Skip to content
Contents
The Public Mind

    Type to search. Planned chapters match by title only.

    Chapter 8.1

    Five axes of the county vote

    In 2024, college share had the largest correlation with the county two-party margin among five county characteristics, +0.66 vote-weighted, and it is not separated from rural share's −0.65.

    Affiliation
    NYU Stern
    Published

    In the 2024 presidential election, counties with more college graduates voted more Democratic and counties with more rural residents voted more Republican, and the two relationships were about equally close. Counting each county by its two-party votes, the correlationCorrelationA number from minus 1 to plus 1 for how closely two measures move together in a straight-line pattern. Zero means no straight-line pattern; a correlation does not say that one measure causes the other. of the county two-party marginCounty two-party marginThe Democratic vote minus the Republican vote, divided by the sum of the two, so a positive value is a Democratic lead. Votes for other candidates are left out., the Democratic votes minus the Republican votes over their sum, was +0.66+0.66 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with college shareInterval95% CI 0.60 to 0.72n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. with the share of adults holding a bachelor's degree and −0.65−0.65 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural shareInterval95% CI −0.74 to −0.56n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. with the share of residents living in rural areas. These characteristics overlap across counties, and so do the non-Hispanic white share, median household income and median age, the other three of the five numbers in this chapter. A correlation with any one of them therefore carries part of the others.

    Which of the five goes most closely with the margin has three answers, because there are three comparisons. A correlation sets the margin beside one number alone. A coefficient from a regression that holds all five together compares counties that match on the other four. The fit of that regression, the share of the variance of county margins the five reproduce as a set, describes them together without ranking them.

    Alone, college share and rural share are not separated

    The first step is the correlation of each number with the margin, one at a time. In 2024, vote-weighted, college share and rural share had the two largest correlations in size, with opposite signs. The white share came next, at −0.57−0.57 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.71 to −0.44n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state., and income and median age trailed. College share and rural share are not separated. The difference in the size of their correlations is +0.01 (95% CI −0.05 to 0.09)+0.01 difference in absolute correlationEstimandabsolute correlation of the first-ranked axis minus that of the second-ranked axisInterval95% CI −0.05 to 0.09Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates., an interval that includes zero, and when whole states are resampled college share has the largest correlation of the five in 60%60% share of resamplesEstimandshare of resamples in which college share has the largest absolute correlation with the county margin of the five axesn3,103Weightstwo-party votes in the electionPopulation2,000 resamples of statesPeriod2024 presidential electionSourceCounty demographics panel of the draws.

    The second step holds the five together in one regression, so that each coefficient describes counties that match on the other four. Here college share has the largest coefficient in 100%100% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,103Weightstwo-party votes in the electionPopulationresamples of statesPeriod2024 presidential electionSourceCounty demographics panel of resamples of states. College share's standardized coefficientStandardized coefficientA regression coefficient measured in standard deviations of the variables involved. It lets effects on differently scaled variables be compared., in standard deviations of the margin per standard deviation of college share, is +0.71+0.71 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on college share in a regression of the county margin on the five axesInterval95% CI 0.63 to 0.79n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state., and rural share's falls to −0.12 (95% CI −0.25 to 0.01)−0.12 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on rural share in a regression of the county margin on the five axesInterval95% CI −0.25 to 0.01n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state., an interval that includes zero. Income changes sign. Alone, counties with higher median household income voted more Democratic in 2024. Among counties with the same college share, white share, median age and rural share, those with higher incomes voted more Republican, with a coefficient of −0.26−0.26 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on median household income in a regression of the county margin on the five axesInterval95% CI −0.40 to −0.11n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state..

    Figure 1

    Held together, college share's coefficient is +0.71 against a correlation of +0.66, and income's is −0.26 against +0.40, vote-weighted in 2024

    Standardized coefficients for the county two-party margin in 2024: each axis alone, all five axes together, and all five with a baseline for each state; two weightings

    • Each axis alone
    • All five together
    • All five with state baselines

    Vote-weighted

    County-equal

    Standardized coefficient (standard deviations of the margin per standard deviation of the axis)

