Friday, October 19, 2012

Another election prediction

There are a lot of formulas that predict the outcome of presidential elections from economic conditions.   One of the best, in my opinion, was devised by Ray Fair, an economist at Yale.  His equation has three economic variables:  growth of per-capita GDP in the first three quarters of the election year, change in prices over the last fifteen quarters, and number of quarters with strong economic growth (over 0.8%) over the last fifteen quarters.  There are also variables for incumbency and how long the party has been in office. 

Fair has estimated the equation using election results for the whole country.  It occurred to me that it would probably be better to omit the south.  Until recently, voting in the south had very little to do with economic conditions--it was about race and religion--so including the South is basically just adding "noise."   While I was at it, I decided to break the US down into regions and see if they reacted the same way.  I divided states into Northeast, Midwest, Plains/Mountain, and Pacific.  The results, where the dependent variable is the incumbent party's share of the two-party vote:

              Incumbent   Office Growth  Prices  Good 1/4   R2
US           2.4   -4.9  .82  -.69   .74   .281
Omit South   3.3   -4.6  .59  -.70  1.58   .445
East         3.6   -4.0  .14  -.58  2.34   .418
Midwest      3.4   -3.9  .51  -.65  1.50   .563

Plains/Mtn   3.0   -5.9 1.01  -.87   .73   .590
Pacific      3.8   -4.0  .67  -.76   .24   .445

The R-square (predictive power) increases substantially when the south is omitted.  The advantage of incumbency becomes larger, as does the advantage of having periods of strong growth.  Comparing the regions, there is an interesting pattern:  in the East and Midwest, the number of good quarters is more important, while in the Western and Pacific states, recent growth is more important.  Maybe westerners have a more short-term orientation?  That seems to fit with popular images of the differences between parts of the country. The regional differences (if they exist) are probably slightly beneficial to Obama--his record is OK in terms of recent growth, but very weak in terms of number of good quarters.  The Fair model for the US predicts a close race.  The regional differences suggest Obama will be a bit weaker than expected in the East, but stronger in the West, which includes more swing states--a few more votes in places like Colorado or New Mexico might make the difference. 

Note:  My estimates for the US are a bit different from Fair's, since my dependent variable is vote by state and his is the vote for the whole US (in effect, the states are weighted by the number of voters). 

No comments:

Post a Comment