What does knowing the heritability of a trait tell me in practice?
post by Kaj_Sotala · 2021-07-26T16:29:52.552Z · LW · GW · 1 commentThis is a question post.
Contents
Answers 11 Jose Blanca 10 Sleeps Furiously 7 dynomight 2 Roko 2 frontier64 1 Daniel V 0 Jay None 1 comment
The concept of heritability gets misunderstood a lot, so there are several articles discussing what it doesn't mean. But reading through all of them leaves me confused about what it does mean in practical terms, outside the technical definition.
For example, as I myself wrote in an old comment [LW(p) · GW(p)] about common misunderstandings:
Caution: heritability, as in the statistical concept, is defined in a way that has some rather counter-intuitive implications. One might think that if happiness is 50% heritable, then happiness must be 50% "hardwired". This is incorrect, and in fact the concept of heritability is theoretically incapable of making such a claim.
The definition of heritability is straightforward enough: the amount of genetic variance in a trait, divided by the overall variance in the trait. Now, nearly all humans are born with two feet, so you might expect the trait of "having two feet" to have 100% heritability. In fact, it has close to 0% heritability! This is because the vast majority of people who have lost their feet have done so because of accidents or other environmental factors, not due to a gene for one-footedness. So nearly all of the variance in the amount of feet in humans is caused by environmental factors, making the heritability zero.
Another example is that if we have a trait that is strongly affected by the environment, but we manage to make the environment more uniform, then the heritability of the trait goes up. For instance, both childhood nutrition and genetics have a strong effect on a person's height. In today's society, we have relatively good social security nets helping give most kids at least a basic level of nutrition, a basic level which may not have been available for everyone in the past. So in the past there was more environmental variance involved in determining a person's height. Therefore the trait "height" may have been less hereditary in the past than now.
The heritability of some trait is always defined in relation to some specific population in some specific environment. There's no such thing as an "overall" heritability, valid in any environment. The heritability of a trait does not tell us whether that trait can be affected by outside interventions.
Some articles that go deeper into the details and math of this include "Heritability is a ratio, not a measure of determinism" (dynomight.net) and "Heritability in the genomics era - concepts and misconceptions" (Nature Reviews Genetics).
However, all of these examples of what heritability doesn't mean have left me very confused about what it does mean. I know that if a trait is 80% heritable, I cannot infer that it is "80% genetically determined", but what can I infer? That 80% of the observed variance in that trait is genetic, yes, but what's the practical thing of interest that having this information allow me to predict, that I couldn't predict before? In particular, what does knowing the heritability of traits such as IQ, subjective well-being, or Big5 scores tell me?
Looking at the Wikipedia article for heritability, I see very little that would help answer this question; the closest that I can find is the "controversies" section, which says that there are people who think the concept shouldn't be used at all:
Heritability estimates' prominent critics, such as Steven Rose,[27] Jay Joseph,[28] and Richard Bentall, focus largely on heritability estimates in behavioral sciences and social sciences. Bentall has claimed that such heritability scores are typically calculated counterintuitively to derive numerically high scores, that heritability is misinterpreted as genetic determination, and that this alleged bias distracts from other factors that researches have found more causally important, such as childhood abuse causing later psychosis.[29][30] Heritability estimates are also inherently limited because they do not convey any information regarding whether genes or environment play a larger role in the development of the trait under study. For this reason, David Moore and David Shenk describe the term "heritability" in the context of behavior genetics as "...one of the most misleading in the history of science" and argue that it has no value except in very rare cases.[31] When studying complex human traits, it is impossible to use heritability analysis to determine the relative contributions of genes and environment, as such traits result from multiple causes interacting.[32] In particular, Feldman and Lewontin emphasize that heritability is itself a function of environmental variation.[33] However, some researchers argue that it is possible to disentangle the two.[34]
The controversy over heritability estimates is largely via their basis in twin studies. The scarce success of molecular-genetic studies to corroborate such population-genetic studies' conclusions is the missing heritability problem.[35] Eric Turkheimer has argued that newer molecular methods have vindicated the conventional interpretation of twin studies,[35] although it remains mostly unclear how to explain the relations between genes and behaviors.[36] According to Turkheimer, both genes and environment are heritable, genetic contribution varies by environment, and a focus on heritability distracts from other important factors.[37] Overall, however, heritability is a concept widely applicable.[9]
Out of those references, the one that sounded the most useful in telling me what heritability might actually mean was the one associated with the sentence "Overall, however, heritability is a concept widely applicable". This is the previously mentioned "Heritability in the genomics era - concepts and misconceptions" (Nature Reviews Neuroscience), which includes a section on "applications":
The parameter of heritability is so enduring and useful because it allows the meaningful comparison of traits within and across populations, it enables predictions about the response to both artificial and natural selection, it determines the efficiency of gene-mapping studies and it is a key parameter in determining the efficiency of prediction of the genetic risk of disease.
