Chapter 4 | Conclusion
This analysis has shown several linkages between American cities. Just as assessing urban form and how it relates to lived experience is subjective, so too would be any assessment of these similarities and clusters. Yet we find evidence that the similarity score captures some relevant aspects of this feel within both how much it agrees with cluster analysis and in how cities in distinct classes and geographies are labeled similar by this algorithm. Generally big, gridded cities are more related than small, messy ones; regions also show similarity. New York is similar to Chicago and Philadelphia—other big, gridded cities—and when we construct a network of relationships from similarity scores, there is a cluster toward the center that is occupied by Orlando, Tampa, and Jacksonville—all Florida cities; Jersey City is close to Newark and many other sensible pairs exist: Garland and Plano, only a few miles apart. The fact that this score is able to hint at an area typology—and within that, the first law of geography1—is assuring. Below we plot samples of cities ranked 1 through 4 in our sample alongside comparison cities, with cities ranked 96 through 100 in our sample following.
everything is related
Los Angeles and New York are world cities that appear related by these measures. Although Pittsburgh and New York are not comparable in size, this analysis is limited to City Hall and its surroundings. In this case, both are at the tip of narrow island—so water might be driving this—and both are in business districts, so the built form is likely similar. Los Angeles and San Diego (California) and Houston and Dallas (Texas) also seem sensible according to the spatial dependencies of the first law.
Four large cities with comparison
Within the sample of less populous downtowns, again, geographic consistencies stand out. Tacoma and Seattle are related (West), as are Boise and Denver (Mountain West); Garland and Arlington—Dallas satellites—are both similar to Baton Rouge, which is in turn not far from them. We see this in clustering analysis as well: geography is an important predictor of how similar any given downtown will be to another.
Four small cities with comparisons
next
Further tuning is necesary: each city is associated with at least one or two others that would not fit subjective exprience. This model was also narrowly focused: we looked only at downtowns and we used one of potentially many ways to identify those downtowns algorithmically. The ideal system would have a better understanding of where the businesses and the people are when locating the window of analysis. One thing is clear from this process, though: Phoenix is the most dissimilar city in America—never connected in our network, little related in our matrix, and with no pairing in our dendrogram. This is likely because our model considers not just its grid, which might be similar to many other cities, but the size of its buildings (large) and the density of its population (sparse). Next we turn to the world: there is already a list of first and second cities by country that could make for a good test for this model. American cities likely all fit within a subset of global taxonomies—intensely gridded and ordered—and thus might be difficult to separate; adding new countries will involve more variance and help answer questions like which European cities are similar to New York, Boston, or Philadelphia.
-
That everything is related but near things are more related than far things. ↩
Updated: