# Data Dictionary: AFOS · United Kingdom 2024 Electoral Divergence

Every file and column, with type, unit and provenance. Party-level: polls measure party vote share, the market prices the probability of winning the most seats (FPTP plurality). Values are never imputed or smoothed, missing values are left **blank**.

---

## `polls/uk-polls.csv`

Party vote-share polling, **long format**, one row per party × poll.

| Column | Type | Notes |
|--------|------|-------|
| `poll_date` | date | Last date of poll (`YYYY-MM-DD`). |
| `fieldwork` | string | Date or window as published. |
| `pollster` | string | Polling firm (YouGov, Savanta, Deltapoll, Opinium, Ipsos, Survation, More in Common, …). |
| `sample` | integer | Sample size, blank when not declared (thousands separators normalized). |
| `party` | string | Party (canonical): Labour, Conservative, Lib Dems, Reform, Green, SNP, Plaid Cymru. |
| `party_full` | string | Full party name (Labour Party, Conservative Party, Liberal Democrats, Reform UK, …). |
| `percent` | number | Party vote-share intention, %. |

## `polls/uk-polls.json`

Full structured polls (pollster, fieldwork, sample, per-party results). Deduplicated by (pollster, date), keeping the most complete row. Parsed from the single national voting-intention table of 2024 (older-year and regional tables excluded).

---

## `data/uk-market-odds-timeseries.csv`

Polymarket **"wins the most seats"** probability per series, daily, from the "Which party wins the most seats after UK Election?" market.

| Column | Type | Notes |
|--------|------|-------|
| `date` | date | `YYYY-MM-DD`. |
| `party` | string | Labour, Conservative, Lib Dems, Reform, Other. |
| `polymarket_pct` | number | Implied probability of winning the most seats, %. |
| `volume_usd` | number | Cumulative traded volume for that series, USD (total market about US$ 1.76M). |

> Post-election resolution points (after `2024-07-04`) were dropped, so the series ends on election day. This removes a settlement spike on the Reform series on 5 July.

## `data/uk-divergence-timeseries.csv`

**Market × poll divergence** per party, each poll's vote share joined to that party's market odds on the poll's date (nearest **on or before**). Only the four parties present in both market and polls cross: Labour, Conservative, Lib Dems, Reform.

| Column | Type | Notes |
|--------|------|-------|
| `poll_date` | date | Poll date. |
| `pollster` | string | Polling firm. |
| `party` | string | Canonical party name. |
| `poll_pct` | number | Party vote share, %. |
| `polymarket_pct` | number | Market "wins most seats" probability on `polymarket_date`, %. |
| `polymarket_date` | date | Market date used (nearest on-or-before `poll_date`). |
| `divergence_pp` | number | `polymarket_pct` minus `poll_pct`, percentage points. |

> **Interpretation caveat:** a poll reports *vote share*, the market prices *probability of winning the most seats* (the FPTP plurality). The two are different quantities, so `divergence_pp` is the spread AFOS tracks, **not** an error metric. In 2024 Labour led the vote around 39% while the market priced it near 99% to win the most seats, and first-past-the-post turned 33.7% of the vote into 411 of 650 seats. Polls before the market opened (`2024-05-01`) are not joined: 523 such poll rows are left out of the divergence file by design, and that is logged when the build runs.

## `data/uk-poly-raw.json`

Raw Polymarket payload (event plus per-series price histories) for the most-seats market, via a US-resolving function, kept for provenance.

## `data/uk-poly-vote-share-raw.json`

Raw payload for the secondary "UK Election: Labour % of the vote?" market, kept for provenance.

---

**Sources:** [Wikipedia: Opinion polling for the 2024 United Kingdom general election](https://en.wikipedia.org/wiki/Opinion_polling_for_the_2024_United_Kingdom_general_election) · Polymarket "Which party wins the most seats after UK Election?". Maintained by [AFOS Analytics](https://afos-analytics.com). Data CC BY 4.0, code Apache 2.0.