Madole & Harden (M&H) compellingly argue that estimated within-family genetic effects can be interpreted as causes in much the same way as randomized controlled trials. Increased focus on within-family molecular genetic designs is needed for many reasons, and M&H illustrate the power of such designs in understanding human neurophysiology and behavior.
Their argument relies on counterfactual accounts of causality, however. Counterfactual accounts have many strengths, but have been equally criticized, usually because of their dependence on states that are by definition impossible to observe (Dawid, Reference Dawid2000). Counterfactual causal accounts attempt to estimate what would have been the case in a different set of conditions that did not and cannot eventuate, leaving a user of such models in a predicament about how to use them: If the causal inference is about something in the past that did not occur, what about the present that did occur, and the future?
Traditional causal accounts and many contemporary accounts (e.g., Granger causality; Dawid, Reference Dawid2015; Granger, Reference Granger1969; Janzing, Balduzzi, Grosse-Wentrup, & Schölkopf, Reference Janzing, Balduzzi, Grosse-Wentrup and Schölkopf2013; Schreiber, Reference Schreiber2000), address a different, actionable question: To what extent does a potential cause provide unique information about expected outcomes following an independent decision, such as a decision to implement a manipulation or intervention? Designs developed within these paradigms, such as randomized controlled trials, directly address this type of question, in that a decision is randomly made so as to instantiate independence from past states, and the outcomes of this decision are then observed.
Counterfactual theory suggests that it also provides this information, by estimating what would have happened if a decision of sorts by nature had been made, assuming it could have been made. However, this relies on a number of assumptions about that alternate possible world that might or might not be true. In an important sense, moreover, it takes an unactionable epistemological stance: Even if a counterfactual account informs about what would have occurred had things been different, it does not inform about what one can do now, given things as they are.
This is a critical distinction given the nature of causal pathways involved in behavior genetics. Usually interest is not actually in the immediate effects of the genotype per se – the nucleotide sequences at different loci and their translation – but the affected neurodevelopmental processes, those effects on neurophysiology at a later time, and their effects on experience and behavior. The black box between gene and behavior is in fact the phenomenon often most of interest in terms of causal explanation. In the polygenic risk regime, where each polymorphism might have an almost unmeasurable unique effect on phenotype, effects of a particular polymorphism or haplotype may be still further removed from the aggregate neurodevelopmental endpoint of primary interest.
There are some ways to construe genetic effects in terms of decision outcome information. For instance, it might be argued that genotype provides additional predictive information about the outcomes of particular decisions for particular individuals – that is, in deciding what intervention to provide to whom, above and beyond any nongenetic data. This is a reasonable argument, but again, with numerous intraindividual and extraindividual inputs into behavioral development, it may be that downstream predictors provide more information about behavior, being causally more proximal mediators of any genetic effects (Morris, Davies, & Davey Smith, Reference Morris, Davies and Davey Smith2020). If a gene is one of many causes of an easily identifiable condition, to treat the condition isn't it more efficient to identify those with the condition, rather than the gene? Moreover, in such a setting the emphasis is still on how to improve efficacy of an intervention, such as an educational or medical intervention.
Another way to construe genetic effects in terms of decision outcome information is in terms of genetic manipulation. Gene editing is a reality (Anguela & High, Reference Anguela and High2019; Saha et al., Reference Saha, Sontheimer, Brooks, Dwinell, Gersbach, Liu and Zhou2021), and randomized controlled trials of genotype manipulation may become salient considerations in the neurobehavioral sciences sooner than is often appreciated. Given developmental cascades (Elam, Lemery-Chalfant, & Chassin, Reference Elam, Lemery-Chalfant and Chassinin press), it is likely that if genotypes are not altered early in neurodevelopment, in many cases genetic effects likely will be irreversible. In that case, the options for intervention are again further downstream from genes both in time and causal proximity to experience and behavior. Alternatively, one could implement preventative gene editing, but that would be a form of eugenics with all the attendant ethical challenges it implies.
M&H's argument for counterfactual causality is telling in that it implies behavior genetic designs have so far often not been seen as causally compelling – otherwise their argument would be unnecessary. Reluctance to construe behavior genetic effects in terms of causes has likely been because of numerous factors, including limitations of common designs, such as lack of precise genotypic information or lack of within-family controls for variables varying between families. However, the reluctance also arguably reflects a perception that behavior genetic studies have generally not provided information about what is changeable or targetable. Translating behavioral genetic effects into interventional hypotheses and designs, where the information they provide can be leveraged to prevent and treat, will likely increase the reception and perceived relevance of such findings. Sometimes causes have effects that, set into motion in a causal chain, are impossible to reverse. But the actionable, effective relevance of the causal chain only goes back so far.
