Episode 63: Brian Boutwell on Twin Studies and Heritability

• Links to this episode: Spotify / Apple Podcasts

Unofficial, reviewed transcript. It may still contain mistakes; check the podcast when wording matters.

Speakers are labeled by first name where identified. Unidentified-speaker labels may refer to different people. Bracketed question marks indicate uncertain wording.

Transcript

[00:00:11]  Bruce: Welcome to The Theory of Anything podcast. Today we have a guest on the show, Brian Boutwell. Did I say that right?

[00:00:21]  Brian: You got it.

[00:00:22]  Unidentified speaker: All right.

[00:00:23]  Bruce: And we got Peter here also. How's it going, Peter?

[00:00:26]  Peter: Hey, Bruce.

[00:00:28]  Bruce: So Brian and I started talking a while back, and he, like a lot of us, he read David Deutsch and became very excited about a lot of the ideas. But Brian is interesting because he actually works in the social sciences. You would be technically considered social sciences or is it just adjacent to social sciences?

[00:00:49]  Brian: Probably either or. Adjacent or in the social sciences, behavioral sciences, I think any of those labels would be appropriate.

[00:00:59]  Peter: It just occurs to me that Brian and I are both the sort of heretics in the David Deutsch world in that I'm a public school teacher and he's a...

[00:01:09]  Unidentified speaker: Yes.

[00:01:10]  Peter: Sociologist who studies genetics.

[00:01:13]  Bruce: So Brian sent me two articles and one is called Behavioral Genetic Methods and that's by Willoughby, Polderman, and Brian himself. And the other one, and that one was in Nature. And then the other one was Meta Analysis of the Heritability of Human Traits Based on 50 Years of Twin Studies by Polderman and a whole bunch of other people whose names I can't pronounce. And I read through those papers and not being in the field, I ended up with lots of questions, but they were very fascinating papers. And as Peter just mentioned, there's definitely a feeling amongst a lot of fans of David Deutsch that certain, that I don't know if I'd say that they dismissed the entire social sciences. Some of them maybe do, but I think some of them take parts of them seriously. You might hear them mention psychology or something like that. But definitely there seems to be some skepticism around a lot of these fields.

[00:02:17]  Bruce: And these papers really kind of lay out what the twin studies were, how those types of studies are done. And I just found it super fascinating. So I wanted to invite Brian to the show and try to ask him questions about stuff so that I can understand this better. And really kind of dig into it. And then Brian and I have also talked a lot about how do you apply [Popper's?] epistemology to social sciences? Which is a really kind of a tough question. And I don't know if we're going to be able to get to that subject. I just asked Brian before the show, maybe we'll need two episodes so we can do that second subject as its own episode. But this is what Brian and I have been talking about. And he's got some really interesting ideas on both of these subjects. And so that's why I invited him to the show. And honestly, it's kind of me just being selfish.

[00:03:07]  Bruce: I wanted a chance to say, hey, Brian, can you like explain these things to me better? There's a lot of things in here that are kind of technical. I'm a layman. I don't understand them. And I really wanted to ask Brian questions and try to understand these articles better. So that was why I invited him to the show. And I thought, hey, let's record this. It can be a podcast episode. And I suspect I'm not the only person who would be interested in this subject. So with that introduction, Brian, why don't you tell us a little bit about yourself and how you discovered David Deutsch and kind of your interest in both David Deutsch and in the field that you're in.

[00:03:45]  Brian: Yeah, certainly. And it really is a pleasure to get a chance to join you guys and talk about these things. Like you mentioned, I found Deutsch's work within the last handful of years. And I sort of arrived at it just via the recommendation, for lack of a better word, of a colleague of mine. And it really did just sort of provide a jolt to my intellectual life in a really wonderful way. So I guess I'll just pause on that briefly to say, so I'm just in terms of professional affiliations, I'm an associate professor in the School of Applied Sciences at the University of Mississippi. I also hold a secondary appointment as a professor of population health in the John D. Bower School of Health at the University of Mississippi Medical Center. I was a visiting scholar at Cambridge University in 2022, and have been at the University of Mississippi now for, this will be, four years.

[00:04:54]  Brian: I've been in academia for, oh gosh, this is when you start to feel old, about 13 years. As Bruce was describing, the work that I do is broadly devoted to trying to understand aspects of human nature. Sometimes that involves kind of the more regrettable aspects, violence and aggression, and then it can, you know, spans over into fields like medicine. So I do work on diabetes type two, general aspects of health and wellness, longevity, health span, and then also components of psychiatry, different forms of psychiatric problems and how they impact functioning and behavior. And all of that is really done using the toolkit of behavioral genetics, which is how I think it's a productive way to think about behavioral genetics, twin studies, things that we're going to discuss as in some ways, they exist as a field unto itself. But in many other ways, it's simply one among many toolkits that try to, or rather that are available to researchers, whether it's medicine or economics, as among other things, a way to improve causal inference capacity when you can't do a randomized experiment, which is pretty common in terms of the types of questions we study, unethical to do those things, which we'll get to all that, but that's sort of a potted career bio.

[00:06:41]  Unidentified speaker: Thank you.

[00:06:42]  Bruce: And that was actually what I found so fascinating about these papers, is let me see if I can kind of explain my own background on this. A lot of my friends online that are fans of David Deutsch, they seem very, very skeptical of any sort of scientific theory if it doesn't have a explanation that's mechanical. So in past episodes, I don't know if you've heard these episodes or not, Brian, but in past episodes, we talked about some of their stances on universality versus IQ, let's say. And they have really big problems with IQ because as a theory, IQ doesn't even attempt to explain why some people would be smarter than others, right? It simply makes the claim that some are, uses maybe tries to measure G factor, but it's not like a highly explanatory theory, if that makes any sense. And so they're very skeptical of any sort of theory that doesn't have that kind of deep mechanical explanation.

[00:07:49]  Bruce: Now, I've used an example on the show of how that just isn't really how I understand critical rationalism. Critical rationalism allows for there to be a hierarchy of explanations. So if I say, and this is the example I've used over and over again, if I say, you know, oranges stop scurvy, that is an explanation, even though there is no mechanical explanation for why oranges stop scurvy. Now, it's not a deep explanation, but it is an explanation, and it's a completely valid scientific explanation because it can be tested. Now, if I were to improve that explanation to vitamin C stops scurvy, I'm now starting to get more into the mechanism, not very far, but a little bit more. So that's a more empirical theory than the first one, and therefore is considered in critical rationalism the better explanation. And then we could go even further. We could get into vitamin C stops scurvy because it affects the connective tissue, blah, and we can come up with this really kind of deep explanation that's way more testable than the other two.

[00:08:51]  Bruce: And yet, if you falsify it, you falsify the other ones. And so they exist as a hierarchy where the explanations get more and more intense and therefore more testable. And that's how I understood critical rationalism. I didn't see it as you can dismiss a theory because it's, quote, non-explanatory. Most theories have at least a little bit of explanation. And therefore, as long as they're testable, they're there. And I've argued this with a lot of people online, and this seems to be fairly consistent that I am on the outside on this, that most of my friends of David Deutsch that consider themselves critical rationalists disagree with me on this. And yet, this is how, when I read Popper, this is what I understand Popper to be saying. I'm not sure where they're coming up with this alternative view because it doesn't seem to be what Popper is saying to me in his books.

[00:09:43]  Bruce: And one of the reasons why I had some concerns with this is they will often say, like I'll mention a theory that is low explanation, like let's say IQ theory. And they'll say, not only does it not have an explanation, but you're just doing correlations. And I'll say, no, I'm not just doing correlations. We're talking about causation. And they'll say, no, it's just correlations. It's no different than, and they'll use examples from the Medieval times, it's no different than trying to look at humors or whatever and trying to come up with explanations. Say, no, this is like nothing like that. We're talking about actual causal explanations. Then I'll usually cite Judea Pearl's work. And I'll say, we can test causation using randomly controlled trials. The fact that I'm listing something as a correlation, if it's done within the right inquiry as part of the right experiments, we can actually test whether something causes something else using randomly controlled trials.

[00:10:48]  Bruce: And Pearl lists out a whole theory of how we do that, how you go about trying to cut the connections. It doesn't require that I have a deep explanation of the mechanism. I just have to know that this causes that. And then I can, through a randomly controlled trial, come up with an appropriate experiment to test if there is a causation there or not. And again, I feel like I'm kind of on the outside on this, that most of my friends who consider themselves critical rationalists disagree with me. And again, it seems to me like I'm just going with what Popper wrote. I don't think Popper had any problem with what Judea Pearl would have said, right? When you sent me these papers, the thing that I found so fascinating was that they're really about how do you test for causation when you can't do a randomly controlled trial? So we're kind of two steps in now.

[00:11:39]  Bruce: First, you've got those who are skeptical of randomly controlled trials, which I am not. Like I totally get that you can test for causation using randomly controlled trials. The people who claim you don't, I don't feel like they're even taking a scientific attitude, to be honest. Or they just don't understand it. They need to go understand Judea Pearl's theory. I'm not sure what it is they're missing. But I can see why there would be more skepticism over if you couldn't do a randomly controlled trial. That is, that's harder for me to wrap my mind around, like I can easily wrap my mind around why a randomly controlled trial tests causation, where it's much harder for me to wrap my mind around how you would do that if you can't do a randomly controlled trial. And that's why I found your paper so fascinating. But they were also very technical.

[00:12:25]  Bruce: And as a layman, I found them a little hard to understand and to get my mind entirely wrapped around them. That's kind of my background on why I found your paper so fascinating. Does that make sense?

[00:12:35]  Brian: Oh, certainly. And it makes good sense and a lot of interesting content in there to unpack. So it'll be fun to try and talk through some of these concepts and see if they become compelling.

[00:12:50]  Unidentified speaker: All right.

[00:12:51]  Unidentified speaker: Sounds good.

[00:12:52]  Bruce: Well, let me just start with, I'm going to quote from your papers. And I'm going to, in some cases, I'm going to ask you interview type questions. In some cases, I'm just asking for myself, can you explain this? Because I didn't understand it and I'd like to understand it better. So I don't think I always wrote down where, I know when I'm quoting you from the articles, but I'm not sure I remember always which of the two articles I'm quoting. So I will do my best to give the quote. And if I need to, I'll try to figure out which paper it came from. But when I was taking notes, I was just like jotting stuff down quickly because I was excited and I didn't maybe make it into great interview questions. To some degree, this show, people, if you're listening to this show, you expect our show to be somewhat technical and to be about trying to figure these things out.

[00:13:40]  Bruce: This isn't strictly like, in many ways, this is not an interview of me to you. This is me trying to understand these topics better and you're here to help me do that.

[00:13:51]  Unidentified speaker: If that makes any sense.

[00:13:53]  Brian: It does. Hopefully, I'm up to the challenger and capable of providing something useful.

[00:14:00]  Unidentified speaker: Okay.

[00:14:01]  Bruce: Right at the beginning of one of the Polderman, both articles are from Polderman. If I say the Polderman article, I really mean the one that you weren't a co-author on.

[00:14:10]  Unidentified speaker: Gotcha.

[00:14:12]  Bruce: It says, the relative importance and specific nature of influence of genes and environment on human traits remains controversial. So right up front, he's saying this is a hotly debated topic. Maybe he also meant politically debated topic.

[00:14:26]  Unidentified speaker: I don't know.

[00:14:26]  Bruce: I didn't get that from reading the article. But could you talk about why this would be, what would be controversial about this topic? So we're kind of just upfront, putting the elephant out in the room. Why is this a controversial topic? Why would scientists feel like this is a debatable topic?

[00:14:47]  Brian: It's an interesting question. So I think in the way that controversy is sort of invoked in that situation, it captures a variety of things related to why behavior genetics can sometimes, you know, spark some debate and sometimes heated debate. And so one aspect of that is really related to, I think, some lingering misunderstandings about certain concepts, in particular the word heritability and sort of the completely understandable, but regrettable tendency to confuse that word with inherited. And we may be [planning to?] talk about both of those, so I won't go into a ton of detail at this point. But they aren't the same, even though they're certainly related. When we're talking about heritability, we're describing something different than when we talk about the inheritance of a particular trait. And so we can talk about that distinction, but it is a really important distinction.

[00:16:06]  Bruce: You're right that I have some questions on that, but why don't you go ahead, and now that you've brought it up, why don't you make the distinction and help everybody understand the difference between heritability and inherited?

[00:16:17]  Brian: Oh, certainly. Probably everybody can remember back to being in a high school biology class or maybe an intro to bio class in college. You're working with or learning about Mendel's work with peas. Remember doing those Punnett squares where you're crossing a smooth pea with a pea that tends to be wrinkled, and you map out the disappearance and reemergence of traits like the smoothness of the pea inside the pod kind of thing. So a lot of time, I think for a lot of folks, that's when we start to talk about genetic influences. Those are the types of examples that come to mind, because it sort of makes sense, and Mendel's work was obviously foundational, especially when it was merged with Darwin's insights about natural selection and sort of made the two mesh well and explain, you mentioned sort of mechanism earlier, it helped better explain the transmission of traits from parents to offspring.

[00:17:32]  Brian: And it helped answer a variety of questions and was one of those great sort of scientific achievements, in my opinion. And so the trouble is, when we talk about heritability, it's easy to imagine inherited, but heritability is simply a statistic. It's one of three parameters in behavioral genetics where we subdivide variance in some trait. So it's a way of understanding, along with two types of environmental effects that we'll talk about, why members of the population differ from one another. So in other words, let's imagine that we've got to, we'll keep the example very oversimplified at this point, a small group of individuals, and we're interested in how tall they are. And one of the obvious things we'll notice is that they are different in how tall they are. Some are relatively taller, some are relatively shorter. There's variation in height.

