Authors: Kathy Lin, Jie Lu, Tin Oreskovic, Jake Snyder
Polling is one of the most robust methods to capture and understand public opinion, especially in politics. However, as with all sampling methods, there are limitations and biases, particularly with regard to the questions asked. We investigate a new question format for political opinion polls with the goal of improving the effectiveness and predictive power of these polls. Using a mobile polling strategy, we ask respondents demographic information, as well as who they plan to vote for in the 2018 midterm elections. Our research uses multilevel regression and poststratification to control for the well-known biases within polling data, such as sampling bias and nonresponse bias. We use Bayesian methods to predict two-party share of the 2018 midterm Congressional vote. Our research shows that, with respect to predictive power, the new question format does not differ much from the traditional format. Since the new question format requires more time and money than the traditional format to implement, it does not seem worth the investment.
For the full report of our research, please read our final report.