The challenge for understanding whether something “causes” something else is that most of our expectations for causal reasoning rely on the availability of experimental evidence (or: comparisons to counterfactuals). When working with human genetics, or in many fields of observational relationships between (for example) environments and phenotypes, we can’t randomly assign folks to different conditions (whether those conditions are genetic, or environmental - for example, we can’t randomly assign people to different families, or to smoke or not smoke, or to different socioeconomic statuses). Behavior genetics provides a variety of methods to potentially address some of the expectations and assumptions that would arise from a causal relationship, even in the absence of direct experimental manipulation.

Each of these methods comes with its own benefits, limitations, things we can learn, and assumptions they rely on. Drawing causal conclusions often relies on the principle of “triangulation” among methods: no single non-experimental method will be conclusive, but if we draw similar conclusions from applying a broad diversity of available methods, we can become reasonably confident that a causal relationship exists even in the absence of experimental research in humans. A classic example of this is the relationship between smoking and cancer. There has never been a randomized control trial of the long-term health consequences of smoking in humans because you couldn’t (either practically or ethically) randomly assign people to smoke or not smoke over the course of decades. Rather, we look at converging evidence from many methods to conclude that either the relationship is causal OR all of the methods (again, based on different assumptions) just happen to be biased in the same wrong direction (a situation that I personally take to be rather unlikely).

In particular in the Briley et al. (2018) paper, pay attention to Figures 1, 2, & 3, which illustrate alternative causal pathways that might result in similar observed correlational patterns. Depending on the true causal pathway, intervening in the genes (if you wanted to) wouldn’t necessarily change the outcome even though a genetic association or heritability would have been robustly observed. Similarly, genetic methods can help us disentangle the causal ordering of phenotypic and environmental characteristics. For example, if we want to improve educational outcomes, should we try to change children (grit, mindset, test training) or change schools (resources, structures) or change families (reading together, resources, parenting approaches)? Whenever our goal is to change outcomes, we need to identify the causal relationships among a whole bunch of correlated variables to predict which features may be most effective to target for interventions (and we must always make choices about where and when to intervene, prevent, or treat; these decisions are costly and often necessarily mean NOT intervening/preventing/treating other areas). Genetically informative research designs don’t only tell us about genes; they offer an opportunity to anchor our understanding of the relationships among non-genetic factors, as well, even when human experimentation is impractical, unethical, or impossible.

Next: 11.1. Substance Use

Previous: 10.2. Activity: Contextualize a Popular Source

Home: Table of Contents