New Data: Can Twitter Predict Elections?





With the 2010 midterm elections coming up on Tuesday, I decided to look into the correlation of candidates Twitter accounts and their recent performance in polls.

I gathered a random sample of 30 senate, house and governor races from RealClearPolitics database of recent election polls and gathered each the number of followers had by each candidate in each race. I used the Twitter accounts linked to by the candidate’s campaign websites (as many of them have multiple accounts, I used the official 2010 campaign accounts). The poll data I used was the RealClearPolitics average, which is an average of recent polls of likely voters from multiple sources.

I found that in 71% of races, the candidate with the most Twitter followers was ahead in the polls.

I also graphed the correlation of “winning the Twitter battle” (having more Twitter followers than your opponent) with “winning the poll battle” (polling ahead of your opponent) and found that while there is a fair amount of variation, there may be a significant amount of “predictiveness” in Twitter following comparisons between candidates.

This research is just a first step into understanding the correlation between election performance and social media usage, but I think it indicates there is an important relationship at work.

If you liked this post, don't forget to subscribe to my RSS feed or my email newsletter so you never miss the science.