Election Forecasting: Big Data, Algorithms, and Baseball

Election Forecasting: Big Data, Algorithms, and Baseball
Larry Sabato, founder and director of the University of Virginia Center for Politics, speaks at a midterm election forecasting seminar in Washington on Aug. 28, 2018. (Samira Bouaou/The Epoch Times)
John Haughey
9/21/2022
Updated:
9/28/2022
0:00

With fewer than 50 days until the Nov. 8 elections, the stage has been set with local, state, and national media churning out a blizzard of stretch-run campaign coverage.

In print media, there’s a uniformity in how election articles are structured. Most routinely note how the race fits into the “big picture,” how much money was raised by candidates, and what the polls say.

This century, another element has become standard with the convergence of “big data” and algorithmic modeling: electorate ratings and forecasts from analysts mining historical statistics, crunching voter data, and aggregating polls to make predictive judgments based on an ever-expanding realm of factors and relationships.

Until the late 1970s, predicting the outcome of an election was a parlor game for academics, political scientists, and pundits, on par with Ouija boards, tea leaves, and the “Washington rule,” which said that if the city’s NFL team wins its last home game before Election Day, the president’s party would win. That “rule” correctly predicted every presidential winner between 1940 and 2000, but since then, has been right just once.

The templates for the “modern era of elections forecasting” were set in 1978 by a Yale University professor’s model based on the economy and incumbency, and in 1979 by a University of Kentucky professor’s model that charted relationships between presidential approval ratings and subsequent votes.

By the late 1990s, the American Political Science Association’s magazine and International Journal of Forecasting were annually introducing new forecasting models developed by universities, analysts, consultants, nonprofits, media outlets, and the gaming industry, based on “econometrics,” public opinion, incumbency, party unity, scandals, poll aggregation, and historical voting patterns, to provide “fluid intelligence” for voters, media, pundits, candidates, and campaigns.

But it wasn’t until a sports writer used the same tools that had been developed to project baseball players’ performances that election forecasting gained ambient credence.

Nate Silver, a KPMG Chicago consultant and co-author of “Baseball Prospectus,” in late 2007, began writing statistical analyses of election data for Daily Kos. Using this baseball model, he correctly predicted presidential outcomes in 49 states in 2008 and all 50 in 2012.

Silver’s FiveThirtyEight site is among dozens online that provide electorate ratings, such as The Cook Political Report, and race projections, such as Larry Sabato’s Crystal Ball, Inside Elections, Electoral-Vote, Politico, RealClearPolitics, and 270 to Win. Here are four oft-cited sites that claim to be independent and nonpartisan, and which don’t issue endorsements.

FiveThirtyEight

Founded by Silver in 2008 as a polling aggregator with an interpretative blog, the name reflects the number of U.S. Electoral College electors. In 2010, The New York Times published the site. ESPN acquired it in 2013, before transferring it to fellow Disney network, ABC News, in 2018.

Since 2014, FiveThirtyEight has focused on poll analysis, politics, economics, science, popular culture, and sports blogging. It ranks pollsters, weighing the accuracy of their surveys.

After correctly predicting outcomes in 49 states in the 2008 presidential election and in all states in 2012, FiveThirtyEight—like most forecasters—got the 2016 election wrong, rating former President Donald Trump’s odds of winning at 28 percent. In 2020, it correctly forecast the results in 48 states, but overestimated President Joe Biden’s margins of victory in Wisconsin, Michigan, and Pennsylvania.

As of Sept. 21, FiveThirtyEight gives Democrats a 71 percent chance to keep control of the Senate, gaining as many as four seats, while Republicans have a 71 percent chance of securing a House majority, adding as many 33 seats in November.

Political analyst Charlie Cook, far right, founder of The Cook Political Report, speaks at the 2022 National Conference of State Legislators Summit in Denver, Colo., on Aug. 1, 2022. (Katie Spence/The Epoch Times)
Political analyst Charlie Cook, far right, founder of The Cook Political Report, speaks at the 2022 National Conference of State Legislators Summit in Denver, Colo., on Aug. 1, 2022. (Katie Spence/The Epoch Times)

The Cook Political Report With Amy Walter

Founded by analyst Charlie Cook in 1984, The Cook Political Report has been an online newsletter analyzing electorates and campaigns for presidential, House, Senate, and gubernatorial races since 2004.

