Which occupation are you?

Validate your stereotypes, or find your future job (or wife)

About a month ago this very interesting paper was published:

• Wolfram, T. (2023). (Not just) Intelligence stratifies the occupational hierarchy: Ranking 360 professions by IQ and non-cognitive traits. Intelligence, 98, 101755. (PDF)

Occupational sorting, the process of individuals actively selecting into and being selected for different occupations, has significant implications for social stratification and inequality. The psychometric view of occupational differentials in ability emphasizes the importance of intelligence for occupational sorting, as it acts as a necessary condition to enter and remain in certain professions due to their high cognitive demand. The resulting cognitive stratification of the occupational hierarchy leads to strong associations between occupational mean IQ and sociological measures of occupational status and pay. Past research has been criticized for lack of representativeness and small sample sizes. In this study, we both confirm the psychometric view in a large representative sample and extend it to a set of nine non-cognitive traits. We show that the psychometric view holds (on a weaker level) for multiple non-cognitive traits, and using small-area estimation, we provide precise mean estimates and rankings of intelligence and non-cognitive traits for 360 occupations, including rare professions. Keywords: Social Stratification, Occupation, Non-Cognitive Traits.

The data are from the neat Understanding Society study, a large (n=40k+) UK survey that measured a variety of psychological traits:

Unfortunately, the measures were pretty short, so reliability is not amazing with omega's from .58 to .91. Good would probably be .90+, and here we have quite a bit lower. It doesn't matter so much at the group level, as the random errors cancel out, so the rank differences between the groups will be the same. However, they will be underestimated in size. Wolfram notes this problem indirectly and adjusts for it:

Each observation is weighted according to the calibrated design weights provided by Understanding Society (Lynn et al., 2010). To correct for measurement error we apply Spearman’s attenuation correction using McDonald’s Omega as a reliability measure (correction using Cronbach’s Alpha led to very similar results), as done in previous studies (most notably Hauser, 2010; Jensen, 1980).

Despite the mediocre quality of the psychological scales, their distributions were fairly normal:

For the purposes of calculating the occupational means, these were rank-normalized to get more normal distributions. But in any case, the differences are mostly fairly minor.

One way to have a broad overview of how much stratification one can explain using some variable is to compute the variance explained from some ANOVA-like model. Looks like this:

Not surprisingly, sex related differences explain the largest share, but intelligence manages to barely beat education. In other words, occupations are slightly more dispersed by intelligence than they are by education level, which is quite amazing in some way. We also see that stratification is lower for younger people. This could be interpreted as people not yet finding or working their way towards their niche or optimal performance level.

In terms of intelligence, here's the top and bottom occupations:

The rank order is probably about right, but the differences are much smaller than previously reported. That's what one would expect based on the low reliabilities mentioned above. Another idea is that one could interpret this as society getting less stratified by intelligence ("the anti-Bell Curve" hypothesis?). Alternatively, one could interpret this as a consequence of the mass university, where top occupations are decreasing in the eliteness of their workforce due to credential inflation. We have previously seen this happen for e.g. Danish PhD students (current mean IQ about 110!), and also for American educational attainment levels. A fourth possibility is that the classification of jobs was suboptimal. I've met a lot of Danish physicists, and I would guess their average IQ is about 125. But "physical scientists" includes probably also geologists and environmental scientists. I know the physicists consider geologists to be bad at math, as they would frequently fail the exams for the math classes both majors take, but physicists found relatively easy. Environmental scientists sound almost like a synonym for climate doomer, though a lot of it is solid science based on complex modeling.

Similarly, we can look at which occupations are highest in the non-cognitive traits:

Again, the results have a lot of face validity. Hairdressers are talkative, sometimes annoyingly so. Legal secretaries are not known for their risk-taking but sports players are. Musicians are open-minded, truck drivers are not. The appendix provides the results for the other traits:

Things generally look sensible. Clergymen are agreeable, people who work in construction are not. Most amusingly, journalists were the 2nd most mentally ill occupation. This probably won't surprise you judging from their tweeting. Too bad that these differences were not explored for their correlations. Presumably some quite juicy findings would be seen if one plotted these by the sex distribution. If someone can acquire the sex and ethnic data for these occupations, that would be great.

A related finding in the literature is that higher occupations have less variation among their employees' intelligence. The reason is simple enough: the more elite people you want on a normal distribution, the less varied they necessarily have to be. There just aren't that many 140+ IQ people you can hire, so if you refuse to hire people below 130, you will get a lot of people with IQs in the 130s, and only a few with IQs in the 140s, and ever fewer with 150s:

Finally, for those who want to apply the stereotypes in real life you can download the supplements and find a specific occupation. Actually, I couldn't find the supplements on the journal website, but a kind Reddittor posted them. They are here:

• wolfram 2023 supplement 1 (docx) (actually, a bunch of tables with sample sizes and statistical estimation details)

• wolfram 2023 supplement 2 (xlsx) (the table below) (Google sheets version)

For those interested in more about the psychology of occupations, my 2015 summary has a lot of information.