Illustrative Problem

I have a number of geographically dispersed stores where I would like to understand what factors predict revenue. Each store has certain properties (sq ft, age, employees, etc.). I also have data for the area surrounding the store by zip code (population, industry spend, competition, etc.).

I am struggling with the right way to incorporate the geographic data into my model. One potential would be to draw radii and create aggregation variables based on distance from the store. For example:

  • Number of households 25km from a store
  • Number of competitors 50km from a store

and then include these in a regression model. My question is:

  • Is this the best technique? Are there other models better suited to do this analysis?
  • Is there a way to statistically determine the distance I should use for each variable?
  • $\begingroup$ This sounds like an "empirical IO" problem in econometrics. $\endgroup$
    – StasK
    Aug 10, 2016 at 19:07
  • $\begingroup$ I'd think that the radius you want will depend on subject matter expertise. A gas station in a suburban area is probably not in any competition with another gas station 25km away -- maybe not even 5km away. On the other hand, a gas station in an extremely rural area might be in competition with a gas station 25 km away. And a different type of store would also be quite different. $\endgroup$
    – Wayne
    Aug 10, 2016 at 19:19

1 Answer 1


"Spatial econometrics" is the key word, I've seen many papers doing just that. The approaches depend on the domain.

For instance, here's a paper pushing the "convergence club" concept in house price modeling. It's often said that there is no national house price market in USA, that the country's so large that what's going in Washington, DC real estate market has nothing to do with prices in Orange County, CA. People kept repeating this as a mantra for decades until the Great recession 2007 happened when all of a sudden house prices crashed everywhere seemingly at once.

So, now there's more interest in studying links between housing markets. There was this old paper on "ripple effect" in UK where you get price shocks travelling from Sougth to North, or something along those lines. Hence, this paper "UK House Prices: Convergence Clubs and Spillovers" by Montagnoli and Nagayasu is one of the studies of geographical links, where they use this idea that house price in some regions tend to converge, or move together, hence, the convergence clubs. They do it with a panel analysis where the house price growth trends are estimated in fixed effects model, and running a test on a parameter that would show whether there is a convergence or not.

This is just one approach, and it demonstrates how spatial analysis can be conducted in a very specific setup. There is not a single universal framework for geographical analysis in econometrics.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.