ChamberCast is the only state-legislative forecasting model we trust with our own clients.
91.2% accuracy across 4,600+ state-house race calls since 2016 — built in-house by coalition principals, audited annually by the methodology team at Marist, and used on every race we touch. Below: who built it, who checks it, what it reads, and where it has missed.
Built by legislators' staff. Audited by a methodology team that doesn't work for us.
ChamberCast was developed in-house beginning in 2015 by three coalition principals — a former Ohio House senior budget staffer, a Florida Senate redistricting counsel, and a Colorado House minority caucus data director — and went live ahead of the 2016 cycle as the firm's internal forecasting instrument. It has been used on every coalition client race since, including all 347 sitting legislators currently on the active roster.
Each January, the model is turned over to the methodology team at Marist for an independent audit: feature review, back-testing on the prior cycle's closed races, holdout validation, and a written methodology letter attached to the firm's annual accuracy report. The 91.2% headline figure is the audited figure — not a self-reported one — and is calculated only against race calls, not against modeled probabilities or directional signals.
— Office of the Chief Methodologist, Chris Jones Coalition LLC · Columbus, Ohio
Audited accuracy, cycle by cycle, since launch.
Headline figure and supporting metrics drawn from the Marist methodology letter, January 2024 review.
A forecasting model is only as honest as the misses it publishes. We publish the misses — every cycle, by district — because that's what an audit is for.
Four data inputs. One model. No black box.
ChamberCast ingests four streams of record-grade data. Each refreshes on a defined cadence. Each contributes a defined weight. The full feature table is reproduced in the annual methodology letter.
-
01
State Secretary of State returns
Certified and unofficial returns pulled directly from each state's SOS or equivalent election division on election night and at the canvass. Refresh cadence: real-time on election night; 24-hour intervals during the canvass window. This is the model's ground-truth anchor — every prediction is back-tested against this stream.
Weight: 38% of feature contribution -
02
Historical district baselines
Twelve-cycle district-level vote histories, normalized for redistricting cycles using a boundary-pairing algorithm cross-checked against the National Institute on Money in Politics index. Refresh cadence: monthly, with an immediate refresh within 72 hours of any certified redistricting order. Anchors the model's read on partisan drift in low-turnout primaries.
Weight: 27% of feature contribution -
03
Incumbency strength modeling
A composite of vote-share trajectory, bill-signing record, committee assignments, leadership tenure, and same-district favorability — assembled from legislative journals, committee vote indexes, and the firm's proprietary Lawmaker Equity Score. Refresh cadence: quarterly, with a real-time update when a client files for re-election.
Weight: 22% of feature contribution -
04
Independent expenditure tracking
Daily pulls from state-level IE filings, cross-referenced with IRS 527 and state ethics disclosures for outside-spending flows into each district. Refresh cadence: 24-hour intervals during the 90-day pre-primary and pre-general windows; weekly outside those windows. Captures the late-cycle spending shock that moves the largest number of legislative races.
Weight: 13% of feature contribution
See what ChamberCast reads in your race.
A 30-minute diagnostic call with a coalition principal. We walk through ChamberCast's current read on your chamber and district, the cycles it tracks closest, and where the model is flagging exposure — before you decide anything.
Diagnostic calls are scheduled within 72 hours of request. Calls are conducted under standard consulting privilege and are not retained on the client's public record.