Wednesday 11 September 2013

Heritability estimates and unexplained variance


Nobody owns unexplained variance. However, standard heritability estimates quite rightly assign the variance which cannot be explained on a genetic basis to “the environment”. For example, in twin studies genetics researchers argue that the higher similarity of monozygotic twins compared with dizygotic twins is down to the greater genetic similarity of the former. Yes, all twins have family environments in common, but the actual observed differences are due to the degree of genetic similarity, and from that the heritability estimate is derived. The rest is “environmental” with little specification as to what that includes, though some distinctions are drawn between family home environments and all other environments. Errors and shortcomings of the genetic model are conceded to the environment.

Now consider whether one can get down to more detail. For example, consider whether one could study environmental conditions so as to specify how much of the overall “environmentality” is due to specific family and school factors, and how much is not attributable to any measureable environmental factor but must be considered an error term. This drilling down to real causes has proved difficult. It is easy to come up with a list of putative causes of “environmentality”: perhaps poorer mothers do not have the time to spend with their babies, or simply do not realise they need to talk to their babies before the babies can talk back to them, or do not have books in the house, or do not read their children night-time stories, or any stories at any time, or just sit their child in front of a TV, or let them play video-games and do not take them to museums. All these are plausible to some degree.

Fostering out children to more intellectually demanding and supportive nurseries or adoptive parents can give estimates of how much can be achieved by enriched environments. There are supportive studies regarding early childhood tutoring (Abecedarian project) early schooling, extra tutoring, and adoption. How much variance can be attributed to each of these enriched environments?  The Abecedarian project boosted IQ by 4 points. Adoption may have boosted intelligence by a like amount. How much of the environmental variance is accounted for by these measures? I do not know for sure, because many of these measures are plagued by covariance confounders. The percentage of environmentality accounted for by specific procedures seems to be rather small. This makes sense, in that boosting child development requires many things to be done, but which are they?

However, we do not usually argue that that such slim pickings prove that the environment does not have much of an effect. We merely assume that something about the environment is effective, but we have not pinned it down yet. It is somewhere out there, like lead in paint, pesticides in food, and perhaps “negative attitudes” (shouting at the child)  in child rearing.

Turning to heritability estimates, it is not yet possible to show which individual genes, or which groups of many genes, bring about the heritability effect. We know that heritability accounts for variance from twin and other relatedness studies. That is not in doubt. What is in doubt is whether the individual components can be named and numbered.

However, some people regard the low rate of variance accounted for by the deeper gene hunting studies (say 1 to 3% of the variance in replications so far) as proof that heritability is only 1 to 3%. Not so. Heritability estimates indicate the total effect. Gene hunts are trying to ascertain the names and numbers of the individual genetic variants. That will take time. However, genetics have one key card: humans start being one cell big. That cell is the starting point. All other cells derive from it. Good to know the moment of creation, the Big Bang of life.

It will also take time to find the individual components of upbringing that make a difference to child outcomes. Environmental factor hunts need to be more specific. Exactly how should one bring up a child so that they do well at school and in life? One starting point is the Home Observation for Measurement of the Environment which has a correlation of about 0.2 with Wechsler intelligence for children (Molfese, DiLalla Bunce, 1997, Table 2) For a recent (2004) paper, look at

I think that the list is pretty comprehensive as regards what you can usefully observe in a home visit. There is no way that this list can be seen as being independent of the genetic quality of the parent. A good score is not an independent measure of the environment. It is a measure of how good a parent is at looking after and caring for a child (most generally, their own child). However, the detailed list is useful in specifying good and poor quality parenting in adoptive and compensatory tutoring environments.

Although any variance heritability estimates can’t account for is counted as a win for the environment, there must be an error term somewhere. By drilling down from the observed general effect to the specific causes, gene hunters are accounting for some of the variance, and that amount may increase as gene analysis becomes cheaper and more powerful.

How well are the environmental factor hunters doing? That is, how well are they doing if we require them to produce pure measures? For example, not talking to your child can be a genetic effect as well as an environmental effect. Similarly, not having many books in the house can be due to the parents having neither the capacity nor the interest to read books. The lack of books in the home can be the emblem of a lack of intellectual curiosity, rather than the cause of it.

There is an error term, and while the “genetics” explanation can be shown to be requiring improvement, so can the “environment” explanation if it is held to the same standards.

No-one owns the unexplained variance. 


  1. Is the programme realistic? You want, I imagine, to know the effect of "procedure" x on IQ at age, say 25. Or 18. Or at least late enough to know that any IQ increase or decrease attributable to x isn't fleeting. But the wait until 25 or whatever will include an accumulation of many other procedures which will muddy the waters. Moreover, just to add to the difficulty, very few research programmes will last from toddlerhood (say) until the subject is 25. The Principal Investigator will have retired, his team will have scattered, and anyway none of the bureaucrats who control research funding would approve of a programme whose results, if any, will happen far too late to reflect any credit on him.

    Anyway, what if there were a successful hunt and it discovered things that are not PC?

  2. Abecedarian project seemed to manage it. Principal investigator still alive, and still getting funding.

  3. And one's enough if you know how to use it. But is it?