Southern WAR methodology

1. Estimand

The score is a race differential, not a pooled candidate-career effect. A residual cannot uniquely divide credit between candidate strength, opponent weakness, and omitted local conditions.

2. Structural model

The selected decaying_lag ridge specification (alpha 100) models the ordinary legislative-ticket gap using ticket margin and its square, state, chamber, baseline office family, election timing, symmetric incumbency, prior presidential margin, ticket change, and a ticket-change-by-years interaction. Specification selection used earlier-cycle forward validation.

3. Historical scoring

Races after 2016 preserve the published Southern WAR v3 same-cycle fitted residual. The strict 2016 races are a backward application of the selected model fitted only on strict races after 2016.

4. Coverage and missing races

The explorer contains all 116 scheduled state/cycle/chamber map slices and every district outline in the exact election-year Census cartographic-boundary file. Only strict observed D–R regular contests receive WAR. Uncontested races, non-D/R races, research-only context, and missing outcomes remain unscored; gray never means WAR zero.

Louisiana, Mississippi, and Virginia retain their actual odd-year election schedules. South Carolina’s staggered Senate schedule is also retained.

5. Fundraising

Fundraising is displayed only when both major-party observations and both candidate identities are complete. The model tested viability gates at every $10,000 from $10,000 through $100,000, plus $250,000. Finance failed the prespecified nested time-forward promotion gate, so it does not enter headline WAR.

Missouri has no usable finance in the warehouse run underlying this map, and Mississippi has very limited electronic coverage. Other states have residual race-level gaps. Missing finance is unavailable, not zero.

6. Geography

Each slice uses the U.S. Census Bureau’s cartographic boundary released for that election year and chamber. A scored race must match one unique district feature in the exact state/year/chamber file. Census geometry is display evidence; it does not overwrite the warehouse’s provider-reported plan-vintage label.

7. State coverage and release limitations

One row per scheduled state: scheduled state/election/chamber maps, maps that contain no strict race, strict D-versus-R races that received WAR, the split between 2016 backcasts and published v3 same-cycle residuals, source-file lineage, excluded outcomes by class, finance overlay completeness, the plan-provenance label, and the exact run identifiers. The machine-readable copy is the state release coverage download linked below.

Historical WAR run

WAR-SOUTH-HIST-V1-6D84680E7B757B057AF1

Upstream Southern WAR v3 run

WAR-POST2016-V3-530FBD4238CC483E557C

Warehouse build run

RUN-504CE4C4DF904D88A5A40D268F3FCEAB

Machine-readable QA: Map join audit · State release coverage · Run manifest · Slice coverage

Score labels

Published v3 same-cycle residual. 3,660 races carry the exact residual from Southern WAR v3 run WAR-POST2016-V3-530FBD4238CC483E557C, fitted in the same cycle. 2016 backcast. 620 strict 2016 races are backward applications of the post-2016 model; the explorer labels them “post-2016-model backcast” and they are descriptive, not contemporaneous fits.

Excluded outcomes

Research-only outcomes (302; Florida, Georgia, Kentucky, Louisiana, Mississippi, Missouri, North Carolina, Oklahoma, South Carolina, Virginia) are outcomes the model carries only as context and never scores. Baseline not strict (151; Virginia): Virginia 2017 and 2021 lower-chamber outcomes whose governor baseline uses 2019-plan cross-election precinct membership. The baseline is not a same-plan observation, so those races are excluded rather than scored, which is why those two scheduled maps contain no strict race. Incumbency experimental (151; Florida, Georgia, Kentucky, Louisiana, Mississippi, Missouri, North Carolina, Oklahoma, South Carolina, Virginia): prior-winner continuity rows without roster evidence. Incumbency is a model input, so a race whose incumbency rests on an unverified continuity inference is excluded. Exclusion is not a claim that a district was uncontested.

Source lineage

All 4,280 strict races resolve to a registered source file.

Plan provenance

Plan-provenance labels by state: “provider-reported district; plan vintage unverified” (Georgia, Kentucky, Mississippi, Missouri, South Carolina, Virginia); “provider-reported district; plan vintage unverified | provider-reported; plan vintage unverified” (Arkansas, Florida, North Carolina, Oklahoma, Tennessee, Texas); “provider-reported; plan vintage unverified” (Alabama, Louisiana). “Provider-reported district” means the district number is the election provider’s; “plan vintage unverified” means the warehouse did not confirm which redistricting plan the provider applied; and the election-year Census boundary drawn in the explorer is display geometry that does not certify the provider’s allocation of votes to districts.

Finance overlay coverage

3,635 of 4,280 strict races (84.9%) display a complete two-candidate fundraising overlay. No finance-complete races: Mississippi, Missouri. Partial coverage: Arkansas 98.2%, Florida 96.6%, Georgia 95.8%, Kentucky 95.2%, Louisiana 94.4%, Oklahoma 95.2%, South Carolina 94.9%, Tennessee 99.7%, Texas 97.6%, Virginia 97.2%. Missing finance is unavailable, not zero, and finance never enters WAR.

Scheduled maps

116 scheduled maps; 2 contain no strict race (Virginia). Empty maps still show every district outline and explain the missing score.

8. Downloads

Race WAR · Candidate orientations · Coverage · State release coverage · Map join audit · Run manifest · Geometry sources

Run WAR-SOUTH-HIST-V1-6D84680E7B757B057AF1