Predicting the Presidential Election
by Chris Bowers, Fri Jun 11, 2004 at 07:48:06 AM EDT
Helmut Norpoth, a professor at SUNY Stony Brook, uses a model that predicts the winner of the general election based on results during the primary season.
Ray Fair of Yale uses a purely macroeconomic model. His model regularly predicts that Bush will win a 1964-1972-1984 type landslide. This is understandable, since candidates with 51% unfavorables facing opponents with 33% unfavorables usually win 61% of the vote.
Allan Lichtman has a model based entirely on how well the incumbent governs. He claims its mathematical, but it seems a riddled with arbitrary abstractions to me.
Alfred Cuzan and Charles Bundrick of the University of West Florida have another model based upon the perceived relationship between economic indicators and election results. The methodology behind this model is quite sizable, and I have not had time to dig through it yet.
The Iowa Electronic Markets use a simulated stock market model.
J. Scott Armstrong of the University of Pennsylvania, has developed a model based upon poll averages and predictor model averages. Well, at least predictor models developed by professors. He probably wouldn't touch amateurs such as Scott Elliot and myself with a ten-foot poll.
Scott Elliot uses a model much like my own, where only the partisan index and current polling data are used in the calculation. His is the only model that projects Kerry in a better position than the General Election Cattle Call.
Randall Jones of the University of Central Oklahoma offers advice on how to predict election results.
It particularly annoys me that a University of Pennsylvania professor named J. Scott Armstrong seems oblivious to the existence of my model. Not only does he have almost exactly the same name as the founder of MyDD, but also I live only three blocks from Penn's campus in University City, Philadelphia.
Why do I think my model will do better than those produced by experts? For starters, I reject the notion that any Presidential election before 1948 has any instructive relevance to contemporary Presidential elections. With wildly different electoral structures, Jim Crow still in full-force, no gender suffrage until 1920, highly different immigration patterns, no television, galacticly different nationwide educational opportunities, nearly unrecognizable industry distribution of the workforce, and no reliable polling information, my assumption is that any similarities between pre-1948 and post-1948 are purely coincidental. The forces that drive elections now are not the forces that once drove them.
My second assumption is that favorable / unfavorable polls are better indicators of the future of the campaign than economic models. This has to remain an assumption, since there are insufficient historical data points to demonstrate this claim one way or the other. However, I simply do not trust most broad economic indicators as useful gauges of the national mood concerning the economy.
My third assumption is that while they never seem to be directly on target, polls are usually pretty good indicators of the state of the race, and no better indicators are available. Also, Gallup does not have a lock on relevant polls, although many political scientists are loathe to use anything else since only Gallup has the long-term history needed to develop sufficient data points that can be used in a reliable study.
Still, over the weekend I will start to add some of these models to the President 2004 page, so that my projections are not the only ones available on MyDD. Welcome to Election Prediction central.