I recently finished reading the back-and-forth on Callie Burt’s article in Behavioral and Brain Sciences (BBS), on the use (or rather against the use) of polygenic indices in the social sciences. There are a few sound arguments in her original paper, of which I know of no opponents. However, what is arguably the most important argument in the paper – namely that within-family estimates of genetic effects are confounded because of “downward causation” – is unfortunately simply wrong. In fact, many of the more forceful arguments in her paper could just as well be used to argue against the possibility of ever establishing that environmental factors influence social outcomes.

The background to her thesis is that conventional genetic association studies are well-known to be confounded by environmental differences between strata of the population that might also differ in the frequencies at which certain genetic variants occur. This is known in the literature as population stratification and can lead to mistaking what are actually environmental effects for genetic effects. The solution to this problem is increasingly understood to be to study genetic differences within families, and more specifically between siblings, since the Mendelian process of random segregation provides an actual randomized natural experiment: genetic differences between siblings are random, meaning that systematic downstream differences in whatever we’re interested in cannot be attributed to environmental confounds. This means that within-family estimates of genetic effects really are causal, not unlike how outcome differences in a randomized controlled trial are causal. Sometimes to the chagrin of those who oppose the notion that genetics could ever be relevant for understanding social outcomes, these causally robust methods are now at a rapid pace finding genetic factors that indeed have such effects, and that occasionally rival the explanatory power of more conventional social science variables.

The argument that Burt provides about downward causation, briefly summarized, is that even these purportedly causal genetic effects are confounded, because the effect could also be transmitted via environmental factors (such as discrimination). That is, some biologically influenced factor elicits a response from the environment, which in turn causes something to change at the individual level. As such, Burt argues, these are actually environmental effects “masquerading” as genetic effects. This is essentially a rehash of an age-old concept in behavior genetics – namely evocative gene-environment correlation, and is as such neither a new argument, nor something that the field hasn’t already debated for decades. The issue with Burt’s argument is that it a) is incompatible with a standard counterfactual understanding of causality and b) at a deeper level just makes no ontological sense.

The counterfactual understanding of causality simply means that we ask the question: what would have happened in the absence of X? Would a given outcome Y look different, on average? If yes, then X is a cause of Y. This is what makes it possible to say that pushing a light switch causes the light to go on (in the counterfactual universe where I didn’t push the button, the light would stay off), but also to reason about the relationship between larger and more abstract phenomena like democracy and growth (democracy is a cause of economic growth if, in the absence of democracy, growth would have been lower, on average) – or indeed between genes and complex social traits like personality or behavior (a given genetic factor is a cause of a behavior if the absence of that genetic factor would lead to a lower propensity for that behavior). This makes the question of establishing the existence of causality fundamentally different from understanding the causal pathway — i.e. what is the cascade of events, or mechanisms, that leads from our cause X to our outcome Y?

Genetic factors are causes of differences in human behavior not because they have a causal pathway that is internal to the physical, biological being of an individual (i.e. from gene to protein to neuron to behavior), but because in the absence of a given genetic factor, behavior would be different, on average. The mechanisms could be a wild plethora of things (and it is, of course, often of great interest to map them out as best we can) – but it is the counterfactual outcome being different that makes it a cause.

In the same way, an environmental factor is a cause of behavior not because all of its subsequent mechanisms are also environmental in nature, but because its absence would lead to an expected difference in behavior. To see why the argument about “downward causation as confounding” is ontologically untenable, thus, we simply have to look at its mirror image. Burt makes the case that if there are environmental layers in the causal pathway between gene and behavior, then it is merely the environment masquerading as genetics. But this argument not only makes genetic effects virtually impossible, it also makes environmental effects even more so. If you were intellectually honest, you would then also have to accept that all environmental effects on human behavior are biologically confounded, since they are by complete necessity mediated by the brain and body of individuals carrying out the resulting actions. All causal chains leading to human action have, at the most proximal spot in the pathway, biology – neurons firing, causing the contraction of various tissues, etc, ultimately forming a behavior. If we are to take the argument about downward causation seriously, all of the social sciences are, ultimately, merely biological effects masquerading as environmental effects.

The fundamental fallacy here, in my view, is to think of the universe as somehow really consisting of discrete categories of phenomena in any meaningful way other than as poor heuristics for our theories. This is not so – reality is continuous and unitary. The mistake lies in thinking that it is generally meaningful to distinguish between purely genetic vs. purely environmental processes at all. Reality is an overwhelmingly immense spiderweb of causes, with a myriad downstream consequences. Whichever node in this mess you choose to prod – whether it be a genetic variant or a type of political system – the resulting ripples are not going to stay neatly within the confines of any crude conception of “environment” vs. “biology”. The sooner we come to grips with this, the better. Rather than arguing about the importance of our favorite bin in this false overarching dichotomy, we would be much better off just pulling on as many nodes in the web as we can think of, and seeing what happens. This is, ultimately, how we can learn the real structure of our shared social reality.