Duranton's notes on Urban Economics

July 11, 2023

The original tweet that inspired this post is here but, in case the “bird” decides to change things up, the slide deck discussed is available on this World Bank URL directly.

My usual doom scroll on the bird this morning took me to this slide deck by Gilles Duranton, “New Frontiers in Urban Economics Research: A personal view”. I wish more big names did this more often. The academic system leaves little formal room for speculation, personal views, and synthesis of the kind that does not fully fit on a journal article (though, incidentally, I think this one would quite nicely!).

Anyhow, here’re my notes from the slides (I was not lucky to be in the room…), most of them repdouced here as the original text with additional comments.A couple of caveats, these are partial in the sense they are biased and in the sense they are incomplete. They contain more my own thoughts as contained in Duranton’s words, much more eloquent I could possibly do anyway.

The talk is structured around two themes: recent improvements in urban economics, and what I’d call “currently hot topics”. The former is then subdivided into: modeling, data, methods. I am these days less close to the applications of Urban Economics, and my focus is more on “infrastructure” that supports the field (as well as a wide range of other areas such as government policy, or other fields such as geography, planning, or architecture). For that reason, I’ll probably say very little about the half that focuses on hot topics, other than I really wish more people in more fields (e.g., quantitative geography) did this kind of exercise as a matter of relative routine).

Modeling

This is the bit I’m less close to. I never wrote a model in my PhD, and there’s a good chance sadly (or not) I never will at this point. But I did find interesting Duranton’s point about a much more iterative approach to empirical research:

I’m pushing typical approaches to the extreme to show neither data nor model should dominate […] Ideally, use an initial model to guide the empirics and revise it as progress is made in the data exploration, etc

Data

This to me is the biggie in the slide deck (and in the last couple of decades of Social Science for that matter…). As Duranton puts it, “we keep winning the lottery data”. Some sources he flags as game-changer in the last few years:

Duranton ackowledges he may be forgetting some sources. To be honest, I spend most of my days thinking just about data and I can’t come up with much more, other that possibly synthetic data simulated or modelled from disparate observed sources…

The other point made in this section relates to the challenges these new sources bring. Not in those words, but his point is these data represent not only a lot more data, but also different kind of data that require different kinds of approaches. And we in the social sciences are being slow to wake up to this realisation (I think).

An important call to not fall prey of the lost keys joke:

Novel data is not a substitute for an important and well-posed research question, insightful modeling, and a good research design

Methods

In Duranton’s view, urban economic research is about:

Of the three, he notes ii. has received notably less attention. I agree with this. Much of the challenge of “democratising” new forms of (unstructured) data in the social sciences is really about finding ways to turn them intro (structured) tables they can then work with. Until then, we can hope folks will become ML wizards but my hunch is the majority won’t. Hence the need to work in teams (also noted in the slides), not only of urban economists.

The second strand of thought in this section is about methods that generate data, think predictions from ML models. His view on this:

Machine-learning approaches have made their splashy entrance into our toolkit both to create data (here) and extract patterns out of them. We need to learn to use appropriate approaches and use them appropriately

The bit that is not mentioned explicitly are LLMs (although hinted at through “textual analysis”), but much of that slide took me back to the “Future of Urban Analytics” panel we organised at the last AAG, where Seth Spielman rounded up his partitipation with the quote “the data is the model, the model is the data”.

Finally, a point about delineation:

Much extent research uses the spatial units it uses “just because”. This leads to a variety of problems (measurement error, MAUP, etc). The cost of delineating appropriate units for a given research project is no longer prohibitive and will increasingly become a requirement

It’s really refreshing to see this view through beyond methodological geography (where it’s been most of its raison d’etre for a long time) and taken so on board. It is one I obviously share, I think we’re at the dawn of a time where thinking about which spatial unit one should use will be a core step in every empirical project, and empirical (spatial) research will be notably better because of that.

Ps.

As I said at the beginning, above are the bits I find most interesting from the slides, but this does not mean they’re the more important ones, so do check out the slides yourself if you’re curious. They are chock full of recent references, which is another aspect I think we do not do enough of in talks (we do of course in papers, but there the focus is usually much more delimitted and less wide ranging than in this case).

Duranton's notes on Urban Economics - July 11, 2023 - Dani Arribas-Bel