Month: February 2020

I have decided to predict the GOP primary. Although I know Trump will almost certainly win, I want to evaluate my model. Many states have canceled primaries and polling data is highly limited which constrains what I can do.

My data source is FiveThirtyEight. I am also finding the margin of error for those polls so I can do my iterative model. My prior is based on the national polls. I will only predict states with three or more polls after December 1st. Texas may be the only other race that will meet that polling requirement. I only include polls after December 1st because I want to exclude the polls with Sanford, who has since dropped out. The candidates I am tracking are President Trump and Bill Weld. These are the only candidates currently in the race who appear in the polls. There are some other candidates on the ballot in some states that I am ignoring. Joe Walsh ran until last Friday and appeared in many polls. I am assuming that 75% of these supporters will vote for Weld. I believe that if you preferred a long-shot candidate over Trump who then dropped out, you are likely to support any candidate that is not Trump, but there is also a chance you will not vote.

I am using the Gaussian Iterative and Gaussian Polls models from my JSM Proceedings paper. The Gaussian Iterative model is my preferred method, but I had to drop two polls because I could not find their margin of error, which is important to the iterative model.

Predictions: Trump will win all the delegates because Weld will likely not surpass the 15% delegate requirement. Below is a table with my predictions. The last column is the approximate margin of error of this model, which since my model is Bayesian, I am predicting a 95% chance that the result is within that margin.