Against the Mutational Load Hypothesis: Part 2

The premises don't imply the conclusion.

Please note that due to issues with the Substack editor, I was forced to split a larger post into three pieces (part 1, part 3).

Is Mutational Load a Plausible Cause for the Leftwards Shift?

So mutational load has increased, and mutational load causes leftism. Ergo, genetic leftism has increased over time as a result of increasing load, making society more left wing. Makes sense, but the next question is “is the effect size of the correct order.” To test this, we need to have a quantitative estimate of societal leftwards shift as well as an estimate of the effects of load on leftism.

A Quantitative Estimate of Leftwards Shift

I used the General Social Survey (GSS) to estimate how far left American society has shifted over the past 70 years. I removed all respondents who were not born in the United States or were not non-Hispanic whites, so as to eliminate racial and minimize immigration confounding. I used a selection of political attitudes on a variety of positions to create a crude index allowing for a quantitative estimate of American political shift over time. 

The selected GSS variables. A 1 in the dictionary means the highest value is the rightmost, and a -1 means the lowest value is the rightmost.

I selected these positions for three reasons:

1. They were asked for at least 30 years. Some politically-relevant questions, like those on most LGBT issues, only began to be asked recently, and many more were only asked for a few years. This means that the index tends to underestimate shift over time, because things that fall out of the right side of the Overton window, such as laws banning interracial marriage, and things that enter the Overton window from the left, such as gay marriage, are not included. It should be understood as a lower bound.

2. They measure the same belief over time. The answers to many relevant questions have different meaning across time. For instance, a question asking if immigration should be raised, lowered, or kept the same means something very different if immigration is 200,000 a year vs 3 million. For most of these, the literal meaning of any answer to the question has shifted left to a degree that is noticeable but difficult to determine, so I elected to exclude them.

3. They are clearly right- or left-coded. Some questions are about idiosyncratic American or foreign policy issues that map onto the Republican/Democrat divide when the question was asked, but do not neatly map onto left vs right.

I then rescaled all of these variables so that the most right-wing response mapped to 1 and the most left-wing response mapped to 0, with intermediate responses in between at the appropriate intervals.

Code for reproduction purposes.

And simply averaged the values of each of these rescaled columns to get a right-wing index, where 0 is maximally left-wing on all of these variables, and 1 is maximally right-wing. The use of a simple average was chosen for simplicity, not for theoretical reasons. I could have tried weighting each issue, but this would have been ad hoc and thus introduced some of my own biases.

Code for reproduction purposes.

The correlation between this index and self-identified political ideology. Note that it rises over time, as expected. In an environment in which everyone is uncomfortable with blacks marrying their neighbor, that question tells you little about their politics. In an environment like today, that’s a sign of extreme rightism.

As expected, we see a nearly monotonic, linear decline in right-wing beliefs by birth year. 

Some of the variance at the beginning is because 19th century politics doesn’t map all that well onto 20th century, some of it is due to low sample size, and some of it due to the fact that many of the index questions only began to be asked in 1975, so the first three years of collection are dominated by a small subset of them. The year is birth year, not collection year.

Note the slope is much steeper than that of American self-reported political ideology by birth year; white Americans have gotten left-wing much faster than we have thought of ourselves as left-wing. This is intuitively reasonable, as left-wing is a relative judgment, the meaning of which shifts left over time.

This is self-identified ideology. It still moves left over time, but it extremely jumpy and moves left much more slowly then beliefs. Also shows Boomer liberalism is real, but mostly a product of self-identification rather than political beliefs.

When we compare the mean political beliefs of Americans born in the 1920s vs those born in the 1990s, we find a 1.15 standard deviation difference in right-wing beliefs. 

I chose the 1920s and 1990s for sample size reasons; sample sizes before and after are much smaller.

If we assume that right-wing beliefs are approximately normally distributed, we can estimate that the median white American born in the 1920s would be at the ~88th percentile of right-wing beliefs among those born in the 1990s. This is intuitively reasonable, and due to aforementioned underestimate, is actually a lower bound.

Right-wing beliefs are approximately normally distributed. You can also see that if the right tail is more cut off then the left tail, perhaps a better measure could discriminate between the far-right and the extremely far-right.

So we can estimate that controlling for immigration and racial composition, America shifted at least 1.15 standard deviations to the left over 70 years starting around World War II (when those born in the 1920s started reaching political maturity). 

