Weekly Updates

Week 1: Uncovering Swing States

Before taking a deep dive into election result predictions, I thought it would be of great interest to investigate some of the parallel universes, or contributing factors if you would like, of election night results. This week, we look at swing states and how we can identify those based on historical trends, and investigate the effect of turnout to the popular vote results.

Week 2: Pennies and Quarters

This week, we start creating our election prediction model by comparing economy data with popular vote results. In particular, we will make and compare a national popular vote prediction model based on the second or third-quarter GDP growth percentage of election years.

Week 3 - Grading Election Polls

This week, we make a new prediction model by comparing FiveThirtyEight’s election poll rankings. In particular, we will make a national popular vote prediction model based on the scaled results of each poll, influenced by FiveThirtyEight’s election poll rankings.

Week 4 - Trump and Incumbency

This week, we make a new linear regression prediction model based on the GDP Quarter Growth we saw in Week 2, and the net approval rating for each incumbent party candidate and use those to predict the popular vote share.

Week 5 - Probability and Electoral Votes

This week, we take a different approach than the typical linear regression model fitting approach we have been following for the last few weeks. Instead, we employ Binomial logistic regression and think of the election outcome for Democrats as a finite draw of voters from the voter-eligible population (VEP) turning out to vote Democrat, modeled as a binomial process.

Week 6 - Considering Turnout

This week, we go back to the different approach than the typical linear regression model fitting approach we followed last week. This time, incorporate a turnout correction for the voter-eligible population (VEP) and then we employ Binomial logistic regression and think of the election outcome for Democrats as a finite draw of voters from the VEP turning out to vote Democrat, modeled as a binomial process.

Week 7 - Considering Turnout Vol. II

This week, we go back to the different approach than the typical linear regression model fitting approach we followed in the last few weeks. This time, we update our turnout correction we created last week by incorporating COVID death rate changes in every state. Then, we calculate the turnout correction for the voter-eligible population (VEP) and then we employ Binomial logistic regression and think of the election outcome for Democrats as a finite draw of voters from the VEP turning out to vote Democrat, modeled as a binomial process.

Final Election Prediction

The election is right around the corner, and we are ready to make our final prediction for the 2020 General Presidential Election. This time we use a three-step model to grapple with the idiosyncrasies of each state. Using our model, we determine the popular vote winner in each state, along with the winner of the Electoral College vote.

Final Election Prediction Reflection

Almost twenty days have passed since Election day, a President-Elect has been called by most networks and the Associated Press, and the country is finally moving on. However, we are taking the time to evaluate our model, understand its shortcomings and propose ways to further develop it for use in future elections.

Election Narrative Analysis

This final week, as states around country certifies the final election results, we move past the analysis of our predictive model, and grapple with narratives presented by media and election analysts. In particular, we formulate a narrative about “nationalized” elections and test out county-level data to understand if this narrative warrants further investigation.