    Show data
    Data for Held together, college share's coefficient is +0.71 against a correlation of +0.66, and income's is −0.26 against +0.40, vote-weighted in 2024
    ElectionWeightingAxisModelCoefficientInterval lowInterval highCounties
    2024Vote-weightedEducationEach axis alone+0.66+0.60+0.723,103
    2000Vote-weightedEducationEach axis alone+0.28+0.18+0.383,113
    2024Vote-weightedIncomeEach axis alone+0.40+0.29+0.503,103
    2000Vote-weightedIncomeEach axis alone+0.10−0.01+0.203,113
    2024Vote-weightedRace and ethnicityEach axis alone−0.57−0.71−0.443,103
    2000Vote-weightedRace and ethnicityEach axis alone−0.52−0.68−0.363,113
    2024Vote-weightedAgeEach axis alone−0.31−0.41−0.213,103
    2000Vote-weightedAgeEach axis alone−0.08−0.19+0.043,113
    2024Vote-weightedRuralityEach axis alone−0.65−0.74−0.563,103
    2000Vote-weightedRuralityEach axis alone−0.46−0.56−0.363,113
    2024County-equalEducationEach axis alone+0.56+0.47+0.643,103
    2000County-equalEducationEach axis alone+0.10+0.01+0.203,113
    2024County-equalIncomeEach axis alone+0.30+0.17+0.433,103
    2000County-equalIncomeEach axis alone+0.02−0.10+0.143,113
    2024County-equalRace and ethnicityEach axis alone−0.50−0.71−0.283,103
    2000County-equalRace and ethnicityEach axis alone−0.40−0.54−0.263,113
    2024County-equalAgeEach axis alone−0.20−0.27−0.123,103
    2000County-equalAgeEach axis alone−0.17−0.28−0.053,113
    2024County-equalRuralityEach axis alone−0.53−0.59−0.463,103
    2000County-equalRuralityEach axis alone−0.27−0.36−0.183,113
    2024Vote-weightedEducationAll five together+0.71+0.63+0.803,103
    2000Vote-weightedEducationAll five together+0.25+0.11+0.393,113
    2024Vote-weightedIncomeAll five together−0.26−0.40−0.113,103
    2000Vote-weightedIncomeAll five together−0.17−0.28−0.063,113
    2024Vote-weightedRace and ethnicityAll five together−0.43−0.57−0.293,103
    2000Vote-weightedRace and ethnicityAll five together−0.50−0.66−0.353,113
    2024Vote-weightedAgeAll five together+0.04−0.13+0.213,103
    2000Vote-weightedAgeAll five together+0.29+0.10+0.493,113
    2024Vote-weightedRuralityAll five together−0.12−0.25+0.013,103
    2000Vote-weightedRuralityAll five together−0.25−0.38−0.113,113
    2024County-equalEducationAll five together+0.56+0.48+0.643,103
    2000County-equalEducationAll five together+0.03−0.09+0.143,113
    2024County-equalIncomeAll five together−0.14−0.23−0.063,103
    2000County-equalIncomeAll five together−0.01−0.12+0.103,113
    2024County-equalRace and ethnicityAll five together−0.47−0.67−0.283,103
    2000County-equalRace and ethnicityAll five together−0.39−0.55−0.223,113
    2024County-equalAgeAll five together+0.13+0.07+0.203,103
    2000County-equalAgeAll five together+0.09−0.02+0.203,113
    2024County-equalRuralityAll five together−0.20−0.30−0.103,103
    2000County-equalRuralityAll five together−0.20−0.28−0.123,113
    2024Vote-weightedEducationAll five with state baselines+0.73+0.65+0.803,103
    2000Vote-weightedEducationAll five with state baselines+0.26+0.14+0.383,113
    2024Vote-weightedIncomeAll five with state baselines−0.41−0.48−0.333,103
    2000Vote-weightedIncomeAll five with state baselines−0.25−0.32−0.173,113
    2024Vote-weightedRace and ethnicityAll five with state baselines−0.62−0.76−0.483,103
    2000Vote-weightedRace and ethnicityAll five with state baselines−0.77−0.90−0.643,113
    2024Vote-weightedAgeAll five with state baselines+0.08+0.02+0.143,103
    2000Vote-weightedAgeAll five with state baselines+0.22+0.11+0.333,113
    2024Vote-weightedRuralityAll five with state baselines−0.09−0.14−0.043,103
    2000Vote-weightedRuralityAll five with state baselines−0.10−0.19−0.013,113
    2024County-equalEducationAll five with state baselines+0.54+0.47+0.603,103
    2000County-equalEducationAll five with state baselines+0.15+0.06+0.243,113
    2024County-equalIncomeAll five with state baselines−0.18−0.23−0.133,103
    2000County-equalIncomeAll five with state baselines−0.15−0.23−0.083,113
    2024County-equalRace and ethnicityAll five with state baselines−0.73−0.84−0.613,103
    2000County-equalRace and ethnicityAll five with state baselines−0.68−0.76−0.603,113
    2024County-equalAgeAll five with state baselines+0.10+0.06+0.153,103
    2000County-equalAgeAll five with state baselines+0.12+0.06+0.173,113
    2024County-equalRuralityAll five with state baselines−0.11−0.15−0.063,103
    2000County-equalRuralityAll five with state baselines−0.11−0.17−0.053,113

    Read each row in the left panel from the grey dot, the correlation of that number alone, to the dark dot, its coefficient with the other four held fixed. The Education row stays close to where it starts, the Rurality row moves toward zero and the Income row crosses it. The third dot in each row adds a baseline for each state, and the right panel counts each county once. Whiskers are intervals with errors clustered by state. Values in Table 3.

