Chapter 12.1
Five axes of the county vote
In 2024 college share, rural share and the white share had vote-weighted correlations with the county two-party margin of +0.66, −0.65 and −0.57, and five county characteristics together reproduce 69% of the variance of the margin, up from 41% in 2000.
In 2024 college share, rural share and the non-Hispanic white share had the three largest correlations with 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. among five county characteristics: +0.66 (95% CI 0.60 to 0.72)+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. for college share, −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. for rural share and −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. for the white share, vote-weighted. The first two are not separated. Their absolute correlations differ by +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 college share has the largest absolute correlation 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 resamples of states, against 38%38% share of resamplesEstimandshare of resamples in which rural 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 for rural share. In 2000 the white share led (−0.52−0.52 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.68 to −0.36n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state., against −0.46−0.46 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural shareInterval95% CI −0.56 to −0.36n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. for rural share and +0.28+0.28 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with college shareInterval95% CI 0.18 to 0.38n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. for college share), in 87%87% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute correlation with the county margin of the five axesn3,113Weightstwo-party votes in the electionPopulation2,000 resamples of statesPeriod2000 presidential electionSourceCounty demographics panel of resamples.
Together the five characteristics reproduce 41% (95% CI 30% to 56%)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 of the county margin in 2000 and 69% (95% CI 63% to 78%)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, vote-weighted, and 20% (95% CI 11% to 33%)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. and 59% (95% CI 49% to 71%)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. when every county counts once. With a baseline margin for each state the vote-weighted shares are 74% (95% CI 65% to 82%)74% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 65% to 82%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. and 87% (95% CI 84% to 90%)87% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 84% to 90%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.. The rise combines a wider fitted part of the margin and a narrower left-over part, by +10.3 (95% CI 7.0 to 13.8)+10.3 percentage points of marginEstimandchange between the two elections in the weighted standard deviation of the margin fitted by the five axes (the same states are drawn for both elections)Interval95% CI 7.0 to 13.8Weightstwo-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. and −2.0 (95% CI −3.6 to −0.7)−2.0 percentage points of marginEstimandchange between the two elections in the weighted standard deviation of the margin left over after the five axes (the same states are drawn for both elections)Interval95% CI −3.6 to −0.7Weightstwo-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 of standard deviation, vote-weighted. Which axis ranks first depends on the election, on the weighting and on whether the other axes are held fixed, and in several elections the data do not separate the leading axes.
This chapter is the common baseline for the geographic chapters that follow. It ranks five axes, education, income, race and ethnicity, age and rurality, as correlates of the county margin, asks how the ranking changes when the axes are held together, and asks how much of the county vote these census characteristics reproduce, a question that the plan for the book once gave its own chapter. Every statement describes an association among counties and not an effect on voters.
Each axis is one measure: the share of adults with a bachelor's degree or more (education), median household income in 2024 dollars (income), the non-Hispanic white share of residents (race and ethnicity), median age (age) and the share of residents living in rural areas as the Census Bureau defines them (rurality). The margin is the Democratic minus the Republican votes over their sum, so a positive correlation means that counties with more of a characteristic voted more Democratic. A vote-weighted statistic counts each county in proportion to its two-party votes in the election, and a county-equal statistic counts each county once. Both are shown throughout, because the largest counties dominate the first and the many small counties the second. Each election uses 3,1033,103 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 to 3,1133,113 countiesEstimandlargest 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 every census measure, and Figure 1 lists which comparisons use the same counties, the same weights and the same other variables.Intervals on correlations and coefficients are 95% intervals with standard 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. [1]. Intervals on R-squared, on spreads, on differences between statistics and on the share of resamples in which an axis leads come from a bootstrap that resamples whole states (2,0002,000 bootstrap replicatesEstimandnumber of bootstrap replicatesWeightsnone (full counts)Populationresamples of statesPeriod2000 to 2024SourceCounty election returns replicates). Returns are full counts, so the intervals say how much a number depends on which states are in the set. The census measures are interpolated between survey years and carry sampling error that no interval includes. Alaska is not in the sample, because it reports returns by election district. Kansas City is added to Jackson County, Missouri in 2024, as it already is in earlier years, so that the unit matches its census measures. The Connecticut counties lack census measures in 2020 and 2024 and are left out, with 1.16%1.16% percent of two-party votesEstimandshare of two-party votes cast in the Connecticut counties left out for a missing census measuren8Weightsnone (full counts)Populationtwo-party votes in counties with returns, Alaska excludedPeriod2020 presidential electionSourceCounty election returns and 1.14%1.14% percent of two-party votesEstimandshare of two-party votes cast in the Connecticut counties left out for a missing census measuren8Weightsnone (full counts)Populationtwo-party votes in counties with returns, Alaska excludedPeriod2024 presidential electionSourceCounty election returns of the two-party votes.
Figure 1
Every comparison in the chapter uses the same outcome and the same counties within an election, and differs in what it holds fixed
The outcome, sample, period, weights and other variables of each comparison; the outcome is the county two-party margin, Democratic minus Republican
| Comparison | Counties | Elections | Weights | Other variables held fixed |
|---|---|---|---|---|
| Correlation of one axis with the margin | Counties with returns and every census measure; the same counties for every comparison in an election | Each presidential election, 2000 to 2024 | Vote-weighted and county-equal | Nothing else |
| Regression of the margin on the five axes | Same counties | Each election, 2000 to 2024 | Vote-weighted and county-equal | The other four axes |
| The same regression with state baselines | Same counties | Each election, 2000 to 2024 | Vote-weighted and county-equal | The other four axes and a baseline margin for each state |
| Share of the variance of the margin reproduced (R-squared) | Same counties | Each election, 2000 to 2024 | Vote-weighted and county-equal | Five axes; five axes with state baselines; state baselines alone; nine measures |
| Checks with fixed census measures | Counties that also have the fixed measures, a few fewer | Each election, 2000 to 2024 | Vote-weighted and county-equal | The 2024 census measures, or the 2000 census measures, used for every election |
| Alternative measures within each axis | The 2024 counties of the main sample | 2024 | Vote-weighted and county-equal | Nothing else |
| Map of the gap between margin and fit | The 2000 and 2024 counties of the main sample | 2000 and 2024 | Fit weighted by two-party votes | The five axes |
Show data
| comparison | counties | elections | held_fixed | weights |
|---|---|---|---|---|
| Correlation of one axis with the margin | Counties with returns and every census measure; the same counties for every comparison in an election | Each presidential election, 2000 to 2024 | Nothing else | Vote-weighted and county-equal |
| Regression of the margin on the five axes | Same counties | Each election, 2000 to 2024 | The other four axes | Vote-weighted and county-equal |
| The same regression with state baselines | Same counties | Each election, 2000 to 2024 | The other four axes and a baseline margin for each state | Vote-weighted and county-equal |
| Share of the variance of the margin reproduced (R-squared) | Same counties | Each election, 2000 to 2024 | Five axes; five axes with state baselines; state baselines alone; nine measures | Vote-weighted and county-equal |
| Checks with fixed census measures | Counties that also have the fixed measures, a few fewer | Each election, 2000 to 2024 | The 2024 census measures, or the 2000 census measures, used for every election | Vote-weighted and county-equal |
| Alternative measures within each axis | The 2024 counties of the main sample | 2024 | Nothing else | Vote-weighted and county-equal |
| Map of the gap between margin and fit | The 2000 and 2024 counties of the main sample | 2000 and 2024 | The five axes | Fit weighted by two-party votes |
Alone, the white share led through 2016 and the leader since 2020 is not settled
Figure 2 plots the correlation of each axis with the margin in each election from 2000 to 2024. Through 2016 the white share has the largest absolute correlation in every election and both weightings in the point estimates, from −0.52−0.52 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.68 to −0.36n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2000 to −0.68−0.68 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.79 to −0.57n3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2016 vote-weighted. It leads in 76%76% share of resamplesEstimandlowest over the elections of 2000 to 2016 of the share of resamples in which the white share has the largest absolute correlationn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2000 to 2016SourceCounty demographics panel to 93%93% share of resamplesEstimandhighest over the elections of 2000 to 2016 of the share of resamples in which the white share has the largest absolute correlationn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2000 to 2016SourceCounty demographics panel of resamples of states vote-weighted (79%79% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute correlation with the county margin of the five axesn3,111Weightstwo-party votes in the electionPopulation2,000 resamples of statesPeriod2016 presidential electionSourceCounty demographics panel in 2016) and in 70%70% share of resamplesEstimandlowest over the elections of 2000 to 2016 of the share of resamples in which the white share has the largest absolute correlationn3,113Weightsnone (each county counts once)Populationresamples of statesPeriod2000 to 2016SourceCounty demographics panel to 98%98% share of resamplesEstimandhighest over the elections of 2000 to 2016 of the share of resamples in which the white share has the largest absolute correlationn3,113Weightsnone (each county counts once)Populationresamples of statesPeriod2000 to 2016SourceCounty demographics panel county-equal (85%85% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute correlation with the county margin of the five axesn3,111Weightsnone (each county counts once)Population2,000 resamples of statesPeriod2016 presidential electionSourceCounty demographics panel in 2016). In 2020 no axis leads clearly. Vote-weighted, rural share (−0.67−0.67 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural shareInterval95% CI −0.76 to −0.58n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state.) leads in 60%60% share of resamplesEstimandshare of resamples in which rural share has the largest absolute correlation with the county margin of the five axesn3,103Weightstwo-party votes in the electionPopulation2,000 resamples of statesPeriod2020 presidential electionSourceCounty demographics panel of resamples, the white share (−0.63−0.63 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.75 to −0.51n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state.) in 23%23% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute correlation with the county margin of the five axesn3,103Weightstwo-party votes in the electionPopulation2,000 resamples of statesPeriod2020 presidential electionSourceCounty demographics panel and college share (+0.64+0.64 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with college shareInterval95% CI 0.58 to 0.70n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state.) in 17%17% 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 statesPeriod2020 presidential electionSourceCounty demographics panel. Counting each county once, college share (+0.53+0.53 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with college shareInterval95% CI 0.45 to 0.61n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state.) and the white share (−0.53−0.53 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic white shareInterval95% CI −0.73 to −0.33n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state.) are level in the point estimates, and the white share leads in 52%52% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute correlation with the county margin of the five axesn3,103Weightsnone (each county counts once)Population2,000 resamples of statesPeriod2020 presidential electionSourceCounty demographics panel of resamples, college share in 34%34% share of resamplesEstimandshare of resamples in which college share has the largest absolute correlation with the county margin of the five axesn3,103Weightsnone (each county counts once)Population2,000 resamples of statesPeriod2020 presidential electionSourceCounty demographics panel and rural share in 14%14% share of resamplesEstimandshare of resamples in which rural share has the largest absolute correlation with the county margin of the five axesn3,103Weightsnone (each county counts once)Population2,000 resamples of statesPeriod2020 presidential electionSourceCounty demographics panel. In 2024 college share has the largest absolute correlation in both weightings, ahead of rural share by +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. vote-weighted and +0.03 (95% CI −0.04 to 0.10)+0.03 difference in absolute correlationEstimandabsolute correlation of the first-ranked axis minus that of the second-ranked axisInterval95% CI −0.04 to 0.10Weightsnone (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. county-equal. Both intervals include zero, and college share leads 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 and 59%59% share of resamplesEstimandshare of resamples in which college share has the largest absolute correlation with the county margin of the five axesn3,103Weightsnone (each county counts once)Population2,000 resamples of statesPeriod2024 presidential electionSourceCounty demographics panel of resamples. The white share, the smallest of the three, exceeds income by +0.18 (95% CI 0.05 to 0.34)+0.18 difference in absolute correlationEstimandabsolute correlation of non-Hispanic white share minus that of median household incomeInterval95% CI 0.05 to 0.34Weightstwo-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. vote-weighted, so income and age are separated from the three largest axes.