[00:18:41]  Brian: So at a basic level, what twin studies do, sort of the classic twin study, as we refer to it, is it partitions the reason for that variation into three factors. The first is heritability. The second is what we call shared environment. The third is also a type of environmental effect. We call it the non-shared environment. And so we can go through those in a moment. But the point about heritability is that it's simply an estimate of how much of that variation is the result or is rather associated with genetic differences that exist between those individuals in our example, our height example. And so some of those differences are going to be linked to genetic differences. Other differences are going to be linked to different environmental experiences.

[00:19:35]  Unidentified speaker: Like nutrition or…

[00:19:37]  Brian: Precisely, that's right. And so the twin studies just work to partition variance. And heritability is one part of variance partitioning. But the reason we don't want to confuse that with inherited is it makes sense logically and mathematically to talk about a trait or some particular trait having about 30% of the variance being heritable. It does not make sense to say that 30% of the trait is inherited. That's incorrect and really even sort of nonsensical.

[00:20:18]  Bruce: Let me see if I can try to summarize what I think you just said. So certain traits may be inherited. Like what's the classic example of blue eyes? I don't even know if that's actually true.

[00:20:30]  Brian: Well, Mendelian traits is sort of how we would refer to it where there's a single gene kind of situation.

[00:20:38]  Unidentified speaker: That causes it. [Overlapping wording unclear.]

[00:20:39]  Brian: Exactly.

[00:20:40]  Unidentified speaker: That's right.

[00:20:41]  Unidentified speaker: So that's what inherited is.

[00:20:43]  Bruce: So the classic example is blue eyes. I don't know if that's actually true that blue eyes are inherited as a Mendelian trait, but that's the one they always use in school. So let's pretend like it is in any case. So if blue eyes is determined by a single trait, and then you either have it or you don't and it's recessive, then that would be an inherited trait, whereas height is not strictly inherited because there's a whole bunch of different factors. It's really more a matter of heritability.

[00:21:13]  Unidentified speaker: Right.

[00:21:14]  Brian: And so there's quite a few things wrapped up in that. When talking about something like eye color or, again, just going back to the pea plant because it's familiar to a lot of folks, smoothness versus wrinkled peas, that is the proper place to use things like inherited. Part of the reason height becomes inappropriate in a situation like that is it's what we call a polygenic trait. There are hundreds, likely thousands of genetic variants that play some small role in explaining why some individuals are taller than others. And they do that in conjunction with the types of environmental differences that you mentioned, things like nutrition or deprivation. And sort of together, that helps us better get a sense of, if it's height we're interested in, why some folks are taller than others in some population of people.

[00:22:12]  Bruce: Now, in the article it also says, we show that a majority of traits, 69%, the observed statistics are consistent with a simple parsimonious model where the observed variation is solely due to additive genetic variation. Now, that wasn't a term I had really heard before. I could probably guess as to what it means just by the choice of words being used there, but I would prefer not to guess. So, what are additive genetic variations and how do they differ from non-additive genetic variations?

[00:22:45]  Unidentified speaker: Yeah.

[00:22:46]  Brian: So, this is also a good point to go back to that word controversy because this is another form that controversy can sometimes be too strong of a word, where some very interesting and useful debate still occurs in quantitative genetics, behavioral genetics, and it is this issue of additive versus non-additive heritability. So, additive is very much like what you might intuit. It refers to, you know, if we have some complex trait, maybe height, say there's a thousand genetic variants, or you might hear me slip and use the word allele. It just refers to different forms of a particular gene, say. There are a thousand alleles linked to variation in height. The effects of those alleles represent the sum of their individual influence. So together with the environment, they have independent and individual effects on whatever outcome we're interested in.

[00:23:56]  Brian: Non-additive refers to interactions between those alleles, so that maybe having one allele magnifies the effect of another on the trait we're interested in. And so really, I think that we can certainly go further into it if you like, but I think that just understanding that is sufficient to sort of get the idea that twin studies certainly are sensitive to the idea that sometimes there are important non-additive effects to think about for genetic variance or heritable variance, and they can calculate those. We call it dominance effect, dominance heritability. And we can estimate if we have reason to suspect that they're present, we can adjust the models and try and get a sense of that. But there is, in returning this idea of debate, there has been a fair amount of work, and we allude to it in the paper, both using just theoretical insights, simulation models, and then a variety of evidence generated from twin studies over the years, where we can be reasonably, at least I'm reasonably comfortable with the idea that most of the heritable variance we observe is from that additive model, where individual alleles are all contributing their individual small effects.

[00:25:31]  Bruce: So it seemed to me that the paper made a big deal out of the fact that the majority of traits were additive genetic variations. But I didn't necessarily understand, like since I'm not in this field, I didn't necessarily understand why that was a big deal. So is it just because that means that the model is quite simple in most cases?

[00:25:53]  Brian: That's, I think, fair to say part of it. It's certainly a nuanced debate. A lot of it sort of hovers around methodological concerns that our models are giving us misspecified or incorrect results or biased estimates, like we may be overestimating the effects of heritability or the effects of some type of environment because of these non-additive effects, these interactive effects that we're not paying attention to. So that's part of the concern is that our results are biased, and that's an important concern. There's no doubt about it. And as I mentioned, we do have some tools to sort of investigate that. And so the other concern is, and it is connected to the first, is that the sort of classic twin studies give a vastly oversimplistic account of how genes influence these complex phenotypes. And the argument is, you know, there are important interactions between genes, and there are important interactions between genes and environment.

[00:27:01]  Brian: And just to partition variance into heritability, shared environment, non-shared environment, is to give such a misrepresented picture that it borders on just a caricature. And again, even though I don't necessarily agree with those arguments, they have been very crucial in the development of the field and sort of testing the assumptions of the models and again, the more methodological side of things. So there's certainly nuggets of very important information in there. And so when folks have debated these issues, they've done a bunch of fields connected to behavioral genetics, a great service because those are very important issues.

[00:27:49]  Peter: So am I understanding your last comment correctly, that you feel that some people who are tied into these twin studies and things take the conclusions about a bit too far?

[00:28:05]  Brian: No, no, I wouldn't say that.

[00:28:06]  Peter: Is that what you're saying?

[00:28:07]  Brian: I would just say that some of the debate revolves around the idea that development of complex traits, whether that's height or perhaps something like schizophrenia or major depressive disorder or depression in general, these are very complex types of things to study. What I think is also important to point out is a lot of times when we're debating the importance of something like additive effects versus non-additive, the debate is not about whether genes are important. Both sides agree that they're important. What they're disagreeing about is the extent to which these interactions between genes or interactions between genes and the environment are being missed in the basic approach of twin studies. But sometimes agreement can get lost in disagreement. There's an agreement that genetic effects are important and interesting for complicated outcomes.

[00:29:14]  Bruce: Okay, so let me just take this statement that we show that the majority of traits 69% are due to additive genetic variation. Let me take that statement at face value. So let's assume for the sake of argument that God comes down and tells us, yes, that was exactly correct or something like that. So we now know it to be correct. Okay, what does it actually mean? So let me see if I can make my question more sharp here. Does that mean that those traits do not have interactions between genes and environment? And that environment may influence it and genes may influence it, but they're not having some weird interaction between them, that it's actually, there's a thousand genes, and if you have 500 of them, you're going to be a certain height. And there's environment, you need certain kinds of nutrition during development, and that will affect your height.

[00:30:09]  Bruce: But it's not necessarily a complex interaction between genes and environment in the case of these 69% of traits.

[00:30:18]  Brian: Yeah, so I think, so correct me at any point if I misstate something you just said, but, so if you were to make the statement that we can dismiss the importance of gene interaction or gene environment interaction for complex traits because most of the variance in complex traits tends to be additive, I would say that is not a good interpretation. What I would say is the assumption of additivity or of additive effects is a good guiding assumption. But for any particular research topic or area of study, let's say, just go back to schizophrenia, we might have, theory might guide us, and this is kind of part of the importance of good theory, good explanation, theory might guide us toward suspecting some interaction between two alleles or three alleles or alleles in the environment. And we can and should investigate that. So it certainly doesn't take that off the table, doesn't make it less interesting or less important.

[00:31:35]  Brian: It really in some ways is sort of pragmatic. It tells us that, you know, if additive genetic effects are the sort of [guide?], if that assumption holds, we're going to get estimates of heritability that are less likely to be biased substantially, if that makes sense.

[00:31:53]  Bruce: Now, kind of related to that was the article said, many genetic variants contribute to the variation of complex traits and the effect sizes are typically small. However, the sum of the [variance?] explained by the detected variants is much smaller than the reported heritability of the trait. This missing heritability, that's in quotes, has led some investigators to conclude that non-additive variations must be important. Can you help me understand this concept of missing heritability?

[00:32:28]  Brian: Absolutely. So, twin studies have been around for, relatively speaking, longer amount of time than the more modern sort of technological advances that allow us to do something like a genome-wide association study.

[00:32:47]  Unidentified speaker: I noticed that.

[00:32:47]  Bruce: Like, they started back in like the 40s or 50s or something.

[00:32:51]  Brian: Yeah, they've been around, have a long history. And whereas to do things like study the, or even to just do simple genotyping with like polymerase chain reactions, PCR, it's relatively new technology. And, you know, going back to the Human Genome Project in the early, which sort of was wrapping up in the early 2000s where we had mapped the genome, at that point we were able to, especially as the technology became cheaper and cheaper and cheaper, to use, to genotype and measure the presence of certain genes, certain genetic variations that folks might have, maybe, you know, which version of a dopamine-related gene that they were carrying, and see if those were linked to these complex traits that twin studies were telling us were heritable. Which, again, even though heritability and inherited aren't the same, heritability does imply genetic influence on a trait, right?

[00:33:58]  Brian: And so we could then go out and try and investigate whether, you know, a certain dopamine gene explained any part of why some individuals tend to be more, perhaps, vulnerable to depressive episodes or major depressive disorder. And we sort of, we mentioned this in our paper, and I think the meta-analysis does as well, there was a flood of what we call candidate gene research. And in general, those were, and I say this as somebody who published candidate gene research, they were underpowered. They were sort of ill-equipped to find real effects. And as a result, a lot of those findings didn't replicate. They were, you know, false positives or, you know, sort of false starts, so to speak. And they were, as a result, I don't do candidate gene studies anymore. You'll still see some candidate gene work published, but there's a much greater emphasis and use of these genome-wide association studies.

[00:35:06]  Brian: And what these studies allowed us to do was investigate possible associations between hundreds of thousands of places in the genome with complex traits. And so we were able to not only look at whether certain places in the genome, certain alleles correlated with something like height, we could also use those genetic variants themselves to calculate heritability. So the way that that's done is interesting and important, but it actually is not super relevant. In this case, the long and short of it is you can calculate heritability estimates using measured DNA. So these looking at different alleles captured in these genome-wide studies. And especially early on when folks were doing this, what they were finding is that the heritability estimates they were getting from using measured SNPs. So I'll explain that term just briefly. If I use the word SNP, it's an abbreviation.

[00:36:19]  Brian: It refers to a single nucleotide polymorphism. So the idea that at a certain point in the genome, one might have a particular base pair, an A or a T, if we think back to basic biology. But that's enough to just describe that it's having a different version of a gene, which is being oversimplified, but that's all right. So you can use that information to calculate heritability. And what missing heritability as a term refers to is, when folks did that, what they found is they got heritability estimates that were considerably lower than what twin studies were providing. So if we conducted a twin study, calculated the variance in height that was heritable, and found that 80% of the variance was attributable to genetic variation, we did the same thing using genome-wide association data. We found that we only explained 5% of the variance in height.

[00:37:20]  Brian: And so I think a couple of things are important there. One, the conclusion was not that genetic effects are irrelevant for height. It was that twin studies are sort of washing over these very complex networks of interactions between genes that explain variance in height and sort of goes back to the idea that maybe assuming that additive effects were the rule rather than the exception was wrong. And so one of the points we make in our paper is that with time and even at the time, some folks were arguing that, look, is missing heritability is really not the confounding issue that you think it is because part of it could be the simple fact that the arrays that we were using to do the genotyping for these genome-wide association studies were missing what we call rare variants or just SNPs that occur at a relatively low rate in the population.

[00:38:26]  Brian: And so when you account for that, you should be able to explain more variance that you thought was missing.

[00:38:33]  Unidentified speaker: I see.

[00:38:35]  Bruce: So in essence, the field moved from thinking that most of the heritability was non-additive to, oh, wow, actually, most of these are additive.

[00:38:46]  Brian: Well, I would say, I think maybe the useful way to think of it is that on a couple of fronts. One, so when we did, we were able to do, and we mentioned a couple of these studies in our paper, you take something like height, a recent study that had the capability of sort of detecting presence of both common and rare alleles, explained, I think it was around 68%. So it had a gene-based heritability estimate of about 68% for height. That's still lower than what you will often get with twin studies, but it's getting much, much closer. So that missing heritability, as we say in the paper, starting just there [unclear]. Yeah, it's getting filled in. But going back to that second point, the idea of complex, non-additive types of interactions, in some ways, that is, we can use additive effects as a good guiding assumption. But for any particular trait, we might be compelled, again, by theory or some other piece of evidence that we want to investigate [gene–gene?] interactions using something like GWAS, and that's completely defensible and important.

[00:39:59]  Brian: So it doesn't mean that you can't look for non-additive interactions. We can and should.

[00:40:05]  Peter: Can I jump in real quick? You've mentioned twin studies a lot. It occurs to me that it's sort of the opposite of a twin study, which seems to me might be more easier to study, which aligns to kind of your experience, is where non-genetically related siblings are raised together, in a shared environment. Are the conclusions of those kinds of studies as important as the twin studies?