    Note: Election returns are full counts of votes, so they carry no sampling error. The 95% intervals use errors clustered by state, so they show how much a number depends on which states are in the set. They do not include the sampling error of the county characteristics, which are census and American Community Survey estimates interpolated between years.Note: Standardization uses the weighted mean and standard deviation of the weighting shown, in the election shown.Source: County presidential returns, 2000-2024 (Algara and Amlani for 2000 to 2020; MIT Election Data and Science Lab for 2024), full counts of votes; county characteristics from the decennial census and American Community Survey, interpolated to election years; estimates weighted by two-party votes or counting each county once, as stated

    Counted votes and interpolated census measures

    The margins are counted votes, not survey answers. They are the county presidential returns for the seven elections from 2000 to 2024, from the Algara and Amlani county file through 2020 and the MIT Election Data and Science Lab for 2024. Alaska is left out because it reports by election district. Each election uses about 3,1003,103 countiesEstimandsmallest number of counties in an election's sampleWeightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 to 2024SourceCounty election returns counties, those with returns and a complete set of census measures, and the report lists which comparisons share the same counties, weights and other variables (Table 1). The five numbers come from the decennial census and the American Community Survey, interpolated to election years, so only 2000 and 2024 are close to a direct measurement. Every statistic is computed two ways, with counties weighted by their two-party votes and with each county counted once, because the largest counties dominate the first and the many small counties the second. A full count has no sampling error, so the intervals here answer a different question, how much a number depends on which states are in the set. They come from errors clustered by stateClustered standard errorsStandard errors that allow answers within the same group, such as the same county or the same week, to be related. They are wider than ordinary errors when members of a group resemble each other. or from resampling whole states, and they do not include the sampling error of the census measures. The regressions are descriptive, and their fitted shares are in-sample, scored on the same counties they were estimated on, and not forecasts. The report gives every estimate with its interval and sample size.

    The fit rose between 2012 and 2016 and then held

    The third step is the share of the variance of county margins that the five numbers reproduce together, the R-squared of the regression. It is a different quantity from any single correlation, and it can rise while the order of the five numbers changes. Vote-weighted, the five reproduced 41%41% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 30% to 56%n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. of the variance in 2000 and 69%69% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 63% to 78%n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2024, and the rise was not steady. Most of it came between 2012 and 2016, +20+19.9 percentage points of the variance of the county marginEstimandchange between the two elections in the R-squared of the five-axis regression (the same states are drawn for both elections)Interval95% CI 15.3 to 23.6Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 to 2016SourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. points of variance. The change from 2016 to 2024 is −3.6 (95% CI −7.1 to 1.5)−3.6 percentage points of the variance of the county marginEstimandchange between the two elections in the R-squared of the five-axis regression (the same states are drawn for both elections)Interval95% CI −7.1 to 1.5Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 to 2024SourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. points, an interval that includes zero, so the data do not separate the 2024 share from the 2016 one. The order changed over the same years. Through 2016 the white share had the largest correlation of the five in every election and both weightings in the point estimates. In 2020 no number led clearly, and in 2024 college share led without being separated from rural share (Table 2).

    One check asks whether the rise is an artifact of the census measures. The measures for the elections between 2000 and 2024 are interpolated, so the early margins could fit worse because they were measured worse. Holding the 2024 census measures fixed for every election makes the 2000 share lower, not higher. The rise from 2012 to 2016 holds with the measures of the year, with the 2024 measures and with the 2000 measures. A second comparison separates the states from the counties within them. For a baseline for each state alone, a fixed effectFixed effectsRegression terms that give each group, such as each country or each year, its own baseline. The other coefficients then compare respondents within the same group. that gives every state its own average margin, the change in the share from 2000 to 2024 is −6.7 (95% CI −16.8 to 4.2)−6.7 percentage points of the variance of the county marginEstimandchange between the two elections in the R-squared of a baseline for each state alone (the same states are drawn for both elections)Interval95% CI −16.8 to 4.2Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 to 2024SourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. points, an interval that includes zero. The share of the variance within states that the five numbers reproduce rose from 61%61% share of the within-state variance of the county marginEstimandshare of the within-state variance of the county margin that the five axes reproduceInterval95% CI 50% to 72%n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. to 82%82% share of the within-state variance of the county marginEstimandshare of the within-state variance of the county margin that the five axes reproduceInterval95% CI 77% to 88%n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates.. Counting each county once instead of weighting by votes gives a lower share in every election, 20%20% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 11% to 33%n3,113Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2000 and 59%59% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 49% to 71%n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2024.