From reading this section, I gather that:
- If I wanted to breed plants or animals that were high on a particular trait, having the heritability estimate for that trait could be useful
- The heritability of a trait can be used to help infer how much statistical power gene-mapping studies targeting that trait need
- If I was trying to predict the genetic risk of something like schizophrenia, then... I don't quite understand this part, but apparently having the heritability estimate would help me know how reliable my prediction was going to be
Usefulness for breeding programs is hopefully an irrelevant consideration when we're talking about humans, which leaves me with the two others; and those also seem to suggest that knowing the heritability of a trait isn't useful on its own, and will only be something that helps me do or evaluate a gene-mapping or genetic risk prediction study better.
This seems to suggest that knowing the heritability of a trait such as IQ, subjective well-being or a Big5 score tells me essentially nothing by itself; is this correct?
(cross-posted to the Psychology & Neuroscience Stack Exchange)
Answers
Heritability is an idea originated in breeding, and it is interesting mainly for breeders. It is defined as the part of the variation for a trait in a population that is due to genetic variation (in contrast to enviromental variation).
The heritability is useful for a breeder because if the variation observed in a population is mainly due to enviromental factors the trait won't be improved by selecting the best individuals in the population.
However, the concept is usually misunderstood. First, it always depends on the population studied. One trait could have high heritability in one population, and low in another because different populations can have different genetic variation. Moreover, for a particular population it also depends on the environment because is the fraction of variation not due to the enviroment, and that, of course, depends on how variable is the environment.
Moreover, calculating the heritability, in practice, is very difficult even in controlled environemnts and populations, and when it is calculated most of the time is restricted to heritability in the narrow sense, taking into account only aditive effects. (This basically means ignoring all medelian gene interactions).
What the heritability does not mean is how much a trait depends on the genes. That idea does not even makes sense because any trait will depend on many genes even when it has no heritability. For instance, the number of fingers in humans has a very low heritability, because most of the variation is due to the enviroment (e.g. accidents), but the number of fingers clearly depends of many genes.
It can tell you something about existing interventions in a variable. In the US, for instance, we spend years of effort and upwards of hundreds of thousands of dollars on primary/secondary school education, and we know that we do a very poor job of making sure that different students have similar education experiences.
So, if SAT scores have low heritability in the US currently, then we would expect that we could figure out which education experiences tend to lead to higher SAT scores and try to do a better job of making sure everyone gets those kinds of experiences. If, on the other hand, heritability is high, then throwing more effort/money at how we do education currently should not be expected to improve SAT scores, and we either need to rethink how we do education, or rethink whether SAT scores measure what we want.
↑ comment by Insub · 2021-07-27T05:14:58.022Z · LW(p) · GW(p)
If, on the other hand, heritability is high, then throwing more effort/money at how we do education currently should not be expected to improve SAT scores
I agree with spkoc that this conclusion doesn't necessarily follow from high heritability. I think it would follow from high and stable heritability across multiple attempted interventions.
