This Model Would Have Beaten the Polls in Every Election Since 2012. Here's What it Says About 2024.

A look at a new way of following elections.

(Credit for the voiceover for this article goes to the great Samuel Lipson)

For nearly 100 years, the most widely-followed metric for predicting U.S. presidential elections wasn’t a poll, a forecast, or a race rating. It was a state—specifically, the state of Maine. Because of the cold autumns in northern New England, the Pine Tree State once held its statewide and Congressional elections in September, two months earlier than the rest of the country. It didn’t take long before political observers in the 19th century realized that these early races could provide a valuable window into the state of the nation’s very polarized political environment. Taking Maine’s Republican lean into account, the general rule was that a strong GOP win would mean a GOP victory in November. On the other hand, a close GOP win would mean a close race, and an outright Democratic win in the state would mean that Democrats were on track to do very well nationally. Thus, the saying went, “As Maine goes, so goes the nation.”

This trend continued unabated as late as 1932, when Maine’s election of a Democratic governor and two Democratic House members presaged Franklin D. Roosevelt’s landslide win two months later. Four years after that, the state once again seemed to provide a sign that major political changes were afoot. In September of 1936, Maine elected a Republican governor, state legislature, and Congressional delegation—a sign that, if history was any indication, the GOP was about to make a major comeback and oust President Roosevelt out of office in November. But once the results that year actually came in, it was immediately obvious that this Republican strength in Maine had only meant one thing: that Republicans were doing relatively well in states like, well, Maine. Alf Landon, the Republican nominee, would carry the state, but it would wind up being one of the two states in the entire country he ended up winning. The state’s longstanding status as a bellwether was completely busted; James Farley, FDR’s campaign manager, would quip that “As Maine goes, so goes Vermont.”

This misfire wouldn’t be a one-off. As the nature of the office of the presidency changed under Roosevelt and local partisan newspapers gave way to new mass media outlets, U.S. elections became something more than the feats of raw partisan strength that they were during the Gilded Age. Voters began to evaluate different candidates for different offices very differently from one another. It was a depolarized era, defined by the new maxim that “All politics is local.” Commentators would shift their attention from off-year elections and towards the new industry of scientific polling, which could more directly evaluate how voters felt about America’s individually unique races. True to form, Maine’s early elections would fail to predict almost any of the presidential elections over the five cycles after 1936; by the time the state moved its elections to November in 1959, it went with little notice. The political world had long since moved on to the polls as the only meaningful pre-election indicators one could get.

This New Deal-era attitude of prioritizing polling above all else still prevails to this day even though the political environment of that era is long gone. In the decades since public polling became king, American politics have steadily become more and more polarized. Ticket-splitting, once so prevalent that it was practically impossible to extrapolate trends in one race onto another, has become near-nonexistent. Instead of evaluating candidates among a countless array of factors, voters often now only evaluate them according to one: partisanship. Altogether, it’s a political world that’s a lot more like the one of the late-19th century than the one of the mid-20th century—one that may be worth looking at in a new light. Although we no longer have early results from Maine to pour over, we still have a myriad of pre-November data points from voters each year, from party primaries or special elections, that we can look at. Although often ignored, these races have repeatedly proven predictive of major trends in U.S. elections, from Republican resilience in 2020 to Democratic strength in 2022. But, despite all of this, hardly any commentators have chosen to incorporate these indicators when creating projections of the national environment.

That is, until now. After correctly using pre-November electoral indicators to predict that 2022 would not be a red wave two years ago, my friend Josh Taft of Dark Horse Politics has worked on creating a model that combines polling, primaries, and special election data to try to present the most accurate possible projection of the national environment. The result is an eerily accurate formula that, when applied to past presidential elections, has almost always been more accurate than the polling averages. Here’s how it works, its historical track record, and what it says about 2024—and why it gives Democrats reason to be optimistic heading into the final week of the election.