Tuesday, December 21, 2010

In Sunday's post re "A Physicist Solves the City", I emphasized the parts that I liked (re Jane Jacobs complexity). But there are now about a half-million links to the article and some objections to this:

After two years of analysis, West and Bettencourt discovered that all of these urban variables could be described by a few exquisitely simple equations. For example, if they know the population of a metropolitan area in a given country, they can estimate, with approximately 85 percent accuracy, its average income and the dimensions of its sewer system. These are the laws, they say, that automatically emerge whenever people “agglomerate,” cramming themselves into apartment buildings and subway cars. It doesn’t matter if the place is Manhattan or Manhattan, Kan.: the urban patterns remain the same.
Yes, that kind of thing is jarring (and is hard to square with Jacobs complexity). I come back to this piece because I have just refereed a paper re urban densities which elaborates the importance of distinguishing between daytime and night time densities. The authors use population and employment densities (for areas in a major city abroad) to illustrate.

That's a fair point and I always worry that the areas chosen for analysis are small enough to be meaningful. Consider that average citywide densities for the top-ten U.S. cities in 2000 varied from 26,401 pop. per square mile (New York) to 2,808 (San Antonio).

But we can get smaller-area data (as in 100 PUMS areas for the Los Angeles metro area). We can then create population density deciles. The densest LA PUMAs had an average population density of 26,738 and an employment density of 9,425. The correlation among the densities of the PUMAs in the decile was 0.95. Areas that housed 10 percent of the population provided 9% of the region's jobs.

The next decile of densest population housed another 9% of the region's jobs. In fact, each of the ten population deciles housed near 10% of the region's jobs (the range was 8.6% to 11.1%). To be sure, the correlations between the densities within each decile varied considerably, from 0.06 to 0.98.

What does it mean? Within and between neighborhoods there is glorious Jacobs complexity. It is way beyond anyone's ability to predict with 85% accuracy.