© 2022 | WUWF Public Media
11000 University Parkway
Pensacola, FL 32514
850 474-2787
NPR for Florida's Great Northwest
Play Live Radio
Next Up:
Available On Air Stations

PollyVote Proves Accurate Again


Since its conception in 2004, the PollyVote political forecasting tool has accurately projected the winner of the two-party vote in the presidential race. In the 2020 election, the project kept the streak going.

“Well, it did amazingly well,” proclaimed Dr. Al Cuzan, UWF professor of government and proud co-founder of the PollyVote.

From the very beginning — many months ago — PollyVote predicted a win for Democrat Joe Biden and a loss for President Donald Trump, the Republican incumbent. Additionally, it accurately projected their individual share of the total votes between them.

“It becomes a percentage of the combined vote for Trump and Biden. So, the percentage of the combined vote for Trump and Biden, that is called the two-party vote,” Cuzan explained. “So, what percentage of the two-party vote did Trump get? They predicted 48 percent.”

With challenges and recounts in some states, the national tally has fluctuated, but it appears the actual results will show Trump with about 48% of the two-party vote, as the PollyVote predicted.

Cuzan says their forecasting project was successful and did better than other more-well-known, more-established forecasting sites. 

“It outperformed the Real Clear Politics average of the polls. It outperformed the FiveThirtyEight average of the polls,” he said, adding the popularity of the two sites.

“So, the PollyVote is like the new kid on the block, so to speak. And, it isn’t just the first year that it has done so (outperformed those other sites). The PollyVote has done better than the FiveThirtyEight for as many elections as we have did it for both of them. So, it’s proving to be a very accurate forecasting tool.”

The PollyVote has accurately projected the winner of the popular vote in every presidential election since 2004, including 2016, when it forecast victory for Hillary Clinton.

When it comes to the Electoral College, it’s been a mixed bag. In 2016, PollyVote wrongly predicted Clinton would also get the 270 electoral votes needed to win.

This time, the PollyVote accurately chose Biden as the winner, but with a greater margin of victory than he actually recorded. The projection was 329 electoral votes for Biden, with 209 for Trump. The current, unofficial total is 306 for Biden and 232 for Trump.

Cuzan credits the overall accuracy of the PollyVote to its structure.

“The PollyVote was constructed on the premise that forecasts have to be combined and that you want to combine forecasts from different methods,” Cuzan said.

The forecasting tool is designed to be an aggregate of numerous forecasting data that fall within three main components, including Polls, Models, and Expectations.

Cuzan points to the work of PollyVote’s primary author, Andreas Graefe, noting that he averages all of the polls on a daily basis. He also averages all the model predictions and expectation data, such as the betting markets.

“The good thing about combining is you’re not relying on any one component, any one poll or model. You’re simply averaging; you average within components and then you average across components,” he said.

According to Cuzan, it’s important to have multiple components of different types. That’s because situations change and particular models that work this year may not work as well the next time around.

“Because you can say that each election is distinctive, unique in some ways, and, different models have different streams of information, different assumptions, somewhat different methods,” he explained.

“Some models are forward-looking. Some models are backward-looking. Some are a mixture of the two. Some models do not include any kind of polling information. Some models do.”

Credit University of West Florida
University of West Florida

Cuzan concedes all of the components have errors, but by combining and averaging them, the errors will cancel each other out.

He says the main lesson here is, “Don’t rely so much on polls. Don’t rely on any one poll. Average out the polls, and even so, look for other methods.”

The PollyVote site is a good resource. It provides a variety of relevant data, such as polls, betting markets, and political science models.

“You can do worse than looking at the PollyVote,” Cuzan says with a chuckle.

Given the wide margin of error among the polls recently, the political scientist was asked if he thinks it’s time for national media to give a look to the aggregated forecast from the PollyVote.

“Well, again, I wouldn’t say only look at the PollyVote, but I think PollyVote is worth a lot more attention that it has received,” he declared.

The bottom line is the forecasting tool has been more successful than some other forecasting sites, and Cuzan thinks it deserves consideration.

“I guess what the PollyVote needs is a breakthrough, to break into the big leagues,” he said with a bit of a laugh. “It’s beating the teams from the big leagues right now. It’s a minor-league player beating the big leagues, like an amateur boxer beating the professional boxer.”

Regardless of the PollyVote’s future, the UWF government professor expressed a level of satisfaction that he was part of creating something that has proven to be so accurate and, at this point, he’s okay with using it as an exercise in the academic world.