Resume IQ

Goal

As we've discussed before, you can estimate someone's IQ a variety of ways: particularly using test scores, and which groups they belong to. However, in the real world, IQ isn't all that matters. Instead, whether you are successful at achieving your educational or professional goals depends on a variety of factors. Here, we attempt to construct a single measure of success that is both highly heritable and highly predictive.

Two Traits - One Common Cause

Educational selectivity is mostly genetic, with very little common environment factors. It is also heavily g-loaded, particularly because of its reliance on standardized test scores.

Earnings is also mostly genetic. It is also fairly heavily g-loaded: both getting the job (because resumes are screened by your educational institution) and also for performing on the job.

Finally, the two factors above are both considered important by society at large and are typically easy to deduce from a resume. For all the above reasons, we'll focus on them here.

One natural question is whether a single "factor" explains both these accomplishments, with the obvious candidate being the g-factor. The place to start is to consider the correlation between someone's intelligence, income, and college quality.

Well, based on the math in our college rankings sequence, the correlation between SAT score and income is about r~0.2. Using Chetty et al's dataset and the known income distribution, we can compute the correlation between college selectivity (measured as average SAT score) and income at the student-level; it is r~0.12. Finally, the correlation between a student's SAT score and their college's average SAT score is about r~0.65.

An interesting observation here is that 0.65 * 0.2 = 0.13, which is awfully close to 0.12. This is consistent with the theory that SAT score explains virtually all the shared variance between college selectivity and income.

If we assume SAT correlates with g at r~0.9, then the above correlations become 0.22, 0.13, and 0.72, respectively. Note that 0.72*0.22 = 0.16, which is a little larger than 0.12, indicating there is probably more to the common factor than just intelligence, but that intelligence still accounts for a majority of the correlation.

This all suggests that, exempting intelligence, the factors that makes someone both earn more and the factors that make someone attend a selective college are generally quite different.

Finally, note that if g is 100% heritable and the only genetic cause of undergraduate quality, it predicts that undergraduate quality should be is 52% heritable (0.72^2). This isn't far from the empirically determined heritability (~62%), which suggests most of the heritability of undergraduate selectivity is actually driven by intelligence heritability.

However, the story is quite different for earnings. If intelligence was the sole cause of earning heritability, it would predict that earnings is 5% heritable (0.22^2). This is too low - by a factor of ~10x, which suggests that other genetic factors drive the heritability of earnings.

The tl;dr here is that intelligence is the main driver of which college you attend, a weak driver of your eventual earnings, and (approximately) the only thing that drives both outcomes. There is some non-intelligence heritability in undergraduate selectivity and lots of non-intelligence heritability in earnings.

Educational Attainment

Based on the NLSY97 dataset, the correlation between SAT score and educational attainment is r~0.19. Based on the model above, one might think that intelligence is the sole common cause of educational attainment and earnings. One would be wrong.

The correlation between educational attainment and earnings is around r~0.26 (TODO: cite). This is much higher than the intelligence-is-the-only-common-cause hypothesis suggests: 0.04 (0.2*0.19). So, what explains the higher correlation? We have only two options: education causes earnings and some confounders cause both. Both options are true.

For a treatment of educational attainment causing earnings, see here, but even after accounting for this causal variance and the variance related to intelligence, most of the correlation between educational attainment and earnings remains unexplained, which suggests other confounders remain. Given that common environment has limited impact on adulthood earnings, the smart money is that the main confounders here are genetic in nature.

Primary Strategy

Our primary strategy is to compute two z-scores: one for earnings and one for undergraduate selectivity. For the former, we'll simply convert income to z-scores using the distribution of income at age 33 Chetty for the bottom 99th percentile and the Pareto distribution for the top 1% Using elasticities to derive optimal income tax rates. For the latter, we'll use the average ACT score at the attended undergraduate institution. Finally, we'll assume a single "resume factor" causes both of these equally.

TODO