< @Polling


Friday, August 10, 2012

BOB GARFIELD:  We have reached the point in our quadrennial cycle when political pollsters begin to descend on the American people by phone, usually during dinner. But there’s another less intrusive way to track your mood, when you're not bothered, when your views drop unsolicited from your fingertips. Twitter. That company has teamed up with Republican and Democratic pollsters as well as a social data company called Topsy to create a new tool called the Twindex, which offers a new and possibly even more accurate way to gauge the political leanings of likely voters.

Adam Sharp is Twitter’s manager for government and politics. He says the Twindex contains a nearly unfathomable amount of data.

ADAM SHARP:  Election Day 2008 there were 1.8 million tweets worldwide. We now see that volume about every six minutes. And so this now gives us a really deep river to pull tweets out of for study.

BOB GARFIELD:  While the sample size is obviously vast, it’s also not randomized. Does the composition of the Twitter user base skew the results? For example, does it skew young? Does it skew left?

ADAM SHARP:  It’s important to note that this isn’t measuring the opinions of every user on Twitter. It is measuring the opinions of those who are actively talking about the campaign or talking about the individual candidates.

BOB GARFIELD:  So it is skewed. It’s skewed toward the politically engaged who are the people who presumably vote to begin with.

ADAM SHARP:  Exactly. And when you recognize that one of the challenges in traditional polling is often how do you define what’s a likely voter, because someone will always say on the phone that they’re going to vote, but then don’t, we would hypothesize that someone who talks about politics every day is more likely to vote than someone who doesn’t.

Also, another challenge in traditional polling has been reaching people who no longer have a land line telephone. Close to 60% of Twitter users use Twitter on a mobile device. So that’s an area where we may actually be a little bit stronger. And we were very encouraged when we found that there was no natural skew. President Obama and Governor Romney, over the last several months – their average Twitter political index values are actually very close to each other. Also, when we looked at our data on the president for the last two years we found a relatively strong correlation with other traditional polling, like the Gallup Approval Poll.

BOB GARFIELD:  I don’t know if that should make you happier or Gallup, which with a small randomized sample, seems to be [LAUGHS] absolutely capturing the zeitgeist. I mean, who should be giving the high fives - you or the polling industry?

ADAM SHARP:  When all the dials point the same direction, that should make both sides excited because that reinforces the findings we each have. When the dials point in different directions, however, that is when the Twindex is surfacing some conversation that is happening that isn’t necessarily being measured in the traditional polls.

BOB GARFIELD:  For example.

ADAM SHARP:  The raid in Pakistan that killed Osama Bin Laden. Both the Gallup Approval number and the Twitter political index shot up following that weekend, but you saw a much quicker drop-off in the daily Twindex values than you did in the Gallup numbers. And when we dug into the tweets for that two-month period we discovered there were more tweets on the economy than on all national security issues combined, including Bin Laden which had been the number one most tweeted news event in Twitter’s history up to that point - that a month after the Bin Laden raid, if a pollster calls you and says do you think the president is doing a good job, you pause, you reflect, you remember the Bin Laden raid and say yes. But you're not necessarily going to the coffee shop with your friends and coworkers every day and saying great news, Bin Laden is still dead.


You’re going back to the same pocketbook impactful issues that have been affecting your life beforehand, and that is where that natural conversation is diverging from the reflective pause that you're measuring in the traditional poll.

BOB GARFIELD:  Were there any other political insights that you would were able to divine from Twindex that neither the traditional polling organizations nor the press were on top of as quickly as you were?

ADAM SHARP:  There’s a large spike for the president in late summer of next year and then it drops off again earlier this year. Each of those shifts were completely contrary to what the public opinion polls are showing at the time. We discovered that last August during the debt ceiling debate in Congress, the president went on national television and urged Americans to tweet their Congressmen. Over a period of several weeks you had very strong investment from the Obama campaign in activating their base on Twitter and essentially turning them out to vote with their tweets.

But he had no general election opponent at that point. When you fast-forward to March and to May of this year, conservatives on Twitter have begun to align more behind Governor Romney, and people who had been tweeting against other Republican candidates were now focusing their attention on the president. You actually saw the president come back to his original numbers.

BOB GARFIELD:  But in terms of understanding the electorate, have you just identified a liability, that it can be manipulated by going directly to the user base and asking them to start a-tweetin’?

ADAM SHARP:  It actually gave us a measure of the ability to motivate the base which, particularly when you get closer to the election and you're looking at the need to drive turnout, it can be an important measure that doesn’t always surface in traditional polling. That lift was a measure of essentially what it looks like when one candidate is able to motivate their base without a response on the other side. As we head into 2012 we wouldn’t expect to see that same pattern because the presumption is that if Obama motivates his people on one side of the issue, the Romney campaign would motivate them on the other side. And if you did see a shift – well, that would be exactly the result you're looking for because it would give you a measure of which candidate was being more effective in that effort.

BOB GARFIELD:  A moment ago I was asking you whether the people at the Gallup organization, for example, should be exchanging high fives because Twindex proves that their methodology has been very good all along, but polling is expensive and crunching Twitter data is cheap, and I wonder if Gallup and the rest of the polling industry should be worried about being squeezed out of business.

ADAM SHARP:  This is an exciting experiment to see how Twitter data can be used. Now having demonstrated that this dataset is available and is large enough and diverse enough to provide credible results, this now creates opportunities for other polling firms and other data analysis firms to enter this space and iterate on this, this further. Hopefully, it gets better and better and better, with every iteration.

BOB GARFIELD:  That was not an entirely direct response, but a very diplomatic one. [LAUGHS] Thank you very much.

ADAM SHARP:  Thank you, Bob.

BOB GARFIELD:  Adam Sharp is Twitter’s manager for government and politics.



Adam Sharp

Hosted by:

Bob Garfield