User Tools

Site Tools


papers:dickens_flynn01_heritability_estimates_environment

Heritability estimates: IQ paradox resolved

William T. Dickens, James R. Flynn
Heritability Estimates Versus Large Environmental Effects:
The IQ Paradox Resolved
Psychological Review, 2001. Vol. 108, No. 2. 346-369

(read this paper)

Some argue that the high heritability of IQ renders purely environmental explanations for large IQ differences between groups implausible. Yet, large environmentally induced IQ gains between generations suggest an important role for environment in shaping IQ. The authors present a formal model of the process determining IQ in which people's IQs are affected by both environment and genes, but in which their environments are matched to their IQs. The authors show how such a model allows very large effects for environment, even incorporating the highest estimates of heritability. Besides resolving the paradox, the authors show that the model can account for a number of other phenomena, some of which are anomalous when viewed from the standard perspective. Abstract

This is an influential paper that challenges the interpretation of heritability measurements, concluding that high levels of heritability for IQ (and other behavior traits) are not evidence that the influence of the environment is small. The paradox results from the conflict between The Flynn Effect and the common interpretation of high heritability and low shared environment effect for IQ as evidence for a low effect of environment differences (such as wealth or school quality) on IQ differences.

See The Flynn Effect for details, but we do find plausible Flynn's argument that broad cultural changes related to modernity changed the environment in a way which has caused people to need a different kind of mental abilities, and that they have learned to do so, concurrent with this cultural change, and mutually reinforcing it.

Because this change was everywhere in these modernizing countries, if you had tested the heritability of IQ at any given time, you would still have found the same small effect of shared environment. This is because families hardly differed in the relevant characteristics of the environment. If, as in Flynn's example, the increasing complexity of entertainment, such as the mass media of movies and radio was part of this change, almost everyone in the country had access to these forms of media. Everyone shared it, so it wasn't a difference in shared environment.

In this view, the Flynn effect certainly does serve as a caution about using heritability and shared effect measurements to rule out the possibility that massive pervasive socio-economic changes could cause significant changes in human behavior and in the capacities that people develop. This is very similar to the common point that height is highly heritable (> 90%), and height is also strongly affected by nutrition, such that in modern times people may be a couple inches taller than their ancestors only a few generations ago. Once again, almost everyone experienced these improvements in nutrition, and also this height increase.

So the paradox of Heritability vs. the Flynn effect can be resolved, without requiring any radical reconceptualization of heritability. However the paper does go on to offer several radical reconceptualizations, of varying merit.

Behavior Changes Environment

The most substantial point is that the environment responds to individual behavior, encouraging some behaviors, and discouraging others, and that people negotiate their way through this social environment, making choices about which part of the environment they are going to engage with, and in turn be shaped by. It is clear that what mental activities you engage in have a strong effect on your abilities (practice). It is plausible that very consistent mental activities such as school, hobbies, and reading for entertainment may change your mind in ways that would increase IQ test results. If, in response to your behavior, the people around you encourage some behaviors and discourage others, and this social feedback changes your behavior in ways that change your mental capabilities, then it makes sense to say that these social interactions have caused that mental change. The authors refer to this feedback between your mind and the social environment as reciprocal causation, meaning that each causes the other. This circular causation is characteristic of a Feedback system.

reciprocal causation produces a multiplier effect that inflates both genetic and environmental advantages by a process in which higher IQ leads one into better environments causing still higher IQ, and so on. Page 346

This point is somewhat weak with respect to IQ itself, since (despite the claims of this article) the evidence that common variations in behavior might have big effects on your adult IQ is pretty weak. The point is weak for IQ, but it does bear strongly on the far broader observation that identical twins are noticeably similar in many many behaviors, and also in important life outcome measures like education level achieved, wealth, and so on. Flynn's remarks about the heritability of Basketball Playing ability (BP) are clearly relevant to understanding this similarity between twins:

If someone's genes predispose them to be good at basketball, then somewhat better play alone is likely to lead him or her into an environment supportive of better performance. The match is not perfect, of course, but the tendency is pronounced. To the extent that environmental quality is matched with genetic endowment, there will be a tendency for identical twins to resemble one another for BP because their shared genes make them likely to have environments that are very similar for the production of BP—whether they are raised together or separated at birth. Despite being raised together in the same home, adopted and unadopted siblings may experience very different quality BP environments, ones matched to their differences in genetic endowment. Page 349

This surely does happen, and it's easy to believe that it is a powerful effect. Of course in another culture lacking basketball, the child's ability would cause them to be steered in a different direction, and it could be that this other direction would not be as prestigious as basketball star.

