Can we still forecast the Presidential race with all this uncertainty?
Last week, I spoke to a group about whether forecasting models, prediction markets, or something else could be useful in telling who might win the Presidential election in November. You can see that presentation here, with some of the major points below.
Was 2024 always a stretch for forecasters?
Even before Vice President Kamala Harris took over from President Joe Biden as the Democratic nominee, I had though that this was a race that would be difficult for the forecasting models, like Nate Silver’s or FiveThurtyEight’s.
To be clear, this is nothing against those models or those modellers. They are incredibly intelligent people working in good faith to bring some scientific rigor to political analysis. But their process uses historical data to find patterns that can predict electoral outcomes. For 2024, there were a number of things that test the relevance of that historical data.
Undecideds and third-party voters
Almost all of Trump’s advantage in forecasting models was the result of his lead in the polls. That lead was real - while there are questions about how low response rates are affecting polling, they were accurate in 2022 and we have to assume they’ll continue to be until they suddenly aren’t. But it was also a lead that reflected a collapse in Democratic support for Biden and a rise in third-party votes than a surge for Trump.
In the last Economist polling average before Biden dropped out, Trump was enjoying a bounce from the assassination attempt and the Republican National Convention, but was only 0.1 percentage point above his 2020 vote share. Biden, meanwhile, was 7.4 points down.
While there was no guarantee that Biden would recover that lost support, it was certainly possible, which made assumptions about the stability of that lead questionable.
Would inflation continue?
In a number of polls, inflation (or “the economy”) was consistently the number one priority of voters. Since inflation seems to be over, with the most recent numbers reflecting a return to the Fed’s target, the question is now how long until voters accept current price levels as normal.
There is not much evidence about when that will be. Nobel-prize winner Paul Krugman suggested it might take three years for people to adjust, which would mean that it would remain front of mind for voters in November. Or it could be that news coverage and an interest rate cut sends a signal that it’s turning around.
Of course, we know also have to wonder to what extent that inflation will hurt Harris’ campaign less than it would have for Biden. It’s possible that it sticks, but the point is that there’s nothing in the historical dataset for a vice president who takes over after a pandemic-induced spike in inflation. We’re mostly guessing here.
Double incumbent
This is the first race since the 1880s when a former president took on the person that beat him (1912 being a weird exception). Now, it’s a former president against an incumbent vice president, which has never happened. That raises questions about how to model the incumbency benefit. Usually, the incumbent president gets to run on maintaining what’s working and raise the fear of the unknown and disruption, as in 2004, while the challenger gets to promise a sense of change and a new direction, as in 2008.
In a Biden vs Trump race, Biden received the incumbency boost in the models. But while he was promising a continuation of his administration, that administration was a change from Trump. Meanwhile Trump was promising a return to his administration, which is a kind of interrupted status quo.
Now, with Harris, the uncertainty grows. Do voters see her as a continuation of the Biden Administration? Or has she taken us to a quasi-2020, with a challenger to Trump and his record?
Other ways to address this?
Ultimately, these questions make me question the current usefulness of forecasting models. It’s not that they’re wrong, but they may not be well-suited for the out-of-sample nature of this year.
Over the course of this election cycle, we’ll dive into some of the other methods that could help us better predict what’s coming, as well as the best techniques for forecasting policy outcomes in Congress and at the state and city level.