Different Forecast, Same Result: More Political Science Models

Categories: Uncategorized

Politico’s Dylan Byers created a minor dust-up in the twitterverse today when he posted an article that appeared to take a shot at the New York Times’  Nate Silver’s prognosticating skills. Byers writes, “Prediction is the name of Silver’s game, the basis for his celebrity. So should Mitt Romney win on Nov. 6, it’s difficult to see how people can continue to put faith in the predictions of someone who has never given that candidate anything higher than a 41 percent chance of winning (way back on June 2) and — one week from the election — gives him a one-in-four chance, even as the polls have him almost neck-and-neck with the incumbent.”  Byers’ article almost immediately created pushback from others who pointed out that the win probability is not the same as projecting a popular vote percentage.

That distinction is likely lost on many Romney supporters who have been criticizing Silver’s forecast for some time now.  But while Silver’s highly publicized work has attracted many Republicans’ ire, it is important to realize that several political scientists have developed their own state-based forecast models that are every bit as good as Silver’s and have the added virtue of being completely transparent and which have been vetted by other political scientists.  Those models are, as of today, also forecasting an Obama Electoral College victory.

In my last post I discussed one such model – the one developed by Emory political scientist Drew Linzer and featured at his Votamatic website.  As of today, Linzer’s state-based polling model continues to forecast an Electoral College victory for Obama with 332 Electoral College votes to Romney’s 206. I’ve discussed some of the assumptions built into Drew’s model in a previous post.

Today I want to discuss a second state-based forecast model created by political scientists Tom Holbrook and Jay DeSart.  Their model is even simpler and more parsimonious than Drew’s. Essentially, they look at three variables: the average Democratic vote share in all trial-heat polls in the field during October, the average Democratic share of the two-party support in national polls taken in the October prior to the election, and the average Democratic two-party state vote share in the four previous presidential elections. Like Drew’s model, then, they are incorporating both a long-term factor – the previous state-level election results that provide a window into a state’s ideological leanings – and short-term factors captured by the October state-level polls during the current election year together with the candidate’s standing in the national polls.

In 2008, the model successfully called all but 3 states: Missouri, North Carolina and Indiana.  (It’s not clear to me what they did for the split vote in Nebraska.) This prediction isn’t quite in Sam Wang territory (Sam nailed everything but Nebraska, as I recall), but it’s not too shabby either.

For those of you interested in playing at home, here’s the equation for their 2012 forecast model:

VOTEi = -29.454 + .575(POLL)i + 0.57(PRIOR VOTE)i + .44(NATIONAL POLL).

As of today that model is also projecting an Obama Electoral College victory, but by a closer margin, 281-257, than the Votamatic projection. Despite the closer Electoral College projection, they still estimate that Obama’s win probability is more than 86%.  Here are the current state-by-state projections:

Obama

Romney

State

Win
Probability

State

Win
Probability

Ohio

67.8

Colorado

50.5

New Hampshire

72.0

Florida

52.7

Nevada

74.4

Virginia

55.0

Iowa

76.8

North Carolina

81.4

Wisconsin

83.6

Arizona

92.0

Pennsylvania

88.7

Missouri

95.0

Michigan

93.6

Georgia

97.5

Minnesota

96.1

South Dakota

97.8

Oregon

97.4

Montana

99.3

New Mexico

98.4

South Carolina

99.7

Maine

99.5

Indiana

99.9

Washington

99.7

Kentucky

100

New Jersey

99.8

Tennessee

100

Connecticut

99.9

Texas

100

California

100

Mississippi

100

Delaware

100

West Virginia

100

Illinois

100

Kansas

100

Maryland

100

Nebraska

100

Massachusetts

100

North Dakota

100

New York

100

Louisiana

100

Rhode Island

100

Arkansas

100

Hawaii

100

Oklahoma

100

Vermont

100

Alabama

100

DC

100

Alaska

100

Wyoming

100

Idaho

100

Utah

100

Note that according to these win probabilities, it is more likely that Obama would “win back” Virginia, Florida or Colorado than it is that Mitt can take Ohio. Indeed, those are the three states that Drew currently has in Obama’s column.  That’s a total of 51 Electoral College votes in those three states alone, and it is the difference between Obama winning 281 Electoral College votes versus 332.  So clearly these forecasts are amenable to change, even with the dwindling number of undecideds.  Given the close nature of the race, several states can tip in either direction.  However, both models suggest that, based on current state polling, Romney has a bigger hurdle ahead of him if he is going to pull this out.

I want to stress that these models use the latest polling in each state to project the winner.   As such, they tend to be more accurate than the structural forecast models political scientists issue by Labor Day based on the “fundamentals” that I’ve discussed in several previous posts.  The drawback, of course, is that these state-based models don’t help us understand why people are voting for a particular candidate.  In effect, they use current support to predict future support.  That works well if all one is interested in is predicting the election outcome, but they aren’t very theoretically satisfying.

Note that both the Linzer and the Holbrook-DeSart models are premised on the assumption that state-based polls this late in the year are generally accurate.  Is that a safe premise?  In fact, as John Sides discusses here citing research by Robert Ericksen and Christopher Wezlien, they are.  As this graph shows, the share of the Democratic vote based on polls in the last week of the election closely aligns to the actual vote percentage received by the Democratic candidate in the period 1952-2008. (If the alignment was perfect, the data points would fall directly on the diagonal line.)

So, using conceptually simple (and transparent!) methods, political scientists still see this race as Obama’s to lose. This does not mean, however, that this election is over.  The key – and as yet unanswered – question is whether Mitt can, through a combination of winning over undecided voters and gaining a turnout advantage, rope in the 1-2% more support he needs to flip Ohio, or some other combination of swing states.   Your answer to that question may depend on whether you think there continues to be movement toward Romney, however slight, during these last eight days. If there is, both models should pick this up, and adjust their projections accordingly.  If there is not, however, and this race has entered a stable equilibrium, the odds seem to be in Obama’s favor.

P.S If you are coming to this site for the first time (there’s been a lot of traffic of late) I encourage you to follow me on Twitter at @MattDickinson44 – I tweet all new posts there.

 

2 Responses to Different Forecast, Same Result: More Political Science Models

  1. Tom Campanella says:

    Hi Professor,

    Saw this article on RCP and thought it was right up your alley. Basic premise – how to reconcile the fact that state polls and national polls are showing us two different races. Anyways, pretty good read. Should be interesting to see come Tuesday which is right.

    http://www.realclearpolitics.com/articles/2012/10/31/whats_behind_the_state-national_poll_divergence_115979-2.html

    In your opinion, should we see the two converge in the last few days… is there even enough time for this to possibly happen?

  2. Matthew Dickinson says:

    Tom,

    I think the polls will move closer as the likely voter screen tighten and the polls become more accurate. They may not perfectly align at the state and national level, but I expect them to both tell a similar electoral story regarding who is ahead.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>