Datasetpublic

AFOS-Analytics1/usa-2026-midterms-divergence

AFOS — US 2026 Midterms Divergence (v1.0.0, pre-electoral) What the market priced and what the polls measured about the same election, day by day, before it happened. ⚠️ Pre-electoral bundle. The election is on November 3, 2026. There is no certified result to score these forecasts against, so this is not the gold-standard version. See DATASHEET.md. The v2, after certification, closes the cycle. What is inside 419 national generic-ballot polls from 71… See the full description on the dataset page: https://huggingface.co/datasets/AFOS-Analytics1/usa-2026-midterms-divergence.

sourceHugging Facecc-by-4.0updated 14h agoView on Hugging Face

1likes1.3kdownloads

Dataset Card

[image]

AFOS — US 2026 Midterms Divergence (v1.0.0, pre-electoral)

![Harvard Dataverse DOI](https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi:10.7910/DVN/XRUT8U)

What the market priced and what the polls measured about the same election, day by day, before it happened.

⚠️ Pre-electoral bundle. The election is on November 3, 2026. There is no certified result to score these forecasts against, so this is not the gold-standard version. See DATASHEET.md. The v2, after certification, closes the cycle.

What is inside

—419 national generic-ballot polls from 71 pollsters, fieldwork 2025-04-19 → 2026-09-17, 417 of them with the institute's own source URL
—25489 market price rows across 9 Polymarket contracts: chamber control, seat distributions, turnout, popular-vote margin, and whether the election happens on schedule
—537 headlines from 54 daily collections over a fixed list of 23 outlets
—One already decided race with its verified outcome: the Texas Republican Senate primary runoff

The one case that is already scored

The Texas Senate Republican runoff resolved on May 26, 2026. Ken Paxton beat John Cornyn with 63.8% of more than 1.38 million votes.

The market had priced Paxton between 57% and 62% for five straight weeks, then moved to 94.5% on May 21, five days before the vote. The price was a probability and the result is a vote share, so the two are not the same quantity and do not subtract. What the series does let you ask is a real question: was the pre-jump price well calibrated, and what moved on May 21.

How to read it without making the classic mistake

⛔ Do not subtract a market probability from a poll margin. One is the chance of controlling a chamber; the other is a lead in vote points. In 2012 Democrats won more votes and fewer seats, and the gap can be entirely geography.

⚠️ Filter by `market_slug`. House and Senate name their outcomes identically.

⚠️ Do not compare `LV`, `RV` and `A` samples as if they were one series.

Reproducing

bash

node scripts/build-us-2026-dataset.mjs    # data
node scripts/build-us-2026-metadata.mjs   # datasheet, dictionary, package, checksums

Both read only files versioned in the AFOS repository. CHECKSUMS.txt carries SHA-256 for every file.

Citation

See CITATION.cff. Harvard Dataverse DOI: [10.7910/DVN/XRUT8U](https://doi.org/10.7910/DVN/XRUT8U). Cite: Felipe, Andre, 2026, "AFOS · USA 2026 Midterms Electoral Divergence Dataset (v1, pre-electoral)", https://doi.org/10.7910/DVN/XRUT8U, Harvard Dataverse.

⚠️ The citation above names NO version number on purpose: the DOI resolves to the latest release, and a hard-coded version goes stale the next time any file changes.

Data CC BY 4.0, code Apache 2.0.