It’s not just one problem with polling. It’s 50.

Given the pace of events and the steady stream of headlines, we might be forgiven for thinking that national forces shape all election outcomes. The reality, shown clearly in our recent polling miss, is that America is still a collection of 50 distinct electorates.

Last February, we launched the Colby Polling Initiative. Our nation and state are divided. Polls like ours perform an important public service. They help us understand voter motivations and concerns. We also knew that the nation would be watching our Senate race – and what a race it was!

For several months, our findings suggested Susan Collins was in the fight of her career. But on Election Day, she defied expectations and pulled out a big win. Our poll, like hundreds of others across the country, missed the mark. We got it wrong, but not for the reason you might think.

We assumed the race would be tight and looked at some other projections with raised eyebrows when they reported double-digit leads for the Democratic challenger, Sara Gideon, over a seasoned political operative who won every county in 2008, the same year Mainers sent Barack Obama to the White House, and once again in 2014. Voters are not that fickle, and just a fraction of the electorate is usually in play in any one election.

In this year’s U.S. Senate race, our final poll showed a statistical dead heat, and we believed it would come down to Maine’s unique ranked-choice process. With the progressive candidate Lisa Savage pulling an impressive 5 percent of the vote (what we predicted), things would have been dire for the four-term senator.

Collins surprised us all – and probably herself – by getting an outright majority.

This year’s round of polling post-mortems tells a predictable story: Pollsters failed to learn the lessons of 2016. Explanations range from the dire – polling is dead, who answers an unknown number anymore! – to the conspiratorial – pollsters want to suppress Trump turnout. Some say there is a secret “hate vote,” although the only people who think a social desirability bias exists in polling are those who find it socially undesirable to vote for President Trump.

We worked to overcome those obstacles. We called landlines and cellphones, and used market-research platforms to locate Mainers – the same tools used by international corporations to field products and test markets. We made sure that every one of our polls was geographically diverse, capturing opinions up and down the state. Trump voters were excited and participated at high rates. We think this approach is responsible for our predicting Donald Trump’s sustained popularity throughout the state. And we were remarkably transparent, uploading our findings, crosstabs and methodology for each of our waves.

So, where did we go wrong in the Senate contest? We will spend many hours on this, but in all likelihood it boils down to something pollsters call the “likely voter model.”

We all knew that voter enthusiasm would be elevated this year. Maine almost always gets out the vote at a higher level, and our initial estimates suggest that the electorate expanded by 3 to 4 percent. With such an important presidential contest and such a high-profile Senate race, we knew Mainers would head to the polls – big – and they did.

And yet, we missed a critical element: the turnout of split-ticket voters. We had glimpses of these folks throughout our samples – a vote for Biden, followed by a vote for Collins. But the rate of split-ticket voting has been rapidly decreasing across the country. Across the nation, we know that people who are devout partisans are motivated and do show up on Election Day. We zeroed in on those voters.

In other words, the last thing we wanted to do was ignore the thousands of passionate Trump supporters and loyal Biden voters. In doing, we may have undercounted the number of Mainers uniquely poised to split their vote.

Nationwide, split-ticket voters have been put on the endangered species list, but in Maine they appear to be the exception that proves the rule. Anywhere from 15 to 20 percent of Mainers bucked that national trend last Tuesday. We are not saying that’s a good thing or a bad thing. It is an exceptional thing.

Only one state in the last two presidential elections has chosen a senator and a president of opposing political parties: Maine. Anywhere else in the country, our voter model would have been on point. But here, we discounted the likelihood of split-ticket voters showing up. We undercounted the number of Mainers who remain open to vote for the candidate, not the party.

American politics has proven especially volatile the last decade, and this raises an important question about the use of polls and election forecasting. Unless we take seriously the idea that people living in different parts of the country experience and think of politics differently, then data-driven polls will always be off. Is there a better way?

Here at Colby, we’ll be asking important questions about how we can help facilitate the conversation between the broader public and governing officials. We will continue to identify the issues that matter most to Mainers. Polls will remain helpful for understanding how individuals perceive government, in identifying hidden issues and in giving average citizens a voice.

Researchers must get things right. For us, we intend to pay closer attention to the spirit of independence here in Maine. Political scientists have wondered whether anything truly was local anymore in American politics. Maybe that’s true far away, but we were reminded that Mainers have always charted their own course, and this election was no exception.