Selection and Analysis of U.S. Election Forecasting ModelsPublic Deposited
Forecasting U.S. elections has been a field of interest for many researchers, with various statistical and mathematical models being proposed. In my research, I analyzed a prior election model, the SIS election model. In this model, a system of differential equations traditionally used in epidemiology to study disease transmission (but also applied to social applications in the past) was instead adjusted to forecast elections. The parameters of the SIS election model, the core of my research, represent state-to-state voting dynamics. For example, one parameter might describe how Democratic voters in Illinois might influence undecided voters in Florida. Another parameter might describe how Republican voters in Ohio influence undecided voters in Texas. My first objective was to study these parameters across the 2004-2016 presidential elections and observe how state-to-state voting relationships change under various election contexts. I did this by applying statistical and data-visualization methods to the model parameters for state-to-state relationships. My second objective was to implement a new means for finding these parameters by applying statistical-modeling techniques to polling data. For my first objective, my work suggests that the voting dynamics of 2016 were significantly different than those in the 2004, 2008, and 2012 presidential races. In addition, I found that Florida, Ohio, and Pennsylvania are key, influential players in elections. For my second objective, I applied statistical methods in Matlab to determine the model parameters more quickly, cutting down the time from a few hours to a few minutes. Ultimately, my findings endorse the use and development of statistical and data-driven methods for analyzing and identifying parameters in election-forecasting models.