The BDN Opinion section operates independently and does not set news policies or contribute to reporting or editing articles elsewhere in the newspaper or on bangordailynews.com
Robert W. Glover is an associate professor of political science and honors at the University of Maine. These are his views and do not express those of the University of Maine System or the University of Maine. He is co-leader of the Maine Chapter of the Scholars Strategy Network, which brings together scholars across the country to address public challenges and their policy implications. Members’ columns appear regularly in the BDN.
A new generation of election forecasting is arriving in Maine politics and we should be cautious about how we use and interpret it.
Recently, researchers used polling data and computer simulations to estimate the likely outcomes of Maine’s Democratic gubernatorial and congressional primaries. The approach is innovative and, frankly, admirable. Rather than simply reporting first-choice preferences, the researchers attempted to account for the complexities of ranked-choice voting by simulating thousands of hypothetical elections and estimating how voter preferences might transfer as candidates are eliminated.
As a political scientist (and supporter of evidence-based approaches to understanding elections), I find this work fascinating. Ranked-choice voting presents challenges that traditional polling was never designed to address. Researchers are right to experiment with new methods that help us better understand how these elections may unfold.
But we should also understand the limitations of such simulations.
Traditional polling asks a straightforward question: “Who do you support?” Polling a ranked-choice election asks a much harder series of questions: Who do you support? Who is your second choice? Your third? What happens if your preferred candidate is eliminated? How stable are those preferences? And will the people answering the survey actually be the people who show up to vote? Every answer introduces another opportunity for uncertainty.
Simulations help researchers explore those uncertainties. They do not eliminate them.
Many voters hear that a candidate has a 97% chance of winning, as the researchers predicted for Nirav Shah in the Democratic primary for governor, and assume the race is effectively over. Yet the simulation is not observing thousands of real elections. It is observing one poll and asking the relatively complicated (and potentially unstable) “what if” questions above.
Repeating an assumption thousands of times does not magically create new information. Imagine taking a blurry photograph and making 1,000 copies of it. You do not suddenly have a clearer picture. You have 1,000 versions of the same blurry photograph.
This is especially important in primary elections, where uncertainty is everywhere. Turnout fluctuates. Candidates gain or lose momentum. Endorsements (or in the case of Maine recently, multi-candidate coalitions) matter. Many voters are only beginning to pay attention in the final days of a campaign. A simulation may accurately reflect what respondents told pollsters last week yet miss the boat on what actual voters do the next week.
None of this means such forecasts are useless. In fact, one of their greatest strengths is helping us understand ranked-choice dynamics. They can reveal which candidates are broadly acceptable across factions, where second-choice support is concentrated, and how coalition-building may matter as much as first-choice enthusiasm.
The problem arises when a nuanced statistical exercise becomes a political narrative. And that transformation can happen remarkably quickly.
Within hours of the forecast’s release, the Shah campaign was circulating the results on social media, accompanied by U2’s “Beautiful Day.” Campaigns are free to celebrate good news, of course. Their job is to project confidence. But there is a difference between “our candidate appears well-positioned” and “the race is effectively over.” The forecast cautiously suggested the former. The social media post implied the latter.
That should concern anyone who cares about democratic participation. Political scientists have long studied bandwagon and demobilization effects. While the evidence is mixed, perceptions of inevitability can shape how citizens think about elections. If voters conclude that an outcome is already determined, some may decide their participation doesn’t matter.
That concern is particularly acute in primary elections, where turnout is already low and relatively small changes in participation can dramatically alter outcomes.
This is not a reason to stop forecasting. Nor is it a reason to dismiss innovative efforts to adapt polling to ranked-choice races. It is simply a reminder that sophisticated models still depend on assumptions, and assumptions can be wrong.
The purpose of election forecasting should be to help citizens better understand elections — not to convince them that elections have already been decided.
The election will not be decided by a simulation. It will be decided by voters.