Heritability estimates will credit genes with creating BP differences that would not exist were genes and environment truly uncorrelated—for example, if everyone irrespective of height played only pickup basketball once a month. Here is something that acts as a mask. Thanks to the matching of environment and genetic endowment, the standard causal model based on heritability estimates can hide the potency of environmental factors.

Here's where it starts to get dicey. It's clear that they're saying that the heritability estimate is in some sense wrong, or at least that it's wrong to understand heritability as meaning “strength of genetic influence”. The crucial weakness developing in this argument lies in the question: “What would an environment have to be like to cause there to be no correlation between a person's genetic endowment and the environment that they experience?”

The environment is responding to the person's behavior, physical shape, appearance, whatever, not directly to their genetic endowment. For the moment, let's consider behavior. What would an environment be like that would not respond to individual behavior? How to we get to that environment where everyone plays pickup basketball exactly once a month?

Environmental Multiplier

They go on to develop the idea of the environmental multiplier. The story is that there is some genuine causal effect of genetics on IQ, or any other heritable behavior, but that most of the effect is caused by social interactions causing the person to experience a distinct environment. That is, heritability of IQ is a large effect which seems genetically caused, but this is an illusion. The true genetic effect, the “direct effect of genes on IQ” is the apparent effect with the social feedback factored out.

Our referring to the multiplier as environment responsiveness and social feedback rather than the bland statistical “gene environment correlation” and “reciprocal causation” does foreshadow where we are going with this. The statistical terms do refer to problematic situations when you are using statistical correlation to infer causality, but it's important to keep in mind the human social meaning of these effects.

Several readers of earlier drafts of this paper have noted that the “direct effect of genes on IQ” may very well be zero, in that there are certainly no genes for IQ scores. This is an overly literal interpretation of what we mean by a direct effect of genes on IQ. We understand a direct effect to be one in which genes predispose one to have some physical characteristic that is useful for performance on IQ tests. Page 353

So the direct effect of genes on IQ is the effect that you would see in the absence of an environment? But genes don't do anything without an environment–no IQ tests without an environment. Perhaps the direct effect of genes on IQ is the effect that you would see in an unresponsive environment. An environment where everyone plays basketball once a month (whether they want to or not.)

We can certainly try to imagine what this unresponsive environment would look like, as a philosophical thought experiment. But the article seems to be saying that this hypothetical environment is the one that we would have to do heritability studies in if we want to measure the true genetic contribution to IQ (or any other trait.) That would never get by Human subject review. Instead, their method is to take the high apparent genetic effect according to heritability measures, and to divide things out to correct the high estimate. But what exactly is the criterion here for what we are going to remove? So far we have the environmental multiplier correction, but why stop there?

Black Boxes

One way to understand heritability is as treating all of life as a Black box, from the moment of conception to the time that we look at whatever traits or outcomes we're measuring. Heritability says that, without knowing anything about what's going on inside the box, we can see that genetic differences are affecting the outcomes that we measure. We can see this because genetically more similar people have predictably more similar outcomes. What it means to be a cause is much less clear-cut than people generally (or this article in particular) suppose, genetics is well situated to be an important cause because it is there from the very beginning, at fertilization of the egg (see Genetic Causes).

What this paper is doing is taking a peek inside the box. We all live inside this box, and we all have personal observations of what life is like in our culture. We also have cultural theories about about what causes individual differences in behavior and social success. Although the paper loosely follows a common model of social science by proposing a mathematical model of what is going on inside the box, they admit that the model has too many unknown parameters to actually fit it to the available data. So they rely on our personal intuitions and social understandings of how our culture works.

In particular, they've identified an important thing about how our culture (and every human culture) works: by far the most important aspects of the environment are social. Except in rare circumstances, we do not live without relying on and constantly interacting with other people. We get social feedback, and this shapes our behavior, influencing what abilities we develop and influencing our overall life course.

They've identified a mechanism inside the box that causally connects the input of the box (the genome) to the output of the box (an individual with particular abilities and social status.) But surely there are other things going on inside the box too. Shouldn't they be subtracted out? They are modelling the output of the box as two things: the “direct effect on IQ” multiplied by the social multiplier.