You have
Access
Madole & Harden (M&H) compellingly argue that estimated within-family genetic effects can be interpreted as causes in much the same way as randomized controlled trials. Increased focus on within-family molecular genetic designs is needed for many reasons, and M&H illustrate the power of such designs in understanding human neurophysiology and behavior.
Their argument relies on counterfactual accounts of causality, however. Counterfactual accounts have many strengths, but have been equally criticized, usually because of their dependence on states that are by definition impossible to observe (Dawid, Reference Dawid2000). Counterfactual causal accounts attempt to estimate what would have been the case in a different set of conditions that did not and cannot eventuate, leaving a user of such models in a predicament about how to use them: If the causal inference is about something in the past that did not occur, what about the present that did occur, and the future?
Traditional causal accounts and many contemporary accounts (e.g., Granger causality; Dawid, Reference Dawid2015; Granger, Reference Granger1969; Janzing, Balduzzi, Grosse-Wentrup, & Schölkopf, Reference Janzing, Balduzzi, Grosse-Wentrup and Schölkopf2013; Schreiber, Reference Schreiber2000), address a different, actionable question: To what extent does a potential cause provide unique information about expected outcomes following an independent decision, such as a decision to implement a manipulation or intervention? Designs developed within these paradigms, such as randomized controlled trials, directly address this type of question, in that a decision is randomly made so as to instantiate independence from past states, and the outcomes of this decision are then observed.
Counterfactual theory suggests that it also provides this information, by estimating what would have happened if a decision of sorts by nature had been made, assuming it could have been made. However, this relies on a number of assumptions about that alternate possible world that might or might not be true. In an important sense, moreover, it takes an unactionable epistemological stance: Even if a counterfactual account informs about what would have occurred had things been different, it does not inform about what one can do now, given things as they are.
This is a critical distinction given the nature of causal pathways involved in behavior genetics. Usually interest is not actually in the immediate effects of the genotype per se – the nucleotide sequences at different loci and their translation – but the affected neurodevelopmental processes, those effects on neurophysiology at a later time, and their effects on experience and behavior. The black box between gene and behavior is in fact the phenomenon often most of interest in terms of causal explanation. In the polygenic risk regime, where each polymorphism might have an almost unmeasurable unique effect on phenotype, effects of a particular polymorphism or haplotype may be still further removed from the aggregate neurodevelopmental endpoint of primary interest.
There are some ways to construe genetic effects in terms of decision outcome information. For instance, it might be argued that genotype provides additional predictive information about the outcomes of particular decisions for particular individuals – that is, in deciding what intervention to provide to whom, above and beyond any nongenetic data. This is a reasonable argument, but again, with numerous intraindividual and extraindividual inputs into behavioral development, it may be that downstream predictors provide more information about behavior, being causally more proximal mediators of any genetic effects (Morris, Davies, & Davey Smith, Reference Morris, Davies and Davey Smith2020). If a gene is one of many causes of an easily identifiable condition, to treat the condition isn't it more efficient to identify those with the condition, rather than the gene? Moreover, in such a setting the emphasis is still on how to improve efficacy of an intervention, such as an educational or medical intervention.
Another way to construe genetic effects in terms of decision outcome information is in terms of genetic manipulation. Gene editing is a reality (Anguela & High, Reference Anguela and High2019; Saha et al., Reference Saha, Sontheimer, Brooks, Dwinell, Gersbach, Liu and Zhou2021), and randomized controlled trials of genotype manipulation may become salient considerations in the neurobehavioral sciences sooner than is often appreciated. Given developmental cascades (Elam, Lemery-Chalfant, & Chassin, Reference Elam, Lemery-Chalfant and Chassinin press), it is likely that if genotypes are not altered early in neurodevelopment, in many cases genetic effects likely will be irreversible. In that case, the options for intervention are again further downstream from genes both in time and causal proximity to experience and behavior. Alternatively, one could implement preventative gene editing, but that would be a form of eugenics with all the attendant ethical challenges it implies.
M&H's argument for counterfactual causality is telling in that it implies behavior genetic designs have so far often not been seen as causally compelling – otherwise their argument would be unnecessary. Reluctance to construe behavior genetic effects in terms of causes has likely been because of numerous factors, including limitations of common designs, such as lack of precise genotypic information or lack of within-family controls for variables varying between families. However, the reluctance also arguably reflects a perception that behavior genetic studies have generally not provided information about what is changeable or targetable. Translating behavioral genetic effects into interventional hypotheses and designs, where the information they provide can be leveraged to prevent and treat, will likely increase the reception and perceived relevance of such findings. Sometimes causes have effects that, set into motion in a causal chain, are impossible to reverse. But the actionable, effective relevance of the causal chain only goes back so far.
Financial support
This research received no specific grant from any funding agency, commercial, or not-for-profit sectors.
Competing interest
None.