I’ve been thinking about how the anti-science label tends to get assigned to anyone who tries to move the GMO debate to the level of political economy. (That is, away from whether GMO food is safe and towards who controls production and prices.) Continue reading
A quick comment by a colleague about how she teaches students to think about the meaning and limitations of the concept of heritability led me to explore her example of coming to speak the language of one’s country of origin.
Consider English-speaking families in the USA and families that have children when they immigrate to the USA from a place where English is not spoken. We know that language “runs in the family” in that children will speak the language of their families. Of course, we suspect that is all about the environment parents provide for the children, but let us examine the heritability of English-speakingness anyway.
We can conduct a classic twin study of 10-year olds to estimate heritability by comparing the similarity of identical (monozygotic or MZ) twins and fraternal (dizygotic or DZ) twins. What we would expect to find is that if one twin speaks English so does the other, whether MZ or DZ. And if one twin doesn’t speak English neither does the other, again whether MZ or DZ. No math is needed to conclude that the heritability is zero. Notice that this result holds even if the twins in some immigrant families have gone to schools that enable them to become bilingual but the twins from other families have not learned English, at least not by age 10.
Suppose however that we go back to the foundations of heritability estimation as it is conceived of in trials of plant varieties grown in a number of locations. Growing in a number of locations for humans requires the thought experiment that identical twins are separated and raised one in an English-speaking family and one in a non-English-speaking family. (Perhaps they could be raised in different countries.)
Suppose that the data looked like Table 1, where 1 = speaks English at 10 y.o., 0 = does not.
|raised in an English-speaking family||raised in a non-English-speaking family|
|Twin pair 1||1||0|
|Twin pair 2||1||1|
|Twin pair 3||1||0|
The heritability estimate for the data in Table 1 is .25.
(The math are as follows: Means for the twin pairs = .5, 1., .5. Variance of those means = 1/18. Variance for the trait over the whole data set = 2/9. Heritability = ratio of those variances. [Any statistician worrying about sample versus population variance can simply imagine that this pattern in the data is repeated many times so that the two estimates converge.])
Why is the estimate not zero this time? After all, nothing changed about the way the world works in relation to language acquisition. We might suspect that the data point for Twin Pair 2 raised in a non-English-speaking family is recorded incorrectly and should be 0. But suppose we have asked for this to be checked and it is correct. What needs to be understood is that the estimate of 0 from the original twin study corresponds to a partial snapshot of the phenomenon compared to the separated twins study, as indicated in Table 2.
|raised in an English-speaking family||raised in a non-English-speaking family|
|MZ or DZ twin pair 1 raised together||1||n.a.|
|MZ or DZ twin pair 2 raised together||n.a.||0|
|MZ or DZ twin pair 3 raised together||Etc.||Etc.|
What does it mean that the heritability estimate is not zero? If the data were from plants not people (and the trait was some plant trait!), then non-zero heritability means that the breeder could expect to increase the average value of the trait in the population by selectively breeding the variety that has higher average value across the locations (i.e., the Twin Pair 2 variety). The plant breeder might be curious about the factors underlying the observed trait values, but would not have to discover those factors before proceeding.
But these are human data; selective breeding is not possible. So the next way to look at the non-zero heritability would be to investigate what the underlying factors are. It turns out, once a sociologist looked deeper at the twins in their families that, for twin-pair 2, the twin raised in a non-English-speaking family watched a lot of English-language TV and learned English even though it wasn’t spoken by the family.
Watching a lot of English-language TV sounds like an environmental factor, so how do we understand that factor resulting in non-zero heritability? From the plant breeder’s point of view, we now have to make sure that the locations have that factor—the families need to have TV with English-language shows and let the child view them—if the heritability value is still to lead us expect to increase the average value of the trait in the population by selectively breeding the variety that has higher average value across the locations. In short, understanding non-zero heritability does not require that we have exposed underlying genetic factors.
Four objections might arise:
- Given that selective breeding is not possible, why then are we interested in heritability for humans? Answer: There is no need to be.
- Is the analysis of Table 1 correct given that being raised in a non-English-speaking family is not being raised in the same location if some children can watch a lot of English-language TV and others can’t? Answer: We don’t know that the others couldn’t watch English-language TV. But, even if that were the case, it was through analysis of the data that we decided to look more deeply at the families (locations). If we had known in advance what all the relevant underlying factors were we wouldn’t have bothered with the twin study.
