college and mobility « Peter Levine

Xiang Zhou has published “Equalization or Selection? Reassessing the ‘Meritocratic Power’ of a College Degree in Intergenerational Income Mobility” in the American Sociological Review–also available here. Like all multivariate statistical models, it’s subject to criticism and should be replicated using other datasets and specifications. But it’s certainly interesting and very well presented.

For background, here is Raj Chetty’s graph from a different source, showing the relationship between Americans’ family incomes when they were growing up and their incomes as adults, both expressed as percentiles.

If it showed a straight line from 0,0 to 100,100 with a slope of one, that would imply a perfect correlation between generations. It would imply zero mobility, although it would be consistent with some random movement up and down the income ladder that would be concealed by displaying averages. Instead, we see a basically straight line from 0,30 to 100,70 with a slope of about .4. People at the very bottom do tend to rise a bunch of ranks (if they survive to adulthood), and people at the very top do average below where their parents were. In short, there is a strong correlation across generations with some reversion to the mean.

The graphs in Zhou’s article use a similar format but they show the data for young US adults depending on whether they graduated from college or not. Graduates are shown in blue, non-graduates in red.

In this first pair, the graphs show the raw data and then the same data reweighted for a bunch of variables that were measured while the subjects were still in high school: their “gender, race, Hispanic status, mother’s years of schooling, father’s presence, number of siblings, urban residence, educational expectation, the Armed Forces Qualification Test (AFQT) percentile score.”

It’s clear that you should go to college. A person whose parents were at the 25th income percentile ends up at the 30th percentile if she doesn’t graduate from college, and above the 60th if she does. But the slope is flatter for college students than non-college students without reweighting; and the lines are closer together and parallel when reweighted.

Zhou interprets this pattern to mean that people who are able to attend college (due to financial resources, a high school diploma, motivation, etc.) end up better off than those who are not able to attend, but college itself adds no noticeable benefit. To put it another way, if we used public policy to move everyone to the blue line by making them all college graduates, the slope of the blue line would remain fairly flat and it would cross the y-axis at a lower point, because it’s impossible that everyone in a society should rise to a higher income percentile than their parents.

Here are the same patterns when Zhou distinguishes between selective and non-selective colleges. The slope for the non-selective colleges is steeper (which could imply that they are better at mobility), but the difference between the selective and non-selective colleges is not statistically significant after reweighting.

If Zhou is right, universal college would not improve economic mobility at all. The line would be fairly flat and lower than the current line for college grads. The economic advantages of college are due to stratification, not the experience of being in college.