After poring over 460,000 tweets from 687 candidates who ran for national office and governorships in the 2010 midterm election, University of Michigan researchers have concluded that conservative candidates (who made significant political gains last fall) made more extensive and coherent use of Twitter than their opponents.
The research team gathered and analyzed the language used in three and half years worth of candidates’ tweets leading up to the election–including crawling the content of 233,000 outgoing links mentioned in the tweets. They found that Republican candidates tended to focus more narrowly on economic messages.
Words like “spending,” “bills,” “budget,” “WSJ,” and “deficit” were among their most popular terms.
Democrats, who tended to post less frequently, spread their 140 characters over a broader range of topics including “education,” “jobs,” “oil_spill,” “Afghanistan,” and “reform.”
Self-identified Tea Party tweeters took things to a new level, exhibiting what researchers referred to as an “organized campaign”–despite their grassroots ethos. They posted almost twice as frequently as Democrats (and substantially more than other Republicans); followed and re-tweeted each other more often; replied to more tweets; and more commonly used hash tags to help organize discussion threads. They also tended to use Twitter as a platform to call out high-profile Democrats by name. Top Tea Party terms included “Nancy_Pelosi,” “Barney_Frank,” “Clinton,” and “Obamacare.”
The study also suggests that candidates’ behavior in the Twittersphere could help forecast election outcomes.
The team was able to predict–with 88 percent accuracy–the winner of a race by looking at the content of the candidates’ tweets, how many followers they had, and whether they were incumbents.
Although the researchers didn’t directly compare this rate with traditional predictors like voter polls, they believe their method could be of interest to pollsters and political pundits.
“It’s much cheaper to run this kind of analysis, compared with traditional polls, so I think it’s something analysts may consider in the future” says Avishay Livne, a graduate student in Computer Science and Engineering who co-authored and presented the study last week at the International Conference on Weblogs and Social Media in Barcelona.
“Eventually this could be developed into a tool that CNN could use, but it would need to be more of a ‘black box’ set-up, where they put in basic info and it generates a prediction.”
In the future, the researchers would like to beef up the analysis–and hopefully the predictive power–by factoring in additional variables.
Accounting for the amount of funding a candidate has, the timeframe of the tweets, and real-world events that focused attention on particular candidates in the Twittersphere and beyond could all help fine-tune predictions.
So what does this mean for candidates and parties looking to better leverage Twitter in upcoming elections?
“The best advice would be to act as a group and coordinate with other candidates across different states,” Livne says “Most importantly, the topics discussed should be similar, but a list of key words would be helpful.”
Christine Hoekenga is a freelance writer based in the southwestern U.S. She’s written for High Country News, Technology Review, and the Smithsonian, among other publications.
Have a story idea or tip for Idea Lab? Send them to Idealab@talkingpointsmemo.com