An exaggerated story for the point I'm about to make: imagine you've never tried to improve SAT scores, and you measure the heritability. You find that, in this particular environment, genetic variance explains 100% of SAT scores. You can predict someone's SAT scores perfectly just by looking at their genome. You decide to take the half of the population with the highest predicted scores, and keep the SAT a secret from them until the day they take the test. And for the lower half, you give them dedicated tutors to help them prepare. Given the 100% heritability, you expect scores to stay exactly the same. But wait! What no one told you was that before your intervention, the learning environment had been magically uniform for every student. There had been no environmental variance at all, and so the only thing left to explain test scores was genetics. What you didn't realize is that your heritability estimate gave you no information at all about how environmental changes would affect scores - because there was no environmental variance at all!
A single heritability measurement only tells you, roughly, the ratio of "[existing environmental variance] times [sensitivity to environmental variance]" to "[existing genetic variance] times [sensitivity to genetic variance]". But it doesn't do anything to disentangle the sensitivities-to-variances from the actual variances. What if there's practically zero variance in the environment, but a high sensitivity of the trait you're looking at to environmental variance? You'd find heritability is very high, but changes to the environment will cause large decreases to heritability. Same thing with genes: what if your trait is 100% determined by genes, but it just so happens that everyone has the exact same genes? You'd find that genetic variance explains zero percent of your trait, but if you then tried some genetic engineering, you'd find heritability shoot upward.
In order to disentangle the "sensitivity of X to environmental variance" from "the level of environmental variance", you'd have to run multiple interventions over time, and measure the heritability of X after each one (or be confident that your existing environment has lots of variance).
Replies from: Jay↑ comment by Jay · 2021-07-29T10:54:09.201Z · LW(p) · GW(p)
True, but "high and stable heritability" across hundreds (perhaps thousands) of attempted interventions is a pretty good description of the real-world results of education research and practice. See Freddie DeBoer's "Education Doesn't Work" for a brief treatment or Kathryn Paige Harden's The Genetic Lottery for a book-length version.
↑ comment by Jay · 2021-07-30T11:10:03.673Z · LW(p) · GW(p)
The evidence indicates that throwing more effort/money at how we do education does not improve IQ scores (for which SAT scores are a thinly-veiled proxy, except that every decade or so they make cosmetic changes to the SAT methodology) or student outcomes. Attempts to rethink education have failed. And IQ is generally useful enough that it is strongly correlated with outcomes we want.
If you're used to the tech sector with rapid change every decade, moving into the human services sector is going to be a very depressing experience. The low-hanging fruit was picked centuries ago and there hasn't been any real progress in the last 50 years.
Replies from: Kaj_Sotala↑ comment by Kaj_Sotala · 2021-07-30T17:22:37.920Z · LW(p) · GW(p)
Attempts to rethink education have failed.
Or at least the particular set of reforms discussed in that article has failed? Even within the context of the US, there do seem to be occasional educational interventions that work, e.g.:
Replies from: JayIn a state once notorious for its low reading scores, the Mississippi state legislature passed new literacy standards in 2013. Since then Mississippi has seen remarkable gains. Its fourth graders have moved from 49th (out of 50 states) to 29th on the National Assessment of Educational Progress, a nationwide exam. In 2019 it was the only state to improve its scores. For the first time since measurement began, Mississippi’s pupils are now average readers, a remarkable achievement in such a poor state.
Ms Burk attributes Mississippi’s success to implementing reading methods supported by a body of research known as the science of reading. In 1997 Congress requested the National Institute of Child Health and Human Development and the Department of Education to convene a National Reading Panel to end the “reading wars” and synthesise the evidence. The panel found that phonics, along with explicit instruction in phonemic awareness, fluency and comprehension, worked best.
↑ comment by Jay · 2021-07-30T20:50:19.851Z · LW(p) · GW(p)
There have been dozens of stories like that; George W Bush got elected on the strength of his education "reforms". Long-term experience justifies a strong belief (confidence over 90%) that the results will ultimately turn out to be due to a combination of selection bias (cherry-picking) and test fraud. The links are just examples; I've been offhandedly following education research and reform for decades. There's a lot more evidence where that came from, and it tells a very consistent story.
