Pessimism of the Intellect, Optimism of the Will     Archive   Favorite posts   Twitter: @steve_hsu

Tuesday, November 24, 2009

Random microworlds: the mystery of nonshared environment

One of the most mysterious aspects of the nature-nurture question is the difficulty in characterizing the nurture component.

Turkheimer and Waldron: When genetic similarity is controlled, siblings often appear no more alike than individuals selected at random from the population. ... it has become widely accepted that the source of this dissimilarity is a variance component called nonshared environment.

... In what may have been the most influential article ever written in the field of developmental behavior genetics, Plomin and Daniels (1987) reviewed evidence that a substantial portion of the variability in behavioral outcomes could not be explained by the additive effects of genotype or the environmental influences of families. They suggested that this residual term, which they called the nonshared environment, had been neglected by environmentally oriented researchers who assumed that the most important mechanisms of environmental action involved familial variables, like socioeconomic status [SES] and parenting styles, that are shared by siblings raised in the same home and serve to make siblings more similar to each other. Indeed, Plomin and Daniels argued, once genetic relatedness has been taken into account, siblings seem to be hardly more similar than children chosen at random from the population.

In other words, despite a lifetime of proximity, your adopted child may bear no more similarity to you (in terms of, e.g., intelligence) than someone selected at random from the general population. The shared family environment that your children (biological or adopted) experience has little or no measurable effect on their cognitive development. While there are environmental effects on intelligence (the highest estimates of heritability for adult IQ are around .8, and some would argue for a lower value; see here for Turkheimer's work suggesting low heritability in the case of severe deprivation), they seem to be idiosyncratic factors that can't be characterized using observable parameters such as the parents' SES, parenting style, level of education, or IQ. It is as if each child experiences their own random micro-environment, independent of these parental or family characteristics.

The nonshared influences are by far the largest environmental (non-genetic) influences on intelligence -- in fact, they are the only detectable non-genetic influences. (Click figure for larger version; from a review by Plomin. More recent overview here.)

Identical twins, whether raised together or apart, turn out to be very similar, but one still finds differences in IQ and personality. The cause of those differences must be the different environments experienced by the twins, but can't be characterized by simple variables of the sort listed above: it is not the case that the twin raised by the higher SES family has, on average, the higher IQ! In fact, twins raised in the same family are about as similar as those raised apart, so family shared environment does not produce a measurable influence. See below for a plausible model that accounts for such outcomes.

By now these results are well understood and accepted by experts, but not by the general population or even policy makers. (See the work of Judith Rich Harris for popular exposition). The naive and still widely held expectation is that, e.g., high SES causes a good learning environment, leading to positive outcomes for children raised in such environments. However, the data suggests that what is really being passed on to the children is the genes of the parent, which are mainly responsible for, e.g., above average IQ outcomes in high SES homes (surprise! high SES parents actually have better genes, on average). Little or no positive effect can be traced to the SES variable for adopted children.

The implications are quite shocking, especially for two groups: high investment parents (because the ability of parents to influence their child's development appears limited) and egalitarians (because the importance of genes and the difficulty in controlling environmental effects seem to support the Social Darwinist position widely held in the previous century).

It is plausible to me that each child tends to create their own environment over time, by selectively seeking out or avoiding stimuli of various types. A bookish kid may end up at the library regardless of whether their father takes them there. An athletic kid may end up on the playground whether or not their mother takes them there. It has been argued that this effect is the reason that the heritability of IQ increases with age: over time, genetic influences assume greater importance as they cause the individual to create or seek out their preferred environment.

In a previous post I discussed individual cognitive profiles as described by an n-vector. Similarly, one could think of an individual's learning profile and learning environment as two more n-vectors. These n-vectors may or may not be well-matched, leading to outcomes with significant and hard to characterize variability. For example, one can imagine that both the environment (provided by parents, siblings, teachers and peers) and a particular child's reactions vary in each of the factors listed below.

Pressure and competition

Stimulation through stories and pretend play; flights of imagination

Ability to learn from repetition and drill / tendency to boredom

Isolated study vs group activities

Visual vs aural vs mechanical stimulation

Level of discipline or structure imposed

Close mentoring vs freedom of exploration

Abstraction vs experimentation

(One can think of many more.)

The factors listed are not intrinsically good or bad for learning -- what matters is whether the learning environment is matched to the nature of the individual child. Some react well to discipline or pressure or story telling, others do not. Further, none of the factors is obviously correlated with SES, parental education level or IQ. Even if they were, it's plausible that a child to some extent creates their learning environment outside the control of parents and teachers (e.g., through peer group or choice of play activities).

An individual whose learning vector (learning style) is well matched to their environment will thrive: the nonshared environmental component in their development will be large and positive. For others, the environment will have a smaller or even negative impact. Because both the learning vector and the environment vector vary in a many-dimensional space, and over time, prediction or control of the overall environmental effect on development is difficult.

Nonshared environmental contributions to development, which are the largest environmental contributions, are effectively random. They are not amenable to control, either by parents or policy makers. Note, this picture -- that each child creates their own environment, or experiences an effectively random one -- does not seem to support the hypothesis that observed group differences in cognitive ability are primarily of non-genetic origin. Nor does it suggest that any simple intervention (for example, equalizing average SES levels) will eliminate group differences. However, it's fair to say our understanding of these complex questions is limited.

Technical remark: if n is large, and factors uncorrelated, the observed environmental variation in a population will be suppressed as n^{-1/2} relative to the maximum environmental effect. That means that the best or worst case scenarios for environmental effect, although hard to achieve, could be surprisingly large. In other words, if the environment is perfectly suited to the child, there could be an anomalously large non-genetic effect, relative to the variance observed in the population as a whole. Of course, for large n these perfect conditions are also harder to arrange. (As a super-high investment parent I am actually involved in attempting to fine tune n-vectors ;-)

Environmental effects cause regression to the mean of a child relative to the parental midpoint. Parents who are well above average likely benefited from a good match between their environment and individual proclivities, as well as from good genes. This match is difficult to replicate for their children -- only genes are passed on with certainty.

blog comments powered by Disqus

Blog Archive


Web Statistics