Figure 2
Vote-weighted, the correlation of college share with the county margin rose from +0.28 in 2000 to +0.66 in 2024, while the white share's moved from −0.52 to −0.57
Correlation between a county characteristic and the county two-party margin in the same presidential election, 2000 to 2024, counties weighted by two-party votes
Correlation with the county two-party margin (Democratic positive)
Show data
| Election | Axis | Measure | Weighting | Correlation | Interval low | Interval high | Share of resamples with the largest absolute correlation (%) | Counties |
|---|---|---|---|---|---|---|---|---|
| 2000 | Education | College share | Vote-weighted | +0.28 | +0.18 | +0.38 | 0 | 3,113 |
| 2004 | Education | College share | Vote-weighted | +0.38 | +0.28 | +0.47 | 2 | 3,113 |
| 2008 | Education | College share | Vote-weighted | +0.43 | +0.34 | +0.51 | 1 | 3,113 |
| 2012 | Education | College share | Vote-weighted | +0.41 | +0.34 | +0.49 | 0 | 3,112 |
| 2016 | Education | College share | Vote-weighted | +0.58 | +0.51 | +0.64 | 1 | 3,111 |
| 2020 | Education | College share | Vote-weighted | +0.64 | +0.58 | +0.70 | 17 | 3,103 |
| 2024 | Education | College share | Vote-weighted | +0.66 | +0.60 | +0.72 | 60 | 3,103 |
| 2000 | Income | Median household income | Vote-weighted | +0.10 | −0.01 | +0.20 | 0 | 3,113 |
| 2004 | Income | Median household income | Vote-weighted | +0.12 | +0.01 | +0.23 | 0 | 3,113 |
| 2008 | Income | Median household income | Vote-weighted | +0.17 | +0.05 | +0.28 | 0 | 3,113 |
| 2012 | Income | Median household income | Vote-weighted | +0.16 | +0.04 | +0.27 | 0 | 3,112 |
| 2016 | Income | Median household income | Vote-weighted | +0.32 | +0.19 | +0.44 | 0 | 3,111 |
| 2020 | Income | Median household income | Vote-weighted | +0.39 | +0.29 | +0.49 | 0 | 3,103 |
| 2024 | Income | Median household income | Vote-weighted | +0.40 | +0.29 | +0.50 | 0 | 3,103 |
| 2000 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.52 | −0.68 | −0.36 | 87 | 3,113 |
| 2004 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.50 | −0.65 | −0.35 | 77 | 3,113 |
| 2008 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.55 | −0.67 | −0.42 | 76 | 3,113 |
| 2012 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.59 | −0.72 | −0.47 | 93 | 3,112 |
| 2016 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.68 | −0.79 | −0.57 | 79 | 3,111 |
| 2020 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.63 | −0.75 | −0.51 | 23 | 3,103 |
| 2024 | Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.57 | −0.71 | −0.44 | 2 | 3,103 |
| 2000 | Age | Median age | Vote-weighted | −0.08 | −0.19 | +0.04 | 0 | 3,113 |
| 2004 | Age | Median age | Vote-weighted | −0.11 | −0.22 | −0.01 | 0 | 3,113 |
| 2008 | Age | Median age | Vote-weighted | −0.22 | −0.32 | −0.12 | 0 | 3,113 |
| 2012 | Age | Median age | Vote-weighted | −0.24 | −0.35 | −0.13 | 0 | 3,112 |
| 2016 | Age | Median age | Vote-weighted | −0.36 | −0.46 | −0.26 | 0 | 3,111 |
| 2020 | Age | Median age | Vote-weighted | −0.34 | −0.43 | −0.24 | 0 | 3,103 |
| 2024 | Age | Median age | Vote-weighted | −0.31 | −0.41 | −0.21 | 0 | 3,103 |
| 2000 | Rurality | Rural share | Vote-weighted | −0.46 | −0.56 | −0.36 | 13 | 3,113 |
| 2004 | Rurality | Rural share | Vote-weighted | −0.46 | −0.55 | −0.38 | 22 | 3,113 |
| 2008 | Rurality | Rural share | Vote-weighted | −0.52 | −0.60 | −0.43 | 24 | 3,113 |
| 2012 | Rurality | Rural share | Vote-weighted | −0.52 | −0.61 | −0.44 | 7 | 3,112 |
| 2016 | Rurality | Rural share | Vote-weighted | −0.65 | −0.74 | −0.56 | 20 | 3,111 |
| 2020 | Rurality | Rural share | Vote-weighted | −0.67 | −0.76 | −0.58 | 60 | 3,103 |
| 2024 | Rurality | Rural share | Vote-weighted | −0.65 | −0.74 | −0.56 | 38 | 3,103 |
| 2000 | Education | College share | County-equal | +0.10 | +0.01 | +0.20 | 0 | 3,113 |
| 2004 | Education | College share | County-equal | +0.21 | +0.13 | +0.29 | 2 | 3,113 |
| 2008 | Education | College share | County-equal | +0.31 | +0.24 | +0.37 | 12 | 3,113 |
| 2012 | Education | College share | County-equal | +0.30 | +0.23 | +0.38 | 1 | 3,112 |
| 2016 | Education | College share | County-equal | +0.46 | +0.38 | +0.55 | 6 | 3,111 |
| 2020 | Education | College share | County-equal | +0.53 | +0.45 | +0.61 | 34 | 3,103 |
| 2024 | Education | College share | County-equal | +0.56 | +0.47 | +0.64 | 59 | 3,103 |
| 2000 | Income | Median household income | County-equal | +0.02 | −0.10 | +0.14 | 0 | 3,113 |
| 2004 | Income | Median household income | County-equal | +0.03 | −0.08 | +0.14 | 0 | 3,113 |
| 2008 | Income | Median household income | County-equal | +0.12 | +0.01 | +0.22 | 0 | 3,113 |
| 2012 | Income | Median household income | County-equal | +0.08 | −0.05 | +0.21 | 0 | 3,112 |
| 2016 | Income | Median household income | County-equal | +0.19 | +0.05 | +0.33 | 0 | 3,111 |
| 2020 | Income | Median household income | County-equal | +0.28 | +0.14 | +0.41 | 0 | 3,103 |
| 2024 | Income | Median household income | County-equal | +0.30 | +0.17 | +0.43 | 0 | 3,103 |
| 2000 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.40 | −0.54 | −0.26 | 98 | 3,113 |
| 2004 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.36 | −0.52 | −0.20 | 92 | 3,113 |
| 2008 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.36 | −0.51 | −0.21 | 70 | 3,113 |
| 2012 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.44 | −0.61 | −0.28 | 88 | 3,112 |
| 2016 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.57 | −0.73 | −0.41 | 85 | 3,111 |
| 2020 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.53 | −0.73 | −0.33 | 52 | 3,103 |
| 2024 | Race and ethnicity | Non-Hispanic white share | County-equal | −0.50 | −0.71 | −0.28 | 30 | 3,103 |
| 2000 | Age | Median age | County-equal | −0.17 | −0.28 | −0.05 | 0 | 3,113 |
| 2004 | Age | Median age | County-equal | −0.14 | −0.25 | −0.03 | 0 | 3,113 |
| 2008 | Age | Median age | County-equal | −0.16 | −0.26 | −0.07 | 0 | 3,113 |
| 2012 | Age | Median age | County-equal | −0.18 | −0.27 | −0.08 | 0 | 3,112 |
| 2016 | Age | Median age | County-equal | −0.26 | −0.33 | −0.18 | 0 | 3,111 |
| 2020 | Age | Median age | County-equal | −0.23 | −0.30 | −0.15 | 0 | 3,103 |
| 2024 | Age | Median age | County-equal | −0.20 | −0.27 | −0.12 | 0 | 3,103 |
| 2000 | Rurality | Rural share | County-equal | −0.27 | −0.36 | −0.18 | 2 | 3,113 |
| 2004 | Rurality | Rural share | County-equal | −0.26 | −0.34 | −0.18 | 6 | 3,113 |
| 2008 | Rurality | Rural share | County-equal | −0.33 | −0.40 | −0.26 | 19 | 3,113 |
| 2012 | Rurality | Rural share | County-equal | −0.35 | −0.42 | −0.29 | 11 | 3,112 |
| 2016 | Rurality | Rural share | County-equal | −0.49 | −0.56 | −0.42 | 9 | 3,111 |
| 2020 | Rurality | Rural share | County-equal | −0.52 | −0.59 | −0.45 | 14 | 3,103 |
| 2024 | Rurality | Rural share | County-equal | −0.53 | −0.59 | −0.46 | 11 | 3,103 |
The new order comes from college share rising. Vote-weighted, college share's correlation changed by +0.38 (95% CI 0.31 to 0.47)+0.38 change in correlationEstimandchange between the two elections in the correlation of the county margin with college share (the same states are drawn for both elections)Interval95% CI 0.31 to 0.47Weightstwo-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. between 2000 and 2024, rural share's by −0.19 (95% CI −0.27 to −0.11)−0.19 change in correlationEstimandchange between the two elections in the correlation of the county margin with rural share (the same states are drawn for both elections)Interval95% CI −0.27 to −0.11Weightstwo-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. and the white share's by −0.05 (95% CI −0.13 to 0.03)−0.05 change in correlationEstimandchange between the two elections in the correlation of the county margin with non-Hispanic white share (the same states are drawn for both elections)Interval95% CI −0.13 to 0.03Weightstwo-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., an interval that includes zero. Income's correlation went from +0.10+0.10 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median household incomeInterval95% CI −0.01 to 0.20n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. to +0.40+0.40 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median household incomeInterval95% CI 0.29 to 0.50n3,103Weightstwo-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 median age's from −0.08−0.08 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median ageInterval95% CI −0.19 to 0.04n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. to −0.31−0.31 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median ageInterval95% CI −0.41 to −0.21n3,103Weightstwo-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.. Zingher [2] documents the rise of the education divide after 2000 in surveys and in aggregate alignment, and Chapter 12.2, The diploma map follows the education gradient across counties in detail. Chapter 12.3, Income and the county vote (planned) takes up income and Chapter 12.5, Rural counties (planned) takes up rurality. Scala and Johnson [3] found that Democratic candidates did well in densely populated counties and poorly on the rural side of the urban-rural interface in 2000 to 2016, which matches the sign of rural share's correlation here.
Held together, college share's coefficient is the largest in 2024 and income's changes sign
The five axes overlap across counties, so a correlation mixes an axis with the others. Figure 3 compares each correlation with the coefficient of the same axis in one regression on all five, in standard deviations of the margin per standard deviation of the axis. 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. for each state gives every state its own baseline margin, so a regression with state baselines compares counties within the same state.
Vote-weighted in 2024, college share goes from +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. alone to +0.71 (95% CI 0.63 to 0.79)+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. with the other four axes held fixed, and the white share from −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. to −0.43 (95% CI −0.57 to −0.29)−0.43 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axesInterval95% CI −0.57 to −0.29n3,103Weightstwo-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.. 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 vote-weighted and 77%77% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,103Weightsnone (each county counts once)Populationresamples of statesPeriod2024 presidential electionSourceCounty demographics panel county-equal. Rural share goes from −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. 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, from +0.40+0.40 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median household incomeInterval95% CI 0.29 to 0.50n3,103Weightstwo-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. alone to −0.26 (95% CI −0.40 to −0.11)−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.: among counties with the same college share, white share, median age and rural share, those with higher median household income voted more Republican. Median age goes from −0.31−0.31 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median ageInterval95% CI −0.41 to −0.21n3,103Weightstwo-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. to +0.04 (95% CI −0.13 to 0.21)+0.04 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on median age in a regression of the county margin on the five axesInterval95% CI −0.13 to 0.21n3,103Weightstwo-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.. In points of margin, one vote-weighted standard deviation of college share (11.811.8 percentage points of adultsEstimandvote-weighted standard deviation of college share across countiesn3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel points of college share) goes with +23.4 (95% CI 20.7 to 26.1)+23.4 percentage points of margin per standard deviation of the axisEstimandcoefficient on college share in a regression of the county margin on the five axes, in points of marginInterval95% CI 20.7 to 26.1n3,103Weightstwo-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. points of margin in the joint model, against +6.7 (95% CI 3.0 to 10.5)+6.7 percentage points of margin per standard deviation of the axisEstimandcoefficient on college share in a regression of the county margin on the five axes, in points of marginInterval95% CI 3.0 to 10.5n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2000, and one standard deviation of the white share (21.121.1 percentage points of residentsEstimandvote-weighted standard deviation of non-Hispanic white share across countiesn3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel points of white share) goes with −14.1 (95% CI −18.7 to −9.6)−14.1 percentage points of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes, in points of marginInterval95% CI −18.7 to −9.6n3,103Weightstwo-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. points of margin, against −13.4 (95% CI −17.4 to −9.4)−13.4 percentage points of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes, in points of marginInterval95% CI −17.4 to −9.4n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2000.
Figure 3
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
| Election | Weighting | Axis | Model | Coefficient | Interval low | Interval high | Counties |
|---|---|---|---|---|---|---|---|
| 2024 | Vote-weighted | Education | Each axis alone | +0.66 | +0.60 | +0.72 | 3,103 |
| 2000 | Vote-weighted | Education | Each axis alone | +0.28 | +0.18 | +0.38 | 3,113 |
| 2024 | Vote-weighted | Income | Each axis alone | +0.40 | +0.29 | +0.50 | 3,103 |
| 2000 | Vote-weighted | Income | Each axis alone | +0.10 | −0.01 | +0.20 | 3,113 |
| 2024 | Vote-weighted | Race and ethnicity | Each axis alone | −0.57 | −0.71 | −0.44 | 3,103 |
| 2000 | Vote-weighted | Race and ethnicity | Each axis alone | −0.52 | −0.68 | −0.36 | 3,113 |
| 2024 | Vote-weighted | Age | Each axis alone | −0.31 | −0.41 | −0.21 | 3,103 |
| 2000 | Vote-weighted | Age | Each axis alone | −0.08 | −0.19 | +0.04 | 3,113 |
| 2024 | Vote-weighted | Rurality | Each axis alone | −0.65 | −0.74 | −0.56 | 3,103 |
| 2000 | Vote-weighted | Rurality | Each axis alone | −0.46 | −0.56 | −0.36 | 3,113 |
| 2024 | County-equal | Education | Each axis alone | +0.56 | +0.47 | +0.64 | 3,103 |
| 2000 | County-equal | Education | Each axis alone | +0.10 | +0.01 | +0.20 | 3,113 |
| 2024 | County-equal | Income | Each axis alone | +0.30 | +0.17 | +0.43 | 3,103 |
| 2000 | County-equal | Income | Each axis alone | +0.02 | −0.10 | +0.14 | 3,113 |
| 2024 | County-equal | Race and ethnicity | Each axis alone | −0.50 | −0.71 | −0.28 | 3,103 |
| 2000 | County-equal | Race and ethnicity | Each axis alone | −0.40 | −0.54 | −0.26 | 3,113 |
| 2024 | County-equal | Age | Each axis alone | −0.20 | −0.27 | −0.12 | 3,103 |
| 2000 | County-equal | Age | Each axis alone | −0.17 | −0.28 | −0.05 | 3,113 |
| 2024 | County-equal | Rurality | Each axis alone | −0.53 | −0.59 | −0.46 | 3,103 |
| 2000 | County-equal | Rurality | Each axis alone | −0.27 | −0.36 | −0.18 | 3,113 |
| 2024 | Vote-weighted | Education | All five together | +0.71 | +0.63 | +0.80 | 3,103 |
| 2000 | Vote-weighted | Education | All five together | +0.25 | +0.11 | +0.39 | 3,113 |
| 2024 | Vote-weighted | Income | All five together | −0.26 | −0.40 | −0.11 | 3,103 |
| 2000 | Vote-weighted | Income | All five together | −0.17 | −0.28 | −0.06 | 3,113 |
| 2024 | Vote-weighted | Race and ethnicity | All five together | −0.43 | −0.57 | −0.29 | 3,103 |
| 2000 | Vote-weighted | Race and ethnicity | All five together | −0.50 | −0.66 | −0.35 | 3,113 |
| 2024 | Vote-weighted | Age | All five together | +0.04 | −0.13 | +0.21 | 3,103 |
| 2000 | Vote-weighted | Age | All five together | +0.29 | +0.10 | +0.49 | 3,113 |
| 2024 | Vote-weighted | Rurality | All five together | −0.12 | −0.25 | +0.01 | 3,103 |
| 2000 | Vote-weighted | Rurality | All five together | −0.25 | −0.38 | −0.11 | 3,113 |
| 2024 | County-equal | Education | All five together | +0.56 | +0.48 | +0.64 | 3,103 |
| 2000 | County-equal | Education | All five together | +0.03 | −0.09 | +0.14 | 3,113 |
| 2024 | County-equal | Income | All five together | −0.14 | −0.23 | −0.06 | 3,103 |
| 2000 | County-equal | Income | All five together | −0.01 | −0.12 | +0.10 | 3,113 |
| 2024 | County-equal | Race and ethnicity | All five together | −0.47 | −0.67 | −0.28 | 3,103 |
| 2000 | County-equal | Race and ethnicity | All five together | −0.39 | −0.55 | −0.22 | 3,113 |
| 2024 | County-equal | Age | All five together | +0.13 | +0.07 | +0.20 | 3,103 |
| 2000 | County-equal | Age | All five together | +0.09 | −0.02 | +0.20 | 3,113 |
| 2024 | County-equal | Rurality | All five together | −0.20 | −0.30 | −0.10 | 3,103 |
| 2000 | County-equal | Rurality | All five together | −0.20 | −0.28 | −0.12 | 3,113 |
| 2024 | Vote-weighted | Education | All five with state baselines | +0.73 | +0.65 | +0.80 | 3,103 |
| 2000 | Vote-weighted | Education | All five with state baselines | +0.26 | +0.14 | +0.38 | 3,113 |
| 2024 | Vote-weighted | Income | All five with state baselines | −0.41 | −0.48 | −0.33 | 3,103 |
| 2000 | Vote-weighted | Income | All five with state baselines | −0.25 | −0.32 | −0.17 | 3,113 |
| 2024 | Vote-weighted | Race and ethnicity | All five with state baselines | −0.62 | −0.76 | −0.48 | 3,103 |
| 2000 | Vote-weighted | Race and ethnicity | All five with state baselines | −0.77 | −0.90 | −0.64 | 3,113 |
| 2024 | Vote-weighted | Age | All five with state baselines | +0.08 | +0.02 | +0.14 | 3,103 |
| 2000 | Vote-weighted | Age | All five with state baselines | +0.22 | +0.11 | +0.33 | 3,113 |
| 2024 | Vote-weighted | Rurality | All five with state baselines | −0.09 | −0.14 | −0.04 | 3,103 |
| 2000 | Vote-weighted | Rurality | All five with state baselines | −0.10 | −0.19 | −0.01 | 3,113 |
| 2024 | County-equal | Education | All five with state baselines | +0.54 | +0.47 | +0.60 | 3,103 |
| 2000 | County-equal | Education | All five with state baselines | +0.15 | +0.06 | +0.24 | 3,113 |
| 2024 | County-equal | Income | All five with state baselines | −0.18 | −0.23 | −0.13 | 3,103 |
| 2000 | County-equal | Income | All five with state baselines | −0.15 | −0.23 | −0.08 | 3,113 |
| 2024 | County-equal | Race and ethnicity | All five with state baselines | −0.73 | −0.84 | −0.61 | 3,103 |
| 2000 | County-equal | Race and ethnicity | All five with state baselines | −0.68 | −0.76 | −0.60 | 3,113 |
| 2024 | County-equal | Age | All five with state baselines | +0.10 | +0.06 | +0.15 | 3,103 |
| 2000 | County-equal | Age | All five with state baselines | +0.12 | +0.06 | +0.17 | 3,113 |
| 2024 | County-equal | Rurality | All five with state baselines | −0.11 | −0.15 | −0.06 | 3,103 |
| 2000 | County-equal | Rurality | All five with state baselines | −0.11 | −0.17 | −0.05 | 3,113 |
The sign of income describes a regression among counties and not households. Counties with higher median incomes also have higher college shares, and the coefficient on income is what remains once college share, the white share, age and rurality are fixed. Gelman and colleagues [4] found that income relates to Republican voting more strongly in poor states than in rich states, and Chapter 3.3, The income ladder reports the income profile of individual voters.
Adding a baseline for each state keeps the order of the point estimates in 2024 and changes some sizes. Vote-weighted, college share is +0.73 (95% CI 0.65 to 0.80)+0.73 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 axes and a baseline for each stateInterval95% CI 0.65 to 0.80n3,103Weightstwo-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., the white share −0.62 (95% CI −0.76 to −0.48)−0.62 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes and a baseline for each stateInterval95% CI −0.76 to −0.48n3,103Weightstwo-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., income −0.40 (95% CI −0.48 to −0.33)−0.40 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 axes and a baseline for each stateInterval95% CI −0.48 to −0.33n3,103Weightstwo-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., rural share −0.09 (95% CI −0.14 to −0.04)−0.09 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 axes and a baseline for each stateInterval95% CI −0.14 to −0.04n3,103Weightstwo-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 median age +0.08 (95% CI 0.02 to 0.13)+0.08 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on median age in a regression of the county margin on the five axes and a baseline for each stateInterval95% CI 0.02 to 0.13n3,103Weightstwo-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.. In 2000 the white share's coefficient is −0.77−0.77 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes and a baseline for each stateInterval95% CI −0.90 to −0.64n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. with state baselines against −0.50−0.50 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axesInterval95% CI −0.65 to −0.35n3,113Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. without them, and it is the largest in 100%100% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regression with state baselinesn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2000 presidential electionSourceCounty demographics panel of resamples with baselines. In 2024 college share has the largest coefficient with baselines in 90%90% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regression with state baselinesn3,103Weightstwo-party votes in the electionPopulationresamples of statesPeriod2024 presidential electionSourceCounty demographics panel of resamples vote-weighted. Counting each county once, with state baselines, the white share's coefficient is −0.68−0.68 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes and a baseline for each stateInterval95% CI −0.76 to −0.60n3,113Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2000 and −0.73−0.73 standardized regression coefficient, standard deviations of margin per standard deviation of the axisEstimandcoefficient on non-Hispanic white share in a regression of the county margin on the five axes and a baseline for each stateInterval95% CI −0.84 to −0.61n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2024, the largest in 98%98% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regression with state baselinesn3,103Weightsnone (each county counts once)Populationresamples of statesPeriod2024 presidential electionSourceCounty demographics panel of resamples in 2024, and college share's is +0.54+0.54 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 axes and a baseline for each stateInterval95% CI 0.48 to 0.60n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state..
Which coefficient is largest is often unsettled (Figure 4). Vote-weighted, college share has the largest coefficient in 65%65% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,111Weightstwo-party votes in the electionPopulationresamples of statesPeriod2016 presidential electionSourceCounty demographics panel of resamples in 2016 and 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 in 2024, but in 51%51% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2004 presidential electionSourceCounty demographics panel in 2004 and 55%55% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2008 presidential electionSourceCounty demographics panel in 2008, when the white share has 49%49% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regressionn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2004 presidential electionSourceCounty demographics panel and 45%45% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regressionn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2008 presidential electionSourceCounty demographics panel. The white share leads in 97%97% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regressionn3,113Weightstwo-party votes in the electionPopulationresamples of statesPeriod2000 presidential electionSourceCounty demographics panel of resamples in 2000 and 73%73% share of resamplesEstimandshare of resamples in which non-Hispanic white share has the largest absolute coefficient in the five-axis regressionn3,112Weightstwo-party votes in the electionPopulationresamples of statesPeriod2012 presidential electionSourceCounty demographics panel in 2012. Counting each county once, the white share has the largest coefficient in the point estimates through 2016 and college share from 2020, with college share first in 56%56% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,103Weightsnone (each county counts once)Populationresamples of statesPeriod2020 presidential electionSourceCounty demographics panel of resamples in 2020 and 77%77% share of resamplesEstimandshare of resamples in which college share has the largest absolute coefficient in the five-axis regressionn3,103Weightsnone (each county counts once)Populationresamples of statesPeriod2024 presidential electionSourceCounty demographics panel in 2024. College share's county-equal coefficient rose from +0.03+0.03 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 −0.09 to 0.14n3,113Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2000 to +0.56+0.56 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 0.48 to 0.64n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2024, and income's is −0.14−0.14 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 −0.23 to −0.06n3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panelNoteInterval: 95% normal interval with errors clustered by state. in 2024.
Figure 4
Held together, college share's coefficient rose from +0.25 in 2000 to +0.71 in 2024 and income's was below zero in every election, vote-weighted
Standardized coefficients from a regression of the county two-party margin on all five axes, 2000 to 2024, counties weighted by two-party votes
Standardized coefficient with the other four axes held fixed
Show data
| Election | Weighting | Model | Axis | Coefficient | Interval low | Interval high | Share of resamples with the largest absolute coefficient (%) | Counties |
|---|---|---|---|---|---|---|---|---|
| 2000 | Vote-weighted | All five together | Education | +0.25 | +0.11 | +0.39 | 1 | 3,113 |
| 2004 | Vote-weighted | All five together | Education | +0.49 | +0.35 | +0.62 | 51 | 3,113 |
| 2008 | Vote-weighted | All five together | Education | +0.51 | +0.40 | +0.62 | 55 | 3,113 |
| 2012 | Vote-weighted | All five together | Education | +0.51 | +0.40 | +0.63 | 27 | 3,112 |
| 2016 | Vote-weighted | All five together | Education | +0.62 | +0.54 | +0.71 | 65 | 3,111 |
| 2020 | Vote-weighted | All five together | Education | +0.66 | +0.58 | +0.74 | 99 | 3,103 |
| 2024 | Vote-weighted | All five together | Education | +0.71 | +0.63 | +0.80 | 100 | 3,103 |
| 2000 | Vote-weighted | All five together | Income | −0.17 | −0.28 | −0.06 | 0 | 3,113 |
| 2004 | Vote-weighted | All five together | Income | −0.31 | −0.42 | −0.20 | 0 | 3,113 |
| 2008 | Vote-weighted | All five together | Income | −0.29 | −0.41 | −0.17 | 0 | 3,113 |
| 2012 | Vote-weighted | All five together | Income | −0.30 | −0.43 | −0.17 | 0 | 3,112 |
| 2016 | Vote-weighted | All five together | Income | −0.25 | −0.37 | −0.13 | 0 | 3,111 |
| 2020 | Vote-weighted | All five together | Income | −0.23 | −0.36 | −0.10 | 0 | 3,103 |
| 2024 | Vote-weighted | All five together | Income | −0.26 | −0.40 | −0.11 | 0 | 3,103 |
| 2000 | Vote-weighted | All five together | Race and ethnicity | −0.50 | −0.66 | −0.35 | 97 | 3,113 |
| 2004 | Vote-weighted | All five together | Race and ethnicity | −0.47 | −0.61 | −0.33 | 49 | 3,113 |
| 2008 | Vote-weighted | All five together | Race and ethnicity | −0.49 | −0.62 | −0.36 | 45 | 3,113 |
| 2012 | Vote-weighted | All five together | Race and ethnicity | −0.56 | −0.69 | −0.42 | 73 | 3,112 |
| 2016 | Vote-weighted | All five together | Race and ethnicity | −0.59 | −0.70 | −0.48 | 35 | 3,111 |
| 2020 | Vote-weighted | All five together | Race and ethnicity | −0.51 | −0.63 | −0.39 | 1 | 3,103 |
| 2024 | Vote-weighted | All five together | Race and ethnicity | −0.43 | −0.57 | −0.29 | 0 | 3,103 |
| 2000 | Vote-weighted | All five together | Age | +0.29 | +0.10 | +0.49 | 1 | 3,113 |
| 2004 | Vote-weighted | All five together | Age | +0.28 | +0.09 | +0.47 | 0 | 3,113 |
| 2008 | Vote-weighted | All five together | Age | +0.20 | +0.02 | +0.38 | 0 | 3,113 |
| 2012 | Vote-weighted | All five together | Age | +0.20 | +0.01 | +0.40 | 0 | 3,112 |
| 2016 | Vote-weighted | All five together | Age | +0.10 | −0.04 | +0.24 | 0 | 3,111 |
| 2020 | Vote-weighted | All five together | Age | +0.07 | −0.09 | +0.22 | 0 | 3,103 |
| 2024 | Vote-weighted | All five together | Age | +0.04 | −0.13 | +0.21 | 0 | 3,103 |
| 2000 | Vote-weighted | All five together | Rurality | −0.25 | −0.38 | −0.11 | 2 | 3,113 |
| 2004 | Vote-weighted | All five together | Rurality | −0.18 | −0.33 | −0.03 | 0 | 3,113 |
| 2008 | Vote-weighted | All five together | Rurality | −0.18 | −0.31 | −0.04 | 0 | 3,113 |
| 2012 | Vote-weighted | All five together | Rurality | −0.14 | −0.29 | 0.00 | 0 | 3,112 |
| 2016 | Vote-weighted | All five together | Rurality | −0.11 | −0.23 | 0.00 | 0 | 3,111 |
| 2020 | Vote-weighted | All five together | Rurality | −0.13 | −0.24 | −0.01 | 0 | 3,103 |
| 2024 | Vote-weighted | All five together | Rurality | −0.12 | −0.25 | +0.01 | 0 | 3,103 |
| 2000 | County-equal | All five together | Education | +0.03 | −0.09 | +0.14 | 0 | 3,113 |
| 2004 | County-equal | All five together | Education | +0.26 | +0.14 | +0.38 | 19 | 3,113 |
| 2008 | County-equal | All five together | Education | +0.34 | +0.23 | +0.46 | 41 | 3,113 |
| 2012 | County-equal | All five together | Education | +0.36 | +0.24 | +0.48 | 18 | 3,112 |
| 2016 | County-equal | All five together | Education | +0.51 | +0.41 | +0.60 | 23 | 3,111 |
| 2020 | County-equal | All five together | Education | +0.55 | +0.46 | +0.64 | 56 | 3,103 |
| 2024 | County-equal | All five together | Education | +0.56 | +0.48 | +0.64 | 77 | 3,103 |
| 2000 | County-equal | All five together | Income | −0.01 | −0.12 | +0.10 | 0 | 3,113 |
| 2004 | County-equal | All five together | Income | −0.15 | −0.25 | −0.05 | 0 | 3,113 |
| 2008 | County-equal | All five together | Income | −0.13 | −0.25 | −0.02 | 0 | 3,113 |
| 2012 | County-equal | All five together | Income | −0.17 | −0.31 | −0.03 | 0 | 3,112 |
| 2016 | County-equal | All five together | Income | −0.17 | −0.28 | −0.05 | 0 | 3,111 |
| 2020 | County-equal | All five together | Income | −0.14 | −0.24 | −0.04 | 0 | 3,103 |
| 2024 | County-equal | All five together | Income | −0.14 | −0.23 | −0.06 | 0 | 3,103 |
| 2000 | County-equal | All five together | Race and ethnicity | −0.39 | −0.55 | −0.22 | 99 | 3,113 |
| 2004 | County-equal | All five together | Race and ethnicity | −0.35 | −0.53 | −0.17 | 80 | 3,113 |
| 2008 | County-equal | All five together | Race and ethnicity | −0.36 | −0.53 | −0.19 | 59 | 3,113 |
| 2012 | County-equal | All five together | Race and ethnicity | −0.45 | −0.63 | −0.27 | 81 | 3,112 |
| 2016 | County-equal | All five together | Race and ethnicity | −0.57 | −0.72 | −0.42 | 77 | 3,111 |
| 2020 | County-equal | All five together | Race and ethnicity | −0.52 | −0.70 | −0.35 | 45 | 3,103 |
| 2024 | County-equal | All five together | Race and ethnicity | −0.47 | −0.67 | −0.28 | 23 | 3,103 |
| 2000 | County-equal | All five together | Age | +0.09 | −0.02 | +0.20 | 0 | 3,113 |
| 2004 | County-equal | All five together | Age | +0.12 | +0.01 | +0.23 | 0 | 3,113 |
| 2008 | County-equal | All five together | Age | +0.14 | +0.04 | +0.23 | 0 | 3,113 |
| 2012 | County-equal | All five together | Age | +0.17 | +0.08 | +0.26 | 0 | 3,112 |
| 2016 | County-equal | All five together | Age | +0.16 | +0.09 | +0.22 | 0 | 3,111 |
| 2020 | County-equal | All five together | Age | +0.14 | +0.08 | +0.21 | 0 | 3,103 |
| 2024 | County-equal | All five together | Age | +0.13 | +0.07 | +0.20 | 0 | 3,103 |
| 2000 | County-equal | All five together | Rurality | −0.20 | −0.28 | −0.12 | 1 | 3,113 |
| 2004 | County-equal | All five together | Rurality | −0.15 | −0.24 | −0.06 | 1 | 3,113 |
| 2008 | County-equal | All five together | Rurality | −0.17 | −0.27 | −0.08 | 0 | 3,113 |
| 2012 | County-equal | All five together | Rurality | −0.19 | −0.30 | −0.09 | 0 | 3,112 |
| 2016 | County-equal | All five together | Rurality | −0.19 | −0.29 | −0.10 | 0 | 3,111 |
| 2020 | County-equal | All five together | Rurality | −0.20 | −0.30 | −0.10 | 0 | 3,103 |
| 2024 | County-equal | All five together | Rurality | −0.20 | −0.30 | −0.10 | 0 | 3,103 |
| 2000 | Vote-weighted | All five with state baselines | Education | +0.26 | +0.14 | +0.38 | 0 | 3,113 |
| 2004 | Vote-weighted | All five with state baselines | Education | +0.47 | +0.33 | +0.62 | 0 | 3,113 |
| 2008 | Vote-weighted | All five with state baselines | Education | +0.49 | +0.36 | +0.62 | 0 | 3,113 |
| 2012 | Vote-weighted | All five with state baselines | Education | +0.47 | +0.34 | +0.60 | 0 | 3,112 |
| 2016 | Vote-weighted | All five with state baselines | Education | +0.62 | +0.53 | +0.70 | 0 | 3,111 |
| 2020 | Vote-weighted | All five with state baselines | Education | +0.67 | +0.60 | +0.74 | 40 | 3,103 |
| 2024 | Vote-weighted | All five with state baselines | Education | +0.73 | +0.65 | +0.80 | 90 | 3,103 |
| 2000 | Vote-weighted | All five with state baselines | Income | −0.25 | −0.32 | −0.17 | 0 | 3,113 |
| 2004 | Vote-weighted | All five with state baselines | Income | −0.37 | −0.42 | −0.31 | 0 | 3,113 |
| 2008 | Vote-weighted | All five with state baselines | Income | −0.38 | −0.44 | −0.32 | 0 | 3,113 |
| 2012 | Vote-weighted | All five with state baselines | Income | −0.39 | −0.45 | −0.33 | 0 | 3,112 |
| 2016 | Vote-weighted | All five with state baselines | Income | −0.35 | −0.40 | −0.29 | 0 | 3,111 |
| 2020 | Vote-weighted | All five with state baselines | Income | −0.37 | −0.43 | −0.30 | 0 | 3,103 |
| 2024 | Vote-weighted | All five with state baselines | Income | −0.41 | −0.48 | −0.33 | 0 | 3,103 |
| 2000 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.77 | −0.90 | −0.64 | 100 | 3,113 |
| 2004 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.77 | −0.89 | −0.66 | 100 | 3,113 |
| 2008 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.76 | −0.88 | −0.65 | 100 | 3,113 |
| 2012 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.81 | −0.91 | −0.70 | 100 | 3,112 |
| 2016 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.79 | −0.88 | −0.71 | 100 | 3,111 |
| 2020 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.69 | −0.81 | −0.58 | 60 | 3,103 |
| 2024 | Vote-weighted | All five with state baselines | Race and ethnicity | −0.62 | −0.76 | −0.48 | 10 | 3,103 |
| 2000 | Vote-weighted | All five with state baselines | Age | +0.22 | +0.11 | +0.33 | 0 | 3,113 |
| 2004 | Vote-weighted | All five with state baselines | Age | +0.23 | +0.11 | +0.35 | 0 | 3,113 |
| 2008 | Vote-weighted | All five with state baselines | Age | +0.15 | +0.05 | +0.25 | 0 | 3,113 |
| 2012 | Vote-weighted | All five with state baselines | Age | +0.14 | +0.05 | +0.23 | 0 | 3,112 |
| 2016 | Vote-weighted | All five with state baselines | Age | +0.10 | +0.04 | +0.16 | 0 | 3,111 |
| 2020 | Vote-weighted | All five with state baselines | Age | +0.08 | +0.03 | +0.14 | 0 | 3,103 |
| 2024 | Vote-weighted | All five with state baselines | Age | +0.08 | +0.02 | +0.14 | 0 | 3,103 |
| 2000 | Vote-weighted | All five with state baselines | Rurality | −0.10 | −0.19 | −0.01 | 0 | 3,113 |
| 2004 | Vote-weighted | All five with state baselines | Rurality | −0.04 | −0.12 | +0.04 | 0 | 3,113 |
| 2008 | Vote-weighted | All five with state baselines | Rurality | −0.05 | −0.12 | +0.02 | 0 | 3,113 |
| 2012 | Vote-weighted | All five with state baselines | Rurality | −0.04 | −0.10 | +0.03 | 0 | 3,112 |
| 2016 | Vote-weighted | All five with state baselines | Rurality | −0.04 | −0.09 | +0.01 | 0 | 3,111 |
| 2020 | Vote-weighted | All five with state baselines | Rurality | −0.09 | −0.14 | −0.03 | 0 | 3,103 |
| 2024 | Vote-weighted | All five with state baselines | Rurality | −0.09 | −0.14 | −0.04 | 0 | 3,103 |
| 2000 | County-equal | All five with state baselines | Education | +0.15 | +0.06 | +0.24 | 0 | 3,113 |
| 2004 | County-equal | All five with state baselines | Education | +0.33 | +0.23 | +0.43 | 0 | 3,113 |
| 2008 | County-equal | All five with state baselines | Education | +0.35 | +0.26 | +0.44 | 0 | 3,113 |
| 2012 | County-equal | All five with state baselines | Education | +0.34 | +0.26 | +0.42 | 0 | 3,112 |
| 2016 | County-equal | All five with state baselines | Education | +0.49 | +0.43 | +0.56 | 0 | 3,111 |
| 2020 | County-equal | All five with state baselines | Education | +0.53 | +0.47 | +0.58 | 0 | 3,103 |
| 2024 | County-equal | All five with state baselines | Education | +0.54 | +0.47 | +0.60 | 2 | 3,103 |
| 2000 | County-equal | All five with state baselines | Income | −0.15 | −0.23 | −0.08 | 0 | 3,113 |
| 2004 | County-equal | All five with state baselines | Income | −0.24 | −0.31 | −0.18 | 0 | 3,113 |
| 2008 | County-equal | All five with state baselines | Income | −0.23 | −0.28 | −0.17 | 0 | 3,113 |
| 2012 | County-equal | All five with state baselines | Income | −0.23 | −0.28 | −0.17 | 0 | 3,112 |
| 2016 | County-equal | All five with state baselines | Income | −0.20 | −0.24 | −0.15 | 0 | 3,111 |
| 2020 | County-equal | All five with state baselines | Income | −0.18 | −0.22 | −0.14 | 0 | 3,103 |
| 2024 | County-equal | All five with state baselines | Income | −0.18 | −0.23 | −0.13 | 0 | 3,103 |
| 2000 | County-equal | All five with state baselines | Race and ethnicity | −0.68 | −0.76 | −0.60 | 100 | 3,113 |
| 2004 | County-equal | All five with state baselines | Race and ethnicity | −0.70 | −0.79 | −0.61 | 100 | 3,113 |
| 2008 | County-equal | All five with state baselines | Race and ethnicity | −0.75 | −0.85 | −0.66 | 100 | 3,113 |
| 2012 | County-equal | All five with state baselines | Race and ethnicity | −0.80 | −0.89 | −0.71 | 100 | 3,112 |
| 2016 | County-equal | All five with state baselines | Race and ethnicity | −0.83 | −0.91 | −0.75 | 100 | 3,111 |
| 2020 | County-equal | All five with state baselines | Race and ethnicity | −0.77 | −0.86 | −0.68 | 100 | 3,103 |
| 2024 | County-equal | All five with state baselines | Race and ethnicity | −0.73 | −0.84 | −0.61 | 98 | 3,103 |
| 2000 | County-equal | All five with state baselines | Age | +0.12 | +0.06 | +0.17 | 0 | 3,113 |
| 2004 | County-equal | All five with state baselines | Age | +0.13 | +0.07 | +0.20 | 0 | 3,113 |
| 2008 | County-equal | All five with state baselines | Age | +0.12 | +0.06 | +0.17 | 0 | 3,113 |
| 2012 | County-equal | All five with state baselines | Age | +0.13 | +0.08 | +0.18 | 0 | 3,112 |
| 2016 | County-equal | All five with state baselines | Age | +0.11 | +0.07 | +0.16 | 0 | 3,111 |
| 2020 | County-equal | All five with state baselines | Age | +0.10 | +0.06 | +0.15 | 0 | 3,103 |
| 2024 | County-equal | All five with state baselines | Age | +0.10 | +0.06 | +0.15 | 0 | 3,103 |
| 2000 | County-equal | All five with state baselines | Rurality | −0.11 | −0.17 | −0.05 | 0 | 3,113 |
| 2004 | County-equal | All five with state baselines | Rurality | −0.04 | −0.10 | +0.02 | 0 | 3,113 |
| 2008 | County-equal | All five with state baselines | Rurality | −0.06 | −0.12 | −0.01 | 0 | 3,113 |
| 2012 | County-equal | All five with state baselines | Rurality | −0.07 | −0.12 | −0.02 | 0 | 3,112 |
| 2016 | County-equal | All five with state baselines | Rurality | −0.08 | −0.12 | −0.03 | 0 | 3,111 |
| 2020 | County-equal | All five with state baselines | Rurality | −0.10 | −0.14 | −0.05 | 0 | 3,103 |
| 2024 | County-equal | All five with state baselines | Rurality | −0.11 | −0.15 | −0.06 | 0 | 3,103 |
Dropping an axis from the five-axis regression lowers the vote-weighted R-squared in 2024 by 16.9 (95% CI 12.1 to 22.8)16.9 percentage points of the variance of the county marginEstimanddrop in R-squared when college share is removed from the five-axis regressionInterval95% CI 12.1 to 22.8n3,103Weightstwo-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. points of variance for college share and 11.0 (95% CI 5.1 to 17.5)11.0 percentage points of the variance of the county marginEstimanddrop in R-squared when non-Hispanic white share is removed from the five-axis regressionInterval95% CI 5.1 to 17.5n3,103Weightstwo-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. for the white share, against 2.72.7 percentage points of the variance of the county marginEstimanddrop in R-squared when median household income is removed from the five-axis regressionInterval95% CI 0.8 to 6.9n3,103Weightstwo-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. for income, 0.70.7 percentage points of the variance of the county marginEstimanddrop in R-squared when rural share is removed from the five-axis regressionInterval95% CI 0.0 to 2.8n3,103Weightstwo-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. for rural share and 0.10.1 percentage points of the variance of the county marginEstimanddrop in R-squared when median age is removed from the five-axis regressionInterval95% CI 0.0 to 2.6n3,103Weightstwo-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. for age. In 2000 the largest loss was the white share's, 15.115.1 percentage points of the variance of the county marginEstimanddrop in R-squared when non-Hispanic white share is removed from the five-axis regressionInterval95% CI 6.9 to 23.8n3,113Weightstwo-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. points, and college share's was 2.42.4 percentage points of the variance of the county marginEstimanddrop in R-squared when college share is removed from the five-axis regressionInterval95% CI 0.5 to 5.4n3,113Weightstwo-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.. Counting each county once, the white share's loss in 2024 (17.617.6 percentage points of the variance of the county marginEstimanddrop in R-squared when non-Hispanic white share is removed from the five-axis regressionInterval95% CI 8.0 to 31.2n3,103Weightsnone (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.) exceeds college share's (12.812.8 percentage points of the variance of the county marginEstimanddrop in R-squared when college share is removed from the five-axis regressionInterval95% CI 8.6 to 16.9n3,103Weightsnone (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 the point estimates, the reverse of the order of the coefficients, because the loss also depends on how much of an axis's variation the other axes already contain.
The five axes reproduce more of the county margin than in 2000
The R-squared of a regression is the share of the variance of county margins that its fitted values reproduce. Figure 5 plots it by election. Vote-weighted, the five axes reproduce 41% (95% CI 30% to 56%)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. in 2000, 52%52% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 44% to 63%n3,112Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2012 and 72% (95% CI 67% to 79%)72% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 67% to 79%n3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2016, then 72%72% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 67% to 80%n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2020 and 69% (95% CI 63% to 78%)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. The rise from 2000 to 2016 is +31.3 (95% CI 22.0 to 39.1)+31.3 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 22.0 to 39.1Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 to 2016SourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. points of variance, and 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., an interval that includes zero. Most of the rise falls between 2012 and 2016, +19.9 (95% CI 15.3 to 23.6)+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. Counting each county once the shares are 20% (95% CI 11% to 33%)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, 33%33% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 24% to 49%n3,112Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2012, 58%58% share of the variance of the county marginEstimandR-squared of the county margin on the five axesInterval95% CI 49% to 70%n3,111Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2016 and 59% (95% CI 49% to 71%)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.
State baselines alone reproduce 34% (95% CI 24% to 43%)34% share of the variance of the county marginEstimandR-squared of the county margin on a baseline for each state aloneInterval95% CI 24% to 43%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 27% (95% CI 19% to 34%)27% share of the variance of the county marginEstimandR-squared of the county margin on a baseline for each state aloneInterval95% CI 19% to 34%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, a change of −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 whose interval includes zero. Adding the five axes to state baselines raises the share to 74% (95% CI 65% to 82%)74% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 65% to 82%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. in 2000 and 87% (95% CI 84% to 90%)87% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 84% to 90%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. The share of the variance within states that the five axes reproduce rises from 61% (95% CI 50% to 72%)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% (95% CI 77% to 88%)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, the five axes with state baselines reproduce 61%61% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 54% to 67%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 82%82% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 79% to 85%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. Nine measures, which add the Black, Hispanic and Asian shares and the foreign-born share to the five axes, reproduce 49%49% share of the variance of the county marginEstimandR-squared of the county margin on the five axes plus the Black, Hispanic and Asian shares and the foreign-born share (nine measures)Interval95% CI 39% to 63%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. in 2000 and 71%71% share of the variance of the county marginEstimandR-squared of the county margin on the five axes plus the Black, Hispanic and Asian shares and the foreign-born share (nine measures)Interval95% CI 67% to 79%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 vote-weighted, so the extra measures add more in 2000 than in 2024.
