Based on Burt's definition of downward causation, phenotypes A and B exhibit downward causation when risk alleles for phenotype A predict phenotype B. A's risk alleles predict B neither because B is a biological consequence of A (e.g., kidney disease can be a biological consequence of diabetes) nor because A and B share risk alleles (e.g., multiple risk loci are shared between autoimmune diseases). Instead, the causal connection between A and B is because of socio-environmental forces (norms, practices, institutions). In addition, as the example of educational attainment (EA) shows, B may be a wholly social construct without its own risk alleles.
Thus characterized, downward causation is equivalent to what sociogenomicists misleadingly refer to as “vertical pleiotropy” (van Rheenen, Peyrot, Schork, Lee, & Wray, Reference van Rheenen, Peyrot, Schork, Lee and Wray2019). Pleiotropy occurs when a single gene plays a causal role in two or more distinct phenotypes. Despite its name, vertical pleiotropy is not a form of pleiotropy at all because phenotypes A and B do not share any causal alleles. Hill and Davis, for example, treat genetic variants that ostensibly predict income as an example of vertical pleiotropy, noting that (Reference Hill, Neil, Ritchie, Skene, Bryois, Bell and Deary2019, p. 19), “genetic variants do not act directly on income; instead, genetic variants are associated with partly heritable traits (such as intelligence, conscientiousness, health, etc.), which have their own complex gene-to-phenotype paths (including neural variables) and are ultimately associated with income.” They also comment that any correlation between a given attribute (e.g., intelligence or health) and income is determined by social institutions: Income could just as well depend on service to the party, and different political policies could alter, if not eliminate, the degree of correlation between, for example, income and health.
As with income, most sociogenomicists would agree there are no alleles for EA per se. Rather, there are alleles for attributes that causally affect EA and it is the socially constructed nature of the educational system that determines what attributes of persons are relevant and rewarded (which may include features of persons that are socially valued, such as attractiveness and height, but not knowingly made criterion of EA). Perhaps in one society, obedience is valued more than critical thinking and rewarded accordingly. Under the assumption that cognitive performance is “strongly” genetically influenced and that socially it exerts a decisive causal influence on EA, sociogenomicists typically treat EA as a “proxy variable” for cognitive performance (Rietveld et al., Reference Rietveld, Esko, Davies, Pers, Turley and Benyamin2014, p. 13791).
Burt implies that in addition to the properties mentioned above, downward causation is characterized by phenotype A being an inappropriate socially mediated cause of EA. All the examples she presents for phenotype A – ethnicity, skin color, attractiveness, height, weight – are examples in which most would agree that it is indeed inappropriate, if not a grave social injustice, that A has a causal effect on EA. Burt notes of such phenotypes, “In a GWAS, such alleles [alleles associated with skin pigmentation of African Americans, but also attractiveness, height, weight, etc.] would be identified as causing differences in educational attainment, but these association signals would, of course, be artificial.” However, the signals would be no less “artificial” if the alleles identified as causing differences in EA were associated with intelligence. Although many would consider this an “appropriate” cause of differences in EA, as noted above, to the extent that it is a cause is because of contingent social and institutional practices and norms.
Normative objections in this context makes one vulnerable to the charge (common enough) that one's objection is not scientific. Such an invocation is unnecessary because Burt herself has already convincingly demonstrated the problem with downward causation in the context of EA (and most other social attributes such as income). The problem lies with the assumption that EA is a proxy variable for intelligence, that is, that in measuring EA, sociogenomicists are measuring intelligence. The error is scientific, not normative, to the extent that this assumption is wrong.
First, as Burt shows, to whatever extent intelligence has genetic influences, the realization of intelligence as a phenotype is so intertwined with so many socio-environmental variables that it is impossible to separate “the genetic influence on intelligence” as some pre-existing, isolated, potential force. Moreover, one of these influences on intelligence may well be education itself. In place of the assumed unidirectional causal pathway from intelligence to EA, EA itself may influence intelligence, resulting in reciprocal or bidirectional causation (Hegelund et al., Reference Hegelund, Grønkjær, Osler, Dammeyer, Flensborg-Madsen and Mortensen2020).
Second, as Burt also shows, there are strong reasons to believe that polygenic EA scores predict not intelligence but ancestry. Population stratification itself is an example of downward causation/vertical pleiotropy. Genetic ancestry bears a socially determined association with any number of social attributes, EA and income being two noteworthy examples. In the latest in a long series of studies showing the enduring impact of population stratification on genome-wide association studies (GWASs) of complex traits, the authors note that “controlling for geographic regions significantly decreased the heritability for socioeconomic status (SES)-related traits, most strongly for educational attainment and income” (Abdellaoui, Dolan, Verweiji, & Nivard, Reference Abdellaoui, Dolan, Verweiji and Nivard2022).
A final word concerning sociogenomicists' repeated assertion that in addition to the heritability of intelligence (and whatever other attributes are considered to have an association with EA), EA is itself heritable (the same is said of income). How can EA be said to be heritable if there are no genetic variants that act directly on it? One might object that nothing in the concept of heritability requires that a trait deemed heritable be influenced by the transmission of parental risk alleles for that trait. It is sufficient that heritable trait A stands in a (socially mediated) causal relationship to trait B. However, if we accept this, we would have no grounds to claim, to use an example of population stratification cited by Burt, that chopstick use is not heritable. Rather, we could say that it is an example of vertical pleiotropy. Being of East Asian descent (phenotype A) is a heritable attribute, and because of social practices (norms, conventions) it is causally associated with chopstick use (phenotype B).
