m going to critique that a little bit, but go ahead.
David Bessis: It depends. Okay. I don’t want to be completely paranoid, but you should not drink a bottle a day when you’re pregnant, that’s for sure. Okay. Can we agree on that?
Russ Roberts: Yes. Listeners can go back to the Emily Oster episodes on parenting. And, I think alcoholism is really bad for a child in the womb. Whether a glass of wine is, is I think, more complicated. But, carry on.
David Bessis: I can agree with that.
Now, what you have with parenting, you can transport that to education. Of course, we all know that story–and it’s a sad one–about the fact that intervention in education rarely scale. Usually, you have a very neat pilot with an expert who invents a new way of making teaching better. And, when you try to scale it, like, it vanishes away.
Okay, this is sad; but does it mean that teachers don’t matter and education doesn’t matter? That’s a tough sell. Maybe there is a good point to say that it doesn’t matter as much as we would like it to matter, but that doesn’t say it doesn’t matter.
And, this is where I think people are stuck with deterministic thinking. And I think this is one of the–again, my theme about the ability to grasp mathematical certainties in these issues is actually very important and goes a long way to explaining all the confusions in the [?] debate.
The Third Law of Turkheimer says things are going to be noisy. Whatever quantified variables you have to describe in a few dimensions–your environment, like the income of your family, whatever–this is just going to be a few correlations with the outcome. But the actual thing that matters is what’s happening in your head, your journey.
And, this is actually the whole topic of the conversation we had about mathematics. And that’s the whole thing. That is, from my perspective, the most important discovery I made in my life, which is to become better at math, which is one of the supposedly most g-loaded activities you can imagine.
Russ Roberts: G-loaded, meaning related to your general intelligence?
David Bessis: Yeah. It’s supposed to be–it looks like, your mathematics looks like a giant IQ test in many ways. But, what I realized in my studies and then in my career as a pure mathematician–and it’s something that most mathematicians agree on–is that what’s really important is how you process feelings in your head. Your invisible metacognitive approach to interacting with mathematical object, playing with your imagination, trying to own your intuition, being patient, being persistent, not succumbing to the fear that this thing is going to be too difficult for you and you should run away from it.
It’s very subtle. It’s invisible. It’s something that is very personal. Therefore, your journey will be very idiosyncratic. And this is super-important. And, yes, it doesn’t scale easily, but it’s one of the most important debates we can have. And, if we say everything is decided on the day you were born, we cancel that conversation. And that’s a big issue.
58:29
Russ Roberts: Yeah. I want to close with a related point and get your thoughts. It really shouldn’t be a controversial idea that the world is complicated–which is another way of saying Turkheimer’s Third Law. The thing that’s happening in your head, your inner journey, your inner narrative, your inner experiences, is never going to be in the dataset, at least in our lifetime. Maybe someday down the road, but I suspect not. The fancy name for this in statistics is there are ‘omitted variables’. And sometimes they’re incredibly important. And that’s just a reality. But we don’t like that reality. So, that’s one. I think that’s a great–an important point.
But, the second thing that I find fascinating related to this is that people really care. And it’s not obvious that they should. This is not important. Nobody lives their life–in a certain dimension, it’s not important–nobody lives their life saying, ‘Well, if it’s only 50%, I’m going to do X, but if it’s 80%, oh my gosh, then I won’t. I’d never bother with doing that.’ We live our lives, we understand the world is complicated, we do the best we can. The idea that we are so passionate about this issue of nature versus nurture strikes at the very heart of how we see our humanity and ourselves. And, I think that is a piece of this debate that is utterly fascinating.
In some sense, it’s irrelevant. It’s meta-meta. It’s once more removed from the reality.
But, it’s not irrelevant. It’s important to us to understand who we are, and we get a position that we care about on this issue. And, we don’t like having it–just like you admitted at the beginning of this–the idea that my notion of my own journey might be an illusion is unbearable. So, you’ve turned out okay. You found out that it turns out you don’t have to give up your illusion. You can actually believe that you’re in control of your fate as a mathematician more than the hereditarians say. But anyway, just react to that.
David Bessis: Yeah, I think you’re spot on. There is something very personal behind these controversies. And, actually, I don’t think I have very good control on my destiny. I just think that it’s non-zero. I just think that. Okay.
And actually, one reason I don’t think that is because I know that my own ability fluctuated a lot depending on my emotional state. That’s the most important discovery of my life.
But, if you look at anecdotes about how people frame their perception of the nature-versus-natural debate, there is a very interesting post called ‘The Parable of Talent’ by Scott Alexander in the old Slate Star Codex. And, it starts by saying, you know: It’s been obvious to me that things are innate to a certain degree because when I was a kid, I was really bad at math. And, I tried hard and I was really bad at math, but I was very good at English and I was not trying hard.
So, that’s interesting because here you see something that’s coming from childhood. And you have this shocking thing: that some people appear gifted and some people appear handicapped–had the same activity. And, we don’t really know what causes that. And, because we don’t really know that, there are certain tendencies to put the blame or sign an act of God somewhere that controls that.
So, some people put the blame on society and the blank slate is saying, ‘Okay, it’s because they were coming from bad parents or whatever–bad family or bad social background.’ Some people put the blame on the genes, you know, saying, ‘Okay, I don’t understand what made me bad at math, so I’m going to assume that this was genetic.’ And, that’s understandable.
Actually, when I was 17, I started some kind of advanced program for kids who were really good at math. And, I was confronted with people who were unbelievably, unbelievingly, smarter than me. One of the kids in my class went on to win a gold medal at the International Math Olympiad, and that was mesmerizing. I could not understand how he was doing it. It was magic. There was not a single problem I could answer that he could not answer, but there are many problems that he could answer that I could not answer. And, it was instant. And he was sleeping during classes. But it looked like magic.
And, at that point, I was pretty convinced that this thing was genetic. Because, I could not imagine a non-genetic explanation. This guy was not coming from a fancy family. Not at all. Quite the opposite.
So, I completely understand where hereditarians are coming from. I was coming from a certain perspective.
Now, we’re in 2025, and we start to understand a few things about the brain. And we start to understand a few things about how you can create machine that emulates some brain function.
So, it’s interesting to try to maybe unpack that mystery. And there’s something to learn. And, so that’s why I really think that we should really go beyond ideology on that, because ideology is really, in that instance, it’s really censorship.
Russ Roberts: My guest today has been David Bessis. David, thanks for being part of EconTalk.
David Bessis: Thank you very much.