[00:40:39]  Brian: So, I want to make sure I've got interpreting your question right. So, are you asking that, is it possible to do sort of the classic design that we would normally do with identical and fraternal twins, but do it with just maybe full siblings and half siblings?

[00:40:57]  Peter: Yeah, yeah, yeah, like going in that direction. I mean, it seems to me that's much more common than twins being raised apart, which would just be the opposite, non-genetically related siblings raised together, right?

[00:41:11]  Brian: Yes, okay. So, just so we are thorough in this answer, you mentioned in your twins raised apart. That is an important design. It's relatively rare. There aren't many of those folks. We do, you're right, you're correct, that classic twin designs rely heavily on the use of identical and fraternal. And I guess at this point, we can also give a term and define it. So monozygotic, dizygotic, the more sort of term you'll see in the literature. Monozygotic refers to identical twins, twins that result from a single fertilized egg splitting. And as a result, those individuals essentially have carbon copies of the same genome.

[00:41:56]  Unidentified speaker: Yeah.

[00:41:57]  Brian: Clones of each other, basically. Dizygotic twins are in some ways the same as full siblings. They're the result of two separate fertilized eggs. And just like full siblings, they share half of their genome, but they also share prenatal environment, obviously. So typically when we're doing twin studies, you're going to see identical and fraternal twins used in the study. You can absolutely do very similar types of analyses using full siblings, half siblings. The important thing is that you have a sample comprised of siblings with types that differ from each other and how related they are to each other. So full siblings share half of their DNA, just like Dizygotic twins, half siblings would share half of that. So as long as you know the genetic overlap between your sibling types, you can conduct behavioral genetic types of analyses. So you talked about maybe the availability of twins or that sort of connected to one of the points you mentioned.

[00:43:13]  Brian: Over the years, and it's not in the paper, but it's actually a resource I'm pretty excited about. It's in our supplementary material on the journal's website. We wanted to offer anyone interested in behavior genetics a resource for how to find data. And what's pretty amazing is that when you sort of look at what's available to anyone who wants to do these types of studies, there's a massive amount of data available, large samples, thousands of twin pairs, and a lot of it's freely available to download. And folks can go out. So that's a large part of what our paper was meant to be, was a resource for anybody who's interested in this stuff because there is really an abundance of twin data. But yes, to go back to your question, we can use full siblings and half siblings. We're a little bit limited in terms of some of the inference capabilities compared to, say, when we have identical and fraternal twins.

[00:44:16]  Brian: But that's an option as well.

[00:44:19]  Unidentified speaker: I see.

[00:44:20]  Unidentified speaker: Thank you.

[00:44:21]  Brian: Oh, absolutely.

[00:44:22]  Bruce: So does the amount of heritability change? Let's take an example of we were trying to measure how much height is heritable today versus, say, in the medieval ages where nutrition was bad. Would the amount of heritability change?

[00:44:40]  Unidentified speaker: Absolutely.

[00:44:42]  Bruce: Heritability is not necessarily stable. It does depend on the actual environment.

[00:44:49]  Brian: You're correct. It's a very important point. It's one of those sort of related to not confusing heritability with inherited. Heritability should not be mistaken to be some type of fixed number. A trait can be heritable at one point in time. Right now, it's highly heritable. But as you point out, if we were to hop in a time machine and be capable of doing a twin study in a prior human epoch, we're probably going to find that most of the reason people differed in their height was because of access to nutrition and not because of genetic differences. When you have severe caloric restriction common with parts of the population, the answer to the question of why are some people taller than others is going to have a lot to do with, well, some folks are not getting the requisite amount of caloric intake in order to grow. And so in that regard, environment becomes a much more salient explanation of why some folks are taller than others.

[00:45:56]  Brian: So yeah, you're 100% right.

[00:46:00]  Bruce: Way back, I did some episodes where we talked about, and we'll do this a little bit in this episode too, about some of David Deutsch's criticisms of these sorts of studies in beginning of infinity. And one of the things I pointed out is that the criticism that he was bringing up for mental traits was really kind of also valid for physical traits, because David Deutsch does not doubt that genes affect height. That wouldn't make sense. Everybody knows genes affect height, right? And yet, they don't really directly affect height. It still depends on some sort of complex interaction with do we have nutrition or not, and maybe other things. I mean, I know about nutrition. I don't know about other causal factors. And so, you can't really ever say this gene, except for the Mendelian ones, this gene directly caused this inheritance. There's always some sort of assumption of given a certain setting, given a certain culture, background, level of knowledge, time period, etc.

[00:47:10]  Bruce: We're talking about the heritability within that setting. Is that correct or have I got that wrong?

[00:47:16]  Brian: So, there are multiple points in there where good points are being made, but I think they're ultimately leading to maybe not the best conclusion regarding how to understand genetic influences on complex traits. And so, I do want to be careful here. I mean, Deutsch is incredibly insightful, and I don't want to attribute anything to an argument he would make that, you know, if he were to hear it, he might say, well, that's unfair, that's not what I think. So, I'll just talk in general terms about those issues. And so, on one hand, the idea that, you know, for really complicated complex traits specifically, there aren't simple, you know, easy to explain mechanisms that get us from genetic variant to trait. And that's entirely correct. We, you know, we certainly, when we show whether using genome-wide association studies or twin studies, that a trait is heritable.

[00:48:24]  Brian: In fact, this is, you know, we actually, in the paper we talk about this, we consider this to be a point captured by the fourth law of behavior genetics, which deals with the idea that for complex traits, there are probably thousands of genetic variants involved. They're all having very small effects, and the effects are not direct. The effects are far upstream of the trait we're interested in. And so, in some ways, though, I don't know that, I think that's a useful way of thinking, even with sort of Mendelian traits, the pathways may be easier in some ways to trace, but even then, you know, if it's a, you know, genes are sort of the old term that used to use are the central dogma of molecular genetics, is that genes code for the production of proteins. Those proteins then have to be assembled into, you know, longer, more complex entities.

[00:49:24]  Brian: And if it's, you know, from there, they may be proteins that build either structural components of our body or functional components of our body, you know, tissues, bones, tendons, all those things. They have to be assembled from the products of what genes are doing. So really, in many ways, any outgrowth of some genetic influence is complicated. So it's not really a criticism of twin studies or GWAS studies to say that, you know, there's no simple path from genetic effect to trait, so that in a way that renders heritability uninteresting or unimportant. And I think that really is an important aspect because it, again, is a more general point. For example, we may, you know, if we've got a, you know, a drug trial, and well, actually Tylenol is a good example. So acetaminophen in some, you know, if we were to, we wouldn't do this at this point.

[00:50:28]  Brian: But let's say we ran a clinical trial to figure out whether or not Tylenol has the effect of relieving headaches more quickly than placebo, right? And let's say that the clinical trial with experimental design so that you're randomly assigned to take Tylenol when you get a headache, but you don't know if you're getting Tylenol, or you're randomly assigned to take a sugar pill, but you don't know. And then we're going to observe the results. And what we find is that the folks who take Tylenol, even though they don't know it, their headaches go away much faster. So we conclude that Tylenol has a causal effect on speed of headache remediation, right? That tells us nothing about the mechanism by which that happens. We assume the mechanism is there because we infer that nothing magical is happening between Tylenol administration and headache remediation.

[00:51:25]  Brian: So if we assume there's no magic, there's a causal pathway playing out. Now, we simply may not know it, but it is there. And so I don't think, I think simple lack of insight about causal pathways doesn't really render the results of twin studies uninteresting or unuseful.

[00:51:46]  Unidentified speaker: Okay.

[00:51:46]  Unidentified speaker: Fair enough.

[00:51:48]  Bruce: So in the Polderman article, that would be the one that you're not an author on, it says the largest heritable estimates for traits classified under the ophthalmological domain, i.e. eyes, followed by the ear, nose, and throat, dermatological and skeletal domains. So that was in order of reducing amount of heritability. Now, that seems fairly straightforward that genes would affect a purely physiological trait like eyes, ears, skeleton, things like that. But I can understand why there would be more skepticism around the idea of cognitive or personality traits being heritable. So now in the Nature paper, so this would be your paper, it says the overarching conclusion at this point in time is that all measured human traits are to some extent heritable and that many genetic variants with each exerting a small effect explain this heritability.

[00:52:51]  Bruce: Now, can we talk about what that means that all measured human traits are to some extent heritable? In context, I think I would have assumed that you meant of the ones we've thought to measure at this point. But the way that's stated, just kind of me taking it out of context, it could be interpreted as meaning literally all traits, including personality traits, are to some degree heritable. Help me understand how I should understand that statement.

[00:53:30]  Brian: Yeah, absolutely. So I mentioned a second ago the fourth law of behavior genetics. What you just described is the first law. And keep in mind too, when I'm talking about laws here, it absolutely is not trying to imply that us in behavior genetics have laws akin to physical law that one might use in astrophysics or something like that, speed of light or whatever. What I mean is a much more, something much more general, but it has to do with the idea that we have found, we have repeatedly found the same thing over and over and over again, that we functionally, we might, you know, we may as well treat that as our version in the behavioral sciences as a law. And so the first law of behavior genetics, and these were, by the way, this idea was proposed in 2000, what would that be, 14, 15 years before the Polderman meta-analysis by a psychologist named Eric Turkheimer, a very important behavioral geneticist.

[00:54:38]  Brian: And the first law of behavior genetics being that all, we should assume all complex traits are heritable. And so why would that be interesting? Well, it's interesting for a variety of reasons. But before we get there to your question, that is, I'm not going to speak for anybody else, but me, that is sort of the guiding assumption that is inherent in my thinking. Whether we've done five studies on a trait or a hundred, we should start with assuming that it's at least partly heritable.

[00:55:12]  Bruce: Here's, by the way, a quote from your article. So consistent were the results that behavioral genetic research has come about as close as possible for a field outside the physical sciences to producing laws. Then I had to look up what the laws were. And here's what I found, because I don't think it actually stated them in the paper, but it said...

[00:55:34]  Unidentified speaker: We did.

[00:55:35]  Brian: It was in a box in the paper.

[00:55:37]  Unidentified speaker: Maybe I missed it.

[00:55:38]  Brian: That's okay. It was totally understandable. But go ahead.

[00:55:43]  Bruce: Law number one, all human behavioral traits are heritable. Number two, the influence of genes on human behavior is greater than the family environment. Three, a significant number of behavioral traits may be explained neither by genes nor by family environment. And four, a typical human behavioral trait is associated with very many genetic variants, each of which accounts for a very small percentage of the behavioral variability.

[00:56:14]  Unidentified speaker: You got it.

[00:56:15]  Unidentified speaker: Okay.

[00:56:15]  Bruce: So those are the four laws.

[00:56:17]  Brian: Those are the four laws. The fourth law was the most recent. It was proposed in, I think, around 2014 as well, kind of in the wake of the Polderman meta-analysis.

[00:56:31]  Bruce: Now, for some reason, it was called four laws and two corollaries. I'm not even sure I caught what the two corollaries were.

[00:56:37]  Unidentified speaker: Yes.

[00:56:37]  Brian: So the two corollaries are something that we in our article and then another individual sort of proposed concurrently two different versions. I'll give them to you. I'll read them to you. And then that might take us in some interesting directions from here regarding why the laws are important in the first place. So from our paper, I've got it in front of me as well. Corollary one, the first law of behavior genetics implies that trait level correlations between phenotypes. Phenotype is anything we can observe or measure. So phenotype can be height. Phenotype could be major depressive disorder. So it's just the trait we're interested in. First law implies that trait level correlations between phenotypes may sometimes reflect overlapping genetic influences across those traits. Simply means that because complex traits are influenced by so many genetic variants, it stands to reason that for some traits, they're going to be influenced by some of the same genetic variants.

[00:57:44]  Brian: So it might be depression and some other type of comorbid disorder that's common with depression. So that if an individual becomes depressed, they're also maybe more likely to develop an anxiety disorder. And so what the corollary one refers to is the idea that probably those two complex traits are impacted by some of the same genetic variants. Corollary two, I think, is if I had to rank order, maybe the more important of the two. Causal inference ability degrades to the degree that the implications of corollary one remain uncorrected in a particular study design. So I recognize that that is a bit cumbersome in terms of its language, but basically what it means is something very practical. And it's that if we assume that not only are two traits correlated, so maybe we find that people who have generalized anxiety disorder also tend to have depressive symptomology.

[00:58:55]  Brian: If we want to make the argument that generalized anxiety disorder causes the likelihood of depression to rise, it causes depression, we have to contend with the fact that there is likely genetic overlap between depression and generalized anxiety disorder. Because if we don't, what we make ourselves vulnerable to is something called confounding. And it basically means, you know, we risk assuming that X causes Y when in fact it doesn't. There's another variable, maybe Z, that impacts both X and Y, and it forces, it creates a correlation between the two. But if we don't know that, it could lead us to a bad conclusion or a spurious conclusion.

[00:59:47]  Unidentified speaker: And so, yeah.

[00:59:48]  Brian: And so this is, you know, gets us to the point. Actually, just briefly, I'll make this point because I think it's important. Whether it's an outwardly stated position or not, this type of point is why experimentation is still considered the gold standard of causation in science in terms of research design. When we randomize individuals to some treatment exposure, so in other words, we create our experimental groups, our treatment group and our control group through randomization. The idea is to make those groups essentially as identical to each other as we possibly can. At the end of the day, the only difference between them is our treatment. In the Tylenol example, the only difference between the two groups in the study was that one group got Tylenol, the other group didn't. When the Tylenol group gets rid of their headache faster, the most defensible conclusion is that the Tylenol caused it.

