Where’s The Beef?

Recently, lots of bloggers have been wringing their hands over the implications of genetically-determined differences in intelligence between various racial and ethnic groups. I think they’re getting a bit ahead of themselves – what’s been conspicuously absent from any of these posts is definitive evidence that the premise is correct.

Let’s start with some (of what I believe any fair-minded observer would agree to be) facts:

1. There are statistically significant differences in IQ test performance between self-identified racial and ethnic groups in the US, and these differences have been sustained over long periods of time. The specific difference that is most widely discussed is the fact that in the US Non-Hispanic whites score, on average, about 15 points (~1 STDEV) higher than African-Americans. (Leaving aside the complication that it matters exactly how we define “long periods of time”, since, for example, there is circumstantial evidence that the black-white IQ gap may have been reduced substantially over the past several decades.)

2. Self-identified racial groups in the US have statistically meaningful differences in genetic content. (Leaving aside the complication that these are fuzzy differences with huge overlaps and no firm boundaries – they are discovered by applying clustering algorithms – rather than clear divisions with sharp borders. These differences are more analogous to something like the distinction in personal lifestyle behaviors between those who self-identify as “conservative” vs. “libertarian” than they are analogous to the simple and clean distinction between the latitude and longitude of residential addresses for “residents of Ohio” vs. “residents of Pennsylvania”.)

3. Operationally, an IQ score measures the degree to which an individual has answered a specified list of questions in conformance with what the tester defines to be correct answers. It turns out that such IQ scores predict, with better-than-random accuracy, the scores on many other tests that ask questions that are intuitively related to aspects of what we mean in normal speech by intelligence. (They also predict, with better-than-random accuracy, measures of many other positive life outcomes, such as school performance, work performance and health.) There is, in a technical statistical sense, an underlying factor that is partially redundant across these tests. Let’s call this factor, I don’t know, g. (Leaving aside the complication that while a statistical “factor” is often an indicator of a common underlying cause, it does not necessarily imply that this must be the case.)

Given these facts, once could develop the hypothesis that genetically-determined differences in intelligence across races exist. But how would we demonstrate that this hypothesis is true? (At least in the scientific sense of “true”)

It seems to me that there are two fundamental, non-mutually-exclusive, potential lines of demonstration.

The first would be the more compelling: to specify the physical mechanism by which genes govern intelligence – what is often called the genotype-phenotype map – and then to demonstrate how this operates differently across races. Let me be more specific about what I mean by this. As a first step, scientists would need to develop the laboratory-replicable demonstration of the chain of biochemical processes that show how some combination of named loci on the human genome deterministically govern named, measurable brain functions that collectively comprise a common-sense, even if necessarily imperfect, definition of intelligence. As a second step, they would then have to demonstrate that the differences in statistical incidences of these genetic phenomena across racial groups accounted for differences in intelligence. At this point the debate would be over in the sense that the debate about Boyle’s Law is over.

There have been remarkable recent advances in understanding how genes might regulate brain function. But none of this is in the same ballpark as being able to accomplish even the first of the two steps that I have delineated – it’s not even in the same county. This is why the writers on this topic are always saying things like “strong preliminary evidence”, or prominently quoting others saying things like “Might it be fair also to say” , or “I can see it coming” or whatever.

I lack the ability to predict future scientific discoveries accurately. Give me a call when we’re able to use the past tense.

In the absence of a physical mechanism, we are left with the second means of demonstration: what I’ll call “econometric” analysis, in which we attempt to infer the presence of genetically-determined differences in innate intelligence between racial groups by trying to hold non-genetic impacts equal and assign the remaining difference to genes. This analysis normally proceeds in two steps, roughly analogous to the two steps I have defined for physical demonstration of the hypothesis. The first is to attempt to show non-trivial implied heritability of IQ at the individual level, and the second is to attempt to show that differences in IQ test performance across races can not be explained completely by non-genetic differences between races, and therefore to assign the remaining difference to genetic differences, given the plausibility of such a cause as per the demonstration of the heritability of IQ in the first step.

