#MakeAmericasPollsGreatAgain: Evaluating Twitter as a tool to predict election outcomes

Abstract

The 2016 Presidential Election revealed the diminishing accuracy of traditional polling methods as evident by Donald Trump's unexpected win over long-assumed winner, Hillary Clinton. Polls projected a large margin Clinton win in key states Michigan and Wisconsin. Due to their inaccuracy, this project seeks other means of political prediction. Twitter is the most active social media outlet in regards to political engagement, so geotagged tweets from Michigan and Wisconsin from the 2016 election cycle were compiled into individual databases for Clinton and Trump. Wordcloud technology was used for each database to produce images based on the most frequent words from each candidate's tweets. Basic statistical comparison was made, but yielded few results; textual analysis indicated unpredicted factors at play, including the media speculation that Clinton's campaign did not resonate within voters. The results of this study suggest that future studies should shift their focus to qualitative analysis rather than strictly quantitative.