Department of
History & Political Science

Presidential Election Forecasting

DeSart and Holbrook Forecast Model

Dr. Jay DeSart, Assistant Professor of Political Science - Utah Valley University
Dr. Thomas Holbrook, Wilder Crane Professor of Government - University of Wisconsin-Milwaukee

UPDATE - The latest run (Nov. 3) of our October Rolling forecast now predicts that Barack Obama will win the election with 53.92%^* of the national 2-party popular vote to to John McCain's 46.08%. The model also predicts that Obama will win the presidency with 349 electoral votes to McCain's 189. Based on these results, there is a 99.994% probability that Barack Obama will win the election. Details

UPDATE: Election Night: The Paths to 270

*NOTE: Given the way our model is calculated, the District of Columbia is not included in the analysis due to its outlier status. Because of this, our national popular vote projections have not included DC. Plugging DC data into the projection model increases the prediction of Obama's share of the two-party national popular vote by 0.1.

October 3, 2008 - The election forecast model we have developed predicts a victory for Barack Obama over John McCain on November 4th.

The model, which is based on state and national polling data as well as state historical voting patterns, predicts that Obama will receive 51.81% of the national 2-party popular vote to John McCain's 48.19%. Furthermore, the model predicts that Obama will win the electoral college vote, 336 - 202.

In March 2008, we presented a paper at the annual conference of the Western Political Science Association in which we presented the updated model that we used in generating this forecast. The analysis that we put forth in the paper looked at data from the elections from 1992 - 2004, and shows that the model was able to correctly predict 90% of state outcomes, predict the national popular vote to within an average of 0.5% of the 2-party vote, and predict the Electoral College vote to within 23 Electoral Votes.

Here are the state predictions, along with their respective win probabilities, as projected by the model:

Obama		McCain
State	Win Probability	State	Win Probability
Colorado	52	West Virginia	53
Missouri	57	Virginia	66
Nevada	60	Arkansas	66
Florida	61	North Carolina	83
Ohio	62	Louisiana	87
New Hampshire	75	Indiana	90
New Mexico	87	Arizona	97
Wisconsin	89	Montana	98
Pennsylvania	92	Georgia	98
Michigan	96	Kentucky	99
Minnesota	96	Tennessee	99
Iowa	97	Mississippi	99
Oregon	97	South Carolina	100
Washington	97	Texas	100
New Jersey	98	South Dakota	100
Maine	100	North Dakota	100
Delaware	100	Alabama	100
California	100	Kansas	100
Connecticut	100	Oklahoma	100
Illinois	100	Alaska	100
Maryland	100	Nebraska	100
New York	100	Wyoming	100
Vermont	100	Idaho	100
Massachusetts	100	Utah	100
Rhode Island	100
Hawaii	100

Based on these state win probabilities, Barack Obama has a 97.5% probability of winning
the necessary 270 Electoral Votes he needs in order to win the election.

Other Academic Models

These are the forecasts of the 2008 Presidential election that
have been generated by various academic forecast models:

Date of forecast	Model	Predicted Dem. 2-Party Popular Vote
8/27/2007	Allan Lichtman - The Keys to the White House	54.0
1/15/2008	Helmut Norpoth - The Primary Model	50.1
6/7/2008	Douglas A. Hibbs, Jr. - "Bread and Peace" model	51.8
7/28/2008	Carl Klarner	53
7/31/2008	Ray Fair	51.5
8/1/2008	Robert S. Erikson and Christopher Wlezien	52.2
8/2/2008	Alfred Cuzan and Charles M. Bundrick	52.0
8/5/2008	Brad Lockerbie	58.2
9/1/2008	Thomas M. Holbrook	55.5
9/1/2008	Michael S. Lewis-Beck	50.1
9/2/2008	Alan Abramowitz - "Time for Change" model	54.0
9/3/2008	Andreas Graefe and J. Scott Armstrong	51.2
9/8/2008	James Campbell	47.3
10/3/2008	DeSart and Holbrook	51.8
10/10/2008	Jerome and Jerome	53.3
Average Forecast		52.4

Composite Polls

These measures attempt to arrive at a more accurate estimate of the candidates' standing in the polls by using a composite average of several polls. By averaging several polls they employ the principle of the central limit theorem. While any one poll is likely to have a certain amount of random polling error, averaging multiple polls theoretically mitigates that problem because the errors found in each poll will work to cancel each other out.

Date	Poll	Dem. Share of 2-Party Popular Vote
11/3/2008	RealClearPolitics	54.1
11/3/2008	FiveThirtyEight	53.0
Average Poll		53.6

While the RCP measure simply averages the polls, FiveThirtyEight employs a complex weighting scheme to arrive at its result.