Forecast methodology

1. Forecast identity

The selected post-2016 Southern WAR model supplies the structural expected gap, including its symmetric incumbency effect. A nominee with a matched prior Alabama race adds 0.43 of that race’s residual WAR, oriented to the Democratic margin; a rematch is counted once rather than twice.

2. National environment

The Silver Bulletin generic congressional ballot average (weighted by pollster rating, sample size and recency, with house-effect adjustment), converted to a two-party margin, is used as a stand-in for the election environment because Alabama-specific polling is sparse. Its change from the 2024 national presidential margin is applied uniformly to every district's 2024 presidential margin. The uniform national-to-Alabama transfer is an owner-selected model assumption; the validation audit does not establish its Alabama-specific validity beyond the single 2022 forward holdout. The Dem and Rep scenario tabs add or subtract one historical national polling-error standard deviation.

3. Candidate history

A nominee with a matched prior Alabama race carries 0.43 of that race’s WAR into this forecast; the rest are evaluated generically. Persistence is estimated on 1,576 repeat-candidate pairs in the Southern v3 panel with candidate-clustered standard errors, and the years-elapsed interaction is not statistically supported, so no decay is applied. Matching is verified-crosswalk first, then exact unique name; ambiguous names and results older than 8 years are left unmatched and enter as zero. 23 of the modeled races carry a candidate adjustment, averaging 1.5 margin points. Ideology and fundraising remain excluded: the payload records finance_used=false for every modeled race.

4. Post-2016-to-2022 forward test

The direct Alabama forward test fits the selected WAR design on 2039 eligible Southern races after 2016 and before 2022, then evaluates 33 Alabama races in 2022, including the WAR model's incumbency term. The generic-ballot district baseline records 7.07 points of MAE and 97.0% winner accuracy. The WAR structural specification alone records 7.65 points, and adding carried-forward candidate WAR records 7.35 points across the 9 holdout races that had a matched prior result. The published model is the WAR specification with candidate history: it is the owner-selected estimand, it improves on the structural model, and it still trails the generic-ballot benchmark on this single holdout. That comparison is published rather than hidden.

5. Probabilities and chamber summaries

Expected margins are converted to conditional probabilities with a Student-t distribution with five degrees of freedom and a 7.75-point scale, the maximum-likelihood scale of the 2022 holdout margin residuals (its nominal 80% interval covers 85% of them). The scale is tuned and evaluated on the same 33 races, so the probability layer has no independent evaluation. Headline chamber summaries use 50,000 correlated simulations with national (2.20), statewide (2.01), chamber (1.03), and district (5.91) error components read from robust_forecast_v1_error_components.csv, the live uncertainty input for this build. Single-major-party seats are fixed in chamber totals.

6. Limitations

• Only one direct post-2016 Alabama forward holdout is available; the selected structural specification still trails the generic-ballot-only benchmark on it.
• The national generic ballot is an imperfect proxy for Alabama's state-legislative environment.
• The probability scale is sample-limited and should not be read as long-run calibration evidence.
• Candidate history is only available for 23 of the modeled races; everyone else is evaluated generically, so the adjustment is uneven across districts.
• Persistence is measured over two-to-six-year gaps and applied up to 8 years, so 2018 results reach 2026 as a one-cycle extrapolation; older results are not carried at all.
• Roster and polling evidence can change before Election Day.

7. Data and audit downloads

District scenariosForward predictionsForward metricsManifestUncertainty componentsAlabama WAR ratings