Does the mystery model generalize?

I recently found a simple statistic that distinguishes with near-perfect accuracy between states that voted for Clinton and states that voted for Sanders. It would be interesting to see whether it generalizes to the Republican primary, or to the 2012 general, though I doubt it will.

State ??? 2016 Dem primary 2016 GOP Primary 2012 General
Mississippi 37.3% Clinton Trump Romney
Louisiana 32.4% Clinton Trump Romney
Georgia 31.4% Clinton Trump Romney
Maryland 30.1% Clinton Trump Obama
South Carolina 28.5% Clinton Trump Romney
Alabama 26.4% Clinton Trump Romney
North Carolina 21.6% Clinton Trump Romney
Delaware 20.1% Clinton Trump Obama
Virginia 19.9% Clinton Trump Obama
Tennessee 16.8% Clinton Trump Romney
Florida 15.9% Clinton Trump Romney
Arkansas 15.8% Clinton Trump Romney
New York 15.2% Clinton Trump Obama
Illinois 14.9% Clinton Trump Obama
New Jersey 14.5% N/A N/A Obama
Michigan 14.2% Sanders Trump Obama
Ohio 12.0% Clinton Kasich Obama
Texas 11.9% Clinton Cruz Romney
Missouri 11.5% Clinton Trump Romney
Pennsylvania 10.8% Clinton Trump Obama
Connecticut 10.3% Clinton Trump Obama
Indiana 9.1% Sanders Trump Romney
Nevada 9.0% Clinton Trump Obama
Kentucky 8.2% N/A Trump Romney
Massachusetts 8.1% Clinton Trump Obama
Oklahoma 8.0% Sanders Cruz Romney
Rhode Island 7.5% Sanders Trump Obama
California 6.7% N/A N/A Obama
Kansas 6.2% Sanders Cruz Romney
Wisconsin 6.1% Sanders Cruz Obama
Minnesota 4.6% Sanders Rubio Obama
Nebraska 4.5% Sanders N/A Romney
Colorado 4.3% Sanders Cruz Obama
Alaska 4.3% Sanders Cruz Romney
Arizona 4.2% Clinton Trump Romney
Washington 3.7% Sanders N/A Obama
West Virginia 3.6% Sanders N/A Romney
Hawaii 3.1% Sanders Trump Obama
New Mexico 3.0% N/A N/A Obama
Iowa 2.7% Clinton Cruz Obama
Oregon 2.0% N/A  N/A Obama
Wyoming 1.3% Sanders Cruz Romney
Utah 1.3% Sanders Cruz Romney
New Hampshire 1.2% Sanders Trump Obama
South Dakota 1.1% N/A N/A Romney
North Dakota 1.1% N/A N/A Romney
Maine 1.0% Sanders Cruz Obama
Idaho 1.0% Sanders Cruz Romney
Vermont 0.9% Sanders Trump Obama
Montana 0.7% N/A N/A Romney

I’ve left out Nebraska and West Virginia from the GOP side, because those races had only one candidate.

The general election is clearly regional; our mystery statistic doesn’t appear to be relevant. The GOP primary is more interesting: the only states Trump lost above our 8.0% cutoff are Ohio (Kasich’s home state) and Texas (Cruz’s), but he won seven states below the cutoff: Rhode Island, Nebraska, Arizona, West Virginia, Hawaii, New Hampshire, and Vermont. Of these, Rhode Island, West Virginia, New Hampshire, and Vermont are East Coast states, and Trump has won every East Coast state except Maine; of the rest, Arizona is an exception for the Democrats as well, for obvious reasons that would be taken into account in a slightly more complex version of this model.

Another way to put this is: above our 8.0% cutoff, 10 Trump states voted for Obama in 2012 and 12 voted for Romney, but below it, every Trump state but Arizona went to Obama. (Below the cutoff, 6 non-Trump states went to Romney and 5 went to Obama.)

So: the mystery model appears to hold for the GOP primary, but there’s a regional effect on top of it.

7 responses to “Does the mystery model generalize?

  1. Alrenous May 12, 2016 at 12:29 am

    Not much of a mystery which statistic is headed by ‘Mississippi.’

    The regional effect is probably severely affected by gerrymandering. Primaries can’t be gerrymandered, can they?

    • nydwracu May 12, 2016 at 1:11 am

      Primaries are usually by county. I’m working on county-level Democratic primary analysis now, because I’d like to be able to call Kentucky. Kansas, Minnesota, and Alaska are by congressional district, so I’m ignoring them. On the Republican side, Wyoming is by groups of two counties, which is sensible: on the Democratic side, about half of the counties in Wyoming were won by a margin of zero or one.

      • nydwracu May 12, 2016 at 3:13 am

        I’m using NYT’s data, which reports >20k voters in Polk County, Iowa.

        Wyoming is probably wrong, but I discarded everything with <1000 voters.

        I did the math on the simple model. It says Sanders wins Kentucky. I wouldn't bet on it either way, but there you have it.

  2. Pingback: Does the mystery model generalize? | Reaction Times

  3. Pingback: Lightning Round – 2016/05/18 | Free Northerner

  4. Pingback: This Week in Reaction (2016/05/15) - Social Matter

  5. Pingback: The mystery model: 2016 is not 2008 | nydwracu niþgrim, nihtbealwa mæst

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

%d bloggers like this: