Dysgenics: Who Called?

Another nothingburger that has been the basis of justification for white supremacist ideology and policy programmes is the alleged phenomenon of “dysgenics”. According to these advocates, the current social structure produces fertility differentials between individuals of different classes (assumed to hold different genetic intelligence) and individuals of different levels of intelligence, leading to a form of directional selection (Woodley of Menie et. al 2016). As such, they argue eugenic policies that ameliorate this alleged problem are justified in the face of such a problematic decline like embryo selection (Lynn 2008), population control (Jackson 1993), changes in immigration policy (Richwine 2009; Chauncey Tinker 2016bWoodley & Dunkel 2015), gene editing (Gyngell & Savulescu 2015), the elimination of welfare (Murray & Herrnstein 1994; Chauncey Tinker 2016b) and fascist birth control policies (Chauncey Tinker 2016a).

But is there actually any basis for the purported decline? Some think so. For example, Emil Kirkegaard held up a recent paper published in Nature by a set of geneticists (Abdellaoui et. al 2019). The paper is controversial for many reasons, but one of the core results is worth looking at. The paper reports:

The standardized effect size of birth year outside of coal mining regions was −0.001 (that is, 0.001 s.d. of the polygenic score decrease per birth year; P=6.8×10−6). Within coal mining regions, the effect size was −0.003 (P<2×10−16) across the entire birth year range and −0.004 after 1945 (P<2×10−16). Across the entire country, the effect of birth year was −0.002 (P<2×10−16; Fig. 5a)

The appendix reports that they normalized their values so the standard deviation of the polygenic score was 1, making our job easier. Let’s do some math.

The important part here is the observation that the effect of birth year (i.e. each year, the amount that PGS for children born that year attain):

the effect of birth year was −0.002

In other words, each year, there was a decline in educational attainment PGS of 0.002. But what does this actually mean in terms of the phenotype? Well, let’s calculate. An increase in the PGS score for EA by 1 standard deviation is associated with an increase in actual educational attainment by about 0.3 standard deviations [1]. So using a simple bit of math:

0.002\text{ SD PGS }\cdot \frac{0.3\text{ SD EA}}{1\text{ SD PGS}}=0.0006\text{ SD EA}

Now we need to convert the standard deviations of educational attainment into actual years of education. The UKBioBank reports that the standard deviation of education in their sample is 2.33844 [2]. Now the final bit of math:

0.0006\text{ SD EA }\cdot \frac{2.33844 \text{ years schooling}}{1\text{ SD EA}}=0.001403\text{ years schooling}.

In essence, the actual figure for the number of genotypic years of schooling lost each year is 0.001403. That’s a lot of zeros! In terms of days of school, we don’t even get one:

0.001403\text{ years schooling }\cdot\frac{180\text{ days}}{1\text{ year of school}}=0.2525\text{ days of school}.

In terms of hours, this is

0.25254\text{ days of school}\cdot\frac{8\text{ hours of school}}{1\text{ day of school}}=2.02032\text{ hours of school}.

Each year, the population’s genotypic value for education decreases by… two and a half hours of school. Guess it’s time to get out CRISPR, right?

Even in terms of longer periods of times, the figure is not worrisome. A generation (~30 years) would lose 7.58 days of school [3]. Even over the course of a century, the country’s average educational attainment would only decrease by ~25 days. Are we supposed to believe there are going to be horrendous effects of not having another month of school? Even though we have observed massive increases in schooling during the same period [4]?

The Other “Evidence”

Despite the most recent study on the question showing an absolute lack of “dysgenics”, hereditarians still posit that this is a worrisome trend in need of immediate remedy (Dutton & Woodley of Menie 2018). What is the evidence that they posit in favor of this view. There are typically two “lines” of evidence. The first is inferring dysgenics from differential fertility by IQ classes, educational attainment or socioeconomic strata [5] (Lynn 1999; Lynn & Van Court 2004; Meisenberg 2010Shatz 2008). The second revolves around the observation that the Flynn effect (the secular increase in IQ in most countries) has reversed in some nations [6] (Dutton & Lynn 2012, 2015Dutton et. al 2016; Pietschnig & Gittler 2015Teasdale & Owen 2008), allegedly a result of dysgenic fertility finally taking over environmental gains in IQ. Despite the rapid growth in these literatures, both inferences are entirely unjustified.

Differential Fertility

Differential fertility has long been a concern of eugenicists (Egendorf 2010), making some posit all sorts of horrendous policies. There are several assumptions underlying the inference from differential fertility to dysgenics:

  1. The assumption that the correlation between the phenotype (IQ, EA) and fertility is on the genetic scale. It is entirely possible that the genetic and phenotypic correlations with fertility for a trait can differ as a result of environmental factors (Morrissey et. al 2012).
  2. The assumption that IQ strata subpopulations are essentially closed, which is an unjustified assumption (Preston & Campbell 1993). It turns out that fertility differentials are compatible with an intelligence equilibrium, and changes in population IQ occur when the rates of differential fertility change.

Anti-Flynn Effects

Amid claims that the world is ‘getting better’ (Pinker 2018), it seems only natural to presume that if a decrease in IQ is being observed, it has to be the result of non-environmental factors: e.g. genetics. This evidence for the proposition that the world is ‘getting better’ is not easily justified [7] (Hickel 2019), and there is actually robust evidence that both the Flynn effect and its reversal are indeed environmentally caused (Bratsberg & Rogeberg 2018).

