South Carolina

In democracies around the world, researchers have used multiple methods to monitor elections using Twitter. These methods, however, have been limited. They often account only for the number of and keywords in relevant tweets. Our research team at the University of South Carolina is working to change this. The team is analyzing support on Twitter for the 2016 presidential candidates. It defines candidate support by four criteria: number of tweets, rhetorical tone, geographical location, and topic. By supplementing preexisting techniques, the team is not only able to say which candidates Twitter users regard more favorably—or unfavorably for that matter—but also where and why Twitter users favor that candidate. These results are promising for politicians and researchers alike hoping to influence and understand public opinion.

This time we shift our attention to South Carolina. South Carolina has been largely a “red state”, and we found interesting differences between public opinion among SC Twitter users and those from NV. An analysis of 111,306 tweets gathered between November 2015 and mid-February 2016 revealed that Trump and Clinton took lead in the SC Twittersphere, like they did in NV. However, quantitative and qualitative differences between candidates’ popularity are less stark, showing a more tightened-up competition.

In aggregate terms, Trump was mentioned most frequently with 65,571 tweets, 61.1% of which were positive. Clinton received the second most tweets (16,545) followed by Rubio (10,874), Sanders (9,220), Cruz (5,865), and Bush (2,398). Attention, however, is not always a good thing. Regarding the overall tone of tweets, Sanders’ (69%) was more positive than Clinton’s (59.5%). Results varied on the Republican side. Cruz (57.1%) and Rubio (52.5%) earned mostly positive tweets, while Bush elicited 52.5% negative tweets.

If we look at month-to-month developments in the Democratic race, we will be seeing a 180 reversal of the NV picture. In SC, Clinton’s aggregate tweet count was larger, while Sanders’ monthly count was more consistent. Sanders have a steady tweet count around 2,000 for all four months, while Clinton lost her momentum in December and January with a 2,000 decrease in tweet count. Moreover, Clinton’s positive tweets declined by 5.0% between January and February, and Sanders’ rose steadily.

A different picture also showed up in SC for the GOP competition. While Trump has been largely predominant in both tweet count and tweet positivity in other states among Republican candidates, his positive tweets dropped considerably in December (by 10%) and February (by 13%), making him the least liked Republican candidate currently in SC Twitter. Formerly dominating in tweet volume, Trump received half the number of tweets in February as he did in December. This trend contrasts the increase in tweets related to Cruz and Rubio.

Additionally, Republican candidates seem to take turns to lead positivity in SC Twitter. Trump had the highest positive rate in November (67.3%), Rubio then took the led in December (78.0%), Cruz surpassed all of them with 77.1% in January. And now in mid-February, we see a strong coming-back of Bush with a 60.4% positive rate, which wins him the second place. Whether or not these tweets come from the real electorate, it is clear that the Republican race has been fierce in SC.

These findings offer a glimpse of the insight that our team’s techniques can produce. While it is too early to say whether these techniques can predict elections, it is clear they can help. Since President Obama pioneered social media politicking in 2008, the strategy has become increasingly prevalent. These techniques thus offer a way to clarify the complex web of politics.