NY-21 · U.S. House

This district's measured terrain −20.485 points

Call the Map's partisan lean on the CURRENT district lines, before any current-cycle evidence. Every pull below is measured from here — not from EVEN.

• Call the Map partisan lean, district geography bound to this forecast edition.
• Prior source: observed_unchanged_lines.

Fitted model, above the terrain shown +9.391 points

What the fitted model adds on top of the starting terrain above — incumbency, past results, campaign finance and the other fitted terms — measured from that terrain rather than from EVEN. It is a comparison with the starting point already shown, not a separate input. The source detail lists the variables the fit uses.

• Fitted baseline -11.095 minus starting point -20.485 = 9.391 points.
• One fitted structural total; this edition does not split it into partisan lean and incumbency.
• Fundamentals: −11.1 — the district's structural lean before this year's polling, money and candidate quality are added.
• This edition publishes that as one fitted total. It does not break it into a partisan-lean and an incumbency term, so this page does not invent that split.
• fundamentals core = 25-term dynamic formula (mass fit over the master feature matrix, held-out-cycle validated; see matrix/feature_dictionary.md + mass_fit_report.json)
• 40-term formula, held-out validated.

This race’s polls +5.398 points

What this race’s own recent polling adds beyond its prior. The forecast weights each poll for recency and sample size and counts each pollster once; the weights listed here are those of the polls’ own average.

• Public Policy Polling (D) (2026-10-13, n=613): R+2.0 · weight 0.57 in the poll average
• Siena University (2026-10-06, n=496): D+4.0 · weight 0.47 in the poll average
• Impact Research (2026-10-04, n=500): R+3.0 · weight 0.46 in the poll average
• Public Policy Polling (2026-09-10, n=557): R+5.0 · weight 0.37 in the poll average
• Impact Research (2026-09-02, n=500): R+3.0 · weight 0.32 in the poll average
• J.L. Partners (2026-08-17, n=500): R+13.0 · weight 0.26 in the poll average
• … and 2 more below.
• Poll average R+1.8, from 8 polls · 8 recorded polls. This is the polls’ own average, each poll at its published margin, the figure this page’s poll summary shows. In House races the forecast then counts a poll sponsored by a party or campaign at full weight and moves its margin 2 points against that party, which puts the average it reads at R+2.8. The forecast gives the polls 86% of the weight against its structural prior for this race, which moves the margin 5.4 points away from Republicans from where it stood before the polls. Eligible polls: latest poll: 2026-10-13 (15 days old).

National environment +2.071 points

The contribution to this race’s margin, not the national polling margin itself. The race’s loading, reference level and polling blend determine how much of the national climate reaches its final forecast.

• Three-source blend, updated 2026-10-14: 53% generic-ballot aggregates (D+7.7) · 28% special-election overperformance (D+9.0 across 46 specials) · 20% approval-implied (D+9.1 from net approval -24.3).
• Latest polls — generic ballot 2026-10-14 (14 days old), approval 2026-10-14 (14 days old).
• Generic-ballot average D+7.7 · support factor 0.87 (the share of that channel's capacity the admitted polls carry) — separate inputs; their product is a capacity diagnostic, not a margin.
• Weights are the 55/25/20 capacities scaled by each channel's evidence and renormalized, so thin polling widens the interval instead of pulling the center toward zero — 12% of the weight is carried that way.
• Blended national environment: D+8.3.
• Applied at full strength to this chamber's races.
• This step’s recorded contribution to the margin is +2.1.
• Blended national environment: D+8.3 (updated 2026-10-14).
• Effective poll weight 88%; fundamentals share 12%, carried by the interval (×1.06) rather than by the center.
• The support factor is the share of the generic-ballot capacity the admitted polls carry; average × support is a capacity diagnostic, not a margin, and is not what the row above multiplies.
• Applied at full strength to this chamber: +2.1 of this margin.

Votes for other candidates +0.024 points

Every step above is measured between the two major-party candidates alone, as if nobody else were on the ballot. The forecast sets 0.7% of the vote aside for other candidates and write-ins, so as a share of every vote cast the gap between the two is smaller by that much. This step is exactly that: the two-party margin times the share going to other candidates. It is the same arithmetic that turns the two-party margin into the two shares printed at the top of the page.