Forecasting Political Executive Approval Using Online Engagement Data

• In Event: Social Media's Impact on Campaigning, Election Outcomes, and Public Opinion

Abstract

Recent studies have tried to forecast electoral outcomes and public opinion trends using social media data with mixed results. Particularly relevant have been supervised machine learning models trying to predict sentiment within texts. This paper employs a novel approach to forecasting modeling. Using data from weekly presidential approval polls of 11 executives in nine countries---from Asia, Europe, North America, and Latin America--, and their respective Twitter accounts, I have trained an unsupervised machine learning model that predicts weekly changes in presidential approval based on online engagement data—the weekly average of retweets, retweets without quote, and favorites. After testing the model, it shows more accuracy in some political contexts than others. However, this approach represents a promising starting point that might be improved as more data become available.

Author

• Juan Salvador Gomez-Cruces, University of Potsdam