[01:00:56]  Brian: That's the only thing that differed between the groups. So when you can do experimental designs, you don't have to worry about either corollary. You don't have to worry about the four laws of behavioral genetics, because the randomization deals with the two corollaries, and it deals with any other possible thing that could confound your findings.

[01:01:18]  Bruce: In Judea Pearl's book, the Book of Why, he talked about this quite a bit, and he used the example of smoking throughout. It would be unethical. You could do a causal study of smoking if you didn't mind the fact that you possibly giving people lung cancer. Instead, you have to use data that wasn't in a randomly controlled trial. Now you've opened yourself up to the possibility of a confounding factor. He gives a history, at least his history, of how scientists tried to deal with that problem. It sounds like it was a very complicated problem that we've only fairly recently started figuring out how to deal with. He talks about two people that were smoking, and one of them was arguing that smoking does cause cancer.

[01:02:20]  Unidentified speaker: I forget who the two people were.

[01:02:21]  Bruce: They were famous scientists. It was causing that it didn't [wording unclear], or at least there would be no way to ever prove it. Then he goes through some of the things that they eventually did that led to the theory that smoking causes cancer to become the better of the two theories. Well, one of them was that they worked out the degree to which genetics would have to cause both smoking and cancer, and it was such a large factor that it was implausible to explain it on its own. In fact, they eventually found that there was a genetic that caused both smoking and cancer, but it had a much more plausible amount of effect, and it was pretty minor by comparison to smoking itself being the causal factor.

[01:03:07]  Brian: I should say, too, that Judea Pearl's work has had also a tremendous impact on just my interest as a scientist and the way I think about things. It's just foundational work, and the Book of Why is just a wonderful book to read, accessible to anybody who's interested in causation, and whether your interests are more philosophical or more technical, it's just a fantastic book. I can't recommend it enough. But yes, and what's interesting is, the smoking example in lung cancer is a wonderful example to use because it illustrates a variety of things. One is the complexity of studying these important traits when you can't do, or rather, the way that the challenge increases when you can't do experimental manipulation. So indeed, you plausibly could assign some people, randomly assign some people to smoke cigarettes and then maybe devise some type of placebo device and have the other group smoke those.

[01:04:20]  Brian: But to do so with the suspicion that it might cause one group to get lung cancer is beyond unethical. You're not going to countenance that type of design. So when you can't do that or you shouldn't do that, you attack the problem in a different way, which is to say you attack it in a variety of ways. One is you do in vitro studies exposing tissue to the carcinogens found in tobacco smoke or in chewing tobacco. And you see if those things tend to induce the types of mutations that lead to cancer. And assuming they do, and they do with smoking, that's a piece of the evidence. Another thing you can do is you can do sort of traditional observational methods where maybe you sample a large sample of individuals living in a country, say the United States, you ask them about their smoking habits. Maybe you follow, when the study begins, you follow them for 10, 20, 30 years, and you, you know, at each time following up to see who in the sample has developed some type of cancer, and you use correlational methods to see if there's a link between self-reported smoking and the development of lung cancer.

[01:05:47]  Brian: And let's say we find that there is. That's another piece of evidence. And this brings us to part of the reason I think twin studies are so important. One of the most powerful designs I would argue for causal inference outside of experimental designs is the discordant, what's known as the discordant twin design. So to give you the sort of the most basic straight ahead version, we said earlier that identical twins have carbon copies of the same genome. Now, what's interesting about that is, and this also reflects the first law, that complex traits are partially heritable, but not entirely heritable. Identical twins differ from each other all the time in a variety of ways for a variety of reasons. But among those reasons, there cannot be the explanation that identical twins differ from each other because of genetic differences, because for all intents and purposes, they don't have any genetic differences.

[01:06:56]  Brian: So when you have that situation, it's an incredibly useful natural experiment. So one thing we can do, and this was done in terms of this smoking situation, and with a variety of other questions to this day, we can use data from twins who are naturally discordant for some outcome we're interested in. So imagine that we have a sample of identical twins. Among that sample, some number of those twins were smokers and their co-twins were not. So they were discordant in their smoking habits. One twin smoked, their sibling did not. We can use that as a type of sort of causal design to then investigate, well, does it seem to be the case that when twins were discordant for their smoking, that the one, the twin who smoked was more often, more typically developed lung cancer. If we find that to be the case, that is pretty compelling evidence of an environmental factor causing some outcome.

[01:08:06]  Brian: In this case, smoking causing, exerting some type of causal effect on the cancer. And notice I said it's evidence of an environmental effect.

[01:08:14]  Unidentified speaker: Right.

[01:08:14]  Brian: Because genetic differences can't explain that effect.

[01:08:17]  Bruce: Right, it can't be, so you've refuted the idea that it was a gene causing both at that point because...

[01:08:23]  Brian: Well, you haven't refuted it. What you've done though is the same thing, something similar to what randomization does. You've held the effects of genes constant.

[01:08:32]  Unidentified speaker: Right.

[01:08:32]  Brian: If that makes sense. So, it's a little different than refuting it. It's still possible that there might be genetic overlap between smoking habits and vulnerability to cancer. That's sort of a separate issue. We can use twin studies to investigate that too, but when we use a discordant design, what we're saying is we're not going to allow genetic effects to confound some effect of an environment on some outcome.

[01:09:00]  Bruce: Let me read a bit more from the articles here. Behavioral genetic designs can be applied in fields as diverse as economics, political science, psychology, and medicine, among others. Then from the other paper, by studying genetically identical twin pairs that differ in some key attribute or diagnosis, the drawback of traditional case control studies can be largely surmounted, a sample of about 300 twin pairs, finding heritability estimates of 41%, 53%, 61%, 41%, and 44% for the five dimensions of neuroticism, extroversion, openness, agreeableness, and conscientiousness respectively. The papers separately [or?] taken together, and the four laws you just mentioned, there is, we are saying that personality traits are heritable.

[01:09:56]  Brian: Yes, there's good evidence of that. Personality traits like other cognitive traits, like other physiological traits, adhere to the four laws of behavior genetics.

[01:10:05]  Unidentified speaker: Okay.

[01:10:06]  Bruce: Now, with that in mind though, there are several things your paper said that I wasn't quite sure I understood. So one of them was this concept of quasi causation. Let me give you the quote. By controlling for common causes, twin studies can test what have been referred to as quasi causal hypotheses. The use of phrases like quasi causality and causal inference as opposed to straightforward causality highlights that twin studies, like other epidemiological approaches, cannot typically demonstrate causality between two variables. They cannot prove that X, the exposure, causes Y, the outcome.

[01:10:45]  Brian: Yeah, so we were actually quoting a study from 2021. There's an excellent paper that is much more sort of narrowly focused on the various twin designs used for causal inference. Fantastic paper that we cite in our article. And so folks who are interested can look that up. That's a quote from that article. And essentially what it's asking is asking a type of rhetorical question in a way or hypothetical question. And the question is this, are discordant twin studies like what we've been talking about as powerful of an engine of causal inference as a randomized clinical trial or randomized experimental design? And the answer is no, they're not. These types of twin studies are vulnerable to certain types of confounding. And there's a really helpful literature out there on that. It's a fairly methodological point, but the important takeaway is pretty straightforward.

[01:11:48]  Brian: And it's that just because you're doing a discordant twin study with identical twins does not mean that you have knock-down proof, which I don't even like the word proof anymore for reasons we can talk about too. It doesn't mean you have knock-down evidence that X causes Y just because you did a twin study. You absolutely do not. What you have is a valuable piece of evidence that is, you know, if it's the initial finding, if you're sort of the first to do that study, it has to be weighted very lightly until there have been efforts to replicate it. So in other words, it needs to keep showing up. If that's if my, you know, trust in the finding is going to increase. But the point that sort of we were trying to draw attention to is that discordant twin studies are exceedingly useful, but they're one among other options in a toolkit, including some options.

[01:12:50]  Brian: And actually, Pearl does a really nice job in his book talking about some of these options that don't require twin data and yet can still give you, if you satisfy certain assumptions, can give you pretty potent causal insights. And so what we're trying to do is avoid the, you know, I'm obviously enthusiastic about these designs, but I don't ever want to convey that they're a panacea for, you know, the problems of causal inference.

[01:13:19]  Bruce: It's interesting that you used the term weighted there in reference to evidence. So in Beginning of Infinity, David Deutsch talks about possible problems with the concept of weight of evidence. In context, he's not talking about what you're talking about. It's a different concept of weight. What you just suggested really was we weight the evidence less because it has yet to be replicated, which makes perfect sense to me. And that's even consistent with critical rationalism, that we don't really consider an observation an observation until really basically you can, anyone can do the experiment and can get the same result and now become something that we have to contend with, that everybody gets the same result when they do this experiment. That's actually a, I would argue, correct and fair way of understanding the concept of weight of evidence within a critical rationalist framework.

[01:14:16]  Bruce: I just wanted to kind of call that out. I know some people may say, well, wait, didn't David Deutsch say that we don't weight evidence?

[01:14:22]  Brian: Yeah, it's a great point. And it's one of those things where as the word left my mouth, I sort of anticipated that it might register with you all for that reason. Some of it's force of habit, because as I mentioned, I was heavily persuaded by Popper and Deutsch sort of collectively and others as well, that, you know, it's not useful to think in terms of something like science verifying truths. So it's not—not how I think about the work that I've mentioned either. But yes, and I do very much just mean it in the, I guess, parochial. Yeah. You know, let's let's not get too excited until other folks start to find the same thing.

[01:15:09]  Unidentified speaker: Right.

[01:15:10]  Unidentified speaker: Right.

[01:15:11]  Bruce: And that makes sense. A study comes out and the media does what they do. And they say this, you know, challenges this belief for this theory or whatever. And a scientist knows we have to wait to see it replicated. But you can't really say that it's not evidence either.

[01:15:33]  Unidentified speaker: Right.

[01:15:33]  Bruce: I mean, like you now have this effect that's been found in a legitimate study. And now the criticism of the study begins and we have to see if people can replicate it. So I think that the metaphor of weight makes sense in that case, that we're giving it less weight until it gets replicated.

[01:15:49]  Unidentified speaker: That's my two cents.

[01:15:51]  Unidentified speaker: Absolutely.

[01:15:52]  Brian: Because I, you know, and I know I can't call it to memory, so I won't say for sure. But if I'm remembering correctly, I think Deutsch has addressed this in a couple articles, if not in the books, that it's certainly possible that you, a particular experiment or a particular study, arrived at some type of result that's erroneous because of a weak or inadequate research design. It might be that your instrument in your physics lab was poorly calibrated. And we find that out after the fact, and that this is why we treat a finding provisionally, because we may well find out next week that, no, our instrument was not calibrated right, so we can't trust the results. Cosmological observation, yeah. And so that's really the same spirit in which I use the point, too, is that, or with any finding, especially those that tend to get sort of splashed into headlines, is that, well, it could be that, you know, the sample was 20 people, and statistically that creates problems.

[01:17:00]  Unidentified speaker: Right.

[01:17:01]  Brian: Yeah, any number.

[01:17:03]  Bruce: And this is another, so I have an episode that we recorded two weeks ago. I don't know if it will come out before or after this episode, where I actually argued that there are definitely certain senses where saying weighted evidence makes sense. And another one would be like p-values, right?

[01:17:20]  Brian: Yeah, absolutely.

[01:17:22]  Bruce: That some things that we call evidence, they just have a much better p-value because maybe the sample size was larger or something like that. So you were fishing with the finer net, as Deborah Mayo would put it. I don't see any way around that. Certain pieces of evidence, if they're statistical, which in your field they would be, some are going to be weighted more than others because they were literally a better study that used a finer net to try to catch, to try to find, severely test the hypothesis. So it makes sense to use the metaphor of weight in that case, or at least as a layman, I'm going to use the metaphor of weight.

[01:17:59]  Unidentified speaker: It makes sense to me too.

[01:18:01]  Bruce: But if you look back at what Deutsch actually said, he was actually talking in context of politics and things like that. I think that there are many cases where we talk about weighted evidence that it doesn't make sense. And I think Deutsch was talking about cases where it did not make sense.

[01:18:18]  Brian: Totally agree. And I think part of that, again, we can save this for a future discussion, but I think part of what was wrapped up in his various discussions of that was this idea of almost being careful not to turn science into an authority was something we hopefully would have projected [possibly rejected] as part of the Enlightenment.

[01:18:42]  Bruce: My next question is about temporary heritability. Let me [take?] a quote from the article here. The data show a decrease in monozygotic and dizygotic twin resemblance after adolescence. So does this mean that heritability traits significantly drop as we become adults, sort of become our own man, so to speak? And how significant is this drop? Is the drop only for personality and cognitive traits, or does it actually also drop for other types of traits? And then here's the thing. Before you answer that, those are my questions. But there's some other statements from the article that I want to take into consideration as you're answering the question. For example, ranging between 57 and 73% depending on measure, such as indicators of spatial and verbal development. Interestingly, these estimates were found to be much lower in children as opposed to older participants.

[01:19:41]  Bruce: That's like the opposite of what I just quoted, often hovering around 45%. That seems very high still, even as talking about older. I just had noticed that it was the opposite of the first quote from the other article. Then it also says, very early in life, heritability estimates can be as low as 20%, which typically rise to around 40% in middle childhood, and then often balloon as high as 80% in adulthood. Again, that seems the opposite of the first quote. Maybe you can shed some light on this concept of temporary heritability based on age.

[01:20:22]  Brian: Yeah, so it's a really interesting topic to discuss. To answer part of your question, not only can heritability estimates change across the life course for certain traits, but in lockstep with that, it implies that the environmental variance components also change. So as heritability estimates go down, either estimates of the shared environment or the non-shared environment have to go up because those three components sum to one or 100%. And so we, again, I guess the more general statement is that heritability and environmental sources of variance can and do change across the life course. It really depends on the trait that we're interested in. And that sort of informs how those different estimates change for certain sort of cognitive traits, as we talk about in the paper. The heritability estimates tend to rise with age. And remember, part of understanding what this is saying means that we need to think back in terms of what heritability means.

[01:21:36]  Brian: And at its core, it is a sort of an answer to the question of what explains differences between folks in a population. And so all that we're saying, it is not saying that genes matter in childhood and they don't matter at all in adulthood, or environment doesn't matter at all in childhood, matters a lot in adulthood. That's not the proper interpretation. Rather, what it's sort of showing is that over time with age, the reasons, the explanation for differences in a population become increasingly more attributable to genetic variation than environmental variation. And that's again, thinking about certain cognitive traits and it's a different sort of a different story for personality traits. Now, it doesn't necessarily, those types of general rules are in many important ways dependent on the trait that we're interested in. Not all, if we're talking about heritability of some complex trait across the life course of a human, those heritability estimates might be relatively static.

[01:22:48]  Brian: They may not change very much, whether we calculate them at age 10, 20, 30, 50, whatever the case might be, or they might change with time. And there are various ways to think about explanations for why and how that happens, mechanisms for how it happens, what explains it. And those are all, we cite a lot of those papers in our paper. And so folks who are interested can track those down as well. But this is sort of the idea of heritability is not a fixed construct. It's not a snapshot measure of how genes causing a trait that never changes. It is part of an explanation for individual differences. And the environment is also part of that explanation.

[01:23:35]  Bruce: One of the articles said the data show a decrease in resemblance after adolescence. But then the other article seems to have twice said that the heritability increases as you go into adulthood. And you said that might be different for cognitive traits versus personality traits.

[01:23:54]  Brian: Yeah. So again, speaking in just very broad terms, you don't tend to see, so rather you see the effects on heritability estimates for more cognitive based traits, indicators of intelligence. So spatial, analytical abilities, memory, certain aspects of recall and memory and processing speed and language ability, those types of things, sort of more indicators, broad indicators of intelligence. The heritability of those things has been seen to increase over time across the life course. The same is not, doesn't seem to be the case with what we would think of as more classic personality traits. And so we take sort of the highly influential model of the big five or OCEAN. So openness, conscientiousness, extroversion, agreeableness, and neuroticism. All are sort of have typically show small to moderate heritability estimates. And they don't seem to get those estimates don't seem to go up with time.

[01:25:06]  Brian: In fact, in some cases, non-shared environmental effects seem to go up across the life course. It's not to say that their, you know, heritability estimates dropped to zero, but they can change with age.

[01:25:21]  Bruce: That actually makes sense to me. Okay, so let's talk about David Deutsch's criticisms of some of these studies that are in beginning of infinity. So obviously, we all know that correlation isn't causation. So Deutsch gives a series of arguments to the idea that we can currently know how much genes really by themselves cause inheritance or personality or cognitive traits. So the example he uses is slavery in the 19th century. If we were to do a study, we would find that it is highly heritable. He also gave an example of attractiveness being heritable, which I would assume it is, but suggests the possibility that culturally we may change what we count as attractive in the future, and therefore the heritability, even though it is heritable, it may not be in the future. Can you maybe talk a little bit about that, particularly the example of slavery?

[01:26:34]  Bruce: Because that does seem like it's kind of relevant to what we're talking about. Slavery would have been a heritable trait in a study for the obvious reasons that there's an explanation for why certain racial groups were in slavery and others weren't.

[01:26:49]  Brian: Yeah, so in regards to Deutsch's points, I don't take issue with any of them, except to say that I think it would be in, you know, in my, I would argue that they don't necessarily, while the points are valid points to make, they're not at odds with, I don't think anything we've talked about. And I don't know that the best conclusion is the one that he reaches in terms of sort of the information provided in a twin study is not something we should really be concerned about.

[01:27:27]  Unidentified speaker: Right.

[01:27:28]  Brian: And so what I'll maybe unpack that a little more, I'll start with, we talked about earlier that heritability estimates in terms of particular, you know, moments in time are very much likely to be different. Sort of the classic example that's been replicated quite a few times concerns the heritability of cognitive functioning, depending on socioeconomic status. And so what this again, it's been replicated a number of times at this point. And the general idea is that the heritability of sort of cognitive functioning may actually be highest in higher SES strata. So in other words, as income rises, as socioeconomic status, which is a broad term you'll see in the social sciences, simply refers to a number of things beyond just income and dollars. But in some ways, captures the functionality of an environment, safety of a neighborhood, you know, again, social status measured in a variety of ways, including income.

[01:28:45]  Brian: As those things go up, we might expect that the heritability of cognitive traits would also go up. And whereas in lower SES strata, where there's more socioeconomic disruption, where there's more economic struggle, we might expect that heritability estimates would be lower. And why would that be? Well, again, this is not unlike our example of nutrition.

[01:29:13]  Unidentified speaker: Right.

[01:29:14]  Brian: When we're asking the question, why are individuals different in a population? Part of that, in this case with heritability, it may well be that the far more salient reason that individuals differ in their cognitive abilities at lower SES levels is because of environmental disruptions, because there's a lack of resources, whether that's quality education access or access to proper nutrition. Any number of things in an environment that might disrupt the development of cognitive and executive function. Whereas as those SES levels increase, what happens is they start to lack the ability to explain those differences. And so that's when we might see heritability estimates rise. And I've used words like might and expect intentionally here, because even though I believe there was a fairly recent meta-analysis of these effects showing that they replicate well, I like to think of all of this stuff as provisional.

[01:30:22]  Brian: And so, but at least at the moment, when you see something like heritability estimates differing as a result of socioeconomic status, it is actually indicative of something interesting and something we might expect. And also keep in mind the classical twin study, which gives us heritability, shared environment, non-shared environment, just by itself in some ways is sort of acausal. The interpretation is differences in the population are associated with either environmental differences or genetic differences. You can take additional steps to sort of investigate causality more, but I think we should sort of resist the idea with the classical twin study anyway to say that genes are causing some outcome or the environment is causing some outcome. We can certainly use twin studies as a jumping off point. For example, let's say we show certain forms of dementia or maybe Alzheimer's disease in particular.

[01:31:33]  Brian: Those are moderately, relatively highly heritable traits, but it's only recently when we've been able to do more powerful types of genome analysis, where we've had large samples like a sample called the UK Biobank, those types of things where we can start to investigate whether certain genetic variants seem to truly have causal effects on the, whether it's by impacting brain structure or function in some way, in ways that sort of maybe create the prototypical plaques and tangles that we see in Alzheimer's patients, things that disrupt cognitive functioning in a physiological way, right, by impacting neural functioning potentially. And so I think while I wouldn't necessarily quibble with the points that Deutsch was making, I would simply add I don't know that it would lead you to sort of have a dim view of the ability of twin studies to help understand causation or explanation.

[01:32:39]  Peter: Can I interject with a quick question?

[01:32:42]  Unidentified speaker: Yes.

[01:32:42]  Peter: I think for myself, I know that as my sort of worldview has sort of, I don't want to say evolved, but sort of shifted from being, you know, very tied into Judith Rich Harris and Steven Pinker, those books and really that were sort of, like I said before, sort of almost like an explosion in my brain kind of a thing. Like, wow, the genetics is just, it influences everything. You know, you look at the family down the street and you assume that the parents are so responsible and the kids are so good. It's just, they think it's really all about genetics, you know, but maybe I was a little extreme in that direction.

[01:33:32]  Unidentified speaker: I don't know.

[01:33:32]  Peter: But as I sort of moved into being more influenced by Deutsch and just getting more into philosophy in a way, I've started to look at things more in terms of epistemology, I guess, rather than just biology, like just seeing human beings as more of our minds rather than brains, if that makes any sense. So it's almost like our minds are, as I understand it, software that's running on hardware. And software, I think, can always be rewritten and might not be easy to do, but you can always change your software.

[01:34:19]  Unidentified speaker: What do you think?

[01:34:20]  Peter: Sorry if this is just getting too much on the tangent, but are we our minds or our brains?

[01:34:30]  Brian: So I love that question, and I've come to love it even more in recent years as my work has sort of expanded to include interest in cognition, cognitive science, consciousness, both from a philosophical standpoint and a sort of more mechanical standpoint, in terms of like even how would you build an artificial mind that would be essentially human in its phenomenal awareness and those types of things. So all of those are tied in interesting ways to the question you just asked. And I think there's a number of ways to answer. The idea that just because we might start to get a better understanding of the way that structures in the brain, differences in those structures either cause one to be more likely or less likely to say develop some form of dementia. And then we, through additional work over the years, we link those differences in brain structure to as being a product in part of differences in certain genetic variants that one might carry or might not carry.

[01:35:56]  Brian: I don't know that that does anything to our ability to think more broadly about, you know, solving really pressing medical and social problems like Alzheimer's disease and dementia. In some ways, the, again, this was one of the sort of, I think, really wonderful points that Deutsch makes in a variety of places, is that the idea that problems are solvable, that unless there's some physical law demanding that, you know, the, you know, something is off limits, it's not in terms of it doesn't make the problem easy. It doesn't matter if the problem is understanding psychiatric outcomes or personality disorder outcomes or medical outcomes and solving problems that may arise in the context. But again, when we're if we're just thinking in terms of cause and effect, just because something may have maybe in part caused by some genetic influences, it does not there is nothing written in that that says, you know, change is impossible.

[01:37:15]  Brian: Disorders of vision are heritable. We correct those every day with environmental intervention. Vulnerability to migraine is, you know, I don't I couldn't call the heritability estimate to mind, but probably low to moderate correctable every day with new medications become available all the time. Actually, you just you see commercials pretty frequently for different migraine interventions that are, you know, whether you once a month or when you have a migraine onset or whatever. The point is, just because a problem is is hard to solve, does not then translate to it's impossible to solve. And just because certain causal effects impact some outcome does not imply that it is beyond the reach of change. And I'll I'll stop rambling in a second. I think the last point I would add, though, is we do. I think there is a temptation to when we start to make distinctions between genetic effects and environmental effects, is to presume that the genetic effects are the hard ones to, you know, to wrestle with, not the environmental ones.

[01:38:34]  Brian: And I think that's quite often can be mistaken. And so one example that I like to, you know, especially with students when I talk about these things in doctoral seminars or whatever, is the example of language as it relates to the accent that one speaks with. So all children, assuming, you know, there are no type of acquired injury or some type of problem preventing it, [excepting?] those cases, all children learn language. That's a capacity that's universal in humans, happens all over the world every day. It's incredible to watch whether you have kids or you have nieces and nephews or doesn't matter. It's wonderful to watch that sort of explosion of language in kids. And what's interesting is that certainly there, you know, we have the biological hardware necessary. And we understand more today about the neural regions associated with language functioning in various ways.

[01:39:39]  Brian: But the the language that we acquire and the regional accents that we acquire are completely rather differences in those things are completely environmental in nature. I have a, you know, at least I'm told, a bit of a southern accent had it my entire life. If I were to move to the northeast or the west coast, that accent not going anywhere. It's going to, I'm going to take it right with me. And the inverse of that is Pinker talks a lot about this in The Blank Slate. Judy Harris talks a lot about it in her book as well in The Nurture Assumption. You know, when linguists study children of, say, folks who have immigrated to a new country, what they find is that while at home they may hear their parents speak in the language of the country where they moved from. So let's say someone moves from the United States to England and their parents speak with American accent in English.

[01:40:43]  Brian: Their children who are exposed to their parents at home, but their peers say at school will speak with an English accent, wholly environmental. Now, that's interesting in its own right. But imagine how easy it might be if someone, whatever accent that you speak with, you were told, I can't speak with that accent anymore. You got to speak with an entirely different accent. It's possible. Actors do it. It can be, you certainly change your accent. I would argue it's probably not particularly easy to do. Nor is it easy as we age to learn a new language from scratch. When we're young children, human children do it, I mean, without any great difficulty. I can remember taking Spanish in high school and struggling mightily to try and get a sense of the language.

[01:41:35]  Unidentified speaker: [Same here?]

[01:41:36]  Unidentified speaker: Yeah.

[01:41:37]  Brian: And so what's interesting is the language we acquire is the result of geography and environmental effects where we're born. That in no way makes it easy to change your language or to acquire new language with age. So just my only point is simply because something may be very much influenced by the environment doesn't necessarily guarantee that it's easy to change. But all of that said, there's no reason, none of what we've talked about today, in my opinion, counsels any type of despair in terms of the changeability of some outcome.

[01:42:16]  Bruce: So let me try to make a steel man version of how I think I understand Deutsch's criticism.

[01:42:23]  Brian: Sure, absolutely.

[01:42:25]  Bruce: So let's actually use the example of accent that you just used. So by explanation, you just explained why it is that accent is actually 100% environment.

[01:42:39]  Brian: Differences in accent, but yes.

[01:42:40]  Unidentified speaker: Yeah.

[01:42:41]  Bruce: But let's say that I set up an experiment. Let's say I didn't know that for whatever reason. I don't have that explanation. So I set up an experiment. I'm probably going to find that there is some sort of heritability. If the experiment doesn't take the right things into consideration, the right confounding factors, I would almost assuredly find that there is a heritability to accent. Because it does in fact match up with genes quite well. So the issue here would be, how can you really know that when you say 30% of accent is explained by heritability, that that's even a meaningful statement? And that it wasn't just the confounding factors that you don't understand or know about at this point, and that you're not taking into consideration?

