Presidential Election
Summer 2015
Title: Modeling the 2012 Presidential Elections Battleground States
Student: Evan Cofer
Advisors: Dr. Eddy Kwessi, Dr. Hoa Nguyen, Dr. Albert Jiang, and Dr. Katsua Nishikawa
Abstract:
The goal of this project is to build a model for prediction of presidential elections based on polls that are available at the time of the election, specifically, starting on a chosen day after the conventions of the two majors political parties. Contrary to existing models in the literature, this model distinguishes between polls collected by different firms and uses Monte Carlo methods to construct polling data for a given firm on a day when the firm has no new polls. Additionally, this model foregoes using uniform swing – the assumption that national polling dictates state polling – and national polling data in favor of state-by-state polling data. Conditions specific to a state, such as a candidate’s home state advantage, are therefore reflected in the collected data rather than in the model, allowing for more abstraction. Polls among likely voters are used rather than as registered voters’ polls, as likely voters most readily translate into actual voters. States were limited to a select few – the so-called battleground states – where no party has a clear advantage. Missing polls are generated under the assumption that changes in polls are determined by the events of that day and are therefore typically moderate but otherwise unpredictable. This day to day change is represented by adding Ɛ, a standard normal distributed random number generated using the Mersenne Twister. A dynamic bootstrap (B = 10,000 samples) is performed on the last day of the election to produce a probability density function for the election outcome. The model is considered successful if the state’s actual popular vote falls within a 95% bootstrap bias corrected and accelerated (BCA) confidence interval. The seed for Ɛ is then shuffled and the whole process is repeated (N = 2,000) and the median of all estimated value is chosen as an estimate of the true poll for the candidate. Biases and mean square errors for different values of N are reported.
--------------------------
Fall 2013
Title: Predicting the presidential election
Students: Crystal Nguyen and Matthew Pearce
Advisors: Dr. Dante Suarez, Dr. Eddy Kwessi and Dr. Hoa Nguyen
Abstract:
The presidential election is decided by the summation of each state’s Electoral College votes, so in order to accurately model the presidential election we must examine the votes of the individual states. Our model uses constrained regressions to model the percentage of votes received by a candidate in a state, with the independent variables being various demographics as well as election poll data. We input these data into the software package Stata in order to run regressions on them. An R code then runs Monte Carlo simulations on this model, using the prediction as a mean and a weighted standard error of the poll data as the standard deviation. We validated our proposed model by comparing it alongside the 2012 campaign. We then collected projected demographic data for 2016 to estimate the next presidential election. Our model produced good fit and supports the outcome of the 2012 election, given state polling data.
--------------------------
Fall 2012
Title: Mathematical modeling of presidential elections and changes in states' political polarity
Students: Christopher Alexander, Ethan Krohn, Vanessa Moreno, Selman Kaldiroglu
Advisors: Dr. Dante Suarez and Dr. Hoa Nguyen
Abstract:
Mathematical models to describe social phenomena have been developing throughout modern history on at least two fronts: as a result of changes in social theories and as a result of changes in modeling methodologies and technology. However, mathematical modeling must ultimately be about reflecting the realities of the world, so the main focus of this work is to develop a way of mathematically expressing all the aspects of reality that are relevant for deciding the outcome of a presidential election and how different states change their voting tendencies (democrat or republican) over time. We use least-squares regressions to mathematically model and understand how demographic parameters such as race, sex, age, etc. affect people's voting and a state's change in its political leaning. The results of this study will provide a simple yet powerful way to analyze the complexities of the democratic system in choosing government representatives.
Student Final Presentation: Final presentation.pptx
Student Final Report: final paper.pdf