    Figure 2

    Vote-weighted, the five axes reproduce 41% of the variance of the county margin in 2000 and 69% in 2024, and 87% in 2024 with state baselines

    Share of the variance of the county two-party margin reproduced by regressions on county characteristics, 2000 to 2024, counties weighted by two-party votes

    Variance of the county margin reproduced (R-squared, %)

    Show data
    Data for Vote-weighted, the five axes reproduce 41% of the variance of the county margin in 2000 and 69% in 2024, and 87% in 2024 with state baselines
    ElectionWeightingCensus measuresRegressionR-squared (%)Interval lowInterval highCounties
    2000Vote-weightedElection-year census measuresFive axes40.829.656.03,113
    2004Vote-weightedElection-year census measuresFive axes43.534.656.73,113
    2008Vote-weightedElection-year census measuresFive axes48.239.659.83,113
    2012Vote-weightedElection-year census measuresFive axes52.344.363.23,112
    2016Vote-weightedElection-year census measuresFive axes72.267.079.33,111
    2020Vote-weightedElection-year census measuresFive axes72.067.179.53,103
    2024Vote-weightedElection-year census measuresFive axes68.663.178.13,103
    2000Vote-weightedElection-year census measuresFive axes and state baselines74.164.882.03,113
    2004Vote-weightedElection-year census measuresFive axes and state baselines74.767.981.93,113
    2008Vote-weightedElection-year census measuresFive axes and state baselines79.974.685.43,113
    2012Vote-weightedElection-year census measuresFive axes and state baselines81.977.186.83,112
    2016Vote-weightedElection-year census measuresFive axes and state baselines88.986.691.73,111
    2020Vote-weightedElection-year census measuresFive axes and state baselines88.385.691.43,103
    2024Vote-weightedElection-year census measuresFive axes and state baselines86.683.690.43,103
    2000Vote-weightedElection-year census measuresState baselines alone34.224.342.73,113
    2004Vote-weightedElection-year census measuresState baselines alone29.521.137.73,113
    2008Vote-weightedElection-year census measuresState baselines alone31.624.138.23,113
    2012Vote-weightedElection-year census measuresState baselines alone31.423.238.63,112
    2016Vote-weightedElection-year census measuresState baselines alone28.918.537.23,111
    2020Vote-weightedElection-year census measuresState baselines alone28.919.236.73,103
    2024Vote-weightedElection-year census measuresState baselines alone27.419.034.43,103
    2000Vote-weightedElection-year census measuresNine measures49.238.662.83,113
    2004Vote-weightedElection-year census measuresNine measures50.141.861.53,113
    2008Vote-weightedElection-year census measuresNine measures52.144.662.63,113
    2012Vote-weightedElection-year census measuresNine measures56.449.366.53,112
    2016Vote-weightedElection-year census measuresNine measures73.869.180.83,111
    2020Vote-weightedElection-year census measuresNine measures73.469.880.53,103
    2024Vote-weightedElection-year census measuresNine measures71.167.279.23,103
    2000County-equalElection-year census measuresFive axes19.611.433.33,113
    2004County-equalElection-year census measuresFive axes19.811.734.33,113
    2008County-equalElection-year census measuresFive axes25.517.439.93,113
    2012County-equalElection-year census measuresFive axes33.123.948.83,112
    2016County-equalElection-year census measuresFive axes57.849.170.23,111
    2020County-equalElection-year census measuresFive axes59.750.671.93,103
    2024County-equalElection-year census measuresFive axes58.749.471.43,103
    2000County-equalElection-year census measuresFive axes and state baselines60.753.966.83,113
    2004County-equalElection-year census measuresFive axes and state baselines63.056.668.43,113
    2008County-equalElection-year census measuresFive axes and state baselines70.965.475.33,113
    2012County-equalElection-year census measuresFive axes and state baselines74.068.978.13,112
    2016County-equalElection-year census measuresFive axes and state baselines81.978.984.83,111
    2020County-equalElection-year census measuresFive axes and state baselines83.480.886.03,103
    2024County-equalElection-year census measuresFive axes and state baselines82.079.284.93,103
    2000County-equalElection-year census measuresState baselines alone33.723.442.63,113
    2004County-equalElection-year census measuresState baselines alone33.122.841.53,113
    2008County-equalElection-year census measuresState baselines alone35.526.243.33,113
    2012County-equalElection-year census measuresState baselines alone35.826.943.63,112
    2016County-equalElection-year census measuresState baselines alone33.024.341.53,111
    2020County-equalElection-year census measuresState baselines alone34.826.742.43,103
    2024County-equalElection-year census measuresState baselines alone34.626.741.63,103
    2000County-equalElection-year census measuresNine measures26.619.638.23,113
    2004County-equalElection-year census measuresNine measures27.721.039.03,113
    2008County-equalElection-year census measuresNine measures30.523.143.13,113
    2012County-equalElection-year census measuresNine measures38.829.852.03,112
    2016County-equalElection-year census measuresNine measures62.254.072.23,111
    2020County-equalElection-year census measuresNine measures65.758.774.13,103
