Measuring Political Effect in Store Chains

erik

Erik Josephson is our resident Data Scientist, and has started to explore our library statistically, looking for interesting trends and conclusions.  He will periodically document his exploits here - part for the sake of curiosity and part to inspire you to find new ways to explore AggData on your own.  If you have any experiment ideas for Erik to try, please contact us with your suggestions.

With the recent election, one of the questions that I really wanted to answer was which stores were leaning most toward Obama and which were most towards Romney.  Using the complete list of all zip codes for each store in aggdata.com, we can attach each store location to the county level election results.  Then, by associating those numbers with each store location’s zip code, we get an aggregate of the percentage of people voting for Obama per store.  By adding up the total number of Obama votes as a ratio of total votes, we get a number for Obama’s share of votes per store. 

The test was run on 1,873 stores.  I decided to count only stores that had more than 50 locations, since a store chain with just two store locations in the most liberal part of the country would make it technically the most liberal store chain.  The problem is that this doesn’t mean much because there are so few stores, so I chose 50 arbitrarily, as a rough estimate of when political leanings may be more evident.  This took the list total down to 1,372. 

Here are the top and bottom ten for Obama’s average percentage:

 

# locations

Obama Share

Pay O Matic

93

0.843419921

Duane Reade

255

0.807170966

C-Town

76

0.77647703

Petland Discounts

58

0.771389402

My Sports Clubs

94

0.743944189

Lucky Supermarket

65

0.736208766

Maui Divers

51

0.72848989

Golden Krust

69

0.719103001

Cosi

100

0.715237435

Equinox

53

0.714656017

     

Our Coop

164

0.327055924

Harps Food

64

0.323809726

Taco Mayo

80

0.319134954

Tennessee Farmers Co-op

119

0.318369622

Chicken Express

171

0.312795213

Brookshire's Grocery

118

0.30615403

Arctic Circle

71

0.29167007

Food City

79

0.286288669

Brookshire Brothers

77

0.270405201

United Supermarkets

50

0.251019143

The problem is that what we could be seeing is just a measurement of how urban a store is, since there’s a clear correlation between population density and left leaning tendencies.  To factor this out, we figure out the rough regression line between population per zip code and Obama share per zip code, and we get something like this:

Using this formula, we calculate the residuals per store location to get a measure of how far above or below the expected Obama share, given the zip code population.  Thus, the higher the residual, the more liberal the store is, independent of the urban bias.  With these adjusted values, here are the top and bottom ten of that list again:

 

# locations

Residuals

Duane Reade

255

0.269663448

Pay O Matic

93

0.251994728

Cosi

100

0.234228975

My Sports Clubs

94

0.232823168

Maui Divers

51

0.227038125

Au Bon Pain

188

0.223941494

Petland Discounts

58

0.215109522

Equinox

53

0.213164318

C-Town

76

0.207240477

Lucky Supermarket

65

0.202645904

     

Game Xchange

56

-0.150127975

Braum's

318

-0.151770819

Taco Mayo

80

-0.154025071

Brookshire's Grocery

118

-0.162281208

Hastings

134

-0.165103232

Chicken Express

171

-0.181092875

Brookshire Brothers

77

-0.185282514

Food City

79

-0.187167318

Arctic Circle

71

-0.214405916

United Supermarkets

50

-0.234106637

This shows how some stores have high Obama percentages mainly because they are urban centered stores.  Famsa, for example, had a high percentage (60.5%) of votes for Obama, but the residual was only slightly above what would be expected, given the urban population where Famsa stores are located.  Thus, Famsa is more of an urban store than an Obama store.  

From all of this, we can conclude that if you had to pick a store chain whose CEO most likely voted for Obama, try Duane Reade.