Figure 5
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
| Election | Weighting | Census measures | Regression | R-squared (%) | Interval low | Interval high | Counties |
|---|---|---|---|---|---|---|---|
| 2000 | Vote-weighted | Election-year census measures | Five axes | 40.8 | 29.6 | 56.0 | 3,113 |
| 2004 | Vote-weighted | Election-year census measures | Five axes | 43.5 | 34.6 | 56.7 | 3,113 |
| 2008 | Vote-weighted | Election-year census measures | Five axes | 48.2 | 39.6 | 59.8 | 3,113 |
| 2012 | Vote-weighted | Election-year census measures | Five axes | 52.3 | 44.3 | 63.2 | 3,112 |
| 2016 | Vote-weighted | Election-year census measures | Five axes | 72.2 | 67.0 | 79.3 | 3,111 |
| 2020 | Vote-weighted | Election-year census measures | Five axes | 72.0 | 67.1 | 79.5 | 3,103 |
| 2024 | Vote-weighted | Election-year census measures | Five axes | 68.6 | 63.1 | 78.1 | 3,103 |
| 2000 | Vote-weighted | Election-year census measures | Five axes and state baselines | 74.1 | 64.8 | 82.0 | 3,113 |
| 2004 | Vote-weighted | Election-year census measures | Five axes and state baselines | 74.7 | 67.9 | 81.9 | 3,113 |
| 2008 | Vote-weighted | Election-year census measures | Five axes and state baselines | 79.9 | 74.6 | 85.4 | 3,113 |
| 2012 | Vote-weighted | Election-year census measures | Five axes and state baselines | 81.9 | 77.1 | 86.8 | 3,112 |
| 2016 | Vote-weighted | Election-year census measures | Five axes and state baselines | 88.9 | 86.6 | 91.7 | 3,111 |
| 2020 | Vote-weighted | Election-year census measures | Five axes and state baselines | 88.3 | 85.6 | 91.4 | 3,103 |
| 2024 | Vote-weighted | Election-year census measures | Five axes and state baselines | 86.6 | 83.6 | 90.4 | 3,103 |
| 2000 | Vote-weighted | Election-year census measures | State baselines alone | 34.2 | 24.3 | 42.7 | 3,113 |
| 2004 | Vote-weighted | Election-year census measures | State baselines alone | 29.5 | 21.1 | 37.7 | 3,113 |
| 2008 | Vote-weighted | Election-year census measures | State baselines alone | 31.6 | 24.1 | 38.2 | 3,113 |
| 2012 | Vote-weighted | Election-year census measures | State baselines alone | 31.4 | 23.2 | 38.6 | 3,112 |
| 2016 | Vote-weighted | Election-year census measures | State baselines alone | 28.9 | 18.5 | 37.2 | 3,111 |
| 2020 | Vote-weighted | Election-year census measures | State baselines alone | 28.9 | 19.2 | 36.7 | 3,103 |
| 2024 | Vote-weighted | Election-year census measures | State baselines alone | 27.4 | 19.0 | 34.4 | 3,103 |
| 2000 | Vote-weighted | Election-year census measures | Nine measures | 49.2 | 38.6 | 62.8 | 3,113 |
| 2004 | Vote-weighted | Election-year census measures | Nine measures | 50.1 | 41.8 | 61.5 | 3,113 |
| 2008 | Vote-weighted | Election-year census measures | Nine measures | 52.1 | 44.6 | 62.6 | 3,113 |
| 2012 | Vote-weighted | Election-year census measures | Nine measures | 56.4 | 49.3 | 66.5 | 3,112 |
| 2016 | Vote-weighted | Election-year census measures | Nine measures | 73.8 | 69.1 | 80.8 | 3,111 |
| 2020 | Vote-weighted | Election-year census measures | Nine measures | 73.4 | 69.8 | 80.5 | 3,103 |
| 2024 | Vote-weighted | Election-year census measures | Nine measures | 71.1 | 67.2 | 79.2 | 3,103 |
| 2000 | County-equal | Election-year census measures | Five axes | 19.6 | 11.4 | 33.3 | 3,113 |
| 2004 | County-equal | Election-year census measures | Five axes | 19.8 | 11.7 | 34.3 | 3,113 |
| 2008 | County-equal | Election-year census measures | Five axes | 25.5 | 17.4 | 39.9 | 3,113 |
| 2012 | County-equal | Election-year census measures | Five axes | 33.1 | 23.9 | 48.8 | 3,112 |
| 2016 | County-equal | Election-year census measures | Five axes | 57.8 | 49.1 | 70.2 | 3,111 |
| 2020 | County-equal | Election-year census measures | Five axes | 59.7 | 50.6 | 71.9 | 3,103 |
| 2024 | County-equal | Election-year census measures | Five axes | 58.7 | 49.4 | 71.4 | 3,103 |
| 2000 | County-equal | Election-year census measures | Five axes and state baselines | 60.7 | 53.9 | 66.8 | 3,113 |
| 2004 | County-equal | Election-year census measures | Five axes and state baselines | 63.0 | 56.6 | 68.4 | 3,113 |
| 2008 | County-equal | Election-year census measures | Five axes and state baselines | 70.9 | 65.4 | 75.3 | 3,113 |
| 2012 | County-equal | Election-year census measures | Five axes and state baselines | 74.0 | 68.9 | 78.1 | 3,112 |
| 2016 | County-equal | Election-year census measures | Five axes and state baselines | 81.9 | 78.9 | 84.8 | 3,111 |
| 2020 | County-equal | Election-year census measures | Five axes and state baselines | 83.4 | 80.8 | 86.0 | 3,103 |
| 2024 | County-equal | Election-year census measures | Five axes and state baselines | 82.0 | 79.2 | 84.9 | 3,103 |
| 2000 | County-equal | Election-year census measures | State baselines alone | 33.7 | 23.4 | 42.6 | 3,113 |
| 2004 | County-equal | Election-year census measures | State baselines alone | 33.1 | 22.8 | 41.5 | 3,113 |
| 2008 | County-equal | Election-year census measures | State baselines alone | 35.5 | 26.2 | 43.3 | 3,113 |
| 2012 | County-equal | Election-year census measures | State baselines alone | 35.8 | 26.9 | 43.6 | 3,112 |
| 2016 | County-equal | Election-year census measures | State baselines alone | 33.0 | 24.3 | 41.5 | 3,111 |
| 2020 | County-equal | Election-year census measures | State baselines alone | 34.8 | 26.7 | 42.4 | 3,103 |
| 2024 | County-equal | Election-year census measures | State baselines alone | 34.6 | 26.7 | 41.6 | 3,103 |
| 2000 | County-equal | Election-year census measures | Nine measures | 26.6 | 19.6 | 38.2 | 3,113 |
| 2004 | County-equal | Election-year census measures | Nine measures | 27.7 | 21.0 | 39.0 | 3,113 |
| 2008 | County-equal | Election-year census measures | Nine measures | 30.5 | 23.1 | 43.1 | 3,113 |
| 2012 | County-equal | Election-year census measures | Nine measures | 38.8 | 29.8 | 52.0 | 3,112 |
| 2016 | County-equal | Election-year census measures | Nine measures | 62.2 | 54.0 | 72.2 | 3,111 |
| 2020 | County-equal | Election-year census measures | Nine measures | 65.7 | 58.7 | 74.1 | 3,103 |
| 2024 | County-equal | Election-year census measures | Nine measures | 66.0 | 59.0 | 74.3 | 3,103 |
| 2000 | Vote-weighted | 2024 census measures for every election | Five axes | 35.1 | 26.7 | 48.1 | 3,102 |
| 2004 | Vote-weighted | 2024 census measures for every election | Five axes | 39.6 | 32.8 | 51.4 | 3,103 |
| 2008 | Vote-weighted | 2024 census measures for every election | Five axes | 45.2 | 37.9 | 56.5 | 3,103 |
| 2012 | Vote-weighted | 2024 census measures for every election | Five axes | 49.7 | 42.2 | 61.0 | 3,103 |
| 2016 | Vote-weighted | 2024 census measures for every election | Five axes | 70.5 | 65.0 | 78.2 | 3,104 |
| 2020 | Vote-weighted | 2024 census measures for every election | Five axes | 71.5 | 66.6 | 79.4 | 3,104 |
| 2024 | Vote-weighted | 2024 census measures for every election | Five axes | 68.6 | 63.1 | 78.1 | 3,103 |
| 2000 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 68.8 | 59.8 | 77.1 | 3,102 |
| 2004 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 70.9 | 62.8 | 79.3 | 3,103 |
| 2008 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 77.8 | 71.7 | 84.2 | 3,103 |
| 2012 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 80.3 | 74.6 | 85.9 | 3,103 |
| 2016 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 88.4 | 85.9 | 91.2 | 3,104 |
| 2020 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 88.3 | 85.6 | 91.4 | 3,104 |
| 2024 | Vote-weighted | 2024 census measures for every election | Five axes and state baselines | 86.6 | 83.6 | 90.4 | 3,103 |
| 2000 | Vote-weighted | 2024 census measures for every election | State baselines alone | 33.8 | 23.2 | 42.9 | 3,102 |
| 2004 | Vote-weighted | 2024 census measures for every election | State baselines alone | 29.4 | 20.9 | 37.9 | 3,103 |
| 2008 | Vote-weighted | 2024 census measures for every election | State baselines alone | 31.4 | 23.7 | 37.6 | 3,103 |
| 2012 | Vote-weighted | 2024 census measures for every election | State baselines alone | 31.2 | 22.6 | 37.6 | 3,103 |
| 2016 | Vote-weighted | 2024 census measures for every election | State baselines alone | 28.8 | 18.3 | 37.2 | 3,104 |
| 2020 | Vote-weighted | 2024 census measures for every election | State baselines alone | 28.9 | 19.2 | 36.7 | 3,104 |
| 2024 | Vote-weighted | 2024 census measures for every election | State baselines alone | 27.4 | 19.0 | 34.4 | 3,103 |
| 2000 | Vote-weighted | 2024 census measures for every election | Nine measures | 43.5 | 34.8 | 57.0 | 3,102 |
| 2004 | Vote-weighted | 2024 census measures for every election | Nine measures | 45.9 | 39.5 | 57.1 | 3,103 |
| 2008 | Vote-weighted | 2024 census measures for every election | Nine measures | 49.1 | 42.5 | 59.6 | 3,103 |
| 2012 | Vote-weighted | 2024 census measures for every election | Nine measures | 53.9 | 47.1 | 64.4 | 3,103 |
| 2016 | Vote-weighted | 2024 census measures for every election | Nine measures | 72.2 | 67.4 | 79.4 | 3,104 |
| 2020 | Vote-weighted | 2024 census measures for every election | Nine measures | 73.0 | 69.2 | 80.5 | 3,104 |
| 2024 | Vote-weighted | 2024 census measures for every election | Nine measures | 71.1 | 67.2 | 79.2 | 3,103 |
| 2000 | County-equal | 2024 census measures for every election | Five axes | 17.8 | 11.4 | 29.5 | 3,102 |
| 2004 | County-equal | 2024 census measures for every election | Five axes | 19.6 | 13.1 | 31.9 | 3,103 |
| 2008 | County-equal | 2024 census measures for every election | Five axes | 24.3 | 16.7 | 37.5 | 3,103 |
| 2012 | County-equal | 2024 census measures for every election | Five axes | 31.2 | 21.7 | 47.0 | 3,103 |
| 2016 | County-equal | 2024 census measures for every election | Five axes | 55.5 | 45.9 | 68.3 | 3,104 |
| 2020 | County-equal | 2024 census measures for every election | Five axes | 58.7 | 49.3 | 71.2 | 3,104 |
| 2024 | County-equal | 2024 census measures for every election | Five axes | 58.7 | 49.4 | 71.4 | 3,103 |
| 2000 | County-equal | 2024 census measures for every election | Five axes and state baselines | 55.1 | 48.3 | 61.7 | 3,102 |
| 2004 | County-equal | 2024 census measures for every election | Five axes and state baselines | 58.6 | 52.1 | 64.5 | 3,103 |
| 2008 | County-equal | 2024 census measures for every election | Five axes and state baselines | 67.7 | 62.5 | 72.1 | 3,103 |
| 2012 | County-equal | 2024 census measures for every election | Five axes and state baselines | 71.4 | 67.0 | 75.4 | 3,103 |
| 2016 | County-equal | 2024 census measures for every election | Five axes and state baselines | 79.9 | 76.6 | 83.0 | 3,104 |
| 2020 | County-equal | 2024 census measures for every election | Five axes and state baselines | 82.5 | 79.7 | 85.3 | 3,104 |
| 2024 | County-equal | 2024 census measures for every election | Five axes and state baselines | 82.0 | 79.2 | 84.9 | 3,103 |
| 2000 | County-equal | 2024 census measures for every election | State baselines alone | 33.7 | 23.5 | 42.8 | 3,102 |
| 2004 | County-equal | 2024 census measures for every election | State baselines alone | 33.1 | 22.5 | 41.5 | 3,103 |
| 2008 | County-equal | 2024 census measures for every election | State baselines alone | 35.4 | 26.2 | 43.1 | 3,103 |
| 2012 | County-equal | 2024 census measures for every election | State baselines alone | 35.6 | 26.8 | 43.4 | 3,103 |
| 2016 | County-equal | 2024 census measures for every election | State baselines alone | 32.6 | 24.2 | 40.7 | 3,104 |
| 2020 | County-equal | 2024 census measures for every election | State baselines alone | 34.8 | 26.6 | 42.4 | 3,104 |
| 2024 | County-equal | 2024 census measures for every election | State baselines alone | 34.6 | 26.7 | 41.6 | 3,103 |
| 2000 | County-equal | 2024 census measures for every election | Nine measures | 26.2 | 19.8 | 36.6 | 3,102 |
| 2004 | County-equal | 2024 census measures for every election | Nine measures | 28.1 | 21.9 | 38.3 | 3,103 |
| 2008 | County-equal | 2024 census measures for every election | Nine measures | 29.7 | 22.3 | 41.2 | 3,103 |
| 2012 | County-equal | 2024 census measures for every election | Nine measures | 37.5 | 28.0 | 50.5 | 3,103 |
| 2016 | County-equal | 2024 census measures for every election | Nine measures | 60.6 | 52.0 | 70.6 | 3,104 |
| 2020 | County-equal | 2024 census measures for every election | Nine measures | 65.1 | 57.8 | 73.7 | 3,104 |
| 2024 | County-equal | 2024 census measures for every election | Nine measures | 66.0 | 59.0 | 74.3 | 3,103 |
| 2000 | Vote-weighted | 2000 census measures for every election | Five axes | 40.8 | 29.6 | 56.0 | 3,113 |
| 2004 | Vote-weighted | 2000 census measures for every election | Five axes | 43.1 | 33.3 | 56.5 | 3,113 |
| 2008 | Vote-weighted | 2000 census measures for every election | Five axes | 47.9 | 38.0 | 60.2 | 3,113 |
| 2012 | Vote-weighted | 2000 census measures for every election | Five axes | 52.8 | 43.4 | 64.5 | 3,113 |
| 2016 | Vote-weighted | 2000 census measures for every election | Five axes | 72.0 | 64.9 | 79.7 | 3,111 |
| 2020 | Vote-weighted | 2000 census measures for every election | Five axes | 71.5 | 65.4 | 79.0 | 3,111 |
| 2024 | Vote-weighted | 2000 census measures for every election | Five axes | 67.6 | 61.1 | 76.6 | 3,110 |
| 2000 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 74.1 | 64.8 | 82.0 | 3,113 |
| 2004 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 74.5 | 67.5 | 82.0 | 3,113 |
| 2008 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 79.2 | 73.7 | 85.1 | 3,113 |
| 2012 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 81.2 | 76.4 | 86.4 | 3,113 |
| 2016 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 88.0 | 85.3 | 90.8 | 3,111 |
| 2020 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 86.4 | 83.2 | 89.8 | 3,111 |
| 2024 | Vote-weighted | 2000 census measures for every election | Five axes and state baselines | 84.0 | 80.7 | 88.1 | 3,110 |
| 2000 | Vote-weighted | 2000 census measures for every election | State baselines alone | 34.2 | 24.3 | 42.7 | 3,113 |
| 2004 | Vote-weighted | 2000 census measures for every election | State baselines alone | 29.5 | 21.1 | 37.7 | 3,113 |
| 2008 | Vote-weighted | 2000 census measures for every election | State baselines alone | 31.6 | 24.1 | 38.2 | 3,113 |
| 2012 | Vote-weighted | 2000 census measures for every election | State baselines alone | 31.3 | 23.2 | 38.6 | 3,113 |
| 2016 | Vote-weighted | 2000 census measures for every election | State baselines alone | 28.9 | 18.4 | 37.2 | 3,111 |
| 2020 | Vote-weighted | 2000 census measures for every election | State baselines alone | 29.0 | 19.3 | 36.8 | 3,111 |
| 2024 | Vote-weighted | 2000 census measures for every election | State baselines alone | 27.6 | 19.2 | 35.1 | 3,110 |
| 2000 | Vote-weighted | 2000 census measures for every election | Nine measures | 49.2 | 38.6 | 62.8 | 3,113 |
| 2004 | Vote-weighted | 2000 census measures for every election | Nine measures | 49.8 | 41.1 | 61.3 | 3,113 |
| 2008 | Vote-weighted | 2000 census measures for every election | Nine measures | 52.2 | 43.7 | 62.8 | 3,113 |
| 2012 | Vote-weighted | 2000 census measures for every election | Nine measures | 57.4 | 49.1 | 67.5 | 3,113 |
| 2016 | Vote-weighted | 2000 census measures for every election | Nine measures | 74.3 | 68.2 | 81.0 | 3,111 |
| 2020 | Vote-weighted | 2000 census measures for every election | Nine measures | 73.6 | 68.6 | 80.5 | 3,111 |
| 2024 | Vote-weighted | 2000 census measures for every election | Nine measures | 70.8 | 65.8 | 78.6 | 3,110 |
| 2000 | County-equal | 2000 census measures for every election | Five axes | 19.6 | 11.4 | 33.3 | 3,113 |
| 2004 | County-equal | 2000 census measures for every election | Five axes | 19.5 | 10.8 | 34.2 | 3,113 |
| 2008 | County-equal | 2000 census measures for every election | Five axes | 25.5 | 16.7 | 40.3 | 3,113 |
| 2012 | County-equal | 2000 census measures for every election | Five axes | 33.0 | 22.5 | 49.6 | 3,113 |
| 2016 | County-equal | 2000 census measures for every election | Five axes | 57.4 | 47.3 | 70.6 | 3,111 |
| 2020 | County-equal | 2000 census measures for every election | Five axes | 59.1 | 48.7 | 72.5 | 3,111 |
| 2024 | County-equal | 2000 census measures for every election | Five axes | 58.4 | 47.8 | 71.8 | 3,110 |
| 2000 | County-equal | 2000 census measures for every election | Five axes and state baselines | 60.7 | 53.9 | 66.8 | 3,113 |
| 2004 | County-equal | 2000 census measures for every election | Five axes and state baselines | 63.1 | 56.7 | 68.6 | 3,113 |
| 2008 | County-equal | 2000 census measures for every election | Five axes and state baselines | 71.2 | 66.0 | 75.4 | 3,113 |
| 2012 | County-equal | 2000 census measures for every election | Five axes and state baselines | 74.7 | 69.9 | 78.5 | 3,113 |
| 2016 | County-equal | 2000 census measures for every election | Five axes and state baselines | 82.5 | 79.9 | 85.1 | 3,111 |