    2024County-equalElection-year census measuresNine measures66.059.074.33,103
    2000Vote-weighted2024 census measures for every electionFive axes35.126.748.13,102
    2004Vote-weighted2024 census measures for every electionFive axes39.632.851.43,103
    2008Vote-weighted2024 census measures for every electionFive axes45.237.956.53,103
    2012Vote-weighted2024 census measures for every electionFive axes49.742.261.03,103
    2016Vote-weighted2024 census measures for every electionFive axes70.565.078.23,104
    2020Vote-weighted2024 census measures for every electionFive axes71.566.679.43,104
    2024Vote-weighted2024 census measures for every electionFive axes68.663.178.13,103
    2000Vote-weighted2024 census measures for every electionFive axes and state baselines68.859.877.13,102
    2004Vote-weighted2024 census measures for every electionFive axes and state baselines70.962.879.33,103
    2008Vote-weighted2024 census measures for every electionFive axes and state baselines77.871.784.23,103
    2012Vote-weighted2024 census measures for every electionFive axes and state baselines80.374.685.93,103
    2016Vote-weighted2024 census measures for every electionFive axes and state baselines88.485.991.23,104
    2020Vote-weighted2024 census measures for every electionFive axes and state baselines88.385.691.43,104
    2024Vote-weighted2024 census measures for every electionFive axes and state baselines86.683.690.43,103
    2000Vote-weighted2024 census measures for every electionState baselines alone33.823.242.93,102
    2004Vote-weighted2024 census measures for every electionState baselines alone29.420.937.93,103
    2008Vote-weighted2024 census measures for every electionState baselines alone31.423.737.63,103
    2012Vote-weighted2024 census measures for every electionState baselines alone31.222.637.63,103
    2016Vote-weighted2024 census measures for every electionState baselines alone28.818.337.23,104
    2020Vote-weighted2024 census measures for every electionState baselines alone28.919.236.73,104
    2024Vote-weighted2024 census measures for every electionState baselines alone27.419.034.43,103
    2000Vote-weighted2024 census measures for every electionNine measures43.534.857.03,102
    2004Vote-weighted2024 census measures for every electionNine measures45.939.557.13,103
    2008Vote-weighted2024 census measures for every electionNine measures49.142.559.63,103
    2012Vote-weighted2024 census measures for every electionNine measures53.947.164.43,103
    2016Vote-weighted2024 census measures for every electionNine measures72.267.479.43,104
    2020Vote-weighted2024 census measures for every electionNine measures73.069.280.53,104
    2024Vote-weighted2024 census measures for every electionNine measures71.167.279.23,103
    2000County-equal2024 census measures for every electionFive axes17.811.429.53,102
    2004County-equal2024 census measures for every electionFive axes19.613.131.93,103
    2008County-equal2024 census measures for every electionFive axes24.316.737.53,103
    2012County-equal2024 census measures for every electionFive axes31.221.747.03,103
    2016County-equal2024 census measures for every electionFive axes55.545.968.33,104
    2020County-equal2024 census measures for every electionFive axes58.749.371.23,104
    2024County-equal2024 census measures for every electionFive axes58.749.471.43,103
    2000County-equal2024 census measures for every electionFive axes and state baselines55.148.361.73,102
    2004County-equal2024 census measures for every electionFive axes and state baselines58.652.164.53,103
    2008County-equal2024 census measures for every electionFive axes and state baselines67.762.572.13,103
    2012County-equal2024 census measures for every electionFive axes and state baselines71.467.075.43,103
    2016County-equal2024 census measures for every electionFive axes and state baselines79.976.683.03,104
    2020County-equal2024 census measures for every electionFive axes and state baselines82.579.785.33,104
    2024County-equal2024 census measures for every electionFive axes and state baselines82.079.284.93,103
    2000County-equal2024 census measures for every electionState baselines alone33.723.542.83,102
    2004County-equal2024 census measures for every electionState baselines alone33.122.541.53,103
    2008County-equal2024 census measures for every electionState baselines alone35.426.243.13,103
    2012County-equal2024 census measures for every electionState baselines alone35.626.843.43,103
    2016County-equal2024 census measures for every electionState baselines alone32.624.240.73,104
    2020County-equal2024 census measures for every electionState baselines alone34.826.642.43,104
    2024County-equal2024 census measures for every electionState baselines alone34.626.741.63,103
    2000County-equal2024 census measures for every electionNine measures26.219.836.63,102
    2004County-equal2024 census measures for every electionNine measures28.121.938.33,103
    2008County-equal2024 census measures for every electionNine measures29.722.341.23,103
    2012County-equal2024 census measures for every electionNine measures37.528.050.53,103
    2016County-equal2024 census measures for every electionNine measures60.652.070.63,104
    2020County-equal2024 census measures for every electionNine measures65.157.873.73,104
    2024County-equal2024 census measures for every electionNine measures66.059.074.33,103
