A ranked-choice analysis · Maine, July 2026

What the slate
was worth

Over one July weekend, 17,279 Maine Democrats chose 500 convention delegates — plus 101 alternates — from a ballot of 3,666 neighbors. Part 1 showed why that weekend, not the convention, will likely decide the Senate nomination. This piece looks more closely at the results themselves: using every published vote total from all 16 counties and the three campaigns' published slates, we can measure exactly what an endorsement was worth — and meet the handful of people who out-ran the odds.

by Philip Englund Mathieu

The model

How we built this

This analysis uses Bayesian statistics to help us understand the uncertainty around any given candidate's vote totals. At its core, Bayesian techniques help you to mathematically quantify every quantity you don't know — how much a slate was worth, how generously a county's room spread its votes, etc. Each piece of evidence narrows the range of uncertainty on our estimates of those factors. A model fit this way states assumptions up front, then uses data to find the parameters that best represent that data.

We model every candidate's vote total like this:

Breaking that down:

By stating these assumptions up front, we can use them to calculate the expected result, as well as the degree to which our observed result differed. Whatever remains for any one candidate once county, ballot position, and slate are priced in — actual votes minus the model's expectation — is fairly called their personal vote. As an added feature of Bayesian statistics, every effect arrives with an uncertainty range instead of just a point guess. See the Method section below for details, or scroll on to see it in practice.

Building a delegate card

The baseline

Penobscot County's ballot was a wall of 341 names competing for 44 delegate seats and 9 alternate spots; each of the county's 1,046 voters could mark up to 53 of them. No parties next to names, no offices, no districts. Just names.

Start with a generic candidate. Random votes - neighbors, friends, town-mates voting for every name they recognize, etc. - earns about 4-5% of ballots on a typical person, roughly 50 votes of Penobscot's 1,046.

The alphabet

But ballots ran alphabetically, and it shows: a name at the bottom of a long ballot collects up to a tenth fewer votes than the same name would at the top. Our candidate — the "S. F." on the card — files under F, high on Penobscot's ballot, so the adjustment here is about two votes. It's a bias too small to see on any one card, and unlikely to shift many votes, yet still big enough to observe. This effect could be easily mitigated by rotating or randomizing the order, but doing so would mean each person gets a different ballot, and looking up a candidate by name becomes more difficult.

The slate

Now the piece that dwarfs the others. Averaged across every county and candidate, appearing on Troy Jackson's slate was worth +31 percentage points of ballots on top of the baseline; Nirav Shah's was worth +15; Shenna Bellows's roster, just +6.

The gap is discipline, not just reach. Jackson's pages said "vote for every name on the slate and no other names," and his voters did: his slate-mates' totals move in near lockstep, while Bellows's listed names barely outran the unslated. Jackson's organization amplified his slate's power by encouraging voters to explicitly avoid voting for anyone beyond his slate.

The personal vote

Adding these three pieces up produces a predicted vote range for every candidate. The difference between that prediction and the actual count is the part no list explains, or the "personal" vote.

Here is the card we've been watching. Sean Faircloth — a sitting state representative from Bangor — was on Jackson's slate, which lifts his prediction to about 410 of Penobscot's 1,046 ballots. He drew 605, the most of any Penobscot candidate. The slate carried him to the line, but his personal vote brought him +195 beyond it.

For most candidates, the final result boiled down to whether they were or were not on the Jackson slate — but not for everyone. Let's look at some real examples to see how different archetypes fared.

The gallery

We'll start with the "median" unslated candidate — a composite of the 2,176 people who ran without any campaign's endorsement (and setting aside the nine sitting legislators that ran; more on that in a moment). Fifty-four votes — the statewide median, in counties that typically cast about 750 ballots. A 0.5% chance of a seat. Most of the nearly 4,000 Democrats who put their names forward this July will look something like this card. With rare exception, no amount of individual name recognition beats the power of the Jackson slate.

The median Jackson-slate candidate is the same person with one different fact: their name was on a list. 450 votes — a shade above what the list alone predicts — and a 97% chance of winning a seat.

Sean Faircloth — the first card we looked at — is the archetype's cleanest case: slate and reputation firing together for Penobscot's top total. He is not alone. Around the state, sitting legislators who joined a slate reliably out-ran their own slate-mates, typically by dozens of votes.

Nicole Grohoski — sitting state senator, Ellsworth — was on no slate at all. Predicted: 38 votes. Actual: 314 — eight times her prediction, a Hancock County seat, and, after fellow senator Cameron Reny, the most improbable overperformance in the state. Statewide, just 12 winners — eight delegates and four alternates — were on no campaign's list, led by exactly this kind of card: two sitting senators, county party officers, longtime local figures (Reny and Grohoski per the Legislature's rosters; the party roles per the candidates' own statements in the County Delegate Guide).

Holly Sargent of York — a sitting state representative — appeared on both the Bellows and Shah slates. Candidates like her are why we can compare slates fairly: same person, two lists, so any difference is the lists. The finding: two endorsements are worth less than the sum of their parts, because the two campaigns were often reaching the same voters. She won her seat anyway.

The last card belongs to Nirav Shah — the Senate candidate himself, who filed as a rank-and-file delegate candidate in Cumberland County. His name recognition is real: 997 votes, five times his prediction, one of the largest personal votes in the state. It still left him 330th of 1,296 in his county — nowhere near a seat. In a room where one campaign's slate votes as a bloc, even the most famous free agent in Maine finishes in the middle of the pack. No single card beats a deck. Had his name been printed on his own list, the model says he'd have polled like any other member — and as it was, every one of the 178 people on it out-polled him.

One twist the sweep built for itself. Jackson's lists named no more than the delegate quota in each county — no spares — and his voters obeyed. So his people took 479 of the 500 delegate seats and just 7 of the 101 alternate spots. The bench belongs to the other campaigns: if Jackson delegates miss the July 25 floor, their replacements come mostly from Shah's and Bellows's decks.

In a field of 3,666, the list you were on mattered more than anything you had ever done — except, for twelve people, everything they had done.

Your turn

Look up a delegate card

All 3,666 candidates from the July 18–19 county meetings are here. You start with a card dealt at random — search a name, deal again, or jump to any county's top vote-getter. Chances are you know someone in the deck.

Cards show official results (Maine Democratic Party) against this model's expectations. The shaded band on the actual bar is the model's 95% expected range for a card like that one; a red tag appears only on the 5% of cards that landed outside their band — everything untagged is within ordinary spread. A card's "personal vote" is a model residual, not a judgment — for slate members, matching the prediction just means the slate was the story. Public roles are noted only for verified current state legislators.