In 2021, it changed its name to The Cook Political Report with Amy Walter after Walter assumed leadership as editor, publisher, and owner.

The site is known for its Cook Partisan Voting Index (PVI), which rates all 435 congressional districts with a seven-category scale: Solid Democratic, Likely Democratic, Lean Democratic, Toss-Up, Lean Republican, Likely Republican, and Solid Republican.

PVI correctly predicted that Republicans would win control of the Senate in 2014 but incorrectly projected Hillary Clinton to win in 2016. In 2020, it forecast that Biden would win with 290 electoral votes, underestimating his margin of victory.

In 35 Senate races this year, Cook rates four as “tossups” and six others as “competitive.” Nine of 14 Senate seats occupied by Democrats, and 16 of 21 controlled by Republicans, are deemed “safe” or “likely” wins for the incumbent’s party.

Cook’s “tossup” Senate seats are held by Democrats in Arizona, Georgia, and Nevada, and Republican Sen. Ron Johnson (R-Wis.). Democratic incumbents in Colorado and New Hampshire are running in “Lean Democratic” races, while Republicans are defending an open seat they won in 2016 in Pennsylvania, which Cook rates as “Lean Democratic.”

Cook rates 162 congressional districts as “Solid Democratic” and 30 as “leaning” or “likely” Democratic. It rates 188 congressional districts as “Solid Republican” and 24 as “leaning” or “likely” Republican. The GOP will gain a House majority, Cook projects, but the margin will depend on the outcome of 31 “tossup” races, including in 22 districts now occupied by Democrats.

Larry Sabato’s Crystal Ball

Created in 2002 by University of Virginia Center for Politics Director Larry Sabato, the site focuses on trends in presidential, congressional, and gubernatorial races, with analyses that include key takeaways for readers.

Senate and House race ratings are outlined in maps and charts showing which candidate won the last presidential election there, among other metrics.

In 2008, Crystal Ball predicted winners in 421 of 435 House races, 34 of 35 Senate races, and all 11 gubernatorial races. Its projection that Barack Obama would win 364 Electoral College votes was one shy of the final result. In 2012, it predicted Obama’s reelection but underestimated his victory margin, and was wrong that the Senate would remain unchanged—Democrats gained two seats.

In the 2016 election, Crystal Ball projected that Clinton would win easily with 322 electoral votes, as well as a 50–50 Senate, and that Democrats would gain 13 House seats to eat into the GOP’s majority. That, of course, didn’t happen.

“We blew it,” Sabato would say at the time.

In 2020, Crystal Ball predicted the Biden–Trump winner in 49 states—missing in North Carolina—and winners in all but five congressional districts. It was the only site to forecast that Trump would win Florida, and one of a few to say that Biden would win Georgia.

In a September “seats-in-trouble” analysis of 2022 midterms, Crystal Ball maintains that Biden’s poor approval ratings put Democrats in the worst position they’ve been in since 2010. It projects the GOP will win control of both chambers in November, gaining as many as 42 seats in the House and at least one Senate seat.

Inside Elections With Nathan L. Gonzales

Founded by Roll Call columnist Stuart Rothenberg as the newsletter “The Rothenberg Political Report” in 1989, Roll Call and CNN elections analyst Nathan L. Gonzales took it over in 2015 and added his name in 2017. Rothenberg remains senior editor.

The site produces newsletters that analyze and handicap House, Senate, gubernatorial, and presidential elections in 24 annual reports that feature candidate interviews, data, electoral history, and trends, as well as public and private polling on individual races.

In 2016, Inside Elections also predicted a win for Clinton, projecting she’d collect at least 332 electoral votes. It also overestimated Biden’s 2020 win.

In 2022, Inside Elections projects that the GOP will gain 12 to 30 House seats, the Senate will remain split, and Republicans will win 20 of 36 gubernatorial races.

John Haughey reports on public land use, natural resources, and energy policy for The Epoch Times. He has been a working journalist since 1978 with an extensive background in local government and state legislatures. He is a graduate of the University of Wyoming and a Navy veteran. He has reported for daily newspapers in California, Washington, Wyoming, New York, and Florida. You can reach John via email at [email protected]
twitter
Related Topics