Good idea! But it’s already been explored by geographers

This post is an attempt to start a Mexican wave among computational social scientists (not only the urban science crowd) for the golden age of structural geography—the 1960s-80s. The main point is not to say geographers should have more credit but that literature from that era is a treasure trove of ideas. Neither is the point to find the true roots of ideas—indeed, many urban data scientists know even older economic geography, like Christaller’s theory of the 1930s or Kohl’s work from the 1850s. Rather, to share my favorite starting points for explorations into geography. Shout out to Carl Nordlund, who reminded me to write this blog post (I’ve thought about it many times before).

Although I have no training in geography, my fascination for the topic goes way back. My dad was an architect, and to escape the doldrums of Swedish suburbia, I occasionally leafed through his old textbooks. Spatial Organization by Abler, Adams, and Gould was a favorite. Especially the variety of illustrations fascinated me, and the fact that this was something considered valuable . . “it’s important for planning,” explained my father. For context, at least then, Swedes used to pride themselves in taking the middle way between market and planned economies. “Fair enough,” I thought, “the Swedish model needs economic geography.” But little did I know I would reach the same type of figures and thought myself much later, and from a very different direction.

Since my first forays into data science, I occasionally ran into geographers who blithely reminded me to check their literature. I guess, about a decade or so ago, I actually listened to them and went back to dust off Spatial Organization. So here are 11 examples that, to me, feel very modern for their time. Some would feel dated if presented at today’s urban data science conferences; some would still feel very fresh. Occasionally, I accompany these ideas with their urban-science soul mates—but in no case are the latter pure reinventions, and sometimes they are well-informed about their geographic predecessors.

1. Degree distributions

Let’s start softly with a degree distribution and an explaining random null model. Because . . hmm, that’s what we do as network scientists. [Ha p49]

2. Statistical network models

How network models match observations. The network structure in question is described by two parameters, I and S, characterizing the distributions of distances in real networks (the plusses). [Ha p45]

3. Street patterns

Maybe the most fundamental urbanology idea to the structuralist mind—patterns of road networks contain a lot of information about the growth and function of a city. Here are figures from [AAG p317] and an instant-classic paper by Geoff Boeing. I think it’s safe to say that the question of how to extract the information hidden in the road structure is still open.

4. Shapes

Also the shapes of contiguous regions appeal to the idea that structure carries information about evolution and function. Here are atolls to the left (from [Ha p71]) and parking lots to the right (by Bogner and Szell).

5. Trade-offs

Real spatial networks (distribution networks and roads) are trade-offs between minimizing the construction cost (total line length) and point-to-point distances. B/W figures from [AAG p260]. Color panels from Gastner and Newman. The FKP model is another model from the complex-networks era with the same premise. (For those of you who also noticed the dual presence of Cincinnati—it is not as simple as that it is the center of mass of the American population.)

6. Network structure

How a network structural measure β (the edge density of railroad networks) predicts an economic indicator, and how countries move within this space. From KJ Kansky’s 1963 Ph.D. thesis [Ka].

7. Network growth models

Stochastic models of the growth of spatial networks, yay! All from [Ha].

8. Life as multigraphs in a space of activities

This shows a brilliant idea I think is not yet fully explored—to represent the course of a person’s day as a directed multigraph in a space of activities. [CPT p87] I guess the closest counterparts in computational social science are the visitation-law of points of interest and the idea that some kinds of activity show bursty behavior.

9. The levels of mobility

Networks all the way down—from [AAG p 409]; somewhat reminiscent of Alessandretti et al.’s work on the scales of human mobility.

10. The constraints of temporal ordering

Similar to some effects in the temporal network literature, geographers have studied constraints from the time ordering of events. [CPT p207]

11. Typologies of movements and distribution

These are from [AAG] and a fascinating topic for someone seeking unifying principles and (down the line) minimal models.

Finally

It’s interesting to imagine how the two paths—structural geography and urban data science—lead to similar ideas. (I’m getting a bit hypothetical and oversimplified now.) Economic geographers were already accustomed to mechanistic modes of thinking from e.g. above-mentioned Christaller et al. The main input they needed was how psychologists translated structuralism into graph theory—e.g., Katz centrality is frequently cited (inspired by Jennings and Moreno’s foundational work on social networks). Computational urban science of today comes from when complexity scientists (many with a background in statistical physics) ready to understand the world with minimal models and networks came across urban data. One side had geographic data and mechanistic models, and needed network science to take the final step; the other had models and networks and needed the data.

Key references

[AAG] Abler, Adams, Gould, 1971. Spatial organization. Prentice-Hall, Englewood Cliffs NJ.
[CPT] Carlstein, Parkers, Thrift, eds., 1978. Human activity and time geography. Edward Arnold, London.
[Ha] Haggett, 1969. Network analysis in geography. Edward Arnold, London.
[Ka] Kansky, 1963. Structure of transportation networks. University of Chicago Press, Chicago IL.