Manifold

Gregory Clark is Distinguished Professor of Economics at UC-Davis. His areas of research are long-term economic growth, the wealth of nations, economic history, and social mobility.

Show Notes

Gregory Clark is Distinguished Professor of Economics at UC-Davis. He is an editor of the European Review of Economic History, chair of the steering committee of the All-UC Group in Economic History, and a Research Associate of the Center for Poverty Research at Davis. He was educated at Cambridge University and received a PhD from Harvard University.

His areas of research are long-term economic growth, the wealth of nations, economic history, and social mobility.

Steve and Greg discuss:

0:00 Introduction
2:31 Background in economics and genetics
10:25 The role of genetics in determining social outcomes
16:27 Measuring social status through marriage and occupation
36:15 Assortative mating and the industrial revolution
49:38 Criticisms of empirical data, engagement on genetics and economic history
1:12:12 Heckman and Landerso study of social mobility in US vs Denmark
1:24:32 Predicting cognitive traits
1:33:26 Assortative mating and increase in population variance


Links:

For Whom the Bell Curve Tolls: A Lineage of 400,000 English Individuals 1750-2020 shows Genetics Determines most Social Outcomes
http://faculty.econ.ucdavis.edu/faculty/gclark/ClarkGlasgow2021.pdf

Further discussion
https://infoproc.blogspot.com/2021/03/genetic-correlation-of-social-outcomes.html

A Farewell to Alms: A Brief Economic History of the World
https://en.wikipedia.org/wiki/A_Farewell_to_Alms

The Son Also Rises
https://en.wikipedia.org/wiki/The_Son_Also_Rises_(book)



Music used with permission from Blade Runner Blues Livestream improvisation by State Azure.


--
Steve Hsu is Professor of Theoretical Physics and of Computational Mathematics, Science, and Engineering at Michigan State University. Previously, he was Senior Vice President for Research and Innovation at MSU and Director of the Institute of Theoretical Science at the University of Oregon. Hsu is a startup founder (SafeWeb, Genomic Prediction, Othram) and advisor to venture capital and other investment firms. He was educated at Caltech and Berkeley, was a Harvard Junior Fellow, and has held faculty positions at Yale, the University of Oregon, and MSU.

Please send any questions or suggestions to manifold1podcast@gmail.com or Steve on Twitter @hsu_steve.

Creators & Guests

Host
Stephen Hsu
Steve Hsu is Professor of Theoretical Physics and of Computational Mathematics, Science, and Engineering at Michigan State University.

What is Manifold?

Steve Hsu is Professor of Theoretical Physics and Computational Mathematics, Science, and Engineering at Michigan State University. Join him for wide-ranging conversations with leading writers, scientists, technologists, academics, entrepreneurs, investors, and more.

Steve Hsu: Welcome to Manifold.

Today, my guest is Greg Clark, distinguished professor of economics at UC Davis, and currently also a visiting professor in the economics department at LSE. Clark is an editor of the European Review of Economics. He is chair of the steering committee of all University of California groups in economic history and a research associate of the center for poverty research at Davis, he was educated at Cambridge University, and he received his Ph.D. from Harvard University.

Now, Greg, welcome to the show.

It's great to have you here. You may not remember, but we met 10 years ago in 2012. When I was visiting UC Davis, I was there to give a physics colloquium and you were kind enough to sit in a cafe with me and discuss your recent work, your then-recent work, on population dynamics in England, which was based on wills and inheritance documents.

Greg Clark: Oh, I remember your visits very well.

Steve Hsu: Yes. So, you had found that the wealthy had more than twice as many children surviving to adulthood than the poor did. And this was true over a period of hundreds of years before the industrial revolution. And I at the time was amazed by your results. And they caused me to revise my own thinking about the pace at which natural selection could proceed in humans and taking, without going into details, taking your empirical results into account, I concluded that you could have significant shifts in population traits, maybe by as much as a standard deviation on timescales, as short as may be less than a thousand years or a few thousand years, which I think to most people's thinking who study evolution is just the blink of an eye.

So since then, I followed your work closely and I'm a great admirer of yours. And I don't think we'll discuss the older work. I think we're going to focus mainly on your more recent work, which is focused on social mobility, genetics, and assortative mating. How does that sound?

Greg Clark: Sounds good. Yep.

Steve Hsu: Great. So first I'd like to give the audience a sense of your background, where you grew up, how you got interested in an academic career and in particular in economics.

Yeah. So, my background is I grew up in a kind of, one of the rougher cities in Britain, Glasgow in Scotland. But I always had a lot of intellectual interests as a kid. And so, I actually went to college intending to do mathematics. But I very quickly realized that Cambridge, that, was a really too dry subject for me.

Greg Clark: So, then I actually studied philosophy for several years, but I fairly quickly concluded that that seemed to be an area without too many possibilities of progress. And so that's how I finally stumbled into economics. But as I say, I've always had this wider kind of intellectual interest. And one thing I would say about economics as a subject is that there's a lot of kind of narrowness to the economics approach to many problems, which, and one of the premises of that approach tends to be that people are the same everywhere.

It's simply the institutions and the incentives that face that change their outcomes. And then the second kind of thing about me that is relevant to the research I'm doing now is that both of my parents came from families of 12 children, and both were number nine in these families. And my father's family, there was this trait that they all mostly seem to be pretty intellectually able, but interestingly, his parents had absolutely no interest in education.

So, the oldest child was a child prodigy that they took out of school at age 14. And so, I was intrigued as to how, given these kinds of backgrounds, people could actually end up expressing certain kinds of intellectual traits, and it kind of biased me in terms of thinking that there must be something else that's transmitting these abilities then the kind of the family circumstances that grew up in.

Steve Hsu: It's interesting that you should mention these large families because although I didn't come from a large family myself, I guess both my mom's and dad's families were large, but, but my own, I only had one brother growing up. But I, from knowing lots of other kids, I grew up in the same town, all through elementary school and high school. So, I knew lots of other families in the town. And I think when you see the range of variation that's present within say a group of siblings, you start to really wonder about issues of, well, what is actually genetically caused and what is caused by the environment, et cetera. And it makes you suspicious that the environmental causes could be everything.

Greg Clark: Well, it's certainly, it raises these intriguing issues. When you see large families like this, as I say, you see then a range of variation, but you also see that they, they do have traits these families, and, and it, it raises these intriguing questions about what is determining people's, social outcomes, and what is being transmitted, between, generations.

And so, I actually say that I talked to my grandmother, who's a crusty, but kind of interesting old woman, and asked her, could she tell me anything about my father as a young boy? He was number nine, remember, in this family. She said, no, I don't remember anything.

Steve Hsu: Yeah. Wow. So, but when you entered into economics at the time you entered, probably I'm guessing almost everyone was a kind of blank slate when it came to genetic influences.

Greg Clark: Yeah, I mean, it is sometimes that the huge premise of economics is it's all about incentives. And so, the first part of my academic life in economics, I spent a lot of time trying to understand why some countries succeed and other countries fail. And it turns out that those differences between countries are now in the range of incomes of 50 to one.

And so, they're absolutely mammoth and they're much more important than anything that economics is actually good at explaining. But I must admit after like 10 or 15 years working on that topic, that it turns out to be almost impossible to make any progress in terms of explaining these differences between countries.

And so, one of the things that actually I found attractive now about working on topics like social mobility and outcome determination in individual families, is that you see absolutely lawlike behavior. And you can actually see something that looks more like science. And, as I say, it turns out to be some of the most important problems in economics. These international differences just seem to defy any kind of systematic explanation.

Steve Hsu: And I guess plenty of economists are content to ascribe those differences to things like a path dependence in history and institutions and all they have. There's no shortage of proposed explanations.

Greg Clark: Oh, yeah, no, absolutely. There are many proposed explanations. But, you know, when you do actual kind of detailed studies of particular industries, then you see that most of these explanations simply don't work. Right? I mean, the puzzle is why when people combine in a textile mill in India, why did they end up having to use five times as many workers as a similar, as an identical textile mill in Lancaster, in England.

And it just is very, very hard, even knowing all about the details of what went on to come up with any explanation about why things work out in that particular way. And I think these kinds of differences must have something to do with the way people interact. But they turn out to be extremely difficult to deal with using the standard tools of economics, because if incentives were the things that really mattered, then that equilibrium would be destroyed, and it would be replaced by the high productivity equilibrium because the incentives would actually produce the results.

And so, so as I say, there’s economics really based on this premise that it's really just incentives that matter. It turns out the most difficult results in economics are actually ones where incentives just don't seem to work very well. And so, it is a somewhat curious subject, economics, where we are hoping for some kind of scientific basis, but it turns out we fail 99% of the time in explaining anything right. In explaining what income per capita will be in individual countries 10 years from now, in explaining, you know, what the rate of growth of income is likely to be, in explaining what regions of countries will prosper or regions will fail. And so, our track record is actually pretty poor.

Steve Hsu: Yes, I, well, I, I guess having, having spent a fair amount of time studying, financial economics, so in the things like modeling, on Wall Street and things like that, I agree with you. It's very, very hard to predict anything.

So let me turn to one of your recent preprints. This is the one which is entitled to whom the bell curve tolls a lineage of 400,000 English individuals, 1750 to 2020 shows genetics determines most social outcomes. And perhaps that's what you were referring to when you described it as having law-like behavior.

Greg Clark: Yeah. And so, here, there's a tremendous resource available to researchers who are interested in social mobility, which is the enormous interest that English people have in their own family histories. And so here we were able to, and we have still wrought on putting this together in assembling from individual researchers, the research they've done on the long lineages that they come from. And so, we're actually able to assemble this complete kind of lineage where we follow everyone with a bunch of rare surnames so that we don't have problems with selecting. And we can then look at things like what was the underlying rate of social mobility in England in 1750 compared to now? And the astonishing answer is that that rate is extremely slow.

The underlying correlated intergenerational correlation seems to be in the order of about 0.8. And the second thing is that it has not changed over time. And yet in that period, we've seen enormous changes. We've seen the introduction of mass public education. We've seen the introduction of significant welfare services. We've seen the introduction of significant transfers between richer people and poor people that didn't exist substantially before. And so, the first impressive thing that stands out in this lineage is that somehow there's an underlying intergenerational correlation in people's social outcomes that is the same across 300 years, in a radically different social structure.

Greg Clark: And then a second thing, that stands out. Is that the rate of upward mobility from the bottom of the distribution is exactly the same as the rate of downward mobility from the top of the distribution. And again, on most social theories of mobility, you would expect to find that there would be differences in that rate of movement towards the mean that were created by, you know, the ability of an elite to embed itself at the top of the structure or by poverty traps, that would trap people at the bottom of the social ladder.

But the very interesting thing in this data is there is no poverty trap and there is no elite privilege. Social mobility is behaving the same way all across the social spectrum.

Steve Hsu: Right. So just for the audience, maybe we can define carefully what you mean by social outcomes or social status.

Greg Clark: Yes. so, it turns out that the way we, the way in practice we measure social status is we have occupational status. We have educational status. We have wealth as measures and also things like health are also correlated with people's social status. And so, in principle, we can measure it using these different attributes and in principle, they could actually all be changing very differently, but it turns out that in this data they're very highly correlated with these different aspects of social status.

And so, in the long-run wealth, education, occupational status, these things really, are, are closely aligned. And so, people behave as though they have some kind of underlying generalized social ability. And then we can get measures of that social ability that are kind of partial measures, but there's some deeper kind of set of kind of social abilities that people have, which seemed to be some combination of their intellectual abilities, but also of their drive, their perseverance, their, their time horizons in terms of thinking about things.

But as I say, the world behaves as though there is just this one kind of underlying set thing, which is a kind of generalized social ability. And then we, another thing we can do with the data is we can, because of the depth of this lineage, we end up with something like 2 million, second cousins, 2 million third cousins, and we can't find outcomes for all of them, but it means that we can actually measure people looking at siblings, cousins, second cousins, third cousins, fourth cousins, and often then people who are at last had a common ancestor, something like 150 or 180 years before. And we can look at that pattern of correlations. And the stunning thing that emerges is that it's exactly the pattern you would predict if this social ability was really being transmitted by additive genetics, but with a very strong correlation in genetics between parents at marriage in the order of 0.6.

Steve Hsu: Right. So, before we get into any particular model, I just want to make sure the audience understands the description of the empirical results. So, you have a lineage of 400,000 people. able to, at least for some subset of those people, measure their social status by their profession or the level of wealth or education. And so, you assign each of them, maybe some value on a scale of one to five or something in terms of what their social outcome is.

Greg Clark: Oh, yes. So, we are able for occupational status, we have 500 separate occupations that we can assign status to. And that ranges from zero to a hundred. And we can actually, the best way we find to assign status is actually to look at who marries whom and what the relationships are between sons and fathers and sons and fathers and law.

And that will actually. Sort of occupations into a very nice kind of occupational scale. And then, for something like educational status, we can just say, did you have higher education, or did you not have higher education? So that's a much cruder scale, but relatively well observed.

Yeah. So, as I say, we can allocate people some kind of social rank within this system.

Steve Hsu: Okay. Got it. And interestingly, in a way, because you're looking at who marries whom, it's sort of what that society at that moment in time also thought of as the social hierarchy, because they're sort of deciding for themselves.

who's going to marry

Greg Clark: That's right. Because it turns out we have another data set, which involves 1.7 million marriages, again, collected by amateur genealogists in Britain. And I estimated I think that this would involve something like a hundred thousand person hours of work to actually have assembled all of this data. But this is one where we can actually then independently then assign a status rank to all kinds of occupations just by looking at who's associating with whom in the social ladder.

The only thing about that ranking is you have to tell it just one thing, which is what's the top-ranked occupation versus the bottom-ranked occupation. But that method is actually standardly used by sociologists, and it turns out it works. Once we have this huge data set, it works actually incredibly well in terms of assigning status.

Steve Hsu: Got it. So, you have a method of assigning status and maybe the first observation is that intergenerationally a high correlation of around 0.8

No. So, here I have to be careful, as these status measures always contain a lot of errors. And so, if you look at the actual intergenerational correlations, they actually tend to fluctuate over time. And for modern data, they would only be in the order of 0.4 or 0.3. Okay. And that suggests that it's actually a very fluid social world.

Greg Clark: And so, to figure out the true underlying social status, we have to correct for that error. And so that number that I quoted you is effectively the error corrected underlying social status correlations. And, and those actually seem to be in the order of 0.8 and to be the same across all periods, the same for people in 2021 as they were in 1700.

Steve Hsu: So, what goes into that correction is, is the claim that the actual observed social status is not actually reflecting the potential of the individual to achieve social status. And you're trying to correct for that? Or, or is it something

Greg Clark: Oh, and now here's, the problem is, suppose I had described someone as a teacher, right. And so, they'll get some allocation on this status scale. But it will still be the case that we're, we don't have an exact measure of where teachers fall with a finite amount of data. We're not going to exactly do that. So that's one source of error. But then within the category of teachers, there will be teachers in England, for example, who teach at Eton, at some incredibly fancy private school, and then there'll be teachers teaching at some job that really doesn't require a lot of, kind of intellectual background. And, and so there are these two sources of error then in terms of that status measure, which is that these are always loose descriptions. And we, on average, haven't assigned exactly the correct status. And then secondly, there's a lot of variation within that description. Okay.

And, and you get the same problem. If you say, you know, if we classify people as, you know, have completed a university degree, there's a big difference between people who complete that degree at MIT versus someone who does it at Sacramento State. Right? And so, any of these measures always contain errors. And these errors are actually changing over time as the nature of jobs has become more, more elaborate in the modern period. There's an incredible number of people who have some kind of manager, but that covers an incredible range of actual social statuses. So, we have to get some corrections for that. And one way we can correct for it is by looking at the relative correlation of you to your grandfather versus you to your father.

And, that relative correlation will actually show you what the true underlying rate of social mobility is because both, both, both of those measures have the same error. And when we divide one by the other, we can cancel out that error, and then we're left just with, well, what one further generation, how much does that reduce the correlation.

Steve Hsu: I see. Okay. So, so far, we've been talking about intergenerational correlation. One of the things I found really interesting is that if you look within a particular generation, but you look at say, first cousins, second cousins’ brothers, so reduced degree of relatedness. I think you also see an equivalent level of correlation. Is that correct?

Greg Clark: Right. I mean, what's happening is that the correlation is declining as we move out on the family tree, but in a very predictable fashion, right? That always, as we move, as we move from second cousins to third cousins, that's two steps away on the family tree, that correlation declines by a factor of 0.64.

Okay. And, and that, it, it, you know, it doesn't matter if you've ever met the other people if you've ever had any connection to them if you ever had any involvement with them that basically, it shows this very regular structure so that we can plot on one axis people's genetic relatedness, and then on the other, the underlying correlation between them. And it'll just fall along the straight line.