[01:43:30]  Brian: Yeah, so it's a good question. There are a couple of things going on there that I think we should probably point out first. One is that we wouldn't want to say that 30% of accent is the result of genes. What we would want to say is, and I'm not trying to pick on you unnecessarily, but the verbiage really is important. What we want to say is that 30% of the variance in accent might be the result of genetic variance or heritability.

[01:44:03]  Bruce: You say the result of, but let me be clear. I have definitely heard it said explained by. They do use the term explained by instead of result of in a lot of cases. I think that's what some people are taking exception to, but continue from there.

[01:44:18]  Brian: Totally worth pointing out. I'm certain guilty of it as well. And using sort of just common words like explained by as opposed to associated with. But so it's fair point. Well, what I would say then is to get those, if what we're after is a heritability estimate, we're not going to calculate that with a true experimental design. What you would be showing with that design is you would be holding everything, presumably if you're doing randomized trial, you're going to be holding everything else constant, including genetic factors in order to investigate whether some treatment or exposure that you're interested in has a causal effect on accent. You could certainly do that. But if we're after a heritability estimate or in theory, you could do it. I don't know that it necessarily be ethical depending on the treatment. But if we're going to calculate a heritability estimate, we're going to do a twin study.

[01:45:17]  Brian: And what you're going to find based on sort of the math of how these heritability estimates and environmental estimates are calculated, very likely that the heritability of accent is zero, at or near zero based on, and again, we take advantage of relatedness to get to that.

[01:45:38]  Unidentified speaker: Well, no, that makes sense.

[01:45:39]  Bruce: So we're not talking about necessarily, you could do a random controlled trial with accent. I don't doubt that, but let's say we're not for whatever reason. You could still do a causal study of accent, even without a random controlled trial, by simply setting up the right kind of natural experiment and taking the right kind of data with twin studies or something like that. And then you would very quickly discover that accents actually determined by environment, not by genetics. And in fact, even though you didn't do a random controlled trial, you would get the right result. You would find zero heritability to accent if you're doing a twin study.

[01:46:22]  Unidentified speaker: Okay.

[01:46:22]  Bruce: I'm with you.

[01:46:23]  Unidentified speaker: Okay.

[01:46:24]  Bruce: So let's talk about then. So we've kind of danced around this, and this might even just be too technical for me. But when I read your papers, it gives a very technical explanation of how they work out causality when you don't have a random controlled trial. And I admit, I just cannot make sense of what the paper is saying. It is too technical. It is stuff that I'm just not familiar enough with. Is it possible for you to try to give me some reasonable examples to help me understand how it is that you can determine accent is caused by environment, not heritability, or slavery is caused by environment, not heritability, and how you would work that out in trials if you can't do a randomly controlled trial?

[01:47:17]  Brian: Absolutely. So I think a useful way to think about this might be with a specific example. So I'll take one, let's see, take maybe a more psychological or medical example. Let's say that we are, well, we'll go back to smoking and lung cancer, because I think that's a useful one. The way that, suppose we have an explanation of lung cancer that involves exposure to cigarette smoke, right? We think that there are a number of lab studies showing that the carcinogens that we know to be present in nicotine smoke are capable of inducing mutations in DNA that then typically form tumors. And so we have that little, we have that piece of evidence in our mind as we're thinking about explanations for why lung cancer tends to happen. And we know, of course, that we're not going to take healthy humans, randomly assign them to smoke or not smoke because that's just, it's off the table and should be.

[01:48:36]  Brian: And so we start to think, okay, what are some of the things I can capitalize on that will let me do a better job at making the argument that smoking causes lung cancer? Or maybe that's not even the best way to phrase it because maybe we're not trying to push an argument, but we're trying to be as harsh as we can on our explanation that smoking causes lung cancer. We're trying to test it.

[01:49:04]  Unidentified speaker: Right.

[01:49:05]  Brian: Exactly. So we're going to do our level best outside of an experiment to subject our explanation to criticism and maybe refute it in some general sense, right?

[01:49:19]  Unidentified speaker: Right.

[01:49:20]  Brian: So one of the ways we go about that is we have a large sample of twins, and what we're going to find is that in that sample, some of those twins, whether they belong to an identical twin pair or a fraternal twin pair, some of those individuals are going to be smokers and their sibling won't be, right? So the term we used earlier when we were talking about this is discordance. There'll be something present in one twin, something not present in the other. And so what we can do, and there are different ways to do this, but I like this example specifically because it sort of cuts right to the heart of it. What we've done when we've made the argument that smoking causes lung cancer, or we've explained lung cancer as being a product of cigarette smoking, we've invoked some element in the environment that is acting as a treatment of a sort and causing some particular outcome.

[01:50:21]  Brian: In this case, the exposure or treatment is smoking, the outcome is cancer. So what we want to do and what experimentation allows us to do in other cases is we want to rule out any other possibility that something else might explain the association between smoking and lung cancer. We want to hold everything else constant, is the language you'll hear researchers use. And when we especially, let's say that we're not, and this actually is a good opportunity to make another point, because it's a theme in our paper. There are reasons to use twin studies that have nothing to do with being interested in genes. You can adopt a discordant twin design and not care one bit about genetic effects. And that's completely defensible. In fact, in some cases, heritability estimates just aren't interesting particularly because we sort of expect to find them.

[01:51:25]  Brian: And in those cases, those heritable factors represent a nuisance that we need to control for, just like anything in the environment might represent a nuisance. So what the identical twin design allows is, of the things that we're going to hold constant, we're going to do, ideally, when we use this design, we're going to rule out the possibility that genetics or shared environment, or family environment, shared environmental experience have anything to do with our outcome of interest because twins share both their genome and their environment, identical twins. 100% genetic overlap, 100% environmental overlap. So what we're doing in this case, in the most basic sense, is we're trying to see if it is the differences in smoking that might be causing the differences in lung cancer. And let's say we do our twin study, and that's what we find. We calculate these effects in the data.

[01:52:30]  Brian: We show that there's a considerable increase in cancer rates for twins who smoke versus twins who don't. It's not a guarantee because in some cases, maybe one twin smokes, the other doesn't, but both develop lung cancer. That suggests that cancer is partly heritable to some extent, right? We may not know what that means mechanistically, but that's what it would imply. But using this design, showing this continued discordance between smokers and nonsmokers for their cancer occurrences should be interpreted, assuming we meet other assumptions, as pretty compelling evidence that the smoking is causing cancer.

[01:53:15]  Unidentified speaker: Right.

[01:53:15]  Unidentified speaker: No, that makes sense.

[01:53:17]  Bruce: Would the following statement then be true? So let's take the example of slavery being heritable. Would the following statement be true? That the only way we would actually find a study that slavery in the 19th century was heritable is if we did the experiment wrong to begin with, and that if we did the experiment right, if we set it up correctly, we would find that it is not heritable even back in the 19th century.

[01:53:40]  Brian: Well, so I think the more general way of thinking about this is to, one, not to necessarily invoke language about experiments versus, remember, when we're doing a twin study, we're not doing an experiment.

[01:53:58]  Unidentified speaker: That's true.

[01:53:59]  Bruce: Experiment may be the wrong word.

[01:54:01]  Brian: I think maybe this might be a helpful way to think of it just in general, right? And it goes back to the idea that heritability estimates are not fixed constructs. They're variance components that can change over time. So thinking about when I see a heritability estimate or an environment, and this holds for shared and non-shared environment estimates as well, is that that by itself does not tell me that gene, there are strong causal genetic effects on a trait, nor does it guarantee that there are strong environmental causes of some trait. What that tells me in and of itself is that in that sample, at that time, this variation or variance in this trait that I'm interested in is partly explained by genetic differences or partly associated with genetic differences, partly associated with environmental differences. That's the extent of what I can say.

[01:55:01]  Brian: If I want to probe that further, investigating various causal effects, I'm going to need to do something else in my research design. That might be a genome-wide association study. That might be a discordant twin study. I'm not talking specifically about Deutsch's example about slavery, but I'm talking as a general principle. I think this is a useful way to think. It's that if we just simply looking at the results of a twin study showing that some trait is partly heritable, it tells you something useful, but that thing that's useful has pretty strong limits on it. If you want to start to pursue other questions, you're going to need to do some additional things.

[01:55:46]  Bruce: Let's talk about a genome-wide association study, because that comes up, and then the other one was a polygenetic score, and I didn't really understand what those were. I've got some quotes about them.

[01:55:57]  Brian: Yeah, maybe go ahead with the quotes, and we can talk through what those maybe mean.

[01:56:02]  Bruce: Okay, so one common use of model comparison is to test a full ACE model against a trimmed AE model in which the C parameter is dropped, A meaning heritable, C meaning common environment, E meaning shared environment [wording as spoken]. This method evaluates each model in terms of parsimony. Does dropping C lead to improved or diminished goodness of fit? MZ twins, discordant for some variable, can be used to evaluate environmental effects with greater specificity. Cross-lag, longitudinal discordant twin design, which makes this particular design desirable, a point which may have been already intuited by readers, is that it improves the ability to test whether changes in one trait cause changes in another over time. So that does seem like... So you're saying that, first of all, maybe help explain what a GWAS, Genome-Wide Association Study is, and does it actually allow me to work out without an experiment causation?

[01:57:08]  Bruce: Or can it? Or how would it?

[01:57:12]  Unidentified speaker: The short...

[01:57:13]  Brian: Okay, so one thing to clarify. The quotes you read were dealing with applications of twin and sibling designs. A genome-wide association study does not necessarily... They can, they do not, by definition, have to include siblings. So like we say in our... We've got a brief little quote in the paper, and this refers to using GWAS studies to calculate heritability estimates, but we say GWAS has helped generate new methods for estimating heritability without the need for siblings. We can use it with, or we can do it with genetic data alone. So what is a GWAS study? A genome-wide association study is sort of followed in the wake of what we talked about as candidate gene studies, where maybe we would take a cheek swab from a sample of 100 or 1,000 participants in a study. We would do some genetic analysis. Maybe we're interested in a particular dopamine gene or a serotonin gene to see if those things increase vulnerability to depression or anxiety.

[01:58:24]  Brian: So we do a cheek swab, collect some buccal cells. We do probably PCR. And that way we're collecting data about individuals in our sample. And let's say that this serotonin-relevant gene that we're interested in comes in one of four varieties. And so what our candidate gene data will tell us is which one of those four varieties each person in our sample carries, right? Version one, two, three, or four. And then two copies of both. So whether they're homozygous or heterozygous. And so what that study would then do is try and correlate to see if there's a correlation between the genes that individuals carry and the outcome that they're interested in, maybe anxiety. A GWAS study, those studies, as we talk about in the paper, had quite a bit of problematic issues with them, small samples, just things that made it very difficult to have, to put a lot of, you know, much of any stock in the findings.

[01:59:29]  Brian: So with moving forward in time and with the advent of new technologies and as those technologies became cheaper, it was possible to, instead of looking at one or two places in the genome, we could look at hundreds of thousands of variants in the genome. And so, and then use that information both to, it can be used to estimate its own type of heritability estimate, or it can be used to construct what you ask about, which is a polygenic score. And so, what a polygenic score, and there's much more. There's a sort of technical treatment in the paper, and I'm going to oversimplify here, but anybody interested more in the technical aspects of it, certainly that's available, and hopefully we made them accessible as we could in the paper. But the idea is basically we're sort of, we're taking the results of GWAS and sort of trying to calculate the collective presence of a genetic propensity for some outcome.

[02:00:37]  Brian: So it might be a polygenic risk score for schizophrenia. So we're, you know, any part of the genome of the, say we looked at a thousand places in the genome and, you know, we found a bunch that seemed correlated with schizophrenia. So then we would create a new variable and that would be the polygenic score. And it would basically capture the level of genetic risk that folks in the sample carried. And so that would differ as well. Maybe I would have higher scores on the polygenic risk measure, meaning I would have potentially higher genetic propensity for some trait. Others would have lower scores on the polygenic risk scale, meaning they would have a less apparent genetic risk for some trait. And then we would have a sample full of information like that. We could then use that polygenic score to investigate it further to say, you know, see if it has direct effects on some type of brain structure or anxiety traits themselves.

[02:01:43]  Brian: That alone does not, that's not a guarantee of causal effects. There are other things that can explain that association. But what's interesting is, and there's some really cool recent papers that have done this. And even if we don't have time to really get to them this time, I can send them your way and potentially talk about them next time. What I think is particularly important about not only GWAS studies, but also sibling studies, twin designs, these data can be combined. You can use information from GWAS, derived from GWAS studies, something like a polygenic score, and put them in a sibling design. And so in that case, you can investigate, say you use just fraternal twins, DZ twins. They might differ in their polygenic risk because they differ from each, they only share half of their DNA. The other half is different from their sibling.

[02:02:36]  Brian: So these types of data can be combined, including with sibling data. And that does start to give you some fairly interesting ways to investigate possible causal routes. But just simply doing a GWAS study, what you've done is potentially reveal associations between some genetic variant and some complex trait. If you want to talk more about causation, especially if you want to begin to understand how that causal effect plays out, there are certainly, you know, it's plausible to do, and there are other very interesting, fascinating designs that then start to allow you to investigate, all right, there seem to be these genetic variants associated with, say, something like schizophrenia. How would that be possible? Well, we might start to take more granular looks at things, and what we find is that this particular set of genes seems to be very much enriched in this particular type of neural tissue, right, that influences an area of the brain related to language functioning, or it disrupts, you know, language functioning in some way.

