(I’ll bury this post later)
Scott Kaufman has an awful piece on Heritability over at Huffpost as per my expectations. To go over 6 of 8 of his points:
1. There is no nature/nurture question
Kaufmann: “Twin studies partition the variance in nature and the variance in nurture. This allows researchers to determine whether differences in genes or differences in the environment in a particular population are associated with more of the differences in observed behavior.
In reality, all biological and psychological characteristics are constructed during development, when genes interact with local environmental factors that can be influenced by the broader environment. Therefore, gene-environment interactions are understood to drive the development of all of our characteristics.”
False! (“In reality,” used here, implies a contradiction)
This statement confuses two different types of gene x environment interactions as defined by two separate behavioral genetic research programs, the biometric and the developmental programs.
It’s trivially true that genes interact with the environment to produce phenotypes and, in that sense, the developmental sense (GE developmental), that all behaviors involve gene x environment interactions. This does not mean that all population variance, the analysis of which is the focus of the Biometric program, is the product of gene x environmental interactions (GE biometric).
For example, the development of general mental ability obviously entails genetic and environmental interactions, but in no way does this imply that differences in general mental ability are the product genetic and environmental interactions.
(Biometric GE refers to situations where: a) a good environment for one geneotype makes for a bad environment for another or b) the same environmental factors have unequal phenotypic effects for different geneotypes. In both, the environment interacts with genotypic differences and produces phenotypic differences.)
2. Heritability does not say anything about individual differences.
Kaufamnn: “Because heritability is a population statistic, it has nothing to say about the individual. It makes no sense to ask whether a particular individual’s intelligence has been more determined by nature or by nurture.“
This is obviously incorrect. As Tal 2009 has shown, heritability estimates allow one to make a probabilistic statement about the factors underlying an individual’s deviation from the population mean. The probability that genetic factors contribute more to an individual’s deviation from the mean is given by the following formula:
Example: For a particular individual with an IQ of 120, whose population has a heritability of IQ of 0.8, there is an 84% chance that the individual’s deviation from the mean is due more to genetic than environmental factors.
3. Heritability doesn’t constrain malleability, Turkheimer’s study and the Flynn effect shows.
Kaufamnn: “ Turkheimer’s study should also be a reminder that just because something is heritable doesn’t mean it’s immutable. The Flynn effect — the dramatic rise in IQ witnessed in the 20th century — is a good example of that. The Flynn effect should be a reminder of just how much the environment matters, even after completely controlling for genes (by looking at IQ changes across generations)….
The heritability of a trait can vary from 0.00 to 1.00, depending on the environments from which research participants are sampled. Because we know that genes play some role in the development of any trait, the precise heritability estimate doesn’t matter in a practical sense.
Misleading, where not false…
This statement is doubly flawed. Turkheimer’s study dealt with the mutability of heritability estimates and not with the malleability of highly heritable traits. The Flynn effect provides no evidence that highly heritable traits are mutable. The Flynn effect describes differences in IQ scores across cohorts. There is no evidence that those differences are of the same nature as individual differences within cohorts (see: Jensen, 2011). Therefore, those differences are not comparable, as yet. The heritability estimates of IQ, of course, pertain only to differences within cohorts. Since the differences across cohorts do not, at present, say anything about the differences within cohorts, and since the heritability estimates of IQ only pertain to differences within cohorts, the Flynn effect says nothing, as yet, about differences, heritability, or mutability.
As for heritability and mutability, if the portion of the variance due to the environment is small, then the source of environmental variance are low given the conditions that prevailed in the population in which h^2 was estimated. As such, existing environmental factors are not strong sources of variance.
To illustrate: Imagine two traits, A and B, for which the H^2 is 0.1 and 0.9, respectively. For the first trait, prevalent environmental conditions can have a large impact on variance and for the latter they can have little. Imagine that we wished to raise a random sub-population’s mean score 1 SD in both traits. To do so, 1.05 SD (1/sqrt .9) versus 3.16 SD (1/sqrt .1) of prevalent environmental effect, respectively, would be needed. That’s a massive difference!
