Nascent doubts may presage conceptual clarity: reply to Surbey

Canadian Psychology, Jul 1994 by Wahlsten, Douglas

It is difficult for anyone to acknowledge that an idea he or she has been teaching may be invalid. I faced this situation several years ago and, after a period of vacillation, concluded that heritability analysis, the dominant paradigm in behavioural genetics, was not merely weak at its foundations but actively misleading to the nonspecialist; hence, it ought to be relegated to the history of science along with phlogiston, penis envy and cold fusion. During a postdoctoral fellowship year with McClearn and DeFries at the Institute for Behavioral Genetics at the University of Colorado in 1968 - 69, I eagerly absorbed the lessons of Falconer and my mentors, who were foremost authorities on heritability analysis. These lessons were acquired easily because mathematically they were so similar to the classic rendition of the analysis of variance presented by Hays and Winer, the statistical gurus of psychology graduate students in that period.

Two events blunted the pleasure of the exercise. The article by Jensen (1969) extolled heritability and pointed to its practical application with the question: "Is there a danger that current welfare policies, unaided by eugenic foresight, could lead to the genetic enslavement of a substantial segment of our population?" (p. 95) What seemed innocent enough when used on the farm took on sinister overtones in political debate. Then came the study by Henderson (1970) showing how the estimate of the heritability coefficient was dramatically altered by rearing conditions in a manner that could not be explained by an additive model. In a review of the genetics of animal learning (Wahlsten, 1972), I cited numerous heritability estimates but expressed nascent doubts. Further evidence of heredity - environment interaction accumulated in the literature and, as a speaker invited by Joseph Royce to participate in the 1978 Banff conference on Theoretical Advances in Behavioral Genetics, I found the earlier doubts giving way to forthright opposition (Wahlsten, 1979). Since then, the many direct demonstrations of control of gene action at the molecular level by environmental conditions have amply confirmed the earlier conclusions.

Evidently the author of this commentary now has doubts about heritability analysis and, while vehemently defending the practice, sees no future for it. This leads to a lack of coherence. It appears the commentator has turned up the heat too far, the sauce has separated, and we are presented with an unappetizing mix of technical misunderstandings seasoned with hyperbole.

Serious Complications

The Intelligence of Heritability" does not erect a straw man by highlighting the additive model. Additivity of genetic and environmental components is the conceptual foundation of heritability analysis. The eminent theorist Sewall Wright, whose professional life spanned the entire history of quantitative genetics from 1917, began his exposition of the topic with this cautionary note: "The determination of the genetic component is simple enough if it can be assumed that the effects of genes are wholly additive in relation to each other and to environmental variations. A first rough approximation may often be obtained on this hypothesis, but there may be serious complications ..." (Wright, 1969a, p. 418) The foundation is not, as the commentator asserts, that h'Symbol not transcribed'2 = V'Symbol not transcribed'G/V'Symbol not transcribed'P; that formula is a conclusion from an algebraic derivation that requires P = G E or some more elaborate equivalent. There is a world of difference between a premise and a conclusion. Without additivity for an individual, heritability in a population is utter nonsense.

It is not true that the model of development in my Figure 4 and population heritability are unrelated because they pertain to different levels. Only if there are two developmentally separate causes adding to yield the phenotypic value of an individual is the phenotypic variance in a population equal to the sum of variances of the two components. I reject the model of an individual that is actually used to justify heritability calculations. It is this additive model which proponents of heritability must somehow defend. Of course, some authors pay lip service to heredity - environment interaction. As my essay states: "Many sources introduce the model with a GXE interaction term attached ..." Once they excise this dangling appendix, which almost always seems to happen in research with humans, they cannot thwart my criticism by claiming they did display an interaction term on page such - and - such and then proceed with a purely additive model. I am quite certain that most students and many of my colleagues for that matter, first at the University of Waterloo and now at the University of Alberta, do not understand the false assumptions inherent in heritability calculations. Are folks at Mount Allison University so much more sophisticated statistically? I am skeptical of this.