The research brief and the white paper report these findings. The paper’s exact supporting tables have not been independently reproduced; these are not independently verified website results.
County characteristics and voting history explain more than local economic change.
The research brief reports that county composition and voting history explain about half the variation in departures. Measured economic and population changes add little.
The white paper describes the sample, election coverage and model, but its tables have not been independently reproduced here. Explanatory power is not vote share; no precise incremental percentage is plotted.
Reported explanatory power · awaiting source verification
Local changes: small addition reported; exact value unavailable.
Economic deterioration did not consistently lead to losses for the incumbent party.
The brief reports slightly pro-incumbent movement, about a quarter of a point relative to the state, among the bottom 5% for relative employment growth. It describes ten tests without a consistent relationship.
The subgroup, employment definition, weighting and ten-test ledger remain unverified. This does not mean hardship benefits incumbents or disprove economic voting nationally.
Toward the incumbent, relative to the state · reported subgroup average, unverified
Volatility is associated with characteristics of places.
The brief identifies smaller electorates, lower educational attainment, transfer-income reliance, previous large swings, population below past peaks, and rapid population growth or decline.
These are reported associations, not individual voter motives or causal effects. Their exact estimates and publication status remain unverified.
The forecasting improvement did not hold up in the next election.
A separate released test; prospective validation remains unavailable. This released out-of-time durability test asks a different question: can public-data models predict later county vote share? It does not verify the brief’s “about half explained” result.
How far were held-out county predictions from the result?
County mean absolute error · percentage points of two-party vote share · lower is better.
Persistence baseline
1.80 pp
Combined public-data model
6.47 pp
Preregistered historical durability test · 2016→2020 · 1,445 test rows. Different outcome, specification and sample from the brief; not a comparison of explanatory power.
Paired harm from the combined model: +4.67 pp MAE (95% interval +4.49 to +4.85). Economics may still matter; this test does not establish a transferable signal.
Source: county_economic_model_v2, exported in model-release.json. No published pre-election county expectations or freeze timestamps are available.
What each measure means
Partisan position
A party’s share of the vote at one election. This does not measure change.
County movement
The change in Democratic two-party share between two elections; pp are percentage points, not percent change.
Relative swing
County vote-share change minus the corresponding statewide vote-share change. Positive means more Democratic relative to statewide movement.
Departure magnitude
The absolute relative swing, with the sign removed. This is the main map measure.
Historical p.Vol
A within-election rank of departure magnitude in the research’s published eligible universe. That universe and ranking convention have not been supplied, so no p.Vol values are displayed. A 90th percentile is a rank, not a 90% probability.
Expected volatility · planned
A released pre-election estimate would require its definition, model version, freeze timestamp, polling benchmark/cutoff and uncertainty. None is available here. Planned research remains separate from historical results.
Competitiveness
How close the parties’ vote shares are, distinct from how much a county moved.
Electoral leverage
How consequential a county might be under a specified election simulation. The existing leverage research is a comparison conditional on its assumptions; it is not p.Vol or a forecast.
Economic conditions
Dated measures of jobs, income, population, housing and migration. Adjacent figures do not establish why people voted.
Illustrative example · not an observed county
A county’s Democratic two-party share rose by 4 percentage points while its state’s rose by 3. Its relative swing was +1 pp; its departure magnitude was 1 pp. Vote-margin change would be twice the share change.
Evidence still needed to verify the paper’s results
The claim-to-table ledger, eligible-universe and tie convention, exact R² specifications, employment-growth subgroup definition, and association estimates are not supplied. The three reported findings retain this status until those sources are checked.
Post-election evaluation would separate shared state polling error from county-specific departure, then compare expected and realized volatility using a published decomposition. No polling error is counted as county volatility here.
Few presidential cycles constrain national inference; detailed economic datasets cover different windows; historical cost-of-living data was unavailable for the reported analysis. County patterns do not establish individual behavior or causation, and a small incremental economic contribution does not make economics irrelevant.