You have
Access
Based on Burt's definition of downward causation, phenotypes A and B exhibit downward causation when risk alleles for phenotype A predict phenotype B. A's risk alleles predict B neither because B is a biological consequence of A (e.g., kidney disease can be a biological consequence of diabetes) nor because A and B share risk alleles (e.g., multiple risk loci are shared between autoimmune diseases). Instead, the causal connection between A and B is because of socio-environmental forces (norms, practices, institutions). In addition, as the example of educational attainment (EA) shows, B may be a wholly social construct without its own risk alleles.
Thus characterized, downward causation is equivalent to what sociogenomicists misleadingly refer to as “vertical pleiotropy” (van Rheenen, Peyrot, Schork, Lee, & Wray, Reference van Rheenen, Peyrot, Schork, Lee and Wray2019). Pleiotropy occurs when a single gene plays a causal role in two or more distinct phenotypes. Despite its name, vertical pleiotropy is not a form of pleiotropy at all because phenotypes A and B do not share any causal alleles. Hill and Davis, for example, treat genetic variants that ostensibly predict income as an example of vertical pleiotropy, noting that (Reference Hill, Neil, Ritchie, Skene, Bryois, Bell and Deary2019, p. 19), “genetic variants do not act directly on income; instead, genetic variants are associated with partly heritable traits (such as intelligence, conscientiousness, health, etc.), which have their own complex gene-to-phenotype paths (including neural variables) and are ultimately associated with income.” They also comment that any correlation between a given attribute (e.g., intelligence or health) and income is determined by social institutions: Income could just as well depend on service to the party, and different political policies could alter, if not eliminate, the degree of correlation between, for example, income and health.
As with income, most sociogenomicists would agree there are no alleles for EA per se. Rather, there are alleles for attributes that causally affect EA and it is the socially constructed nature of the educational system that determines what attributes of persons are relevant and rewarded (which may include features of persons that are socially valued, such as attractiveness and height, but not knowingly made criterion of EA). Perhaps in one society, obedience is valued more than critical thinking and rewarded accordingly. Under the assumption that cognitive performance is “strongly” genetically influenced and that socially it exerts a decisive causal influence on EA, sociogenomicists typically treat EA as a “proxy variable” for cognitive performance (Rietveld et al., Reference Rietveld, Esko, Davies, Pers, Turley and Benyamin2014, p. 13791).
Burt implies that in addition to the properties mentioned above, downward causation is characterized by phenotype A being an inappropriate socially mediated cause of EA. All the examples she presents for phenotype A – ethnicity, skin color, attractiveness, height, weight – are examples in which most would agree that it is indeed inappropriate, if not a grave social injustice, that A has a causal effect on EA. Burt notes of such phenotypes, “In a GWAS, such alleles [alleles associated with skin pigmentation of African Americans, but also attractiveness, height, weight, etc.] would be identified as causing differences in educational attainment, but these association signals would, of course, be artificial.” However, the signals would be no less “artificial” if the alleles identified as causing differences in EA were associated with intelligence. Although many would consider this an “appropriate” cause of differences in EA, as noted above, to the extent that it is a cause is because of contingent social and institutional practices and norms.
Normative objections in this context makes one vulnerable to the charge (common enough) that one's objection is not scientific. Such an invocation is unnecessary because Burt herself has already convincingly demonstrated the problem with downward causation in the context of EA (and most other social attributes such as income). The problem lies with the assumption that EA is a proxy variable for intelligence, that is, that in measuring EA, sociogenomicists are measuring intelligence. The error is scientific, not normative, to the extent that this assumption is wrong.
First, as Burt shows, to whatever extent intelligence has genetic influences, the realization of intelligence as a phenotype is so intertwined with so many socio-environmental variables that it is impossible to separate “the genetic influence on intelligence” as some pre-existing, isolated, potential force. Moreover, one of these influences on intelligence may well be education itself. In place of the assumed unidirectional causal pathway from intelligence to EA, EA itself may influence intelligence, resulting in reciprocal or bidirectional causation (Hegelund et al., Reference Hegelund, Grønkjær, Osler, Dammeyer, Flensborg-Madsen and Mortensen2020).
Second, as Burt also shows, there are strong reasons to believe that polygenic EA scores predict not intelligence but ancestry. Population stratification itself is an example of downward causation/vertical pleiotropy. Genetic ancestry bears a socially determined association with any number of social attributes, EA and income being two noteworthy examples. In the latest in a long series of studies showing the enduring impact of population stratification on genome-wide association studies (GWASs) of complex traits, the authors note that “controlling for geographic regions significantly decreased the heritability for socioeconomic status (SES)-related traits, most strongly for educational attainment and income” (Abdellaoui, Dolan, Verweiji, & Nivard, Reference Abdellaoui, Dolan, Verweiji and Nivard2022).
A final word concerning sociogenomicists' repeated assertion that in addition to the heritability of intelligence (and whatever other attributes are considered to have an association with EA), EA is itself heritable (the same is said of income). How can EA be said to be heritable if there are no genetic variants that act directly on it? One might object that nothing in the concept of heritability requires that a trait deemed heritable be influenced by the transmission of parental risk alleles for that trait. It is sufficient that heritable trait A stands in a (socially mediated) causal relationship to trait B. However, if we accept this, we would have no grounds to claim, to use an example of population stratification cited by Burt, that chopstick use is not heritable. Rather, we could say that it is an example of vertical pleiotropy. Being of East Asian descent (phenotype A) is a heritable attribute, and because of social practices (norms, conventions) it is causally associated with chopstick use (phenotype B).
Financial support
The author has received no funding for this article.
Competing interest
None.