I guess as more people are actually reading the Chetty et al (2013) study study, they are finding what I discussed yesterday: that study doesn’t have any urban form variables. Talen and Koschinsky from Arizona State have a short feature up at BetterCities that suggest there is a connection.
It’s a short piece, so there’s only so much to see, but there are problems here. They use WalkScore as a proxy for suburbanization, which is defensible if you think that clusters of amenities are likely to also to represent job clusters–I don’t see anything particularly wrong with that. However, I do have problems with the finding that there is a “strong correlation” when they don’t report the correlation, test it, or test its robustness to outlier removal. Sure, fine, don’t show us a full suite of regressions; you just started fiddling with the data, but we should know what type of correlation was used, particularly since one variable is an index.)
I also do not understand why they are aggregating to the city level in their reporting, other than the rather compulsive desire people have to rank cities on the internet. They have block group data; what we are talking about here is a neighborhood-within-a-region phenomenon–not strictly a region or a metropolitan phenomenon in and of itself. With that aggregation, they have a couple outliers that appear to be pulling the line in various directions, and that’s likely to be the case with 82 percent of their sample in the inaccessible category. There is also uneven spread in the data; that is, they have more scatter in some parts of the distribution than others. It may not matter, but here I suspect it does.
Anyway, none of that reassures me that the analysis shows what they want it to show.
We also do want to be careful. Intergenerational income and individual income mobility are longitudinal phenomena, conceptually. Income data attached to places as units measures the incomes of the people in that place at some aggregation. People move in and out, and it may be that places remain poor while people move in and out of that place and achieve higher incomes along the way. If you have a trajectory for an individual’s income, matching that to a place in time assumes they have always lived there, and that’s a dangerous assumption. We also have reverse causation problems here. Places with high walkscores and lots of amenities are more desirable places to live to begin with, so it shouldn’t surprise us that people who wind up there have income and wealth. We’ll need an experiment or a good instrument to suss these issues.
Beyond that, people, for heaven’s sakes: we already know that lack of access affects income and wealth because it affects job prospects. Why would we think otherwise? If there is one literature that has been empirically tested, it’s the spatial mismatch literature. Did John Kain live in vain? The very best work on this subject that I have ever encountered comes from Evy Blumenberg and Michael Stoll at UCLA. The story being told here is not particulary new. It stands to reason: if you grow up in a place where it’s hard for your parents to get ahead because they don’t have access, it’s not likely that you are going to get ahead, either. The Chetty et al. 2013 project is trying to look at opportunity from a wide variety of variables, and that is new.