American Presidential Elections

Background

Methods to predict the winner of American Presidential Elections have been studied for some time. As discussed in Gelman & King (1993), most of the earlier models predicted the popular vote. However, in both the 2000 and 2016 elections, the popular vote winner did not win the presidencies. The Electoral College is what ultimately decides the winner of the election. In most states, the electors from that state vote for the winner of that state. Increases in both the quantity and quality of state level polling data has made it easier to predict the winner of presidential elections in individual states. There has also been an increase in poll data for Senate and House races. There are several notable groups that predict American

If a senate candidate is judged to be a minor candidate and is expected to get less than 5% the vote, they will not be included in the prediction. This is because it is very difficult to predict the support of minor candidates, since minor candidates often don’t appear in the poll data. Most of the senate races in 2018 are expected to have two major candidates: a Republican and a Democrat. There are a few races (Vermont & Maine) that currently have three major candidates. Since the 2018 election is a few months from the start of the fellowship, the prediction will use the Gaussian conjugate prior. The Gaussian conjugate prior is a relatively simple calculation that would be easy to quickly implement. Huffington Post’s Pollster will be used to provide the polling data. All races will have a poll data from another race to act as a the prior. The following factors will be considered when prior are chosen: incumbency, political ideology of the candidates, political ideology of the state, and major issues of the state. Current events and the favorability ratings of the President, Congress, and the candidates will be monitored throughout the election cycle. The poll data will be adjusted for agency bias using Five Thirty Eight’s pollster rating. Since most polls include undecided respondents, each poll will be normalized so that the proportions sum to 1. The first prediction will be run and

I made a weekly blog post about statistics and the presidential election. I also presented my research at a nearby community college to a public audience. This project would take a similar approach and also have a blog. During the campaign seasons (August 2018-November 2018, January 2020-November 2020), the blog will have weekly updates. During the off-season posts will not be regularly made, but occasion updates on progress of the research will be made. The exact methodology will be posted to the blog, and computer programs will also be made available after the elections. Senate and Presidential elections are important events in American culture, and this research could help promote statistical literacy. Better methods to analyze poll data can be used to help estimate public opinion on certain issues. This research would also contribute to the field of applied Bayesian statistics. Many other fields including pharmacology have similar issues with disparity of the amount information between various similar objects or groups. The group with the most information could serve as a prior and then the estimated proportion would be adjusted with the data of the group being studied. Approaches similar to the one proposed could help create better estimations of proportions and better ways to accurately compare the estimated