- It is possible that all the twins were allowed to watch English-language TV in the non-English-speaking families, but only in Twin Pair 2 did the twin learn English by the age of 10. The underlying factor is no longer having English-language TV to watch, but choosing to watch it and learning from that. That no longer sounds like an environmental factor. Response: a. From the plant breeder’s point of view, nothing has changed; the label is unimportant; b. From the human sociologist’s point of view, the situation has become more interesting: What leads a child to choose to watch and learn from the English-language TV when it is not spoken in the family? This is an interesting question, but not one that demands that the factors we investigate are genetic.
- What if we learned that, contra the twin study described at the start, DZ twins are less similar than MZ twins in choosing to watch and learn from English-language TV in the non-English-speaking families? That is, the heritability estimate is non-zero. Answer: We shouldn’t be any more likely to search for underlying genetic factors than we would based on the non-zero estimate based on Table 1 (which, as noted earlier, is a complete not partial snapshot of the situation).
The impulse to look for the underlying factors is understandable if we are interested in changing the situation (in this case, to produce English-language speakers even in non-English-speaking families). What heritability estimation does not warrant is taking values of heritability as an indication that the factors to look for are genetic. Indeed, heritability estimation is a snapshot of a situation at one point of time, so it does not warrant a subsequent search for underlying factors at all.
When breeders use the estimates to make predictions about advances under selection they know from experience that the outcomes do not always match the predictions. If they care enough about the discrepancy, they might go on to investigate a) the underlying genetic factors and how they are getting recombined through bi-parental matings (unless they are cloning offspring); b) the underlying environmental factors to see whether they are truly reproducing the locations from generation to generation; and c) the ways that those genetic and environmental factors combine to influence the trait.
A clear understanding that heritability does not measure the relative influence of genetic versus environmental factors may lead us not to teach students to think about the meaning and limitations of the concept of heritability through human examples that involve modifying underlying factors. (The classic case of this approach involves not language learning, but the high heritability human trait height. One points to the increase in average height of Japanese from the pre-WWII to the post-WWII generation and suggests that changes in the quality of diet led to the change. The problem with this approach is that it invites us to imagine that the explanation is probably genetic factors if we encounter a trait in which there is a large average difference between groups but no obvious single environmental factor explains the difference.)
In recent publications (Taylor 2010, 2012) I show how the estimates made in human quantitative genetics are unreliable and typical interpretations, including interpretation of non-shared environmental influences, are unjustified. Some readers may deem it implausible, given decades of debate among methodologically sophisticated scholars, that some fundamental problems in quantitative genetic estimation have been overlooked (Kendler 2005). With a view to moving at least some skeptical readers to consider the full set of problems and conceptual themes I present, let me sketch the background that allows me to see the study of heredity and variation differently from most researchers and philosophers of science who have addressed quantitative genetics.
My initial research work in the mid-1970s involved the statistical analysis of large plant breeding trials, in which many cultivated varieties would be tested in each of many locations around the world. A first step in the analysis was to partition the variation in a given trait, say, yield of wheat plants, into components related to the averages or means of the varieties (across all locations), the means of the locations (across all varieties), and so on. (Indeed, agricultural breeding was where partitioning of variation and measuring heritability originated.) The challenge for the plant breeders with whom I worked was to go beyond the partitioning and hypothesize what it was about any variety that led to its pattern of response across locations and what it was about any location that led to the varieties’ responses in that location compared to others. Knowing what aspects of, say, the pedigree of the variety or of the environment conditions in the location could inform subsequent breeding or cultivation decisions. Yet, hypothesis generation was not easy even though we had large and complete data sets to work from. A lesson from that experience was that the limits to hypothesizing about genetic and environmental factors must be even greater when researchers partition variation for human traits. In human studies any genetically-defined type is, at best, replicated twice—as identical twins—and different genetic types cannot be systematically raised across the same range of “locations”—families, socio-economic conditions, and so on.
