Genes and Economic Ability

Income

Twin studies suggest that 85% (75%) of variability in American male (female) lifelong income is due to genetic variability, while ~0% is due to shared environment variability.

Chetty et al find intergenerational correlations of $r \approx 0.32$ for men and $r \approx 0.26$ for women Kline. However, this is biased downwards because (1) they separate out races and (2) while they measure parent and child income as the average over 5 and 2 years respectively, rather than lifetime income.

The first problem is likely minor since race doesn't account for more than a trivial amount of overall income inequality, but the latter is quite important.

For instance, the correlation between male Swedish fraternal twins' incomes increases from $r \approx 0.16$ to $r \approx 0.27$ if you change from a 1 year snapshot to a 20 year window Benjamin. Some naive math suggests removing this source of error would increase Chetty et al's estimates from 0.32 (0.26) for men (women) to 0.40 (0.32). This is more consistent with the twin estimates from the twin studies, which makes sense because I adjusted the twin studies for the same issue.

Another serious problem is that heritability generally increases with age, which makes sense since many people spend their 20s in various educational institutions, internships, or probationary roles - not to mention "finding themselves". On study found that adjusting for both those confounders suggests the correlation for men may be nearly 0.60 and that estimates of around 0.50 are common in the literature Mazumder. Ultimately, then, Chetty et al's study (despite its many virtues) yields a quite biased estimate of income mobility.

The main point here is that the idea that earnings are ~80% heritable is both implied by the twin studies and consistent with parent-child heritability estimates.

Social Status Mobility

The most common measure of income mobility is $b$.

The basic idea is that we (1) transform the income distribution to make it roughly standard normal, (2) perform linear regression between parental and child income, and (3) interpret the slope as $b$.

Typically the chosen transformation is logarithmic, which makes interpreting $b$ easy: every 1% increase in parental income is associated with a $b$ percent increase in child income. This is effectively (but not precisely) the model used by Chetty et al (referenced above). Naively, this model suggests an exponential decay regression to the mean income across generations.

However, this naive interpretation turns out to be false. To see how, consider this example. Suppose Alice grows up and becomes a doctor. Bob is equally gifted/privileged but chooses to become a professor. Alice will earn much more than Bob, but their social status is similar and it wouldn't be unreasonable to expect their children to have equal chances in life. Indeed, we'll see later that raw money is a relatively small way parents economically benefit their children.

More generally, there are ways parents affect their childrens' incomes besides their own income - connections, genes, upbringing, etc. These positive affects tend to go along with higher incomes and so the naively expected exponential decay doesn't really happen.

Attempts to estimate social status mobility typically assume status are normally distributed but that the mean and spread of status is different for people with some specific or collections of surnames. They then compare the proportion of people with those surnames who become doctors (or other high-status professions) relative to the proportion of the general population who do. These proportions are used to estimate how the mean status of the surname group changes over time. Estimates of decay in status (as measured in this way) typically find that $b$ is around 0.74 Clark.

These estimates are much higher than the correlations of simple income heritability found between parents and their children's in the first section. If you assume no assortative mating and no shared environment effects, $b=0.5$ is the highest possible value attainable. Given the lack of shared environment in predicting income, it seems likely that its effects here are small. Hence, the hypothesis most consistent with this data is that (a) social status is entirely genetic and (b) there is a lot of assortative mating.

Going forward, I'm going to use the term "ability" rather than status. Ability as in "ability to get and hold a high-status job", since that is really more what the above studies measure.

Merging the Two

Over many generations, we should expect the decay of status to equal the decay in lifetime income. Over one generation, though, they are clearly not equal. To my mind the obvious model is a hidden Markov model, where the hidden state is some notion of ability/status and the visible state is lifetime income:

This model implies

Based on the above studies, b=0.74 and the implied a is 0.77 (0.72) for men (women) in the US.

Why is the relationship between ability and income higher for men? I suspect the reason is primarily that men more uniformly choose to prioritize monetary compensation, whereas women are more heterogenous in their priorities, frequently attaching more weight to things like leisure and mothering.

Genes

Recall that twin studies find lifetime income is almost entirely heritable and have virtually no shared environment component.

Suppose ability was 100% heritable. Then, in the absence of assortative mating, we'd expect a child's ability to be 0.707 times their parents' average ability plus noise. Given the existence of assortative mating by income Schwartz, the actual prediction is more like 0.74. Note, this perfectly matches our estimate for b.

The fact that shared environment has ~0 effect on lifetime income strongly suggests it has ~0 effect on ability. The only other possibility is that it has an effect on ability and an equal but opposite effect on lifetime-income-conditioned-on-ability; this alternative seems implausible. For these reasons, I think we have good evidence that ability is ~100% genetic.

As the twin studies already showed, however, lifetime income has significant (~20%) of its variance attributable to unshared environment. The resolution is straightforward: how much people are able to earn is entirely genetic; how much they do depends on unshared environment. Most likely, imo, it depends mostly on raw luck and on how much they prioritize earnings versus other factors.