WAN-IFRA

A publication of the World Editors Forum

Date

Fri - 27.05.2016


Twitter takes on real-time political forecasting with Twindex

Twitter takes on real-time political forecasting with Twindex

“Ignore the ‘Twindex’ at your peril,” warned Chris Cillizza yesterday on The Fix, a Washington Post blog.

The which?

Twindex (n.) is a catchy abbreviation for the Twitter Political Index, a joint endeavour by the identity-shifting microblogger, data analysis company Topsy and pollsters Mellman Group (Democrat-leaning) and NorthStar Opinion Research (Republican-tilting) to take the political pulse of the United States of America by indexing the 400 million 140-character statements pronounced each day – by sentiment.

That is, Topsy sifts through this onslaught of daily tweets from around the world to establish a neutral baseline of positive and negative sentiment. It then sifts through them to find all of the tweets that express opinions about President Barack Obama and soon-to-be-anointed Republican rival Mitt Romney, runs a “sentiment analysis” on them, and weighs them based on how enthusiastic or disparaging they are compared to the baseline. It looks at the last three days of tweets, weighting the more recent ones higher.

These are plotted on a sentiment index ranging from 1 to 100, with 1 representing the most negative thing said on Twitter that day by any user in any country on any subject (as Cillizza said, that must be quite the tweet) and 100 representing the most positive sentiment expressed.

At 8 p.m. Eastern time each night, the two candidates are ranked on this index based on how many people had good and bad things to say about them that day. Here are some samples:

Obama has SCREWED UP so many things in such a big way that it's really hard not to hate the guy! ... #obama #failure

— Again America .com (@AgainAmerica) July 29, 2012

If #Romney visit to #UK meant to show foreign relations skills, still has some learning to do. Olympics comments not gone down well here

— Robin Bew (@RobinBew) July 26, 2012

Yesterday, Obama was given a numerical score of 44, which means that tweets about him had been more positive in tone than 44 percent of all other tweets. Romney, meanwhile, entered August with a Twindex score of 26, meaning that 74 percent of all tweets on all subjects were more positive than tweets about Romney yesterday.

“Welcome to the age of big political data,” pronounced Mat Honan of Wired’s Gadget Lab on the subject.

Twitter's data supply has certainly grown since the last US presidential race: on Election Day in November 2008, Twitter had one of its biggest traffic days ever, with 1.8 million tweets flooding its servers. These numbers now seem “quaint,” as Honan puts it; these days, 1.8 million tweets are posted every six minutes.

It was back in 2008 that Twitter first tested the idea of serving as a political barometer, hiring a company called Small Batch to construct a site that could give an idea of what people were saying about the election. At the time, the sample group was too small to give an accurate or useful picture of what voters were thinking.

This is no longer the case.

Indeed, the figures between August 10, 2010 and now hold tight to polling data from Gallup approval ratings (see chart below). The difference is that Twitter witnesses opinion shifts in real-time, not after the fact.

Twitter reportedly realized some time last year that the things users were saying about candidates “accurately foreshadowed voter sentiments showing up in traditional polls.”

Wired cites two examples: firstly, Twitter saw a real-time spike in kind feelings toward Newt Gingrich during a Fox News debate broadcast, and sure enough, a few days later Gingrich climbed in the polls. Secondly, before the Michigan and Arizona primaries, Mitt Romney was pulling in Twitter followers while Santorum was watching them run for the hills; later, the results of the primaries showed that voters had done the same things at the polls that they had already done on Twitter.

At that point, Twitter partnered with Topsy to create an algorithm capable of scrutinizing the language of tweets using a database of thousands of words, including colloquialisms, and understanding which posts tilted negative and which ones swung positive. Then Topsy and Twitter created a keyword engine together, and following extensive human testing, found that 90 percent of the time their algorithm would generate voter-accurate results. From here the companies continued to refine their algorithm, and found that each time they did they grew closer to the Gallup data.Twindex vs. Gallup

While the proximity helped them be sure they were “onto something,” the team was more interested in the places where the figures were disparate.

“If the dials are pointing in different directions, people are saying one thing to pollsters, and another in conversation,” said Adam Sharp, head of government, news and social innovation at Twitter, to Wired. “That is where the Twitter index is providing a real service to journalists, because it’s where we are saying we don’t have a complete picture, and need to be asking better questions.”

The idea is not to replace a traditional pollster, according to Sharp, but to add a different, real-time dimension to analysis of the country’s shifting feelings toward its prospective leaders.

“We believe the Twitter political index reinforces the transitional models of research,” he told The New York Times. “By providing more signals, more dials — that can agree or disagree — these new technologies give a more complete picture of crafting a political forecast.”

Sources: Wired, The New York Times, The Washington Post, Twitter

Image courtesy of Cain and Todd Benson via Flickr Creative Commons

Author

Emma Knight

Date

2012-08-02 14:06

The World Editors Forum is the organization within the World Association of Newspapers devoted to newspaper editors worldwide. The Editors Weblog (www.editorsweblog.org), launched in January 2004, is a WEF initiative designed to facilitate the diffusion of information relevant to newspapers and their editors.


© 2015 WAN-IFRA - World Association of Newspapers and News Publishers

Footer Navigation