How district lines turn votes into seats

Results

One deciding race per district per election: the runoff where one was held, otherwise the November general (Louisiana’s November all-party primary; Texas’s court-ordered 1996 and 2006 races). Ranked-choice races use the final round; candidates on several party lines have them summed. 1976–1980 and 1998–2024 come from FiveThirtyEight’s candidate-level returns; 1982–1996 from MIT’s Election Data and Science Lab, which digitized the Clerk’s statistics.

Verification. For 1998–2024 every district was compared with the Clerk of the House’s official Statistics of the Congressional Election: 98.9% match exactly and no winner differs. Where the Clerk shows the compiled file is wrong (a missing candidate, a mislabeled party, a partial count, a typo) the Clerk’s figure is used, 13 districts in all (11 of them in 1998–2024, the others in 1976). For 1982–1996, 14 candidates cross-nominated by both parties were filed under the wrong party and are corrected; with that, every election reproduces the House’s party divisions. North Carolina’s 9th district in 2018 counts as vacant: that result was never certified.

Efficiency gap

A vote is wasted if it was cast for a loser, or for a winner beyond the half needed to win. The gap is Democratic minus Republican wasted votes over all votes; positive numbers favor Republicans. Multiplied by a state’s seats, it reads as seats gained from the asymmetry. Races without both major parties are imputed at 75–25 on that district’s turnout (or the state’s average contested turnout when no votes were reported). One-seat states are excluded.

Neutral simulated maps

From 2012 the headline compares each enacted map with the ALARM Project’s computer-drawn maps for that state, 5,000 to 30,000 per state (McCartan et al., Scientific Data, 2022), which follow the state’s legal criteria but use no partisan data. All maps are scored on the same 2016–2020 statewide races; seats from the map are the neutral median’s expected Democratic seats minus the enacted map’s. Each simulation is used only for elections held on the plan it simulated, confirmed by matching district presidential results: all 2020-census plans match the 2022 lines; for the 2010 census, Florida, North Carolina, Pennsylvania and Virginia do not match their post-remedy lines and are left out, as are the five states that redrew again for 2024. The result depends on how the rules are encoded, and expected seats describe the map, not one election.

Presidential-vote baseline

The same measures on the presidential vote in each district (2012–2024), which no incumbent, challenger or uncontested seat distorts. District figures are The Downballot’s calculations on the lines each House election used: the same year’s presidential result where one exists, otherwise the most recent earlier one on identical lines. They are checked against the FEC’s official statewide totals and, for 2020, district by district against the ALARM Project’s precinct-based figures (425 of 429 districts within half a point). Earlier maps have no complete district results; North Carolina’s 2020 lines have none.

District maps

Lines for 1982–2010 come from Lewis, DeVine, Pritcher and Martis’s congressional district boundaries; 2012–2024 from the Census Bureau’s cartographic boundary files (North Carolina 2020 from Lewis et al.); 2026 from the Census Bureau’s 2026 TIGER files. They are simplified for the page, so very small islands are not drawn, and New York’s 33rd district is missing from the 1982–1990 source.

2026 redraws (projections)

The plans in force for November 2026 come from Loyola Law School’s plan table (as of September 10, 2026): a state counts as redrawn when its plan’s passage date or drawer differs from 2024’s, and the build checks that exactly those states have new district figures. The projection counts districts carried by Trump in 2024 on the old and new lines, using The Downballot’s 2024 presidential results on both; every state’s statewide votes must agree between the two within half a percent. It measures the lean of the lines, not who will win, and plans in litigation may still change before the election.

Mean − median

The average district’s Democratic share minus the median district’s. When one party’s voters are packed into a few lopsided districts, the median falls below the mean. Positive numbers favor Republicans. It is most informative in competitive states.

What this cannot show yet

Asymmetry is not proof of intent: Democratic voters cluster in cities, which wastes votes under any map, and wave elections swing the efficiency gap by themselves. The neutral-map comparison separates geography from line-drawing better than either asymmetry measure, but it exists only for the 2010- and 2020-census plans, not for the maps redrawn for 2026. The 2026 figures are projections on the 2024 presidential vote until that election is held and certified.