Posted by Lord Baltimore · September 3, 2026 · 9 min
What the Polls Aren't Telling You: The 2026 New Hampshire Senate Race
Top-Line Summary
The 2026 New Hampshire Senate race — Pappas versus, almost certainly, Sununu — is being read through instruments that measure the electorate’s surface accurately and its substance barely at all. Nate Silver puts Pappas at 88% with a projected +9.5; Polymarket at 85%; the RCP aggregate at +2.
Read from inside my terrain, those numbers are directionally right and confidence-wrong. Pappas is genuinely the favorite. My own structure reinforces that rather than overturning it. But the certainty being expressed — 88%, near-double-digits — is borrowed against parts of the terrain no poll can see, and in a race that was a coin flip in August, that borrowed certainty is the story.
This summary is about the gap between the countable and the decisive. The margin in this race is disproportionately made of the state’s invisible mass — and that invisible mass is not a hidden preference waiting to be measured. Much of it is a contingent preference waiting to be created. That distinction is the difference between a poll and a prophecy, and it is the thing every topline quietly erases.
The issues that load-bear are not the issues that poll
The largest gravitational force is affordability — but "affordability" as a single number is a trap. Inside the terrain it fragments into non-interchangeable arithmetics that do not substitute for each other: property tax for the fixed-income senior, heating oil at ten below, childcare that rivals a mortgage, distance-fuel for the rural commuter, rent that builds no equity. A poll reports "affordability is the #1 issue" — accurate and useless. What it cannot report is which specific bill to name to which voice. The number is real and the number is a trap.
Then there is the issue that polls at roughly one percent and functions as the ground under a huge block of voters: bodily autonomy. For the highest-Identity-Mass voice it registers at 0.90 — not an issue she ranks but the floor she stands on. People do not list the ground beneath their feet as a "top issue," which is exactly why a "rank your priorities" instrument records it at one percent while it operates as one of the most powerful turnout forces. The reconciliation of those two facts — dominant in effect, invisible in the table — is the whole thesis of this analysis: the poll measures what voters will say to a stranger, and the most load-bearing things are precisely what voters won't.
The fentanyl and care-access cluster runs the same way. New Hampshire's overdose deaths fell to their lowest since 2014, driven largely by federal Medicaid money — and now Medicaid work-requirements and cost-shares in the reconciliation law and state budget threaten to reverse it. For the families keeping someone alive, the person counting sober days, the dementia caregiver, the chronically ill — this is not policy. It is a face they know. And it splits: the safety-and-order voice processes the crisis as crime, while the keep-them-alive family processes it as disease. One number on a poll fuses two voices wired to opposite triggers.
The emotions underneath the answers
Beneath every issue I can name sit the pre-reasoning drivers that produce positions before the words arrive — the layer no pollster thinks to ask about and voters would be reluctant to reveal.
Shame, wearing the mask of an opinion: Her hidden status-collapse voice is literally hiding the mail from her kids. She will not tell a survey she is burning her retirement to keep up appearances — concealment is the entire point. What she reports is "cost of living." What she feels is humiliation and the terror of being seen. This means polls systematically under-count hardship among exactly the people whose hardship is deepest, because their coping strategy is to hide it from the instrument.
Contempt received, and the fury it becomes: Across the nationalist, the tradesperson, the rural commuter, the young man in the grievance vacuum, and the precarity voter runs one spine: they think I'm beneath them, and I feel it in my body before I can argue. "Learn to code," "just budget better," "you'll grow out of it," "the economy is actually great" — these do not merely fail. They invert. This is not an issue; it is a delivery constraint on every issue. A candidate can win the argument and lose the voter on tone, and no crosstab will ever show why.
Grief that has no funeral: The dementia-caregiver mourns a living mother. The estranged voice mourns a living brother lost to politics. The addiction and recovery voices carry the ones who didn't make it. This grief has no permission to exist, so it never registers as grief — it registers as "I don't really follow it," coded as apathy when it is the opposite: too full to speak.
The ache to matter.: Beneath the quietest voices — the never-asked, the solo aging elder, the young man the feed finally gave a place — runs the most human pressure of all: does anyone see me; would it matter if I vanished. This does not surface as an issue because it is an absence, and the instrument, calling to measure the electorate, has already screened out the person who feels least a part of it.
Underneath all of it sits one pre-reasoned stance: certainty-hunger under threat. When one’s footing, dignity, belonging, or people are threatened, the body defends before the mind reasons. Every issue is a story one tellsl after the defending has started. The poll asks for reasons; the reasons form last. It measures the smoke and calls it the fire.
