Spatial evolution of the US urban system

We examine spatial features of the evolution of the US urban system using US Census data for 1900-1990 with non-parametric kernel estimation techniques that accommodate the complexity of the urban system. We consider spatial features of the location of cities and city outcomes in terms of population and wages. Our results suggest a number of interesting puzzles. In particular, we find that city location is essentially a random process and that interactions between cities do not help determine the size of a city. Both of these findings contradict our theoretical priors about the role of geography (physical and economic) in determining city outcomes. More detailed study suggests some solutions that allow us to restore a role for geography but a number of puzzles remain.