Education simply isn't a green field - the space of potential solutions has been fairly thoroughly explored (for the set of non-medical solutions that are broadly consistent with Western values and American cultural practices). If you are active in this space without learning that history, you are almost certain to repeat it.
↑ comment by spkoc · 2021-07-26T19:24:31.897Z · LW(p) · GW(p)
But isn't this exactly the mainstream intuition that the OP dissolves? My understanding:
a) Heritability measures don't seem to make sense for really complex traits like intelligence.
b) Heritability measures are not stable outside the environmental conditions in which they were measured.
For instance, some people have sickle cell anemia, which helps them better survive malaria(but otherwise is slightly harmful). If you measure heritability of infant mortality in environment with malaria and then in environment without malaria you get opposite effects. You get a lighter case of malaria with it, so sickle cell probably positively correlates with intelligence if there's malaria in the environment.
There's also weird epigenetic multigenerational effects: https://www.nature.com/articles/s41467-018-07617-9 (quote is about a reference in this study)
The findings from the Överkalix cohorts imply that grandparental access to food during their slow growth period can modify diabetes and all-cause mortality in grandchildren. Cardiovascular mortality on the other hand was associated with parental, but not grandparental, nutritional experience. The authors interpret their results11 as “proof-of-principle that a sex-specific male-line transgenerational effect exists in humans”, which they consider likely to be epigenetic rather than genetic, cultural or social. A summary of the Överkalix findings is available in19. Their findings have been discussed in renowned peer-reviewed journals20,21,22,23,24 and are cited over 2,000 times (October 2018).
I have this vague memory about reading an article claiming that grandparents' access to food influenced their grandchildren's height, moreso than their immediate childrens', but I can't seem to find it. Perhaps nonsense.
Replies from: Sleeps Furiously, deepthoughtlife↑ comment by Sleeps Furiously · 2021-07-26T21:19:21.881Z · LW(p) · GW(p)
It is not that intuition. The mainstream intuition is that a high heritability means that variations in the thing that you're measuring (height, SAT scores, infant mortality) is primarily affected by genetics and cannot be influenced by the environment. A better framing would say that high heritability means that variations in the thing that you're measuring are not well explained by existing variation in the environment. To reframe, then, what I was going for above, a high heritibility is an upper bound on how much you can expect to improve a specific measure by playing around with existing policy levers, versus having to think outside the box and develop new policy levers.
So, again, pretend that SAT scores have a high measured variability in the US. (I think it's actually closer to 50 than either extreme, but I don't have off-hand an actual example of a measure that we might want to increase/reduce the variance of as a society that does have a high heritability.) We might look at nutrition, SES and education and ask whether we can manipulate these to significantly reduce the variance in height and/or SAT scores. Under the high heritability assumption, the answer is that this is very unlikely for both stats.
Again, this is without saying that high heritability means the existing policy levers are not having an effect right now. If we, as a country, stoppd feeding kids at age 5 and left them to fend for themselves, I would expect the survivors to be both shorter and have lower SAT scores than preceding generations. If we stopped sending kids to school, I would expect average SAT scores to drop and the variance to increase significantly. And nothing about heritability argues against any of that.
It also doesn't mean that there are no policy levers that could have an effect. I am not biologist enough to know if the application of HGH to children on a large scale would increase average height, but what I can say is that heritability has nothing to say on the question, since it is currently an environmental variable with almost no variance itself.
Replies from: Sleeps Furiously↑ comment by Sleeps Furiously · 2021-07-26T21:28:50.516Z · LW(p) · GW(p)
Oops. Accidentally dropped a paragraph about whether you could increase height/SAT scores across the board with the existing policy levers. In that case, a high heritability doesn't directly say that that's unlikely, but you would at least have to expect to increase your policy setting to the point where most people are getting more of the nutrition or education or whatever than is currently a few standard deviations above average, since otherwise if less than that much of your policy was having an effect, you would expect to already see it in the heritability measurement. So heritability doesn't say that feeding kids 10,000 calories a day or sending them to school for 12 hours a day won't have effects on height/SAT scores, but as a general rule you'd be looking at such extreme interventions that it's probably not a good direction to go.
