Against the Mutational Load Hypothesis: Part 1
The premises.
Please note that this post is the first of three (part 2, part 3). These were written as a single piece, but I was forced to split them up due to issues with the Substack editor.
What has caused the massive leftwards shift seen in every Western country since 1945? This shift is almost monotonic, and if we allow for brief periods of reaction such as Nazi Germany, it arguably goes back to at least the French Revolution. The content of leftism changes wildly, but Cthulhu always swims left. What’s more, it has some eerie parallels with historic imperial declines (feminism, LGBT, dysgenics, mass immigration, declining ethnocentrism in the state forming peoples) that occurred under very different economic and technological conditions and ideological frameworks. There have many proposed explanations for this process, most of which have long since been debunked, but recently Joseph Bronski has begun promoting a new theory: the Mutational Load Hypothesis (MLH).
What is Mutational Load?
Mutational load is the accumulation of deleterious mutations1. As most mutations are neutral or harmful, this is a function of the number of recent mutations and the strength of purifying selection against them. In turn, these terms are mainly a function of paternal age and infant mortality. Mutational load impacts everything from IQ to lifespan to mental illness to facial asymmetry. In the modern era, increasing mutational load is mainly a function of decreased infant mortality, which means relaxed purifying selection and thus a higher mutation-selection equilibrium.
What is the Mutational Load Hypothesis?
The mutational load hypothesis (MLH) is composed of three parts:
Premise 1: Mutational load has increased since the Industrial Revolution as a result of technological and economic advances.
Premise 2: Mutational load causes leftism2.
Conclusion: Increasing mutational load is the principal cause of society’s leftward shift3.
Has Mutational Load Increased in the Modern Era?
Yes. While paternal age, somewhat surprisingly, did not increase across the board in the 20th century,4 infant mortality plummeted all over the world, with it crashing first and hardest in Northwestern Europe and Anglosphere settler colonies.
With this relaxation of purifying selection, theory would lead us to believe that mutational load would increase. And indeed, this is exactly what we find with direct measurements of known disease-causing alleles. So premise (1) of the MLH is correct.
Does Load Cause Leftism?
Possibly. To a rightist, the idea that leftists are deranged mutants certainly has some emotional appeal. And there’s some evidence for it. Leftists are more mentally ill, less attractive, less altruistic, more sexually deviant, and physically weaker. On the other hand, they are also slightly smarter, so not every indicator of mutational load correlates with leftism. In general, outside of a handful of monogenic diseases, the correlation between mutational load and any trait is low, because broken gene variants are such a tiny fraction of the genome (which itself explains only a portion of the variance), so trying to parse out the effects of mutational load on politics in this manner is not going to be reliable. Is there better evidence than just correlations?
The gold standard of mutational load effects is direct measurement, showing that rare gene variants affect the trait conditional on otherwise equal genetics. This exists for IQ, but for little else, and does not exist for political beliefs. The silver standard is paternal age effects. This is because paternal age is the primary determinant of intracohort load variation. If, conditional on the characteristics of the parents and known environmental confounders, children of older fathers display more of a trait, that is strong evidence that that trait is partially caused by genetic load. For instance, this has been shown to be the case for various mental illnesses.
The question of whether left-wing political behavior is literally a genetic disorder is understandably not one researchers are eager to investigate, so the evidence here is limited, but I was able to find two relevant investigations and construct two myself. First, we have Woodley of Menie et al’s finding that, when adjusted for various potential confounders5, paternal age predicts lower church attendance in the (mostly born in the 1970s) AddHealth survey, but not in the (mostly born in the 1930s) Wisconsin Longitudinal Study. They interpret this finding as evidence that mutational load reduces religious behavior when societal expectations of religiosity weaken. Unfortunately, this is not directly measuring the outcome we care about (political beliefs), merely something loosely related to it, the effect size is tiny and (as the authors note) possibly due to uncontrolled confounders (the AddHealth survey does not have the same composition as the WLS; an effect existing in AddHealth but not in the WLS could reflect cohort differences or it could reflect composition differences).
Both the Wisconsin Longitudinal Study and AddHealth also recorded respondent’s self-reported political ideology (on a scale of very liberal to very conservative). Woodley of Menie et al control for this in their analysis on religious behavior, but they do not report the effects of paternal age on self-reported ideology, so I downloaded the public versions of both of these datasets to check this question myself.
Without controls, there is a weak, but real correlation between political ideology and paternal age (N = 9155, r = -0.029, p = 0.0055) in the WLS.
However, this result is not robust. I originally intended to control for the same set of known confounders as Woodley of Menie et al do in their paper (see footnote 4 for a list), but simply removing entries which have no information for one of these confounders (without controlling for anything) eliminates the result entirely (N = 5690, r = -0.009, p = 0.466).
Running an ordinary-least squares regression shows no variance in political ideology explained by paternal age once IQ, sex, and paternal education are controlled for.
When analyzing the AddHealth dataset, I restricted analysis to whites so as to minimize racial confounding. As in the WLS, the uncontrolled data shows a very weak correlation between paternal age and left-wing political ideology, though in this case it is not statistically significant (N = 2327, r = 0.021, p = 0.308).
Controlling for respondent’s birth year, sex, maternal age, IQ, and parental education using a squared semi-partial correlation weakens the result even further (r2 = 1.84e-06, p = 0.687), and performing an ordinary least squares regression finds paternal age explaining ~none of the variance once birth year, sex, and IQ are controlled for.
These null results are not definitive, for two reasons. First, while self-identified ideology is closer to what we care about than religious behavior, it is still not the outcome of interest, which are political beliefs. These are correlated, but not the same thing. Second, as Woodley of Menie et al point out, paternal age effects are very small and thus sensitive to both sample size and unmeasured confounding. I may have failed to control for something that would shift the result, or the sample size may simply be too small to see a true effect.