Population-Wide Effects of Load on Intelligence

In a perfect world, we would have a good estimate of the relationship between mutational load and politics (ie x additional de novo mutations produces, on average, y shift to the left) as well as a good estimate of how much mutational load has risen since the Industrial Revolution. If those existed, we could simply calculate the approximate leftwards shift predicted by accumulated mutational load and see how closely it matches to the observed political shift. Unfortunately, neither of those exist. We know mutational load has risen since the Industrial Revolution, but not how much¹, and it isn’t even known for certain if there is a relationship between load and leftism, much less the precise magnitude. So to get an estimate of how large we could reasonably expect load effects on politics to be, it makes sense to look at a different, better-studied trait. If mutational load has caused a sizable decline in a trait we know it affects, this makes it more plausible that it has a similar effect in politics. If it hasn’t, this makes a major effect on politics implausible, even if the leftism-load relationship is real. 

To be a good proxy, the trait needs to have the following properties:

1. It must be known to be directly affected by mutational load. This excludes traits such as height or handedness.

2. It should be a continuous variable like politics, not a binary one. This excludes almost all mental illness diagnosis rates, as well as monogenic genetic disorders.

3. It should be a mental, rather than physical, trait so as to better approximate politics.

4. It should not have large, known environmental confounders that could change the trend over time, since the MLH posits that mutational load is much more important than any environmental factor.

5. It should have a heritability equal or higher than that of political beliefs (which have a heritability of around 0.4).

Fortunately, the best-studied construct in psychology, intelligence, possesses these properties². Note that by intelligence, I’m referring to g rather than IQ scores, as these are confounded by the Flynn effect, which is not on g. Once the Flynn effect is taken into account, intelligence has been approximately constant in the West since rigorous measurements began in the early 20th century. So is there any room for a mutational load induced decline?

IQ declines per decade in countries where IQ declines have been detected.

Beginning in the 1990s, IQ scores have declined in some countries by around 1.7 points per decade. This is quite a lot; if this decline were on g, and it was mostly the result of mutational load, it would be of the correct magnitude for the political shift (around 0.8 standard deviations over the course of 70 years). But neither of these things is true. The two extant meta-analyses on the subject find the following [1][2]:

1. IQ declines, like the Flynn effect, are mostly not on g³. In other words, we are getting worse at IQ tests faster than we are getting dumber.

2. IQ decline is positively related to immigration, and more g-loaded declines are correlated with higher immigration. Given immigrants to almost all Western countries have lower IQ than the natives, this is unsurprising.

3. IQ decline is negatively related to maternal age (used as a proxy for paternal age in the absence of better data).

4. There is no correlation between IQ declines and “Index of Biological State,” a measure of both relaxed selection (mutational load) and selection against intelligence.

Maternal age is a good proxy for paternal age. Maternal age is strongly negatively correlated with IQ decline. Ergo, paternal age, and thus increasing rates of load accumulation, is probably not a factor in IQ decline.

A table showing the correlations between various proposed causes of IQ decline and IQ decline by country. Note that this table underestimates the impact of Third World immigration, because “immigration” is simply net immigration, which includes intra-European immigration (very high in much of the EU). The important parts of this chart are the fact that g-loading is negatively correlated with IQ decline, immigration is positively correlated with IQ decline, and IBS, which measures both load and selection, has almost no correlation, suggesting load doesn’t play a major role in IQ declines.

In other words, IQ declines are mostly not intelligence declines, and to the extent they are, this is largely the result of Third World immigration. There is no detectable positive relationship between mutational load and IQ declines, and possibly a negative one⁴. The effect of mutational load accumulation on intelligence is undetectable and probably ~0, certainly nowhere remotely near the magnitude of the political shift over the same timeframe. Given mutational load accumulation’s lack of a major effect on a trait it is known to affect, it having such a massive effect on politics is implausible⁵.

A table showing the biggest cause of intelligence declines in the West: immigrants, even 2nd generation immigrants benefitting from a First World environment and education, are dumber then the natives in every Western country except Australia. This is what you’d expect, since intelligence is the biggest determinant of national wealth in the modern world, and people move from poor countries to rich ones.

Selection for Right-Wing Beliefs in the United States and Europe

The MLH is a subset of genetic hypotheses of leftism. As such, evidence against genetic hypotheses of leftism in general also apply to the MLH. We’ve seen that mutational load has not caused major changes in intelligence (and thus is unlikely to have caused major changes to politics), but this comparison actually understates the case against the MLH. When it comes to intelligence, load accumulation is working in the same direction as selection: dumber people have more children ([1][2][3]) and accumulating mutational load makes people dumber. And, as expected, polygenic scores for intelligence have fallen (at a rate equivalent to around 1 IQ point per generation in 20th century Iceland).