    2000Vote-weighted2000 census measures for every electionFive axes40.829.656.03,113
    2004Vote-weighted2000 census measures for every electionFive axes43.133.356.53,113
    2008Vote-weighted2000 census measures for every electionFive axes47.938.060.23,113
    2012Vote-weighted2000 census measures for every electionFive axes52.843.464.53,113
    2016Vote-weighted2000 census measures for every electionFive axes72.064.979.73,111
    2020Vote-weighted2000 census measures for every electionFive axes71.565.479.03,111
    2024Vote-weighted2000 census measures for every electionFive axes67.661.176.63,110
    2000Vote-weighted2000 census measures for every electionFive axes and state baselines74.164.882.03,113
    2004Vote-weighted2000 census measures for every electionFive axes and state baselines74.567.582.03,113
    2008Vote-weighted2000 census measures for every electionFive axes and state baselines79.273.785.13,113
    2012Vote-weighted2000 census measures for every electionFive axes and state baselines81.276.486.43,113
    2016Vote-weighted2000 census measures for every electionFive axes and state baselines88.085.390.83,111
    2020Vote-weighted2000 census measures for every electionFive axes and state baselines86.483.289.83,111
    2024Vote-weighted2000 census measures for every electionFive axes and state baselines84.080.788.13,110
    2000Vote-weighted2000 census measures for every electionState baselines alone34.224.342.73,113
    2004Vote-weighted2000 census measures for every electionState baselines alone29.521.137.73,113
    2008Vote-weighted2000 census measures for every electionState baselines alone31.624.138.23,113
    2012Vote-weighted2000 census measures for every electionState baselines alone31.323.238.63,113
    2016Vote-weighted2000 census measures for every electionState baselines alone28.918.437.23,111
    2020Vote-weighted2000 census measures for every electionState baselines alone29.019.336.83,111
    2024Vote-weighted2000 census measures for every electionState baselines alone27.619.235.13,110
    2000Vote-weighted2000 census measures for every electionNine measures49.238.662.83,113
    2004Vote-weighted2000 census measures for every electionNine measures49.841.161.33,113
    2008Vote-weighted2000 census measures for every electionNine measures52.243.762.83,113
    2012Vote-weighted2000 census measures for every electionNine measures57.449.167.53,113
    2016Vote-weighted2000 census measures for every electionNine measures74.368.281.03,111
    2020Vote-weighted2000 census measures for every electionNine measures73.668.680.53,111
    2024Vote-weighted2000 census measures for every electionNine measures70.865.878.63,110
    2000County-equal2000 census measures for every electionFive axes19.611.433.33,113
    2004County-equal2000 census measures for every electionFive axes19.510.834.23,113
    2008County-equal2000 census measures for every electionFive axes25.516.740.33,113
    2012County-equal2000 census measures for every electionFive axes33.022.549.63,113
    2016County-equal2000 census measures for every electionFive axes57.447.370.63,111
    2020County-equal2000 census measures for every electionFive axes59.148.772.53,111
    2024County-equal2000 census measures for every electionFive axes58.447.871.83,110
    2000County-equal2000 census measures for every electionFive axes and state baselines60.753.966.83,113
    2004County-equal2000 census measures for every electionFive axes and state baselines63.156.768.63,113
    2008County-equal2000 census measures for every electionFive axes and state baselines71.266.075.43,113
    2012County-equal2000 census measures for every electionFive axes and state baselines74.769.978.53,113
    2016County-equal2000 census measures for every electionFive axes and state baselines82.579.985.13,111
    2020County-equal2000 census measures for every electionFive axes and state baselines83.581.286.13,111
    2024County-equal2000 census measures for every electionFive axes and state baselines82.580.185.23,110
    2000County-equal2000 census measures for every electionState baselines alone33.723.442.63,113
    2004County-equal2000 census measures for every electionState baselines alone33.122.841.53,113
    2008County-equal2000 census measures for every electionState baselines alone35.526.243.33,113
    2012County-equal2000 census measures for every electionState baselines alone35.626.743.43,113
    2016County-equal2000 census measures for every electionState baselines alone32.924.341.53,111
    2020County-equal2000 census measures for every electionState baselines alone35.226.843.23,111
    2024County-equal2000 census measures for every electionState baselines alone35.126.942.43,110
    2000County-equal2000 census measures for every electionNine measures26.619.638.23,113
    2004County-equal2000 census measures for every electionNine measures26.819.738.73,113
    2008County-equal2000 census measures for every electionNine measures29.621.843.13,113
    2012County-equal2000 census measures for every electionNine measures37.727.852.13,113
    2016County-equal2000 census measures for every electionNine measures61.252.172.13,111
    2020County-equal2000 census measures for every electionNine measures64.256.373.93,111
    2024County-equal2000 census measures for every electionNine measures64.356.773.63,110