Steve Hsu: And is it true that, so I think in your paper you say that the person who first formulated a model for what these correlations would be under the assumption of some level of assortative of mating, which we'll define in a second. And then, under the assumption that the underlying trait obeys rules of additive genetic inheritance, that person was Fisher.

Greg Clark: Yes. So, Fisher, I think in 2019 had a famous paper, which interestingly was so badly written that the Royal Society in England rejected it. And so, it was actually published in Scotland in a journal. But it turns out that that paper absolutely correctly predicted what all of these correlations would be.

And you could then also infer from the set of correlations, what the underlying marital correlation was and the trait, right. That the data then would also tell you what the underlying implied marital correlation was. there's some history in this, which is I, I gave a, you know, it takes me a long time to finish any of these projects. I gave, seven years ago, an early version of this paper. And, said here, it's amazing this actually follows exactly the pattern that Fisher would have predicted a month in this genealogy.

And someone said, well, that's very interesting, but the implied underlying correlation of 0.6, genetic correlation, that's just completely implausible. It turns out since then, that has actually been shown to be for educational prediction status predictors, to be the actual correlation of people in modern England. And, and so it turns out that the, as I say, the model makes these predictions and now we can actually test that against, whole-genome evidence on what are the determinants of people's educational outcomes.

Steve Hsu: Right. So let me, I have the reference right in front of me. So let me give it to the audience. And what we could put in the show notes. So, it's Fisher 1918, the correlation between relatives on the supposition of Mendelian inheritance transactions of the Royal Society of Edinburgh.

So that's the paper you referenced, and it contains the results in that paper. I'm sure that paper's difficult to read, but I was discussing this with one of my postdocs and he mentioned that in one of his population genetics textbooks, I think he was Faulkner or something. He said that the exact set of formulas appears in the textbook now. So, it's a very famous result.

Greg Clark: Yes, it was from Faulkner that I actually got the act, because as I said, the Fisher paper is very, very difficult to interpret. I had to look at it, but it's very, very difficult. But Faulkner has a simplified version of that.

Steve Hsu: Yes. And so given a certain level of assortative mating, which means that the husband chooses the wife, or they choose each other in such a way that in a way their genotype is more correlated than by random chance. So, there's some measure of that which appears in Fisher's formulas. And the second is the narrow-sense heritability of the trait in question. And I think given those two parameters, correct me if I'm wrong, given those two parameters, then one can predict for any degree of relatedness between two individuals, what the consequent correlation should be in, in this case, in terms of their status potential.

Greg Clark: That's right. And, and there's only one final complication, which is Fisher contemplates the case where people actually match on the genotype and the cases where they match on the phenotype. But I think, you know, it's not plausible the case where they match on the genotype. It has to be that they're actually matching on some kind of phenotype.

Steve Hsu: Yes, so when I first saw some of your early work on just the rate of regression to the mean intergenerationally, and, you know, your results show that this was much slower than people would have expected. I myself thought this can't be right because even if I assume a very high heritability for the underlying traits surely the degree of assortative and mating is not high enough to actually give the results that you found empirically.

And so, I was just amazed when I saw this more recent paper by you and also looked at the genomics results from UK Biobank, which you just mentioned, which actually show that the level of genetic assortativeness is extremely high. It's something like 0.65, the correlation between the polygenic score for educational attainment in married couples.

So, I was actually amazed that your empirical work implied this result many years before we were actually able to measure it in actual genomics.

Greg Clark: Yeah, and I should, for the listeners here, I should explain. I, you know, have had no background in genetics before I started looking at some of the thing’s material. And so, I was getting some results and then I went and learned a little more about genetics. And then it turns out in my earlier book, which looked at social mobility using surnames. It turns out that empirically, we formulated a simple model that seemed to capture what was determining social status over multiple generations. It turns out that that model is the one that additive genetics actually would imply.

And so, this was a kind of convergence between observing certain empirical facts about the world and then, learning that well, if we want to try and explain these facts, what potential models could actually work. And it turns out that there's a congruence between what we're observing and what would be predicted by additive genetic transmission. But as I say, with this one requirement of a very high degree of genetic assortment on social status abilities in marriage, and so this latest paper that I've been working on, we actually find that at the phenotypic level, the correlate, the implied correlation of people matching in marriage is in the order of 0.8, it's even greater than that underlying implied genotype correlation of about 0.6.

Steve Hsu: I see, I see, I think that for my own personal history with this is I think I read your surname analysis papers, which I think might be almost 10 years old now, is that right?

Greg Clark: 2014.

Steve Hsu: Yeah. So, I saw those and then I saw your pretty high correlations, which I, I vaguely remembered, remember, it's like 0.7 or 0.8. And I just, I did a back of the, I guess I did know something about additive genetics. So, I did a back-of-the-envelope estimate and I said, well, this can't be right, because surely the implied level of assortation in mating is just way too high. It's higher than, for example, we observe in height or something like this.

And it was then to my surprise, a few years ago, realizing that no, you are absolutely right that people do actually assort that strongly. We can now see it in large-scale genomic studies for cognitive things related to cognitive abilities, and actually, it all fits together and supports what you found.

Greg Clark: Yeah. And the one interesting thing about the book that I'm working on now is that it really is going to explain social history as being a very important combination of a completely social feature, which is how do people decide who to marry. Right. And why do they match up in terms of social abilities as opposed to youth or beauty or wealth?

And it's a combination of that, but also with the fact that these abilities are actually stemming from additive genetics, and consequently are transmitted in certain very predictable ways. And you can imagine a society where people match very strongly in marriage, but only the males’ characteristics actually matter to the outcome for the children.

But the interesting thing is that there's substantial evidence in marriages here in England, that men and women actually mattered equally for things like occupational status or educational outcomes, even in a world where women were actually denied access to education, formal education, and where women didn't have occupations, particularly if they were upper-class women.

And so, as I say, the story here really is an interesting one where it's a socio-genetic explanation of social outcomes here with this very interesting issue about why people marry in the way they do. And then the other interesting question that comes up is, is this something that's unusual or is this something that's found in all human societies?

Steve Hsu: Yes now, you know, one of the hypotheses I had was that as women started to get more and more education, it would make it easier for men to, you know, for assortment to happen because it would be easier for men to find women of similar capability, maybe having attended both having attended the same elite university or something.

But I think what your results show is that actually even before women were admitted to higher education, there were ways in which, in selecting a mate, men were actually sensitive to whatever these attributes are. Even if the woman had not had a chance, for example, to go to college.

Greg Clark: Yeah, no. And so, the important thing is, in England, the general assumption in the very limited assortment literature that there is out there that assortment really has been an increasing feature of the modern world as women have acquired more education, more income, more occupational status. And then also, if we look at the data measure, the assortment is actually relatively, quite modest, even for education. It's only in the order of 0.4 to 0.5 correlation.

What we're able to do is figure out a way by looking at the father and the father-in-law to figure out how closely assorted the marital couple were. Right. And essentially saying, if there's going to be a very strong assortment, then the relationship of the groom to his own father will be very close to the relationship between the groom and his father-in-law.

And that will actually be an implicit measure of just how tightly the assortment is in marriage. And we can then implement that measure from England from 1837, all the way to 2021. And the answer is that it's an absolute constant and that the number is 0.8 is the degree of assortment. And then you could say, well, how can that happen in a world where women don't have formal education before about 1920 in England?

And here, I think the important answer is that women actually had all kinds of informal education and that in making matches, men and women seem to have valued these traits very significantly in partners. And that you can actually observe in your own life, you can tell within a relatively rapid amount of time, roughly how kind of intelligent, capable, driven, another person is and what your kind of impression of that person is. And somehow that seems throughout this period to have been the important thing in determining matches in Britain.

But as I say, it turns out to be something that has a very important influence on the rates of social mobility. And also, it turns out in the long run also on the society-wide distribution of social abilities.

Steve Hsu: Yes. So, I think we're getting into the more recent paper, which let me read the title of it. And we'll put a link to it in the show notes. The title is assortative mating and the industrial revolution, England 1754 to 2021. And this is with your collaborator Neil Cummings. And the data set here is a database of 1.7 million marriage records for England 1837 2021.

Greg Clark: Yeah. And so here, the British government reformed the marriage register in 1837. And it's essentially kept the same format all the way down to the present time. And so, the British government is actually sitting on records of 106 million marriages over this period. But by an act of parliament, it costs $15 to see any one of those marriage records.

