PredictWise Blog

In addition to providing comprehensive election news and real-time results, Bing Elections also provides unique social media insights and analysis. And as we count down to Election Day, social media buzz on the candidates, parties, issues, results and more are only intensifying. As a way to help people make sense of the noise, we created and report discussion levels for every major elections topic popping across Twitter on Bing Elections.

This discussion level scale is sourced by something that we call the Bing Social Media Interest Richter Scale, which categorizes the level of buzz on the presidential candidates across key issues popping in Twitter. The three levels are chatter (every day levels), discussion (an event), and hot (a major event). Generically, the discussion levels provide you with an easy to understand scale for how much social media is discussing a given person, sets of people, topics, etc. To provide transparency to these values, below is a snapshot of the Bing Social Media Interest Richter Scale, the proprietary tool, which we use to gauge discussion levels. The internal scale puts current values into historical context and provides people with an easy way to compare events. The scale is a powerful resource for anyone interested in benchmarks comparing the current and past social media interest levels on a given person or topic.

“Big Data” is a popular buzzword these days and there are certainly many places to find streams of data about all sorts of things; the trick is finding data streams that answer interesting and meaningful questions for their readers. Raw data totals are not helpful for most readers, because they lack context. What does X number of tweets or “likes” mean anyway? Opaque indexes hardly do any better. If you do not know what is going into the index, it is impossible to understand what the index means.

The discussion levels, outlined above in the Bing Social Media Interest Richter Scale, provide helpful context and are very transparent. The scale today focuses on Twitter data. We call this a “Richter Scale” because we use the same mathematical principles as the famous earthquake scale to convert raw data into an easy scale. Each higher number means 10 times the interest!

The scale can display multiple entities at a time to provide head to head comparisons, or entities can be combined to show the discussion levels around groups of people or topics. Above is the scale for both Barack Obama and Mitt Romney during the hectic few weeks around the Democratic Convention through the first presidential debate.

The historical data puts the magnitude into perspective. Major set-pieces of the campaign, Obama’s speech at the end of the DNC and the first debate, scored 5.6 and 5.5 respectively; only these huge events enter the realm of hot! Unknown major events, such as the assassination of the US ambassador to Libya and release of the hidden video of Romney at a fundraiser, scored lower on the scale with 4.2 and 4.0 respectively; these memorable events are in the realm of discussion. Of course, those events also unfolded over a long period of time, with less concentrated spikes. So far, the last few days of the campaign have lingered in this elevated discussion level.

Knowing how much social media buzz is happening is important and meaningful, because it may be a bellwether for the impact of major events over time. This is true both in terms of magnitude at the time and whether or not a major event will be discussed in the days and weeks to come.

Future iterations of this scale will provide more details on who is talking: gender and geography. As we work to evolve and deepen our analysis, it can continue to grow in providing a critical pulse on the conversations that are happening across many segments of voters – helping to provide an early indication on the true impact of social conversations.

To see our discussion scale and more social media analysis around the elections, please visit Bing.com/elections. This is joint work with Emre Kiciman of Micrsoft Research.

Click Here for the Full Text on Bing.com

In the last two days, six new polls of Ohio voters have been released, five of which favor President Barack Obama, by leads of 6 points, 4 points, 4 points, 3 points and 2 points. The last reports a tie, from right-leaning Rasmussen, which gave former Gov. Mitt Romney a 2-point advantage last week. Obama now has a 3.1 percentage point lead in Pollster's average for Ohio and a 2.9 percentage point lead in the RealClearPolitics average for Ohio.

A 3 percentage point lead is hardly an insurmountable margin for Obama. But it has been so consistent over the past several weeks that Romney ought to be hoping that the polls favor Obama on a systematic basis by overestimating the turnout of respondents with demographics that favor him. This is certainly possible, but it would mean that many different polling institutions are making similar mistakes. The outcome of Ohio, then, will be as much a referendum on the art of polling as it will be on the art of the presidency.

Click Here for the Full Text on Yahoo!'s The Signal

In the last two days, six new polls of Ohio voters have been released, five of which favor President Barack Obama, by leads of 6 points, 4 points, 4 points, 3 points and 2 points. The last reports a tie, from right-leaning Rasmussen, which gave former Gov. Mitt Romney a 2-point advantage last week. Obama now has a 3.1 percentage point lead in Pollster's average for Ohio and a 2.9 percentage point lead in the RealClearPolitics average for Ohio.

A 3 percentage point lead is hardly an insurmountable margin for Obama. But it has been so consistent over the past several weeks that Romney ought to be hoping that the polls favor Obama on a systematic basis by overestimating the turnout of respondents with demographics that favor him. This is certainly possible, but it would mean that many different polling institutions are making similar mistakes. The outcome of Ohio, then, will be as much a referendum on the art of polling as it will be on the art of the presidency.

Click Here for the Full Text on Yahoo!'s The Signal

In the latest YouGov/Xbox poll, the pivotal state of Ohio showed slightly more Romney supporters than Obama supporters. But, when asked who they expect to win Ohio, the same respondents predicted Obama would win their state. Justin Wolfers and I have conducted an exhaustive research project on the power of expectation polling, and the evidence is overwhelming: when the intention (i.e., support) and expectation of respondents in a poll point in opposite directions, the expectation is correct over 75 percent of the time.

The YouGov/Xbox poll on Monday, October 29 asked respondents which candidate they were likely to support (their intention) and which candidate they thought would win their state (their expectation). We interviewed 12,479 respondents throughout the United States, 642 of who live in Ohio. Of the Ohioan respondents who supported either Obama or Romney, 48.3 percent supported Obama (versus 51.7 for Romney). Yet a notably larger 51.1 percent of these same Ohioan respondents, who reported an expectation that either Obama or Romney will win, predicted Obama to win their state of Ohio (versus 48.9 percent for Romney).

One explanation for why the expectation question is more meaningful for forecasts is that it incorporates more information. Every person possesses a batch of both public and private information about the election, and when a pollster asks them about their intention, the pollster is extracting only a small portion of that information; the pollster is just getting the respondent's support on that day. The expectation question captures information about the respondents social network (e.g., who the respondent thinks her friends and family will vote for) and more public information (e.g., what the respondent is seeing in the news), along with the respondents' intention to vote and support for one or the other candidate. Our paper shows that the expectation question has the forecasting power, relative to the intention question, of asking 10 times as many random respondents.

The standard intention question has dominated the polls for 75 years because it implicitly looks like the expected vote share, which is an outcome people can easily wrap their heads around. If we find that Romney is up 51.7 to 48.3 percent in our poll of the intentions of Ohio voters, one could implicitly assume that the likely vote share will be 51.7 percent for Romney to 48.3 percent for Obama. But most of us do not care about the vote share in Ohio; we care about who is going to win Ohio. If we state that Obama is up 51.1 to 48.9 percent in our poll of the expectation of Ohio voters, this number does not correspond directly to any one single outcome, but is much more likely to point towards the eventual winner than the poll of voter intention!

Of course, Ohio is just one piece in a bigger puzzle, the Electoral College; while our nationwide respondents are nearly even on whom they will vote for, they decisively expect Obama to win the election. Our October 29 poll had 50.4 percent support for Obama to 49.6 percent support for Romney (in major party support), a razor thin margin. However, 54.3 percent of the same people, on the same day, in the same poll, expect Obama's reelection, while just 45.7 percent expect Romney to win.

In the next few election cycles, randomly selected respondents from representative samples will get more and more expensive to contact, and we are very cognizant of that when we consider the goals of YouGov/Xbox poll. We want to provide meaningful insight into the current election, while also testing cutting edge graphical interfaces and questions that might promote even more accurate forecasts of election outcomes down the road. We encourage you to keep refreshing Huffington Post's Pollster's list of the Ohio intention polls all day (I know I do), but also look forward to a cycle or two from now when new questions and techniques break into the mainstream. We hope that our research being conducted in this poll will help to lead the way.

Mike Malecki, Doug Rivers, and Brian Stults (YouGov) contributed to the data work of this article. In the interest of disclosure, I help run the YouGov/Xbox poll in my capacity as an economist at Microsoft Research.

This article is syndicated on the Huffington Post.

Since the final presidential debate, 15 polls have surveyed voter opinion in Ohio, the state that is more likely than any other to determine the election. President Barack Obama leads former Gov. Mitt Romney in 13 of them. The candidates tied in one, and Romney leads in one. Those last two polls were both conducted by Rasmussen, one of the more right-leaning polling institutions, as FiveThirtyEight's Nate Silver has documented.

No one is saying Ohio is a walk for the president. The Huffington Post's Pollster and the RealClearPolitics average both have Obama leading with 51.2 percent among those who express a preference for either major candidate. There is still time for a shift toward Romney, and it's always possible that there is a systematic bias in the polls.

But I don't think it's likely. Historically, polls have been pretty accurate this close to the election. Based only on these surveys, the Signal gives Obama a 75 percent chance of victory in Ohio. When we factor in prediction markets, that figure ticks up a small amount, to 77 percent.

Click Here for the Full Text on Yahoo!'s The Signal