    The line for the five axes climbs most between 2012 and 2016 and runs level from 2016 to 2024, while the line for state baselines alone stays below it and close to level. Each line is the share of the variance of the county two-party margin reproduced by one regression, vote-weighted, and the bands come from a bootstrap over states. The line that adds the five axes to the state baselines sits above both. The other views count each county once or hold the census measures of one year fixed. Values in Table 5.

    Note: Election returns are full counts of votes, so they carry no sampling error. The 95% intervals come from a bootstrap that resamples whole states, so they show how much a number depends on which states are in the set. They do not include the sampling error of the county characteristics, which are census and American Community Survey estimates interpolated between years.Note: In the last four views the counties that lack the fixed census measures are dropped from every election.Source: County presidential returns, 2000-2024 (Algara and Amlani for 2000 to 2020; MIT Election Data and Science Lab for 2024), full counts of votes; county characteristics from the decennial census and American Community Survey, interpolated to election years; estimates weighted by two-party votes or counting each county once, as stated

    What a fitted share means, and what it does not

    The three steps answer different questions, and which number ranks first depends on the election, on the weighting and on whether the other numbers are held fixed. A correlation says how a county's margin moves with one number across all counties. A coefficient in the joint regression says how it moves with that number among counties that match on the other four. The sign of income is a property of that comparison among counties and not of households. Counties with higher median incomes also have higher college shares, and the coefficient is what remains once college share and the rest are fixed. Gelman and colleagues [1] found that income relates to Republican voting more strongly in poor states than in rich states, and Chapter 3.5, The income ladder reports how individual voters of different incomes voted. The fitted share says how much of the spread of margins the five numbers reproduce as a set, and the part they leave over runs by region. Figure 3 maps the gap between each county's margin in 2024 and the margin the five numbers fit. Zingher [2] documents the education divide after 2000 in surveys and in aggregate alignment, and Companion 8.1a, The diploma map follows the college gradient across counties.

    Figure 3

    In 2024 the mean gap between a county's margin and the five-axis fit is −10.8 points in the Census South and +6.7 points in the rest of the country, vote-weighted

    County two-party margin minus the margin fitted by a regression on the five axes weighted by two-party votes, 2000 and 2024, in percentage points of margin

    Loading county values

    County margin minus the margin fitted by the five axes (points)

    More RepublicanMore Democratic

    Values outside −40 to +40 take the end colors.

    No data (3,143 counties)

    Select a class to show only those places.

    Hover or tap a county for its value, or search by name. Zoom with the buttons on the map, then drag to move it.

    • Alaska draws as no data. Its election returns are reported by election district, and the district codes cannot be matched to boroughs and census areas.
    Show data
    Data for In 2024 the mean gap between a county's margin and the five-axis fit is −10.8 points in the Census South and +6.7 points in the rest of the country, vote-weighted
    ElectionPositionCountyMargin minus fit (points)Margin (points)Fitted margin (points)Two-party votes
    2024Largest gaps below the fitForsyth, GA−52.9−33.5+19.4136,790
    2024Largest gaps below the fitWilliamson, TN−52.1−32.9+19.2142,257
    2024Largest gaps below the fitShelby, AL−47.4−41.3+6.1112,753
    2024Largest gaps below the fitComal, TX−46.8−46.0+0.9102,436
    2024Largest gaps below the fitMiami-Dade, FL−46.8−11.5+35.31,085,945
    2024Largest gaps below the fitLubbock, TX−46.2−39.9+6.2123,695
    2024Largest gaps below the fitMontgomery, TX−45.2−45.9−0.8304,241
    2024Largest gaps below the fitCollin, TX−45.0−11.4+33.5501,649
    2024Largest gaps above the fitClayton, GA+52.2+69.6+17.4111,080
    2024Largest gaps above the fitSonoma, CA+48.2+47.8−0.4243,026
    2024Largest gaps above the fitPrince George's, MD+48.1+77.0+29.0392,046
    2024Largest gaps above the fitSanta Cruz, CA+46.8+56.6+9.8128,976
    2024Largest gaps above the fitBaltimore City, MD+42.7+74.9+32.2223,093
    2024Largest gaps above the fitMultnomah, OR+42.5+64.3+21.8396,686
    2024Largest gaps above the fitMarin, CA+40.1+65.7+25.6140,206
    2024Largest gaps above the fitSt. Louis, MN+36.8+14.0−22.8116,400
    2000Largest gaps below the fitCollin, TX−49.0−49.9−0.9171,063
    2000Largest gaps below the fitMontgomery, TX−40.5−55.2−14.7103,886
    2000Largest gaps below the fitUtah, UT−39.5−71.3−31.8114,700
    2000Largest gaps below the fitGreenville, SC−37.0−35.8+1.2136,524
    2000Largest gaps below the fitSarasota, FL−36.9−6.6+30.4155,986
    2000Largest gaps below the fitFort Bend, TX−36.9−21.5+15.4121,136
    2000Largest gaps below the fitLee, FL−35.9−18.1+17.7179,722
    2000Largest gaps below the fitDenton, TX−35.7−43.6−7.9142,315
    2000Largest gaps above the fitBristol, MA+45.7+36.9−8.8199,173
    2000Largest gaps above the fitBronx, NY+42.9+76.0+33.1302,046
    2000Largest gaps above the fitPhiladelphia, PA+40.6+63.3+22.7550,141
    2000Largest gaps above the fitKings, NY+40.0+67.5+27.5594,122
    2000Largest gaps above the fitSuffolk, MA+38.3+55.4+17.1199,329
    2000Largest gaps above the fitBaltimore City, MD+37.5+70.8+33.3185,915
    2000Largest gaps above the fitDistrict Of Columbia , DC+36.8+81.0+44.2189,996
    2000Largest gaps above the fitProvidence, RI+36.6+39.8+3.1203,847

