Last week at Politics and Prose, I had the opportunity to hear Robert Putnam’s book talk for Our Kids: The American Dream in Crisis. In his book, he focuses on data that purport to show a growing divide in income inequality and social unraveling.
Putnam told a personal anecdote about his deteriorating home town in the Rust Belt. If Putnam were able to prove that the shrinking economy from the loss of manufacturing jobs in the Rust Belt proportionately reflected the larger economy in all other sectors, he might have a strong data point, but a personal anecdote is not enough. The availability heuristic is not strong evidence.
So much of the talk concerning income inequality pertains to an unstated premise about social mobility. The widespread fear is not just that the rich are getting richer, but that the rich are getting richer at the expense of the poor. The mental model assumes some fixed share of wealth that exists in the world should be divvied up fairly so as to avoid predation by the strong on the weak. Often, the evidence presented for a “fixed pie” theory is to show the shrinking share of income among the lower quintiles and the growing share of income among the higher quintiles. The problem with this methodology is that it doesn’t actually account for social mobility. To prove that capital is flowing from individuals in the bottom quintile to individuals in the top quintile, we need panel data.
If we don’t analyze with panel data, we might observe the top quintile, profiting from some entirely new high-tech sector, drastically increasing their income by 20% while lower quintiles still increase at 2%. More wealth generated at the top wouldn’t imply material loss for the bottom quintile.
In the Q&A, I pressed Professor Putnam on his methodology, specifically, to what extent he used panel data to show decreased social mobility. After his book signing, he elaborated for me.
Putnam claimed that there was some good panel data for income, but that it couldn’t be used to show current trends in social mobility.
We might suppose that the relevant panel data to measure social mobility would include income, for .
At or so, we might expect people to be generating the most amount of income for their life.
The methodological issue Putnam pointed out was that individual incomes over a lifetime are highly nonlinear. If you were to track a random sample of individuals starting at t=20, very different kinds of individuals would look very similar, but both would appear in the bottom quintile. Specifically, could be -$100,000 for a Harvard pre-law student who’s taking out student loans, and could be $16,000 for a minimum wage job. However, might be $450,000, while might be something like $25,000.
Putnam pointed out that because of how this panel data is measured, the data is always intrinsically 30-40 years out of date. Wait, is this a cop out? Is this methodological laziness?
Just because any one particular study requires 40 years doesn’t mean that we couldn’t observe multiple concurrent staggered studies, with different individuals to show panel data over time. We can imagine Study A starting in 1945 with a batch of individuals at , Study B that tracks at 1950, Study C that tracks at 1955, and so on and so forth. Then, despite nonlinearity in lifetime earnings, we would still be able to see trends in how individuals are or aren’t moving up, out of their birth quintiles.