A voting-systems experiment
Senators are elected by statewide popular vote. This re-runs all 100 sitting senators' most recent elections under a different rule — winner-take-all by congressional district, the way the Electoral College awards states — to see where, and how often, the counting rule and the popular vote pick different winners.
Congressional districts hold roughly equal population by law, so awarding a seat to whoever carries the most of them is winner-take-all weighted by people — the Electoral College's logic, one level down. This leaves out the College's +2 small-state bonus, which has no district equivalent, so it captures only the winner-take-all-plus-districting part.
The same 100 elections, counted three ways. Reading down the list, each rule weighs a little less of how many voters and a little more of where they were — and the partisan balance moves in one direction the whole way down.
The Democratic total slips one seat at each step — 45, then 42, then 41 — as the rule moves from counting voters to counting districts, and the Republican total rises to match. The middle rule (vote-total) is the less district-dependent of the two: it ignores district lines and weighs only turnout, and it splits the five tied states both ways — giving Montana and Maine to the Democrats and Maryland to the Republicans.
Tammy Baldwin won 28,781 more votes than Eric Hovde — but carried only two of Wisconsin's eight districts, dense Madison and Milwaukee, while Hovde carried the other six. Both district rules therefore give the seat to Hovde. It's the clearest single-state look at how the three rules can diverge.
In most states the district method reseats the same senator. The exceptions cluster in closely divided states with lopsided district maps.
Tiles show each state's district split. Hover for detail. Six single-district states (AK, DE, ND, SD, VT, WY — plus Montana in 2020) have one at-large district, so the method is identical to the popular vote and always holds.
A seat flips exactly when a candidate wins the state but carries a minority of its districts — every point below the line. With one exception they are the same party, which is what makes the divergence look systematic rather than random.
Five of 100 seats flip; five more survive only because ties break to the statewide winner.
How much of this is just gerrymandering?
Less than you might expect. The district-count rule can't separate the winner-take-all effect from how the lines are drawn. The vote-total rule can: it ignores district boundaries and weighs only turnout — and it still overturns seats. That points to a cause beyond any single map: how each party's voters are distributed across the state.
Is five out of a hundred a meaningful number?
It's small, and worth keeping in perspective. Two things make it notable anyway: the reversals are directionally consistent rather than scattered across both parties, and control of the Senate has repeatedly come down to one or two seats.
Senators aren't chosen by district — so what does this measure?
It's a controlled counterfactual: real returns, one rule changed, everything else held fixed. It measures what winner-take-all-by-district does to a set of known outcomes — the same family of rule the Electoral College uses to award states in a presidential race.
What's measured. For each state's two current senators, we take their most recent election and ask who carried each congressional district. The district-count method awards the seat to whoever carried the most districts (winner-take-all; exact ties break to the statewide popular-vote winner). The vote-total method credits each district's entire turnout to whoever won that district and sums it — a turnout-weighted variant with no ties. Round 1 is each state's more-recently-elected senator; Round 2 is the other seat. 2020-cycle races use the district maps in force then (e.g., Montana was a single at-large seat in 2020).
The vote-total data. Wikipedia's district tables give percentages, not vote counts, so per-district turnout for the 18 competitive seats was collected separately from official state canvasses, Daily Kos/The Downballot datasets, and county→district crosswalks, each validated to sum to the certified statewide total. Landslide seats (a clear district majority) are the same winner under all three methods. A handful of split-county allocations are estimates, flagged in the data but never close enough to change an outcome.
Sources. District-level Senate results are from each race's Wikipedia "By congressional district" table, which cite official state election agencies and Dave's Redistricting App; every flip, tie, and close call was independently re-verified against those primary sources. Four blowout races have no published district table (Louisiana's 2020 & 2022 jungle primaries, Arkansas 2020, Massachusetts 2020); their splits are estimated but the outcome — an easy hold — is not in doubt. Maine 2020's district split carries a Wikipedia sourcing caveat but matches the state's known geography.
Caveat. This isolates the winner-take-all + districting effect and omits the Electoral College's +2 small-state bonus, which has no sub-state equivalent — so it is a partial analog, not the full system.