Correlation, causation and the American racial divide

Too much of the social justice discourse fixates on correlations and fails to arrive at policies that address the cause of social injustice. 

A recent example is an article by Ta-Nehisi Coates in The Atlantic that rattles off a bunch of correlations about blackness in America and then advocates for slavery reparations, free college tuition, a single-payer healthcare system and affirmative action. There is very little any of these policies will achieve for the racial divide in America; that much is clear to anyone who understands causation.

To say that something is correlated with something implies that they move together. There is a consistent relationship or association. For example, having a college education is correlated with higher earnings than having only a high school diploma. Correlations are statistical artefacts derived from averages. A consequence of this is that you can typically express them in terms of a likelihood. For example, there is a 70% chance that a black male in America without a high school diploma will end up in jail by his mid-30s. This is because there is a strong correlation between each of being black, being male and being a high school drop-out and being in jail before you’re 30. Put all of those together and you get a huge number.

However, one of the dangers of correlations is that they might be spurious. For example, there is a 99.26% correlation between the divorce rate in Maine and per capita consumption of Margarine. It’s spurious because margarine consumption does not cause divorce.

When we put too much stock in correlation without actually determining whether there is a causal relationship we can often end up misleading ourselves. For example, an influential study by Catalyst found that having more women on corporate boards increased the profits of those firms. They had a footnote that read: ‘correlation does not imply causation’. Recent work by Renee Adams has demonstrated that taking fixed effects or simply controlling for firm size causes this effect to disappear. The gender of the board members is not what is causing the difference in financial performance. The causality lies somewhere else.

Establishing causality is very difficult. The gold standard is an experimental design. This involves establishing a control group and a test group and administering a treatment to the test group. You could, for example, randomly allocate women to the boards of some firms and then measure the impact over time. Such practices are relatively straight forward in the laboratories of the hard sciences, but much more difficult in the social sciences. Few firms, for example, would consent to simply having some women randomly placed on their boards to test the effects.

There are numerous other techniques, like instrumental variables, panel data models, natural experiments (like the Norwegian quota for women on boards) and synthetic control methods, though these all have their shortcomings. One of the most pronounced is that they all require a good grasp of mathematics, something that most stakeholders do not possess or care to acknowledge. Nonetheless, without this painstaking work you can’t be sure you’re targeting the root cause of a problem. You may just completely waste your time and money.

Here we return to the racial divide in America. There may be moral arguments for reparations payments, but these certainly will not fix the outcomes gap between blacks and whites, they will merely breed welfare dependence. Australia’s experiences with Aboriginal welfare make that abundantly clear. Moving from the minimum wage grind in the ghetto to government handouts in the ghetto is hardly an empowering result.

Affirmative action is a rather bizarre call given that America has had substantial affirmative action for decades with only limited results. Statistical analysis suggests that while some discrimination remains in the American labour market, the reason why blacks are not getting jobs at the rate of whites isn’t prejudice but a skills deficit. There is nothing prejudicial about ignoring the colour of someone’s skin and hiring the better qualified candidate. As whites are better qualified on qualified on average, firms higher more whites on average. This correlation isn’t racist. For firms to do otherwise would be financial suicide.

The answer to this skills gap is not free college tuition because black students aren’t qualifying for college at the rates of whites in the first place. The academic achievement gap starts much earlier.

Here we come to one of the few bright spots in this area of research, which is Heckman’s work on early childhood interventions. This body of research suggests that the racial skills gap, on average, starts before children even reach school. A range of factors, notably terrible wages and absent fathers, means that disadvantaged single mothers are unable to inculcate soft skills like delaying gratification, sitting quietly, interacting peacefully with other children and the like, into their kids prior to them reaching school. They subsequently struggle to learn, feel marginalised by the system and become delinquent. Note that the emphasis is on soft skills, not employable skills, not cognition, not even numeracy or literacy. Soft skills are the prerequisites to learning those other skills.  

The solution is a targeted intervention based off the Perry Pre-school program that specifically targets relevant households to address this soft skills gap. This will get at the cause not just the correlation. Note that many disadvantaged households across all races could benefit from this program, so it needs to be targeted not on the basis of race. This runs contrary to what Nehisi advocates (racially targeted policies), but is the morally correct option on any criterion (Rawlsian, Kantian, Utilitarian) other than pure identity politics. Moreover, if the Australian experience is anything to go by, then having racially targeted policies would inevitably run afoul of its own set of complaints about racial profiling, so a sophisticated class-based targeting system seems the most politically viable.

Early childhood interventions are a very promising policy. Enormous effort went into the foundational research, causal analysis, design and promotion of this policy. You had several dozen researchers across multiple disciplines working on it. They used decades of formal theory, cutting edge statistical techniques and a 40-year study to demonstrate their conclusions beyond reasonable doubt. This kind of thing will actually help end entrenched disadvantage and close gaps in minority outcomes, but it requires stacks more effort and allows for far less moral indignation and self-flagellation than cartoons about white privilege.

Killer Mike was on Colbert the other night making some good points about addressing the racial divide in America. One thing he said was that the themes that are being discussed now were being discussed in the 90s, the 80s, the 70s and the 60s, and yet they persist. But his solution was vote for Sanders because Sanders has good intentions. This is misguided.   

These problems have persisted not because of a lack of will on the part of America’s largely well-meaning citizens but because the causes behind the problem are extremely difficult to pin down and very difficult to draft solutions for that are effective and politically palatable. As Nehisi points out, ‘something, something socialism and then a miracle happens’ is not going to work.

Social justice advocates are right to emphasise the injustice of the racial divide in America but they are unconstructive in their readiness to explain such phenomena as the product of human evil rather than human ignorance. Progress will require empirically verified causal explanations and sophisticated policy responses, something critical theoretical excoriations of unconscious white supremacists and a handful of correlations will never be able to provide.