But as their footnote makes clear, there is no actual direct effect of genes on IQ in the real world. They've created a model that has a variable they call that, but it's really a residue or error term. It's everything about the effect of genetics on the output that isn't explained by the social multiplier theory. Clearly whatever's rolled up in that term has a lot to do with biology, so maybe we should look inside the body. Though we don't understand very well how it is that genetics participates in the maintenance of the adult body or the developmental processes up to then, we do know that gene regulatory networks exhibit feedback and sometimes strong nonlinearity. So the biology of the human body magnifies the effect of whatever minor variations there is in our DNA. There is no doubt that a change to just one out of the 3 billion bases in our genome can have dramatic effects on our health and well being, but it is pretty clear that the effect of a mutation also depends on interactions with the rest of our genome (the genetic background). For a simple example, see sickle-cell anemia. Only if both copies of the gene (from mother and father) have the mutation does disease develop. If a gene ever has a non-additive effect, then let's call this the somatic multiplier effect, and divide that out too.

If we continue with this logic, and if we actually developed the necessary biological understanding, then we'd eventually get to the point where we would discover that there is no “direct effect of genetics on IQ” at all. Or if we continue where their story is leading us, since the direct effect is the true effect (all else is inflation) then genes have no genuine effect on anything. What they have done is to observe a mechanism that generates life outcomes and to observe that this mechanism works by social interactions. They rely on our intuition that social events are not genetically caused, implicitly introducing the conclusion that they are trying to justify.

Causation

A lot of this depends on what it is that we mean by “cause” or by saying that something is “determined”. Unfortunately, although these are important words, and common everyday words, philosophy has shown that causation resists rigorous definition (see David Hume: Causation), and has to be regarded as a fuzzy evolved intuition about the structure of the world. We constantly try to infer causation because it gives us useful leverage for predicting and manipulating the world, but causality is not something we can directly perceive. Mostly we only perceive a coincidence: that two things happen at the same time.

Genetics is not a sufficient cause for the child or the adult. That is, the the child's genome does do not single-handedly cause there to something rather than nothing. The genes have to be in the nucleus of a suitable cell, with all that cellular machinery, inside a woman, who exists in a sufficiently supportive social and material environment. Yet in comparison to those other supportive factors (necessary causes), it is clear the genome has an important role in patterning the person who may later develop (as long as the environment remains sufficiently supportive and responsive.) See Genetic Causes.

Innateness and Social Construction

We can clarify the position of the Dickens/Flynn model in the existing narrative of nature/nurture debate and social criticism by renaming the term “direct effect of genetics on IQ” to “innate” and “social multiplier” to “social construction”. That is, they are saying that part of the apparently genetic contribution to an outcome is socially constructed, and that we should model the combined effect of the socially constructed factors taken together with the innate factors as a multiplication (a gene/environment interaction).

outcome = innate * socially_constructed

Of course a “social multiplier” theory demands a multiplication, but what are the consequences of this model choice, and do they make sense? bla bla, extreme values, yes.

The important point is that because life always takes place in a social context, an outcome can be simultaneously socially constructed and genetically determined. It has frequently been pointed out that the idea of “genetic determinism” independent of any environment is silly, but it is far less widely appreciated that claiming an outcome can be “socially constructed” (independent of any genetic effect) is equally oversimplified. This is the inexorable logic of the “it's both” interpretation of heritability.

We can better understand the position of the Dickens/Flynn model in social criticism by dividing up the socially constructed part:

outcome = innate * (racism + ...ism + positive_feedback + negative_feedback + other)

Strong social negative feedbacks that reduce outcome variation.

The social multiplier model is a rectangle model of gene/environment interaction (area = height*width). A rectangle model is a step in the direction of realism, because it acknowledges that there are interacting factors that create the outcome, but the author's own argument of the complex correlation between genetics and social interaction makes clear that the rectangle model cannot actually be correct. It is not possible to independently vary the innate and socially constructed components because each is correlated with the other.

For realism we need a more general model which admits arbitrarily complex gene/environment interactions. There is a well-developed scientific theory of these sorts of interactions, but it comes from a discipline that is well outside the usual scope of social science.

Fitness landscape

Ever since Darwin, biologists have known that genetics and environment interact in complex ways. In particular, there is no such thing as “the environment” because the environment varies from one place to another. It has also long been clear that any organism which has behavior experiences a strong coupling between behavior and important outcome measures such as health and number of surviving offspring. This is why behavior evolved in the first place: organisms who took control of their environment increased their success.

You could leave a comment if you were logged in.
papers/dickens_flynn01_heritability_estimates_environment.txt · Last modified: 2014/07/13 14:43 by ram