Statistical Modeling, Causal Inference, and Social Science

I knew there was a reason I liked New England :-)

along the same lines - where all the mcdonalds, burger king, jack in the box, pizza hut, wendy's, kfc, hardess, carl's jr, and in-n-out's are: http://www.fastfoodmaps.com/static.html

Ah, but what is the density of Wal-Marts with Starbucks IN THEM?

Great example of pictures speaking for themselves!

For the scatter plot, I find "people per Walmart/Starbuck" easier to understand than the other way.

Hmmm.... According to Wikipedia, there are just under 11000 Starbucks locations in America, compared to just under 4000 Walmarts in the US. I'm surprised the graph looks so symmetrical, considering that they're on the same scale.

The population-weighted average of the Walmarts should be about 13, and Starbucks about 36. Starbucks looks right, but it looks like the *lowest* Walmart numbers are around 13. Even with NY, NJ, CA at the bottom I don't see how the averages come out right. I think your y-axis is mislabeled.

Some quick googling suggests that as of a couple years ago Mississippi had just over 50 Walmarts for 3 million people, while the graph suggests it has 50 per million. Arkansas, about 75 Walmarts for 3 million people; the graph suggests over 75 per million.

You should try this at lower geographic areas. It is even more striking.

Both companies are justly renowned for their site selection, but their core markets are different. (space requirements, too, but it's income that really drives this)

Around the office we've joked about categorizing block groups based on the relative density of Wal-marts and dollar stores within a 5 minute drive time -- versus Starbucks. Haven't found a client who's willing to pay us to do it yet.

Given that Starbucks started in the pacific northwest and wal-mart in the south-central US and both followed the standard outward growth from the initial base pattern, the only possible thing that could be considered interesting here are the outliers.

Why Virginia & Illinois for Starbucks? I am guessing that this is due to the existence of large metro & suburban areas (Chicago and the D.C. region) that were compact enough for easy franchise growth and where the bulk of the population is also concentrated in those areas, making the rest of the state statistically insignificant. This seems to hold if you look at Colorado (the I25/Denver corridor), and Arizona (Phoenix & Tuscon) but Texas and Florida are a bit more work for a chain franchise.

Starbucks... founded in Seattle
Walmart ... founded in Arkansas

What happens if you repeat this for e.g. Dunkin' Donuts (founded Massachusetts) or Safeway (founded Idaho)?

Might be fun to see how true geography compares with the "axes of Starbucks/Dunkin'" - and if coffee does a better job than groceries in this regard.

Of course, if you check these same two against states that have gone for Hillary and states that have gone for Obama, I wonder what you'll find... :)

And, what does this say about Texas and Ohio for Tuesday?

What a surprise. There are more Starbucks near Seattle where Starbucks was founded, and there are more Wal-Marts around Arkansas where Wal-Mart was founded.

Nathan: Cool graph; thanks for the link. It's hard to make comparisons though, since the biggest factors are the total #stores for each and the general population distribution of the U.S.

Alex: Thanks for the note. I'll ask the student who collected these data to figure out what's going on.

Z: Good point. State-level is just how I got the data.

Evegen, Ken: Exactly. I thought it was interesting, first that the spatial pattern was so stron; second, the outliers from the pattern.

Mark: Thanks for the sarcasm. For a brief shining moment, I'd forgotten I was on the internet.

I want to live in Vermont...

what happens if you transform both axes?

Regarding proximity of store origin in the last graph, suppose the graph were changed to distance from Arkansas on the vertical and distance from Washington on the horizontal, and kept the lower 48 as data points, would the graph look the same?

Looks like an indifference curve!

Fair enough about my sarcasm. I was irritated by the implication (not stated, true) that by counting the number of Wal-Marts or Starbucks, you can say something meaningful about the people in those two parts of the country. I don't think you can. It's a not-so-implicit form of stereotyping. Yes, there are differences between customers of the two (for the record, I frequent both), but your maps, because they plot by state, strike me as misleading.

@mark: all andrew did was show the data. he didn't make any implications or even attempt to explain anything other than axis labels. interpretation of the graphics are left up to you, the viewer.

It would be interesting to survey how many people had the same interepretation. Whatever the results, I wouldn't mind to see Canadian maps for the Starbucks and Walmarts per capita.