| 2020 | County-equal | 2000 census measures for every election | Five axes and state baselines | 83.5 | 81.2 | 86.1 | 3,111 |
| 2024 | County-equal | 2000 census measures for every election | Five axes and state baselines | 82.5 | 80.1 | 85.2 | 3,110 |
| 2000 | County-equal | 2000 census measures for every election | State baselines alone | 33.7 | 23.4 | 42.6 | 3,113 |
| 2004 | County-equal | 2000 census measures for every election | State baselines alone | 33.1 | 22.8 | 41.5 | 3,113 |
| 2008 | County-equal | 2000 census measures for every election | State baselines alone | 35.5 | 26.2 | 43.3 | 3,113 |
| 2012 | County-equal | 2000 census measures for every election | State baselines alone | 35.6 | 26.7 | 43.4 | 3,113 |
| 2016 | County-equal | 2000 census measures for every election | State baselines alone | 32.9 | 24.3 | 41.5 | 3,111 |
| 2020 | County-equal | 2000 census measures for every election | State baselines alone | 35.2 | 26.8 | 43.2 | 3,111 |
| 2024 | County-equal | 2000 census measures for every election | State baselines alone | 35.1 | 26.9 | 42.4 | 3,110 |
| 2000 | County-equal | 2000 census measures for every election | Nine measures | 26.6 | 19.6 | 38.2 | 3,113 |
| 2004 | County-equal | 2000 census measures for every election | Nine measures | 26.8 | 19.7 | 38.7 | 3,113 |
| 2008 | County-equal | 2000 census measures for every election | Nine measures | 29.6 | 21.8 | 43.1 | 3,113 |
| 2012 | County-equal | 2000 census measures for every election | Nine measures | 37.7 | 27.8 | 52.1 | 3,113 |
| 2016 | County-equal | 2000 census measures for every election | Nine measures | 61.2 | 52.1 | 72.1 | 3,111 |
| 2020 | County-equal | 2000 census measures for every election | Nine measures | 64.2 | 56.3 | 73.9 | 3,111 |
| 2024 | County-equal | 2000 census measures for every election | Nine measures | 64.3 | 56.7 | 73.6 | 3,110 |
A higher R-squared can come from a wider fitted part of the margin or from a narrower left-over part. Figure 6 separates them. Vote-weighted, the standard deviation of county margins rose from 26.526.5 percentage points of marginEstimandweighted standard deviation of the county marginInterval95% CI 24.0 to 28.6n3,113Weightstwo-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. points in 2000 to 35.935.9 percentage points of marginEstimandweighted standard deviation of the county marginInterval95% CI 33.3 to 37.9n3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2016 and was 32.832.8 percentage points of marginEstimandweighted standard deviation of the county marginInterval95% CI 30.6 to 35.0n3,103Weightstwo-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. The fitted part went from 16.9 (95% CI 13.4 to 20.9)16.9 percentage points of marginEstimandweighted standard deviation of the margin fitted by the five axesInterval95% CI 13.4 to 20.9n3,113Weightstwo-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 27.2 (95% CI 24.7 to 30.4)27.2 percentage points of marginEstimandweighted standard deviation of the margin fitted by the five axesInterval95% CI 24.7 to 30.4n3,103Weightstwo-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., a change of +10.3 (95% CI 7.0 to 13.8)+10.3 percentage points of marginEstimandchange between the two elections in the weighted standard deviation of the margin fitted by the five axes (the same states are drawn for both elections)Interval95% CI 7.0 to 13.8Weightstwo-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. The left-over part went from 20.4 (95% CI 18.0 to 21.9)20.4 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axesInterval95% CI 18.0 to 21.9n3,113Weightstwo-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 18.4 (95% CI 15.5 to 19.7)18.4 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axesInterval95% CI 15.5 to 19.7n3,103Weightstwo-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., a change of −2.0 (95% CI −3.6 to −0.7)−2.0 percentage points of marginEstimandchange between the two elections in the weighted standard deviation of the margin left over after the five axes (the same states are drawn for both elections)Interval95% CI −3.6 to −0.7Weightstwo-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. Holding the left-over part at its 2000 value would put the 2024 share at 61% (95% CI 53% to 71%)61% share of the variance of the county marginEstimandR-squared in 2024 if the left-over spread of the five-axis regression stayed at its 2000 valueInterval95% CI 53% to 71%Weightstwo-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. instead of 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., so the narrower left-over part accounts for 7.2 (95% CI 2.5 to 13.3)7.2 percentage points of the variance of the county marginEstimandpart of the 2000 to 2024 rise in R-squared that comes from the narrower left-over spreadInterval95% CI 2.5 to 13.3Weightstwo-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. of the +27.8 (95% CI 18.9 to 37.3)+27.8 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 18.9 to 37.3Weightstwo-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 of the rise. With state baselines added, the left-over part went from 13.513.5 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axes and a baseline for each stateInterval95% CI 11.7 to 14.7n3,113Weightstwo-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 12.012.0 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axes and a baseline for each stateInterval95% CI 10.4 to 12.9n3,103Weightstwo-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, the standard deviation of margins rose from 24.524.5 percentage points of marginEstimandweighted standard deviation of the county marginInterval95% CI 22.3 to 26.3n3,113Weightsnone (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. to 31.731.7 percentage points of marginEstimandweighted standard deviation of the county marginInterval95% CI 29.5 to 33.7n3,103Weightsnone (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., the fitted part went from 10.910.9 percentage points of marginEstimandweighted standard deviation of the margin fitted by the five axesInterval95% CI 8.2 to 14.3n3,113Weightsnone (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. to 24.324.3 percentage points of marginEstimandweighted standard deviation of the margin fitted by the five axesInterval95% CI 21.3 to 27.8n3,103Weightsnone (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. and the left-over part from 22.022.0 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axesInterval95% CI 19.2 to 23.9n3,113Weightsnone (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. to 20.420.4 percentage points of marginEstimandweighted standard deviation of the margin left over after the five axesInterval95% CI 16.9 to 22.6n3,103Weightsnone (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..
Figure 6
Vote-weighted, the part of the county margin that the five axes leave over narrowed from 20.4 to 18.4 points while the part they fit widened from 16.9 to 27.2
Standard deviation across counties of the two-party margin, the margin fitted by the five axes and the margin left over, 2000 to 2024, counties weighted by two-party votes
Standard deviation across counties (percentage points of margin)
Show data
| Election | Weighting | Part of the margin | Standard deviation (points) | Interval low | Interval high | Counties |
|---|---|---|---|---|---|---|
| 2000 | Vote-weighted | Margin | 26.5 | 24.0 | 28.6 | 3,113 |
| 2004 | Vote-weighted | Margin | 27.4 | 25.2 | 29.1 | 3,113 |
| 2008 | Vote-weighted | Margin | 28.8 | 26.5 | 30.7 | 3,113 |
| 2012 | Vote-weighted | Margin | 31.1 | 28.7 | 33.1 | 3,112 |
| 2016 | Vote-weighted | Margin | 35.9 | 33.3 | 37.9 | 3,111 |
| 2020 | Vote-weighted | Margin | 34.1 | 31.8 | 36.1 | 3,103 |
| 2024 | Vote-weighted | Margin | 32.8 | 30.6 | 35.0 | 3,103 |
| 2000 | Vote-weighted | Fitted by the five axes | 16.9 | 13.4 | 20.9 | 3,113 |
| 2004 | Vote-weighted | Fitted by the five axes | 18.0 | 15.0 | 21.4 | 3,113 |
| 2008 | Vote-weighted | Fitted by the five axes | 20.0 | 17.0 | 23.1 | 3,113 |
| 2012 | Vote-weighted | Fitted by the five axes | 22.5 | 19.5 | 25.7 | 3,112 |
| 2016 | Vote-weighted | Fitted by the five axes | 30.5 | 27.7 | 33.3 | 3,111 |
| 2020 | Vote-weighted | Fitted by the five axes | 28.9 | 26.4 | 31.8 | 3,103 |
| 2024 | Vote-weighted | Fitted by the five axes | 27.2 | 24.7 | 30.4 | 3,103 |
| 2000 | Vote-weighted | Left over after the five axes | 20.4 | 18.0 | 21.9 | 3,113 |
| 2004 | Vote-weighted | Left over after the five axes | 20.6 | 18.4 | 21.7 | 3,113 |
| 2008 | Vote-weighted | Left over after the five axes | 20.7 | 18.1 | 22.2 | 3,113 |
| 2012 | Vote-weighted | Left over after the five axes | 21.5 | 18.8 | 23.0 | 3,112 |
| 2016 | Vote-weighted | Left over after the five axes | 18.9 | 16.5 | 20.3 | 3,111 |
| 2020 | Vote-weighted | Left over after the five axes | 18.0 | 15.5 | 19.3 | 3,103 |
| 2024 | Vote-weighted | Left over after the five axes | 18.4 | 15.5 | 19.7 | 3,103 |
| 2000 | Vote-weighted | Left over after the five axes and state baselines | 13.5 | 11.7 | 14.7 | 3,113 |
| 2004 | Vote-weighted | Left over after the five axes and state baselines | 13.8 | 11.9 | 14.9 | 3,113 |
| 2008 | Vote-weighted | Left over after the five axes and state baselines | 12.9 | 11.3 | 13.9 | 3,113 |
| 2012 | Vote-weighted | Left over after the five axes and state baselines | 13.2 | 11.5 | 14.3 | 3,112 |
| 2016 | Vote-weighted | Left over after the five axes and state baselines | 12.0 | 10.4 | 13.0 | 3,111 |
| 2020 | Vote-weighted | Left over after the five axes and state baselines | 11.7 | 10.2 | 12.6 | 3,103 |
| 2024 | Vote-weighted | Left over after the five axes and state baselines | 12.0 | 10.4 | 12.9 | 3,103 |
| 2000 | County-equal | Margin | 24.5 | 22.3 | 26.3 | 3,113 |
| 2004 | County-equal | Margin | 25.4 | 23.2 | 27.1 | 3,113 |
| 2008 | County-equal | Margin | 28.1 | 25.7 | 30.1 | 3,113 |
| 2012 | County-equal | Margin | 30.1 | 27.4 | 32.3 | 3,112 |
| 2016 | County-equal | Margin | 32.3 | 29.8 | 34.5 | 3,111 |
| 2020 | County-equal | Margin | 32.6 | 30.3 | 34.8 | 3,103 |
| 2024 | County-equal | Margin | 31.7 | 29.5 | 33.7 | 3,103 |
| 2000 | County-equal | Fitted by the five axes | 10.8 | 8.2 | 14.3 | 3,113 |
| 2004 | County-equal | Fitted by the five axes | 11.3 | 8.6 | 14.9 | 3,113 |
| 2008 | County-equal | Fitted by the five axes | 14.2 | 11.7 | 17.7 | 3,113 |
| 2012 | County-equal | Fitted by the five axes | 17.3 | 14.6 | 21.0 | 3,112 |
| 2016 | County-equal | Fitted by the five axes | 24.6 | 21.8 | 27.9 | 3,111 |
| 2020 | County-equal | Fitted by the five axes | 25.2 | 22.4 | 28.8 | 3,103 |
| 2024 | County-equal | Fitted by the five axes | 24.3 | 21.3 | 27.8 | 3,103 |
| 2000 | County-equal | Left over after the five axes | 22.0 | 19.2 | 23.9 | 3,113 |
| 2004 | County-equal | Left over after the five axes | 22.7 | 19.8 | 24.5 | 3,113 |
| 2008 | County-equal | Left over after the five axes | 24.3 | 20.9 | 26.5 | 3,113 |
| 2012 | County-equal | Left over after the five axes | 24.6 | 20.8 | 27.1 | 3,112 |
| 2016 | County-equal | Left over after the five axes | 21.0 | 17.6 | 23.2 | 3,111 |
| 2020 | County-equal | Left over after the five axes | 20.7 | 17.1 | 23.0 | 3,103 |
| 2024 | County-equal | Left over after the five axes | 20.4 | 16.9 | 22.6 | 3,103 |
| 2000 | County-equal | Left over after the five axes and state baselines | 15.4 | 14.0 | 16.4 | 3,113 |
| 2004 | County-equal | Left over after the five axes and state baselines | 15.4 | 14.2 | 16.4 | 3,113 |
| 2008 | County-equal | Left over after the five axes and state baselines | 15.2 | 14.0 | 16.1 | 3,113 |
| 2012 | County-equal | Left over after the five axes and state baselines | 15.3 | 14.2 | 16.3 | 3,112 |
| 2016 | County-equal | Left over after the five axes and state baselines | 13.7 | 12.5 | 14.7 | 3,111 |
| 2020 | County-equal | Left over after the five axes and state baselines | 13.3 | 12.1 | 14.3 | 3,103 |
| 2024 | County-equal | Left over after the five axes and state baselines | 13.4 | 12.1 | 14.4 | 3,103 |
The census measures are interpolated, which could change the early R-squared if the early measures are noisier. The last four views of Figure 5 hold the measures of one year fixed for every election. With the 2024 measures, the vote-weighted five-axis share is 35%35% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2024 census measures used for every electionInterval95% CI 27% to 48%n3,102Weightstwo-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. in 2000, 71%71% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2024 census measures used for every electionInterval95% CI 65% to 78%n3,104Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. in 2016 and 69%69% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2024 census measures used for every electionInterval95% 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 with the 2000 measures it is 41%41% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2000 census measures used for every electionInterval95% 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., 72%72% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2000 census measures used for every electionInterval95% CI 65% to 80%n3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panelNoteInterval: 95% percentile interval from a bootstrap over states, 2,000 replicates. and 68%68% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2000 census measures used for every electionInterval95% CI 61% to 77%n3,110Weightstwo-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 the shares are 18%18% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2024 census measures used for every electionInterval95% CI 11% to 29%n3,102Weightsnone (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 axes, with the 2024 census measures used for every electionInterval95% 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 with the 2024 measures, and 20%20% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2000 census measures used for every electionInterval95% 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. and 58%58% share of the variance of the county marginEstimandR-squared of the county margin on the five axes, with the 2000 census measures used for every electionInterval95% CI 48% to 72%n3,110Weightsnone (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. with the 2000 measures. The rise from 2012 to 2016 holds under each choice, at +19.9+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 with the measures of the year, +20.9+20.9 percentage points of the variance of the county marginEstimandchange in the R-squared of the five-axis regression with the 2024 census measures used for every electionn3,104Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 to 2016SourceCounty demographics panel with the 2024 measures and +19.3+19.3 percentage points of the variance of the county marginEstimandchange in the R-squared of the five-axis regression with the 2000 census measures used for every electionn3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 to 2016SourceCounty demographics panel with the 2000 measures. Small differences do not: the change from 2016 to 2024 is −3.6−3.6 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, −2.0−2.0 percentage points of the variance of the county marginEstimandchange in the R-squared of the five-axis regression with the 2024 census measures used for every electionn3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 to 2024SourceCounty demographics panel and −4.5−4.5 percentage points of the variance of the county marginEstimandchange in the R-squared of the five-axis regression with the 2000 census measures used for every electionn3,110Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 to 2024SourceCounty demographics panel under the three choices. The census definition of rural also changes with the 2020 census, and the county-equal mean rural share goes from 59.6%59.6% percent of residentsEstimandmean rural share of residents across counties, each county counted oncen3,112Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 presidential electionSourceCounty demographics panel in 2012 to 64.0%64.0% percent of residentsEstimandmean rural share of residents across counties, each county counted oncen3,103Weightsnone (each county counts once)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panel in 2020, while the population-weighted mean goes from 19.4%19.4% percent of residentsEstimandmean rural share of residents across counties, weighted by county populationn3,112Weightscounty populationPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2012 presidential electionSourceCounty demographics panel to 20.0%20.0% percent of residentsEstimandmean rural share of residents across counties, weighted by county populationn3,103Weightscounty populationPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panel.