    The Census South reads red on balance and the rest of the country blue. Vote-weighted, the mean gap between a county's margin and the five-axis fit in 2024 is −10.8−10.8 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression among counties in the Census Southn1,422Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel points in the Census South and +6.7+6.7 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression among counties outside the Census Southn1,681Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel points elsewhere. Red counties voted more Republican than the five numbers fit and blue counties more Democratic. A gap compares votes with the characteristics of residents, so it can reflect candidates, campaigns, turnout and characteristics the five axes omit as well as place. The slider shows the same gaps for 2000. Values in Table 9.

    Note: Election returns are full counts of votes, so the map draws no intervals. The map is drawn on 2020 county boundaries and a county with no fit draws as no data. Alaska is not in the sample, because it reports returns by election district; the Connecticut counties draw in 2000 and not in 2024, because they lack 2024 census measures; Kansas City is added to Jackson County, Missouri in 2024.Note: The data table lists the eight counties at each end of the gap for each year, among counties above the vote cut named in the caption, not the whole map.Source: County presidential returns, 2000 and 2024 (Algara and Amlani; MIT Election Data and Science Lab), full counts of votes; fitted values from a regression weighted by two-party votes on county characteristics from the census and American Community Survey

    What these county statistics cannot tell you

    Every statistic here is a property of counties, and a pattern across places can differ from the pattern among the people who live in them, a limit known as ecological inferenceEcological inferenceDrawing a conclusion about individuals from data on groups or places, such as counties. A pattern across counties can differ from the pattern among the people who live in them.. Robinson [3] showed that a correlation across areas can have the opposite sign from the correlation across individuals. The five numbers are five chosen measures, and a different measure of the same axis, such as per capita income instead of median household income, can change the ranking. The numbers describe residents, including residents who cannot vote, and the margin describes voters. The chapter has no data on turnout by group, on candidates or on campaigns, so it cannot say why the order of the axes changed after 2016 or why the fit improved.

    Go deeper

    • The report: every estimate with its interval and sample size.
    • Tables: every figure of the report as a table.
    • Neighborhoods: the explorer for this part.

    Data sources

    • County election returnsCounty presidential returns, 2000-2024: Algara and Amlani (2000-2020) and MIT Election Data and Science Lab (2024), repaired fileYears: President 1868-2024 (40 elections); House and Senate 1990-2024 · Weights: none (full counts of votes); statistics weighted by two-party votes in the election, or county-equal · License: The Heideman file is CC BY 4.0 (doi:10.5281/zenodo.22976553), but the state records behind it keep their own terms, and the sites of Florida, Indiana, Colorado and Wisconsin restrict reuse.
    • County demographics panelCounty demographics panel, 2000-2024: education, income, race and ethnicity, age and rural share, with the rural-urban continuum code and its editionYears: 2000-2024 (election years, interpolated between census and survey anchors) · Weights: none (census and American Community Survey estimates, interpolated to election years; sampling error not carried) · License: US government source data, public domain. The interpolated panel is derived work.

    All data sources

    References

    1. Gelman, Andrew; Shor, Boris; Bafumi, Joseph; Park, David (2008). Rich State, Poor State, Red State, Blue State: What's the Matter with Connecticut?. Quarterly Journal of Political Science 2(4): 345-367. doi:10.1561/100.00006026
    2. Zingher, Joshua N. (2022). TRENDS: Diploma divide: Educational attainment and the realignment of the American electorate. Political Research Quarterly 75(2): 263-277. doi:10.1177/10659129221079862
    3. Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review 15(3): 351-357. doi:10.2307/2087176

    How to cite

    @misc{singh2026fiveaxesofthecountyvote,
  author = {Singh, Vishal},
  title = {Five axes of the county vote},
  year = {2026},
  url = {https://vishalsingh.org/politics/five-axes-of-the-county-vote},
  note = {In The Public Mind. Published 3 October 2026.}
}