From the British Biobank. This shows a variety traits being selected for. We are getting genetically fatter, sicker, more impulsive, and dumber.

But the reverse is true for politics. The MLH posits that society-wide leftwards shifts are the result of a genetic shift leftwards caused by accumulating mutational load. If there has been selection for right-wing views, however, the effects of mutational load would have to be large enough to not only explain the leftwards shift itself, but also to overcome selection. This is difficult, because, when it exists, directional selection is generally more powerful than load unless mutation rates are very high. And there has been selection for right-wing views in both the United States and Europe (charts are from the GSS unless otherwise noted, the ones that look like they were made using matplotlib were made by me, and only include non-immigrant, non-Hispanic whites to control for the changing demographic composition of the United States).

Rightists (by belief, rather than self-ID) have been having more children then leftists for a long time in the US, but note that the correlation get much bigger for those having children after the Baby Boom, probably because of the collapse of the broadly pro-natal, patriarchal, religious society of 1940s-1960s America.

But maybe we don’t care about the whole population. After all, people with political agency are a tiny, elite group. If smart leftists have more children than smart conservatives, perhaps this group has seen selection for leftism, even if the broader population has not. 

The same chart as above, but restricted to people who scored 9 or 10 on the WORDSUM, a simple 10 question vocabulary test that serves as a measure of IQ, particularly the verbal IQ more important in politics. Note the correlation is higher then the previous chart, and that there is not a single high IQ far-rightist with children born 1970-1979.

But this is also not true. In fact, the fertility difference among high IQ right- and left-wingers is even higher than that among the general population. This makes sense, since high-IQ people are more ideological and have more agency than low-IQ people.

Correlations between right-wing beliefs and children, controlling for birth year. Code provided for reproduction purposes.

Controlling for the confounder of birth year (those born earlier being both more conservative and more fertile), we can see that the correlation between children and right-wing beliefs is positive (N = 28410, p ~ 0, r = 0.077), and that this correlation is doubled among the high IQ (N = 2499, p ~ 0, r = 0.146).

Correlation between WORDSUM and number of children, controlling for birth year.

These correlations are of the same order as the correlation between fertility and intelligence (N = 14232, p ~ 0, r = -0.098).

The same is true in Europe. Right-wingers (as measured by attitudes towards gays and the importance of motherhood for women) in almost every European country outbreed left-wingers.

Table showing correlations between various variables and number of children and grandchildren. Rightists have significantly more of both.

Right-wingers have more children and grandchildren than left-wingers in Europe, and the proportion of the population with right-wing views on these two issues have increased in most European countries (though this is somewhat confounded by Third World immigration).

Chart that shows the fraction of rightists on these issues across generations. It’s increased in almost every European country. Cthulu doesn’t swim left on everything, everywhere.

This finding - that right wing views on homosexuality and sex issues are increased by differential fertility - has been replicated in the US, where differential fertility has increased opposition to gay marriage by 7.9% and abortion by 6.8% 2004-2018 (compared to the counterfactual of no differential fertility).

Chart showing actual opposition to abortion and gay marriage vs the counterfactual in which there was no differential fertility.

So we can see that there has been selection for rightism with real demographic consequences across the First World, undermining genetic theories of the leftwards-shift.

Conclusions

Even granting that load has increased and affects political beliefs, the conclusion that load accumulation has caused the political shift does not hold. The magnitude of the leftwards shift is far larger than the measurable effects of load accumulation on a similar, but much better-studied and more heritable trait (~1.15 d vs ~0), and the credibility of any genetic hypothesis, including the MLH, is further undermined by selection for right-wing beliefs in both the US and Europe. But we don’t have to look solely at theory, analogies, and the case of the modern West to evaluate the MLH. 

(Part 3)

1. It is quite difficult to directly measure overall levels of load at all; it isn’t even known if different continental-scale racial groups have different levels of load, despite multiple measurement attempts.

2. The negative effects of load on intelligence have been shown with the highest quality evidence possible: direct genetic measurement, which shows that adults with rare gene variants have lower cognitive function conditional on identical polygenic scores for intelligence.

3. The non-g component of IQ doesn’t generalize well and is less important.

4. Given that declines are negatively correlated with maternal age, and that there is no correlation with a measure combining mutational load and negative selection (which by itself almost certainly have a positive relationship with decline). If this is the case, it’s due to confounding, as mutational load is known to be causal in lower IQ.

5. Not impossible. The load-politics link could, in theory, be many times stronger than the load-intelligence link. But there’s no reason to believe this is the case a priori; it needs to be proven. Null hypothesis is that it is weaker, given the lower heritability of political beliefs compared to intelligence.