Chapter 10.1, Nationalization of the county vote measures the spread of county swings and Chapter 10.2, Distance from the national vote, 1868 to 2024 the distance of counties from the national margin, which are different quantities from the spread of margins here. Chapter 15.5, Composition and behavior in county change (planned) separates change in composition from change in behavior.
Weighting changes the level of the correlations and the likeliest leader in 2020
Figure 7 compares the vote-weighted and county-equal correlations for 2000 and 2024. In 2000 the vote-weighted correlation of college share exceeds the county-equal one by +0.18 (95% CI 0.09 to 0.26)+0.18 difference in correlationEstimandvote-weighted correlation of college share with the county margin minus the county-equal correlationInterval95% CI 0.09 to 0.26Weightstwo-party votes minus none (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., and that of rural share is more negative by −0.19 (95% CI −0.26 to −0.12)−0.19 difference in correlationEstimandvote-weighted correlation of rural share with the county margin minus the county-equal correlationInterval95% CI −0.26 to −0.12Weightstwo-party votes minus none (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 2024 the differences are +0.11 (95% CI 0.02 to 0.20)+0.11 difference in correlationEstimandvote-weighted correlation of college share with the county margin minus the county-equal correlationInterval95% CI 0.02 to 0.20Weightstwo-party votes minus none (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. and −0.13 (95% CI −0.20 to −0.05)−0.13 difference in correlationEstimandvote-weighted correlation of rural share with the county margin minus the county-equal correlationInterval95% CI −0.20 to −0.05Weightstwo-party votes minus none (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., and the white share's is −0.08 (95% CI −0.17 to 0.04)−0.08 difference in correlationEstimandvote-weighted correlation of non-Hispanic white share with the county margin minus the county-equal correlationInterval95% CI −0.17 to 0.04Weightstwo-party votes minus none (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., an interval that includes zero. For income the difference goes from +0.08 (95% CI −0.05 to 0.19)+0.08 difference in correlationEstimandvote-weighted correlation of median household income with the county margin minus the county-equal correlationInterval95% CI −0.05 to 0.19Weightstwo-party votes minus none (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. to +0.09 (95% CI −0.04 to 0.22)+0.09 difference in correlationEstimandvote-weighted correlation of median household income with the county margin minus the county-equal correlationInterval95% CI −0.04 to 0.22Weightstwo-party votes minus none (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. and for median age from +0.09 (95% CI −0.03 to 0.20)+0.09 difference in correlationEstimandvote-weighted correlation of median age with the county margin minus the county-equal correlationInterval95% CI −0.03 to 0.20Weightstwo-party votes minus none (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. to −0.11 (95% CI −0.20 to −0.01)−0.11 difference in correlationEstimandvote-weighted correlation of median age with the county margin minus the county-equal correlationInterval95% CI −0.20 to −0.01Weightstwo-party votes minus none (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.. The R-squared of the five axes is higher vote-weighted in every election, by +21.2 (95% CI 9.5 to 31.9)+21.2 percentage points of the variance of the county marginEstimandvote-weighted R-squared of the five-axis regression minus the county-equal R-squaredInterval95% CI 9.5 to 31.9Weightstwo-party votes minus none (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. points in 2000 and +9.9 (95% CI 0.8 to 18.5)+9.9 percentage points of the variance of the county marginEstimandvote-weighted R-squared of the five-axis regression minus the county-equal R-squaredInterval95% CI 0.8 to 18.5Weightstwo-party votes minus none (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, and by +13.4 (95% CI 5.3 to 20.9)+13.4 percentage points of the variance of the county marginEstimandvote-weighted R-squared of the five-axis regression with state baselines minus the county-equal R-squaredInterval95% CI 5.3 to 20.9Weightstwo-party votes minus none (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. and +4.7 (95% CI 1.2 to 8.6)+4.7 percentage points of the variance of the county marginEstimandvote-weighted R-squared of the five-axis regression with state baselines minus the county-equal R-squaredInterval95% CI 1.2 to 8.6Weightstwo-party votes minus none (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. points with state baselines.
Figure 7
The vote-weighted correlation of college share exceeds the county-equal one by +0.18 in 2000 and by +0.11 in 2024
Correlation of each axis with the county two-party margin when counties are weighted by two-party votes and when each county counts once, 2000 and 2024
- County-equal
- Vote-weighted
Show data
| axis | ce_high | ce_low | county_equal | election | n_units | vote_weighted | vw_high | vw_low |
|---|---|---|---|---|---|---|---|---|
| Education | 0.195 | 0.005 | 0.100 | 2000 | 3,113 | 0.282 | 0.381 | 0.182 |
| Income | 0.137 | −0.101 | 0.018 | 2000 | 3,113 | 0.096 | 0.198 | −0.005 |
| Race and ethnicity | −0.259 | −0.541 | −0.400 | 2000 | 3,113 | −0.519 | −0.357 | −0.681 |
| Age | −0.054 | −0.283 | −0.169 | 2000 | 3,113 | −0.076 | 0.036 | −0.187 |
| Rurality | −0.180 | −0.357 | −0.269 | 2000 | 3,113 | −0.461 | −0.357 | −0.565 |
| Education | 0.638 | 0.474 | 0.556 | 2024 | 3,103 | 0.663 | 0.722 | 0.604 |
| Income | 0.434 | 0.173 | 0.304 | 2024 | 3,103 | 0.397 | 0.504 | 0.290 |
| Race and ethnicity | −0.285 | −0.707 | −0.496 | 2024 | 3,103 | −0.574 | −0.440 | −0.707 |
| Age | −0.118 | −0.272 | −0.195 | 2024 | 3,103 | −0.310 | −0.214 | −0.406 |
| Rurality | −0.462 | −0.592 | −0.527 | 2024 | 3,103 | −0.653 | −0.564 | −0.741 |
A vote-weighted statistic describes the counties where the votes are, and a county-equal statistic describes the typical county. The 135135 countiesEstimandcount of counties with 250,000 or more two-party votesn3,103Weightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty election returns counties with 250,000250,000 two-party votesEstimandvote cut for a large countyWeightsnone (full counts)PopulationcountiesPeriod2000 to 2024SourceCounty election returns or more two-party votes cast 46%46% share of two-party votesEstimandshare of two-party votes cast in counties with 250,000 or more two-party votesn3,103Weightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty election returns of the votes in 2024 and 7676 countiesEstimandcount of counties with 250,000 or more two-party votesn3,113Weightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty election returns such counties cast 35%35% share of two-party votesEstimandshare of two-party votes cast in counties with 250,000 or more two-party votesn3,113Weightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty election returns in 2000, so the vote-weighted figures describe the large counties more than the county-equal ones do. Solon, Haider and Wooldridge [5] discuss the reasons for weighting a regression aimed at causal effects and find that each sometimes does not apply where practitioners assume it does. The regressions here are descriptive, and the chapter reports both weightings and does not choose between them. They agree on the sign of every correlation in every election. They differ in the leader of the point estimates in 2020, where rural share leads vote-weighted and college share and the white share are level county-equal, and in the share of the margin that the five axes reproduce in the early elections.
Each axis is several measures
Figure 8 replaces the single measure of each axis with alternatives, for 2024 and the same counties. Within education, the share of adults with a bachelor's degree has a vote-weighted correlation of +0.66 (95% CI 0.60 to 0.72)+0.66 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with bachelor's degree or moreInterval95% 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. and the share without a high school diploma has −0.03 (95% CI −0.20 to 0.14)−0.03 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with no high school diplomaInterval95% CI −0.20 to 0.14n3,103Weightstwo-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, so the education axis here is the top of the schooling distribution. Within income, median household income has +0.40 (95% CI 0.29 to 0.50)+0.40 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median household incomeInterval95% CI 0.29 to 0.50n3,103Weightstwo-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., per capita income +0.55 (95% CI 0.47 to 0.64)+0.55 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with per capita incomeInterval95% CI 0.47 to 0.64n3,103Weightstwo-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 the poverty rate −0.03 (95% CI −0.14 to 0.09)−0.03 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with poverty rateInterval95% CI −0.14 to 0.09n3,103Weightstwo-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. Within age, median age has −0.31−0.31 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with median ageInterval95% CI −0.41 to −0.21n3,103Weightstwo-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., the share aged 6565 years of ageEstimandage at which the older-adult share startsWeightsnone (full counts)PopulationresidentsPeriod2000 to 2024SourceCounty demographics panel or older −0.34−0.34 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with share aged 65 and overInterval95% CI −0.45 to −0.23n3,103Weightstwo-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 the share under 3030 years of ageEstimandage below which the younger share countsWeightsnone (full counts)PopulationresidentsPeriod2000 to 2024SourceCounty demographics panel +0.12 (95% CI 0.00 to 0.25)+0.12 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with share under 30Interval95% CI 0.00 to 0.25n3,103Weightstwo-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 8
The correlation depends on the measure within an axis in 2024, from −0.03 for the share without a high school diploma to +0.66 for college share, vote-weighted
Correlation of fifteen county measures, grouped by the five axes, with the county two-party margin in 2024, vote-weighted and county-equal
- Vote-weighted
- County-equal
Education
Income
Race and ethnicity
Age
Rurality
Correlation with the county two-party margin (Democratic positive)
Show data
| Axis | Measure | Weighting | Correlation | Interval low | Interval high | Counties |
|---|---|---|---|---|---|---|
| Education | Bachelor's degree or more | Vote-weighted | +0.66 | +0.60 | +0.72 | 3,103 |
| Education | Bachelor's degree or more | County-equal | +0.56 | +0.47 | +0.64 | 3,103 |
| Education | No high school diploma | Vote-weighted | −0.03 | −0.20 | +0.14 | 3,103 |
| Education | No high school diploma | County-equal | −0.15 | −0.26 | −0.04 | 3,103 |
| Income | Median household income | Vote-weighted | +0.40 | +0.29 | +0.50 | 3,103 |
| Income | Median household income | County-equal | +0.30 | +0.17 | +0.43 | 3,103 |
| Income | Per capita income | Vote-weighted | +0.55 | +0.47 | +0.64 | 3,103 |
| Income | Per capita income | County-equal | +0.39 | +0.27 | +0.51 | 3,103 |
| Income | Poverty rate | Vote-weighted | −0.03 | −0.14 | +0.09 | 3,103 |
| Income | Poverty rate | County-equal | +0.05 | −0.09 | +0.20 | 3,103 |
| Race and ethnicity | Non-Hispanic white share | Vote-weighted | −0.57 | −0.71 | −0.44 | 3,103 |
| Race and ethnicity | Non-Hispanic white share | County-equal | −0.50 | −0.71 | −0.28 | 3,103 |
| Race and ethnicity | Non-Hispanic Black share | Vote-weighted | +0.39 | +0.24 | +0.54 | 3,103 |
| Race and ethnicity | Non-Hispanic Black share | County-equal | +0.44 | +0.35 | +0.52 | 3,103 |
| Race and ethnicity | Hispanic share | Vote-weighted | +0.24 | +0.10 | +0.38 | 3,103 |
| Race and ethnicity | Hispanic share | County-equal | +0.10 | −0.09 | +0.30 | 3,103 |
| Race and ethnicity | Non-Hispanic Asian share | Vote-weighted | +0.53 | +0.36 | +0.69 | 3,103 |
| Race and ethnicity | Non-Hispanic Asian share | County-equal | +0.45 | +0.31 | +0.58 | 3,103 |
| Race and ethnicity | Foreign-born share | Vote-weighted | +0.54 | +0.36 | +0.72 | 3,103 |
| Race and ethnicity | Foreign-born share | County-equal | +0.37 | +0.22 | +0.53 | 3,103 |
| Age | Median age | Vote-weighted | −0.31 | −0.41 | −0.21 | 3,103 |
| Age | Median age | County-equal | −0.20 | −0.27 | −0.12 | 3,103 |
| Age | Share aged 65 and over | Vote-weighted | −0.34 | −0.45 | −0.23 | 3,103 |
| Age | Share aged 65 and over | County-equal | −0.20 | −0.28 | −0.12 | 3,103 |
| Age | Share under 30 | Vote-weighted | +0.13 | 0.00 | +0.25 | 3,103 |
| Age | Share under 30 | County-equal | +0.13 | +0.05 | +0.21 | 3,103 |
| Rurality | Rural share | Vote-weighted | −0.65 | −0.74 | −0.56 | 3,103 |
| Rurality | Rural share | County-equal | −0.53 | −0.59 | −0.46 | 3,103 |
| Rurality | Rural-urban continuum code | Vote-weighted | −0.51 | −0.58 | −0.44 | 3,103 |
| Rurality | Rural-urban continuum code | County-equal | −0.41 | −0.49 | −0.34 | 3,103 |
Race and ethnicity is measured by five shares here. The white share has −0.57 (95% CI −0.71 to −0.44)−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 the Black, Hispanic, Asian and foreign-born shares have +0.39+0.39 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic Black shareInterval95% CI 0.24 to 0.54n3,103Weightstwo-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., +0.24 (95% CI 0.10 to 0.38)+0.24 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with hispanic shareInterval95% CI 0.10 to 0.38n3,103Weightstwo-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., +0.52+0.52 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic Asian shareInterval95% CI 0.36 to 0.69n3,103Weightstwo-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 +0.54+0.54 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with foreign-born shareInterval95% CI 0.37 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.. Counting each county once, the Hispanic share's correlation is +0.11 (95% CI −0.09 to 0.30)+0.11 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with hispanic shareInterval95% CI −0.09 to 0.30n3,103Weightsnone (each county counts once)Populationcounties 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. Per capita income (+0.55+0.55 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with per capita incomeInterval95% CI 0.47 to 0.64n3,103Weightstwo-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.), the Asian share (+0.52+0.52 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with non-Hispanic Asian shareInterval95% CI 0.36 to 0.69n3,103Weightstwo-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 the foreign-born share (+0.54+0.54 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with foreign-born shareInterval95% CI 0.37 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.) have correlations about the size of the white share's. The ranking of the five axes in the first figure is therefore a ranking of five chosen measures, and a different measure of an axis can change it. Chapter 3.4, Hispanic, Black and Asian voters (planned) and Chapter 12.7, Education, ethnicity and the county shift, 2016 to 2024 follow the group shares, and Chapter 3.6, Age and generation takes up age.
