I know there are lots of gravity models in economics that take as an input distance between countries. My understanding is distance is typically coded based on the distance between capital cities, either in $L_2$ norm or great-circle distance. Most countries are small relative to the distances between them, and so maybe it doesn't matter much how they code the center of these countries for purposes of trade. But in my application I'm looking at US states, and many states are large enough that the internal placement of the location for calculating distances between states makes a difference. Additionally, state capitals are often not the economic center of the state (Carson City, NV and Albany, NY come to mind) nor the geographic center (North Folk, CA is hours away from Sacramento, CA), so I don't know of obvious choices.

Are there any standard choices I should know about here?

As a resource to others here are some sample data:

  1. Average Latitude and Longitude for US States
  2. US Census Centers of Population
  3. List of geographic centers of each U.S. state

Would anyone provide a list of the economic center of each state?

  • 2
    $\begingroup$ In many GIS applications, people use population-weighted centroids. If you have data on locations of the economic phenomena you are working on (say factories or retail locations), you can do something similar. There's a GIS stack site where you may get a better answer. $\endgroup$
    – dimitriy
    Feb 11 '15 at 18:07
  • $\begingroup$ I think the Census Centers of Population uses the centroid method. $\endgroup$
    – BKay
    Feb 11 '15 at 18:27

I think the answer depends on who is impacted by your measure of distance and for what purpose. Kennan and Walker (2011) Econometrica measure distance between states as "the great circle distance between population centroids" in an attempt to model moving costs. They also include an indicator for whether or not the state is adjacent.


Instead of using population weights, you could do a spatial average of county centroids using county-level personal income as the weight. The residence adjustment that is applied (LAPI is calculated on a place of residence basis, rather than place of work) will screw this measure up a little bit (fortunately, the adjustment mostly just shifts around income within metropolitan areas), and it doesn't account for differences in the capital intensity of various industries, but it's likely a better proxy than using population as a weight.

More broadly, I'd really recommend checking out Jim LeSage's spatial econometrics work (he wrote the toolbox for Matlab), as much of it deals with spatial weight structures.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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