There’s a high likelihood that PredictIt will no longer be allowed to offer U.S. election exchange wagering after February 2023, and that has quite a few political scientists saddened because they view these markets as a useful predictive tool.
How useful? That’s a question that will be much easier to answer after next week’s midterm elections.
The dollars being wagered on the site indicate something decidedly different than the polling indicates in several key races. After a spring that had most experts expecting a “red wave” and a summer that had many believing Democrats would buck the trend of the party in power taking a midterm beating, the fall has seen a Republican resurgence. But the strength of that resurgence varies from emphatic on PredictIt to uncertain on a polling-focused site like FiveThirtyEight.com.
While there are countless House of Representatives races, gubernatorial contests, and local campaigns worthy of attention, the heaviest national focus one week from Election Day is on control of the Senate.
Here’s a breakdown of the four closest Senate races as well as the overall picture in that chamber, and where PredictIt’s and FiveThirtyEight’s numbers stood as of mid-day Tuesday, Nov. 1.
Arizona: Mark Kelly vs. Blake Masters
At PredictIt, Democrat Kelly was priced as a clear favorite all the way until Oct. 27, when Republican Masters’ price suddenly nudged above Kelly’s. On Sunday, Masters cost 53 cents (with a return of $1 if he wins and $0 if he loses, suggesting about a 53% likelihood he wins and an equivalent sports betting odds price of -113) to 50 cents for Kelly.
It flipped again on Monday, with Kelly back on top at 57 cents to 48 cents — presumably powered by a Siena College/New York Times poll that had Kelly 6 points ahead, 51%-45%. On Tuesday, with Libertarian candidate Marc Victor dropping out and endorsing Masters, PredictIt’s prices flopped again, and as of publication of this article, Masters was 54 cents and Kelly 48 cents.
All the while, FiveThirtyEight has never wavered in showing Kelly as the favorite, and he leads in every recent poll (albeit by just 1 or 2 points in all but Siena/NYT). On Tuesday, FiveThirtyEight’s model saw Kelly winning 67 of every 100 simulations.
This is a significant disagreement. The bettors see the Republican claiming the seat, while the polls unanimously favor the Democrat. It’s a fascinating test case for which system is more useful in 2022.
Georgia: Raphael Warnock vs. Herschel Walker
This has been a strange and unpredictable race in a state that tilted red for decades but just barely went blue in 2020 — both in the presidential election and in the runoffs for two Senate seats. One of those two Senate seats is already up for grabs again, with incumbent Warnock challenged by ex-football player Walker.
Democrat Warnock was far ahead on PredictIt — as high as a 68-cent to 36-cent split — until Oct. 20, when Republican Walker swung into the lead. That lead shrunk slightly after the Siena/NYT poll came out (with Warnock up by 3 points), but it grew again Tuesday, with Walker at 61 cents and Warnock at 44 as of press time.
FiveThirtyEight’s projections aren’t dramatically different here. It had Warnock ahead until Oct. 27, and the forecast had Walker winning 53 of 100 simulations at last check. FiveThirtyEight’s latest polling averages have the candidates dead even, at 49.2% apiece.
It may all be moot, however, because Georgia races go to a runoff if nobody comes in over 50%. That would mean four more weeks of campaigning, and four more weeks of potential momentum changes, before a Dec. 6 re-run — to determine a seat that may just decide which party controls the Senate.
Nevada: Catherine Cortez Masto vs. Adam Paul Laxalt
This race was nip and tuck on PredictIt all summer, but Republican Laxalt took the lead on Sept. 21 and has been running away with it ever since. At last look, he was trading at 72 cents while Democrat Cortez Masto was at 31 cents.
But the polling is still split, as the Siena/NYT poll had the candidates dead even and FiveThirtyEight has 54 of 100 simulations going Laxalt’s way. According to Nate Silver’s site, this election is still close to a tossup, whereas the wagering on PredictIt suggests it’s all but decided.
Pennsylvania: John Fetterman vs. Mehmet Oz
This is a highly contentious race for a seat being vacated by Republican Pat Toomey that, like in Georgia, sees a Donald Trump-endorsed celebrity candidate taking on a Democrat already elected to statewide office.
At PredictIt, Democrat Fetterman was ahead all the way until Oct. 17, but Oz has led the last two weeks. In keeping with other patterns, that lead shrunk following the Siena/NYT poll — which had Fetterman +5 — then stretched a bit once again. On Tuesday, Oz was trading at 64 cents and Fetterman at 43.
FiveThirtyEight, meanwhile, still has Fetterman winning 58% of the time — although that’s markedly down from his high of 83% success in simulations on Sept. 14.
With the vice president serving as a tiebreaker, Democrats need 50 Senate seats (the number they currently have) to retain control, and Republicans need 51. To get to 50, assuming no major upsets elsewhere (not an entirely safe assumption, of course) Democrats need to win three of the four tight races outlined above. Republicans would need to win two of the four to get to 51 seats.
In answer to the question, “Which party will control the Senate after 2022 election?” PredictIt had Democrats in the lead throughout August and September, but flipped red on Oct. 17. It wasn’t all that close anymore as of Tuesday, with Republicans at 73 cents and Democrats at 34.
Nevertheless, FiveThirty Eight is calling the race for Senate control an absolute 50-50 dead heat. In 50 of every 100 simulations, there are at least 50 Democratic senators. In 50 of every 100 simulations, there are at least 51 Republican senators.
So who is correct?
The divide is significant when it comes to several key races. The latest polling suggests a solid chance that the Democratic Party staves off the traditional wave in favor of the party out of power, while the majority of PredictIt users are buying into the red wave notion.
One thing both sides can presumably agree on, based on various races and results in 2016 and 2020, is that polling is flawed. There are countless theories as to why, but there’s no denying that polls were way off in Michigan and Wisconsin in 2020 (President Joe Biden won both states by much narrower margins than the polls projected) as well as in Georgia (where Trump was favored but Biden prevailed).
The question is which polls in 2022 have been most flawed. Is it the Siena/NYT poll, that made supporters of Democrats feel better? Or is it several of the polls that favored Republicans just prior to that? One theory circulating is that the right wing has pushed possibly inaccurate polls showing Republicans doing well so they can contest results that don’t go their way, as some party officials have done in competitive recent elections.
And are PredictIt’s customers just reacting to polls and betting accordingly? Or are they out ahead of the polling, seeing trends more clearly than the pollsters? In both 2016 and 2020, the swings in the betting markets on election night (and/or deep into the morning) anticipated the final result before any news outlets were ready to make a call.
Depending on how some of these critical races play out next Tuesday, we’ll know whether PredictIt, in what may be its last major election, is still a relevant forecasting tool.