And what they did in 1837 was for reasons that are not totally clear to me, they required people at marriage to state their rank or occupation or occupation for both the groom and the bride. And then for the father of the groom and the father of the bride. In most of those records, only about 10 or 15% of them, is there any occupation recorded for the bride. And as I say, that's why it's very hard to think about measuring assortment directly in these records. But it's pretty complete for the other parties. And it turns out that when these records were made, if people got married in a church, there was a copy of the record in the church register.

Those have more, many of them have been deposited in local, record offices. And this anarchist group of genealogists who believe that all genealogical information should be available for free, have gone to these offices and systematically transcribed all the details in these records. And we were able to find on their webpage FreeREG, 1.7 million such records. And then for the more recent years, we had to add about 10,000 records of our own, because most of these people are interested in more historical records. And as I say, they allow for an estimation of the degree of assortment in marriage going all the way from 1837 to the present. And the interesting thing is that the implied assortment is extremely strong and that it is unchanged over this period.

And so somehow a feature of marriage in England that, you know, people's names have changed. People's clothing has changed. All kinds of things have changed about Britain, but not this one kind of social constant, which is how strongly people match in marriage.

Steve Hsu: It's an amazing result. Now that I think you said that one point is rough, most of the 1.7 [million] items that you have in the database, those were only people who are married in a church?

Greg Clark: That's right. And so, in the 19th century, almost everyone was married in a church. So, there's not a cell activity issue. As we get to the current time, it's a minority of people who actually get married in a church. And so, there is some possibility of some kind of cell activity in the records. Though I, you know, I, I can't think of any obvious way where, you know, it might be that slightly higher-class status or lower-class status people are getting married in churches. But as long as they're matching in the same way as other people are in this society in terms of who they match, it shouldn't influence the results, very much. But as I say, it's just a limitation. The government is sitting on the other records, and it seems to have no intention of actually making those public. Because even though this is a public record, there are growing kinds of worries about information now in modern England, like in many modern societies. And so, they're not going to make it any easier to get access to those records.

Steve Hsu: You know, it's funny if you had, this is, this is meant humorously, but if you have the Raj Chetty superpower, you could get the British government to appoint you an official researcher with access to these records.

Greg Clark: Yes. Yes. I mean, they, the amazing thing is I don't understand why the government hasn't digitized these records and started selling them. Currently, some ancient civil servant has to go to a paper file somewhere deep in an archive somewhere if you request one of these records, pull it out, and then scan it, and then create a copy for you.

And so, you think that this would be an obvious project for, you know, some, you know, a set of people just to transcribe the whole set and then finance it by charging people instead of, you know, $15 a copy, $1 a copy.

Steve Hsu: Yes. let me, let me try to back up, just zoom out a little bit for our audience, because we've covered a lot of ground here And, I want to try to break down your results. I guess, you know, we could say in a way, this is sort of summarizing, what's going to go in, in your new book. But I think there are a number of claims here, which are counterintuitive to say most economists and economic historians, but for which I think you have strong empirical evidence for now.

So, one of them is that an additive genetic model, of, of fairly hair, you know, pretty highly heritable trait on which people assort very strongly when marrying, is the best model to describe social status in England for hundreds of years.

Greg Clark: Yes, and the other thing that I've been trying to do, but not with much success, is to see if we can develop simple, alternative kinds of social models of transmission. And say could we find models that would have similar predictions that came from a purely social system of transmission. And it turns out there is, it really is exposed that economics has no very clear model of how status gets determined. How it gets transmitted from one generation to the next. There is this whole idea of human capital of investments in human capital, but that is so elastic that you could effectively have almost any pattern of transmission if you just assume the right functional forum, and the right parameters.

And so, so one of the frustrating things here is that the simple additive genetic model is extremely well specified and has extremely clear predictions, but in some sense, it would be nice to have alternatives that you could test that against, but such clear alternatives just really don't exist when you think about say cultural transmission or social transmission or investments and children.

And so, the one thing, the only thing you can do in the book is we can test the power of this simple model using the genealogy in all kinds of interesting ways. And so, one simple example is what happens if one of your parents dies while you're young? Right. And, and if you think this is all about additive genetics, then potentially that would have very little effect. And the answer is very clear from the data. There's almost no effect on people's social outcomes from losing a parent. And it turns out there's a nice study that was done in Sweden, looking at the 1918 influenza epidemic. And the losses of parents that came from that as a kind of a quasi-random intervention, which essentially finds the same thing for Swedish society that people losing a parent actually had surprisingly little impact on their lives and future.

Greg Clark: And so, I'd say there's, there's lots of, kind of interesting things that the attitudes genetic model also implies. It should imply that birth order is not important in terms of outcomes. It should imply that family size on its own is not important in terms of outcomes. It should imply a kind of symmetry between men and women in terms of their determination of outcomes. It should imply that contact with grandparents would not change social outcomes for grandchildren. So, there are lots of things. And so, this part of the book is going to explore all of these other predictions of this kind of a simple, but quite elegant model of transmission.

Steve Hsu: Yes, I think, at least in the working papers of yours, that I've looked at, you, you address a number of these issues. And it seems to me, the evidence is very strong in favor of, you know, the Fisher model only really has, I think, two or maybe two or three parameters, depending on whether you have assortment by genotype only or assortment by genotype and phenotype. But it's a very small number of parameters and the number of tests you have of the model seems to be much larger.

Greg Clark: Yeah. Yeah. And, now I must admit, I am still somewhat amazed that you know, things like losing a parent have so little effect in terms of what, when we look at the data here. And I'm still kind of amazed that the implication is that, you know, basically, the important contribution of the parents is made at the time of conception in terms of the outcome from the children.

I mean, it does seem startling this, and I, and I presume, you know, that there would be other sites where this would not be the case, right? But at least in English society all the way through this period, there seem to have been sufficient compensating mechanisms that parents actually don't seem to have mattered in a direct fashion in terms of providing nurture to their own child.

I think you meant, providing nature to their children.

But yeah, so that's the ambition of the book then, is going to be, to find as many of these tests as possible. Now one of the things that people would find very disturbing about this if it's the correct description of English society, is that it just immediately leads to this implication that the relative fertility of different groups within the society would then potentially have an impact on the overall levels of ability within the society.

Greg Clark: And secondly, that the nature of migration, the nature of immigration, then immigration to the society would also have long-lasting effects, in terms of the nature of outcomes within the society.

But it turns out that we have some ability to look at that. In another paper, we look at it within England because, in recent years, the south has been the area of higher income and higher output per person than the north has. And then there's a long discussion about, you know, is this the result of locational externalities? Is this the result of, now, geography, disfavor in certain regions and favoring other regions? But we're actually able to show again, using a kind of surname analysis, that the people who came from the north of England who have a distinctive set of surnames actually are doing just as well as the people who came from the south of England overall.

And that disparity between north and south seems to entirely have been created by the selective migration of more talented people out of the north, into the south, and then a reverse migration of less talented people from the south to the north, mainly in the 19th century, when the industrial revolution industries offered all kinds of opportunities.

Greg Clark: And so that even things like regional disparities in England actually potentially have a genetic underpinning. And recent work using the UK Biobank actually finds that the predictor for education, is actually, has a higher value for people from the south of England than it does for people from the north of England. So, we actually know that there is a kind of a genetic correlation to which are the. successful economic areas in England now on which are the unsuccessful areas. Though, as I say, I think that kind of thing is, or that kind of implication is something that a lot of people would find themselves uncomfortable with.

Steve Hsu: Yes, I think so. I think people who think deeply about the implications of your results, are likely to really not like them. So, we can get into that a little bit later. But before we get into that, I'm just curious at the level of w without, having any particular prior or, or, or ideological motivation, who just look at the empirical data that you've collected or the methods used to analyze it, who actually have criticisms about just, you know, the steps you took to reach your

Greg Clark: Oh, I'm actually sure there will be significant criticisms of the underlying data. But sometimes I think people are going to be forced into those criticisms if they don't want to accept the conclusions from the data. And so, I know in the earlier work we did on surnames that people just said, oh, this is just selective, this is something that's unique to the top of the social distribution.