Replies from: spkoc, Jay↑ comment by spkoc · 2021-07-27T08:24:53.711Z · LW(p) · GW(p)
I'm still so confused(through no fault of your own, I think you're right, it just doesn't fit in my head). Let me try to walk through my thought process.
I assume heritability of SAT score is probably different if you sample across USA, or just upper-mid class suburbs or just South Side Chicago, or just rural Eastern Europe, or just Malawi during a famine. Right? Given that environments are pretty radically different.
What heritability score are we using to determine if policy interventions matter or not? Is the first step to make sure that the region we want to improve has an environment that mimics 'successful' regions? Heritability would be very high in a homogenous environment(since that's the only variation), but it goes down as more varied environments are added to the sample. Heritability is very high if we just look at rich area USA schools, lower if you sample all USA and even lower if we sample the whole world?
Also how is this linked to amplitude of effect? Super high heritability of SAT/IQ in say homogenous Denmark, but presumably the actual variation in scores is lower than in a global sample. Is there a way to say genetics account for +/- 5 points of IQ? So if you're measuring IQs of 95-105 in your area that's probably all genetic effects and policy interventions can't do much?
Edit. I realize now that this is mostly Insub's point below, but less coherent.
↑ comment by Jay · 2021-07-30T00:16:23.923Z · LW(p) · GW(p)
Pretty much. If an intervention is well outside of the set of experiences of your population, there's probably a reason for that. Perhaps it's just too new, but it's likely that it's inconsistent with the way the culture usually functions (its values as actually implemented) and/or has fairly obvious side effects.
↑ comment by deepthoughtlife · 2021-07-26T22:07:24.950Z · LW(p) · GW(p)
The commenter you are replying to is right. Heritability shows how much of the current variation is genetic in origin, which shows whether the variability of nurture matters within that particular culture. If you want to successfully intervene, it needs to be either in something not very heritable, or with an intervention that is not a significant part of the current landscape of the society. (The fact that outside of context interventions may exist means heritability isn't a measure of how genetic it is, but that wasn't Sleeps Furiously's point.)
Edit: Note, Sleeps Furiously had not replied to spkoc when I saw this and I wrote my response without refreshing.
My view is that people should basically talk about heritability less and interventions more. In most practical circumstances, what we're interested in is how much potential we have to change a trait. For example, you might want to reduce youth obesity. If that's your goal, I don't think heritability helps you much. High heritability doesn't mean that there aren't any interventions that can change obesity-- it just means that the current environments that people are already exposed to don't create much variance. Similarly, low heritability means the environment produces a lot of variance, but it doesn't tell you anything specific you can actually do!
If you goal is to find interventions, all heritability gives you is some kind of vague clue as to how promising it might be to look at natural environmental variation to try to find interventions.
↑ comment by dynomight · 2021-07-26T21:24:57.049Z · LW(p) · GW(p)
On the other hand, there is some non-applied scientific value in heritability. For example, though religiosity is heritable, the specific religion people join appears to be almost totally un-heritable. I think it's OK to read this in the straightforward way, i.e. as "genes don't predispose us to be Christian / Muslim / Shinto / whatever". I don't have any particular application for that fact, but it's certainly interesting.
Similarly, schizophrenia has sky-high heritability (like 80%) meaning that current environments don't have a huge impact on where schizophrenia appears. That's also interesting even if not immediately useful.
what can I infer?
If you are missing a finger from an accident and you want children with the normal 5 digits, you can use heritability to work out that finding a wife with 6 fingers isn't going to get you back to normal.
Another one is if you are gay and you want gay children, the low heritability of sexual orientation indicates that finding a gay gamete donor won't help much.
On the other hand if you are short and want taller children, you might put a lot of effort into finding a wife who is taller than average (heritability of height tells you that this will work).
