The 2016 US Primary Elections and Twitter: A methodological study of econometrics, machine learning and their intersection

Sammanfattning: In this thesis modern methods in using social media data from Twitter to predict the 2016 US primary elections are investigated. The collected data is processed via a simplified sentiment analysis and later used in modelling election results with less traditional linear models and methods within statistical learning. This investigation uses a limited methodology in sentiment analysis since a rigorous analysis would require methods within natural language processing, which is a thesis topic on its own. The methods used in this thesis achieves significant results and there is potential for improvement using more exact analysis. Further, it is clear that the more refined methods yields more precise estimates. In conclusion, the results suggests that the approach is highly plausible for future research and under less bold assumptions. This is concluded from significant results despite a relatively simplified approach.