[02:03:53]  Brian: So that starts to give us, it starts to give us a picture of mechanisms and pathway. So that is, I know that's a very long answer, I apologize for rambling, but it sort of echoes that point earlier of in isolation, showing a trait to be heritable, finding GWAS [correlates or?] hits with genetic variants in a GWAS study in some trait, that alone is not dispositive for really anything except the finding in that study. It has to be, you know, you need to continue to put the pieces of the puzzle together. And actually, the other thing I would add is, to me, it makes a very strong case for the need of guiding theory or explanation, if you want to use more Deutsch's terms, you need to be working with good, hard to vary explanations that guide you in terms of where to look, what pathways and that kind of thing.

[02:04:54]  Bruce: Let's go back to the 19th century slavery example. Let's pretend, obviously, in this case, part of the reason why he's using this as an example is because we already know what the explanation is. But let's pretend like we didn't, right? We're living in the 19th century, we're aliens observing the humans or something, and we're missing a lot of things that the humans would understand. And we're trying to figure out if slavery, to what degree it is heritable. But they're smart aliens, they're good scientists, they know how to set up a good natural situation like this to try to study it. How would these aliens go about trying to look at the heritability of slavery? And what would they find? Would they find that it is or isn't heritable?

[02:05:40]  Brian: So, I mean, it's sort of a perplexing way to think about the issue. Well, again, what I would say is that one, experimentation is something different than trying to...

[02:05:56]  Bruce: Yeah, I tried to drop the word experiment this time. You're looking... It's maybe a natural experiment would be as far as we can go.

[02:06:04]  Brian: Okay, and in regards to heritability, it's a tricky thing to speculate about. But let's say that I think Deutsch says that in his thought experiment, which we'll just take on face that you happen to arrive at some non-zero heritability estimate. I mean, the obvious implication is that it would be a mistake to think that that tells you anything about the causes of slavery. And I think, but again, this is kind of why I said earlier that I don't necessarily take issue with those points on their face, because the same is true for showing that, you know, there are correlations between genes in a GWAS study and some other complex trait. That's not enough. Or showing that a trait is heritable, you know, were you to take from that and infer that, you know, the environment is sort of irrelevant, that it's a genetically determined trait. That's an erroneous conclusion for, you know, for all of the things we talked about.

[02:07:13]  Brian: And so that's why heritability estimates just represent a piece of the puzzle in the same way that, you know, environmental sources of variance represent another. And then to build more insightful types of explanation, it requires a variety of things, other moving parts that go much beyond a single type of study design. And I think that's what makes sort of, I think, behavioral genetics a tricky field to contemplate in some ways, is that all of the points that we've talked about apply in studies that, you know, just normal social science studies that you see every day. Maybe you might find a correlation between someone's level of agreeableness and their job success, just to pick a personality example, right? So agreeableness is a broad personality trait. And let's say we do a study. We've got a thousand individuals. Nobody's related to anybody else.

[02:08:14]  Brian: It's a sample of just individuals drawn from the population. You got a sample of a thousand of them. We measure their level of agreeableness. We somehow measure their occupational success over time, right? And we find that more agreeable people tend to do better in their chosen occupation, right? So there are so many reasons why we can't, just based on those correlations, infer very much at all about causal effects, genetic effects, environmental effects. They're all wrapped up in one package that's impossible in that setting to pull apart. If you could do an experiment where you randomly manipulated someone's or individual's level of agreeableness, which is impossible, but just for the matter of a thought experiment, imagine we could do that. Let's say we could just turn dials and randomly set people at various places of agreeable or disagreeable on that personality trait.

[02:09:16]  Brian: Then we, you know, that would be a much stronger step in the direction of saying that agreeableness causes occupational success. Now, if we thought about that, obviously we can't do that. We thought about it in terms of the discordant twin design we've been discussing. What we might look at is differences in agreeableness that emerge between identical twins, because that's certainly plausible. One twin may be more agreeable than the other. Identical twins don't correlate perfectly for their personality styles. Obviously, this is easily observable for anyone who knows twins. They can often be quite different from each other in terms of their personalities. They're similar, but there are also differences. In identical twins, the more agreeable twin, on average, tends to have better occupational success. That's, I would argue, a stronger piece of evidence than the first example where we just used the 1,000 unrelated individuals.

[02:10:18]  Brian: But as these things start to accumulate, that's when we start to understand our phenomena of interest, but just sort of overly simplistic reliance on something like a heritability estimate. Deutsch rightly rejects that, and I think in general, we should reject any type of lazy conclusions about heritability.

[02:10:41]  Bruce: Could we do something similar with the 19th century slavery example, though? Could we come up with a way of collecting the data using discordant twins or something like that, that would allow these aliens who don't understand to realize this is not actually a heritable trait?

[02:11:01]  Brian: Perhaps if I sat down and thought about it long enough, because I would need to think through the design. Honestly, it sort of fits in the category of one of those examples. I'd prefer not to speculate too far about, because then you run the risk that people might think you're actually serious about contemplating an explanation of that particular outcome.

[02:11:26]  Bruce: Well, in this case, I'm assuming that it's that isn't the explanation, that heritability isn't the explanation. I'm just wondering how the aliens could figure this out, that that isn't the explanation.

[02:11:36]  Brian: Yeah, so no, it's a completely reasonable question in the sense of how do we just a broader sense of how do we avoid erroneous conclusions.

[02:11:46]  Unidentified speaker: Right.

[02:11:47]  Brian: And in some ways, as we've already said, that just calculating heritability and environmental estimates on its face is not enough.

[02:11:56]  Bruce: On its own, right. But could they even reasonably figure out that it's not really heritable? Or does heritability as a concept mean nothing more than correlation?

[02:12:10]  Brian: Well, so let's go back to the example of that we used for cognitive, the heritability of cognitive traits.

[02:12:18]  Unidentified speaker: Right.

[02:12:18]  Brian: And the finding in sort of that's emerged several times in the past that certain cognitive traits become more heritable as socioeconomic status increases.

[02:12:31]  Unidentified speaker: Right.

[02:12:31]  Brian: And so in a case like that, we, let's imagine we just, we conducted the first twin study of, you know, some cognitive trait. And we find that it's about 40% of the variance is heritable variance and about 60% of the rest of the variance is explained by the non-shared environment. Just looking at that, again, that alone tells me some interesting things, but those interesting things have limits. So let's say that I, one of the ways that we might expect that we might, you know, sort of conjecture and refute certain ideas to go back to Popper's vernacular is, let's say, all right, think about what these estimates are telling us. These are telling us about where, what we can attribute variance in this sample to. But again, that goes back to some of the differences are seen linked to genetic differences. Other other aspects of these differences seem to be a product of the environment.

[02:13:33]  Brian: In fact, 60% of the variance in this case was explained by environmental variation. So that might prompt us to think, because we know some other things about cognitive traits that we've amassed, knowledge we've amassed over time. One of which is cognitive development in general is vulnerable to environmental disruption, lack of calories, environmental adversity of different kinds, discrimination, racism, a whole bunch of environmental things that vary in the population that can impede cognitive growth in human beings. That would lead us to then expect, all right, let's look at, because we also know that heritability estimates are not fixed constructs. They can and do change in certain settings. So now that could get us to the point of thinking, all right, let's look, let's specifically investigate these heritability estimates across different strata of socioeconomic environment.

[02:14:36]  Brian: And then, you know, we sort of arrive at the finding that's that's emerged several times, which they do change. Those heritability estimates vary depending on socioeconomic effects. And again, I don't want to harp on this too much in the sense of there might be a study that comes out tomorrow that finds relatively stable heritability across the strata or it replicates the original finding again. That's in this case sort of secondary. The point is, we have to do some additional work. And also remember sort of the basic tenets that we talked about earlier, that these things aren't, you know, sort of concrete, irrefutable pieces, qualities of whatever trait we're studying. There are many more moving pieces to the puzzle.

[02:15:23]  Unidentified speaker: Right.

[02:15:25]  Peter: We've kept Brian a long time. Do you have another question, Bruce? Or can I ask a quick one?

[02:15:35]  Bruce: No, go ahead.

[02:15:39]  Peter: Sorry if this takes us on a little different path, but I think it's related. You know, on the podcast last week, I was kind of mentioning how, you know, my field is education. And, you know, oftentimes when I see sociological studies, or studies about education or kinds of related soft science fields, my BS detector just goes crazy.

[02:16:05]  Unidentified speaker: I don't know, I can't help it.

[02:16:08]  Peter: Oftentimes, I think that it might be because they are making assumptions about humans, whether related to behaviorism or Marxism or other political kinds of things. It just kind of does not ring true for me. And then after that, just the conclusions just don't really go anywhere. And as I've gotten more into Popper and things, I think I might have amplified that in a way. I'm not going to, I'm certainly not saying anything that you've said here or in your papers or your excellent Quillette articles. I meant that I would encourage everyone to go and read. I think you've written about 10 for Quillette, which are just great. But just to sort of play devil's advocate, I just wonder if it's just possible that there are assumptions in here in this kind of research that just maybe aren't completely valid. If I may just read something I wrote on Facebook.

[02:17:21]  Peter: It didn't get too many likes. Maybe it's not as clever as I thought. But let me just read it and see what you think.

[02:17:28]  Unidentified speaker: Go for it.

[02:17:28]  Peter: Person one. Hello, friend. Here's a thousand books and a thousand movies that you must experience. Also, I'm going to keep you in my house for two decades and explain to you in minute detail my philosophy on life until the thought of me speaking to you makes you groan. Person two. Sounds good. Nothing else going on. Person one. Oh, but scientific studies indicate I don't actually have any influence over you. That comes from nature. Person two. WTF. So the idea is that I think just on a human level, I mean, does it even make sense to say that parents don't influence our children? We're keeping them in our house for 20 years. We're interacting with them in the most intimate way pretty much. You can interact with another person. I mean, does it even make sense that you can parse out nature and nurture and the influence of genes and all this?

[02:18:32]  Peter: Do you see where I'm going with this?

[02:18:35]  Brian: Oh, certainly. Absolutely. It's a great question.

[02:18:37]  Unidentified speaker: I would just love your thoughts.

[02:18:41]  Brian: Well, I think first of all, I'll start by agreeing in some ways to the idea that it's not intuitive at all, that something like shared family environments might not matter in terms of, and I'll put matter in sort of air quotes. We'll get back to what that means in a second, but may not be as impactful on human development as we presume them to be. But I think that in and of itself sort of implies something that I think is pretty widely appreciated in science, certainly by you all and probably anybody listening, is that the nature of reality can often be pretty severely at odds with our intuitions about the world. Sometimes they align and other times they don't. And in this case, if we're going to sort of scrutinize those intuitions, there are different ways we can do it. And by different ways, I mean different research designs. One way is the twin method we've been talking about.

[02:19:40]  Brian: Another way is sort of the classic type of method that's often referenced with twin studies, and those are studies of monozygotic twins separated at birth and raised apart. That's another way to sort of scrutinize this particular idea about parenting effects. Another way is adoption studies. So to look at adopted siblings, in this case, siblings who are unrelated to each other and unrelated to their adoptive parents, but are raised together in the same home to see how similar they are to each other. And so we can sort of start with that. In general, what you find is that while there might be some degree of similarity in terms of their personalities, their interests, political orientations, those types of things, there might be some superficial similarity early in life. Most of it's gone by the time they're in adolescence or adulthood. So if we presume any effects of the parenting environment that function to make those adopted kids similar to each other, those effects have pretty much dissipated with age.

[02:20:56]  Brian: Another, again, going back to the MZ twins reared apart, what we tend to find is that we might expect those exposures to sometimes vastly different homes, not always, but quite often very different experiences would create quite a bit of difference. In fact, if we think especially in terms of parenting effects being sort of the foundational things that shape human development, we would expect those individuals to be quite different because they don't share any parents, don't share any formative years in the same environment. What we find is that they're remarkably similar. Personality traits, cognitive traits, educational attainment, all the way down to sort of more superficial things like hobbies and extracurricular interests, those types of things, incredibly similar. And then you can combine that and the findings of twin studies where the shared environmental component is where we would expect sort of parenting effects that make kids similar to each other.

[02:21:58]  Brian: That's where those effects would live. And, you know, as we mentioned with the laws of BG, behavior genetics, and then just sort of repeated findings, those shared environmental effects can sometimes show up in childhood, maybe small to moderate effects, but for quite a few traits, they're gone by adulthood. And then for other traits, they're never really there in strong effect at all. There are a few exceptions to this, but this is sort of a general rule. And then that's, you know, so we sort of pieced together this puzzle in a variety of ways. And I think there are a couple of things that create the consternation that folks feel when they're sort of coming to this literature for the first time. One is that there are all of these studies showing correlations between what parents are doing and some outcome in the kid. Well, what you have to, one thing to remember there is unless we're using some type of twin or sibling design, it's as likely that the reasons parents' outcomes correlate with child outcomes is because of genetic overlap.

[02:23:09]  Brian: Parents also share genes with their kids. And so we need to account for the correlations that might exist between parents and children that are linked to genetic correlations, genetic overlap that they have. So twin studies, sibling studies allow us to account for that. Adoption designs allow us to account for that in a certain way. And we just don't find these strong parenting effects. And it sort of has led to the general conclusion, at least the argument I would endorse, is that most of the reason that biological children who are raised with their biological parents tend to be similar to each other is because they share genes with each other. The additional shared home environments not contributing much more on top of that. Now, the second reason this can prove frustrating, I think, has to do with forgetting that most of the time heritability estimates and shared environmental estimates are often much smaller than non-shared environmental estimates.