What About The Effect?

If we are observing some level of dysgenesis, shouldn’t we worry? There are several reasons for skepticism about the existence and magnitude of the effect observed in the paper. The first was already noted by the authors:

It is also consistent with year-of-birth-associated ascertainment bias, such that the oldest participants in the UK Biobank were more selected on traits associated with EA (for example, health and longevity) than the youngest participants

It is well-known that individuals with higher levels of education live longer, and subsequently those with less education live shorter lives (Davies et. al 2019). As a result, individuals farther in the past with lower polygenic scores for education will die earlier and not be ascertained in the sample, creating a continuous bias for higher polygenic scores in the past, and larger ones more recently. This inflates the decline over time, and could potentially even entirely account for it.

There are several other issues with the metric of polygenic score that the authors employed. It has been known that using PGS within-family decreases the magnitude of the association significantly (Lee et. al 2018; Selzam et. al 2019). Moreover, there have been documented environmentally mediated genetic effects on education (Cheeseman et. al 2019Willoughby et. al 2019). Different types of gene-environmental correlation, including passive (Krapohl et. al 2017) and active gene-environment correlation (Beam & Turkheimer 2013; Beam et. al 2015; Briley & Tucker-Drob 2013; de Kort et. al 2014) can cause environmental effects to be captured by genes, when in fact the effect size is partially environmental [8].

Finally, known issues with population stratification and cryptic relatedness could potentially play a role in inflating the effect [9] (Haworth et. al 2019; Holland et. al 2017Lawson et. al 2019Richardson & Jones 2019; Shadrin et. al 2019).

Conclusion

What does the failure of dysgenic hypotheses to born out by actual data tell us about hereditarian scientific hypotheses? It gives us strong reason to be skeptical about the allegedly robust “meta-analyses” they collect [10] (te Nijenhuis et. al 2015; te Nijenhuis & Van de Hoek 2016), and the “methods” of “inference” they employ to conclude “genes” are the cause of racial gaps in IQ scores [11] (Lasker et. al 2019). It also gives us an insight into the way that white supremacists interpret scientific evidence as well as the way that science functions in legitimizing white supremacist ideology. It is clear that the rise of “academic” hereditarianism and the alleged evidence used to support it has sparked a psuedoscience movement from within the academy, one that must be stopped.

Footnotes

[1] Consider the fact that a generous estimate of the R^2 for educational attainment is ~11% (Lee et. al 2018), indicating a r=\sqrt{0.11}\approx 0.3.

[2] At first, I was unable to find adequate data for the standard deviation of the years of education in the United Kingdom, so I substituted with data from a very similar society: the United States. The General Social Survey has a large corpus of data, including on years of education completed, and a simple calculation will confirm a value for the standard deviation of about 3 (slight overestimate). The education systems of the UK and US seem to be very similar in terms of educational attainment, so we shouldn’t worry about large deviations from the true value of the SD.

[3] This can be calculated simply by multiplying the estimate number of days for a single year by the years in a generation; \frac{0.25254\text{ days of school}}{1\text{ year}}\cdot\frac{30\text{ years}}{1\text{ generation}}=\frac{7.5762\text{ days of school}}{1\text{ generation}}

[4] This could be explained by the co-occurrence model (Figueredo & Woodley 2013), of course, but the absolute quantity is not nearly enough. For instance, Woodley of Menie et. al (2015) predict a decrease in IQ of .262 points per decade, and given that educational attainment PGS predicts IQ just as well as intelligence PGS do, the observed decline should make us wonder whether the assumptions in the breeder’s equation apply (Pujol et. al 2018; Witting 2000), whether the heritability estimates are wrong (Richardson & Norgate 2005), or whether the estimates of differential fertility are wrong (Etikan et. al 2016Kruskall & Mosteller 1980Peterson et. al 2005), also see here.

[5] See [3].

[6] They seem to be occurring almost entirely in regions like Scandanavia (e.g. Denmark, Finland, Iceland, etc), and perhaps in France, Britain or Estonia.

[7] See also Fischer (2019), Jerven (2013)Reddy & Pogge (2002), Reddy & Pogge (2005), Reddy & Minoiu (2007)Reddy (2009), Reddy & Lahoti (2015), Sundaram & Chowdhury (2011)this series of posts by UnlearningEcon.

[8] Despite claims that gene-environment correlation can be subsumed under heritability (Sesardic 2005), gene-environment correlations in fact constitute a violation of the equal environment assumption (Carey 2019) wherein phenotype→environment (or P→E [see Beam & Turkheimer 2013; Beam et. al 2015Beam et. al 2016; Briley et. al 2018Dolan et. al 2014; de Kort et. al 2014], a.k.a. the smorgasbord model [Eaves et. al 1986; Eaves et. al 1990]) transmission—wherein early differences between individuals create feedback loops because of how those differences affect choices in environments (Scarr & McCartney 1983)—induces gene-environment correlations.

[9] It is worth noting that Holland et. al (2017) did not find an inflation for educational attainment, but did find one for intelligence, while Shadrin et. al (2019) found inflation for both. It is unclear why, however.

[10] See here.

[11] An article on why this paper gives us no information is forthcoming, but for the time being, read this thread.