Akashic
1892–2024
akashic / research / coalition cycles

How coalitions move between elections

Election maps show where margins changed. They do not, by themselves, show whether voters switched, stayed home, entered the electorate, moved, or were replaced by a new generation. This research program separates the observable raw-vote components first, then tests the historical explanation.

Validated national build · 2026-07-15

What the six-period diagnostic found

The build evaluated 6,108 county-period rows and 24 cohort-period summaries. The compact result below is tied to workflow run 29423927515; its artifact digest is retained with the committed data contract.

2020→2024 cohortCountiesD vote changeR vote changeTwo-party swingD-loss shareDiagnostic
New Netherland core proxy16−799,128+336,164−14.15 pts70.4%persuasion or electorate replacement signal
Biden high-income metropolitan suburbs14−364,400+229,582−8.28 pts61.3%persuasion or electorate replacement signal
Southern majority-Black counties93−149,961−27,671−2.29 pts100.0%broad turnout decline
Midwestern German, Norwegian, Dutch, French, or French-Canadian counties895−410,602+360,028−2.76 pts53.3%persuasion or electorate replacement signal
New Netherland · two similar maps, different arithmetic

2004 was growth; 2024 was withdrawal plus growth

From 2000 to 2004, Democrats still added 174,665 votes in the 16-county proxy, but Republicans added 529,805, producing a −10.29 pts swing. Obama then added 496,475 Democratic votes while Republican votes fell 111,202 from 2004 to 2008.

From 2020 to 2024, Democratic votes fell 799,128while Republican votes rose 336,164. Democratic loss was 70.4% of the two gross arithmetic drivers. That makes the 2024 result more dependent on nonparticipation than the visually similar 2004 swing.

Long sorting cycle

The Midwest break preceded Clinton, then deepened

The 895-county ancestry proxy moved −6.67 pts from 2008 to 2012 as Democratic votes fell 1,145,800 and Republican votes rose 557,345. The 2012→2016 movement was larger—−9.29 pts—with a Democratic loss of 1,524,768 and a Republican gain of 720,194.

The data therefore supports acceleration and completion under Clinton, not a clean 2016 starting line. Candidate effects still matter, but the longer sorting cycle was already observable in Obama’s reelection.

The 2020 bridge

The most Biden places expanded both parties

In the 14-county high-income suburban diagnostic, Democratic votes rose 948,974 from 2016 to 2020 and Republican votes also rose 372,943. The net result was a +5.23 pts Democratic swing inside a rapidly growing electorate.

Turnout before conversion

The Black Belt result was broad turnout decline

Across 93 Southern majority-Black counties, Democratic votes fell 149,961 from 2020 to 2024—but Republican votes also fell 27,671. The cohort moved −2.29 pts because the Democratic decline was larger, not because the aggregate Republican raw vote grew.

This supports the turnout interpretation for this fixed cohort while remaining compatible with real Republican persuasion in other Black electorates or demographic subgroups.

The decomposition

For each fixed county or cohort, the tool computes the change in Democratic votes, Republican votes, other votes, total votes, and two-party margin. A material change exceeds the larger of 25 votes or 0.5 percent of baseline two-party votes.

DiagnosticObserved conditionInterpretive limit
Persuasion or electorate replacement signalDemocratic votes fall materially while Republican votes rise materially.Consistent with switching, but also migration, mortality, first-time voters, and cohort replacement.
Democratic withdrawal signalDemocratic votes fall materially while Republican votes remain approximately flat.A margin shift driven primarily by nonparticipation, not Republican raw-vote growth.
Republican mobilization signalRepublican votes rise materially while Democratic votes remain approximately flat.Consistent with new Republican participation; it does not identify the source population.
Broad turnout declineBoth major-party vote totals fall materially.Compare the magnitude of each decline and population change before assigning partisan meaning.
Electorate growthBoth major-party vote totals rise materially.A share swing inside a growing electorate may describe differential growth rather than defection.
Reproducible cohorts and periods

The committed analysis runs 2000→2004, 2004→2008, 2008→2012, 2012→2016, 2016→2020, and 2020→2024 for four cohorts. Every cohort definition is machine-readable; the complete CI artifact retains county-level CSV/JSON, while the compact reviewed result committed here contains all 24 cohort periods without adding the 7.8 MB county payload to the application bundle.

Related instruments: Political Twins, the data table, and the Foresight election-night desk.

County aggregates cannot identify individual voter transitions. Driver shares are arithmetic components, not causal persuasion rates. The output is a disciplined hypothesis generator, not an individual-level causal estimate.