From The Economist, we get a link to William Easterly writing that the original racial neighborhood tipping point model of Schelling and others is unstable:
The basic prediction is that mixed neighbourhood are unstable but segregated neighbourhood are stable. Data on American neighbourhoods from 1970 to 2000 rejected these predictions – it was the segregated neighbourhood that were unstable. There was as much “white flight” out of all-white neighbourhoods as there was out of mixed neighbourhoods, and there was a white influx into segregated non-white neighbourhoods. Neighbourhoods are still very segregated in the year 2000, but not because of tipping. Maybe segregation exists because most whites really do want segregation, not because of a chain reaction due to herd behaviour.
So I want to make sure this doesn’t approach SIWOTI territory, but it involves some of my favorite econometrics graphs. This paper is actually a reaction to a paper by Card, Mas, and Rothstein (2008), Tippin and the Dynamics of Segregation, as the author notes in his paper:
The advantage of this paper’s methodology in estimating the entire distribution (2) is that it allows for the “classic” tipping point story to be compared to two alternatives: (a) there are no tipping points, and (b) there are tipping points but the CDF does not fit the “classic” global tipping story. The Card et al. 2008 approach, in contrast, can only rule out (a), not (b).
So Card et al. found that there are some tipping points in the racial composition of neighborhoods which Easterly concedes; their stability across the entire CDF is in question. Since looking at finding a single tipping point is more fascinating to me than trying to derive the full equilibrium story, I find this difference to be less interesting than the original (the Easterly paper obviously goes into this in detail).
Let’s check out what Card et al. found. This is one of my favorite econometric graphs ever:
On the x-axis is the racial composition of a neighborhood in a tract of housing in Chicago by minority, non-white or hispanic, in 1970. On the y-axis is the net percent change in white non-hispanic population between 1980 and 1970. We go to war with the data army we have, and we get census tract racial compositions in 10-year blocks. See the story it tells?
If the racial composition in 1970 is 5% minority, it gains 10% in white population over the next ten years when we check the data again in 1980. If the racial composition is 7% minority, it loses 20% in white population. How crazy is that! Something about that extra 1 percent minority share causes a chain reaction within a ten year time frame that causes a massive flight of whites from the neighborhood.
(Extra-geek sidebar: I replicated this paper a few years ago using quantile regressions. I wanted to see the interactions between RD design that the paper uses and quantile methods. I was hoping to try and use quantile regression to tease out tail end correlations of risky portfolios, and that there would still be time to save capitalism – I failed there. I was also hoping to use this replication as bait in an application to become David Card’s star padawan. Sigh. Good times.)
Now Chicago is a city famous for, ahem, being conscious of the racial composition of its neighborhoods. How does your favorite city match up? Also from the Card paper (click for full-size):
This effect is very obvious in the data during the 70s-80s time frame, but becomes less obvious in the 90s-00s. I’m excited for the 2010s data to get here already, so I can get to town seeing if this effect exists for college degrees or income.
Very related: Conor Clarke has an interview with Thomas Schelling, who is one of the people responsible for originating the tipping point model, this week (Part One, Part Two) that is really worth your time.
Rortybomb 
Thank you for pointing out this paper – the graphs are fascinating.
I’m curious about the differentials between the tipping points. Do you know of any studies that have compared the tipping points in given cities (i.e. around 7% in Chicago versus 18ish % in L.A.) with other data?
It would be interesting to try to tease out some empirical evidence of, say, economic inequality leading not just to increased segregation, but to a lower tipping point (i.e. less comfort with mixed living). Or perhaps searching for some fallout from the infamous “southern strategy” – if there were cities where the political advertising was more overtly racist, does that correlate over time with a lower tipping point? Obviously you would have to watch out for the direction of that causal arrow, however.