My understanding is that heritability is a measure of predictive ability. Meaning if a trait is 80% heritable and you want to guess whether or not Bob has that trait then you'll be 80% more accurate if you know whether or not Bob's parents have the trait than if you didn't have that information. Likewise, for a very low heritability trait like having 2 legs, knowing whether or not Bob's parents have two legs doesn't improve your guess much if it improves it at all.
As you mentioned, environmental factors can at times subsume the genetic factors (e.g. heritability of height can be subsumed with very low nutrition). So if environmental factors for the dataset that you're trying to predict from are significantly different from the factors which you used to determine heritability, then the heritability estimate may not be as accurate and it should be reassessed for the different factors.
You can still make very concrete predictions about traits based on heritability even though very different environmental circumstances could reduce how heritable the trait is. Heritability of IQ has been determined in an environment exceedingly similar to that faced by kids in the modern schooling system. Let's say IQ has shown to be 80% heritable in circumstances not dissimilar to the US school system (as far as I am aware this is the current state of the art). Now if you want to predict the IQ of 20,000 parents of 10,000 US schoolchildren you'll do 80% better if you know the kids IQ than if you were just guessing randomly. Similarly, if you know Bob is smart you should update your prior estimate that Bob's parents are also smart significantly in favor of their intelligence.
↑ comment by dynomight · 2021-07-28T18:24:16.235Z · LW(p) · GW(p)
if a trait is 80% heritable and you want to guess whether or not Bob has that trait then you'll be 80% more accurate if you know whether or not Bob's parents have the trait than if you didn't have that information.
I think this is more or less correct for narrow-sense heritability (most commonly used when breeding animals) but not quite right for broad-sense heritability (most commonly used with humans). If you're talking about broad-sense heritability, the problem is that you'd need to know not just if the parents have the trait, but also which genes Bob got or not from each parent, as well as the effect of dominant genes, epistatic interactions, etc.
Assuming you're talking about broad-sense heritability, I think a better way of looking at it would be to say that you'll be 80% more accurate if Bob has an identical twin raised by a random family and you know if that twin had the trait. This isn't quite right either, but I think it's valid if you assume that phenotypic traits are the sum of genetic effects and environmental effects and also that genetic effects are independent of environmental effects.
Of course, few people have identical twins raised by random families, and most phenotypes probably aren't additive in genetic and environmental effects, and those effects probably aren't independent! Which... is a lot of caveats if you want to know practical applications of heritability numbers.
Heritability is an explanation. It's R-squared. As useful as "percent of variance explained" can be in some situations, and as useless in others.
↑ comment by cubefox · 2021-07-28T15:36:26.048Z · LW(p) · GW(p)
I'm skeptical of this "percentage of variance explained" expression. In this setting, the only candate for explanation is causal explanation. But as far as I understand one cannot infer causation from statistical data (and heritability seems to be a purely statistical measure) without any additional counterfactual information.
The simplest and most useful answer is that heritability tells you the amount of variation that environmental factors don't control*. Traits with very high heritability** are generally going to be worse targets for intervention than traits with low heritability.
*In the range of environments over which the data was collected. The heritability of a trait as measured in Somalia or North Korea may be much lower that as measured in America. You can interpret this as meaning that there is much more hope for useful intervention in Somalia or North Korea, although the practical difficulties may be considerable.
**Some relevant traits are nearly 100% heritable, unfortunately. This includes executive function, which governs working memory and impulse control. Any non-biochemical intervention aimed at improving these traits is unlikely to succeed.
1 comment
Comments sorted by top scores.
comment by Aaro Salosensaari (aa-m-sa) · 2021-07-26T17:20:51.185Z · LW(p) · GW(p)
My take is that the scientific concept of "heritability" has some problems in its construction: the exact definition (Var(genotype)/Var(phenotype)), while useful in some regard, does not match the intuition of the word.
Maybe the quantity should be called "relative heritability", "heritability relative to population" or "proportion of population variance explained", like many other quantities that similarly have form A/B where both A and B are (population) parameters or their estimates.
Addendum 1.
"Heritable variance"? See also Feldman, Lewontin 1975 https://scholar.google.com/scholar?cluster=10462607332604262282