The final directly-relevant study of the effects of paternal age on politics is by Joseph Bronski. This is the single strongest piece of evidence in either direction, because it was directly constructed to check this.
He finds a paternal age effect in that children of fathers above 35 are noticeably woker than children of fathers below 35. He also finds that this holds true no matter the year in which the father was born (though with overlap in confidence intervals due to very small sample size per father birth year), so this is not due to older fathers simply being born earlier and thus less left wing. Bronski’s wokeness metric is constructed off of actual beliefs rather than simple self-reported ideology (he asks if BLM, LGBT, and feminism are good), and so is a closer measure to what we’re actually looking for. He coded those who answered “yes” to all three of his questions as leftists, and found that they had fathers about a year older, on average, then non-leftists6.
While he did not measure the political beliefs of the parents of his sample, he did survey a separate set of American fathers of the appropriate generation, and found that those who had children above the age of 35 were no more left wing then those who didn’t. Furthermore, he asked these fathers whether or not their wives were left wing using the same questions, and found that older fathers did not believe that their wives were more left wing.
There are a few obvious confounders and problems. First, he asked fathers about their wives. Marriage is increasingly a right-wing and upper-middle-class lifestyle choice, and single mothers are much more left wing than married mothers. Furthermore, married women have similar politics to married men, but unmarried women are far to the left of unmarried men. As such, polling only married men will systematically underestimate how left-wing mothers are, even if the measurement of the mother’s political beliefs is accurate. Second, his metric of mother’s political beliefs is not very accurate, as it is based on the father’s report, not on asking her directly, and so maternal beliefs may be quite confounded. Educated women are known to be more left wing and have children later, so this confound is likely to weaken the relationship.
Third, he does not adjust for known confounders such as IQ (associated with later birth and more leftism), parental education (same), maternal age (which can cause non-mutational load developmental problems7), or birth order (likewise - firstborn are systematically different). Fourth, due to data limitations, he is forced to binarize his paternal age variable rather than treat it as a continuous variable and he does the same when coding leftists vs non-leftists. Fifth, his sample of fathers is not necessarily representative of the actual fathers of his first sample, as his first sample is deliberately evenly split between self-identified liberals and conservatives and his sample of fathers is married. Sixth, his measurement of political beliefs fails measurement invariance across time, and so may not work across cohorts8.
Furthermore, the effect sizes are implausibly large. He finds that having a father over 35 as opposed to under 35 matters about as much as being born an entire generation (~25 years) later. While he doesn’t post the average age of the fathers in each bucket, the difference is probably 15 years or less. Given the small size of paternal age effects, this suggests his results are confounded.
Conclusion
While I would very much like to see Bronski’s study replicated with a larger sample size and the problems I mentioned addressed, in the absence of better evidence, we must go on what we have: one weak, loosely relevant positive result, two more relevant null results, and one highly relevant medium-strength positive result. The evidence is not definitive, but it weakly supports the claim that leftism is partially caused by mutational load. For the sake of testing this hypothesis, let’s assume the relevant evidence has been collected and we know for certain that mutational load causes leftism.
This differs slightly from the technical definition, which focuses on fitness. In this case, I’m referring to de novo mutations that degrade function (adaption execution) rather than fitness (number of surviving offspring).
For the purposes of this article, leftism is the decay of group-selected behavior. Feminism (particularly the decline of absolute monogamy and the loosening of sexual restraints on women), irreligiosity, lowered ethnocentrism, socialism, and promotion of sexual deviancy are all indicators of leftism. This definition is motivated by Chapter 7 of Modernity and Cultural Decline, and can be applied outside of the usual post-French Revolution taxonomy of leftism to premodern cases.
To be precise, this post is aimed at this version of the hypothesis, as represented by the quote: “Both the elite and masses suck, and Cthulhu swims left because mutational load is accumulating at all socio-economic status levels. If this continues, we can predict that 20 years from now we will be even more woke, and it hardly matters how power flows.” Some of the evidence presented is also relevant to the much more nuanced SEAM model from Modernity and Cultural Decline, but as Woodley et al don’t claim mutational load is the largest factor, or give any quantitative estimate at all of its relative importance, some of it is not.
In the early 20th century, mothers began having children earlier but continued to have them later then became typical after the Baby Boom, with these two effects essentially canceling each other out.
Mother’s age at birth, birth year, sex, race, mother’s and father’s education, mother’s and father’s religiosity, mother’s and father’s income, IQ, education, income, political attitudes, and birth order
Note this contradicts WLS and AddHealth quite strongly.
But, crucially, these problems don’t build intergenerationally, unlike those caused by paternal age.
In other words, “Is LGBT good?” means something different to those who associate LGBT with gay marriage or the legalization of homosexuality then to those that associate it with Drag Queen Story Hour and rapid onset gender dysphoria. If we assume people form their beliefs on “LGBT” early on in their lives and mostly don’t deviate from them, this would cause systematic bias when comparing cohorts, since a member of an older cohort saying “yes” may mean something well to the right of a member of a younger cohort saying so. The same holds true for feminism, though probably not for BLM, as this is a very novel issue.
It was quite common in the 1800s for older men to marry much younger women and produce children with them. If paternal age were a guarantee of high mutational load, there would have been an epidemic of autism and down syndrome in the 19th century, but we find no evidence for this.
I think it's a mistake to focus on paternal age. The majority of mutations any of us have will be inherited from previous generations. The number of mutations that "matter" at all (e.g. to coding regions) that come from mutations during one man's lifetime will be very small, not much more than about 1 or 2, and most of these won't have any observable phenotypic effect.