Fast forward to a decade ago: I was learning about three disparate areas of quantitative research that attempt to make sense of the complexity of biological and social factors that build on each other in the development of the given trait over the life course (Taylor 2004). I was impressed by what had been accomplished, but had some reservations about the models used in one of the areas, namely, Dickens and Flynn’s (2001) attempt to resolve the IQ paradox, in which researchers find large generation-to-generation advances in IQ test scores even though the trait is held to have high heritability. I explained my reservations to Dickens, digested his responses, and explained my reservations about his subsequent responses. In the course of this I found myself digger deeper into the conceptual foundations of heritability estimation and partitioning of variation. In order to present a picture that differed from what Dickens, Flynn, and others accepted without second thought, I was explicating first principles, not disputing specialized models or mathematics. Making extensive use of perspectives and examples from the earlier plant breeding research, my exposition took a pedagogical style (Taylor 2006, 2007, 2010).
Meanwhile, my investigation continued of the other two areas—life events and difficulties research (Brown and Harris 1989) and developmental origins of chronic diseases (Barker 1998). Barker’s work led me to life-course epidemiology (Kuh and Ben-Shlomo 2004), so I spent time with Ben-Shlomo and the active social epidemiology research group at Bristol University. Davey Smith is a leading figure in that group and co-edits the International Journal of Epidemiology based at Bristol. While visiting in 2007 I gave a talk on “new and old debates about genes and environment,” which touched on some of the questions about heterogeneity raised in this article. Davey Smith’s spoken response was along the lines of his subsequent “gloomy prospect” article: epidemiologists have to accept considerable randomness at the individual level and keep the focus on modifiable causes of disease at the population level. In his ensuing article, Davey Smith (2011) links this perspective to claims from quantitative genetics, thus providing me an opportunity to address social epidemiologists and human quantitative geneticists at the same time as I respond to his account. In an as-yet-unpublished article bringing my interest in heterogeneity to the attention of those audiences, I extend the pedagogical style and first-principles emphasis of the other recent work and thereby speaks to philosophers of science. My contribution to philosophy takes the form, however, of articulating conceptual themes, not dissecting specific cases on empirical, analytical, bioethical or policy grounds. The expository approach reflects the background reviewed here, with its roots in plant breeding trials, as well as the idea that contributing to the conceptual toolbox of readers will prepare them to make their own contributions to wider discussion of heredity, variation, and heterogeneity.
Barker, D. J. P. (1998). Mothers, Babies, and Health in Later Life. Edinburgh: Churchill Livingstone.
Brown, G. W., & Harris, T. O. (Eds.) (1989). Life Events and Illness. New York: Guilford Press.
Davey Smith, G. (2011). Epidemiology, epigenetics and the ‘Gloomy Prospect’: embracing randomness in population health research and practice. International Journal of Epidemiology, 40, 537-562.
Dickens, W. T., & Flynn, J. R. (2001). Heritability estimates versus large environmental effects: The IQ paradox resolved. Psychological review, 108, 346-369.
Kendler, K. S. (2005). Reply to J. Joseph, Research Paradigms of Psychiatric Genetics. American Journal of Psychiatry, 162, 1985-1986.
Kuh, D., & Ben-Shlomo, Y. (Eds.) (2004). A Life Course Approach to Chronic Disease Epidemiology. Oxford: Oxford University Press.
Taylor, P. J. (2004). What can we do? — Moving debates over genetic determinism in new directions. Science as Culture, 13, 331-355.
Taylor, P. J. (2006). Heritability and heterogeneity: On the limited relevance of heritability in investigating genetic and environmental factors. Biological Theory: Integrating Development, Evolution and Cognition, 1, 150-164.
Taylor, P. J. (2007). The Unreliability of High Human Heritability Estimates and Small Shared Effects of Growing Up in the Same Family Biological Theory: Integrating Development, Evolution and Cognition, 2, 387-397.
Taylor, P. J. (2010). Three puzzles and eight gaps: What heritability studies and critical commentaries have not paid enough attention to. Biology & Philosophy, 25, 1-31.
Taylor, P. J. (2012). A gene-free formulation of classical quantitative genetics used to examine results and interpretations under three standard assumptions. Acta Biotheoretica, 60, 357-378.
My first, mostly unscripted take of a response to an invitation to prepare a very short video where Activists and Scholars respond to the question: “What can Europe learn from the World?” http://www.youtube.com/embed/9Jn0GMweuTs I shift the question to what can European researchers learn.
Takes 2 or 3 or… are needed. My video is far longer than the others on the current playlist: http://www.youtube.com/playlist?list=PLXGOQzlnH7jRLCQoyqIpubpBcz-FSVie8