[02:24:18]  Brian: So we forget that. This is not saying that biology is all that matters, and quite the contrary. In fact, it's saying most of the time the environment explains most of the differences between humans. It's just not the family environment. It's the non-shared environment. All of the things that happen to us that make us different from our siblings. So it could be anything from one sibling finds themselves in a particular peer group at school and the other one doesn't.

[02:24:44]  Unidentified speaker: Peer influence.

[02:24:46]  Brian: Yeah, exactly. And so the other thing I would add to that, though, is there is a completely defensible, plausible mechanism by which the way parenting effects could manifest themselves. And that's through the non-shared environment. And I think in this case, imagine that, you know, and maybe folks who have kids might imagine this too. It's completely plausible that parents can act in ways that treat their children differently. And that can create differences between the children. And so in that case, parenting would be an environmental effect, but it would be a non-shared effect. I'm not saying there's a ton of evidence for that. I'm just saying it's completely plausible that parenting could function to increase differences between kids. Although one thing that needs to be kept in mind there is that in many cases, parents, the differential treatment that can emerge across kids is more a result of the differences sort of pre-existing in the child that the parents are reacting to versus the differential treatment causing the differences in the two siblings.

[02:25:57]  Brian: But that aside, it is, you know, sort of there are varieties of pieces of evidence that sort of arrive us at the idea that developmentally speaking, personality speaking, those types of things, the shared environment effects are not as large frequently as either genes or non-shared environment. Now, the final thing that creates consternation is people, I don't want to say that too accusatorily, there can be a temptation to conflate what I've just talked about with the idea that parents don't matter. And I emphasize that word. And in that regard, I think that is just a very, that would be a conclusion that would make me fairly despondent because we were talking about earlier before we started recording. I'm fortunate to have a wonderful family. So many good memories with my parents, mom, dad, brother. But my brother and I happened to be adopted.

[02:27:00]  Brian: So we fit within that framework of adoption design. We're not biologically related to each other or to our parents. And we're very different from each other. Our personalities are exceedingly different. We're different people. But that doesn't mean that we don't, you know, both have wonderful, you know, lasting memories that we can call up of, you know, fun times with parents or them helping us get through difficult times in our life and just a wellspring of good things that we can call to mind. And those things absolutely matter. And the converse is also true. It's completely plausible that, you know, and happens sadly a fair amount. Parents do terrible things to their children and put them in psychological or physiological path of harm.

[02:27:57]  Peter: It seems like that's the caveat in a lot of these studies. Trauma is kind of like you want to like isolate that, right? The trauma.

[02:28:05]  Unidentified speaker: Yeah, certainly.

[02:28:06]  Unidentified speaker: We'll see.

[02:28:06]  Unidentified speaker: So, all right.

[02:28:07]  Brian: This actually has a methodological component to it. It's the idea of restricted range. Most of what parents, what it's also capturing is what parents do in sort of a normal range variation. Too far out in that curve to where you're talking about abuse, neglect, maltreatment, that does not apply to what we've said. Those are sort of qualitatively different in terms of parenting effects, and there is good evidence that those matter for different outcomes in child development. And so, the other point I like to make, this actually, I wish I had thought of it, Judith Harris and Pinker both provided the different versions of it. I think Judy's conception might have been first. I think it was because Blank Slate came later from Pinker. But this idea of, we imagine that part of the moral obligation parents have to kids is that they are playing a causal role in shaping various components of their development, their psychological development, their moral development, the nature of their character, those types of things.

[02:29:20]  Brian: So we assumed that that's part of the moral obligation. But the first point to make there is, even if those effects are not as strong as we think, it doesn't damage the moral obligation at all. There's still a moral obligation of one human to another to not cause grievous harm and to provide, especially when you have parents and children, right? Because it would also probably be a mistake, and I don't know if it would be one a lot of folks make, to assume that, say, one spouse who, let's say they met when they were in their 20s, met at college or whatever, get married, have a long-term relationship, whatever, it would probably be a bit mistaken to think, I'm going to cause significant personality change in my spouse over time. I'm going to remold them into the person I want them to be. That's probably a mistake that you don't make many times in relationships.

[02:30:24]  Brian: What you find is that part of the reason that attracted you to the person was the nature of who they were already. And so I don't think we would ever be tempted to conclude that just because we don't have strong causal effects on their personality development, that we can treat them any way we can or want to, or that our relationship doesn't matter because I'm not causing their cognitive traits to develop. I mean, that's a recipe for disappointing and disastrous relationships.

[02:30:56]  Peter: That's an excellent perspective.

[02:30:58]  Bruce: So I wanted to actually ask about, I read a book, and I want to say it was Pinker, but it may not have been Pinker, where they made the case that while when we talk about parental influence, what they're trying to measure is what your actions in the environment you're creating as a parent, how it impacts the child, but that isn't actually the main influence of a parent. The main influence of a parent is who the parent is. So for example, and this is what the book claimed at least, I don't know if it's true or not, that like if the parent has a high education or is wealthier or something along those lines, that the very fact that you were born into that environment will have an influence on that child compared to if they weren't born into that environment. Even if it's not specifically what you as a parent is doing, that is, you know, you decide you're going to read them Good Night Moon every night may have no impact at all on them.

[02:31:59]  Bruce: But whether that you were, you know, the very fact that they were born in a marriage or in a more wealthy environment, or they know that you have a higher education, that those things do in fact have an impact. Is that true or false? And I wish I could remember well enough which book I'm thinking of.

[02:32:20]  Brian: I didn't find much disagreeable about that at all. I mean, again, this goes back to the idea that the no one, certainly people might make this argument. I would always like to be on record as not making the argument that biology is all that matters. I mean, the evidence is far to the contrary. The environment definitely matters. That twin studies provide some of the best evidence of that. But yes, it can include things like what you talk about in terms of some things that we might even equate with just good luck. The idea that we have good health because we were born not in a time before modern medicine really took hold. Absolutely. And then the examples you give of yes, even within the course of a lifetime of a child, so maybe two parents decide to have children later in life when their finances are more settled as opposed to earlier in life when things are much more economically unsettled.

[02:33:31]  Brian: And I wouldn't rule out the possibility that could certainly change a number of things in terms of the experiences of the kids. So yeah, all of that to say, yeah, I think what you say is plausible.

[02:33:46]  Bruce: If I recall correctly, the book was claiming that there is absolutely an influence of parents on children, but it has more to do with who the parent is, not what actions the parent chooses to make. That was the point the book was trying to make. And I wish I could remember which book it was.

[02:34:05]  Brian: I think in some ways that could be seen as a version of sort of what I said earlier is that most of the similarities or correlations that we see emerge between biological parents and their biological offspring when they're all raised in the same home is more the result of the genetic overlap that exists between parents and offspring. So in other words, part of sort of who they are in that regard, that the additional things that happen in the shared environment don't do much to alter, you know, make things more similar between kids, if that makes sense.

[02:34:43]  Peter: I think this perspective can really make people a better parent, really. You know, that's how, when I really got into this, these twin studies and read the books we mentioned, it felt kind of liberating in a way, because I didn't feel like I was molding my children into perfect humans. I just felt like I could just do what made sense to me and just enjoy my children, really. And, you know, I read to them every day, not because I think it's terribly influential to them, it's just because I love to read to my children. And, you know, I just try to think about what I enjoy, too. And, you know, just enjoying life together and not like, you know, children change us, too.

[02:35:41]  Unidentified speaker: And I think it's just...

[02:35:44]  Unidentified speaker: I don't want to cut you off.

[02:35:45]  Brian: It's funny you mention that, though, because it has a more... It's a very interesting, good, nuanced point. It goes by the name often in sort of the developmental literature as child-driven effects. But it is part of what you described, that children impact their parents.

[02:36:02]  Unidentified speaker: Yeah, yeah.

[02:36:03]  Brian: Sometimes in unexpected ways. My wife and I don't have kids yet. We hope to one day. But I can still imagine the idea that, you know, the experience of parents vastly changes whenever they have, you know, another human being that they become completely responsible for. So, yeah, I have no doubt that that's fairly landmark change in a person's life.

[02:36:29]  Bruce: The other thing I wanted to ask concerning this, though, was like, let's say we took religion as a factor, OK? Now, undoubtedly, there is a genetic factor to religiosity. I don't doubt that. But what religion you are is definitely not random, right? If you do go on to be a religious person, there's a very good chance it's the religion of your parents, because that's the one you're familiar with and that you know and that you grew up with. So it would be hard to say that that wasn't an influence of your parents, right? If that makes any sense. But that may not be like a trait that these sorts of studies would even be interested in necessarily.

[02:37:11]  Brian: No, it's a very good example, and it would fit in the category of something I might give a general exception to, depending on what precisely we were talking about. So you used an interesting term first, and that was religiosity. So if we think of religiosity as sort of the general combination of personality styles and traits that lead one to be sort of inclined to seek out spiritual experiences, religious experiences, those types of things, that does show sort of small to moderate heritability estimates, not surprising like every other trait, huge environmental effects. But if what we're talking about instead, if we be a little careful with the language, and we say maybe denomination, so one might be Baptist, Methodist, whatever the case might be.

[02:38:03]  Unidentified speaker: Christian versus Hindu.

[02:38:05]  Unidentified speaker: Sure.

[02:38:06]  Brian: Certainly. Absolutely. Yes. Depending on where, much like accent, depending on place of birth and elements of family culture and those types in broader national culture, time period in which one was born. No doubt about it. That's, I would say, very much akin to accent that we were talking about and language that one speaks. Yeah.

[02:38:32]  Bruce: I guess what I'm trying to say, though, is that that would be a case where you would expect a very high influence from parents.

[02:38:41]  Brian: Potentially, yeah. So if you did, let's say you did a twin study of religious affiliation somehow, you know, and quantified it, this would prove methodologically tricky, just the nature of how you would measure it. But imagine we could do it. Yes, I would imagine large and significant shared environmental effects, nothing really in terms of heritability.

[02:39:05]  Unidentified speaker: Right.

[02:39:08]  Bruce: I guess my point is that that's probably not a trait you'd be interested in looking at scientifically to begin with. If for no other reason than because you already have a really good explanation as to what, that it's going to be a shared environment trait.

[02:39:22]  Brian: I don't know that I would go that far, because we shouldn't necessarily forget about the non-shared environment. And there are studies, not a ton of them, but there are several studies, happy to share those with you too, looking at it's a type of [twin-difference?] design, looking at differences in religiosity, predicting differences in overt delinquent behaviors. And at least some number of studies have found some evidence suggesting that higher religiosity corresponded to less delinquent involvement. So more pro-social behavior. So that's, yeah, it's still interesting, certainly in a social science framework of the effects of exposure to religion in various ways and how that manifests as behavior. Twin studies are quite useful for that.

[02:40:12]  Bruce: No, I could see that. I was thinking more about the denomination or...

[02:40:15]  Brian: Sure, yes. That, in that regard, just pure denomination, yeah, I think your intuition is pretty good. I would, at least speaking for me, and if I'm, what I want to focus on or behavioral outcomes or whatever, I'm gonna, it's more the religiosity type of thing. I would be interested.

[02:40:32]  Bruce: What I'm suggesting though is, as an answer to Peter's question, it may be the types of things that parents do in fact influence children on happen to fall often into a category where we wouldn't be that interested in asking about them scientifically to begin with. So that wouldn't be what the study would be about. The study would want to study religiosity, not which denomination you're in.

[02:40:56]  Brian: Yeah, I think that's fair in a lot of cases. Maybe if we sat here long enough, we might think of some exceptions to its other examples [wording unclear], but I'm, yeah, it makes sense.

[02:41:05]  Bruce: Okay, let's wrap this up. I guess, Peter, do you have any final questions before we call it quits for today?

[02:41:13]  Peter: Keep going for six hours, but I think we've kept Brian long enough and I've truly enjoyed it. It's just, especially for someone who is more of a lay person who doesn't, I mean, I'm not in the world of academia or anything, and just to be able to talk to someone as smart as you are, Brian, it's a thrill.

[02:41:36]  Unidentified speaker: I have to pinch myself.

[02:41:37]  Peter: It's a low amplitude event in the multiverse.

[02:41:43]  Brian: You both are being too generous and too kind with your compliments. I enjoyed it. I had a blast. I'm looking forward to part two down the road. But no, it was a pleasure. And I apologize for my inadequacy at explaining things. I know there are other folks who could have given you better examples and clarity than I could. But I still very much appreciate the chance to come on and talk about this stuff.

[02:42:10]  Unidentified speaker: All right.

[02:42:11]  Unidentified speaker: Thank you.

[02:42:12]  Unidentified speaker: Thank you so much.

[02:42:13]  Brian: Absolutely.

[02:42:17]  Bruce: The Theory of Anything podcast could use your help. We have a small but loyal audience, and we'd like to get the word out about the podcast to others so others can enjoy it as well. To the best of our knowledge, we're the only podcast that covers all four strands of David Deutsch's philosophy as well as other interesting subjects. If you're enjoying this podcast, please give us a five-star rating on Apple podcasts. This can usually be done right inside your podcast player, or you can Google The Theory of Anything podcast Apple or something like that. Some players have their own rating system, and giving us a five-star rating on any rating system would be helpful. If you enjoy a particular episode, please consider tweeting about us or linking to us on Facebook or other social media to help get the word out. If you are interested in financially supporting the podcast, we have two ways to do that.

[02:43:03]  Bruce: The first is via our podcast host site, Anchor. Just go to anchor.fm slash four dash strands, F-O-U-R dash S-T-R-A-N-D-S. There's a support button available that allows you to do reoccurring donations. If you want to make a one-time donation, go to our blog, which is fourstrands.org. There is a donation button there that uses PayPal.

[02:43:28]  Unidentified speaker: Thank you.