Of course, there might be yet unknown factors that are not prevalent which can greatly increase trait B, but they aren’t prevalent. And, as they are unknown, they are not readily usable. How does that not have practical significance?
4. True heritability can never be know; heritability can not be generalized.
Kaufmann: The heritability of a trait can vary from 0.00 to 1.00, depending on the environments from which research participants are sampled. Because we know that genes play some role in the development of any trait, the precise heritability estimate doesn’t matter in a practical sense.
The problem is that our understanding of the factors that contribute to the development of human traits in general — and to IQ in particular — is currently so deficient that we typically do not know if the environmental factors important in the development of a particular trait are stable across testing situations, vary somewhat across those situations, or vary wildly across those situations.
There are two questions here which need to be disentangled: “Can valid population heritability estimates be made,” and “Can population heritability estimates be generalized across rather different environmental conditions.” The first question asks if heritability estimates can be generalized from the sample to the populations. The second asks if these estimates can be generalized to environments which have substantially different conditions than those that prevailed in the population in which h^2 was estimated.
The first can be answered in the affirmative. Multiples lines of evidence support the validity of heritability estimates. (Twins reared apart; mid parent-offspring correlations; classical twin studies; adoption studies).
The strategy for undermining heritability values a la Kaufman is to imagine a possible world in which they don’t apply and then sow doubt about the validity or practical significance of the estimates. As noted above though, heritability values have practical significance given the conditions that prevailed in the population in which h^2 was estimated. The more conditions tested, the more generalizable the results.
5. Heritability is not the same as heredity
Heredity = breeding value = the additive genetic component of heritability.
6. Heritability studies do no reveal the cause. (GE correlations)
Kaufman: It’s important to keep in mind that the route from genotype (genetic makeup) to phenotype (observed behavior) is hardly ever clear-cut. It’s possible for many traits to involve gene-environment correlations. The idea here is that environments set off an appetite in the genes that nudges individuals to engage in certain experiences, and the environment then responds in a reciprocal fashion that reinforces an individual’s nature. The genes and environment eventually become correlated.
In general, BUT…
Since rGE explanations of the H^2 of IQ (g) are in vogue, it’s worthwhile going over the multiple lines of evidence which disconfirm a strong case of rGE for general intelligence. [1-6 from Sesardic (2005)]
rGE explanations run accordingly:
G (genetic difference) –> C (characteristic not related to trait) –> E (environmental influence) –> P (trait differences)
A GE model…
1. a. Implies C will correlate with P
Scarr and Carter-Salzmann (1982): Compared MZ twins to DZ twins incorrectly classified (because they were phenotypically very similar) as MZ twins. Contrary to GE predictions, incorrectly classified DZ twins were less phenotypically similar than MZ twins
2. Implies P differences are systematically related to E differences
Loehlin and Nichols (1976): Collected information about twins’ different treatment and found that average correlations between a composite and these factors and psychological traits was close to zero.
3. Predicts a correlation between a given environmental measure of one twin and another’s phenotype, since G –> E–> P.
Loehlin 1992; Neale & Cardon (1992): compared twins reared apart. Found no cross correlation between E and P. Plomin 2001: compared parent-Child pairs. Found little evidence of a active or reactive G-E covariance.
4. Would predicted that similarity would decrease with age of separation, if shared MZ twin environment causes the phenotypic similarity:
Lykken (1995): found no effect.
5. Is disconfirmed by multivariate genetic analysis: Look simultaneously at the impact of genetic differences on two variables (environment and phenotype) in an attempt to estimate whether the two effects overlap (and to what extent).
Klump et al. (2000); Kendler et al., (2000). No strong G-E correlations.
6. Is limited. Many GE interactions can be ruled out a priori as the between family environmentality is close to zero which means that between family differences are uncorrelated with Intelligence.
A GE model for IQ (g) ….
7. Implies E will correlate with endophenotyes. (see: van Leeuwen et al., (2009) A genetic analysis of brain volumes and IQ in children for a discussion of this)
Posthuma et al. (2003), De Moor et al. (2008), van Leeuwen et al (2009), and Betjeman et al. (2009) found this not to be the case.