Approval Ratings and Predicting United States Presidential Elections

Abstract

Predicting the outcomes of United States Presidential Elections has become a favourite project of many data scientists and academics. Creating a model which can accurately predict who the next president will be is a sure-fire way to solidify one’s name as a reputable source of analysis. The objective of this research paper was to look at various data, and see if there are any important variables which modellers have ignored. A dataset was organized, which included many variables, including economic indicators like GDP growth, average presidential approval ratings, and changes in party composition in the House of Representatives. This included information for every presidential election from 1948 to present. The dataset was then bootstrapped and analyzed through a model which used gradient boosted trees. The findings of the model found a surprisingly high significance for the variable of average approval rating. In 5546 out of 10,000 bootstrap runs (55.46%), this variable was the most significant in predicting the outcome of the election. Given the model’s accuracy rating of 79.52% on test sets, there is validity to the claim that average approval rating is an important predictor of election outcome (though not by itself, as interaction with the other variables was necessary for predictions). The model was then used to predict the outcome of the 2020 presidential election, based on current data. After running the model 10,000 times with the 2020 election as the test set, it was found that Donald Trump was re-elected in 366 trials, indicating a 3.66% chance of re-election given current data (which, especially regarding the economic indicators, is subject to change). A sensitivity analysis of the average approval rating found that for his odds of re-election to exceed 50%, his average approvals must climb to at least 60%. Based on the findings of this report, it can be concluded that approval ratings, while generally ignored or minimized in other models, are actually a very significant indicator of election outcome. If this is correct, president Trump is facing serious challenges in 2020, though with one year left, and economic data left to be reported, his odds could change.