The rural-urban continuum code is the other common measure of rurality. Its 2013 edition covers the elections through 2016 and its 2023 edition covers 2020 and 2024. Vote-weighted, its correlation with the margin is −0.51−0.51 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with the rural-urban continuum code (edition 2013)n3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panel in 2016 and −0.52−0.52 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with the rural-urban continuum code (edition 2023)n3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panel in 2020, and that of rural share is −0.65−0.65 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural share of residentsn3,111Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2016 presidential electionSourceCounty demographics panel and −0.67−0.67 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural share of residentsn3,103Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2020 presidential electionSourceCounty demographics panel, so the change of edition leaves both series without a break. The code's correlation in 2024 is −0.51 (95% CI −0.58 to −0.44)−0.51 correlation coefficient, from −1 to +1Estimandcorrelation of the county two-party margin with rural-urban continuum codeInterval95% CI −0.58 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., smaller in size than rural share's −0.65 (95% CI −0.74 to −0.56)−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..
The gaps from the fit run by region
Figure 9 maps the gap between each county's margin and the margin fitted by the vote-weighted five-axis regression, estimated separately for 2000 and 2024. A negative gap means the county voted more Republican than the five axes fit. The standard deviation of the gap is 20.420.4 percentage points of marginEstimandvote-weighted standard deviation of the five-axis residualWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panel points in 2000 and 18.418.4 percentage points of marginEstimandvote-weighted standard deviation of the five-axis residualWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel in 2024, and counties more than 2020 percentage points of marginEstimanddistance from the fitted margin used to flag a countyWeightsnone (full counts)PopulationcountiesPeriod2000 to 2024SourceCounty election returns points from the fit cast 35%35% share of two-party votesEstimandshare of two-party votes in counties whose margin is more than 20 points from the five-axis fitWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panel of the votes in 2000 and 25%25% share of two-party votesEstimandshare of two-party votes in counties whose margin is more than 20 points from the five-axis fitWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel in 2024.
Figure 9
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
| Election | Position | County | Margin minus fit (points) | Margin (points) | Fitted margin (points) | Two-party votes |
|---|---|---|---|---|---|---|
| 2024 | Largest gaps below the fit | Forsyth, GA | −52.9 | −33.5 | +19.4 | 136,790 |
| 2024 | Largest gaps below the fit | Williamson, TN | −52.1 | −32.9 | +19.2 | 142,257 |
| 2024 | Largest gaps below the fit | Shelby, AL | −47.4 | −41.3 | +6.1 | 112,753 |
| 2024 | Largest gaps below the fit | Comal, TX | −46.8 | −46.0 | +0.9 | 102,436 |
| 2024 | Largest gaps below the fit | Miami-Dade, FL | −46.8 | −11.5 | +35.3 | 1,085,945 |
| 2024 | Largest gaps below the fit | Lubbock, TX | −46.2 | −39.9 | +6.2 | 123,695 |
| 2024 | Largest gaps below the fit | Montgomery, TX | −45.2 | −45.9 | −0.8 | 304,241 |
| 2024 | Largest gaps below the fit | Collin, TX | −45.0 | −11.4 | +33.5 | 501,649 |
| 2024 | Largest gaps above the fit | Clayton, GA | +52.2 | +69.6 | +17.4 | 111,080 |
| 2024 | Largest gaps above the fit | Sonoma, CA | +48.2 | +47.8 | −0.4 | 243,026 |
| 2024 | Largest gaps above the fit | Prince George's, MD | +48.1 | +77.0 | +29.0 | 392,046 |
| 2024 | Largest gaps above the fit | Santa Cruz, CA | +46.8 | +56.6 | +9.8 | 128,976 |
| 2024 | Largest gaps above the fit | Baltimore City, MD | +42.7 | +74.9 | +32.2 | 223,093 |
| 2024 | Largest gaps above the fit | Multnomah, OR | +42.5 | +64.3 | +21.8 | 396,686 |
| 2024 | Largest gaps above the fit | Marin, CA | +40.1 | +65.7 | +25.6 | 140,206 |
| 2024 | Largest gaps above the fit | St. Louis, MN | +36.8 | +14.0 | −22.8 | 116,400 |
| 2000 | Largest gaps below the fit | Collin, TX | −49.0 | −49.9 | −0.9 | 171,063 |
| 2000 | Largest gaps below the fit | Montgomery, TX | −40.5 | −55.2 | −14.7 | 103,886 |
| 2000 | Largest gaps below the fit | Utah, UT | −39.5 | −71.3 | −31.8 | 114,700 |
| 2000 | Largest gaps below the fit | Greenville, SC | −37.0 | −35.8 | +1.2 | 136,524 |
| 2000 | Largest gaps below the fit | Sarasota, FL | −36.9 | −6.6 | +30.4 | 155,986 |
| 2000 | Largest gaps below the fit | Fort Bend, TX | −36.9 | −21.5 | +15.4 | 121,136 |
| 2000 | Largest gaps below the fit | Lee, FL | −35.9 | −18.1 | +17.7 | 179,722 |
| 2000 | Largest gaps below the fit | Denton, TX | −35.7 | −43.6 | −7.9 | 142,315 |
| 2000 | Largest gaps above the fit | Bristol, MA | +45.7 | +36.9 | −8.8 | 199,173 |
| 2000 | Largest gaps above the fit | Bronx, NY | +42.9 | +76.0 | +33.1 | 302,046 |
| 2000 | Largest gaps above the fit | Philadelphia, PA | +40.6 | +63.3 | +22.7 | 550,141 |
| 2000 | Largest gaps above the fit | Kings, NY | +40.0 | +67.5 | +27.5 | 594,122 |
| 2000 | Largest gaps above the fit | Suffolk, MA | +38.3 | +55.4 | +17.1 | 199,329 |
| 2000 | Largest gaps above the fit | Baltimore City, MD | +37.5 | +70.8 | +33.3 | 185,915 |
| 2000 | Largest gaps above the fit | District Of Columbia , DC | +36.8 | +81.0 | +44.2 | 189,996 |
| 2000 | Largest gaps above the fit | Providence, RI | +36.6 | +39.8 | +3.1 | 203,847 |
The gaps run by region. In 2024 the vote-weighted mean gap 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, the Census Bureau's South region, 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 in the rest of the country, against −11.5−11.5 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression among counties in the Census Southn1,423Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panel and +6.0+6.0 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression among counties outside the Census Southn1,690Weightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2000 presidential electionSourceCounty demographics panel in 2000. By state, Texas is −21.7−21.7 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression in TX countiesWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel and Vermont +42.6+42.6 percentage points of marginEstimandvote-weighted mean residual of the five-axis regression in VT countiesWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel in 2024. State baselines absorb the state-level part of these gaps, and the vote-weighted five-axis R-squared goes from 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. to 87%87% share of the variance of the county marginEstimandR-squared of the county margin on the five axes and a baseline for each stateInterval95% CI 84% to 90%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 when they are added. Among counties with 100,000100,000 two-party votesEstimandvote cut for naming countiesWeightsnone (full counts)PopulationcountiesPeriod2000 to 2024SourceCounty election returns or more two-party votes, the three largest gaps above the fit in 2024 are Clayton County, Georgia (+52.2+52.2 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel), Sonoma County, California (+48.2+48.2 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel) and Prince George's County, Maryland (+48.1+48.1 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel). The three largest below are Forsyth County, Georgia (−52.9−52.9 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel), Williamson County, Tennessee (−52.1−52.1 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel) and Shelby County, Alabama (−47.4−47.4 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel). Miami-Dade County, Florida, is fitted at +35.3+35.3 percentage points of marginEstimandcounty margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel and voted −11.5−11.5 percentage points of marginEstimandcounty two-party margin, Democratic positiveWeightsnone (full counts)Populationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel, a gap of −46.8−46.8 percentage points of marginEstimandcounty margin minus the margin fitted by the vote-weighted five-axis regressionWeightstwo-party votes in the electionPopulationcounties with returns and a complete set of census measures, Alaska excludedPeriod2024 presidential electionSourceCounty demographics panel points.
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. Chapter 15.1, Party and place compares partisans in counties with different margins, and Chapter 12.7, Education, ethnicity and the county shift, 2016 to 2024 follows the high-Hispanic counties across 2016 to 2024.
What these data cannot show
Every statistic above is a property of counties. 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., and Robinson [6] showed that a correlation across areas can have the opposite sign from the correlation across individuals. The characteristics describe residents, including residents who are not citizens or not of voting age, and the margin describes voters. A sign such as income's negative coefficient is a feature of one regression among counties and says nothing about how households of different incomes voted.
The five axes are five chosen measures. The ranking changes with the measure within an axis (Figure 8), and the axes omit religion, occupation, density, the age of housing and every characteristic that is not in the county panel. A standardized coefficient depends on the spread of the axis across counties, which differs between the weightings and between elections, so a coefficient that rises can reflect a change in spread as well as a steeper slope. The R-squared values are in-sample fits of linear regressions and not forecasts. State baselines absorb every difference between states, including differences in composition that sit at the state level, so the within-state results describe counties compared inside states and not the effect of removing state context.
The census measures are decennial census and American Community Survey estimates interpolated to election years. Only 2000 and 2024 are close to a direct measurement, and the elections between are blends of adjacent releases, so the measures for those years carry sampling error and interpolation error that no interval includes. Rural share is anchored to the decennial censuses and has one value for 2020 and 2024. The county set is not harmonized across boundary changes, a few small counties lack measures in some years, and the counties are the units of the returns and not of any community.
The intervals show dependence on which states are in the set. They are not sampling error of the vote, which is a full count, and the bootstrap intervals for differences between statistics treat the states as the only source of variation. 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.
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.
Methods
Sample: presidential returns for the 7 elections from 2000 to 2024 (Algara and Amlani for 2000 to 2020, MIT Election Data and Science Lab for 2024), from the repaired county file whose 2024 totals are corrected for 173 counties in South Carolina, Wisconsin and West Virginia. The 41 Alaska rows (election districts) are dropped before any calculation; in these years no other row lacks a FIPS code, a vote count or two-party votes. Kansas City, Missouri is a separate row (FIPS 2938000) in 2024 only; its votes are added to Jackson County (29095), which holds them in 2000 to 2020 and whose census measures cover the whole county. The 2024 returns carry Oglala Lakota County, South Dakota, under its former code 46113, and it is recoded to 46102 to match the census panel. An election's sample is the set of counties with two-party votes and a complete set of the census measures below (3,103 to 3,113 counties); every comparison in an election uses that same set, and the sample is not harmonized across boundary changes. Counties that lack a census measure are left out: 10 county rows in 2020 and 9 in 2024. Among them are the eight old Connecticut counties, for which the census panel has no 2020 or 2024 values (it fills the new planning-region rows only at the 2024 anchor); they hold 1.16% and 1.14% of the two-party votes in 2020 and 2024. The rest are single counties with a missing income or rural-urban code, among them Clifton Forge city, Virginia, in 2000 (no rural-urban code, which every county in the sample needs) and Oglala Lakota County in 2012 and 2016. Outcome: the county two-party margin, 100 times the Democratic votes minus the Republican votes over their sum, so a positive value favors the Democrats. Axes and measures: education is the share of adults 25 and older with a bachelor's degree or more; income is median household income in 2024 dollars; race and ethnicity is the non-Hispanic white share of residents; age is median age; rurality is the share of residents living in rural areas. The nine-measure model adds the non-Hispanic Black, Hispanic and non-Hispanic Asian shares and the foreign-born share to the five axes. The 2024 comparison of measures adds the share without a high school diploma, per capita income, the poverty rate, the share aged 65 and over, the share under 30 and the rural-urban continuum code (1 most urban to 9 most rural). Census measures come from the county demographics panel, which interpolates between anchors: 2000 is the decennial census long form, 2004 and 2008 interpolate between it and the American Community Survey release for 2012, 2012, 2016 and 2020 are midpoints of adjacent releases, and 2024 is the 2024 release; so only 2000 and 2024 are close to a direct measurement and the other years carry interpolation and sampling error that no interval includes. Rural share is anchored to the decennial censuses and has one value for 2020 and 2024; the census definition of rural changed with the 2020 census, and the mean rural share in the sample (county-equal, and weighted by county population) is reported for 2012, 2016 and 2020. The rural-urban continuum code uses the 2013 edition through 2016 and the 2023 edition for 2020 and 2024. Weights: two-party votes in the election (vote-weighted) or one per county (county-equal); both are computed for every statistic. Coefficients: weighted least squares of the standardized margin on standardized axes, standardized with the weighted mean and standard deviation of the weighting used, so a one-variable coefficient equals the correlation; the joint regression uses the other four axes, and the state-baseline regression adds an indicator for each state and the District of Columbia. R-squared is the weighted share of the variance of the margin reproduced by the fitted values; the share within states is the R-squared with baselines minus that of baselines alone, over one minus the latter; the unique contribution of an axis is the R-squared of the five-axis regression minus that of the regression without the axis, in points of variance. Spreads are weighted standard deviations of the margin, of the five-axis fitted values and of the residual, with and without state baselines; the 2024 R-squared with the left-over spread held at its 2000 value is one minus the squared ratio of the 2000 left-over spread to the 2024 spread of the margin. Fixed-measure checks: the same statistics with the 2024 census measures, and with the 2000 census measures, used for every election, on the counties that have them. Ranks: the share of resamples in which an axis has the largest absolute correlation (alone), the largest absolute coefficient in the five-axis regression, or the largest absolute coefficient with state baselines; and the difference in absolute size between two axes, with its interval. Intervals: normal 95% intervals with errors clustered by state (49 to 50 clusters) for correlations and coefficients; 95% percentile intervals from a bootstrap over whole states for R-squared, spreads, unique contributions, differences between two statistics, rank shares and rank differences (2,000 replicates, seed 12345; a state drawn twice counts as two clusters; the weightings share draws; changes between elections draw the same states for both elections). Returns are full counts, so the intervals show dependence on which states are in the set and not sampling error of the vote, and they do not include the sampling error of the census measures. Map: the residual of the vote-weighted five-axis regression, estimated separately for 2000 and 2024, in points of margin; the Census South is the Census Bureau's South region (16 states and the District of Columbia). Suppression: none, since all figures are county-level counts or fits.
References
- Cameron, A. Colin; Miller, Douglas L. (2015). A Practitioner's Guide to Cluster-Robust Inference. Journal of Human Resources 50(2): 317-372. doi:10.3368/jhr.50.2.317
- 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
- Scala, Dante J.; Johnson, Kenneth M. (2017). Political Polarization along the Rural-Urban Continuum? The Geography of the Presidential Vote, 2000-2016. The ANNALS of the American Academy of Political and Social Science 672(1): 162-184. doi:10.1177/0002716217712696
- 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
- Solon, Gary; Haider, Steven J.; Wooldridge, Jeffrey M. (2015). What Are We Weighting For?. Journal of Human Resources 50(2): 301-316. doi:10.3368/jhr.50.2.301
- 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.}
}