Anticipated Date of Graduation

Spring 2024

Document Type

Thesis

Degree Name

Master of Science in Mathematical Sciences

Department

Mathematical Sciences

First Advisor

Doug Darbro

Abstract

The error between presidential polling and the results in recent elections have been abnormally large compared with relatively better polling in the decades prior. The science behind how to create a political poll remains difficult given the impossibility of knowing the sampling frame of the election before its occurrence. The public wants to know who leads and where. Media wants to provide the details and makes its money doing so. Political organizations spend hundreds of millions of dollars and countless hours of human capital because polls pointed them in a certain direction. The public and its relationship to democracy goes hand-in-hand with the idea of a fair election, and polls play a large role in it. This study investigates why polls have been missing recently by looking at key characteristics of the composition of nearly 200 polls. The researcher examines polls from presidential elections in 2012, 2016, and 2020 and compared the absolute difference of the poll and the election results to the corresponding year. Then that difference was predicted using a multiple linear regression method with seven independent variables: age, race, education, proximity to election date, poll margin of error, undecided vote share, and poll mode. Results revealed that days away from the election had statistically significant results, and that mixed-methods samples were nearly significant when compared to phone-only polls. These results imply that pollsters should continue to publish more polls as the election draws nearer. Additionally, it would be worth looking in-depth about poll modes, as mixed mode could have some relationship to the current threat of nonresponse that has recently plagued polling firms.

Download

Share

COinS