pollyvote

Wharton marketing professor J. Scott Armstrong has been working on forecasting elections since 2004.

Photo: Katie Zhao / The Daily Pennsylvanian

As the presidential primaries progress, the question of who will end up as the Republican and Democratic nominees has reached a fever point. Wharton marketing professor J. Scott Armstrong has the answer — or at least he’s working on it.

Armstrong, along with a group of political scientists, created PollyVote.com in 2003 to predict presidential elections, beginning with the 2004 election. Like ESPN writer Nate Silver’s popular FiveThirtyEight blog, PollyVote uses evidence-based methods of forecasting to make political predictions.

While the website focuses on the presidential election, its main purpose is to compare different methodologies of forecasting.

“What we were trying to do is to demonstrate to the world that if you use what we call ‘evidence based principles forecasting,’ you can improve just about any forecast in the world,” Armstrong said. He chose the presidential election for the purpose of gaining attention from the press and the general public.

Armstrong believes his staff does a far more comprehensive job than the more noteworthy FiveThirtyEight.

“They also do combining, not nearly to the extent we do,” Armstrong said, adding that Silver “uses his judgment and that’s not a good idea, but it makes it more interesting for people.”

PollyVote, in contrast to other businesses or government agencies, combine forecasts for improved accuracy. Their most up-to-date forecast of the presidential election has the Democratic Party taking the general election with 52.7 percent of the vote.


“Most people think if you look at a forecast, you should try to pick the best one, but that’s false,” Armstrong noted. “In fact, if you use a combined forecast, you can very often, very often, do better than the best component in that combination.”

“Combination” entails the inclusion of multiple statistical factors together to predict outcomes.

Armstrong uses about 10 econometric models, political polls, expert opinions, prediction markets and citizen forecasts.

One model he uses is called a “biographical model,” which draws from the biographical information about the candidate to predict the outcome of the election.

“If we have a lot of information about the background of that candidates, if that’s all the information we have, we can get a pretty good forecast of who is going to win,” Armstrong said.

1968 Wharton graduate and Republican presidential candidate Donald Trump scores very poorly on the bio-index.

“We’ve never had anybody so low, it’s astonishing,” Armstrong said. “That should be an indication to the party that they are going to have a real problem if they select Trump.”

On the PollyVote website, users can actually calculate their own biographical index to see how well they would fare against Trump or Democratic candidate Hillary Clinton.

PollyVote includes the predictions and news, along with complete explanations of the process used to forecast. As for the future, Armstrong said that the site will add factors to assist people in choosing a candidate and choosing which issues to advertise and help people improve the persuasiveness of their advertisements.

“Getting rid of subjectivity and combining are the two key things,” Armstrong said. “So far, it’s been the most accurate way to forecast the election.”

Comments powered by Disqus

Please note All comments are eligible for publication in The Daily Pennsylvanian.