When attempting to accomplish the first step, we can start by recognizing that, tautologically, all differences in IQ test performance between any two individuals or groups must be explained by the combination of genetic and environmental differences between them, as long as we define everything other than genetics as “environment”, and recognize that genes and environments interact. The trick, of course, is that these interactions are so hard to disentangle that we must resort to “natural experiments” that hold many other factors (approximately) constant to try get limited insights into the relative importance of genetic vs. environmental factors. Twin studies and adoption studies which hold genetic and environmental factors (approximately) constant are the central means of such analysis. The key estimate developed from these studies is “heritability” of IQ. Heritability of 0.0 implies that 0% of individual IQ variation within a given population is genetically determined, and heritability of 1.0 implies that 100% of individual IQ variation within that population is genetically determined.

While there are problems with each research design, scholars have come to the consensus viewpoint that individual heritability of IQ is about 0.5 in childhood, and (somewhat counter-intuitively) rises to more like 0.75 by late adolescence. For our purposes, it is sufficient to conclude that heritability is meaningfully greater that 0.0 and meaningfully less than 1.0.

We then proceed to the second step: demonstrating that systematic environmental differences between races can not account for differences in IQ performance. We can see the fundamental difficulty with distinguishing between correlation and causality in such analysis by first considering an idealized case. Suppose that exactly one negative environmental driver of IQ performance impacts everyone in group A equally, and does not impact anyone in group B at all. In this case, we would observe high heritability of IQ at the individual level within both groups A & B as well as lower average IQ performance for group A. We would have no mathematically valid way to separate the impacts of genetics vs. environment on the differential performance of the two groups unless we had some independent estimate for the impact of the adverse environmental impact (that is, unless we already knew the answer). Precisely because many such environmental factors are theorized to operate explicitly at the group level (e.g., racism, culture, etc.), it means that we will have an inherent problem of confounded variables.

Further, if you think about the earlier observation that there are huge overlaps in the genetic make-up of the individuals who comprise various self-identified racial groups, any actual underlying differences that are driven by specific combinations of genes that regulate intelligence will be correlated with group membership only (very) imperfectly. So when we evaluate differences in intelligence by self-identified racial group, we are, in effect, adding a huge amount of noise to the measurement of any true underlying effect.

It’s important to keep both of these observations in mind when considering econometric arguments. In combination, I believe that they are the key reasons that scholars have been able to get to consensus on the existence of a genetic component to IQ differences at the individual level within groups, but not at the level of group-to-group differences.

Historically, researchers first began investigating the potential genetic basis of racial disparities in IQ scores by evaluating whether “degrees” of racial membership corresponded with degrees of IQ difference. This work led to no compelling results. In more recent decades, researchers have analyzed various natural experiments relevant to this question. The most famous of these is the Minnesota Transracial Adoption Study in which various high-IQ white parents adopted biologically black, white and mixed race children. In theory, this should allow us to isolate the genetic influence on intelligence by evaluating the IQs of each group of children after they have all been raised in (approximately) equal environments. In fact, as opposing interpretations (pivoting on the potential confounding of age-of-adoption with racial group) demonstrate, it actually provides a good illustration of why it is so difficult to segregate genetic from environmental effects accurately by racial group – no natural experiment is sufficiently controlled to do this. Given our current datasets and analytical tools, when we use econometric methods to try to understand the causes of group differences in intelligence, we are like cavemen trying to figure out how a computer works by poking at it with sharpened sticks.

Do genetic differences accounts for any material portion of the difference in IQ scores by self-identified racial groups in the US? The only honest answer is that we don’t know. This, not political correctness is why the American Psychological Association’s formal consensus point of view on this question is stated without qualification: “At present, this question has no scientific answer.”

The chattering classes should stop spinning Eloi-and-Morlock fantasies, and come back to this issue if and when it ever becomes real. Resolution will almost certainly require advances in understanding of the physical mechanism of intelligence, not more clever econometric analysis. There is only one thing that we can be confident about what these new scientific advances will reveal: surprises.