And that's in part what motivated this idea. Well, let's get data on individuals that doesn't rely on surnames and actually covers the whole of the distribution. And the results here are completely consistent with the surname evidence, right. They're producing the same underlying numbers. So, then the other criticism will likely be that look, you collected this data from hobbyists, a lot of it, most of it. and they are going to be selectively only including in the genealogies people who are relatively successful over time. Right. And so, there'll be, so for example, if you look at China it has a whole set of genealogies, but there's always this fear that they're omitting the relatives who didn't do well and only selectively counting certain relatives and the other ones just get forgotten about.

So that's as to say, why we deliberately actually went for people who in principle had actually collected, they were interested in everyone with a given surname. So that there wouldn't be any kind of an issue of exclusion of particular individuals. But, but, you know, having, you know, spent a lot of time assembling this data, I'm pretty confident that it's, it's a pretty representative set of English families that we now have in the genealogy.

Steve Hsu: So that, I mean, the first, the earlier paper for whom the bell curve tolls, I think you, you posted that over a year ago online. And the newer one, which you sent me just before this interview, I hadn't seen, but maybe that hasn't been out so long, but, and I don't know how economic history works, but I mean, have you had a serious engagement with your critics over, over the details of the calculations?

Here, I have to see it's no one or very, very few people in economic history are, people find this work kind of interesting and challenging, but no one in economic history is interested in pursuing the implications of genetic transmission of attributes. Right? It's not a topic that anyone really wants to pursue. Right. Particularly in the field of economic history, right. People are interested in long-run history. And so, as I say, it's, it's, whatever it is, it's, it's the nature of the field, people are only interested in models of some kind of social transmission. And so, I think, I'm not sure for me, it's a little bit of a puzzle. I mean, I'm not sure what the audience is for a result like this.

Steve Hsu: Well, I'm, I'm definitely in your audience.

But I'm curious, but let's say you give a seminar at a good economics department, and you know, maybe a few of the people in the audience are actual economic historians, but the other people are just old economists, micro, macro what have you, financial economists.

I would imagine they can follow your talk, and they are pretty sharp people. So, I would think, you know, something gets through, right. Even if they don't want to think about genetic transmission, you're actually making a claim that genetic transmission actually fits the data better than anything else. And do you not, do you feel like people don't take you seriously?

Greg Clark: Uh, now. So, I've given this talk, what was the reason I gave it to Moscow State recently? I, NYU, at Bocconi in Italy, everywhere. If I talk about this stuff, people say this is very, very interesting. And everywhere they can, they raise issues, you know, but not, I think, you know, super significant, issues about the estimations.

But all I'm saying is that the way economics is set up, it's not that someone says, hey, I want to go and see if we have similar results for France or for Italy or for someone else somewhere else. The subject is just not set up to really think about this as a major determinant of people's social outcomes, right.

No one is contemplating that as a move and people significantly fear the social implications of even thinking about such an explanation of social outcomes.

Steve Hsu: I see. So would you, would you say even before they get into the nitty-gritty of looking at your calculations and, you know, details of the empiric, how you collected the empirical data, they actually want to stop their program to stop short immediately when they realize you're actually starting to entertain some genetic, you're allowing some genetic causation to slip in.

Greg Clark: No. I think people are perfectly happy to hear someone discuss this. They're very interested in it. But it's just a path that no one has an interest in following. Right. That they, they have seen all too clearly that this, this has implications that will just lead to trouble. Right.

And so, and so I, I think, you know, so I haven't, you actually raised for me something that I hadn't really thought about before, which is, you know, I've given this talk and then on a number of occasions and, and always, you know, people have found it kind of interesting and very unexpected, and puzzling. But I had the thought before, you know, in some senses, why don't people then say, you know, what should we, how should we be redoing stuff to think about this? Right. And part of that I think is because people have very strong prior beliefs in economics that social interventions are powerful and important.

And I think here, I would say that we have a whole bunch of funders in economics who want to fund research that shows how to improve the conditions of poor people. And once you set up like that, it's not like we're discussing quirks and electrons where no one has an individual stake in these things. Once you set up a system like that, there's going to be an inherently, very strong bias towards publishing only results that show the effect of social interventions. And you know, once you have a whole industry of people out there looking for such effects, then you're selectively going to see a bunch of papers that happened to show these effects because the ones that didn't show the effects will just never get published.

Steve Hsu: Yeah, there's still in this file.

Greg Clark: And even in doing this research, we applied for some funding, and it got referred to the sociology panel at NSF. And there was an explicit reviewer who said, we should not fund research that shows that it's not possible to change social outcomes. We should only fund research that shows how to change social outcomes.

Yeah, I mean, so as I say, the whole enterprise, and it's a mammoth enterprise, the social science enterprise is all about changing social outcomes. So as part of this project, we have another paper where we actually, you know, because if you show that, you know, in practice, additive genetics seems to be determining social outcomes.

The comeback that people could make is well, look, myopia is also probably largely determined by genetics. But once we have corrective lenses, genetics doesn't matter anymore. Right. And so really, you know, a second question is, well, it might be that additive genetics is determining this, but can you intervene significantly and change these outcomes?

And so, we had another paper where we were able to generate data where we could test what happened to people in England who got an extra year of education in specific years because of the extension of compulsory education. And so, I think it's 1973, 1946, 1919. These were all years when the government extended the amount of education people needed to have and where suddenly a cohort of people got an extra half a year of schooling.

And what we control is that that had zero effect in terms of life outcomes for those people, right. It didn't improve their health. It didn't improve their house value. They were observed in 1999. But when we were still attempting to publish that paper, it was notable that getting this kind of zero result, it's harder to publish in this area than if we had found a 10% gain in people from a year of extra education. I'm absolutely confident that the paper would have, quite happily, glided into very respectable journals immediately.

And then when we looked and said, well, what's the previous literature on this? What you could actually see is that there are very clear signs that there is a very strong kind of publication bias, here again, that there were papers that made mistakes earlier that got published, but no one caught the mistake, which exaggerated the effect of education, because it produced an effect that was in the range that people expected. Whereas if someone had come up with a negative effect of education, that paper would never get published. Right.

Greg Clark: And so again, it's just this, this problem that the whole thinking the whole apparatus is set up to find these results. And so, I don't know how you, you have kind of independent social science inquiry, because of this fact that the journals just do not want to publish, and in particular, they don't tend to like papers that have no effect of policies.

Steve Hsu: Well, I'm starting to appreciate the situation you're in, your research is in conflict with some sacred or fundamental assumptions of your colleagues. And so, I guess you can only get a limited hearing from them.

Greg Clark: Yes. And so, it really is a kind of a double bind in terms of if you think that say genetics is mattering here, which is, first of all, it's against this kind of strong presumption that social interventions really are going to matter. And as I say, the people funding this research when the Ford Foundation comes in or the Rockefeller Foundation or whatever, they want results. They want something that they can then bring back to the public and to the board to say, look, we're improving people's lives here.

And so that's one problem. And that's the second problem is that people have a generalized fear of any explanation involving genetics.

Steve Hsu: Yes, I can appreciate that. Especially in the social sciences.

You know, one comment I would make is that, when it comes to additive genetic architecture and prediction of phenotypes, human height is extremely heritable. And we now have extremely good predictors for height, and we can measure the level of assortation at the genetic level and the phenotypic level, for couples.

And so that's a system where one could actually, you know, using Fisher's formulas and with some, you know, for example, the data that's, I think already 23andme or ancestry.com are sitting on, we could validate the entire Fisher framework very easily.

Greg Clark: I mean, one thing we do actually have data from on a very large scale is longevity. And there, the interesting thing though is longevity is so weakly inherited that you very soon run into problems of UV and estimating the effects, the intergenerational correlation, the phenotype correlation is 0.1 on the data.

And so again, you know, there we have masses of data and people have actually written papers using this data collected by amateur genealogists. But the thing is, for occupational status in the 19th century, the phenotype correlation can be as high as 0.7. And so there you're getting just a much stronger signal. Right. In terms of being able to then go to second cousins, third cousins, fourth cousins, and so on.

So, as I say, it's a difficult area of research, I think. And what I would say, anyone who's a younger person who to write on this topic that I'm writing on now, I would say to them that's suicide. Honestly, yeah.

Steve Hsu: I feel really bad that you're saying that because to be honest, I have the opposite view that, you know, assuming, you know, if say a bunch of other researchers came and validated your results and show that they hold up, with even higher quality datasets and further analysis and maybe across other countries, I think you deserve the Nobel Prize in Economics because you're actually revolutionizing the way that people think about these social dynamics. And you do actually obtain results, which are law-like which actually follow very simple, easy-to-derive equations. So, it's an incredible triumph.

