Measuring Grad Unemployment Properly

Graduation Age Matters to Educational Comparisons

This is a very data-wonky piece, intended for people who are interested in diving deeper into properly measuring the “college grad crisis”.

TL;DR: The deterioration in labor market outcomes for young four-year-university graduates relative to their same-age peers is real, but an overestimate of the “true” effect. That’s because young college graduates are less seasoned in the workforce than non-college-graduates.

1. Context

If you’ve been following econ wonk discourse (or my writing), you’ve probably seen a chart like this floating around. It shows an increase in the unemployment rate for people in their early 20s across educational groups, but also a relative deterioration in outcomes for college graduates of that age relative to their peers.

The explanation for the behavior of these time series is hotly contested. Is it cyclical or structural? (I think it’s some of both.) Is it AI or the business cycle? (I think AI is probably laying a role, but not a dominant one.)

But this piece is about something slightly different: proper measurement of labor market outcomes across different educational attainment subgroups of the same age. To give an obvious example, compare a 22 year old who just finished their bachelor’s degree to one who finished high school 4 years earlier. The college grad is much more likely to be starting their career at that age. And that may almost certainly means more vulnerability to an adverse hiring environment (like the present).

An implication of this graduation-year effect is that comparisons of labor market outcomes within the 20-24 or 22-27 age groups, across educational attainment subgroups, are going to mash together two different effects:

1. A graduation age effect for 4 year college grads

2. Whatever differential exists in outcomes between education subgroup 1 and subgroup 2

When we’re talking about whether young college grads are being particularly hard hit by recent labor market developments, we’re very interested in 2 but much less interested in 1.

I’ll spend the rest of this piece providing some evidence for this. In my opinion there’s more work to be done here.

2. Graduation Age Effects Matter

If you’ve looked at the unemployment data sliced up by subgroup, you’re probably aware of two key “gradients”: people with higher educational attainment tend to have lower unemployment rates; people who are older also tend to have lower unemployment rates.

But there are exceptions to this general pattern. In 2022-23 and 2018-19, periods that were exceptionally good for the US labor market, the unemployment rate for 22 year olds with a bachelor’s degree (or more) had a higher unemployment rate than 22 year olds without one. Was this indicative of a bad labor market or of a “penalty” for young college grads? Obviously the answer is no. Recent education completers are much more likely to need a new job. By the time you get to the mid-20s, the typical educational gradient reasserts itself.

A related observation is that this “graduation year” effect also applies to other educational groups. For folks without 4 year degrees, unemployment rates peak in their late teens.¹

So far we’ve talked about what the labor market looked like in good times. In the past 2-3 years we’ve seen a substantial cooling in the labor market, happening mostly through much weaker hiring. And weaker hiring hits people who need new jobs particularly hard. Combined with ongoing low layoffs, this means bad news for young people. And not surprisingly, their unemployment rates have increased.

But that increase has not been uniform. For people without a college degree, unemployment rates have risen a fair amount for ages 18-22; for people with a college degree, the effect happens from ages 23 onward.²

3. What This Means

One basic conclusion: if you do a comparison of recent changes in unemployment rates within the aggregated 20-24 age group, you will detect a college-grad-specific effect of weak hiring that doesn’t really exist - because you are excluding some of the non-college-grad increase. If you do this for the ages 22-27 age group, the mismeasurement will get even worse.

My initial thought was that this would also have bearing on the excellent “Canaries in the Coalmine" paper by Erik Brynjolfsson , Bharat Chandar and Ruyu Chen, but I’m not sure. The core conclusion of this paper: employment in the last few years declined for people ages 22-25 in occupations with high AI exposure relative to those in other groups (either older, or in occupations with low AI exposure). But imagine the following set of relationships:

1. General hiring weakness hits early-career folks harder than later-career people

2. Early-career folks without a college degree are likelier to be below the age of 22 than those with a college degree (this post provides evidence of 1 & 2)

3. Prevalence of non-college-grad workers within occupations is inversely proportional to AI exposure (outside the scope of this post)

In that case, you might detect an inverse relationship between age 22-25 employment and AI exposure that has nothing to do with AI exposure.

The reason I’m not convinced that the authors’ paper is undermined by what I’ve written about in this post: they split occupations into college-grad-heavy (70% of workers within the occupation or more) and college-grad-light (30% of workers within the occupation or less) and find similar AI exposure results. But I would like to see this analysis across a more granular slicing of industries, and would also like to see the ages 18-21 (call it “early non-college career” or “very early career”).

Anyway, I hope this is useful!

1. To avoid noise from small subsamples, I lumped “some college” and “associate degrees” (two fairly distinct groups) together. The flatter age profile of this intermediate group probably reflects the fact that associate degree completers are likely to search for a “first job out of school” at an older age than college dropouts.

2. It’s a good question why we see the college grad effect kicks in at 23 rather than 22. My guess is that college graduation happens sufficiently late in the year that, when the labor market cools,