Is Twitter Sentiment A Good Predictor of Election Results?

Our team at Content Harmony paired up with WiseMine to take a look at how Twitter volume and sentiment for 10 of the top 2014 midterm races compared with actual election results. Our goal was to see if Twitter sentiment positively correlated with the Midterm election results.
In the graphic below, you’ll see volume of tweets we recorded over a 10-day period leading up to November 4th for each candidate’s twitter handles and campaign website. Next to the volume chart, you’ll see the percentage of positive versus negative sentiment tweets that we found using WiseMine’s sentiment analysis algorithm. We removed neutral sentiment tweets, which averaged 40-50% of tweet volume for most candidates.
Next, we made a basic prediction of winner for each race, based upon comparing pure number of positive tweets per candidate. While this obviously isn’t a great proxy, we wanted to see how a simple analysis would compare to traditional polling numbers.
We found that this calculation of Twitter sentiment matched FiveThirtyEight.com’s Senate Race Forecasts in just 40% of the top 10 races that we looked at.
The TLDR? They seem pretty poorly correlated.
Keep an eye on the WiseMine blog later this week to see the rest of the data we collected and which metrics correlated well with actual results!
U.S. Senate Election 2014: Democrats Win by a Large Margin … on Social Media | Wisemine Blog
[…] In fact, if you were to predict the outcome of the races, purely based on social media favorability, then about 80% of democrats would have won their races. Here’s an infographic showing these results on the Content Harmony blog: [Infographic:] Is Twitter Sentiment A Good Predictor of Election Results? […]