How the forecast works

Every input is public, and the model is a short Python script (forecast.py) that anyone can rerun. Win chances are model estimates with stated assumptions, not predictions of certainty.

1. Polls, adjusted for sponsor

Polls are weighted by recency (a 45-day half-life) and by the square root of sample size. Each poll from a partisan-aligned sponsor is shifted 3 points against its sponsor's party. That adjustment is an assumption, not an estimate. Polls with documented data fabrication are excluded. Current average: R+4.3 raw, R+7.3 adjusted.

2. Fundamentals

We start from Mississippi's 2024 presidential margin (R+22.9) and move it 0.8 points for each point the national environment has shifted since 2024 (R+1.5 nationally), measured by the generic-ballot average (D+8.9). The 0.8 is the average of Mississippi's 2016–20 and 2020–24 elasticities. Result: R+14.6. Applied to 2018, the formula gives R+12.6, compared with an actual average of R+13.1 across that year's two Senate races.

3. Blend and uncertainty

The Senate projection is 40% polls and 60% fundamentals, or R+11.7. The forecast error standard deviation is assumed to be 7 points for the Senate, 8 for the House and 10 for 2027. Win chance is the normal CDF of margin divided by SD, and the 90% range is ±1.645 SD. The blended figure averages the model, the rater consensus (Likely R, treated as 92%) and the two prediction markets.

4. County map and turnout

County projections start from the Nov. 2018 Hyde-Smith–Espy runoff, which had the same incumbent in a midterm under a Republican president, and shift it uniformly to the statewide projection. A vote-weighted ecological (Goodman) regression of that runoff on each county's Black population share estimates Democratic support at about 98% among Black voters and 14% among other voters. These estimates come from aggregate data and carry ecological-inference error. The turnout slider changes each county's Black share of the electorate, and its vote weight, using those rates.

5. 2027 offices

The average 2019/2023 down-ballot statewide margin (R+18.5) is the low-profile-Democrat baseline. Hood (2019) and Presley (2023) ran 14.3 points better than their party's down-ballot average, which sets the strong-Democrat case. The mid-tier case uses half that gap. County projections for governor shift the 2023 county map to each scenario.

6. House

Each district's 2024 presidential margin on current lines, moved by the same national shift.

What we don't model

• Candidate-specific regional surges beyond the 2018 map
• Weather, late news, or ballot-access changes
• Precinct-level effects (planned for Pro)

Sources

Downloads: county_results_long.csv · forecast_county.csv · forecast.json