Election Central Methodology

Source-backed public records

Election Central reads version-controlled JSON records from this site's public repository. A record enters public output only when it is marked both published and verified. Each record carries an original source, retrieval time, verification time, freshness status, and update time.

Race scope and filing information prioritize the Texas Secretary of State and other responsible election authorities. Federal finance summaries use Federal Election Commission records. State finance summaries use Texas Ethics Commission exports. We do not publish placeholder candidates, invented poll numbers, estimated vote totals, or person-level voter records.

Polling averages

A qualifying poll receives a multiplicative weight. The implemented formula is:

poll weight = recency × sample size × population × pollster quality × independence × methodology

Recency decays with a 21-day half-life. Sample-size weight grows with the square root of the sample and is capped so one very large poll cannot dominate the average. Likely-voter and primary-voter samples receive more weight than registered-voter or adult samples. Published pollster grades, internal or partisan sponsorship, survey mode, methodology disclosure, margin-of-error disclosure, and weighting disclosure also affect the final weight.

Candidate averages divide the weighted sum of reported percentages by the total usable weight for that candidate. Missing candidate responses are never imputed. The output includes the poll count, field-date range, weighted averages, weighted margin, an uncertainty range, recalculation time, and a per-poll explanation. A race without qualifying public polls displays an honest no-poll state.

Forecast model

The deterministic model combines the weighted polling margin, previous election margin, district partisan lean, incumbency, fundraising, the statewide election environment, and a documented candidate-quality adjustment. When usable polling exists, the current model gives polling the largest share of the expected-margin calculation. When it does not, the forecast is explicitly labeled fundamentals-based and polling averages remain blank.

Each output includes expected vote share, a rounded win probability, race rating, confidence level, change from the prior run, model version, sources, timestamp, and a fundamentals-based flag. Probabilities are derived from the expected margin with a fixed logistic conversion. They are model estimates, not facts or guarantees.

The daily repository workflow appends one forecast snapshot per race per calendar day. It does not overwrite the historical series. A forecast is generated only when the race, candidates, and required input sources are published and verified.

Candidate comparison

Comparison tables use the same structured candidate, finance, polling, endorsement, statement, and voting-record data shown elsewhere on Election Central. Positions are displayed only when a source documents them. The table says “No documented position found” instead of inferring a candidate's view from party affiliation or outside commentary.

Ballot geography

County and district associations are included only from authoritative mappings. ZIP-code coverage is displayed only when the imported source identifies it as authoritative, and ZIP codes are never presented as a substitute for an exact official ballot. County election pages and official sample-ballot links are shown when available. Address geocoding is not part of the launch model.

Validation, freshness, and results

Every production build validates unique identifiers and slugs, candidate-race relationships, HTTPS source links, required metadata, dates, poll percentages, forecast probabilities, fundamentals sources, publication status, and the absence of person-level voter data. Automated QA also checks launch-scope race coverage, duplicate candidates, generated routes, source freshness, polling and forecast fixtures, sitemap output, and unofficial-results labeling.

Election-night totals remain unofficial until certified by the responsible authority. The public homepage takeover remains disabled until the verified directory is populated, validation and QA pass, major routes contain real records, and polling and forecast pages show sourced information or accurate empty states.

See the corrections policy to report a factual issue.