Greg Clark: Right. Yeah, no, I must say, as I've gotten older and I realized then that you have declining abilities as we get older. You have to kind of say to yourself, well, how can I still make a contribution here? Right. I mean, there are a lot of very, very, very smart people in economics. And you think, well, maybe one way you can actually make a contribution is by, in some sense, being more ordinary and more immune to the pressures of the discipline. Because it's the unpopular results that are more likely to be correct and, and discoverable now, than the ones that would, that would lead to kind of immediate popularity.

Right. And so I can see that, you know, if someone like Raj Chetty finds evidence that, you know, your geography is actually significantly determining your rate of social mobility, that's a result that immediately people applaud and immediately are going to, kind of, it'll open all kinds of doors.

And when there is all this gain from that, then of course in this industry, there are going to be people who effectively, not deliberately, but effectively are going to produce these kinds of results. Right. Because if they find them, they'll get published. And if it's a mistake, the mistakes will tend not to be discovered. And so, as I say, I think because one of the questions that you would ask here is that look, Galton already in the late 19th century effectively thought that this was how social status was being transmitted. How was it the case that we're now in 2022, and it's been the case all along, and we've had a lot of data all along that we're only now coming around to thinking that maybe this is the correct thing here? Right.

Greg Clark: I mean, remember the Fisher article was 1918, right. So how could it have taken us to this point to realize — if it is correct — how could it take those to this point to realize something that, I'm saying, was absolutely the same in the 19th century as it is in the 20th century and the 21st century?

Steve Hsu: Well, it's incredible. I mean, yes. Names like Galton and Fisher are in, you know, under attack in the current environment. But on many things, of course, it will turn out they were absolutely correct. Well, and Fisher of course did tremendous things in statistics and genetics. So, I think we're just in a very wrong-headed mode right now in not being willing to look realistically at certain aspects of human society.

Greg Clark: Yeah, no. And I, as I say, I think it is a problem that is kind of almost unique to the social sciences and the way the social sciences are practiced. But also, I think, I can't think of any easy way to get out of this. Right. I mean, there's going to be a bureau of unpopular results that will find what people don't want to hear about their societies.

Steve Hsu: Well, I'll give you a, I don't know if this will encourage you or discourage you, but let me tell you how I perceive, in particular, in the specific area of cognitive trait prediction in genomics, what the current situation is — which is directly related to your interest in social mobility and things like this.

So, the predictors, polygenic predictors, for things like educational attainment and cognitive ability are already good enough that if you go to some longitudinal study, say that you have a longitudinal study, that's been going for 50 years, and they have genotypes for all 10,000 participants. One can meaningfully predict upward or downward social mobility, say between two brothers in the same family that are in this longitudinal study using these polygenic predictors.

And the data is just very clean. Anybody in genomics, who's looked at this data agrees. They know there's no disagreement about how the results are interpreted or the level of predictive power of the polygenic predictor. And that quality is just going up and up. It'll just continue getting better.

Now, it is true that within genomics you can't get any money directly to study cognitive trait prediction. You get it as a side effect of people wanting to study some medical condition or some other more fundamental issue in genomics. But gradually the data accumulated. And so scientifically there's no dispute that there is at least an additive genetic contribution to the traits that make people either successful or unsuccessful in society. And I think that's totally beyond dispute in the field of genomics.

At what point regular social scientists accept that is true and start to kind of noodle around with it, I don't know. Maybe it'll take another generation, but on the purely genomic side, that none of this is in question.

Greg Clark: Yeah. If I'm thinking about this study we've done in England. I mean, one interesting feature they have in modern societies is that we're spending about 10% of national income on education in modern societies. And I have a hunch having taught many years in universities and seen other universities operate that a bunch of that expenditure is pure waste. That is, people are, the people subjected to this education are getting very little from it.

But the very strong belief is that this is the great increase of social mobility. And so, in some practical sense, if genetics really has this very important role in some, they will say, look, you could spend that money on healthcare, much more productively than on endless expansions of the educational system.

And, and so I think there are actually significant social gains that would be made if we had a correct appraisal of what is actually determining social outcomes. And, and so, so it's not as say without some kind of, of social benefits though, as I say, I'm not sure if the current crop of researchers and social science really are going to be running out to say, okay, close a bunch of the universities.

Greg Clark: It's not doing anything, for us. And, and so, yeah, so as I say, it's an interesting question. Just that, which makes you realize the importance of the research environment.

Steve Hsu: Yeah, on this last point, just before we started recording, I mentioned to you a recent paper by James Heckman, who for the audience is a Nobel Laureate in economics with a specialization in statistical techniques. His papers with a Danish researcher and they compare inequality in educational and cognitive outcomes, comparing the United States to Denmark.

And of course, Denmark has, according to the paper, free pre-K education, effectively free college, and all sorts of other massive transfer payments to low-income families. So, it's very different. I believe beyond what any progressive American could think that they could achieve politically in the near term in the United States.

You know, the dream of, you know, a very far-left progressive in the United States, wouldn't reach the level of social programs that Denmark already has in place. And yet in this Heckman paper, what they show is that if you look at, for example, cognitive test scores of, I think they have mandatory one year of military service. So, everyone is tested upon induction, at least all the men are when they are brought in for military service. So, they have cognitive test scores for everybody, or at least all the males. And if you ask, how does that vary according to parental income rank, and how is that inequality, how is it different from the United States? It's almost identical. So, your IQ conditional on parental income in Denmark is almost the same as in the United States.

And they have other statistics like the probability of college completion as a function of parental income rank. Also doesn't vary appreciably between Denmark and the United States.

So, these are results saying, which effectively to me say, I don't think they say it very explicitly in their paper, but it seems to me, it sort of argues against the return on these investments in social programs in Denmark relative to the United States.

Greg Clark: Oh, yeah, no, I have [unclear] with Denmark and so. The student support and support mechanisms are elaborate in Denmark. And correspondingly there's a very high tax rate on people to pay for all of these benefits. And I think Heckman is a very interesting example. I mean, he is a stunningly smart researcher, and he really has kind of taken up strongly the case that early life intervention can substantially change people's social outcomes. But it turns out that they've spent a lot of time going over and re-examining some very modest kinds of experiments that were done with such interventions and, and finding results there, but not necessarily finding ones that were entirely consistent.

So, one aspect of people's lives got improved, but not other aspects. And, and then as I said, yes, this paper from Denmark really seems to kind of undercut a lot of that line of research. And I then had done my own work in the surnames book, where if you look at measured rates of, you know, social mobility in somewhere like Sweden, they seem to be much higher than, than the United States or Britain.

But then if you look at the persistence of elite and underclass surnames over time. It's at exactly the same rate in Sweden as it is in England or the United States. and, and so I'm not surprised by this, Heckman, result. And as I say, it is interesting because he has been so committed to the belief that early life interventions can really substantially change people's outcomes.

I mean, one puzzle that actually shows up in the data that we have in the lineages for Britain is, if you really thought early life interventions were very significant, then you would expect that the people who do worse in each generation would actually come from the bottom 10%. But it turns out that on average, those people are going to move upwards.

And because of the nature of regression to the mean, they move upwards more than any other group. And so, the data doesn't suggest any kind of significant poverty trap. It says that even the people at the bottom will eventually make their way to, they average their descendants in this data.

And so, as I say, there's a lot of reasons in this data also to think that really bad adverse circumstances of childhood actually have surprisingly small effects on people's final outcomes.

Steve Hsu: Yeah, I think, I mean, I, well, maybe I'm biased, but I think if you look at the totality of all this economic mobility data, social mobility data, I don't see anything that's inconsistent with at least some partial genetic influence on outcomes. And, you know, you can ask what the balance is between, you know, the environmental and the genetic component, but certainly, it would be hard to deny that there isn't a significant genetic component.

Greg Clark: Oh, yeah. I mean, that's very clear from the twin data and other data. But as I say, this question still comes up, well, you know, it may be there in practical data, but what if we launched a real Danish type of intervention, right? What have we actually supplied everyone in the United States with quality schooling and quality university access and stuff like that, could we actually very significantly change rates of social mobility?

Steve Hsu: Well, that's why this recent, this is a 2021 paper from Heckman. I think that's why it's so interesting because, as I said, you know, even Bernie Sanders can't imagine us reaching Danish levels of social intervention here in the United States. And yet you can then look at and see how much they've reduced, at least on these measures, cognitive measures, and years of education measures, they haven't there, they're no more egalitarian in those outcomes than the United States. That's an amazing result.

