What is the Correlation between Twitter, Polls and the Popular Vote in the 2012 Presidential Election?

Abstract

This paper contributes to the ongoing discourse on social media and elections, by evaluating Twitter activity during the 2012 United States presidential campaign. We specifically ask: what is the correlation between Twitter, polls and the popular vote? In response to current debates on methods, we evaluated multiple configurations of Election Day Twitter activity. The Twitter activity in our study included a commercial proprietary sentiment analysis, dictionary-based sentiment analysis, and volume of messages analyzed at multiple time periods. We normalized all data by converting values into binary proportions for the two major candidates. Bivariate statistical analysis using a chi-square test provided comparative analysis. Our findings suggest that Twitter data analyzed with sentiment analysis better reflects the popular vote than the volume of messages. For further analysis, we compared the binary proportions of the positive sentiment data. Having a Mean Absolute Error (MAE) of 3.11% and a 1.7% point-bias, the Twitter activity analyzed with the proprietary sentiment analysis was within the same proximity as public opinion polls that accurately reflected the popular vote.