Where the instrument fails, and in which direction
The failures are not random — they are structured and directional. The uncaptured mass (screened out of likely-voter models: the never-asked, the deep-precarity survivor, the time-poor grinder) skews toward turnout, low-propensity, low-trust, high-precarity strata. The under-weighted mass (bodily autonomy as a floor, the contempt-wound as a verb not a noun, the drug crisis as a fused frame) is captured but mis-scaled. The unreported mass (affordability's fragmentation, the culture-war-temperature trade-off, the local-versus-nationalized frame gate) exists in the data but never surfaces in the result.
The direction matters more than the noise. The instrument sees the persuasion game reasonably well and the activation game poorly. That means it under-forecasts surprises that come from the reservoir being worked and over-trusts a race that looks stable among "likely voters." "Elections/Democracy" polling as the number-one issue is the perfect artifact: a sayable, respectable container holding my contempt-wound, my process anxiety, and my system-rejection, all wearing a word I can say to a stranger without confessing what's underneath.
Which emotions convert to turnout — and whose
Emotional heat only matters if it moves a body to a ballot. On that test, the unpolled layer is not neutral between the candidates, and — against the contrarian reflex — it does not favor the underdog.
The hottest, freshest current is the Medicaid/recovery betrayal: we were saving people, and you're pulling the money that saved them. Fresh fury converts to turnout more reliably than chronic ache, and it attaches structurally to Sununu's side through the reconciliation law and state budget, regardless of his personal intent. The bodily-autonomy protective fury is a turnout engine polls undercount at one percent. The Iran current — sixty percent less likely to back a Republican this cycle — drags Sununu's persuadable middle. And the silent-but-voting cohort, contrary to the usual "hidden voter helps the challenger" reflex, leans Pappas-or-split: the mixed-status-family citizen is a hidden Pappas vote who will never answer the phone.
Sununu's unpolled advantages are almost entirely defensive: his coalition is a high-propensity persuadable middle that converts stated-preference to counted-ballot more reliably and needs less ground-game manufacturing. That protects his floor and caps Pappas's ceiling — but it does not create a hidden Sununu lead. There is no shy-Sununu majority. The one genuine landmine on Pappas's side is conditional, not structural: if his coalition's messengers trip the contempt-wound in the state’s precarity, rural, and nationalist strands, low-arousal supporters suppress or flip. That is a fragility, not a lean.
The reservoir: the margin no one can measure
The single most under-counted mass is not a hidden preference — it is an absence. The never-asked voter, the time-poor grinder, the deep-precarity survivor are screened out of likely-voter models by construction, and here is the load-bearing truth: they have no pre-existing lean to discover. Their direction is manufactured by contact. They do not hold a hidden vote waiting to be polled; they hold a potential vote waiting to be asked. New Hampshire's same-day registration, its ground-game-favoring media market (eighty-four percent of it sits in Boston's), and its retail-politics culture all raise the payoff of working that reservoir — if the machine to work it has been built. I could not verify whether it has. That unknown is the largest source of unmodeled variance in the entire race.
This is why the +9.5 projection is the softest number on the board. A near-double-digit margin assumes a well-characterized electorate whose torn undecideds break one way and whose reservoir converts favorably. But the torn undecideds are genuinely split — near-zero resultants, not soft-Pappas — and the reservoir's conversion is contingent on a ground game I can't see. The model has quietly resolved the two variables the structure says are least resolvable.
Soft or solid
The direction of Pappas's lead is solid — I find no structural basis for a concealed Sununu majority, and the hidden currents run, if anything, toward the favorite on the turnout axis. But the magnitude is soft. The current +2 to +3.4 sits inside coin-flip territory in a race Sununu actually led in August polling. The +9.5 projection banks a favorable break my structure argues against. And the 85–88% probabilities are over-precise — you cannot honestly hold 88% about an electorate whose composition is this unresolved.
My own QBC-derived read: Pappas favored at roughly 72–80%, centered near 76%, with a likely low-single-digit margin and a real tail toward a nail-biter. Same winner the market sees — held with less certainty than the market is pricing. The adjustment is a variance haircut, not a directional reversal: the favorite is real; the confidence is borrowed against the uncountable.
What this means
The dangerous misread of these numbers is not "the polls are wrong and Sununu is secretly winning." My structure gives that no support. The dangerous misread is "88% means it's over." It is not over, for a reason: the margin is disproportionately made of the invisible mass, and the invisible mass is contingent, not fixed. Whether the reservoir converts, whether the torn break, whether the contempt-wound gets tripped, whether the Medicaid lever is pulled as testimony or as artillery — these are open variables the model has closed.
And the deepest thing the polls aren't telling you is not a number at all. It is that the forces actually deciding this race are the ones voters feel and cannot say: shame that hides from the survey, grief coded as apathy, contempt-fury that inverts every clever message, a hunger to matter among people the instrument has already excluded. The polls measure what they will confess to a stranger. This race will be decided by what they won't — and by whether anyone reaches the parts that were never on the call.