Greg Clark: Well, it's of some interest to me because I'm about to head to Denmark to take up a position there where the proposed project we're going to look at is one where we can actually identify for a lot of families their status in the 19th century. And then we can hopefully link them to the most contemporary data in Denmark. And then the Danes actually have a very kind of sensible attitude about allowing anonymous access to this modern data. And so, you know, we should be able to have some very nice study of, you know, what has happened over 150 years to people who started off in relatively privileged positions and underclass positions in terms of height, you know, very lots of different social outcomes that people have at the contemporary time in Denmark. And actually, my prediction is that it will be exactly like Britain, that there will be no difference between these societies.

Steve Hsu: Yes. Well, I want to congratulate you on that move to Denmark. And I agree with you there. I think the record-keeping there is extremely strong. For example, I mentioned that they still have military service, and they have IQ test scores for everyone that is inducted into the military. So that's effectively all the males in Denmark.

So, in all kinds of ways, I think they also have very good medical records as well. So, in all kinds of ways, I think you can do really, really good there.

Greg Clark: I noticed that one thing I was looking at when we were doing the surname evidence, I thought it would be interesting to link up stuff to people's criminal records. But it turns out you can't do that in somewhere like Britain, because deliberately they actually want to suppress those records for some, in some ways, very good reasons.

But I know in Denmark they have a complete catalog in the government records of everyone's convictions, and that people have done research about how, how it links to adopted versus non-adopted children. And I, I think there's been no release of any individual information through this kind of policy they've had of allowing researchers access.

And so really, you know, Scandinavia has really become the laboratory for doing a lot of research now into things like social mobility.

Steve Hsu: Yes. I know that Finland and Estonia and a number of these countries have really good digital records across many different fields for each individual. So, ranging from, I didn't know about the criminal record part, but medical records, things like IQ scores, incomes, all these things are actually I think, readily accessible.

Greg Clark: Yup. And so, I say it is a little problematic, but there's one group of societies, which has a very distinctive social structure, is the one that you can actually then study things like social mobility. So, we'd like to do this, you know, in places like China or Russia, but hopefully, it's still in some ways representative of the human condition.

Steve Hsu: Yes, but you know, the interesting thing about the Scandinavian countries is even though they are distinctively progressive in terms of their social programs, if like Heckman you find that the impact wasn't as large as people thought, that's an incredibly important finding.

Greg Clark: Absolutely.

Steve Hsu: Yes. Well, one thing I must say, Greg is you seem very upbeat for somebody who's done what I consider incredibly important work but hasn't really gotten the recognition that I think you deserve. You seem very upbeat and unbowed by the fact that your colleagues are ignoring you.

Greg Clark: Well, I think again of selection mechanisms, if I really wasn't upbeat, I wouldn't be here talking to you. I would have buried this work or gone in some other direction. And so, I think, for some of us, I mean, it's, it's just interesting. It's intellectually exciting. I end up meeting a lot of people from other disciplines outside of economics, and that's actually interesting and stimulating. And I do still think that the data is good enough and the methods are straightforward enough, that there is actually a contribution to be made here that will actually have some kind of lasting, lasting value.

Steve Hsu: Yeah, this is why I kept sort of probing you on whether people had really actually confronted you with really substantive criticisms that made you revise your work. In the absence of that, then, you know, one can only hope that in the long run, serious people who are interested in these questions will find your papers and books and realize someone figured it out back in the 2010s.

Greg Clark: Yeah. Yeah. I thought when the work on the whole genome studies would actually significantly affect the way people thought about some of these issues, but now I'm, I'm not so optimistic because I mean, for things like education, it's still the case that you get as good a predictor of education as just asking what was your mother and father's education, as going to these genomic studies. And it looks like you're going to need an enormous amount of data to get a predictor that's even just as good as the simple one of looking at people's relatives. Looking at some of the other things like political attitudes or other stuff like that, it didn't seem to me that that was really going to make its way into economics as a kind of a viable control in terms of, of, doing studies on people or as a predictor in terms of people's outcomes.

Steve Hsu: Well for cognitive traits in genomics, as I mentioned, no one's really trying very hard to get the answers because again, even there, so even in genomics, there are sensitivities when you come to cognitive traits. There are no sensitivities when you're talking about medical conditions or height or things like this.

But for cognitive traits, there is sensitivity and hence the field is not trying very hard at
all to solve that problem. And hence they don't really have good data. So, if they had, I mean, based on the mathematical modeling that we've done, I'm pretty confident that if they had just 1 million genotypes where they had a decent IQ score, like an SAT score or something, or even your, what is the thing in England GCSEs? Even if you had that for a million people in their genotypes, you could build a predictor that correlates maybe 0.6 or something with the actual IQ score.

And yeah, that would be huge. Now it is true. What you said is that if I know nothing about you and I either tell you what your mom and dad's level of education is, or I give you the current polygenic score, based on your genotype, those two are comparable in terms of the level of predictive power that you might get.

But keep in mind this other scenario where you're comparing siblings. So, they have the same mother and father, so that information about the mother and father's educational attainment is already there. It's controlled for. And then I ask which of these two brothers is going to be upwardly-mobile, and which of these two brothers is going to be downwardly-mobile, for that polygenic prediction is useful.

And, as you know, I'm a co-founder of a company that does embryo genetic testing, and we don't offer cognitive testing for the embryos because it's too controversial. We offer only medical, you know, specific risk scores for medical conditions and things like this. But the level of interest, and this is what you economists would call revealed preferences or revealed understandings about how the world actually works. The level of interest among parents who are doing this kind of genetic testing and IVF in getting the cognitive scores is through the roof. And actually, the more highly educated and knowledgeable the parents are, the more they want to know what the cognitive scores are of their embryos. But we don't give that information, but it's a reveal preference there. So, people understand the reality, even if your colleagues in the economics department don't want to know about it.

Greg Clark: Yeah, now one thing I have seen in this genealogical data is that you would think that dating services would be very interested in the correlation you observe there because it turns out that you want to predict a child's outcomes then it's not just the parents. The grandparents, the uncles, the aunts, and the cousins all contained significant protective power.

And so, you will think that someone would be setting up and say, I can score the breeding value of your potential mate here if you just give me a set of genealogical information. And, because as I said, I mean, the services are already kind of scoring people on all kinds of other metrics.

Steve Hsu: You might see that eventually. I think people do it the old-fashioned way where, when I, you know, you introduce your girlfriend to your mom, your mom immediately starts asking her about her parents and family tree. But yeah, you might see it as part of a habit.

Greg Clark: Right. Yeah. No, as I say, the ideology of our society is that, that you, that all of that irrelevant, right. That it would just be these, these potential [unclear] partners would, would, they would have all the information.

One of my students has an interesting paper, you know, Quebec has very good genealogical records and those records actually tell you in the 19th century whether the parents or the children are literate or not.

And because of the accidents of childbirth and early death, there are lots of families where there'll be two mothers, you know, three children with one mother and then four children with a second mother. And he's actually able to show those records that even controlling for the father, if one mother's literate and the other one is not, then that gets transmitted to her biological children, but not to the other children.

So that even within the family, the mother who raises you has less influence often than the mother who biologically bore you. And so, so as I say, there's some kind of fascinating information in these records of what it actually matters to child outcomes. And the other thing that's nice about that data is that the effect of fathers and mothers is symmetrical in terms of transmitting literacy.

Steve Hsu: Yeah. I mean, it always comes back to something which is more or less consistent with the additive genetic model. So, in the same way, these huge large-scale adoption studies with, you know, hundreds of thousands of adoptees, they always show that the adoptees are much, much more similar to their biological parents than to the adoptive that they are.

Greg Clark: Yeah. One thing I should say with the English data is the one thing that is different in terms of its transmission is wealth, but for families that have wealth. And so, their family size matters and they're also, it's your father's wealth that's more important than the wealth that's coming from the mother's side of the family.

And so there, you can actually see that there are other types of transmission that we can observe in the data here. And, and ones which clearly have been influenced by some kind of social rules. But the important thing about wealth is that it's actually physically transmitted from parents to children.

Steve Hsu: Yeah, it's something you can just directly hand over. It's not like that when the sperm and the egg meet after that, there's nothing you can

Greg Clark: Right. But, interestingly in the data, wealth tends to adjust in the long run to people's social status. And, and so if you come from a big family, you didn't get much wealth transmitted. You'll tend to accumulate that if you have relatively high social status. But if you get a lot of wealth and you have relatively low occupational status, that wealth will tend to dissipate.

