Twitter says it has developed a way to measure how its users feel about the presidential candidates, drawing on the nearly 2 million weekly posts on the micro-blogging site about President Obama and Republican challenger Mitt Romney.
The company joined forces with the data analysis firm Topsy and two campaign pollsters--Democrat Mark Mellman and Republican Jon McHenry--to launch the new Twitter Political Index, which it says "evaluates and weighs the sentiment of tweets mentioning Obama or Romney relative to the more than 400 million tweets sent on all other topics" each day.
(RELATED ANALYSIS: Obama, Regaining the Momentum)
The effort is designed to supplement conventional ways of measuring public opinion, Twitter says, and is not a replacement. But as the political survey research industry is confronting unprecedented challenges, many are looking to non-survey approaches to fill the gaps.
Topsy developed an algorithm that assesses the sentiment of a tweet in the same way that a random individual would more than 90 percent of the time. And Adam Sharp, the leader of Twitter's government, news, and social innovation team, says that the algorithm can be altered and refined to reflect the changing rhetoric of the campaign. "It is a collection of key words, phrases, and patterns that is ever expanding what is positive and negative," he said.
The initial installment of the Twitter Political Index, called the "TwIndex" for short, shows Obama with a score of 34 and Romney with 25, based on tweets posted on Tuesday. Since the TwIndex compares tweets about the candidates to all tweets on other topics, that means that tweets about Obama are on average more positive than 34 percent of tweets not mentioning him. It also means that tweets about Obama are generally more positive than tweets about Romney. The plan is for the latest Twitter Political Index will be posted each day at 8 p.m. at election.twitter.com.
The TwIndex can also be used to show changes in perceptions of each candidate among Twitter users. Sharp shared charts and graphs with National Journal that showed the changes in TwIndex scores for Obama closely paralleling movements in Gallup's daily-tracking poll. And because the TwIndex is a daily, immediate measure, compared with the three-day rolling sample used by Gallup to form its tracking poll, tweets often foreshadowed changes in the president's approval rating, he said.
But Sharp also pointed to instances when tweets and polling did not align. In August 2011, as Congress and the White House battled over extending the federal debt limit, the president called on his supporters to write or tweet their members of Congress to urge them to support raising the debt ceiling. That led to an increase in his TwIndex, while his approval ratings in Gallup's polling fell.
Tweets after the killing of terrorist Osama bin Laden turned back to the economy fairly quickly, meaning that Obama's TwIndex rating returned to normal faster than the bump he received in Gallup's polling receded.
More recently, Sharp said, Obama and Romney saw ratings change around Obama's announcement that he supports same-sex marriage and the Supreme Court decision to uphold the health care law, although neither event led to significant changes in their standings in the polls.
The development of the Twitter Political Index comes amid fast-approaching changes in public-opinion research. Traditional telephone polling is becoming more difficult and prohibitively more expensive for most news organizations as Americans move away from landline phones toward mobile devices. Voters are increasingly harder to reach over the phone, and they are less likely to participate in polls when they are reached.
But Sharp stressed that the companies' analysis of publicly available tweets is not designed to replace traditional polls but is rather an addition to existing survey research. "There are definite strengths to polling that we can't match," he said, including the ability of pollsters to obtain a random sample of Americans.
Twitter does have advantages over phone polls in certain areas, including the difficulty that pollsters have in identifying which respondents are likely to vote in elections. Many respondents say they will vote but do not end up casting ballots. "Someone who tweets about politics regularly is probably a likely voter," Sharp said.
"We believe that the Twitter Political Index helps give a more complete picture" of the election, Sharp added.
McHenry, the GOP pollster working on the project, agreed. "This isn't polling," he said. "It's a useful public-opinion tool, we hope."
"Seeing what people are spontaneously talking about as it relates to the candidates.... I think that will add something to the conversation going forward," McHenry added. "And it will be super interesting during the debates."