DATA_DICTIONARY.md

Data Dictionary — AFOS USA 2024 Electoral Divergence

🌐 EN — Data dictionary for the AFOS USA 2024 electoral-divergence dataset (polls × prediction markets × press). · PT — Dicionário de dados do dataset AFOS EUA 2024 (pesquisas × mercados de previsão × imprensa). · ES — Diccionario de datos del dataset AFOS EE. UU. 2024 (encuestas × mercados de predicción × prensa). Column names and definitions below are kept in English (CSV/academic standard). · Os nomes e definições de coluna seguem em inglês (padrão dos CSVs). · Los nombres y definiciones se mantienen en inglés.

Poll figures trace to a named pollster's published release, compiled from FiveThirtyEight's public 2024 president-poll database (Wayback snapshot of 4 Nov 2024). Market odds come from public Polymarket markets; press anchors are public articles archived in the Wayback Machine. Nothing is imputed or smoothed; missing values are left blank.

polls/usa-national-polls.csv (long format)

One row per candidate per poll. National general-election polls only (state = blank, stage = general, cycle = 2024), Harris & Trump. 3,691 rows.

Column	Type	Notes
poll_id	integer	FiveThirtyEight poll identifier.
pollster	string	Polling firm (e.g. "Siena/NYT", "Marquette Law School", "Emerson").
end_date	date	End of fieldwork (YYYY-MM-DD).
sample_size	integer	Sample size.
population	string	Sampled population: lv (likely voters), rv (registered voters), a (all adults).
numeric_grade	number	FiveThirtyEight pollster quality rating (0–3 scale), where published.
methodology	string	Survey mode (e.g. "Online Panel", "Live Phone", "IVR/Online").
candidate	string	Harris or Trump.
pct	number	National voting intention for that candidate, %.
source_url	string	Primary source URL for the poll, as recorded by FiveThirtyEight.

data/usa-winner-market-timeseries.csv — PRIMARY AXIS (electoral college)

Daily Polymarket win-probability from the "Presidential Election Winner 2024" market (who becomes president; total volume ≈ US$ 3.7 bn — the largest election market in history). 613 rows.

Column	Type	Notes
date	date	YYYY-MM-DD (daily fidelity).
candidate	string	Trump or Harris.
win_prob_pct	number	Implied probability of winning the presidency, % (0–100).

data/usa-popularvote-market-timeseries.csv — COUNTERPOINT (popular vote)

Daily Polymarket win-probability from the "Popular Vote Winner 2024" market (who wins more popular votes; total volume ≈ US$ 628 M). This is the market that favored Harris and erred. 615 rows. Same columns as the winner series; win_prob_pct = implied probability of winning the popular vote.

data/usa-poll-aggregate-timeseries.csv

Trailing 7-day simple average of national poll vote share. 1,322 rows.

Column	Type	Notes
date	date	YYYY-MM-DD.
candidate	string	Trump or Harris.
poll_vote_share_pct_7dma	number	7-day trailing average of national vote-share polls ending on or before date, %.

data/usa-divergence-timeseries.csv

Market × poll divergence for the Harris–Trump head-to-head period (after Biden withdrew, 21 Jul 2024). 218 rows. Axis = the winner (electoral-college) market.

Column	Type	Notes
date	date	YYYY-MM-DD (market date).
candidate	string	Trump or Harris.
market_win_prob_pct	number	Winner-market implied probability of winning the presidency, %.
poll_vote_share_pct	number	National poll vote share (7-day trailing average), %.
naive_gap_pp	number	market_win_prob_pct − poll_vote_share_pct, in percentage points. Different units (probability vs share): reported raw, not scale-reconciled. The sign and trajectory are the signal, not the magnitude.

press/usa-2024-press-timeline.csv

Curated, Wayback-archived press anchors cross-referencing the market×poll divergence. 7 rows.

Column	Type	Notes
date	date	Article publication date (YYYY-MM-DD).
outlet	string	Publishing outlet (Fortune, Al Jazeera, NBC News, NPR, PBS NewsHour, CNN).
headline	string	Exact article headline.
theme	string	market_poll_divergence · poll_near_tie · market_conviction_whale · result_validator · market_vindicated · poll_accuracy_postmortem.
axis	string	Which AFOS axis the anchor maps to (market, poll, market×poll, result).
tier	string	anchor (core) or secondary.
primary_source	string	Underlying primary source where the carrier differs (e.g. "Associated Press (race call)", "Wall Street Journal (original)").
url	string	Live article URL (verified HTTP 200).
wayback_url	string	Internet Archive Wayback Machine snapshot (permanent citation).

data/usa-case-summary.json

Machine-readable summary of the validated case: result (electoral college + popular vote), both markets with eve-of-vote probabilities and verdict, final poll average, and the honesty note (two markets disagreed; the validator is the real result).

Raw provenance

data/usa-winner-poly-raw.json and data/usa-popularvote-poly-raw.json — raw Polymarket payloads (event metadata + per-candidate daily price histories) for both markets, kept verbatim for provenance.

data/usa-structural-context.csv

Structural country context from the World Bank, complementary to the divergence data: it frames the country, it does not predict the electoral outcome. Long/tidy format, one row per indicator, latest available year per indicator.

Column	Type	Notes
category	string	governance, economy, or education.
indicator	string	Machine code (e.g. political_stability, gdp_usd, expected_years_schooling).
label	string	Human-readable indicator name (English).
value	number	Governance on a 0-100 scale; economy in US$ / %; education in % of GDP or years.
unit	string	index_0_100, USD, percent, or years.
year	integer	Reference year of the value (latest available).
source	string	World Bank WGI (governance, via Data360) or WDI (economy & education).
iso3	string	ISO 3166 alpha-3 country code.

Both sources are open-licensed (CC BY 4.0) and keyless. Governance = Worldwide Governance Indicators; economy & education = World Development Indicators.