And so, it's true in the short run that wealth shows that strong social effect, but actually, in the long run, it remains actually highly correlated with people's other measures of people's social status.

Steve Hsu: Yes. So, I've taken up a lot of your time. I think we've been on for over 90 minutes. I, I just wanted to mention one thing, which we, you mentioned very briefly in the, at the beginning and, we, we, we diverted into other topics. But I want to emphasize this again.

In your most recent paper, one of the consequences you point out of assortative mating is that it increases the variance of whatever these traits are in the population. Because if below-average men on whatever this trait is, tend to marry below-average women, that creates more population on the left side of the distribution. And if above-average men tend to mate with above-average women, it pushes things further to the right, in the right tail of the distribution.

Steve Hsu: And that may, I think you're pointing out that may have important consequences, for example, for how the industrial revolution got going.

Greg Clark: Yeah, no this is, and this is purely in the realm of speculation. But we've had a long struggle trying to understand why the long delay in the industrial revolution and why in Europe, as opposed to China. And it turns out when I saw this data on assortment in marriage, in England, it did occur. It's possible then to kind of plot out, well, what happened if that type of assortment actually only occurred, say, in 1400. That people started assorting in that way. And earlier they had only half the level of assortment. It turns out that it will have very big effects on the distribution of abilities, not the average level of abilities, but the distribution. But it will take something like 400 years for half of that effect to come in.

It has to distribute through the population and get to a new equilibrium. And so, it's an interesting kind of teasing possibility that maybe one of the features that generated the modern world, and the possibility of the modern world is something as mundane as what type of people got married. Right. And then they actually did some investigation and looked at some pre-industrial societies. And so, there's somewhere, this is actually common in the Islamic world still, but it's also true of Amerindian groups in South America where very commonly it's a particular cousin who's the desirable marriage partner. And you can actually calculate in equilibrium what would be the correlation in social abilities with such a type of marriage. And we normally think of cousin marriage as, oh my God, these people are inbreeding. It turns out that the correlation in social status would be 0.24 if you just were to marry a random cousin, and that was the pattern of marriage and the society.

And so, you would actually get much less assortment and people were achieving in England through just the ordinary mechanisms of marriage there. And then also you can calculate, well, what would happen if the parents were the ones who match in marriage, but they do it when the children are still very young.

And so, the children haven't revealed really what their kind of social potential is. And it turns out that will still only get you to a correlation of about 0.4. And so, it turns out a marriage pattern for the, as in England for the parties to choose each other independently of the parents and actually chooses based on their social attributes is historically actually relatively unusual and would actually have these consequences in terms of the population distribution of abilities.

And so, as I said, it's, it's purely a hypothesis, but I think it's, it's an intriguing possibility again for how we might have transitioned from this long pre-industrial static world into the dynamic modern world that we now find ourselves in.

Steve Hsu: Right. So, independent of the speculative part about how it influenced economic development or technological development. I think just the simple mathematical point that you made, which is that having a free marriage market where young adults can choose their mate and assuming they assort the way that you see them as sorting in the data, that is the best system for populating the far right

Greg Clark: Yes. And an important thing about the nature of marriage and Northwest Europe is that people delay. And so, the average woman's only getting married at age 25 and the average man at 27. And so, it means that they're substantially independent of their parents, and they also have some experience in the world before they get married. And, and so that actually has to say, may have been a factor in producing this particular type of assortment. Whereas, you know, in the Roman world, many girls get married at age 10. And so that's a very different type of marriage.

Steve Hsu: Right. And I just wanted to couple it with the older work, which is what first got me into contact with you. Is that, okay, you're populating the right tail. You're increasing the variants, but you're also populating the left tail under these conditions. But if the right tail is out reproducing the left tail by two to one or something, which is maybe the result of your earlier work, then it's pushing everything in the right direction.

Greg Clark: No, no. So, another part of the book that we haven't talked about that I've been working on though, is it's possible for people who went to Oxford and Cambridge to actually get pretty good estimates of their net fertility, all the way from 1700 up to the 1980s. People born in the 1970s, sorry, would be the last decade. And you can definitely see, and it's still the case in 18th century England, that people with these higher levels of education were reproductively much more successful.

But then you see that for people born in the generation of the decade of the 1850s, that process suddenly changes dramatically. And this upper tail in terms of educational attainment actually becomes extremely unsuccessful reproductively. And so, they go from the 18th century of having three children per family, per person, down to having one by the time we get to the decade of the 1850s.

And, and so as I say, if it is additive genetics that's determining the kind of social outcomes, then you actually see, as I say, very different phases in British history in terms of what's happening to the top tail of abilities within English society.

Steve Hsu: Right. I mean, just to spell that out for the audience, I think what you're saying is that, so we still have in place the assortation that leads to a wider variance in populations of both tails, but now the elites have very low fertility, and the non-elites have, I think, much higher fertility, at least in some countries. I don't know about the UK. But, but anyway, that could push things in exactly the opposite

Greg Clark: Yes. So, essentially in British history, that is definitely a phase, a long phase leading up to the industrial revolution where you've got super fecundity and population level, that's having big effects. And then, there's a phase going from people born in the 1850s up until the 1920s. When these elites have very, very low fertility and then actually in England, it's a little hard to tell now it looks like it may have kind of equalized, right?

And so, these things mainly are kind of historical patterns, but they're big enough effects that they would actually have left a trace in terms of the distribution of abilities and the society, and also the average level of abilities. And then also what's happening at the upper tail of these abilities.

Steve Hsu: I see, probably you're familiar with these decode results, in Iceland, but I think they found the loss of something like one or two IQ points per generation because of quote dysgenic trends.

Greg Clark: I think I saw that study. There's someone who's done a seminar study using the Biobank data in Britain, and he was finding negative effects recently, but they look pretty. Right. And in terms of their effects. So, thank God. This is not something that we have to confront, at least, in modern Britain.

But I'd say when you look at the data, so I have a chart which shows how many descendants someone who was a laborer would have from 1700 versus someone who was an Oxford graduate, at some stage, I think there are eight times as many descendants for the Oxford graduate as for the laborer, but then it actually reverses.

But it's still the case relative to 1700, that there are more descendants, from that assuming their children remained in that class all the way through which is a counterfactual. There would still be more descendants of the elite group around in Britain now than there would be of the laboring class.

Steve Hsu: Yeah. When I encountered those numbers like that from your work 10 years ago, and I just put in very, very mild levels of heritability of traits, I could see very, very rapid evolution happening due to these differential reproductive rates.

And that just shocked me because I had, you know, been educated in the United States that there had been no evolutionary change in humans in the last 50 years or so. So, when I realized like, well, identified effects, like yours could lead to things on timescales of a thousand or few thousand years easily, I realized that this had to be rethought.

Greg Clark: Yeah, no, I mean, it was additive genetics going through, you know, a thousand or 2000 different [unclear]. It's easy to get a very rapid change in the distribution at those [unclear], and then the average values. And so, it really does say that you know, with the social structures we have, these changes could actually come within a couple of hundred years. You can actually have very significant changes in the population.

Steve Hsu: Yes. If I take the higher-end numbers, not the sort of really conservative numbers. Yes. A few hundred years is even possible.

And those higher-end numbers might actually be the realistic ones. I was being conservative when I made my first estimates. But you might know this already, but for the listeners, there's something called Fisher — the same Fisher — Fisher's fundamental theorem of natural selection in which he writes out a dynamical equation for how fast natural selection can act to shift traits in additive traits or traits in general, in a species. And it's dominated by the additive variance, the number of variants of that trait that's controlled by additive effects.

Steve Hsu: And because as you just pointed out, just changing the frequencies of those, [unclear], which have the additive effects is the fastest way to evolve a species. And Fisher knew that it was his most famous. If you're an evolutionary biologist, that's his most famous equation.

Greg Clark: I should go back and look at that again. That sounds interesting.

Steve Hsu: Yeah. It's called Fisher's fundamental theorem of natural selection. So, in a certain community, it's extremely famous, but most people outside that community don't know of it.

Well, so I think I've kept you long enough and we've covered quite a lot of topics. So, I think maybe we should just end here, and I want to thank you again for being a guest. And I think the audience will really enjoy this conversation.

Greg Clark: Oh, thanks a lot. It's been a lot of fun rehearsing these things, and it reminds me that I really got to get back to work, to finish this damn book.

Steve Hsu: Yes. We're all waiting to see it.